Pseudogenes

"The question is not whether some DNA regions that look like pseudogenes have a function.  This is a fact.  The real question is whether these regions are exceptions to the general rule that a pseudogene is a pseudogene or whether most pseudogenes have been mislabeled." — Laurence Moran, biochemist, 2020¹

"The strongest statement that could be warranted by the evidence would be that we do not currently know of a function for the vitamin C pseudogene." — Jonathan Wells, The Myth of Junk DNA, 2011²

Summary

A pseudogene is a stretch of DNA that looks like a gene but carries damage that should stop it from making the gene's product:  a stop signal in the middle, a shifted reading frame, or a missing start.  For forty years pseudogenes were the textbook example of junk DNA.  Since about 2000, papers have reported pseudogenes that are copied into RNA, that regulate their parent genes, and in a few cases that are needed for a cell to work.  Design writers now say that most pseudogenes probably have a function.  Their critics say that a handful were co-opted and the rest are dead.

This article counts.  The short answer is that both slogans outrun the data.  A small number of pseudogenes have a function shown by experiment.  A larger minority shows signs of activity or of constraint.  Most have never been tested by any experiment at all, and the tests that have been run on them as a group look like what neutral decay would produce.

1. There are about 14,700 human pseudogenes, and together they are a tiny slice of the genome. GENCODE's current human annotation, release 50 of June 2026, lists 14,702: 10,634 processed, 3,535 unprocessed (duplicated) and 296 unitary.³ Counted for this article from the same release, their exons cover about 11 million bases, 0.36% of the genome, and 1.3% if the introns of duplicated copies are included.⁴ If every pseudogene in the genome turned out to work, the total amount of functional DNA would barely move. The fight over pseudogenes is about what they show, not about how much DNA they are.
2. Somewhere between one in twelve and one in five is copied into RNA, depending on the method. The GENCODE team found 876 transcribed pseudogenes among 10,523 studied, and called that conservative.⁵ The ENCODE pilot found 38 of 201, 19%.⁶ A three-phyla comparison found about 15% transcribed in humans, worms and flies alike.⁷ Long-read sequencing finds many transcribed pseudogenes the reference annotation misses.⁸ Transcription is not function, and researchers on both sides of the junk-DNA debate say so.⁹
3. A few are translated, and a few of those are under selection. Ribosome profiling found 155 of 426 expressed pseudogenes translated into peptides longer than ten amino acids.¹⁰ But when Xu and Zhang compared human and macaque copies, only five of 34 translated pseudogenes showed the signature of protein-level selection. Across all pseudogenes the rate was 4.9% of transcribed ones and 0.49% of untranscribed ones.¹¹
4. Most processed pseudogenes are young, and few are shared with mice. Only 129 human pseudogenes, about 1%, have mouse orthologs.⁷ For ribosomal-protein pseudogenes, the largest family, 1,282 sit at matching places in human and chimpanzee and 6 in human and mouse.¹² The GENCODE team counted 1,019 human pseudogenes as conserved, 28.1% of duplicated ones and 3.4% of processed ones.⁵
5. A widely repeated design argument inverts its own source. The 2017 Crossway volume Theistic Evolution says that 60% of human processed pseudogenes "are very similar in mouse and human" and that this implies "a sequence-dependent essential function."¹³ It cites a 2012 review, which cites a 2004 paper. The 2004 paper's abstract says the reverse: of about 5,000 mouse processed pseudogenes, "approximately 60% are lineage specific, created after the mouse and human diverged."¹⁴
6. Measured function exists, but the list is short and uneven. The cleanest cases change DNA and watch what happens. In chickens, 25 broken antibody-gene copies are the spare-part donors from which the bird builds its antibody variety;¹⁵ deleting them in a chicken B-cell line stops the process.¹⁶ In human red-cell precursors, deleting the region of the beta-globin pseudogene HBBP1 switches fetal haemoglobin back on,¹⁷ and a 2021 study found HBBP1's RNA needed for red cells to form.¹⁸ The famous PTENP1 tumour-suppressor result replicated only in part when an independent team repeated it in 2020.¹⁹ The famous Makorin1-p1 result in mice did not survive: a 2006 study found that the pseudogene "is not transcribed".²⁰ A 2021 screen that silenced 850 expressed pseudogenes in a breast-cancer cell line found 69 whose loss slowed growth.²¹ No database of experimentally confirmed functional pseudogenes was found. Individually tested cases number a few dozen, not thousands.
7. Of the cases named in this article's brief, one is a real relic-turned-useful, and three are not. HBBP1 is the real one, shown by deletion. The human vitellogenin (egg-yolk gene) remnant has no measured function; the claim that it is an enhancer rests on a gene prediction and a patent.²² Enamelin has no design-side function claim at all. For GULO, the vitamin C gene, no design-side writer offers data for a function, the leading creationist paper on it says the region is "transcriptionally inactive",²³ and Wells, quoted above, concedes the point.
8. Each side has a habit to break. Dan Graur's team wrote that pseudogenes "by definition" are nonfunctional, so any pseudogene found to work is "by definition" a gene.²⁴ That makes the junk claim true by definition and untestable. On the other side, design writers have treated transcription as function and counted conservation that is not there. Jerry Bergman called transcription one of the "levels of function" ENCODE documented,²⁵ and the 2017 volume's mouse-human figure, point 5, is the wrong way round.
9. The design case does not need most pseudogenes to work. A young-earth model expects some genes to have broken since creation. This site's Functional DNA is Evidence of Design says so: "The only junk DNA would be that which has been destroyed by mutations since the original design." Unitary pseudogenes like GULO fit that reading. What a young-earth model owes is an account of why these broken copies share positions and lesions across species, and that question belongs to Shared Genetic Mistakes and Common Ancestry and Shared Retroviruses and Pseudogenes.

Verdict in one paragraph.  A fraction of pseudogenes with measured function, well under 1%.  A larger fraction, roughly 10% to 30% depending on the measure, with transcription, chromatin activity or constraint.  A remainder, most of the total, with no evidence of function, and never tested one by one.  The claim that most pseudogenes are functional is not supported.  The claim that pseudogenes are nonfunctional "by definition" is not science.  The honest statement is the one Moran and Wells each made about the part of the question they were willing to concede.

Part 1: What a pseudogene is, and how many there are

Three ways to make a broken copy

Laurence Moran, a University of Toronto biochemist and the most persistent public defender of junk DNA, describes four kinds.²⁶

1. Duplicated (unprocessed) pseudogenes. A stretch of chromosome holding a gene is copied, and one copy picks up a disabling mutation. The copy keeps its introns and often its promoter, so it may still be transcribed. It usually sits near its parent gene.
2. Processed pseudogenes. A cell's own reverse transcriptase, mostly the machinery of LINE-1 elements, copies a finished messenger RNA back into DNA and pastes it somewhere else in the genome. The copy has no introns and no promoter, so it is usually "dead on arrival". For a processed pseudogene to be inherited, the copying has to happen in the germ line, so most come from genes that are busy in sperm or egg precursors.
3. Unitary pseudogenes. A single-copy gene breaks in place and is not replaced. There is no working copy left in the genome. The human vitamin C gene, GULO, is the classic case.
4. Polymorphic pseudogenes. A unitary pseudogene that is broken in some people and working in others.

The difference matters for the function question.  A processed pseudogene never had a working promoter in its new home, so if it does anything, it acquired that job after it landed.  A duplicated or unitary pseudogene inherited a working gene's controls, so it can keep doing part of the old job, or some of its DNA can keep a role that never depended on the protein.

The three kinds also carry different expectations for a young-earth reading.  A unitary pseudogene is a created gene that broke, so that reading expects no function, and GULO shows none (Part 4).  A duplicated pseudogene could be a created copy with a job of its own, and the data lean that way:  the GENCODE team found conservation in 28.1% of duplicated pseudogenes against 3.4% of processed ones,⁵ and the two strongest measured cases, the chicken antibody donors and HBBP1, are duplicated copies that sit beside their working relatives (Parts 3 and 4).  The mainstream reading expects the same pattern, because duplicates keep their promoters, so the pattern does not decide between the two.  Processed pseudogenes, the majority, are the hard case for a young-earth model, for reasons set out in Part 5.

How they are found

Pseudogenes are found by computer, not by experiment.  A program takes the known proteins of a species, searches the genome for stretches that resemble them, and labels as a pseudogene any match that has a premature stop codon, a frameshift, a missing start, or broken splice sites.  Processed copies are recognised by their missing introns and the remains of a poly-A tail.  GENCODE, the reference annotation used by Ensembl and ENCODE, combines two automated pipelines with manual curation by the HAVANA team.⁵

So the label "pseudogene" is a prediction from sequence, not a measurement of activity.  That is not a scandal;  every gene annotation starts the same way.  It does mean the count depends on the program and on what it is told to look for.  Moran notes that early efforts on the human genome "gave results ranging from more than 18,000 to a little over 11,000 depending on the program that was used", and that because the different programs overlapped so little, "the actual number of pseudogenes could have been as low as 7,000".²⁶

The counts

For comparison, the same GENCODE release lists 19,442 protein-coding genes.³  So the human genome holds roughly three pseudogenes for every four working protein genes.  The counts have converged near 14,000 to 15,000 for pseudogenes of protein-coding genes.  Pseudogenes of RNA genes are harder to see and mostly uncounted;  Moran notes that the Alu repeats began as copies of a small RNA gene, 7SL.²⁹

How much DNA they are

Moran has given two figures for how much of the genome pseudogenes occupy.  In 2012 he wrote "Pseudogenes represent about 1% of the genome".³⁰  In 2020 he wrote that "There are roughly 15,000 annotated pseudogenes in the human genome and they cover about 5% of the genome."¹

This article measured it.  A short script, research/gencode-pseudogene-count.ts, reads the GENCODE release 50 annotation file and merges the coordinates of every gene typed as a pseudogene.  It reproduces GENCODE's own published counts exactly:  10,634 processed, 3,535 unprocessed and 296 unitary.  The pseudogenes' own exon sequence covers 11.0 million bases, 0.36% of a 3.1 billion base genome.  Adding the introns that duplicated and unitary pseudogenes keep gives 41.2 million bases, 1.33%.⁴  Moran's 2012 figure is right;  his 2020 figure is about four times too high even on the generous measure.

This number cuts both ways, and both sides should hear it.  For the design side, it means pseudogenes cannot carry much weight in the argument over how much of the genome is functional.  This site's Functional DNA is Evidence of Design argues from transcription, protein binding, disease association and conservation across the whole genome;  if every pseudogene were proved functional tomorrow, those totals would rise by less than half a percentage point of exon sequence.  For the junk side, it means pseudogenes are not a large part of the junk either.  The reason both sides care about them is not their bulk.  It is that a broken gene, if it really is broken, looks like a scar left by history rather than a part put there for a purpose.

Unitary pseudogenes and genes that came back

Unitary pseudogenes are the ones most often argued about, because each marks a function the species lost.  Zhang and colleagues at Yale searched for genes that work in other mammals but are broken in humans, and found "76 unitary pseudogenes" in the human lineage since the split from rodents.  They found that the losses were spread out "at a fairly uniform rate throughout primate evolution".  And they found 11 that are polymorphic, broken in some people and intact in others, of which "two of them are in fact pseudogenes in non-human primates, suggesting that they represent cases of a gene being resurrected in the human lineage."³¹  GENCODE now annotates 296 unitary pseudogenes, after better methods raised the count.³

The resurrection finding is a reminder that "pseudogene" is a label for the state of a sequence, not a verdict on its history.  A gene can break and be repaired by a later mutation.

Part 2: Transcribed, translated, conserved

Three kinds of evidence are offered for function across pseudogenes as a class:  they are copied into RNA, some are translated into protein, and some are conserved between species.  Each is real.  Each is weaker than it sounds, for reasons the researchers who measured it state themselves.

Transcription: one in twelve to one in five

The spread comes from where and how hard each team looked.  Cancer cells and testis switch on a great deal of DNA that is silent elsewhere, and long-read sequencing catches transcripts that short reads assign to the parent gene.  Troskie and colleagues, using long reads, found transcripts from 521 processed pseudogenes, including 220 transcribed in the same direction as their parent and independent of any known gene, "only 43 of which were previously annotated as transcribed pseudogenes in Gencode."⁸  So the true transcribed fraction in some tissue at some stage is probably higher than the GENCODE flag.

Some of these transcripts are pieces of longer non-coding RNAs.  Michael Milligan, Leonard Lipovich and colleagues laid the human pseudogene catalogues over the catalogues of long non-coding RNA genes and found "2277 loci containing exon-to-exon overlaps between pseudogenes, both processed and unprocessed, and long non-coding RNA genes", 1,167 of them with direct evidence of transcription.  Their merged catalogue holds about 20,945 pseudogene loci, so that is roughly 5% to 11%, in line with the table above.  From the overlaps they concluded that pseudogenes "are pervasively transcribed on both strands and are common drivers of gene regulation."³⁴  The count is a useful measurement.  The conclusion goes further than an overlap can carry:  a pseudogene that sits inside a non-coding RNA is transcribed, and whether it drives anything has to be tested one case at a time.  Overstatement on the function side is not confined to design writers.

Transcription is not function.  Svensson, Arvestad and Lagergren, who were looking for functional pseudogenes, wrote it plainly:  "That a pseudogene is transcribed is not sufficient evidence of biological function."⁹  RNA polymerase transcribes a great deal of DNA at a low rate, and a pseudogene that lands inside an intron is transcribed along with its host gene whether it does anything or not.  The ENCODE pilot also tested whether transcribed pseudogenes are more constrained than untranscribed ones.  They are not:  "the transcribed pseudogenes did not show significantly different evolutionary constraints compared to those not transcribed".⁶  If transcription marked function, it should have marked constraint too.

Design writers have not always kept this line.  Jerry Bergman, writing in the Creation Research Society Quarterly, cited ENCODE as having "documented several levels of function in many pseudogenes (Birney, 2012), including transcription".²⁵  Transcription is one of the things ENCODE measured.  Calling it a level of function assumes the point at issue.

Translation: real, but mostly not selected

Some pseudogene transcripts are read by ribosomes.  Ji, Song, Regev and Struhl used ribosome profiling and found that "out of 426 expressed pseudogenes (~3% of those annotated), 155 (36%) are translated into peptides longer than 10 aa."¹⁰  Mass spectrometry has found peptides from 107 human pseudogenes.¹¹

Jinrui Xu and Jianzhi Zhang at the University of Michigan asked whether those peptides matter.  If a pseudogene's protein does a job, mutations that change its amino acids should be weeded out, so human and macaque copies should differ more at silent sites than at amino-acid sites.  "Five of 34 translated pseudogenes amenable to evolutionary analysis" show that signature.  Across the genome, they found it in "4.9% of transcribed pseudogenes and 0.49% of nontranscribed pseudogenes".¹¹  Their conclusion is the fairest sentence in this literature:  "neither ORF disruption nor presence of protein product disproves or proves gene functionality at the protein level."¹¹

Conservation: mostly lineage-specific

If a pseudogene had an important job, mutations that damaged it would be weeded out, and its sequence would change more slowly than neutral DNA.  This is the argument from conservation, and it is the strongest evidence either side has, because it does not depend on any one experiment.

Measured across the class, pseudogenes mostly decay at the neutral rate.  Torrents and colleagues judged "95% ± 3%" of 19,724 pseudogene regions to "evolve neutrally".²⁷  The ENCODE pilot found that "The overall sequence decay rate of pseudogenes is very similar to that of neutrally evolving DNA."⁶  Sisu and colleagues found no excess of rare variants in human pseudogenes that would mark selection in the population.⁷

Both sides should note what these tests can see.  Torrents's test compares protein-changing with silent substitutions.  It detects selection on a protein.  A pseudogene whose job is done by its RNA or its DNA would be constrained at both kinds of site alike, and this test would score it as neutral.  The ENCODE pilot's decay-rate comparison and Pei's divergence test do not have that blind spot, and they find constraint in a minority.

The blind spot is the creationist writer John Woodmorappe's main point about pseudogenes.  In 2004 he wrote that "the potential or actual function of a pseudogene can only be determined by direct experiment, and it is high time that evolutionists abandoned KA/Ks as a ground for the a priori discounting of pseudogene function."  KA/KS is the ratio of protein-changing to silent substitutions that Torrents used.  In Woodmorappe's view, pseudogenes "appear to lack function primarily because very few of them have been carefully studied for function.  It is no more complicated than that!"³⁵  He is right that the ratio test cannot see a job done by RNA or DNA, and right that experiment decides.  But the tests without that blind spot also find most pseudogenes decaying at about the neutral rate.  His point keeps the question open for the untested majority.  It is not evidence that they work.

The minority is real.  Pei and colleagues compared each human pseudogene with its counterparts in chimpanzee and mouse and found "1,019 conserved pseudogenes", enriched for transcribed ones, "strongly implying biological function".  The split by type matters:  "28.1% of duplicated pseudogenes and 3.4% of processed pseudogenes are conserved", because "most processed pseudogenes are lineage-specific, and also that most of them are dead on arrival."⁵  Khachane and Harrison found 1,750 transcribed pseudogenes with mid-sequence damage;  "About half of the human TPAs are conserved in rhesus monkey, but strikingly, very few in mouse (~3%)", and they named "68 strong candidates" for function across several mammals.³⁶  Balasubramanian and colleagues reported a point for the function side in the largest pseudogene family:  "syntenic pseudogenes have a lower rate of nucleotide substitution than their surrounding intergenic DNA."¹²

How many human pseudogenes are shared with mice at all?  Very few.  Sisu and colleagues:  "only 129 (∼1%) of the human pseudogenes have mouse orthologs."⁷  For ribosomal-protein pseudogenes, Balasubramanian and colleagues counted 1,282 at matching places in human and chimpanzee, 6 in human and mouse, and 11 in human and rat.¹²  Svensson's search for pseudogenes conserved since before the primate-rodent split found 30 candidates and concluded that "functional pseudogenes are relatively rare on a long evolutionary timescale, they nevertheless exist."⁹

A "60 percent" that is the other way round

The 2017 Crossway volume Theistic Evolution, in a chapter by Ann Gauger, Ola Hössjer and Colin Reeves, gives the design side's conservation argument in one paragraph:

"There are more than eight thousand processed pseudogenes in the human genome; 60 percent are very similar in mouse and human.  The fact that their sequences are so similar means that they are likely to have a sequence-dependent essential function, so that their sequence cannot be changed without harm."¹³

The note cites Wen and colleagues' 2012 review "Pseudogenes are not pseudo any more".  Wen's sentence is:  "comparative analysis of processed pseudogenes in the mouse and human genomes has surprisingly demonstrated that 60% of the processed pseudogenes are conserved in both mammalian species."³⁷  Wen cites Zhang, Carriero and Gerstein's 2004 paper in Trends in Genetics.  That paper's abstract says:  "We have systematically identified approximately 5000 processed pseudogenes in the mouse genome, and estimated that approximately 60% are lineage specific, created after the mouse and human diverged."¹⁴

Lineage-specific means not shared.  The source says 60% of mouse processed pseudogenes arose after the split and so have no human counterpart;  the review turned that into 60% conserved in both, and the book into 60% very similar.  At most the remaining 40% predate the split, and nothing in the abstract says they are similar, let alone constrained.  The later counts above, 129 human pseudogenes with mouse orthologs and 6 shared ribosomal-protein copies, show how few are.  The first half of the book's sentence is fine:  Zhang's 2003 catalogue did find ">8000 processed pseudogenes".²⁸  The second half should be withdrawn, and the inference built on it with it.

The error originated with Wen and colleagues, who are mainstream biologists;  Francisco Ayala, a well-known evolutionary geneticist, is a co-author.  The volume's authors repeated a peer-reviewed review accurately.  But the claim carried the paragraph's argument, and checking one citation back would have caught it.

What the class-wide numbers support

Put together, the class-wide evidence says three things.  About one pseudogene in ten is transcribed in normal adult tissue and perhaps one in five somewhere.  A few percent of processed pseudogenes and about a quarter of duplicated ones show conservation beyond the neutral rate.  Protein-level selection is found in about one transcribed pseudogene in twenty.  The rest, the large majority, look on every class-wide test like copies decaying without constraint.

Two cautions keep that from being a verdict.  First, conservation is a sign of function only on the assumption that the species compared share an ancestor and that enough time has passed for unconstrained DNA to drift.  A young-earth reader does not grant the timescale, and this site's Functional DNA is Evidence of Design argues that a lack of conservation implies nothing unless evolution is assumed.  Second, most processed pseudogenes are young even on the mainstream timescale, and a young copy has had little time to show either constraint or decay.  So the conservation tests can find function in old pseudogenes, but they can say little either way about most of the young ones.

Part 3: Measured function, case by case

What counts as a measurement

A pseudogene has a measured function when someone changed it and something else changed as a result.  The tests differ a great deal in strength.  From strongest to weakest:

1. Delete it in a living animal, and see a defect.
2. Delete it in cells, at its own place in the genome, and see a defect.
3. Silence it in cells with RNA interference or CRISPR interference, and see a defect. Pseudogenes resemble their parents, so the reagent may hit the parent too.
4. Overexpress it, often far above its natural level, and see an effect. This shows what the molecule can do, not what it does at its natural dose.
5. Remove something general, such as the Dicer enzyme that makes all small interfering RNAs, and infer which pseudogenes were involved.

And on top of each:  did anyone else repeat it?

Xu and Zhang add an evolutionary test.  Because "biochemical activities may have no fitness benefit, proof of a true biological function requires the demonstration that the activity or the pseudogene is under natural selection".  By that standard, they wrote of PTENP1 in 2015, "No such proof has been provided".¹¹  A young-earth reader need not accept that selection is the test of function.  But a design advocate who points to conservation as evidence of function has already accepted half of it.

The cases

Chicken antibody genes are the oldest and clearest case, and they work at the DNA level.  Chickens have a single working light-chain antibody gene.  Upstream of it, Reynaud and colleagues found in 1987 "25 V-hybridizing elements, all of which are pseudogenes".  During development, B cells copy short stretches from these pseudogenes into the working gene "at a frequency of 0.05-0.1 per cell generation", and that is how the bird builds its antibody variety.¹⁵  In 2004 Arakawa, Saribasak and Buerstedde deleted the pseudogenes in a chicken B-cell line:  "the stepwise removal of the ψV donors not only reduces and eventually abolishes Ig gene conversion, but also activates AID-dependent Ig hypermutation."¹⁶  The pseudogenes are spare-part libraries.  Their stop codons do not matter, because they are never meant to be read on their own.  This is function in the strictest sense, and it has been textbook immunology for decades.

HBBP1 is the human case that holds up best.  It is covered in Part 4 with the named cases.

PTENP1 is the famous case, and it replicated only in part.  In 2010 Laura Poliseno, Pier Paolo Pandolfi and colleagues reported in Nature that PTENP1, a processed copy of the tumour-suppressor gene PTEN, soaks up the microRNAs that would otherwise shut PTEN down:  "We find that PTENP1 is biologically active as determined by its ability to regulate cellular levels of PTEN, and that it can exert a growth-suppressive role."⁴¹  A 2016 review said this "competing endogenous RNA" idea "has gained substantial attention as a unifying function for long non-coding RNAs, pseudogene transcripts and circular RNAs".⁴²  In 2020 the Reproducibility Project:  Cancer Biology repeated the key experiments in the same cell line with the original siRNA pools.  Its editors' summary:  "This Replication Study has reproduced some parts of the original paper but it also contains results that are not consistent with other parts of the original paper."¹⁹  Silencing PTENP1 still made the cells grow faster as a main effect, "F(1,16) = 13.8, p=0.0019".  But the fall in PTEN messenger RNA that was the whole proposed mechanism was small and not significant, a relative level of 0.883.¹⁹  A separate objection concerns the dose.  Denzler and colleagues counted microRNA molecules and binding sites in mouse liver cells and concluded that "a ceRNA effect mediated through a single miRNA family in a physiological or disease setting of the liver is unlikely", because a decoy would need to add tens of thousands of binding sites per cell to matter.⁴³  That was not measured on PTENP1, but it applies to any low-abundance decoy.

Braf-rs1 caused cancer in mice, at an unnatural dose.  Karreth, Pandolfi and colleagues built mice that make the mouse BRAF pseudogene on command.  On the drug that switched it on, they "became moribund and had to be sacrificed after a median survival of 421 days".³⁸  The authors are candid that the natural pseudogene is far rarer than the transgene:  "While endogenous Braf-rs1 expression was between 6–115-fold lower than B-Raf, expression of transgenic Braf-rs1 was comparable to B-Raf".³⁸  That shows what the RNA can do in excess.  It does not show what the pseudogene does in a healthy mouse.

Mouse eggs make small RNAs from pseudogenes.  Tam, Hannon and colleagues sequenced small RNAs from about 6,000 mouse oocytes and showed that "a subset of pseudogenes generates endogenous small interfering RNAs (endo-siRNAs) in mouse oocytes."  When the Dicer enzyme was removed, target genes rose:  "Loss of Dicer, a protein integral to small RNA production, increases expression of endo-siRNA targets, demonstrating their regulatory activity."³⁹  Watanabe and colleagues reported the same system in the same issue of Nature.⁴⁴  This is a real regulatory role, but no single pseudogene was removed, and the authors say "we cannot distinguish whether these small RNAs direct target cleavage or whether the act of siRNA production per se, which consumes the coding mRNA, is sufficient for repression."³⁹

The CRISPRi screen is the largest test, and the hardest to read.  Sun and colleagues silenced the promoters of 850 pseudogenes that are expressed in breast cancer cells and grew the cells for three weeks.  "We identified 69 pseudogene hits (out of 850 pseudogenes) that were negatively selected in MCF7 cells", 8.1%.  None helped growth when silenced.  They tested four one at a time and rescued the effect by adding back the pseudogene's cDNA.²¹  But in 15 of the 69 hits the parent gene was a hit too, which raises the question whether the reagent silenced both.²¹  And a cancer cell line in a dish is not an organism:  it shows that switching off these transcripts hurts that line's growth, not what they do in a body.

Makorin1-p1 is the cautionary tale.  In 2003 Hirotsune and colleagues reported in Nature that a transgene landing near the mouse pseudogene Makorin1-p1 reduced its transcription and so destabilised the parent gene's RNA "in trans", giving mice cystic kidneys and bone deformity.⁴⁵  It became the founding example of a regulatory pseudogene.  In 2006 Gray, Wilson, Fortin and Nicholls tested it directly and found the opposite on every count:  "Mkrn1-p1 is not transcribed and therefore cannot stabilize Mkrn1 transcripts in trans", and "mice in which Mkrn1 has been directly disrupted show none of the phenotypes attributed to a partial reduction of Mkrn1."²⁰  Wells reported both papers.  He called the result "controversial" and argued that "Invalidating a report of one function in one pseudogene cannot exclude other possible functions in that pseudogene, much less possible functions in other pseudogenes."²  That is logically correct, and it is also no evidence for any other function.  The creationist case leaned on this result more than any other.  Woodmorappe's 2003 note "Pseudogene function:  more evidence" was built on it, reporting that Hirotsune's team "had inadvertently disrupted the regulatory effects of the Makorin1-p1 pseudogene upon the expression of the Makorin1 gene."⁴⁶  His 2004 paper still used it as the example of a function no substitution test would find.³⁵  Both papers came before Gray's 2006 test, so Woodmorappe was reporting the literature as it then stood.  The lesson for both sides is the same:  one unrepeated experiment is a lead, not a result.

How many?

No one keeps a list.  The search for this article found no database of experimentally confirmed functional pseudogenes:  the resources that exist, such as PseudoFuN and dreamBase, are built from computed links or high-throughput data, not from curated experiments.  The reviews that argue for function list cases one by one.  Pink and colleagues' 2011 review in RNA names about a dozen, and its own conclusion is careful:  "While not all pseudogenes (or even all transcribed pseudogenes) will have biological functions, it is likely that, where an unexpected regulatory benefit results from the formation of a pseudogene, the effect will be conserved."⁴⁷  Cheetham, Faulkner and Dinger's 2020 review in Nature Reviews Genetics says only that "growing numbers of pseudogenes are being found to play important biological roles."⁴⁸

Counting generously, every individually tested pseudogene in this Part, Part 4 and the reviews comes to a few dozen, and the CRISPRi screen adds 69 hits in one cell line.  Against about 14,700 human pseudogenes, that is well under 1%.

That number measures how little has been tested as much as how little works.  Most pseudogenes have never been deleted, silenced or overexpressed by anyone.  Cheetham and colleagues argue that the label itself discourages the work:  calling something a pseudogene "has led to their frequent dismissal from functional assessment and exclusion from genomic analyses."⁴⁸  Moran answers that this is a reasonable use of scarce money:  "If some researcher suspects that the label is incorrect then the burden of proof is on them to show that the DNA has a function."¹  Both are right about part of it.  The untested majority is untested, not tested and found empty.

The definitional dodge

Dan Graur, an evolutionary biologist at Houston, wrote in 2013 that pseudogenes "by definition" are nonfunctional, so "The measly handful of 'pseudogenes' that have so far been assigned a tentative function ... are, by definition, functional genes, merely pseudogene look-alikes."²⁴  This makes the claim that pseudogenes are junk impossible to test, because any counter-example is reclassified.  Moran is more careful:  "The question is not whether some DNA regions that look like pseudogenes have a function.  This is a fact."¹  The fair way to frame the question is his:  are the working ones exceptions, or is the label wrong for most?  On present data they are exceptions, and the large untested remainder leaves the question open in principle.

Part 4: The named cases

Four pseudogenes carry most of the public argument.  Each is also used as evidence that humans share ancestors with other animals, and that question is argued in Shared Genetic Mistakes and Common Ancestry and, for humans and chimpanzees, in Shared Retroviruses and Pseudogenes, which reviews all three human cases in detail.  This Part asks only whether each does anything.

HBBP1: the relic that does something

HBBP1 sits in the beta-globin cluster on chromosome 11.  It is a copy of a globin gene with a broken start codon, a premature stop and a frameshift, and it makes no globin.  Jeffrey Tomkins of the Institute for Creation Research argued in 2013 that it "appears to be a highly functional and cleverly integrated feature of the human genome that is intolerant of mutation."⁴⁹  The 2017 volume made the same case more briefly, saying its similarity "is not due to protein coding but appears to involve regulation of developmental changes via chromosomal interactions."¹³

Four lines of evidence now support them.

1. Population constraint. Moleirinho and colleagues sequenced the cluster in 1,092 people from the 1000 Genomes Project and found that "HBD and its neighbor pseudogene HBBP1 have mainly evolved under purifying selection, suggesting that their roles are essential and nonredundant."⁵⁰
2. Deletion of the region. Huang, Blobel and colleagues cut the HBBP1 region out of adult human red-cell precursors with CRISPR. The locus folded more like a fetal one, and these changes were "accompanied by strong increases in γ-globin transcription."¹⁷ HBBP1's DNA helps keep fetal haemoglobin switched off in adults.
3. The RNA. In 2021 a Beijing group led by Yanni Ma reported in Developmental Cell that HBBP1 "is essential for erythropoiesis", by binding a protein called HNRNPA1 and raising a master regulator of red-cell development, TAL1.¹⁸ The RNA is plentiful where red cells are made: it is "expressed exclusively in bone marrow and particularly in bone marrow erythroblasts".¹⁸ Deleting the whole locus, or only its start site, in human embryonic stem cells cut the number of early red-cell colonies from about 58 to about 2.¹⁸ To tell the RNA from the DNA, the team broke the RNA's splicing while leaving the locus in place, and red-cell development still failed; HBBP1 RNA made from elsewhere in the genome partly rescued cells lacking the locus.¹⁸ A correction to the paper was published in 2025.
4. Disease. The same team measured HBBP1 in the blood of 303 patients with β-thalassaemia, an inherited anaemia, and found that "higher levels of HBBP1 and TAL1 transcripts were associated with milder β-thalassemia symptoms". The authors call these "preliminary data".¹⁸ Tomkins had cited, in 2013, "three different reports that link point mutations in the HBBP1 gene with β-thalassemia disease pathologies."⁴⁹ The patient data point the same way.

So the design side was right that HBBP1 works, right that its RNA does something, and right before the decisive experiments were published.  Three parts of the story still need care.

First, the laboratories disagree on one point.  Moleirinho had found HBBP1 "either marginally or not transcribed at all" in the cells they studied,⁵⁰ which Ma's data from developing red cells now contradict for those cells.  Huang removed only the transcript's start site in a cultured adult red-cell line and saw no change in globin switching:  "the HBBP1 genomic region, but not the HBBP1 transcript per se, is necessary for maintaining γ-globin silencing in adult erythroid cells."¹⁷  Ma's team used different mutations and measured a different outcome, red-cell formation rather than globin switching, so both results may stand.  No one has yet tested both kinds of mutation side by side.

Second, every RNA result comes from one laboratory, in cells grown in a dish or human cells grown in mice.  The case that the RNA matters in a whole human body rests on the thalassaemia association, which its authors call preliminary.

Third, the job is human-specific.  Ma's team found HBBP1 RNA absent from chimpanzee blood and concluded that "HBBP1 has been functionless for the majority of its evolution but has recently acquired its function as a human-specific essential regulator of erythropoiesis."¹⁸  That cuts against the 2017 volume's suggestion that the human and chimpanzee copies are alike because they do the same job.  So does Moleirinho's own table:  human and chimpanzee HBBP1 exons differ at 1.45% of sites, more than the working genes beside them, HBB at 0.26% and HBD at 0.80%.⁵⁰  What is unusually low is variation among humans, not divergence between species.  A creationist can read the chimpanzee result as a designed difference between two kinds.  Either reading leaves the chimpanzee copy carrying the same disabling lesions as the human one, which is the ancestry question.

HBBP1 also shows the limit of the function argument.  A sequence can be a disabled copy of a globin gene and a working piece of chromatin architecture at once.  The second fact does not erase the first, and it says nothing either way about how humans and chimpanzees came to share it.

Vitellogenin: no function shown

Birds make egg yolk with three vitellogenin genes.  In 2008 Brawand, Wahli and Kaessmann looked for them in mammals at the matching place in the genome, beside the same neighbouring genes, and found broken fragments in humans, dogs, armadillos and other placentals.  They reported that "Exon 3 of VIT1 reveals shared indels between human and dog", which they read as inactivation before those lineages split.⁵¹  Dennis Venema made it the centrepiece of the BioLogos case for common descent.⁵²

Tomkins answered in 2015 that the human fragment "is not a pseudogene remnant at all, but a functional enhancer element in the fifth intron of a 'genomic address messenger' (GAM) gene."²²  His evidence for the host gene is a computer prediction, "the gene prediction software Genscan".  His evidence that GAM genes work is that they "have been demonstrated to selectively inhibit the translation of known target genes", for which his reference is a United States patent held by Rosetta Genomics.²²  A patent is not an experiment on this locus.  No study that deleted, silenced or tested the human VIT1 fragment was found.

Venema conceded part of this:  "Even though the evidence that this sequence is functional in humans is rather thin, the main issue is that even if its function were convincingly demonstrated in the future, it would not remove the evidence that this sequence was once part of a functional VIT gene".⁵³  On the function question, "rather thin" is generous.  The claim of function is a hypothesis with no measurement behind it.  The relic reading is the only one with data, and a future enhancer in that intron, if one is found, would be a new job for some of the DNA, not proof that it was never a yolk gene.

Enamelin: no function claimed

Enamelin is a protein needed to build tooth enamel.  Meredith, Springer and colleagues predicted in 2009 that toothless and enamel-less mammals would carry broken copies, and found "'molecular fossils' of the enamelin (ENAM) gene in four different orders of placental mammals (Tubulidentata, Pholidota, Cetacea, Xenarthra)".  They concluded that "The molecular decay of ENAM parallels the morphological degeneration of enamel in the fossil record of placental mammals".⁵⁴  In 2014 the same group found inactivating mutations in tooth genes shared across modern bird orders, "indicating that the common ancestor of modern birds lacked mineralized teeth."⁵⁵  Al-Hashimi and colleagues identified the chicken's broken enamelin copy.⁵⁶

No design or creationist author was found who claims a function for any enamelin pseudogene, and no study testing one was found.  Wells's book does not discuss enamel.  So enamelin is not a "relic turned out to be doing something" case.  It is an argument from decay, and whether its pattern shows common ancestry belongs to the ancestry articles.  A young-earth reading would treat a broken enamel gene in a toothless animal as loss after creation, which needs no function at all.

GULO: no function shown, and the leading creationist paper agrees

Most mammals make their own vitamin C.  Humans cannot, because the gene for the last enzyme, L-gulono-γ-lactone oxidase, is broken.  Nishikimi and colleagues cloned the human remnant in 1994 and concluded that it "has accumulated a large number of mutations without selective pressure since it ceased to function during evolution."⁵⁷  The human copy keeps about five of the rat gene's twelve exons, with stop codons and frameshifts in what remains.  Drouin and colleagues, reviewing vitamin C loss across vertebrates, admit that "The nature of the initial mutations which inactivated the anthropoid primate and guinea pig GLO genes is not known."⁵⁸

What does the design side claim?

1. Don Criswell, in an Institute for Creation Research article of 2007, proposed that the remnant "might have an entirely different function than the rat gene", while granting that "the exact role of pseudogene sequences in the genome are not known by any scientist."⁵⁹
2. Jonathan McLatchie, in 2013, proposed that GULO might be switched on in early life with its RNA edited past the stop codons, noting that "there are in fact some data to suggest that vitamin C may in fact be produced endogenously" in infants.⁶⁰ No edited GULO transcript has been reported.
3. Wells, in 2011, argued that function had not been ruled out, but conceded: "The strongest statement that could be warranted by the evidence would be that we do not currently know of a function for the vitamin C pseudogene."²
4. Tomkins's 2014 paper in Answers Research Journal, the fullest creationist treatment, does not claim a function at all. It reports that "no GULO transcripts have been detected" and that the region is "polycomb-repressed" and "transcriptionally inactive".²³ His argument is about whether humans and apes lost the gene separately, which belongs to the ancestry articles.

The public expression data agree with Tomkins.  In GTEx, the largest survey of gene activity across normal human tissues, the GULO pseudogene's median level is exactly zero in 19 of 54 tissues and at most 0.49 transcripts per million, in the fallopian tube.⁶¹  That is below the cut-off of 1 transcript per million that Jiang, Borsari and Gerstein used to call a pseudogene expressed,³³ in every tissue.  A few reads at that level can come from mapping noise.

So a claim sometimes repeated online, that creationists have shown GULO is "partially transcribed" or functional, is not supported by any creationist primary found for this article.  One secondary essay attributes the "partially transcribed" idea to Tomkins;⁶² Tomkins's own paper says the opposite.  On present evidence GULO is what both sides' primary sources say it is:  a broken gene the body tolerates.  Tomkins's title calls it evidence for "genetic entropy", the young-earth idea that genomes decay over time,²³ which is a reading that needs no function.

Part 5: What the numbers support

The tally

Read down the table and the numbers grow as the evidence weakens.  About four in five pseudogenes show some biochemical sign of activity, in Sisu's words "Contrary to the actual definition of pseudogenes ('dead genomic elements'), this group comprises only ∼20% of the total pseudogenes."⁷  About one in five is transcribed somewhere.  About one in fifteen is conserved beyond the neutral rate.  Well under one in a hundred has a function shown by experiment.

That gives the answer to the article's question in three layers:

1. Measured function: a few dozen pseudogenes, under 1%. Real, including at least one human case, HBBP1, that matters for a common disease.
2. Activity or constraint without a test: roughly 10% to 30%, depending on whether one counts transcription in normal tissue, transcription anywhere, or conservation. These are candidates. Some will turn out to work, and some are transcriptional noise or copies too young to have drifted.
3. No evidence either way: the large remainder. These pseudogenes have never been deleted, silenced or overexpressed. Class-wide tests make most of them look like copies decaying without constraint, but the sharpest of those tests sees only protein-level selection, and none can see a job done by DNA structure alone.

What each side may fairly claim

The junk side may fairly say that most pseudogenes show no sign of selection, that the processed majority are mostly lineage-specific copies that look "dead on arrival",⁵ and that the best-known regulatory claim, Makorin1-p1, collapsed when tested.  It may not say that pseudogenes are nonfunctional "by definition".  It should drop the habit of treating the untested majority as tested.

The design side may fairly say that function has been found where people looked, including in a case its own writers predicted (HBBP1);  that the pseudogene label discouraged testing for decades;  and that most pseudogenes are untested.  It may not say that "most" pseudogenes are functional, which no measurement supports.  It should stop counting transcription as function and stop citing the "60 percent" mouse-human figure.

What a young-earth reading predicts

A young-earth model does not need pseudogenes to work.  On that model, genomes were created whole and have lost information since.  Robert Carter, who writes on human genetics for Creation Ministries International, put it this way in 2009:  "young-earth creationists do not require the entire genome to be highly functional", and some stretches may "just add temporal structure to the functional parts."⁶³  Unitary pseudogenes like GULO, and broken enamel genes in toothless animals, fit that expectation without any function at all.  Tomkins's GULO paper reads the vitamin C gene that way.²³

Processed pseudogenes are harder.  Sisu and colleagues describe the human set as reflecting "a massive burst of retrotranspositional activity at the dawn of the primates".⁷  No creationist account of where these copies came from was found.  Woodmorappe states the working assumption:  "Creationist scientists (including me) generally assume that God would not create purposeless genes in different primates, and that God did not independently disable the same genes in humans and nonhuman primates during the Curse."  He also says plainly that "Not enough is yet known about eukaryotic genomes to construct a comprehensive creationist model of pseudogenes."⁶⁴  No later creationist model was found.  That leaves two options, and neither is free.

Created in place.  This is where Woodmorappe's assumption leads.  It predicts that most processed pseudogenes have jobs, and for most of them the evidence is absent.  Carter's "temporal structure" would let them be spacers whose exact sequence does not matter, which would fit their neutral-looking decay.  But it does not say why a spacer would look exactly like a messenger RNA copied back into DNA:  no introns, the remains of a poly-A tail, and a working parent gene elsewhere in the genome.

Inserted since creation.  No creationist model proposes this for most of them, so the figures that follow are for scale only.  The human genome holds 10,634 processed pseudogenes.³  Six thousand years is about 240 generations of 25 years, so inserting them all would take about 44 new copies in every generation.  Today's rate has been measured directly.  Feusier and colleagues sequenced 599 people in three-generation Utah families and found 26 new insertions of mobile DNA of every kind in 437 births, about one per 17 births.  None was a processed pseudogene;  such events, they write, "are rare".⁶⁵⁶⁵  Across the living population, Abyzov and colleagues catalogued 174 gene copies that are present in some people and absent in others, among almost 1,000 genomes.⁶⁶  So even if every new insertion of every kind were a processed pseudogene, today's rate falls short by a factor of more than 700, before any copy has to spread from one person to everyone.⁶⁷  A model of this kind needs a burst of gene copying far above today's rate, concentrated in the first generations, and no mechanism for one has been proposed.  It also has to explain why apes carry so many of the same copies at the same places, which is the question Shared Retroviruses and Pseudogenes takes up.

A young-earth reader should also notice that the conservation evidence cuts both ways.  This site's Functional DNA is Evidence of Design argues that a lack of conservation implies nothing unless evolution is assumed.  By the same logic, conservation between humans and mice, where it exists, is evidence of function only on the assumption that the two share an ancestor and have been separate long enough for neutral DNA to drift.  On a young-earth timescale, neither the presence nor the absence of conservation says much about pseudogene function, and the case has to be made by experiment.  The experiments so far support function for a few dozen and say nothing about the rest.

Objections and responses

1.  "Most pseudogenes have simply not been tested.  Every time someone looks, a function turns up."  The first sentence is true and the second is not.  When Sun and colleagues silenced 850 expressed pseudogenes at once, 69 slowed the growth of one cancer line and 781 did not.²¹  When Gray and colleagues tested Makorin1-p1, the function vanished.²⁰  When the PTENP1 experiments were repeated, part of the result held and the proposed mechanism did not.¹⁹  Looking turns up function in a minority of the cases looked at.  That is a reason to keep looking, and no reason to assume the answer for the untested majority.

2.  "Transcription in specific tissues at specific times is the mark of regulated, functional RNA."  Specific patterns are what one would expect of functional RNA, and they are also what one gets when a pseudogene sits inside or near a gene that is itself switched on in that tissue.  The ENCODE pilot tested the difference directly and found that transcribed pseudogenes are no more constrained than untranscribed ones.⁶  About half of the pseudogenes expressed in normal tissue are seen in only one tissue.³³  Transcription marks a candidate.  It does not mark a function.

3.  "Pseudogenes are nonfunctional by definition, so any example is just a misnamed gene."  This is Graur's position.²⁴  It defines the question away.  If "pseudogene" means "sequence that looks like a broken gene", then whether pseudogenes work is an empirical question with a partly positive answer.  If it means "broken gene with no function", then no one can know whether a sequence is a pseudogene until every possible function has been excluded, which Cheetham and colleagues point out is never done.⁴⁸  Moran's framing, that the question is how common the exceptions are, is the scientific one.¹

4.  "Pseudogenes are conserved across species, which proves they have a function."  A minority are.  Most processed pseudogenes are not shared even with mice:  only about 1% of human pseudogenes have mouse orthologs,⁷ and the 60% figure sometimes quoted for this is the wrong way round.¹⁴  Conservation also rests on the mainstream timescale, which a young-earth reader does not grant.

5.  "HBBP1 shows that the whole pseudogene argument for evolution has collapsed."  HBBP1 shows that one pseudogene's DNA does a job, and on one laboratory's evidence its RNA does too.  It does not show that the other 14,700 do, and it does not settle how humans and chimpanzees came to share it.  Its RNA job is also human-specific on that laboratory's evidence, so it cannot explain why the chimpanzee carries the same broken copy.  Moran's point applies:  that some pseudogenes work "is a fact";¹ the question is how many.

6.  "The design side predicted function for pseudogenes, and evolutionists predicted junk.  Function was found, so design wins."  The prediction ledger is set out in this site's Functional versus Junk DNA Predictions.  For pseudogenes specifically, both predictions have partial support:  most pseudogenes look inactive on class-wide tests, and some have important jobs.  A prediction that "many of these mysterious features will turn out to have important functions",¹³ made without a number, cannot be scored against a count until someone says how many "many" is.

7.  "Function does not matter.  Shared broken genes prove common ancestry whether or not they do anything."  The ancestry argument is a separate argument, made in Shared Genetic Mistakes and Common Ancestry.  But function does matter to it in one way.  The ancestry argument's force comes from the claim that the shared feature is pointless, the way a shared misspelling is.  Where a pseudogene turns out to work, as HBBP1 does, that rhetorical force is gone, even though the shared position still needs explaining.  Where it does not work, as GULO does not, the argument stands or falls on the ancestry evidence alone.

8.  "The 2017 book's mistake about mouse pseudogenes is a small slip in a long book."  It is one sentence, but it was the sentence that carried the paragraph's argument, that similarity implies "a sequence-dependent essential function".  The book's authors repeated a peer-reviewed review accurately;  the review is where the inversion happened.  The error should be fixed in any future edition, and design writers should not repeat it.

9.  "Creationists say GULO is transcribed and may still work."  No creationist primary found for this article says that with data.  The leading creationist paper on GULO says the region is "transcriptionally inactive",²³ and the GTEx data put it below the usual expression cut-off in every tissue.⁶¹  The proposals of a function, Criswell's and McLatchie's, are hypotheses without measurements.

Issues and unanswered questions

1. No one has counted the cases. No curated database of experimentally confirmed functional pseudogenes was found. The "few dozen" in this article is a count of the cases found in the primary papers and reviews read here, and it could be low. A systematic literature count, with each case scored by the strength of its test, would settle it.
2. Several key papers were read only as abstracts. The full texts of Cheetham 2020, Watanabe 2008, Hirotsune 2003, Balakirev and Ayala 2003, Zhang 2004 and Meredith 2014 were not obtained, nor was the text of the 2025 correction to Ma 2021. Nothing is quoted from them beyond their abstracts. Zhang 2004's body might qualify its "60%"; the abstract is clear, but the full text should be read.
3. The CRISPRi screen cannot tell RNA from DNA, or pseudogene from parent. In 15 of its 69 hits the parent gene was also a hit,²¹ and silencing a promoter can also quiet neighbouring genes. The screen is in one cancer line. A screen in normal cells, with deletion rather than silencing, has not been done.
4. The ceRNA dose problem is unresolved for pseudogenes. Denzler's count was made on one microRNA in liver cells.⁴³ Whether PTENP1 is ever abundant enough in a real cell to matter has not been measured in the papers read here.
5. The transcribed fraction depends on the denominator. Sisu's "∼15%" cannot be reproduced from its own human count,⁷ and estimates range from 8% to 20% depending on tissue, method and threshold.
6. No test sees DNA-level function in bulk. Substitution-ratio tests detect protein-level selection only, and knockdowns remove RNA, not DNA. HBBP1's job is done by its DNA region, and only a deletion found it. How many other pseudogenes have such a job is unknown, and no genome-wide deletion screen of pseudogene DNA exists.
7. GULO transcription has not been settled with long reads. The short-read GTEx signal is near background. A long-read or unique-mapping study of the locus would close the question.
8. The young-earth origin of processed pseudogenes has no published model. Woodmorappe said in 2000 that none could yet be built,⁶⁴ and none was found since. Part 5 prices the two options. Created copies predict a function that the evidence does not show for most of them, and leave their marks of reverse transcription unexplained. Copies inserted since creation need more than 700 times today's total rate of new insertions, even on assumptions that favour the model.⁶⁷ A young-earth model of pseudogenes would have to choose between these, or find a third option, and test it.

Verdict

Are pseudogenes broken relics, or do many of them do something?  Both, in proportions that can now be stated.  A few dozen human pseudogenes have a function shown by experiment, and at least one, HBBP1, matters for human disease.  Roughly a tenth to a third show transcription or constraint that marks them as candidates.  The rest, most of the 14,700, have no evidence of function, and the tests that have been run across them look like what neutral decay would produce.  They have also never been tested one by one.

Of the named cases, HBBP1 is a true "relic turned out to be doing something".  Vitellogenin, enamelin and GULO are not:  no function has been measured for any of them, and for GULO the design side's own best paper agrees.

Neither side's slogan survives.  Bergman's conclusion that "most of them have a function"²⁵ outruns every count.  The claim that pseudogenes are nonfunctional "by definition" is not a finding.  The defensible statement is narrower, and both Moran and Wells have made half of it:  some pseudogenes clearly work, and for most of them no function is known.

For the debate over design, pseudogenes are a small prize.  Their exons are 0.36% of the genome, and the case that most human DNA is functional, made in Functional DNA is Evidence of Design, does not depend on them.  A young-earth model expects some genes to have broken since creation and loses nothing if most pseudogenes are dead.  What it cannot do is claim pseudogenes as evidence of design on the present data, or treat the shared positions of broken genes as answered by function.  That second question is argued in Shared Genetic Mistakes and Common Ancestry.

Sources