Genetic Redundancy | Berean Archive

"Such a redundant genetic organization is sensible from an engineer's point of view: important functions require backup devices that can take over in case of failure. But can natural selection favour the emergence and stability of redundant genes?" — Martin Nowak, Maarten Boerlijst, Jonathan Cooke and John Maynard Smith, Nature, 1997¹

Summary

A gene is redundant when knocking it out does little harm because something else does its job. The claim tested here has three parts. Knockouts are often harmless because something covers for the missing gene. Evolution predicts that the cover comes from duplicate copies of the same gene. Much of the cover actually found comes instead from unrelated genes and whole networks, which no duplication route explains and which a conservation test cannot see. Each part was checked against the papers that measured it.

Knockouts often look harmless, but mostly not because of a backup. In rich laboratory medium about 80% of yeast gene deletions have "no obvious phenotypic consequence."² Tested across 1,144 drugs and stresses, 97% of yeast deletions showed "a significant fitness defect" somewhere.² Most silent knockouts are genes needed under other conditions, or genes whose small cost only a competition experiment can see.³⁴
Where true backup has been measured, duplicates carry a large share of it. In yeast, at least a quarter of silent deletions are covered by a duplicate, with an upper bound of 59%.⁵⁶ In yeast glucose metabolism, duplicates supplied 75% of the backup and alternative pathways 25%.⁷
Cover by unrelated genes is real, but it is mostly not hidden spare parts. Thousands of single-copy genes can be removed without a visible effect.⁶ Andreas Wagner, the leading advocate of "distributed robustness," judged unrelated back-up genes "probably rare" and describes the cover as a property of the network.⁸⁹ The one paper that claimed networks outperform duplicates was corrected by its authors because of a software bug.¹⁰
Evolutionary theory expects a perfect spare to decay, and it has models for partial ones. Nowak and colleagues wrote that a truly redundant gene "would not be protected against the accumulation of deleterious mutations," then gave three conditions under which redundancy is stable anyway.¹ Their simulation grew redundancy between genes that were never copies of each other.¹ So theorists do offer a route that needs no duplication. What the route leaves out is the molecular step: the model lets a gene pick up a second job at a rate the modeller chooses.
A conservation test sees partial backups and misses perfect ones. Authors from the ENCODE project write that sequence alignment misses "mutations whose effect is buffered by redundancy."¹¹ But the backup enhancers that Osterwalder and colleagues studied in 2018 were chosen because they are highly conserved.¹² A backup that carries any share of the load shows up as conserved.
The main creationist paper on the subject has four checkable errors. It cites the withdrawn networks paper as current evidence. It paraphrases a Nature abstract's setup and drops its finding. It uses a family of recognizable duplicates as evidence against duplication. And it relies on a yeast result that its author, Wagner, later attributed to thin data.¹³⁶¹⁴

Verdict. Life is built with far more backup than a lean design would need, and the backup takes both forms engineers use: identical copies and different parts that can do the same job. The catalogue claim overstates two things. Duplicates account for more of the measured backup than the claim allows. And evolutionary theory does offer a non-duplication route, though that route assumes its hardest step. What survives is narrower and still real. A perfect backup, one whose loss costs nothing until the primary fails, is exactly what evolutionary theory says should decay. It is also exactly what a conservation test cannot see. How many perfect backups exist is unmeasured, and that is where the question should now go.

Part 1: How much of the genome can be knocked out

The numbers

Every well-studied organism tolerates the loss of most of its genes, one at a time, under laboratory conditions.

Organism	Test	Result
E. coli	single-gene deletions	303 of 4,288 genes could not be deleted¹⁵
B. subtilis	single-gene inactivation	192 of about 4,100 indispensable, 79 more predicted¹⁶
Yeast	deletions in rich medium	19% lethal, 15% slower growth²
Worm	RNA interference	89% of single-copy and 96% of duplicate genes showed no detectable effect⁶
Mouse	1,751 knockout lines	about 35% lethal or subviable¹⁷
Human cells	CRISPR screen, one cell line	1,878 of 18,166 genes needed for growth, about 10%¹⁸
Living humans	141,456 sequenced people	1,815 genes likely tolerated with both copies broken¹⁹

Two cautions apply to the table. The worm figure comes from RNA interference, a temporary and often incomplete knock-down that Conant and Wagner say detects fewer effects than a true deletion.⁶ The human figure counts genes seen broken in adults in a medical reference sample. It says those people were alive, not that they were unaffected, and "selection is largely blind to phenotypes emerging after reproductive age."¹⁹

Why a silent knockout is usually not a backup

A knockout can look harmless for three different reasons. The gene may be needed only in conditions the lab did not supply. It may help only a little, below what a plate or a cage can show. Or something else may really be covering for it. Only the third is redundancy.

The first explanation is the largest. Maureen Hillenmeyer and colleagues grew the yeast deletion collection under 1,144 treatments. Only 205 of about 6,000 strains failed to show a defect in any of them. They concluded that "genetic redundancy is therefore limited in providing tolerance of the tested conditions."²² The 97% figure is generous, because it pools two kinds of test and carries a false-positive rate of about 10%. But the direction is clear. Balázs Papp, Csaba Pál and Laurence Hurst modelled yeast metabolism in 2004. They found that genes needed only in untested conditions account for 37–68% of the dispensable metabolic genes, duplicates for 15–28%, and rerouting through the network for only 4–17%.⁴

The second explanation is also common. In 1998 John Thatcher and colleagues grew yeast strains with disrupted nonessential genes in direct competition with the normal strain. At least 19 of 27 strains "that have no conventional phenotype do have significantly reduced fitness," mostly by less than 5%.³ A 1% disadvantage is invisible on a plate and decisive over a few hundred generations.

So the headline numbers used on both sides of this debate, 80% of yeast genes "silent" and 89% of worm genes without effect, measure the limits of the laboratory. They are not a count of backup genes.

Part 2: Duplicates or networks

Genes can cover for each other in two ways. A duplicate copy can do the same job. Or, in Wagner's words, cover can come from "interactions between genes with unrelated functions."⁸

The case for networks, and what happened to it

Wagner tested this in 2000 with about 250 yeast knockouts, 45 of them duplicated genes, and concluded that "interactions among unrelated genes are the major cause of robustness against mutations."⁸ This is the paper the creationist literature cites for redundancy that is "not due to duplications." The quotation is accurate.

Three years later Zhenglong Gu and colleagues repeated the test on nearly the whole yeast genome. They found "a significantly higher probability of functional compensation for a duplicate gene than for a singleton," and estimated that "at least a quarter of those gene deletions that have no phenotype are compensated by duplicate genes."⁵ Their upper bound was 59%.⁶ Wagner, writing with Gavin Conant in 2004, said his own earlier null result for duplicates was "caused by the limited amount of gene-knockout data available at the time."⁶

Direct measurements of metabolism favour duplicates. Lars Blank, Lars Kuepfer and Uwe Sauer traced carbon flow through 367 yeast mutants. Once genes idle on glucose were set aside, "network redundancy through duplicate genes was the major (75%) and alternative pathways the minor (25%)" source of backup.⁷ Alexander DeLuna and colleagues found that for yeast's whole-genome-duplication pairs, "the functional overlap between paralogs sufficiently explains the apparent dispensability."²⁰ Mammals show the same pattern. Lethal mouse knockouts are "much less likely than viable genes to have a paralog."¹⁷ In human cells, "genes with paralogs are indeed less likely to be essential."¹⁸ Human genes with a 90%-identical relative are about three times less likely to carry known disease mutations.²¹

The only large study claiming the reverse for mammals, Toshimori Kitami and Joseph Nadeau's 2002 paper, was corrected the same year: "redundant networks provide no greater genetic buffering than do gene families."¹⁰

What is left for the networks

The network side keeps one strong fact. Many dispensable genes have no duplicate at all. In the worm, 7,872 single-copy genes showed no knock-down effect, more than the duplicates did.⁶ Conant and Wagner considered whether each could have an unrelated backup gene with the same function. They judged that "the massive scale—more than 7000 genes—at which such convergence would have to occur makes this seem unlikely," and favoured "interactions of unrelated genes in genetic networks."⁶ Wagner had already written in 2000 that unrelated back-up genes "may exist, but are probably rare."⁸

In 2005 Wagner defined this "distributed robustness" as a system where "many parts of a system contribute to system function, but all of these parts have different roles," and the system compensates "but not because a ''back-up'' redundant part takes over."⁹ A metabolic network rerouting around a missing enzyme is the standard example. The 2016 yeast genetic-interaction map shows how pervasive such relationships are: over 23 million double mutants, with the average gene showing a harmful interaction with about 2% of the genes tested.²²

So the catalogue clause is half right. Much robustness does not come from duplicates. But that part is mostly network rerouting and genes that serve other conditions, and the largest measured share of dedicated backup is duplicates.

Part 3: What evolutionary theory says about keeping a spare

The problem, stated by its own theorists

Nowak and colleagues put the difficulty plainly in 1997: "If a gene was truly redundant then it would not be protected against the accumulation of deleterious mutations. A widespread view is therefore that such redundancy cannot be evolutionarily stable."¹ Michael Lynch and John Conery found that duplicates arise often, about 0.01 per gene per million years, and that "the vast majority of gene duplicates are silenced within a few million years."²³ Yet duplicated yeast genes "can maintain substantial redundancy for extensive periods of time following duplication (∼100 million years)."²⁴ Lynch put the design side's question in his own words in 2012: "How can selection promote the establishment of additional layers of fitness-enhancing mechanisms if the established primary lines of defense are already highly refined?"²⁵

The answers on offer

Theorists answer in four ways.

Slow decay. Even with no selection at all, Nowak's first model gives redundancy an average lifetime "about 10⁷ generations" at a loss rate of one per million genes per generation.¹
Partial redundancy. Redundancy is stable if the backup is less efficient than the primary and also mutates less often, if each gene has a second job of its own, or if developmental errors knock out genes more often than germline mutations do.¹ Thatcher's result supplies the needed selection: "a selection coefficient approximately the inverse of the effective population size is sufficient to prevent loss of function."³
Population size. David Krakauer and Joshua Plotkin showed in a model that small populations favour redundancy and large ones favour mechanisms that remove damaged copies. They note that they "do not describe explicitly how redundancy will have arisen or is encoded."²⁶²⁶
By-product. Wagner argued that robustness accumulates "via an indirect mechanism," so that "evolution of robustness is not an adaptive phenomenon."⁸ Lynch doubts that selection built robustness at all and asks of adaptive explanations, "then where is the evidence?"²⁷ In a second paper that year he wrote: "Contrary to widespread belief, there is no compelling empirical or theoretical evidence that complexity, modularity, redundancy or other features of genetic pathways are promoted by natural selection."²⁸

Subfunctionalization, the leading account of why duplicates survive, is not an account of redundancy. Allan Force and colleagues propose that each copy loses different parts of the original job, so that both become necessary.²⁹ That keeps both genes by ending the backup.

Does any route build redundancy without duplication?

Yes, on paper. Nowak's team started a simulation in which "each function is performed by one gene, and each gene performs one function," and ended with "complex redundancy–pleiotropy networks in which each function is performed by several genes and each gene performs several functions."¹ Gerald Edelman and Joseph Gally named the phenomenon "degeneracy," "the ability of elements that are structurally different to perform the same function," and called it "an inevitable outcome of natural selection."³⁰ Nowak's team had already said in one line that redundancy "cannot only be a consequence of gene duplication because very different genes can also show overlapping redundancy."¹ So the claim that evolution expects backup only from duplicates is Terborg's reading of Ohno, not the position of the theorists themselves.

So "no duplication route explains" distributed redundancy is wrong as a report of the literature. It is fair as a criticism of the route's depth. Nowak's simulation lets a mutation change how well a gene performs any function to a random value, so a gene gains a new job in one step.¹ Edelman and Gally assert the result rather than model the chemistry. Whether a real protein can pick up another protein's job by a few mutations is an empirical question. These papers do not test it.

Part 4: The design argument in its makers' words

The creationist statement

The fullest creationist treatment is Peer Terborg's 2008 paper in Journal of Creation. Terborg argues that "truly redundant genes are impossible paradoxes because natural selection cannot prevent the accumulation of harmful mutations in these genes."¹⁴ He states that "Ohno's theory predicts an association between genetic redundancy and gene duplication," and concludes that "the predicted association between genetic redundancy and gene duplication is nonexistent."¹⁴¹⁴ His positive claim is that redundant networks are "an intrinsic, engineered characteristic of genomes," and his timing claim is that "redundant genes cannot have resided in the genome for millions of years."¹⁴¹⁴

Randy Guliuzza of the Institute for Creation Research makes the engineering inference general: "When humans can identify the true source of fail-safe redundancy, it always is an indicator of good design and a good designer."³¹ Casey Luskin of the Discovery Institute uses Lynch's layering question to ask "How can a system that is only rarely used, or only occasionally needed, evolve to such a high and efficient level of complexity?"³² Luskin also quotes Lynch's sentence that there is "no compelling empirical or theoretical evidence" that redundancy is promoted by natural selection, and that quotation is accurate to the page.²⁸

Where Terborg's paper fails its sources

A withdrawn result cited as current. Terborg's support for "cooperating networks of unrelated genes contribute significantly more to robustness than gene copy number" is his reference 23, Kitami and Nadeau 2002.¹⁴ Its authors withdrew that conclusion in 2002, six years before Terborg wrote.¹⁰
A finding inverted. Terborg writes that the prediction that nonessential genes evolve faster "has not been confirmed."¹⁴ His wording closely follows the opening of Aaron Hirsh and Hunter Fraser's 2001 Nature abstract, which goes on: "Our analysis reveals a highly significant relationship between protein dispensability and evolutionary rate."¹³ The mouse study Terborg does cite explained its null result by "strong selection acting against even very subtle deleterious phenotypes," which is the opposite of genes escaping selection.³³ Human cell data later showed essential genes more conserved.¹⁸
Duplicates offered as evidence against duplication. Terborg's lead example, the SRC gene family, has members "only sharing 60–80% of their sequences."¹⁴ That is a family of recognizable duplicates. The quotation he relays about it explains why their redundancy is kept: damaging point mutations are selected against.
A superseded yeast result. The "no association" rests on Wagner 2000, which Wagner later attributed to limited data.⁶ Terborg cites the 2004 paper for its worm numbers but not for that correction.

Guliuzza's reading of Edelman and Gally also runs backwards. He says they "recognize that backup, or functionally redundant, systems are indicative of design."³¹ They wrote that in engineering "redundancy is built into design," that "this is not the case for biological systems," and that Darwin laid "the argument by design to rest."³⁰ They contrast engineering with biology. They do not concede design.

One design-side outlet has been candid. An unsigned 2015 Evolution News article compared the two views' predictions about nonessential genes and wrote: "The distinctions are not clear cut."³⁴

The engineering analogy, fairly stated

The analogy is not a creationist invention; Nowak's team opened with it. Real engineering practice adds a detail that matters here. For the Curiosity rover's landing, NASA engineer Gerard Holzmann writes, "Running the same landing software on two CPUs in parallel offers little protection against software defects," so the backup computer ran a different, simplified version.³⁵ A copy protects against random failure. A backup built differently protects against a shared flaw. Biology uses both. Duplicates are copies, and Kafri, Levy and Pilpel showed that many are "responsive backup circuits" that switch on when their partner is lost. They concluded that "compensation for gene loss is merely a side effect of sophisticated design principles using functional redundancy."³⁶ Their "design" means engineering logic, not a designer, but the circuit they describe is the kind an engineer draws.

Part 5: Can a constraint test see a backup?

A constraint test finds DNA that has changed less between species than neutral DNA has, and infers that selection kept it. The inference assumes the species share an ancestor. The Functional DNA article covers why that assumption matters for this site's own count of function.

The ENCODE authors state the blind spot themselves. Alignment methods "are not well suited to capture" mutations "whose effect is buffered by redundancy or epistatic effects. Thus, absence of conservation cannot be interpreted as evidence for the lack of function."¹¹

But the blind spot covers less than the catalogue claim implies. Backups that carry any share of the load are conserved, and constraint often sees them better than a knockout does.

Osterwalder and colleagues deleted ten mouse limb enhancers, "each with strong evolutionary conservation." None changed limb shape alone, and pairs did.¹²¹² The backups were found because they were conserved.
Nadav Ahituv's team deleted four ultraconserved elements in 2007 and saw nothing.³⁷ In 2018 a related team deleted ultraconserved enhancers near the gene Arx and found "neurological or growth abnormalities." They concluded that strong conservation "likely results from fitness deficits that appear subtle in a laboratory setting."³⁸
Human population data flagged 3,230 genes as intolerant of loss, "72% having no currently established human disease phenotype."³⁹ Constraint found them before medicine did.

The reason is the one Thatcher gave. Selection notices any gene whose loss costs more than about one over the population size, which is far below what a lab can measure.³ A partial backup is conserved. Only a perfect backup, one whose loss costs nothing in any condition until the primary fails, would drift and score as unconstrained. Evolutionary theory predicts such backups decay, and the constraint test would miss any that exist. Nobody has counted them.

So "a constraint test cannot see it" is true of perfect redundancy and false of partial redundancy. A constraint-based count of functional DNA is a lower bound in this respect, which is how Functional DNA already uses it.

Objections and responses

"Most knockouts do nothing, so most genes must be backups." Most silent knockouts are genes needed under other conditions or genes with small costs.³⁴ The honest form of the argument uses measured backup, which is a minority of silent knockouts.

"Natural selection cannot keep a gene it never sees, so redundancy refutes evolution." That holds for a perfect backup, and evolutionary theorists say so.¹ Most measured backups are partial, and selection maintains them. The argument has force only where perfect redundancy is shown.

"Redundancy between unrelated genes cannot come from duplication, so evolution cannot explain it." Evolutionary models offer a non-duplication route.¹ The better objection is that those models assume the key molecular step.

"Engineers build redundancy, so redundancy shows engineering." Engineers also build without redundancy when cost matters, and Wagner notes that for any system with redundant parts "one can design a system without redundancy but the same" behaviour, so the two may be impossible to tell apart from outside.⁹ The analogy is suggestive, not decisive.

A skeptic's objection: "Redundancy is just leftover duplication, the mess an unguided process leaves." Leftovers should decay within a few million years, and many duplicate pairs have kept overlapping functions for about 100 million years.²³²⁴ Some are wired to switch on when their partner fails.³⁶ That is regulation, not debris.

A skeptic's objection: "The design case rests on perfect backups nobody has found." Largely correct. The measured examples are partial, and the design argument has not identified perfect ones. It is the design side's burden to find them.

Issues and unanswered questions

The key class is unmeasured. No study counts genes whose loss has zero cost in every condition while a partner is present. Competition assays across many environments, as Thatcher and Hillenmeyer used, combined with double knockouts, could do it.
A young-earth prediction exists and has not been tested. Terborg's own claim is that redundant genes cannot have lasted millions of years.¹⁴ If perfect backups were created a few thousand years ago, they should still be nearly intact in every population.⁴⁰ Evolution predicts they should be mostly broken or gone. That is a real test, and neither side has run it.
The human numbers are ambiguous. The 1,815 genes seen broken in living adults could be backups, genes whose loss causes late or subtle harm, or genes already decaying, which fits Genetic Entropy as well as it fits dispensability. The data do not say which.
Several key papers were read as abstracts only. Gu 2003, Papp 2004 and DeLuna 2008 are paywalled. Their numbers are cited from abstracts, and Gu's bounds from two later papers that report them.

Verdict

Genetic redundancy is real and abundant, and the design reading of it is weaker than the creationist literature claims. Most "silent" knockouts are not backups. Duplicate genes supply more of the measured backup than the design argument allows. Evolutionary theorists have models for keeping partial redundancy and one for building it without duplication, though that model assumes the step that matters most.

What survives is a narrower argument that has not been refuted. Perfect backups are the one kind evolutionary theory expects to disappear and the one kind a conservation test cannot see. If they exist in number, they are evidence for design and against using conservation to measure function. Whether they exist in number is unknown, and the young-earth model makes a testable prediction about them that no one has checked.

Sources

1.
Nowak, Martin A. et al. “Evolution of genetic redundancy.” Nature 388(6638):167–171. 1997. Read in the Nowak lab reprint, recovered from the Wayback Machine. Page numbers printed on the page. Text extraction drops superscripts; the exponents are restored from the sense. Mirror: Archive.org
1. 1a.
  Page 167: “Yet there is a problem: genes encoding functional proteins must be under selection pressure. If a gene was truly redundant then it would not be protected against the accumulation of deleterious mutations. A widespread view is therefore that such redundancy cannot be evolutionarily stable.” ↩ ↩ ↩
2. 1b.
  Page 167: “Such a redundant genetic organization is sensible from an engineer's point of view: important functions require backup devices that can take over in case of failure. But can natural selection favour the emergence and stability of redundant genes?” ↩
3. 1c.
  Page 168: “For a mutation rate of 10⁻⁶ per gene per generation, and a 10% difference between ua and ub, the average lifetime of redundancy is about 10⁷ generations. Therefore, a certain amount of redundancy in our genomes could be the consequence of recent gene duplication events. We note, however, that redundancy cannot only be a consequence of gene duplication because very different genes can also show overlapping redundancy.” ↩ ↩ ↩
4. 1d.
  Page 168, Figure 2 caption: In the simulation, a mutation can remove one of a gene's functions or change the efficacy with which it performs a function to a random value between 0 and 1, at a rate of one-twentieth of the gene's inactivation rate. Paraphrased, because the caption's two columns interleave in the text extraction. ↩
5. 1e.
  Page 169: “The starting configuration is neither redundant nor pleiotropic: each function is performed by one gene, and each gene performs one function. During the simulation, mutations that lead to functional overlap arise spontaneously and can be favoured by selection. Genes evolve to perform additional functions. … Final configurations often consist of complex redundancy–pleiotropy networks in which each function is performed by several genes and each gene performs several functions.” Model 2, a less efficient backup with a lower mutation rate, is on page 168; model 4, developmental errors, on page 169: "Thus the mutation rate in each gene has to be smaller than the developmental error rate in the other gene." ↩ ↩ ↩ ↩
2.
Hillenmeyer, Maureen E. et al. “The chemical genomic portrait of yeast: uncovering a phenotype for all genes.” Science 320(5874):362–365. 2008. Read in the NIH author manuscript, which has no page numbers; paragraphs are counted in the main text. Mirror: PMC
1. 2a.
  Section Abstract: “because complete deletion of most yeast genes (~80%) has no obvious phenotypic consequence in rich medium, it is difficult to study their functions.” ↩
2. 2b.
  Paragraph 3: “Previous studies revealed that 34% of homozygous deletion strains display a distinct phenotype (19% lethality and 15% fitness defect) when grown in rich medium (2, 3, 9).” ↩
3. 2c.
  Paragraph 4: “Only 205 strains (3%) failed to exhibit a phenotype under either homozygous or heterozygous conditions. Thus, considering all ~6000 yeast genes, 97% manifested a significant fitness defect when deleted. In a control analysis of rich-medium experiments, only about 10% of strains manifested a false positive fitness defect”. ↩ ↩
4. 2d.
  Paragraph final: “We have found conditions that render the remaining genes essential for optimal growth in some condition and show that genetic redundancy is therefore limited in providing tolerance of the tested conditions (5-7).” ↩
3.
Thatcher, John W. et al. “Marginal fitness contributions of nonessential genes in yeast.” Proceedings of the National Academy of Sciences 95(1):253–257. 1998. Mirror: PMC
1. 3a.
  Section Discussion, paragraph 1: “In contrast, and consistent with the marginal benefit model, at least 19 of 27 strains (70%) that have no conventional phenotype do have significantly reduced fitness. The fitness defects relative to the parental stock ranged from about 0.3% to almost 22%, but most were below 5% and several below 1%.” ↩ ↩ ↩
2. 3b.
  Section Discussion: “The theory of nearly neutral replacement (8, 15) indicates that a selection coefficient approximately the inverse of the effective population size is sufficient to prevent loss of function by mutation and drift. Because effective population sizes for yeast are very large (10, 16), fitness contributions much smaller than those we measured would be important.” ↩ ↩
4.
Papp, Balázs et al. “Metabolic network analysis of the causes and evolution of enzyme dispensability in yeast.” Nature 429(6992):661–664. 2004. Abstract only, from PubMed; the full text is paywalled.
1. 4a.
  Section Abstract: “Our model indicates that this is the dominant explanation for apparent dispensability, accounting for 37-68% of dispensable genes, whereas 15-28% of them are compensated by a duplicate, and only 4-17% are buffered by metabolic network flux reorganization.” ↩ ↩ ↩
5.
Gu, Zhenglong et al. “Role of duplicate genes in genetic robustness against null mutations.” Nature 421(6918):63–66. 2003. Abstract only, from PubMed; the full text is paywalled. The 23% to 59% bounds are as reported by Conant and Wagner 2004 and Wagner 2005.
1. 5a.
  Section Abstract: “Here we show that there is a significantly higher probability of functional compensation for a duplicate gene than for a singleton, a high correlation between the frequency of compensation and the sequence similarity of two duplicates, and a higher probability of a severe fitness effect when the duplicate copy that is more highly expressed is deleted. We estimate that in S. cerevisiae at least a quarter of those gene deletions that have no phenotype are compensated by duplicate genes.” ↩ ↩
6.
Conant, Gavin C., and Andreas Wagner. “Duplicate genes and robustness to transient gene knock-downs in Caenorhabditis elegans.” Proceedings of the Royal Society of London B 271(1534):89–96. 2004. Read in the PMC scan. Pages from the running heads: PDF page 2 is headed 90, page 6 is 94, page 7 is 95. Mirror: PMC
1. 6a.
  Page 89: “We found that 89% of single-copy and 96% of duplicate genes show no detectable phenotypic effect in an RNAi knock-down experiment.” ↩
2. 6b.
  Page 89: “This analysis, based on over 250 synthetic null (gene-knockout) mutations, found that more than 40% of mutations with no phenotypic effect occurred in single-copy genes. It also showed little support for the role of gene duplications in robustness, a result caused by the limited amount of gene-knockout data available at the time.” "This analysis" is Wagner 2000. ↩ ↩ ↩
3. 6c.
  Page 94: “However, the massive scale—more than 7000 genes—at which such convergence would have to occur makes this seem unlikely. A second possibility is that much mutational robustness is the result of interactions of unrelated genes in genetic networks.” ↩
4. 6d.
  Page 94: “The main caveat to this lower bound is that RNAi detects fewer phenotypic effects than does gene knock-out in yeast, biasing the estimate. … In summary, the available yeast data suggest that the contribution of duplicate genes to weak phenotypic effects is between 23% and 59%, whereas the corresponding range for the worm is 3–36%.” ↩ ↩ ↩
5. 6e.
  Page 94: “Although the absolute number (7872) of single-copy genes with no knock-down effect is higher than the number of duplicate genes with no knock-down effect, proportionally more duplicate genes have no knock-down effect than do single-copy genes.” Page 95: “In summary, the worm genome contains thousands of single-copy genes with absent phenotypic effects.” ↩ ↩
7.
Blank, Lars M. et al. “Large-scale 13C-flux analysis reveals mechanistic principles of metabolic network robustness to null mutations in yeast.” Genome Biology 6(6):R49. 2005. CC BY. Mirror: PMC
1. 7a.
  Section Abstract, Conclusions: “The apparent dispensability of knockout mutants with metabolic function is explained by gene inactivity under a particular condition in about half of the cases. For the remaining 207 viable mutants of active reactions, network redundancy through duplicate genes was the major (75%) and alternative pathways the minor (25%) molecular mechanism of genetic network robustness in S. cerevisiae.” ↩ ↩
8.
Wagner, Andreas. “Robustness against mutations in genetic networks of yeast.” Nature Genetics 24(4):355–361. 2000. Read in the author's reprint, recovered from the Wayback Machine. Page numbers printed on the page. Mirror: Archive.org
1. 8a.
  Page 355: “The second mechanism has its origin in interactions between genes with unrelated functions, and has been documented in metabolic and regulatory gene networks. … My results demonstrate that interactions among unrelated genes are the major cause of robustness against mutations. This type of robustness is probably an evolved response of genetic networks to stabilizing selection.” ↩ ↩
2. 8b.
  Page 359: “If so, at least one back-up gene with identical biochemical function, but completely unrelated sequence and perhaps even unrelated structure, should exist, for each dispensable gene. Such back-up genes may exist, but are probably rare34.” ↩ ↩
3. 8c.
  Page 359: “Increased robustness against mutations in genetic networks would evolve via an indirect mechanism in which robust networks do not confer higher fitness on their carrier, but accumulate in a population because mutations within networks are less likely to have deleterious effects. As a consequence, enhanced robustness increases the mean fitness of a population by an amount that may only be of the order of the mutation rate36,37. Thus, evolution of robustness is not an adaptive phenomenon, even if it increases resilience to mutations within a network by orders of magnitude.” ↩
9.
Wagner, Andreas. “Distributed robustness versus redundancy as causes of mutational robustness.” BioEssays 27(2):176–188. 2005. Read in the author's reprint, recovered from the Wayback Machine. Page numbers printed on the page. Mirror: Archive.org
1. 9a.
  Page 177: “In distributed robustness, many parts of a system contribute to system function, but all of these parts have different roles. When one part fails or is changed through mutations, the system can compensate for this failure, but not because a ''back-up'' redundant part takes over the failed part's role.” ↩ ↩
2. 9b.
  Page 186: “General theorems of control theory imply that for any system with redundant parts, one can design a system without redundancy but the same transfer function, i.e. the same behavior (ref. 95, Chapter 4). … Beyond such enumeration, however, it may be impossible to distinguish distributed robustness and redundancy, because the two have indistinguishable signatures on the variables necessary to understand a system's function and robustness.” ↩
10.
Kitami, Toshimori, and Joseph H. Nadeau. “Corrigendum: Biochemical networking contributes more to genetic buffering in human and mouse metabolic pathways than does gene duplication.” Nature Genetics 32:681. Dec 2002. Corrects Nature Genetics 32:191–194 (2002), doi 10.1038/ng945.
1. 10a.
  Paragraph 1: “To the authors' dismay, a software bug adversely affected some but not all of the main conclusions of this article. … Reanalysis of the entire data set with the corrected software shows that redundant networks do not contribute more to genetic buffering than do gene families. One of the main conclusions from the data should now be that redundant networks provide no greater genetic buffering than do gene families.” ↩ ↩ ↩
11.
Kellis, Manolis et al. “Defining functional DNA elements in the human genome.” Proceedings of the National Academy of Sciences 111(17):6131–6138. 2014. Mirror: PMC
1. 11a.
  Section Evolutionary Approach: “More generally, alignment methods are not well suited to capture substitutions that preserve function, such as compensatory changes preserving RNA structure, affinity-preserving substitutions within regulatory motifs, or mutations whose effect is buffered by redundancy or epistatic effects. Thus, absence of conservation cannot be interpreted as evidence for the lack of function.” ↩ ↩
12.
Osterwalder, Marco et al. “Enhancer redundancy provides phenotypic robustness in mammalian development.” Nature 554(7691):239–243. 2018. Read in the NIH author manuscript. Mirror: PMC
1. 12a.
  Section Summary: “Surprisingly, none of ten deletions of individual enhancers caused noticeable changes in limb morphology. In contrast, removal of pairs of limb enhancers near the same gene resulted in discernible phenotypes, indicating that enhancers function redundantly in establishing normal morphology.” ↩
2. 12b.
  Section main text, first results paragraph: “We used CRISPR/Cas9 genome editing to individually delete ten embryonic enhancers, each with strong evolutionary conservation and robust limb activity in transgenic mouse reporter assays”. “these results suggest that a substantial proportion of limb enhancers, even if highly conserved in evolution, are not individually essential for normal limb morphogenesis.” ↩ ↩
13.
Hirsh, Aaron E., and Hunter B. Fraser. “Protein dispensability and rate of evolution.” Nature 411(6841):1046–1049. 2001. Abstract only, from PubMed.
1. 13a.
  Section Abstract: “The rationale for this prediction is that relatively dispensable proteins should be subject to weaker purifying selection, and should therefore accumulate mildly deleterious substitutions more rapidly. Although this argument was presented over twenty years ago, and is fundamental to many applications of evolutionary theory, the prediction has proved difficult to confirm. … Our analysis reveals a highly significant relationship between protein dispensability and evolutionary rate”. ↩ ↩
14.
Terborg, Peer. “Evidence for the design of life: part 1—Genetic redundancy.” Journal of Creation 22(2):79–84. Aug 2008. Published by Creation Ministries International. Read in CMI's PDF; page numbers printed on the page. Eric's research notes attribute this paper to Peter Borger.
1. 14a.
  Page 81: “Genetic redundancy falsifies several evolutionary hypotheses. Firstly, truly redundant genes are impossible paradoxes because natural selection cannot prevent the accumulation of harmful mutations in these genes.” Page 82: “This argument, which was presented over twenty years ago, is fundamental to many theoretical applications of evolutionary theory, but despite intense scientific scrutiny the prediction has not been confirmed.” ↩ ↩
2. 14b.
  Page 81: “Yet the proteins of the SRC family are distinctly different, only sharing 60–80% of their sequences.” ↩
3. 14c.
  Page 82: “Hence, Ohno's theory predicts an association between genetic redundancy and gene duplication.” ↩
4. 14d.
  Page 82: “More recent studies have shown that cooperating networks of unrelated genes contribute significantly more to robustness than gene copy number.23 … The predicted association between genetic redundancy and gene duplication is nonexistent. Ohno's interesting idea of evolution by gene duplication therefore cannot be right.” Reference 23 is Kitami and Nadeau, Nature Genetics 32:191–194, 2002. ↩ ↩ ↩
5. 14e.
  Page 83: “Scale-free genetic and protein networks are an intrinsic, engineered characteristic of genomes and may explain why genetic redundancy is so widespread among organisms.” ↩
6. 14f.
  Page 84: “It is also important to realize that redundant genes cannot have resided in the genome for millions of years, because natural selection, a conservative force, cannot prevent their destruction due to debilitating mutations.” ↩ ↩
15.
Baba, Tomoya et al. “Construction of Escherichia coli K-12 in-frame, single-gene knockout mutants: the Keio collection.” Molecular Systems Biology 2. 2006. Article number 2006.0008. Read from the PMC copy. Mirror: PMC
1. 15a.
  Section Abstract: “Of 4288 genes targeted, mutants were obtained for 3985. … We were unable to disrupt 303 genes, including 37 of unknown function, which are candidates for essential genes.” ↩
16.
Kobayashi, K. et al. “Essential Bacillus subtilis genes.” Proceedings of the National Academy of Sciences 100(8):4678–4683. 2003. Mirror: PMC
1. 16a.
  Section Abstract: “Among ≈4,100 genes of the organism, only 192 were shown to be indispensable by this or previous work. Another 79 genes were predicted to be essential.” ↩
17.
Dickinson, Mary E. et al. “High-throughput discovery of novel developmental phenotypes.” Nature 537(7621):508–514. 2016. International Mouse Phenotyping Consortium. Read in the NIH author manuscript. Mirror: PMC
1. 17a.
  Section Results, first paragraph: “further support that approximately 35% (24% lethal and 11% subviable) of null mutations across the genome are essential for survival at normal Mendelian ratios.” ↩
2. 17b.
  Section Results, incomplete-penetrance paragraph: “Prior work has shown that lethal genes are much less likely than viable genes to have a paralog, and thus less potential for functional redundancy”. ↩
18.
Wang, Tim et al. “Identification and characterization of essential genes in the human genome.” Science 350(6264):1096–1101. 2015. Screens in the KBM7 human leukemia cell line. Read in the NIH author manuscript. Mirror: PMC
1. 18a.
  Section main text: “Of the 18,166 genes targeted by the library, 1,878 scored as essential for optimal proliferation in our screen, although this precise number depends on the cutoff chosen (Fig. 2A; Table S2-3). Overall, this fraction represents ∼10% of genes within our dataset or roughly 9.2% of the entire genome”. ↩
2. 18b.
  Section main text: “Essential genes should be under strong purifying selection and should thus show greater evolutionary constraint than non-essential genes (12). Consistent with this expectation, the essential genes found in our screens were more broadly retained across species, showed higher levels of conservation between closely related species, and contain fewer inactivating polymorphisms within the human species”. ↩
3. 18c.
  Section main text: “Using the results from our genome-wide screens, we revisited this question and observed that genes with paralogs are indeed less likely to be essential, consistent with the idea that paralogs can provide functional redundancy at the cellular level (Fig. 2J).” ↩
19.
Karczewski, Konrad J. et al. “The mutational constraint spectrum quantified from variation in 141,456 humans.” Nature 581(7809):434–443. 2020. Pages from the footers: PDF page 1 is 434, page 3 is 436, page 7 is 440.
1. 19a.
  Page 437: “we subjected this set to additional filtering and deep manual curation before defining a set of 1,815 genes (2,636 high-confidence variants) that are likely to be tolerant to biallelic inactivation”. ↩
2. 19b.
  Page 440: “these metrics primarily identify genes undergoing selection against heterozygous variation, rather than strong constraint against homozygous variation45. … Furthermore, selection is largely blind to phenotypes emerging after reproductive age”. ↩
20.
DeLuna, Alexander et al. “Exposing the fitness contribution of duplicated genes.” Nature Genetics 40(5):676–681. 2008. Abstract only, from PubMed; the full text is paywalled and was not read.
1. 20a.
  Section Abstract: “We find that the functional overlap between paralogs sufficiently explains the apparent dispensability of individual WGD genes.” ↩
21.
Hsiao, Tzu-Lin, and Dennis Vitkup. “Role of duplicate genes in robustness against deleterious human mutations.” PLoS Genetics 4(3):e1000014. 2008. CC BY. Mirror: PMC
1. 21a.
  Section Abstract: “Genes with a 90% sequence identity homolog are about 3 times less likely to harbor known disease mutations compared to genes with remote homologs.” ↩
22.
Costanzo, Michael et al. “A global genetic interaction network maps a wiring diagram of cellular function.” Science 353(6306):aaf1420. 2016. Read in the NIH author manuscript. Mirror: PMC
1. 22a.
  Section Abstract: “We generated a global genetic interaction network for Saccharomyces cerevisiae, constructing over 23 million double mutants, identifying ~550,000 negative and ~350,000 positive genetic interactions.” Section Genetic interaction network connectivity: “Although the average gene participated in ~100 negative interactions (2% of genes tested) and ~65 positive interactions (1% of genes tested)”. ↩
23.
Lynch, Michael, and John S. Conery. “The evolutionary fate and consequences of duplicate genes.” Science 290(5494):1151–1155. 2000. Read in the Lynch lab reprint; the first page of the article is headed 1151. Mirror: Archive.org
1. 23a.
  Page 1151: “Observations from the genomic databases for several eukaryotic species suggest that duplicate genes arise at a very high rate, on average 0.01 per gene per million years. … However, the vast majority of gene duplicates are silenced within a few million years, with the few survivors subsequently experiencing strong purifying selection.” ↩ ↩
24.
Dean, Eric J. et al. “Pervasive and persistent redundancy among duplicated genes in yeast.” PLoS Genetics 4(7):e1000113. 2008. CC BY. Mirror: PMC
1. 24a.
  Section Abstract: “We demonstrate that duplicated genes can maintain substantial redundancy for extensive periods of time following duplication (∼100 million years).” ↩ ↩
25.
Lynch, Michael. “Evolutionary layering and the limits to cellular perfection.” Proceedings of the National Academy of Sciences 109(46):18851–18856. 2012. Mirror: PMC
1. 25a.
  Section Introduction: “Although these layered lines of defense are clearly advantageous and in many cases essential to cell health, because the simultaneous emergence of all components of a system is implausible, several questions immediately arise. How can selection promote the establishment of additional layers of fitness-enhancing mechanisms if the established primary lines of defense are already highly refined?” ↩
26.
Krakauer, David C., and Joshua B. Plotkin. “Redundancy, antiredundancy, and the robustness of genomes.” Proceedings of the National Academy of Sciences 99(3):1405–1409. 2002. Mirror: PMC
1. 26a.
  Section Abstract: “Assuming a cost of redundancy, we find that large populations will evolve antiredundant mechanisms for removing mutants and thereby bolster the robustness of wild-type genomes; whereas small populations will evolve redundancy to ensure that all individuals have a high chance of survival.” ↩
2. 26b.
  Section Introduction: “We do not describe explicitly how redundancy will have arisen or is encoded (13), nor do we estimate the time over which redundancy is expected to be preserved (14).” ↩
27.
Lynch, Michael. “The frailty of adaptive hypotheses for the origins of organismal complexity.” Proceedings of the National Academy of Sciences 104(suppl. 1):8597–8604. 2007. Mirror: PMC
1. 27a.
  Section final section: “If complexity, modularity, evolvability, and/or robustness are entirely products of adaptive processes, then where is the evidence?” ↩
28.
Lynch, Michael. “The evolution of genetic networks by non-adaptive processes.” Nature Reviews Genetics 8(10):803–813. Oct 2007. Read from the author's reprint archived from the Lynch lab site. Copyright. Mirror: Archive.org
1. 28a.
  Page 810, section Concluding remarks: “Contrary to widespread belief, there is no compelling empirical or theoretical evidence that complexity, modularity, redundancy or other features of genetic pathways are promoted by natural selection.” Folio anchored on reprint pages 1 (803) and 11 (813); the sentence opens the closing section on reprint page 8. ↩ ↩
29.
Force, Allan et al. “Preservation of duplicate genes by complementary, degenerative mutations.” Genetics 151(4):1531–1545. 1999. Mirror: PMC
1. 29a.
  Page 1531: “The duplication-degeneration-complementation (DDC) model predicts that (1) degenerative mutations in regulatory elements can increase rather than reduce the probability of duplicate gene preservation and (2) the usual mechanism of duplicate gene preservation is the partitioning of ancestral functions rather than the evolution of new functions.” ↩
30.
Edelman, Gerald M., and Joseph A. Gally. “Degeneracy and complexity in biological systems.” Proceedings of the National Academy of Sciences 98(24):13763–13768. 2001. Read in the PMC scan; page numbers printed on the page. Mirror: PMC
1. 30a.
  Page 13763: “Degeneracy, the ability of elements that are structurally different to perform the same function or yield the same output, is a well known characteristic of the genetic code and immune systems. Here, we point out that degeneracy is a ubiquitous biological property and argue that it is a feature of complexity at genetic, cellular, system, and population levels. Furthermore, it is both necessary for, and an inevitable outcome of, natural selection.” ↩
2. 30b.
  Page 13763: “The contrast between degeneracy and redundancy at the structural level is sharpened by comparing design and selection in engineering and evolution, respectively. In engineering systems, logic prevails, and, for fail-safe operation, redundancy is built into design. This is not the case for biological systems. Indeed, not the least of Darwin's achievements was to lay the argument by design to rest.” ↩
31.
Guliuzza, Randy J. “Major Evolutionary Blunders: The 'Degenerate' Genetic Code?” Acts & Facts 45(8). Institute for Creation Research. Jul 29, 2016. Read from a curl download of the page.
1. 31a.
  Section The Genetic Code Is a Design Marvel: “It is also telling how Gerald Edelman and Joseph Gally recognize that backup, or functionally redundant, systems are indicative of design.” ↩
2. 31b.
  Section The Genetic Code Is a Design Marvel: “When humans can identify the true source of fail-safe redundancy, it always is an indicator of good design and a good designer.” ↩
32.
Luskin, Casey. “Natural Selection Struggles to Fix Advantageous Traits in Populations.” Evolution News (now Science and Culture Today). Oct 23, 2014. Luskin is research director of the Discovery Institute's Center for Science and Culture. Read from a curl download of the page.
1. 32a.
  Section Complex Redundancy: “In Lynch's view, there are many cellular systems that aid in survival, but are redundant. As a result, they serve as backup mechanisms that are only used when a highly effective primary system fails. … How can a system that is only rarely used, or only occasionally needed, evolve to such a high and efficient level of complexity?” ↩
33.
Hurst, Laurence D., and Nick G. C. Smith. “Do essential genes evolve slowly?” Current Biology 9(14):747–750. 1999. Abstract only, from PubMed. Terborg 2008 cites this paper.
1. 33a.
  Section Abstract: “After correction for this bias, we found that the rate at which genes evolve does not correlate with the severity of the knockout phenotype. … Our findings most probably reflect strong selection acting against even very subtle deleterious phenotypes”. ↩
34.
Evolution News. “Why Would Evolution Produce Non-Essential Genes?” Evolution News (now Science and Culture Today). Dec 22, 2015. Unsigned. Published by the Discovery Institute's Center for Science and Culture. Read from a curl download of the page.
1. 34a.
  Section Contrasting Predictions: “Let's integrate this information by contrasting the predictions of design and Darwinism about essentiality. The distinctions are not clear cut. … So who's winning this debate on essentiality? It's too early to tell. Not enough is known yet.” ↩
35.
Holzmann, Gerard J. “Mars Code.” Communications of the ACM 57(2):64–73. Feb 2014. Read in the author's reprint; the page footer reads 73. Mirror: Archive.org
1. 35a.
  Page 73: “Running the same landing software on two CPUs in parallel offers little protection against software defects. Two different versions of the entry-descent-and-landing code were therefore developed, with the version running on the backup CPU a simplified version of the primary version running on the main CPU.” ↩
36.
Kafri, Ran et al. “The regulatory utilization of genetic redundancy through responsive backup circuits.” Proceedings of the National Academy of Sciences 103(31):11653–11658. 2006. Mirror: PMC
1. 36a.
  Section Abstract: “many such backed-up genes were shown to be transcriptionally responsive to the intactness of their redundant partner and are up-regulated if the latter is mutationally inactivated. … We thus challenge the view that such redundancies are simply leftovers of ancient duplications and suggest they are an additional component to the sophisticated machinery of cellular regulation. In this respect, we suggest that compensation for gene loss is merely a side effect of sophisticated design principles using functional redundancy.” ↩ ↩
37.
Ahituv, Nadav et al. “Deletion of ultraconserved elements yields viable mice.” PLoS Biology 5(9):e234. 2007. Public-domain declaration. Read in PMC HTML. Mirror: PMC
1. 37a.
  Section Abstract: “Remarkably, all four resulting lines of mice lacking these ultraconserved elements were viable and fertile, and failed to reveal any critical abnormalities when assayed for a variety of phenotypes including growth, longevity, pathology, and metabolism.” ↩
38.
Dickel, Diane E. et al. “Ultraconserved enhancers are required for normal development.” Cell 172(3):491–499. 2018. Read in the NIH author manuscript. Mirror: PMC
1. 38a.
  Section Summary: “Mice with single or pairwise deletions of ultraconserved enhancers were viable and fertile, but in nearly all cases showed neurological or growth abnormalities, including substantial alterations of neuron populations and structural brain defects. Our results demonstrate the functional importance of ultraconserved enhancers and indicate that remarkably strong sequence conservation likely results from fitness deficits that appear subtle in a laboratory setting.” ↩
39.
Lek, Monkol et al. “Analysis of protein-coding genetic variation in 60,706 humans.” Nature 536(7616):285–291. 2016. Exome Aggregation Consortium. Read in the NIH author manuscript. Mirror: PMC
1. 39a.
  Section Summary: “identifying 3,230 genes with near-complete depletion of truncating variants with 72% having no currently established human disease phenotype.” ↩
40.
Scale arithmetic, not a published model. Nowak and colleagues use a loss-of-function rate of 10⁻⁶ per gene per generation; rates up to 10⁻⁵ are sometimes used. At 25 to 30 years per generation, 6,000 years is about 200 to 240 human generations. The expected share of copies disabled by new mutations in that time is about 240 × 10⁻⁶ to 240 × 10⁻⁵, or 0.02% to 0.24%, before drift. Against that, Nowak's model gives an average lifetime of about 10⁷ generations.¹ Species with short generations would show more decay, and a test would have to account for that. ↩