Lenski's Long-Term Evolution Experiment

"No population evolved the capacity to exploit citrate for >30,000 generations, although each population tested billions of mutations." — Blount, Borland and Lenski, PNAS, 2008¹ ¹

"We conclude that the rarity of the LTEE mutant was an artifact of the experimental conditions and not a unique evolutionary event. No new genetic information (novel gene function) evolved." — Van Hofwegen, Hovde and Minnich, Journal of Bacteriology, 2016² ²

"But that's like saying a new book — say, Darwin's Origin of Species when it first appeared in 1859 — contains no new information, because the text has the same old letters and words that are found in other books." — Richard Lenski and Zachary Blount, replying, 2016³ ³

Summary

In 1988 Richard Lenski put twelve flasks of Escherichia coli on a shaker and started transferring one percent of each culture into fresh food every morning. He has not stopped. The experiment left Michigan State for Jeffrey Barrick's lab in Texas in 2022 and came home with him in 2025; by November 2025 it had passed 82,000 generations.⁴ It is the longest controlled evolution experiment ever run, and it is the closest thing biology has to a direct measurement of what mutation and selection can build when nobody is helping.

This article works through the numbers. Every claim rests on the primary papers, which are listed in the literature review; the verbatim quotations on both sides are collected in quotes. The conclusions, in order of confidence:

The scale is genuinely enormous, and it is smaller than it sounds. Each population makes about half a billion new cells a day and has produced roughly 6 × 10¹² cells over the whole run; all twelve together, about 7 × 10¹³.⁵ At the measured mutation rate of 8.9 × 10⁻¹¹ per base pair per generation,⁶ that is about 2.5 billion point mutations per population — enough that every one of the 13.9 million possible single-letter changes in the 4.63-million-base genome has been made roughly 180 times over in each flask.⁵ But those cells never coexist. The standing population is only 5 × 10⁸ at its daily peak and 5 × 10⁶ after each transfer,⁷ so the experiment samples single mutations exhaustively and specific pairs of mutations hardly at all. That distinction does most of the work in every dispute below.
In 38 years the experiment produced exactly one new metabolic ability, in one of twelve populations. After roughly 31,000 generations, population Ara–3 learned to grow on citrate in the presence of oxygen — something E. coli's inability to do has been part of the species' clinical definition for a century.¹ Through 75,000 generations, none of the other eleven has done it, and that figure comes from a peer-reviewed paper by Lenski's own group rather than from a press release.⁸ ⁸ Citrate had been sitting in the medium at twelve times the molar concentration of the glucose the entire time.¹ ⁹
The citrate ability is a real gain of function and a rearrangement of existing parts. Both statements are true, and the argument is mostly about which one to call it. The actualizing mutation is a 2,933-base tandem duplication that placed a copy of the silent citrate transporter gene citT downstream of the rnk promoter, which does fire in oxygen.¹⁰ ¹⁰ Nothing new was written; something old was rewired. The cell can now do something no ancestor could do, and can grow on citrate as its sole carbon source, which even anaerobic E. coli cannot.³
The Discovery Institute's recent "stepwise" claim is correct as written, and the measurements behind it were made by Lenski's own former students. Jonathan McLatchie, writing in August 2026, says the path to citrate use "is stepwise, conferring a selective advantage at each stage," and credits Scott Minnich's 2016 paper.¹¹ ¹¹ The trait-creating mutation really is separately selectable, and the measurement is not Minnich's: Dacia Leon and Jeffrey Barrick built a single-copy mimic of the evolved rnk-citT module, put it into the chromosome of the unevolved 1988 ancestor, and measured a 1.0% fitness advantage over an isogenic partner carrying only the antibiotic marker.¹² ¹² ¹² Erik Quandt and Barrick had already shown that citT plus one dctA mutation are sufficient for the full-strength trait, so "the earlier mutations thought to have potentiated the initial evolution of Cit+ are not required."¹³ ¹³ Anyone who says the citrate trait needed a string of neutral mutations before it could work is out of date. Note carefully what was measured, though. Leon's construct is the promoter capture — the step that creates the phenotype. The bare citT amplification that McLatchie names as step one has never had a fitness value put on it under LTEE conditions, by anybody.
What the claim leaves out is not a step but a delay, and that is where the interesting number is. McLatchie's "stages" are the three mutational steps — amplify, capture a promoter, tune dctA — and each of the three does carry an advantage in the system he is citing. Leon and Barrick say so about that system by name: starvation conditions "dramatically increase the fitness benefit of evolving the rudimentary Cit+ phenotype," and activating citT "would be universally beneficial in any genetic background in these types of experiments."¹² What the four sentences do not say is that the same second step, measured in the LTEE's own daily-transfer regime, swung with the genetic background it landed in: +1.7% in the earliest cluster, −5.4% for a three-clone stretch between 10,000 and 20,000 generations, +0.4%, then +2.4% by the time the trait appeared.¹² That is an answer to why the LTEE took fifteen years, not to whether each step pays, and it is the single most interesting number in the dispute.
Minnich's headline "12 generations" is a real result reported in a misleading unit. His group grew E. coli in citrate medium and recovered 46 independent citrate-using mutants, some after as few as twelve generations of growth.² But those twelve generations are two weekly transfers of about six generations each, with seven days of starvation in stationary phase in between; the fastest mutant took 19 days and most took a month or two.² ³ Twelve generations of LTEE time is a day and three quarters. The correct comparison is days and cells, not generations, and stated that way the gap between Minnich's weeks and Lenski's fifteen years is still large — but it is a gap between two different selection regimes, which is exactly what Minnich says it is.
Loss of function dominates the ledger by count, and the numbers are not close. All twelve populations deleted the ribose operon within 2,000 generations, and the deletion is worth 1–2% fitness.¹⁴ ¹⁴ Total catabolic breadth across 64 carbon sources fell to 58% of the ancestor's by 20,000 generations.¹⁵ ¹⁵ Six of twelve populations broke a DNA repair or nucleotide-sanitising gene and became about a hundred times more mutable.¹⁶ ¹⁶ Average genome length shrank by 63 kilobases, about 1.4%, over 50,000 generations.¹⁶ Michael Behe's "first rule of adaptive evolution" — break or blunt any functional coded element whose loss would yield a net fitness gain, which he labels a rule of thumb rather than a law¹⁷ ¹⁷ — describes this experiment well, and Lenski says so in print.¹⁸ ¹⁸
But "all of it is degradation" is not established, and the best count does not settle it. Couce and colleagues, with Lenski among the authors, classified the LTEE's adaptive mutations by type and found that presumptive knockouts — nonsense codons, frameshifts, indels — supply more than half of the early adaptive mutations and about a quarter over the long run.¹⁹ ¹⁹ More than half is a real concession to Behe. The quarter is a floor, not a ceiling: the other three-quarters are mostly single amino-acid changes whose effect on their proteins nobody has measured, so the count neither confirms nor refutes his "very likely that all of the identified beneficial mutations worked by degrading or outright breaking" the ancestral genes.²⁰ What tells against "all" is narrower. Rohan Maddamsetti and Lenski found that ten of the 57 most-selected genes show the same amino-acid substitution recurring in different populations, at protein interfaces — the signature of tuning, not of knockout.²¹ Sixteen of the 57 are essential genes, and only one of those carries a genuine knockout.²¹ And Behe himself files the citrate duplication under modification-of-function, not loss.²⁰ Behe reads "fine-tune" as "probably degrade";²⁰ ²⁰ that is an assumption, not a measurement, and so is the opposite reading. The one system where the enzymology was measured runs his way for most of its length: three loss-of-regulation events on the path to citrate use, then a second wave of mutations that throttles citrate synthase below the ancestral level once the trait arrives.²² The article reports that in full below rather than calling it a draw.
Fitness rose 70% and showed no ceiling through 60,000 generations. That refutes one creationist forecast, and not the one usually cited. Relative fitness rose 37% in the first 2,000 generations⁷ and about 70% by 50,000.²³ ²³ A power law with no upper bound fits better than a hyperbola with one, and the fit made a correct out-of-sample prediction ten thousand generations ahead.²³ Mean fitness never declined in any population that was measured — and that qualification matters, because Lenski's 2015 study covers nine of the twelve populations through 60,000 generations, not all twelve through 82,000, and it is the last published measurement.²³ Three were excluded, including Ara–3, the citrate population this article spends most of its length on. Three of the nine that were measured are hypermutators carrying an enormous load of unselected mutations, and they rose too. The leading creationist writers did not predict a decline here. John Sanford, the author of Genetic Entropy, called the LTEE's improvement "not unexpected or remarkable in any way" and put his own forecast, that the strains shed unused functions and eventually go extinct, "hundreds of years" out.²⁴ ²⁴ The creationist forecast the data did refute is Behe's, from 2007: "After a while, beneficial changes from the experiment petered out."²⁵ ²⁵ They did not. Sanford's forecast has a difficulty of its own, priced in the objections below.²⁶
The experiment is not a test of macroevolution, and neither side should treat it as one. Lenski designed it to be as simple as possible and says plainly that it "was not intended to mimic the complexities of nature, nor was it meant to be a test-bed for the evolution of new functions."¹⁸ There is no sex, no gene transfer, no predation, no immune system, one carbon source, one temperature. In 82,000 generations no new protein fold has appeared and no new species has been named. Nine stretches of formerly silent non-coding DNA have started being transcribed, five of them because an insertion element parked its promoter upstream, and one of those is translated — but none has a demonstrated function, and the authors say so.²⁷ A creationist who says the LTEE shows evolution cannot build is over-reading a rigged-easy environment; an evolutionist who says it shows evolution can build eyes is over-reading the same environment in the other direction.

The verdict: The LTEE is the best available measurement of what unguided mutation and selection do to a bacterium in a constant environment, and what they do is mostly tune and prune. One innovation appeared, it was built by re-wiring existing parts, it was hard enough that eleven other identical populations still have not managed it, and it was easy enough that a different laboratory can produce it in a month when the environment rewards it directly. Those two facts are not in tension. They say the barrier was ecological and genetic context, not chemistry — a real answer to a real question, and a narrower one than either side usually claims.

On the specific dispute this article was asked to examine, the Discovery Institute has the better of it, and it should be said plainly. Each of the three mutational steps to aerobic citrate use does carry an advantage; the first one was measurable in the 1988 ancestor; the potentiation story that made the result look like deep contingency has been retired by Lenski's own scientific descendants. What the Discovery Institute's account leaves out is why the LTEE was slow, and there the answer is a fitness valley in the middle of one lineage's history plus competition from cheaper mutations — which is Behe's own mechanism, measured, and which the Discovery Institute could claim if it wanted to. The place where the mainstream case is strongest is elsewhere entirely: fitness rose 70% and was still rising at the last published measurement, 60,000 generations, and the share of adaptive mutations that are outright knockouts fell from more than half early on to about a quarter as the easy ones ran out.

What each side gets right

The Discovery Institute's two lines of argument are one position, and Behe carries both without strain. Darwin Devolves endnote 29 to chapter 7 records, flatly: "The citrate mutation can be generated in weeks rather than decades by altering the bacterial growth regimen," citing Van Hofwegen, Hovde and Minnich.²⁰ So the man who calls the citrate result a "sideshow" also accepts, in the same chapter's notes, that the trait is easy to get when the regimen rewards it. That is consistent, and it is the position this article ends up at too.

The Discovery Institute is right that the LTEE corpus is dominated by degradation, and right that "potentiation" has been quietly demoted. Van Hofwegen, Hovde and Minnich argued in 2016 that the fifteen-year delay was an artifact of the transfer regime rather than a requirement for rare antecedent mutations.² John Roth and Sophie Maisnier-Patin — mainstream Salmonella geneticists with no stake in the design debate — published a commentary in the same issue agreeing: "the delay may not reflect the necessity of a nonselective potentiation event... but rather... selective events with a low probability of serial transfer."²⁸ ²⁸ Two years later Barrick's own lab confirmed the mechanism was competition, not inaccessibility.¹² On the science, Minnich won a point.

Lenski is right that "no new information" is a definition doing the work of an argument — though the sharper charge against Behe is a different one, and this article's earlier draft made the weaker one. Van Hofwegen et al. assert the "no new information" claim three times without offering any measure of information.² Behe's scheme, by contrast, is falsifiable and he said in advance what would falsify it. Writing in 2010, before the mechanism was known, he named the test: "If the phenotype is due to one or more mutations that result in, for example, the addition of a novel genetic regulatory element, gene-duplication with sequence divergence, or the gain of a new binding site, then it will be a noteworthy gain-of-FCT mutation."¹⁷ His own definition of the category lists "the construction by mutation of a new promoter" first.¹⁷

Two years later Blount showed the actualizing event was a duplication junction that created an rnk-citT transcription unit which had not previously existed, expressed from a promoter that had never driven that gene.¹⁰ In 2019 Behe classified it as modification-of-function anyway, "because no new functional coded element was gained or lost, just copied."²⁰ Whether moving a promoter next to a silent gene counts as "the addition of a novel genetic regulatory element" is a real question. What is not in doubt is that Behe posed the test himself and then decided the answer was no. That is a charge about consistency, and it is one the evidence supports; "the criterion cannot be failed" is a charge the evidence does not.

Both sides have quietly changed position and neither has said so loudly. Blount et al. in 2008 proposed that potentiation worked by raising the mutation rate to Cit+.¹ By 2012 the same group had reframed it as epistasis;¹⁰ by 2014 Barrick's lab showed the potentiating alleles were not required at all;¹³ by 2018 the explanation was clonal interference plus an intervening fitness valley.¹² Meanwhile Minnich in 2016 said the LTEE's slowness "merely reflects experimental conditions,"² but in 2025 he and Hovde argued the timing was predictable from a joint mutation probability of 10⁻¹⁴ against 10¹⁴ cells screened — an argument that treats the first step as effectively unselected in the LTEE's regime, while the stepwise-advantage claim made on his behalf describes the direct-selection regime. The two fit together, but the emphasis has moved from "the delay was an artifact" to "the delay was predictable."²⁹ ²⁹ ³⁰

The experiment

The design, in one table

Everything in the LTEE was chosen to be boring. Most of the numbers below come from the founding paper;⁷ the medium's exact glucose and citrate concentrations and the 500-generation freezing interval come from Blount et al. 2008's Methods, which restate the design after twenty years of it.¹

Feature	Value
Started	1988, University of California Irvine; moved to Michigan State in 1991
Populations	12, from one clone of E. coli B
Ancestors	REL606 (Ara⁻, six lines) and REL607 (Ara⁺, six lines); the marker is neutral
Medium	DM25 — Davis minimal broth, 139 µM glucose (25 mg/L), 1,700 µM citrate, thiamine
Volume	10 mL in a 50-mL Erlenmeyer flask
Conditions	37 °C, shaken at 120 rpm, aerobic
Transfer	0.1 mL into 9.9 mL of fresh medium, every day
Growth per day	100-fold, therefore log₂(100) = 6.64 generations
Standing population	~5 × 10⁶ after transfer, ~5 × 10⁸ at stationary phase
Archive	Every population frozen at −80 °C every 500 generations
Genome	4,629,812 bp (REL606, GenBank NC_012967.1)³¹
Point-mutation rate	8.9 × 10⁻¹¹ per bp per generation = 4.1 × 10⁻⁴ per genome per generation⁶
Generations reached	75,000 by 2022; 82,000 by November 2025³² ⁴

The frozen archive is the feature that makes the LTEE different from every other evolution experiment. Every 500 generations of every population is alive in a freezer. Any ancestor can be thawed and made to compete directly against any descendant, so "fitness" here is not inferred from a model — it is a measured ratio of growth rates in a head-to-head race.⁷

The arithmetic of scale

The daily cycle is what generates the enormous numbers. A population goes from 5 × 10⁶ cells to 5 × 10⁸ overnight, so about 4.95 × 10⁸ cells are born per population per day — Lenski's own shorthand is "around half a billion."³³ At 6.64 generations a day, 82,000 generations is 12,342 days of transfers, or a little under 34 years of them. That works out to:⁵

Quantity	Per population	All twelve
Cells born, through 33,127 generations (when Cit+ took off)	2.5 × 10¹²	3.0 × 10¹³
Cells born, through 82,000 generations	6.1 × 10¹²	7.3 × 10¹³
Point mutations produced, through 82,000 generations	2.5 × 10⁹	3.0 × 10¹⁰
Distinct point mutations possible (4,629,812 × 3)	1.4 × 10⁷	—
Times each possible point mutation has been made	~180	~2,200

Blount and Lenski put the same point in words: each population "experienced billions of mutations... far more than the number of possible point mutations in the ≈4.6-million-bp genome. This ratio implies, to a first approximation, that each population tried every typical one-step mutation many times."¹

Two cautions on these figures. Evolved LTEE cells are larger than their ancestors and reach lower cell densities, so the later years produced somewhat fewer cells per day than the design figure implies.¹ And a mutation that arises in the last hour before transfer has almost no chance of surviving the 1:100 dilution, so "tried" is not the same as "given a fair test."

Why 82,000 generations is not a million years of mammal evolution

The comparison is made constantly — Van Hofwegen et al. say the LTEE's generations are "equivalent to over 1 million years of human evolution,"² and Behe uses the same framing.²⁰ It is a useful hook and a bad measure. The LTEE has no sex and no recombination, so every beneficial mutation must arise in a lineage that already carries the previous ones; favourable variants that appear in different cells compete rather than combine, which is why clonal interference dominates the dynamics.³⁴ ³⁴ Lenski's group later tested the transfer regime itself, running 48 populations at 2-, 8-, 100- and 1000-fold daily dilution, and found that adaptation started earlier and went further at the larger dilutions — the signature of a regime where beneficial mutations are common enough to compete with each other.³⁵ The environment never changes, so there is almost nothing to adapt to after the first few thousand generations. And the effective population size is around 10⁷,³⁴ which is a thousand times larger than any vertebrate's — so drift is nearly absent, and mutations that would fix by chance in a mammal are simply removed here.

The right way to use the LTEE is as a lower bound on what mutation and selection can do with a large, asexual, constant-environment population. That is a genuinely useful bound. It is not a scale model of anything with a backbone.

What has been gained

Fitness, and the ceiling that is not there

Mean fitness relative to the frozen ancestor rose 37% in the first 2,000 generations and kept rising, more and more slowly.⁷ Wiser, Ribeck and Lenski fit two models to the trajectory through 50,000 generations:²³

Hyperbolic (a hard ceiling): w = 1 + 0.7007 t / (t + 4431), predicting 1.6437 at 50,000 generations and 1.6525 at 60,000.
Power law (no ceiling): w = (0.00515 t + 1)^0.0950, predicting 1.6951 and 1.7246.

Lenski's group then ran more than 1,100 new competition assays out to 60,000 generations and found the power law fit better.²³ This is a real prediction, made in advance, tested against new data, and confirmed — which is more than most claims in this dispute can say. Whether fitness truly increases without bound is unknowable from 60,000 generations of data; what is established is that no ceiling has been reached and the deceleration is slower than a hard limit would produce.

The mechanism behind the deceleration has now been measured too. Couce and colleagues used transposon libraries to ask what fraction of all possible gene-disrupting insertions are beneficial in a given background, and watched that fraction collapse: 6.8% in the ancestor, 4.3% by 2,000 generations, 3.2% by 15,000, while the deleterious fraction held flat near 20%.¹⁹ The supply of easy improvements is visibly running out. Fitness keeps rising anyway, more slowly, on what is left.

For young-earth readers this result should be stated plainly, together with what it does and does not contradict. A population that has accumulated tens of thousands of mutations, six of whose twelve lines have a broken proofreading system, has not lost fitness in its own medium. All nine measured populations improved at each of the three time points through 60,000 generations, the last published measurement.²³ That contradicts Behe's 2007 statement that the experiment's beneficial changes "petered out,"²⁵ which went further than the source he cited for it: de Visser and Lenski had reported in 2002 that the rate of improvement had slowed, not that it had stopped. It does not contradict Sanford, who expected the rise: "Over a period of a few decades, a strain of E. coli bacteria became slightly better-adapted to an artificial nutrient medium. This is not unexpected or remarkable in any way."²⁴ One qualification, and it is Lenski's group's own: measured against a counterfactual in which its partner is absent, the Cit− clade of population Ara–3 actually lost fitness in the LTEE medium over time.⁸ That is a clade adapting to a cross-feeding partner, not a population decaying, and it is not a counterexample to the fitness trajectory — but a careful reader will find it, and should find this article has already found it.

The citrate ability, step by step

The one qualitative novelty is worth going through carefully, because every argument about the LTEE eventually lands on it.

Why E. coli cannot eat citrate in air. It is not that the cell cannot metabolise citrate — it runs a complete citric-acid cycle and makes citrate itself. The single missing piece is transport: the citrate carrier CitT is an antiporter that swaps citrate in for succinate out, and its operon is switched on only in the absence of oxygen.¹ Growing on citrate aerobically therefore needs the transporter expressed under the wrong conditions. Failing the citrate test on Simmons agar is one of the standard bench criteria for calling an isolate E. coli.¹

Actualization — a duplication that captures a promoter. All nine sequenced Cit+ genomes carry two or more tandem copies of the same 2,933-base segment. The duplication junction fuses the upstream end of rnk to the downstream end of citG, which puts the second copy of citT under the rnk promoter — and rnk fires in oxygen.¹⁰ Blount tested this directly: a reporter under the native citT control region showed no expression above background in any strain, while the evolved rnk-citT module did.¹⁰ Inserting one copy of the module into a Cit− clone made it Cit+, and all thirteen spontaneous Cit− revertants of a Cit+ clone had lost the amplification.¹⁰

The first Cit+ cells were nearly useless. The engineered single-copy strain lagged more than 60 hours after glucose ran out before growing on citrate at all, and grew inconsistently. It still beat its parent by 1.0% in competition.¹⁰ That number matters: it is the measured selective value of the first step, made by Lenski's own lab in 2012, and it is the foundation of the "stepwise" claim now advanced against him.

Refinement — more copies, then a second transporter. The five Cit+ genomes sequenced from before the population expansion carry between two and nine copies of the module, and later Cit+ genomes settle back to four;¹⁰ cloning it onto a high-copy plasmid reproduced the strong phenotype and the population expansion.¹⁰ The decisive refinement was elsewhere. Because CitT pumps a four-carbon acid out for every six-carbon citrate in, a Cit+ cell bleeds succinate. A mutation upstream of dctA switched on the aerobic C4-dicarboxylate transporter, letting the cell take the succinate back.¹³ Quandt and Barrick then backcrossed away everything else and found that citT and dctA alone give the full-strength trait.¹³

Potentiation — what the earlier mutations actually did. Blount's 2008 replay experiments took clones from many points in Ara–3's history and re-ran evolution from each. Across three experiments, 3,212 replicate cultures produced 17 independent Cit+ mutants, and every one came from a clone sampled at generation 20,000 or later.¹ No Cit+ mutant appeared among 8.4 × 10¹² ancestral cells or 9.0 × 10¹² cells from sixty clones through 15,000 generations.¹ The measured mutation rate to Cit+ was 6.6 × 10⁻¹³ per cell-generation in potentiated clones against an upper bound of 3.6 × 10⁻¹³ in the ancestor — "at least 2-fold, and probably much more."¹

That factor of two is thin, and the paper says so. The third and largest replay gave a Monte Carlo P value of 0.0823, which is not significant at the conventional threshold, and one 20,000-generation clone produced two Cit+ mutants when the model said it should produce none.¹ The combined result across three experiments is highly significant; the individual result that mattered most was not.

The Cit+ lineage did not take over. Cit− cells persist at about 1%, and Blount's 2008 explanation was that they are simply better at glucose while the Cit+ cells are better at citrate.¹ Fifteen years later Caroline Turner and colleagues found the mechanism, and it is more interesting: cross-feeding. Every citrate molecule CitT imports is paid for with a four-carbon acid exported, so a Cit+ cell leaks succinate, fumarate and malate into the flask — and the Cit− clade evolved to eat them.⁸ One limiting resource became at least five, shared or divided between two clades. Whether that is the first step toward two species is exactly what Blount et al. asked and did not answer.¹

And then the lineage broke its DNA repair. The Cit+ clones from generation 36,000 onward carry a premature stop codon in mutS, which is why later Cit+ genomes accumulate point mutations far faster than the Cit− clones beside them.¹⁰ The innovation and the hypermutability are independent events, but they happened in the same lineage, and Behe is entitled to point at the pairing.

How many mutations it took depends on who is counting. Four counts are in print, and they answer different questions. The minimum for any aerobic citrate use is one: Leon and Barrick's single-copy rnk-citT module makes the 1988 ancestor weakly Cit+ and 1.0% fitter.¹² The minimum for the full-strength trait is two, the citT promoter capture plus the dctA mutation.¹³ Behe, counting only the citrate module itself, gets "perhaps three to four mutations — the original duplication placing the oxygen-tolerant promoter near the citT gene, plus several rounds of duplication of that region."³⁶ ³⁶ Counting Ara–3's actual history, including the "at least two" potentiating mutations Blount inferred,¹⁰ the BioLogos biologist Dennis Venema arrived, as Behe reports it, at "roughly a half-dozen."³⁶

By Behe's own rule, the citrate pathway sits inside his "edge of evolution." The Edge argument counts only the steps selection cannot see: "Only required deleterious and neutral mutations count against the limit to Darwinian evolution."³⁶ Both kinds of step in Behe's own count have measured benefits: the first duplication is worth 1.0% in a potentiated clone and in the ancestor, and extra copies explain the stronger growth that let the population expand.¹⁰ ¹⁰ Behe argued in 2012 that the result stays inside his edge, and Don Batten of Creation Ministries International said the same in 2016: "As I have said, there is nothing here that is beyond the 'edge of evolution' that Michael Behe described in his book on the topic."³⁷ ³⁷ That cuts both ways. The LTEE's one innovation does not refute the Edge, and it does not confirm it either, because it is not the kind of event the Edge is about. One line of Behe's 2012 essay has aged well. He wrote that, shown a cartoon of the duplication beforehand, "I would have assumed that would be sufficient — that a single step could achieve it," and took the long delay as evidence that "even skeptics like myself overestimate the power of the Darwinian mechanism."³⁶ Six years later Leon and Barrick showed that the single step does pay in the ancestor, and traced the delay to competition from other mutations,¹² the mechanism Behe himself put at the centre of Darwin Devolves.²⁰

What young-earth creationists said about citrate, and how it held up

Young-earth writers answered the 2008 paper within weeks, and their claims can now be scored against the genome data.

The prediction that held. Georgia Purdom of Answers in Genesis wrote in June 2008, before the mechanism was known: "Since E. coli already possess the ability to transport and utilize citrate under certain conditions, it is conceivable that they could adapt and gain the ability to utilize citrate under broader conditions. This does not require the addition of new genetic information or functional systems."³⁸ ³⁸ Blount 2012 found that kind of change: an existing citrate transporter, expressed under new conditions. In 2016 the creationist writer Jay Wile counted Minnich's rapid Cit+ mutants as a confirmed creationist prediction: "If Dr. Purdom is correct, it shouldn't be at all difficult to produce citrate-using Escherichia coli in an oxygen environment."³⁹ ³⁹ Robert Carter's rule that a function made by gene duplication "must necessarily be related to the function of the old" fits as well, since the duplicated gene still moves citrate.⁴⁰ ⁴⁰

Two limits on that credit. The prediction was not distinctive. Blount, Borland and Lenski had written in the same 2008 paper that "a more likely possibility, in our view, is that an existing transporter has been coopted for citrate transport under oxic conditions,"¹ and Purdom quotes that sentence. The same paper cites a 1998 finding that citT alone, on a multicopy plasmid, already lets E. coli grow on citrate in air.¹ And "easy" is Wile's word, not Purdom's; she wrote "conceivable."

The claim that did not hold. Both ministries describe the mechanism as broken regulation. Carter: aerobic citrate use "involves the loss of control of the normal anaerobic citrate digestion."⁴⁰ Purdom in 2012: "There was a definite loss of regulation as the original citT gene was only expressed under low oxygen levels."⁴¹ ⁴¹ Her 2008 article had predicted the same, "loss of regulation and/or specificity."³⁸ Blount's reporter test says otherwise. The native citT control region "showed no expression (above background) in any strain," and the strains included a weakly Cit+ clone as well as the ancestor, so the oxygen switch still works in a Cit+ cell. What changed is that a copy of the gene picked up a different switch, the rnk promoter.¹⁰ Don Batten, whose 2008 article Carter cites for the claim, conceded the point in a 2016 addendum, "the existing controls and the citrate transporter gene were not actually broken," and then concluded in the same paragraph that "the ability of the cell to control the production of the citrate transporter was indeed broken."³⁷ That is a relabelling, not a finding. The accurate creationist description is Behe's: modification of function, "because no new functional coded element was gained or lost, just copied."²⁰ It concedes what the experiment shows and keeps the point that matters to the design argument, that no new gene was written.

The argument that lost its premise. Purdom's 2012 reaction to the genome paper was amazement "at just how many changes were necessary to allow the bacteria to perform a pre-existing function."⁴¹ Two years later Barrick's lab showed that two changes give the full trait, and four years after that, that one gives a weak version in the ancestor. The creationist reading that survives is the one Purdom's own 2008 article, Behe and Minnich share: the trait is a rearrangement that put an existing transporter under a new switch, it arises quickly under direct selection, and it was slow in the LTEE for reasons of ecology and competition.

The rest of the ledger

Beyond citrate, the gains are quantitative. Two genes, pykF and nadR, acquired substitutions in all twelve populations, which is what a shared selective target looks like.⁴² The rest is faster growth on glucose, shorter lag on transfer into fresh medium, larger cells, altered DNA supercoiling in ten of twelve populations, and adaptation to acetate, the waste product the cells excrete and then re-consume.¹ ²² Nine of the twelve populations spontaneously split into clades that have coexisted for more than 10,000 generations, most likely by cross-feeding, so ecological diversity has increased even inside a single flask.³⁴

One more gain deserves its own paragraph, because it is the nearest thing in the LTEE to the birth of a new gene. Deyasi uz-Zaman and colleagues mined the LTEE's RNA-sequencing and ribosome-profiling data for proto-genes — stretches of previously silent, non-coding DNA that have started being transcribed.²⁷ They found nine confirmed cases, and five of the nine were created by an IS150 element landing upstream and pointing its own promoter at the sequence.²⁷ One of the nine shows ribosome-profiling evidence of being translated, into a 117-amino-acid open reading frame.²⁷ Their estimated rate is about one proto-gene per 60,000 generations, with the earliest appearing by 4,000.²⁷

Two things are true of that result at once. It is new transcription arising from DNA that encoded nothing, which is more than the usual ledger allows. And the authors are candid that none of the nine has been shown to do anything: "Future experimental studies will be required to establish whether the novel transcripts and peptides we detected affect E. coli fitness or are byproducts of mutations that are beneficial because of their effects on nearby, existing genes."²⁷ What has not been reported in the LTEE, still, is a new protein fold or a new enzymatic chemistry.

What has been lost

Ribose, in all twelve populations, in under 2,000 generations

Every one of the twelve lines deleted part or all of the ribose operon early. The deletions differ in extent but all involve an IS150 element sitting immediately upstream of rbs: the element transposes into the operon, then homologous recombination between the two copies removes everything between.¹⁴ Three things make this the cleanest loss-of-function case in the literature:

The mutation rate from Rbs⁺ to Rbs⁻ is about 5 × 10⁻⁵ per cell generation — roughly a hundred thousand times a typical point-mutation rate, because it is a repeat-mediated deletion rather than a base change.¹⁴
The deletion is worth 1 to 2% in fitness, confirmed by transducing the functional operon back in and by building the deletion independently without IS150.¹⁴
It happened in 12 of 12 populations, which is what selection looks like.

Deleting a working gene made these cells measurably better at their job. That is Behe's rule, demonstrated cleanly, and no one disputes it.

Catabolic breadth

Cooper and Lenski tested the evolved lines against 64 informative carbon sources. Nine showed parallel decay by 2,000 generations and sixteen by 10,000. Averaged over all 64, total catabolic function fell from 1.00 to 0.68 by 10,000 generations and to 0.58 by 20,000.¹⁵ Most of the decline happened early, in parallel across populations, and before any population had become hypermutable — the signature of selection, not of mutations quietly piling up in unused genes.¹⁵

Marx's group later argued the opposite for a broader set of substrates: that metabolic erosion in E. coli runs mainly through mutation accumulation rather than trade-offs.⁴³ Both can be true of different substrates, and the disagreement is worth knowing about; it does not change the fact that diet breadth fell by 42%.

DNA repair

Three populations were mutating between ten and a hundred times faster than the ancestor by 10,000 generations, through defects in mutS, uvrD or mutL.⁴⁴ By 50,000 generations six of twelve were point-mutation hypermutators, and those six carried 96.5% of all the point mutations in the experiment.¹⁶ A seventh population, Ara+1, became a different kind of mutator: 31.8% of all its mutations are IS150 insertions.¹⁶

Hypermutability then partly reversed. Populations that were hypermutable at 10,000 generations accumulated synonymous substitutions far more slowly between generations 40,000 and 50,000 than between 10,000 and 20,000, because the deleterious load favoured antimutator alleles.¹⁶ Population Ara–1 is the case that has been worked out. A frameshift broke mutT, whose product sanitises oxidised guanine nucleotides, and the point-mutation rate rose about 150-fold.⁴⁵ Two independent lineages later acquired mutations in mutY, "which encodes a DNA repair glycosylase that excises mis-paired bases from DNA helices," and the rate fell back by 40 to 60%.⁴⁵ Behe describes this as damage partly offsetting damage, and on the mechanism he is right — both alleles are broken genes.²⁰ What he does not report, and what this article's earlier draft repeated after him, is that Wielgoss and colleagues hedge their own adaptive reading: "it is also possible that these mutY mutations were not adaptive."⁴⁵

What happened inside those mutator genomes is the strongest piece of pro-decay evidence in the whole corpus, and it is not Behe's — it is Lenski's, with Tenaillon and Schneider. Couce and colleagues asked whether selection or mutational bias better predicts which sites in a mutator genome actually changed. Their title is the answer: "Mutator genomes decay, despite sustained fitness gains, in a long-term experiment with bacteria."⁴⁶ Selection is not absent, but it is weaker than the bias — and that holds even when the analysis is restricted to genes that are essential under LTEE conditions.⁴⁶ Their explanation is genomic draft: in an asexual population, a strongly beneficial mutation sweeping to fixation drags every passenger in its genome along with it, so linkage rather than population size is what limits selection's ability to purge damage.⁴⁶ This is a mainstream demonstration that fitness gain and genome decay can run at the same time in the same cells, and any honest account of the LTEE has to carry it.

Genome size

Across 264 sequenced genomes, Tenaillon and colleagues catalogued 14,572 point mutations, 500 IS-element insertions, 726 small deletions, 1,132 small insertions, 267 large deletions and 45 large duplications. Average genome length fell by 63 kilobases — about 1.4% — over 50,000 generations.¹⁶ ¹⁶ Insertions and deletions make up nearly half the mutations in non-mutator clones at 50,000 generations.¹⁶

The genome is smaller, its diet is narrower, its repair systems are worse, and it grows 70% faster. Any account of the LTEE that leaves out either half of that sentence is not describing the experiment.

How strong is the evidence that the gains are degradation?

This is where the argument is live. Behe's claim in Darwin Devolves is that "it's very likely that all of the identified beneficial mutations worked by degrading or outright breaking the respective ancestor genes."²⁰ His grounds are that Tenaillon's two tables of most-mutated genes show many different mutations per gene, which is what knockout looks like.

The counter-evidence is in the paper Behe cites in his own endnote. Maddamsetti and colleagues examined the 57 genes with two or more non-synonymous mutations across the non-mutator lines — genes holding 50.1% of all non-synonymous mutations in 2.1% of the coding genome.¹⁶ They found:²¹

Sixteen of the 57 are essential genes. Only topA and mrdA carry any candidate knockout, and the topA one is a frameshift in the penultimate codon that adds three residues — almost certainly not a knockout.²¹
Only five of the 57 have nonsense mutations anywhere in the non-mutator genomes, and all five are dispensable genes.²¹
Ten of the 57 show the same amino-acid substitution recurring across populations, several mapping to functional interfaces: infC R132C and rpsD D50G at the contact surface between those two proteins, nadR Y294C on the helix face that binds NAD, pykF A301S at the multimerisation interface implicated in allostery.²¹

Repeated hits on one specific residue at a binding interface is not what breaking a gene looks like; breaking a gene produces scattered nonsense codons and frameshifts, which is what these same populations show in nadR and pykF in other lineages. Behe's response is that fine-tuning is "probably" degradation.²⁰ Lenski names the assumption behind it: it only works if the ancestral protein was already optimal, which is exactly what a new environment makes false.¹⁸ The Science review of Darwin Devolves, co-written by Lenski, makes the complaint in one sentence: Behe emphasises "the many mutations that arose that degraded function — an expected mode of adaptation to a simple laboratory environment, by the way — while dismissing improved functions and deriding one new one as a 'sideshow'."⁴⁷

The number Behe's claim needed, and what it leaves uncounted

Until 2024 the fairest thing to say was that nobody had counted. Enzyme activity had been assayed for citrate synthase and for essentially nothing else. Then Alejandro Couce and colleagues, with Lenski and Tenaillon among the authors, published a genome-scale answer in Science.¹⁹ They built transposon-insertion libraries in the LTEE ancestor and in evolved clones and measured the fitness effect of hundreds of thousands of insertions in every gene, in bulk competition.¹⁹ Then they classified the LTEE's own adaptive mutations by type, treating nonsense codons, frameshifts, deletions and insertions as presumptive knockouts. The result:¹⁹

"these presumptive inactivating mutations contribute most (>50%) of the early adaptive mutations in the LTEE, and they continue to be a sizable fraction over the long run (~25%...)."

That is the number the whole dispute has been missing, and it is a count of one kind of mutation, not of every kind. More than half of the early adaptive mutations look like breakage — a substantial concession to Behe's rule, made by Lenski's own co-authors. About three-quarters of the long-run ones are not of a breaking type, which is not the same as saying they do not degrade. Most of that three-quarters are single amino-acid changes, and Couce and colleagues did not classify them; their own Discussion says only that point mutations and rearrangements "may be more likely to generate gains or changes of function."¹⁹ So the count shows the kind of adaptive mutation shifting, from cheap knockouts early to finer changes later. It does not contradict Behe's "very likely that all of the identified beneficial mutations worked by degrading or outright breaking the respective ancestor genes,"²⁰ because the missense changes his claim turns on are exactly the ones nobody assayed. The evidence against "all" is the structural pattern in the previous section and Behe's own classification of the citrate duplication as modification-of-function, not this count.

The caveat runs in Behe's favour: "presumptive inactivating" is an inference from mutation class, not an enzyme assay. A frameshift almost certainly breaks a gene; a missense change at a binding interface, counted here as not inactivating, may still reduce activity. The 25% figure is a floor for degradation, not a ceiling.

The one case where the enzymology was measured, in full

By this article's own standard — that almost nothing in the LTEE has had its enzyme activity assayed — the citrate-synthase story is the decisive test case, and it should be reported whole rather than in the half that suits either side. Behe reports it whole in Darwin Devolves, and everything he says about it checks out against Quandt's paper.²⁰ The trajectory:

gltA1 knocks out the NADH allosteric brake on citrate synthase, raising flux. A destroyed regulatory site producing more enzyme activity — the case is genuinely ambiguous, and it is the only step in the chain that is.²²
iclR and arcB are, in Quandt's own words, "loss-of-function mutations that derepress expression of enzymes needed for acetate assimilation." Mutations in arcB or arcA are found in eleven of the twelve LTEE populations.²²
Then the trait arrives, and the metabolic optimum inverts. Flux-balance analysis says that "when citrate is the sole carbon source ... any CS activity is detrimental."²²
So a second wave of gltA mutations cuts the enzyme back down, one of them by reducing transcription roughly tenfold, ending below the ancestral level. Quandt's own section heading is "Most secondary citrate synthase mutations reduce enzyme activity."²²

Three loss-of-regulation events on the way up and a deliberate throttling on the way down is not a draw, and calling it one would be the kind of even-handedness that costs a reader's trust. What it is not is a demonstration that the cell got worse. Every one of those steps was selected because it made the cell better at the job in front of it, and the endpoint is an organism doing something no ancestor could do. Behe's ledger of the mechanism is right. His inference from it, that this is all evolution can ever do, goes further than the LTEE can test in either direction.

The Discovery Institute's stepwise-citrate claim

What Minnich actually did

Van Hofwegen, Hovde and Minnich reasoned that if aerobic citrate use were rewarded directly, rather than as a side-benefit in a glucose-limited flask, it would appear fast. They ran three protocols on wild-type E. coli K-12 and B, on REL606 itself, and on knockout derivatives:²

Direct selection — citrate as the only carbon source, incubated for weeks with no transfers.
Modified direct selection — citrate plus a trace of glucose or glycerol (enough for about six generations), then seven days in stationary phase, then a 1:100 transfer, weekly.
Low-citrate variants to test concentration-dependence.

They recovered 46 independent Cit+ mutants. None arose in the ΔcitT strain. Genome sequencing showed the same class of change as the LTEE: strain DV130 had citT amplified four-fold and was weakly Cit+; its descendant DV133 gained an IS30 transposase inserted upstream of citT — a promoter capture — and became strongly Cit+; strain DV159 fused citT to the promoter of the stress gene uspG, which is what the LTEE mutant did with rnk.²

Then they did the experiment Minnich himself puts first, and this article's earlier structure buried it. They made a P1 phage lysate on DV133 and used it to move the Cit+ determinant into wild-type K-12 that had never seen citrate selection. Four days later the transduced culture was turbid; streaked on Simmons citrate agar it threw blue colonies in two days.² Minnich's own summary of what that proves: "using phage transduction, we show that the Cit+ mutation can be transferred into wild-type E. coli K12 without prior citrate exposure, genetic proof that the aerobic Cit+ phenotypes do not require contingent single nucleotide polymorphism accumulation."²⁹

That is a real result and it is the cleanest thing in the paper. Its limit is equally real: the recipient is K-12, not REL606, so it shows the determinant is background-independent in K-12. Leon and Barrick's anti-potentiation measurement is the direct counterpoint, and it is measured in Ara–3's own history. Both are true; neither settles the other.

The experiments are good and Lenski said so: "the actual science that was done and reported looks fine and interesting."³ What he disputes is the packaging.

What the 2026 article claims, and how it holds up

The context matters, and this article's earlier draft left it out. McLatchie is answering the YouTuber Dave Farina, who had argued that the citrate trait "qualifies as irreducibly complex according to Behe's own definition," because "all of these changes were required, and none of them can have been selected for individually. In fact, the critical duplication is actually deleterious on its own without the other mutations."¹¹ On that specific question McLatchie is right and Farina is wrong, and the number that settles it is Leon and Barrick's: the citT-activating module was worth +1.0% in the unevolved ancestor, so the critical duplication is not deleterious on its own.¹² An article that grades McLatchie's claim without saying he won the argument he was actually having would be scoring points rather than reporting evidence.

McLatchie states the claim in four sentences:¹¹

"While multiple changes are required for the efficient Cit+ trait, biologist Scott Minnich and colleagues showed that the process is stepwise, conferring a selective advantage at each stage. An initial weak benefit is realized by amplification of citT via gene duplication. This allows limited citrate uptake during aerobic conditions. Aerobic expression of citT is then strongly activated by promoter capture, yielding a functional Cit+ phenotype. Subsequent refinement of the dctA gene yields a further advantage by bringing succinate back into the cell that was leaking out."

Point by point:

"Stepwise" — correct, and not contested. Everyone involved describes the same three-stage path. The disagreement was never about whether there were steps.

"An initial weak benefit... by amplification of citT" — demonstrated in Minnich's environment, unmeasured in the LTEE's. Minnich's DV130 carried four citT copies and no promoter capture and was weakly Cit+, so amplification alone did something under citrate-only starvation.² Roth and Maisnier-Patin supplied a mechanism: dense stationary-phase cultures go locally anoxic, so the normally silent cit operon may be slightly expressed, and more copies then mean more transport.²⁸ The nearest measurement pointing the other way is Blount's luciferase reporter, which found no expression above background from the native citT control region.¹⁰ That is weaker evidence than it looks. The reporter sat on a plasmid held at one to two copies per cell, so what it establishes is that native citT expression falls below the detection floor of a single-copy assay.¹⁰ It does not establish zero, and it does not measure what four or nine copies would do in a flask that spends most of every day in dense stationary phase.

What of the order of operations? In the LTEE the sequenced Cit+ genome shows a duplication that was the promoter capture, in one event;¹⁰ in Minnich's cultures DV130 amplified first and DV133 captured a promoter afterwards. That looks like a difference and may not be one. Roth and Maisnier-Patin's model predicts exactly this pattern: a transient amplification is unstable, so it "may explain why the posited simple duplications were not apparent in sequences of cells from early tubes from the LTEE — they may have been lost during nonselective growth prior to sequencing."²⁸ Genome data taken after the fact cannot distinguish "there was never an amplification" from "the amplification did not survive to be sequenced." Roth and Maisnier-Patin themselves judge the two experiments to run "essentially the same series of events."²⁸ The question is open, and the LTEE's sequences do not close it in either direction.

"Conferring a selective advantage at each stage" — correct as written, and the article's earlier draft rebutted a claim McLatchie did not make. His stages are mutational steps, not points in Ara–3's history, and this article concedes all three: DV130 shows amplification alone gives a weak phenotype under direct selection; Leon measured the promoter capture at +1.0% in the unevolved ancestor; the dctA refinement is Barrick's own result. Two regimes are in play, though, and the whole dispute turns on not mixing them.

In a citrate-only culture the claim is not in doubt, and Leon and Barrick say so about Minnich's system by name:¹²

"this difference in environmental conditions relative to the glucose-limited transfer regime of the LTEE, in which cells are in stationary phase for only ~16–18 hours each day, dramatically increases the fitness benefit of evolving the rudimentary Cit+ phenotype. Therefore, these stark conditions are expected to completely mask and overwhelm the dependency on potentiating genetic and population factors found in the LTEE. Activating citT expression would be universally beneficial in any genetic background in these types of experiments. Any increase in lag phase or other trade-off with respect to growth rate that might accompany this intermediate step in the pathway to Cit++ is irrelevant when cells without the mutation simply cannot replicate at all."

That last sentence matters more than it looks. The penalty that makes the module deleterious in the LTEE is a lag-phase penalty — Leon and Barrick measured it directly¹² — and a lag phase costs nothing when the alternative is not growing at all. So McLatchie's claim survives in the environment it was made about, and the people who supplied the measurement are the ones who said so.

Inside the LTEE the same module behaves differently. Leon and Barrick built it into twenty-five backgrounds spanning Ara–3's history and measured competitive fitness against isogenic Cit− partners. Reading down the lineage in time order:¹²

Background	Fitness effect of one chromosomal Prnk-citT module
REL606, the 1988 ancestor, measured on its own	+1.0%
Earliest cluster: REL606 with ZDB409 and ZDB429	+1.7% [0.0%, +3.3%]
Anti-potentiated cluster, 10,000 to 20,000 generations	−5.4% [−7.2%, −3.6%]
25,000 to 27,000 generations	+0.4% [−1.3%, +2.0%]
Fully potentiated, 29,000 generations onward	+2.4% [+1.4%, +3.4%]
ZDB706, the potentiated Cit− revertant	+2.4%
ZDB564, carrying the authentic evolved duplication	+2.2%
ZDB564 again, competed inside the whole 31,000-generation population	+0.9%

Quandt's earlier experiment shows the same thing from the other direction:²² revert the gltA1 citrate-synthase mutation out of the first Cit+ clone and the citT duplication becomes "highly deleterious," producing a long growth lag that "would have almost certainly led to the rapid extinction of any newly evolved Cit+ cells."²²

Three cautions before that table is used as a rebuttal of anything. Leon and Barrick found five other strains where activating citT cost more than 20% of fitness, but they place those clones off the line of descent, attributing the effect to "private" mutations not shared with the pre-Cit+ lineage; the honest number for the dip on the path is −5.4%, not −20%.¹² They add their own hedge: they "cannot rule out that other changes in the impact of citT activation may have occurred on the main line of descent within this interval."¹² And the whole table is measured under LTEE conditions, which the same authors say do not apply to Minnich's.

So the claim splits cleanly, and mostly in the Discovery Institute's favour. No neutral waiting period was required — correct, and Leon's +1.0% in the ancestor establishes it. Each of the three mutational stages carries an advantage under direct selection — correct, and Leon says so explicitly about that class of experiment. What the claim does not address, because it was never asked to, is why the LTEE was slow. There the answer is that the same second step swung to −5.4% for a stretch of the line of descent lasting something like ten thousand generations. That is not a rebuttal of McLatchie. It is the most interesting number in the dispute, and it belongs to a different question.

"Refinement of the dctA gene yields a further advantage" — correct, and this is Barrick's result.¹³

"No new genetic information (novel gene function) evolved" — a definition, not a finding. Minnich's parenthesis equates information with novel gene function, and no gene acquired a new catalytic function, so the conclusion follows from the premise. Lenski's counter is that a genome encodes when to express what, and the cell now expresses a transporter under conditions no ancestor could.³ Neither side has an operational measure of biological information that the other accepts; the exchange is therefore not resolvable as stated, and the article recommends dropping it in favour of what can be measured.

The generations problem

The single most repeated figure from Minnich's work is "as few as 12 generations." Its context is worth quoting in full:²

"After inoculation (9 × 10⁵ to 7 × 10⁶ CFU/ml), the turbidity of the cultures rose to an OD₆₀₀ of 0.02 to 0.03 overnight, or the equivalent of approximately six generations based on plate count. These cultures were maintained for 7 days in stationary phase, after which cells were inoculated into fresh media at a 1:100 dilution... All but one E. coli Cit+ strain were recovered within two to five weekly transfers, representing 12 to 30 generations."

Twelve generations therefore means two weekly cycles: six generations of growth, seven days of starvation, six more generations, seven more days. In the direct-selection protocol the cultures sat for up to 134 days in stationary phase, and generations were not counted at all.² Lenski's reading of Table 1 is that the fastest Cit+ appeared at 19 days and most took a month or two.³

Nothing here is hidden. The paper reports the protocol, and it states a range — "12 to 100 generations" — not a floor.² Minnich and Hovde give the range and the mechanism in the same breath in 2025.²⁹ It is McLatchie's summary that compresses it to "as few as 12 generations," and even he offers it as a claim about what potentiation requires rather than about elapsed time.¹¹

There is also a defence of the generation unit that deserves stating, because the article's first draft skipped it. The claim Minnich is answering is that Cit+ needed a long series of prior mutations accumulated through replication. Mutations accumulate per cell division, so generations are the right unit for that claim, and days are not. Where days matter is a different question — how fast a laboratory can produce Cit+ — and on that question twelve LTEE generations is a day and three quarters, while Minnich's twelve took two weeks. Both units are correct for their own question and neither is correct for both.

Stated in days, the contrast is that a directly selected culture produced Cit+ in two to six weeks while a glucose-limited serial-transfer culture with a daily 1:100 dilution took fifteen years. That is still striking, and it is the contrast Minnich's own discussion draws: amplifications carry a fitness cost, are lost during the non-selective glucose phase, and rarely survive the daily bottleneck long enough for the rarer promoter capture to occur.²

The probability arithmetic, checked

In 2025 Minnich and Hovde restated the case as a numbers argument in The Journal of Physiology:²⁹

"the combined probability of getting both aerobic CitT expression and the dctA point mutation suppressing the need of dcuS is ∼1 × 10⁻¹⁴. This figure is derived from: (i) gene amplifications are the most common mutations in bacteria... and we estimate that getting aerobic expression of CiT via gene amplification/recombination in the order of ∼1 × 10⁻⁵; and (ii) a point mutation to suppress dcuS⁻ is in the range of 1 × 10⁻⁹ to 10⁻¹⁰. The fact that the LTEE Cit+ mutation was acquired after 33,000 generations (∼1 × 10¹⁴ cells...) likewise argues against contingency."

Their ~10¹⁴ is not an independent count — it is Elizabeth Pennisi's figure from a 2013 Science news feature — but it is the right order of magnitude. Recomputed from Lenski's own design parameters, twelve populations had produced 3.0 × 10¹³ cells by generation 33,127, and 7.3 × 10¹³ by 82,000.⁵ Against 3.0 × 10¹³ cells, a joint probability of 10⁻¹⁴ predicts 0.30 events; against Minnich's rounder 10¹⁴ it predicts one. Either way the order of magnitude fits the single observed event, and the argument is not frivolous.

Three problems with it, in increasing order of seriousness.

First, the 10⁻⁵ figure is for amplification, not for aerobic expression. The rate at which a given locus duplicates is indeed around 10⁻⁵ to 10⁻⁴; the rate at which a cell acquires aerobically expressed CitT is what Blount measured directly, and it is 3.6 × 10⁻¹³ or lower in the ancestor.¹ Minnich's step (i) is off by seven or eight orders of magnitude from the only direct measurement of it, and the coincidence between his 10⁻¹⁴ and the observed timing depends on that number.

Second, the only rate anybody has measured was measured in the wrong regime, and that cuts both ways. Blount's figure comes from harvesting cultures onto minimal-citrate agar and incubating them for 45 days — sustained direct selection, the regime Minnich says produces Cit+ readily. Blount et al. say their own number is "probably far too high," and their jackpot analysis shows the second of the two required mutations happened on the plate, over those weeks.¹ A rate whose second step occurs during six weeks of selection is not a per-generation rate that can be multiplied by LTEE cell-generations. Do it anyway and twelve populations over 31,500 generations expect about 19 independent Cit+ originations, or 10 at the ancestral upper bound; one established.⁴⁸ That arithmetic is worth seeing, and it points at Leon and Barrick's conclusion that the barrier was survival against competitors rather than mutational access.¹² But it is the same category error this article charges Minnich with, and the honest verdict is symmetrical: nobody has measured the rate that matters — the per-generation rate of citT activation in a shaken flask under a daily 1:100 dilution — and both "it should have happened many times" and "it happened exactly on schedule" depend on it.

Third, the second term is priced as a point mutation and it is not one. Minnich's step (ii) puts the dcuS workaround at 10⁻⁹ to 10⁻¹⁰, "a point mutation."²⁹ But the ancestral defect is a five-base-pair insertion that leaves three tandem copies of the same five bases, so deleting one copy restores the reading frame.⁸ Slipped-strand events in short tandem arrays run orders of magnitude faster than base substitutions; the rbs deletion this article has already quoted, also repeat-mediated, runs at 5 × 10⁻⁵.¹⁴ And this is not a hypothetical. The LTEE performed that repair. Turner and colleagues found the five-base deletion in the Cit− clade, reconstructed it in the ancestor's own genome, and showed the reconstructed strain grows on succinate, fumarate and malate where REL606 cannot.⁸

Fourth, a note on an apparent contradiction that is not one. Multiplying two probabilities to get 10⁻¹⁴ assumes the intermediate is not selected, and Van Hofwegen et al. say so in terms — "under nonselective conditions."² The 2026 stepwise claim says the intermediate is selected, at every stage.¹¹ Read side by side those look incompatible. They are not. Van Hofwegen et al. supply the reconciliation on the same page: an amplified citT gene has "only a brief advantage in the use of citrate once glucose was expended" and is "likely lost with each daily dilution/transfer into fresh glucose-containing media due to the fitness cost of the maintenance of multiple gene copies."² The step is selectable for a few hours a day and counter-selected for the rest, so under the LTEE's regime it does not persist and the two events must effectively co-occur; under direct selection it does persist and they need not. One mechanism, two protocols. What is missing is not consistency but the qualifier — McLatchie's four sentences drop the regime dependence that makes Minnich's two claims hang together, and a reader meeting the claim there will not know it is there.

Minnich's strongest point: the wrong strain

The strain argument is usually dated to the 2025 letter. It is nine years older, and it was published with a prediction and three supporting results.² It is also not about probability at all. E. coli B REL606 carries a five-base-pair frameshift in dcuS, the sensor kinase that switches on the succinate transporter DctA.² ⁴⁹ K-12 strains do not. So a Cit+ K-12 cell recaptures its leaked succinate automatically, while a Cit+ REL606 cell needs an extra mutation to override the broken sensor — which is exactly the dctA promoter mutation Quandt identified.¹³ ²⁹ Minnich's conclusion: "Had he chosen E. coli K-12, rather than E. coli B (Lenski's starting strain which lacks a functional key gene in its genetic deck), it is likely that all 12 flasks would rapidly have become Cit+."²⁹

The premise is correct and it is sourced to Lenski's own collaborators. The conclusion is weaker than it looks, for a reason published two years before the letter and not cited in it. The LTEE fixed the broken gene, twice, by two different routes. In the Cit+ clade, the dctA promoter mutation Quandt identified made DctA constitutive and bypassed the sensor.¹³ In the Cit− clade, a five-base deletion restored dcuS itself to frame, and Turner and colleagues proved it by rebuilding that allele in the ancestor's genome and watching the reconstructed strain grow on succinate — which REL606 cannot.⁸ A defect that a population repairs spontaneously within its own history, by two independent mechanisms, is not the kind of defect that explains a fifteen-year delay.

What survives of the argument is smaller and still worth having. REL606 does need one more mutation than K-12 would, and Van Hofwegen et al. tested that within their own paper: a K-12 strain artificially crippled at the same point in the pathway — ΔdctA, phenotypically equivalent to REL606's broken dcuS — took 134 days to go Cit+, the slowest single result they report, while intact K-12 needed only a dctA duplication.² What remains untested is the stronger counterfactual, that twelve K-12 populations run for fifteen years would all have gone Cit+. And notice what the argument concedes even if it wins: if the reason the LTEE was slow is that its founder carried a broken gene, then the LTEE's citrate result is a measurement of one strain's peculiar starting point — and so is every creationist argument built on the rarity of that result.

Common objections

"Thirty-eight years and all they got was a bacterium that eats a different sugar. That is not evolution." It is evolution; it is not much of it. The honest complaint is about rate, and the rate is set by the environment. A flask with one food source, one temperature, no predators and no competitors offers almost nothing to adapt to. Lenski designed it that way on purpose and says the experiment "was not intended to mimic the complexities of nature, nor was it meant to be a test-bed for the evolution of new functions."¹⁸ Arguing from the LTEE's meagre novelty to a limit on evolution generally is arguing from a rigged-easy case.

"It is still E. coli." True, and also true of every 82,000-generation lineage of anything. Bacterial species are defined by clusters of traits rather than reproductive isolation, and inability to grow on citrate aerobically is one of the traits used to define this one — so the Cit+ line has crossed a diagnostic boundary while remaining, by descent, the same organism. Lenski's position is that speciation is a process, not an event, and that the coexisting Cit+ and Cit− lineages are at the beginning of one.³ Whether they finish is an empirical question nobody can answer yet.

"Almost every mutation is a loss. Evolution can only break things." The first clause is close to true in this experiment. The usual answer is Lenski's: in an environment where almost every ancestral capability is useless, discarding capabilities is the cheapest available improvement, and and — the analogy is Lenski's — "just as loss of sight in cave-dwelling organisms is a special case that won't tell us how eyes evolved, one must be careful when extrapolating from this experiment to evolution writ large."¹⁸ That answer is correct and incomplete, because Behe anticipated it and built a sharper argument on the very arithmetic this article opened with.

Behe's argument is about competition, not scarcity, and Leon and Barrick confirmed it. His claim in Darwin Devolves is that "it's not so much the rarity of constructive mutations that undermines Darwinian evolution — it's the frequency of damaging but helpful ones," which "appear quickly even on short time scales, even in small populations" and "compete fiercely with any gain-of-FCT mutations that might eventually arrive on the scene."²⁰ That is exactly what Leon, D'Alton, Quandt and Barrick measured in Lenski's own system. Their abstract: the evolution of citrate utilization "was blocked at this point due to competition with other mutations that improved fitness in the original niche," so "intense competition and the types of mutations that it favors can lead to short-sighted evolutionary trajectories that hide a stepping stone needed to access a key innovation."¹² The mutations doing the outcompeting are the ones this article has already catalogued as degradative — the rbs deletion, spoT, malT, pykF, nadR. Behe named the mechanism; Barrick's lab measured it; this article cites that paper repeatedly and its earlier draft never said whose prediction it matched.

Nor does Behe rest the general claim on the LTEE, as an earlier draft of this article asserted. He rests it on a meta-analysis: Hottes and colleagues pooled genome-wide fitness data across 144 growth conditions and found "at least one beneficial null mutation was identified in all but five of the 144 conditions considered."⁵⁰ Whether that generalises beyond bacteria in defined media is arguable. That it is a general argument, and not one borrowed from a single flask, is not.

Behe's saturation argument, stated properly. If a population has made every possible single-nucleotide change 180 times over, then selection has had every chance to find any single-step constructive mutation worth having. Behe: "one would expect to have observed simple gain-of-FCT adaptive mutations that had sufficient selective value to outcompete more numerous loss-of-FCT or modification-of-function mutations in most experimental evolutionary studies, if they had indeed been available."¹⁷ He also pre-empts the two stock replies: that the laboratory is artificial (he points to Yersinia pestis gene loss and the anti-malarial mutations in wild human populations), and that the timescales are short (his point is not that complex gains should appear quickly but that whatever arrives first is what selection sees first, in the wild as in a flask).¹⁷

The reply that actually works. Behe's saturation figures count point mutations, and the one innovation the LTEE produced is not a point mutation. It is a 2,933-base tandem duplication that had to land at one particular junction to fuse rnk to citT.¹⁰ The space of possible structural rearrangements in a 4.6-million-base genome is vastly larger than the 13.9 million possible point mutations, and the per-genome rate of any specific one is far lower — which is why Blount could measure the rate to Cit+ at 10⁻¹³ while the point-mutation rate is 10⁻¹⁰.¹ Saturating the point-mutation space 180-fold says nothing about whether the rearrangement space has been sampled even once at most of its addresses. So the honest form of Behe's argument is narrower than he states it: single-nucleotide constructive mutations really do appear to be scarce in the accessible neighbourhood, and the LTEE's one innovation came from the class of mutation his arithmetic does not cover.

"The citrate mutant only switched on a gene it already had." Correct, and the same is true of most regulatory evolution in every organism. The question this raises is a good one and it is not about citrate: where did citT come from? Rewiring an existing transporter demonstrates that rewiring is easy. It demonstrates nothing about the origin of transporters. The LTEE was never going to answer that, and neither side should pretend it did.

"The Cit+ population is a sick puppy in a death spiral." Behe's 2020 article draws on real data.⁵¹ Blount and colleagues stained stationary-phase cultures and scored 10.7% of ancestral cells as dead, against 29.6% for the 33,000-generation Cit+ clone and 39.9% for the 50,000-generation one in the same medium, rising to 44.2% in citrate-only medium.⁵² ⁵² Three qualifications. The authors say mortality "varied considerably even between replicate assays for reasons that we do not understand."⁵² The 10.7% baseline shows substantial death is normal for this organism in stationary phase, so the change is a threefold rise rather than a switch from health to sickness. And the same population's fitness, measured in competition, never fell. Behe's word "devolution" describes the gene tally accurately and the fitness trajectory not at all.

"Blount's potentiation result was statistically weak, so the whole contingency story was wrong." Half right. The third replay experiment gave P = 0.0823, and a 20,000-generation clone produced two Cit+ mutants against the model's expectation of zero.¹ The two-fold potentiation of the mutation rate was, in the paper's own words, the most that could be shown.¹ But the story was not simply abandoned; it was replaced by a better-supported one. Quandt showed the potentiating alleles are not required for the trait;¹³ Leon showed the constraint was competition plus a temporary fitness valley.¹² Contingency in the strict sense — the outcome depended on the population's particular history — survives that revision intact, because the fitness valley is itself a historical accident of which mutations that lineage happened to fix.

"Fitness rising for 60,000 generations refutes genetic entropy." Not as its author states it. John Sanford's Genetic Entropy discusses the LTEE by name and expects the rise: "This is not unexpected or remarkable in any way."²⁴ He makes two other claims about it. One is that "in every case" the adaptation came from "loss-of-function or loss-of-regulation mutations," and that the strains will keep shedding unused functions "until the bacterial genome is stripped down to its minimal functionality."²⁴ The ribose deletions, the 42% fall in catabolic breadth and the 63-kilobase shrinkage support the direction of that claim; "every case" is the same universal Behe makes, and it meets the same evidence. The other is a forecast: "Over hundreds of years, Lenski's 'evolving' strains of bacteria must logically go extinct," as nearly-neutral harmful mutations make the genome "rust out."²⁴ Thirty-eight years cannot test that directly, but the arithmetic behind it can be run at the LTEE's numbers, and by drift alone it gives almost nothing. With an effective population size near 10⁷³⁴ and 4 × 10⁻⁴ mutations per genome per generation,⁶ a non-mutator genome whose every new mutation sat just inside the band selection cannot see would lose about 0.01% of its fitness per thousand years, and a 150-fold mutator about 1.5%.²⁶ The forecast's real support is elsewhere, in the mutator lines, where linkage lets harmful mutations ride along with beneficial ones ("genomic draft," next paragraph). Those genomes are decaying, and their measured fitness still rose through 60,000 generations.²³ So the fair conclusion is narrow: genetic entropy is not a universal property of replication, the LTEE refutes no forecast Sanford actually made, and his extinction forecast does not follow from drift at these population sizes. Whether it operates in small sexual populations, where the arithmetic is very different, is treated in the genetic entropy article.

But rising fitness does not mean the genome is fine, and Lenski's own group has shown it. Couce et al. put the two facts in one title — mutator genomes decayed despite sustained fitness gains — and traced it to genomic draft: strong linkage in an asexual population means a beneficial sweep carries its passengers to fixation whether they help or not.⁴⁶ ⁴⁶ The creationist argument that survives this experiment is therefore not "fitness must fall." It is the narrower and better-evidenced claim that selection cannot see most of what mutation does to a genome, even in a population of ten million with a 1% advantage on offer — and that claim has a PNAS paper behind it with Lenski's name on it.

"Six of twelve populations broke their DNA repair. Surely that is decay." It is loss of function that was favoured by selection, because a higher mutation rate meant more beneficial mutations while there were still easy ones to find.⁴⁴ It also partly reversed: the same populations later acquired antimutator alleles as the deleterious load rose.¹⁶ Whether that reversal was selected is not settled — Wielgoss and colleagues, who measured it, say "it is also possible that these mutY mutations were not adaptive."⁴⁵ That is a more interesting result than either "decay" or "improvement" and it belongs to neither side's slogan.

"You credit the Discovery Institute with a measurement Minnich never made." A fair hit, and the article should own it. Every fitness number behind the stepwise claim comes from Barrick's lab, not from Van Hofwegen, Hovde & Minnich, whose paper reports the mutational steps and measures no fitness effect for any of them.² ¹² What Minnich's group established is that the trait evolves readily under direct selection and by the same class of mutation as in the LTEE. What they did not establish is that any single step pays its way, and McLatchie's four-sentence summary states it as though they had.¹¹ The claim is right; the citation under it is not.

"Why has no other population evolved Cit+ in 50,000 further generations?" This is the best remaining argument for real difficulty, and it cuts against Minnich. Since Ara–3 turned cloudy, the other eleven populations have run another 49,000 generations and produced roughly 4 × 10¹³ more cells.⁴⁸ At Blount's measured rate that is enough for another 26 origination events. None established. Whether the block is the dcuS defect Minnich identifies, or clonal interference, or both, is unresolved. Note who proposed the experiment that would settle it. Van Hofwegen et al. wrote in 2016 that the frozen archive "could be used to test this possibility. They could be revived and the experiments replayed from any generational landmark with longer incubations extended into deeper stationary phase ... to determine if the pathway to citrate use has been preserved or lost among all 12 E. coli cultures."² Roth and Maisnier-Patin seconded it. Until somebody runs it, the eleven-population silence is evidence about the LTEE's transfer regime, not about whether those populations could do it. That is the sharpest open question in the whole dispute, and the cheapest experiment in it.

Where this leaves the argument

The LTEE settles less than either side wants and more than either side admits.

It settles that a large asexual population, given tens of thousands of generations, will keep improving in its own environment even while its genome sheds functions it does not use, and that most of that improvement will be achieved by discarding or retuning what it already has. It settles that a genuinely new metabolic capability can arise by re-wiring existing parts, without any new gene, and that such an event is rare enough to happen once in twelve replicate populations over 38 years. It settles that the rarity was contingent — dependent on which mutations that particular lineage happened to acquire — even though the specific mechanism proposed for that contingency in 2008 has since been replaced twice.

It does not settle whether the citrate transporter itself could have arisen this way, which is the question that actually divides the two camps. Nothing in 82,000 generations of E. coli speaks to the origin of a transporter, a fold, or a code. Both sides keep citing the LTEE for that purpose. Neither should.

Sources

Full source-by-source treatment in the literature review. Verbatim quotations from all parties are in quotes.

A note on method. No quotation in this article was established through an AI page-summarising tool. Every quotation was read out of bytes downloaded locally — PDFs converted with pdftotext -layout, HTML stripped locally, and in the case of Lenski et al. 1991 (a scanned JSTOR offprint with no text layer) read off 300 dpi rendered page images via OCR and checked by eye. Library keys refer to sources/index.tsv.

1.
Blount, Zachary D. et al. “Historical contingency and the evolution of a key innovation in an experimental population of Escherichia coli.” PNAS 105(23):7899–7906. 2008. A study of how the citrate-using trait (Cit+) arose in population Ara–3 after more than 30,000 generations. The authors replayed the experiment from clones saved at different times, and found that later clones were more likely to evolve the trait, which they read as historical contingency: the outcome depended on what had evolved before. The article uses it for the citrate story, covering the long wait, the replay tests and the potentiating mutation. ↩
1. 1a.
  Page 7899, section Abstract: “No population evolved the capacity to exploit citrate for >30,000 generations, although each population tested billions of mutations. ... We observed no Cit+ mutants among 8.4 × 10¹² ancestral cells, nor among 9 × 10¹² cells from 60 clones sampled in the first 15,000 generations. However, we observed a significantly greater tendency for later clones to evolve Cit+, indicating that some potentiating mutation arose by 20,000 generations.” ↩ ↩
2. 1b.
  “The inability to use citrate as an energy source under oxic conditions has long been a defining characteristic of E. coli as a species.” ↩ ↩
3. 1c.
  Section Materials and Methods: “The twelve populations have been propagated for almost 20 years by daily serial dilution in DM25, a minimal salts medium that has 139 μM glucose and 1,700 μM citrate.” ↩ ↩
4. 1d.
  “Nevertheless, E. coli is not wholly indifferent to citrate. It uses a ferric dicitrate transport system for iron acquisition, although citrate does not enter the cell in this process. It also has a complete tricarboxylic acid cycle, and can thus metabolize citrate internally during aerobic growth on other substrates. E. coli is able to ferment citrate under anoxic conditions if a cosubstrate is available for reducing power. The only known barrier to aerobic growth on citrate is its inability to transport citrate under oxic conditions.” ↩
5. 1e.
  “During that time, each population experienced billions of mutations, far more than the number of possible point mutations in the ≈4.6-million-bp genome. This ratio implies, to a first approximation, that each population tried every typical one-step mutation many times. It must be difficult, therefore, to evolve the Cit+ phenotype, despite the ecological opportunity.” ↩
6. 1f.
  Table 1: First experiment: 72 replay populations, 4 independent Cit+ mutants. Second: 340 plates, 5 mutants. Third: 2,800 cultures totalling ≈4.0 × 10¹³ cells, 8 mutants. Totals 3,212 replicates and 17 mutants. Blount et al. 2012 restate this as: “270 clones sampled over the population's history. The replays produced 17 Cit+ mutants that derived from 13 clones, all from generation 20,000 or later.” ↩
7. 1g.
  Section Results: How the fluctuation tests were run: “We prepared 40 replicate cultures for each potentiated clone, and 280 for the ancestor. Potentiated and ancestral cultures had, on average, 1.1 and 1.5 × 10¹⁰ cells, respectively, which were harvested and incubated on MC agar plates for 45 days.” The authors' own caution on the ancestral bound: it. “is also probably far too high because no mutations were actually observed for the ancestor ... and because some cell turnover and other DNA activity probably occurred during the many days that plates were incubated.” On where the second mutation happened: “This conclusion, taken together with the jackpot distribution of Cit+ mutants, indicates that the phenotypic change required two mutations, one of which occurred during the culture growth before plating and the other after plating.” ↩
8. 1h.
  Section Results: “With no more than three mutations among the 8.4 × 10¹² cells tested here and in the third replay experiment, the upper bound on the ancestral mutation rate to Cit+ is 3.6 × 10⁻¹³ per cell per generation. ... Cit+ mutants arose in 2 of the 280 new cultures, giving an estimate of 6.6 × 10⁻¹³ for the mutation rate, with the 95% CI extending from 7.9 × 10⁻¹⁴ to 2.4 × 10⁻¹². ... The potentiated genetic background thus increases the mutation rate to Cit+ at least 2-fold, and probably much more.” For comparison, the paper measured the Ara⁻ → Ara⁺ point-mutation rate at 2.3 × 10⁻¹⁰ in both ancestor and potentiated clones, showing the potentiation was specific and not general hypermutability. ↩ ↩ ↩ ↩
9. 1i.
  “Although the Cit+ cells continued to use glucose, they did not drive the Cit– subpopulation extinct because the Cit– cells were superior competitors for glucose. ... the Cit− cells gradually approached an equilibrium frequency of roughly 1%, regardless of their initial frequency.” ↩
10. 1j.
  Section Discussion and Future Directions: The paper's own closing question, in its Discussion: “Will the Cit+ and Cit– lineages eventually become distinct species?” ↩
11. 1k.
  Section Results: “the potentiation effect in later generations is underestimated by these tests, because the number of cells trended lower in later-generation cultures owing to the evolution of larger cells that reach lower population density.” ↩
12. 1l.
  “All evolved higher maximum growth rates on glucose, shorter lag phases upon transfer into fresh medium, reduced peak population densities, and larger average cell sizes relative to their ancestor. Ten populations evolved increased DNA supercoiling.” ↩
13. 1m.
  Table 2: Monte Carlo P values 0.0085, 0.0007 and 0.0823 for the three replay experiments. The text says: “Although the third experiment was the largest, it was the least significant, owing primarily to the production of two Cit+ mutants by a 20,000-generation clone.” Combined by the Z-transformation method the result is P < 0.0001. ↩ ↩
14. 1n.
  Section Abstract: “This potentiating change increased the mutation rate to Cit+ but did not cause generalized hypermutability.” The 2008 hypothesis is explicitly mutational access, not epistatic rescue. ↩
15. 1o.
  Page 7904, section Discussion and Future Directions: “One possibility is that the Cit+ lineage activated a "cryptic" transporter, that is, some once-functional gene that has been silenced by mutation accumulation. … A more likely possibility, in our view, is that an existing transporter has been coopted for citrate transport under oxic conditions. This transporter may previously have transported citrate under anoxic conditions or, alternatively, it may have transported another substrate in the presence of oxygen.” ↩
16. 1p.
  Page 7900: “Hall reported the only documented case of a spontaneous Cit+ mutant in E. coli. … Pos et al. identified an operon in E. coli K-12 that apparently allows anaerobic citrate fermentation, and which includes a gene, citT, encoding a citrate–succinate antiporter. High-level constitutive expression of this gene on a multicopy plasmid allows aerobic growth on citrate”. ↩
2.
Van Hofwegen, Dustin J. et al. “Rapid Evolution of Citrate Utilization by Escherichia coli by Direct Selection Requires citT and dctA.” Journal of Bacteriology 198(7):1022–1034. 2016. An experiment that used direct selection, which picks out bacteria able to grow on citrate, to see whether citrate-using mutants appear as quickly as the LTEE's did after 33,000 generations. It isolated 46 independent citrate-using mutants, the earliest within as few as 12 generations, and sequenced four of them. Minnich is a Discovery Institute fellow and was an expert witness for the defence in Kitzmiller v. Dover. The article uses it for Minnich's headline result, 46 citrate-using mutants within as few as 12 generations, and for the claim that the rarity of the LTEE mutant was an artifact of the experimental conditions. ↩ ↩
1. 2a.
  Section Abstract, page 1022: “Forty-six independent E. coli Cit+ mutants were isolated from all E. coli derivatives except the E. coli ΔcitT::kan mutant. Potentiation/actualization mutations occurred within as few as 12 generations, and refinement mutations occurred within 100 generations.” ↩ ↩
2. 2b.
  Page 1022, section Importance statement: The final two sentences of the statement: “We conclude that the rarity of the LTEE mutant was an artifact of the experimental conditions and not a unique evolutionary event. No new genetic information (novel gene function) evolved.” ↩
3. 2c.
  The claim appears in the importance statement and twice in the discussion, including: “because this adaptation did not generate any new genetic information and required expanded expression of only two existing transporters (citT and dctA), generation of E. coli Cit+ phenotypes in our estimation does not warrant consideration as a speciation event.” ↩
4. 2d.
  Section Results, pages 1024–1028: Media: M9 minimal salts with 6.8 mM citrate (M9C), or with 0.005% glycerol (M9CG50) or 0.0025% glucose (M9C25); 50 mL in 250-mL flasks; parallel experiments at 1.7 mM citrate to match the LTEE. Strains: E. coli K-12 wild type, ΔrpoS::kan, ΔcitT::kan, ΔdctA::kan, E. coli B, and REL606. ↩
5. 2e.
  Page 1027: The article quotes this passage in full. ↩ ↩
6. 2f.
  Page 1027: “All Cit+ E. coli strains were recovered within 40 days of the beginning of the selection incubations, with one exception, the E. coli ΔdctA::kan mutant, which arose after 134 days.” That sentence does not agree with the paper's own Table 1, which lists recoveries at 41, 42, 43, 56, 63, 68 and 77 days as well as the 134-day outlier. The 46 rows that carry a day value give a minimum of 19, a median of 49.5, and 40 of 46 falling between 28 and 63 days, which confirms Lenski's reading. The paper contradicts itself here, not this article. The day values are 19, 21, 21, 28, 30, 31, 35 (×6), 40 (×4), 41, 42 (×5), 43, 56 (×2), 63 (×20), 68, 77 and 134. That count of forty-six also independently corroborates the paper's. “46 independent Cit+ mutants.” ↩
7. 2g.
  Pages 1029, 1031: “The read depth profile of the genome of strain DV130 showed 4-fold coverage of citT and 2-fold coverage of dctA above the normalized coverage of 100× for the rest of the chromosome.” And: “Because our E. coli Cit+ cells were recovered after a minimum of 12 generations, potentiation requires only citT amplification, as determined for strain DV130. Increased expression of the CitT transporter allowed minimal but sufficient access to citrate and a low level of cell division.” Note that DV130 also carried a single-base-pair deletion in dctA 26 nucleotides from the translation start, five bases from the mutation Quandt et al. identified. ↩
8. 2h.
  Pages 1029–1030: DV133 gained. “insI1, an IS30 transposase ... inserted 5′ to the citT gene, a presumptive promoter capture”. DV159 showed. “a duplication and a deletion to generate a fusion of citT with the neighboring uspG gene ... analogous to that of the original Cit+ strain identified in the LTEE”. and both carried a 140-kb duplication containing dctA, generated by recombination between rhsA and rhsB. ↩
9. 2i.
  Page 1033: “the E. coli strain used, REL606, has a dcuS five-base-pair deletion that prevents dctA expression. This strain is defective in the very pathway required for competitive citrate utilization.” This is where the five-base figure comes from. Neither Minnich & Hovde 2025 nor Quandt et al. 2014 states a size. ↩
10. 2j.
  Page 1028: “we made a P1(vir) bacteriophage lysate based on the strain DV133 isolate. This lysate was used to transduce E. coli K-12 wild-type cells at an MOI of 0.1:1. ... After 4 days, the transduced culture displayed turbid growth with an OD600 of 0.7 and a sample from this flask was streaked directly onto Simmons citrate agar. Blue colonies appeared after 2 days. ... This E. coli K-12 wild-type Cit+ transductant was designated strain DV133T.” ↩
11. 2k.
  Pages 1032–1033: “the E. coli strain used, REL606, has a dcuS five-base-pair deletion that prevents dctA expression. This strain is defective in the very pathway required for competitive citrate utilization. ... Nonetheless, we predict that Cit+ mutations, even in REL606, might have occurred sooner and repeatedly in the LTEE protocol if the dcuS in this strain had been repaired, because then, only gene amplifications (citT and dctA) would be required, as we found to be the case. The following three results support this reasoning about the dcuS defect: (i) we rapidly isolated Cit+ mutants in E. coli REL606 and another E. coli B strain; (ii) our characterized E. coli K-12 Cit+ mutants required only dctA duplication, a more common process than the dctA-specific point mutation required to suppress a dcuS defect; (iii) we isolated a Cit+ mutation in the E. coli K-12 ΔdctA::kan mutant, a strain phenotypically equivalent to the REL606 strain with a dcuS mutation (DctA negative), in 134 days.” The third of those is a purpose-built control and it is the experimental leg of the argument. ↩ ↩
12. 2l.
  Page 1033: “Because Cit+ E. coli REL606 has been identified in only 1 of the 12 parallel cultures of the LTEE, it may well be that a weak selection inhibits access to a 'citrate-use solution.' Frozen fossils from the 12 LTEE cultures could be used to test this possibility. They could be revived and the experiments replayed from any generational landmark with longer incubations extended into deeper stationary phase, the selection conditions described by us, to determine if the pathway to citrate use has been preserved or lost among all 12 E. coli cultures.” Roth & Maisnier-Patin close their commentary in the same spirit: “Other aspects are eminently testable.” No such replay has been published as of August 2026. ↩
13. 2m.
  Page 1032: “First, potentiated and actualized cells were routinely diluted away by the LTEE daily 1:100 dilution transfers. Cells with an amplified citT gene had only a brief advantage in the use of citrate once glucose was expended and were likely lost with each daily dilution/transfer into fresh glucose-containing media due to the fitness cost of the maintenance of multiple gene copies.” ↩ ↩ ↩
14. 2n.
  Page 1033: “This indicates that the 33,000 generations to potentiate the evolutionary resources for the Cit+ phenotype do not reflect a direct requirement but merely experimental conditions.” ↩
15. 2o.
  Page 1033: “The frequency of the occurrence of both amplification/promoter capture (citT) and a point mutation (dctA) in the same cell under nonselective conditions is on the order of 1 in 10¹⁴. This is almost exactly the number of cells Lenski screened to find Cit+, and the frequency Hall predicted for his Cit+ mutation.” The article adds emphasis to the words under nonselective conditions. ↩
16. 2p.
  Page 1022: “To date, the Lenski LTEE cultures exceed 63,000 generations, equivalent to over 1 million years of human evolution.” ↩
3.
Lenski, Richard E., and Zachary D. Blount. “On the Evolution of Citrate Use.” Telliamed Revisited. Feb 20, 2016. A blog post, not a peer-reviewed comment, by Richard Lenski and Zachary Blount of Lenski's lab, replying to the 2016 paper by Van Hofwegen, Hovde and Minnich. It was published three days after the accepted manuscript appeared, and no journal reply was published. The article uses it for the authors' answer to the 'no new information' charge and for their praise of the experiments. ↩
1. 3a.
  “The claim that 'no new genetic information evolved' is based on the fact that the bacteria gained this new ability by rearranging existing structural and regulatory genetic elements. But that's like saying a new book—say, Darwin's Origin of Species when it first appeared in 1859—contains no new information, because the text has the same old letters and words that are found in other books.” ↩
2. 3b.
  “In an evolutionary context, a genome encodes not just proteins and patterns of expression, but information about the environments where an organism's ancestors have lived and how to survive and reproduce in those environments ... So when natural selection—that is, differential survival and reproduction—favors bacteria whose genomes have mutations that enable them to grow on citrate, those mutations most certainly provide new and useful information to the bacteria.” ↩
3. 3c.
  “these Cit+ strains can grow on citrate as a sole carbon source—when E. coli grows anaerobically on citrate, it requires a second substrate for growth in order to use the citrate (a phenomenon called 'co-metabolism').” ↩
4. 3d.
  “Moreover, the actual science that was done and reported looks fine and interesting, though we have a few quibbles with some details that we will overlook for now. By and large, the work confirms many of the findings that were reported in our papers cited above.” ↩
5. 3e.
  “the fastest time that Van Hofwegen et al. saw for the Cit+ function to emerge was 19 days (from their Table 1), and in most cases it took a month or two. While that's a lot faster than 15 years, it's still much longer than typical 'direct selections' used by microbiologists where a readily accessible mutation might confer, for example, resistance to an antibiotic after a day or two.” ↩ ↩
6. 3f.
  “In short, speciation is not an event. ... In our first paper on the citrate-using E. coli that arose in the LTEE, we clearly emphasized that becoming Cit+ was only a first step on the road to possible speciation.” ↩
4.
Mauger, Bethany. “'Evolution under a microscope' going strong at MSU.” MSUToday. Nov 11, 2025. “In 2022, the LTEE left MSU's campus and was placed in the hands of Lenski's former postdoctoral researcher, Jeffrey Barrick, at the University of Texas at Austin. His group continued the experiment there to 82,000 generations. Now, the E. coli are home again.” A university press release rather than a peer-reviewed source, by Bethany Mauger for MSUToday, the news site of Michigan State University. The article uses it only for the generation count and the custody history. ↩ ↩
5.
Berean Archive. In-house calculation. n.d. Stationary-phase population = 5 × 10⁷ cells/mL × 10 mL = 5 × 10⁸; after a 1:100 transfer, 5 × 10⁶; so 4.95 × 10⁸ cells are born per population per day. Generations per day = log₂(100) = 6.6439. Through 82,000 generations that is 12,342 days, giving 6.11 × 10¹² cells per population and 7.33 × 10¹³ across twelve; through 33,127 generations (4,986 days), 2.47 × 10¹² and 2.96 × 10¹³. The per-genome mutation rate is 8.9 × 10⁻¹¹ × 4,629,812 = 4.12 × 10⁻⁴ per cell division, so 6.11 × 10¹² cells give 2.52 × 10⁹ point mutations per population at 82,000 generations. Distinct possible point mutations = 4,629,812 × 3 = 1.39 × 10⁷, so each has been made about 181 times per population. These are upper estimates: evolved cells are larger and reach lower densities than the ancestor, and the experiment lost some days to freezing, contamination restarts and the 2020 pandemic shutdown. A calculation made for this article from the design parameters in Lenski et al. 1991, not a published study. The article uses it for the cell and mutation totals. ↩ ↩ ↩ ↩
6.
Wielgoss, Sébastien et al. “Mutation Rate Inferred From Synonymous Substitutions in a Long-Term Evolution Experiment With Escherichia coli.” G3 1:183–186. 2011. “The resulting molecular estimate was 8.9 × 10⁻¹¹ per base-pair per generation.” A genome study of 19 clones, lines grown from single cells, taken from eight of the experiment's twelve populations after about 40,000 generations. It counted the changes that leave the protein unchanged (synonymous substitutions) and found the rate, which the authors call the most accurate bacterial estimate available to date. Multiplied by the 4,629,812-bp REL606 genome, that is 4.1 × 10⁻⁴ mutations per genome per generation. The article uses that rate to estimate how many point mutations each population has made. ↩ ↩ ↩
7.
Lenski, Richard E. et al. “Long-Term Experimental Evolution in Escherichia coli. I. Adaptation and Divergence During 2,000 Generations.” The American Naturalist 138(6):1315–1341. 1991. The founding paper of the long-term experiment, in which twelve E. coli populations were founded from one clonal ancestor and grown for 2,000 generations. Mean fitness rose by about 37% and the rate of gain slowed. The article takes its design and its early fitness figures from it. ↩
1. 7a.
  Pages 1317–1318, section Culture Conditions: “The culture medium used in all experiments was Davis minimal broth ... supplemented with 2 × 10^-6 g thiamine hydrochloride and 0.025 g glucose per liter, which permits a stationary phase bacterial density of about 5 × 10⁷ cells per mL. Culture volume was 10 mL maintained in 50-mL Erlenmeyer flasks set in a shaking incubator at 37°C and 120 rpm. Cultures were propagated daily by transferring 0.1 mL of each culture into 9.9 mL of fresh medium. ... The resulting 100-fold daily growth of each bacterial population represents ~6.64 generations of binary fission.” ↩
2. 7b.
  Page 1315, section Abstract: “Twelve Escherichia coli populations were founded from a single clonal ancestor and allowed to evolve for 2,000 generations. Mean fitness increased by about 37%. However, the rate of increase in mean fitness was slower in later generations.” ↩ ↩
3. 7c.
  Page 1318, section Fitness Assay Methods and Definitions: Relative fitness is the ratio of Malthusian parameters of the two competitors over one daily growth cycle, with the neutral arabinose marker used to tell the colonies apart on tetrazolium-arabinose plates. The marker's neutrality was measured at 1.00 ± 0.01 (95% CI). ↩
8.
Turner, Caroline B. et al. “Evolution of a cross-feeding interaction following a key innovation in a long-term evolution experiment with Escherichia coli.” Microbiology 169:001390. 2023. A research article from Lenski's group on what happened after one population of the experiment gained the ability to grow on citrate. It found that the citrate-using cells release succinate, fumarate and malate, and that the cells that cannot use citrate evolved to eat them, so the population went from one limiting resource to at least five. The article uses it for the count that only one of the twelve populations had evolved aerobic growth on citrate by 75,000 generations, and for the cross-feeding that keeps the two lineages coexisting. ↩
1. 8a.
  Section Discussion: “The ability to grow aerobically on citrate has so far evolved in only one of the twelve LTEE populations, even after 75 000 generations and billions of spontaneous mutations in each population.” ↩
2. 8b.
  Section Abstract: “This coexistence was shaped by the evolution of a cross-feeding relationship based on C4-dicarboxylic acids, particularly succinate, fumarate, and malate, that the Cit+ variants release into the medium. Both the Cit− and Cit+ cells evolved to grow on these excreted resources. The evolution of aerobic growth on citrate thus led to a transition from an ecosystem based on a single limiting resource, glucose, to one with at least five resources that were either shared or partitioned between the two coexisting clades.” ↩
3. 8c.
  Section Discussion: “The LTEE ancestor does not express DctA, owing to a 5 bp frameshift mutation that disrupts the gene, dcuS, that encodes its positive regulator. We have shown that a 5 bp deletion in dcuS occurred in the Cit− lineage, which restored the reading frame and enabled growth on C4-dicarbolyates (Fig. 7). By contrast, Quandt et al. showed that a mutation in the promoter region of the dctA gene — one that caused constitutive, high-level expression of DctA — conferred the ability to grow on C4-dicarboxylates in the Cit+ lineage.” ("dicarbolyates" is the paper's typo.) Strain ZDB1052 is REL606 carrying only the evolved dcuS allele from the 34,000-generation Cit− clone ZDB86; it grows on succinate, fumarate and malate where REL606 does not, and grows more slowly on glucose. ↩ ↩
4. 8d.
  Section Results: “It is perhaps relevant that the 5 bp insertion in dcuS in the ancestor creates a tandem array with three repeats of the same 5 bp sequence. The deletion of one copy of this repeat yields a two-copy array and a functional dcuS sequence. Such mutations in tandem arrays often occur at high rates.” ↩
5. 8e.
  Section Discussion: “We also showed that the fitness of the Cit− ecotype increased over time in the LTEE medium, but only in the presence of the Cit+ ecotype (Fig. 6). In the absence of the Cit+ ecotype, the Cit− ecotype actually lost fitness in that medium (Fig. 6).” This is one clade measured against a counterfactual without its cross-feeding partner, not the mean fitness of a population, so it is not a counterexample to the grand-mean trajectory in Lenski et al. 2015. It is included because a reader hunting for a fitness decline in the LTEE will find this sentence, and should find it here first. ↩
9.
Berean Archive. In-house calculation. n.d. 1,700 µM citrate against 139 µM glucose is a molar ratio of 12.2. Both molecules carry six carbons, so the citrate pool holds about twelve times the carbon of the glucose the cells were actually eating. A calculation made for this article, not a published study, comparing the citrate and glucose in the growth medium. The article uses it for the twelve-fold figure. ↩
10.
Blount, Zachary D. et al. “Genomic analysis of a key innovation in an experimental Escherichia coli population.” Nature 489(7417):513–518. 2012. A genome study of the bacteria that learned to use citrate, written by the experiment's own team, Lenski among them. It sequenced their genomes and found that the trait came from a duplication, in which a stretch of DNA is copied so that the two copies sit side by side, that placed a switch that works in oxygen (a promoter) in front of a citrate transporter gene that had been silent. The article uses it for the step that switched the trait on. ↩
1. 10a.
  “We constructed a low-copy (1-2 per cell) plasmid, pCDrnk-citTlux ... We made two other plasmids, pCDrnklux and pCDcitTlux, where the reporter was under the control of the native upstream regulatory regions of rnk and citT, respectively.” A reading at background on a one-copy reporter cannot distinguish "no transcription" from "transcription below the assay floor," and the whole amplification hypothesis is about what happens when a sub-threshold transcript is multiplied. ↩
2. 10b.
  “All 9 sequenced Cit+ genomes have two or more tandem copies of a 2933-bp segment that includes part of the citrate fermentation (cit) operon. ... The amplification joined upstream rnk and downstream citG fragments, producing an rnk-citG hybrid gene expressed from the upstream rnk promoter. Because the citT and citG genes are normally monocistronic, the downstream copy of citT should therefore be co-transcribed with the hybrid gene.” ↩ ↩ ↩ ↩ ↩
3. 10c.
  “The native citT regulatory region showed no expression (above background) in any strain, indicating that citT is normally silent under oxic conditions. The native rnk regulatory region was expressed in all three strains, with a peak around the transition into stationary phase. Expression from the evolved rnk-citT module was much weaker, but there were small spikes in expression in ZDB30 and ZDB172.” ↩ ↩ ↩
4. 10d.
  “This construct, ZDB595, indeed has a Cit+ phenotype, although it is extremely weak, similar to the earliest evolved Cit+ variants. ... We also isolated 13 Cit− revertants of a 33,000-generation Cit+ clone, CZB154. All had lost the cit amplification.” ↩
5. 10e.
  “In the same medium used in the LTEE, ZDB595 experienced a lag of >60 hours after glucose depletion, followed by a short period of abortive and inconsistent growth on citrate. These data imply that the initial Cit+ type was too weak to allow exploitation of citrate under the daily-transfer regime of the LTEE. Nonetheless, ZDB595 had a small (1.0%) but significant competitive advantage over ZDB30 in the same environment (n = 10, t = 3.09, two-tailed P = 0.0128).” ↩ ↩
6. 10f.
  “The earliest one had a tandem duplication, whereas later genomes had a three-copy tandem array within a larger tandem duplication, a four-copy tandem array, a tandem duplication in a larger three-copy tandem array, and a nine-copy tandem array. ... The increased number of rnk-citT modules can thus explain the refinement of the Cit+ phenotype that allowed the population expansion.” ↩ ↩ ↩
7. 10g.
  “These differences demonstrate epistatic interactions between the rnk-citT module and mutations that distinguish the backgrounds.” The 2012 paper's phylogenetic analysis also concluded that. “potentiation involved at least two mutations, with one arising before these three clades diverged and another in C3.” ↩ ↩
8. 10h.
  “The sequenced Cit+ clones from generation 36,000 and later have a SNP in the mutS gene that produces a premature stop codon and truncates the MutS protein, thereby causing a defect in methyl-directed mismatch DNA repair. ... this mutation arose after the origin of the Cit+ lineage. Also, none of the 20 Cit− genomes has this mutation.” ↩
11.
McLatchie, Jonathan. “Answering Dave Farina on the Lenski Experiment.” Science and Culture Today (Discovery Institute). Aug 18, 2026. One article in a series in which McLatchie answers the YouTuber Dave Farina about the Lenski experiment, and in which he states the stepwise path to citrate use. The article uses it as the Discovery Institute's current statement of that stepwise claim. ↩
1. 11a.
  “The prior neutral mutations are not necessary for potentiation — the only requirement is citT amplification. As Minnich and his colleagues showed, this can happen in as few as 12 generations.” The claim is about what potentiation requires, not about how long a laboratory takes. ↩
2. 11b.
  McLatchie quoting Farina: “A particular strain evolved an attribute that qualifies as irreducibly complex according to Behe's own definition ... This experiment produced a strain of bacteria that was able to digest citrate in an aerobic environment, and in order to do this it required multiple mutations in both the dctA gene and citT gene, and a duplication of the citT gene into a new region of the genome. All of these changes were required, and none of them can have been selected for individually. In fact, the critical duplication is actually deleterious on its own without the other mutations.” The last sentence is the one Leon and Barrick's +1.0% measurement contradicts. ↩
3. 11c.
  The claim is quoted in full in the article's body. The article's endnote 3 attributes the claim to Van Hofwegen, Hovde & Minnich 2016; the four-stage description given does not appear in that form in the cited paper, which reports amplification, promoter capture and dctA activation but measures no fitness effect for any of them. ↩ ↩ ↩ ↩
12.
Leon, Dacia et al. “Innovation in an E. coli evolution experiment is contingent on maintaining adaptive potential until competition subsides.” PLoS Genetics 14(4):e1007348. 2018. A laboratory study of the LTEE's citrate-using lineage that rebuilt the first mutation needed for citrate use, in strains isolated from the first 31,500 generations, and measured its fitness effect. The mutation was already slightly beneficial early on, but competition with other mutations blocked it until later changes restored the benefit. The article uses it to explain why citrate use took so long to evolve. ↩ ↩
1. 12a.
  Section Abstract: “We found that there was already a slight fitness benefit for this mutation in the original ancestor of the evolution experiment and in other early isolates. However, evolution of citrate utilization was blocked at this point due to competition with other mutations that improved fitness in the original niche. Subsequently, an anti-potentiated genetic background evolved in which it was deleterious to evolve rudimentary citrate utilization. Only later, after further mutations accumulated that restored the benefit of this first-step mutation and the overall rate of adaptation in the population slowed, was citrate utilization likely to evolve.” ↩ ↩ ↩ ↩ ↩ ↩
2. 12b.
  “We found a slight fitness benefit of 1.0% for the Cit+ mutation”. The measurement was made in. “the ancestral strain, REL606.” ↩ ↩ ↩
3. 12c.
  Section Methods, Development of Prnk-citT knock-in assay for potentiation: Earlier work had tested the module on a multicopy plasmid, “but this approach is problematic because these plasmids are multicopy, whereas only a single activated copy of the citT gene was present in the initial Cit+ strains,”. and building the authentic duplication into a chromosome fails because. “it readily collapses via homologous recombination.” So they made. “an activated citT module linked to an antibiotic selection marker in which the rnk promoter is upstream of the truncated rnk-citG fusion ORF formed by the duplication followed by the complete citT reading frame,”. integrated in place of the lac operon, “which is unrelated to citrate or glucose metabolism,”. and competed it against a null module carrying only the resistance gene. Validation: in the fully potentiated Cit− revertant ZDB706 the knock-in gave citT mRNA levels statistically indistinguishable from the naturally evolved Cit+ isolate ZDB564 (P = 0.65), and a fitness advantage of 2.4% indistinguishable from the authentic mutation's (P = 0.71). ↩
4. 12d.
  “For five of these strains, activating citT expression was extremely detrimental, decreasing competitive fitness by >20% (one-tailed bootstrap test, P < 0.05).” ↩
5. 12e.
  Section Discussion, pages 12–13: The article quotes the sentence in full in its text. The sentence is about the class of experiment Minnich ran: “Cit++ mutants of E. coli ... have been isolated in other studies ... because they involve starving E. coli cells for days to weeks under conditions in which citrate was present as a potential carbon source.” Van Hofwegen, Hovde & Minnich 2016 is one of the four studies cited there. ↩ ↩
6. 12f.
  Section Results: “we found that activating citT expression drastically increased the lag phase of growth in the severely anti-potentiated strains (Fig 2B). This additional lag time can explain the sizable competitive disadvantage”. versus the Cit− strain, “even though the Cit+ variants are able to reach a higher final cell density if cultured alone.” This is the mechanism, and it is why the penalty vanishes in a citrate-only culture: a slow start is fatal in a race and irrelevant in a famine. ↩
7. 12g.
  Section Results: “ZDB425 at 10,000 generations and ZDB458 at 20,000 generations have only one and two 'private' mutations not shared with the main pre-Cit+ lineage, respectively, though we cannot rule out that other changes in the impact of citT activation may have occurred on the main line of descent within this interval.” And of the worst-affected clones: “Other strains are not classified into these major groups. It is less likely that they are representative of how potentiation evolved in the lineage leading to Cit+. For example, the four most highly anti-potentiated clones (ZDB467, ZDB483, ZDB14, ZDB18) appear to have evolved this property independently and due to 'private' mutations not shared with the main pre-Cit+ lineage.” ↩
8. 12h.
  Figures 1 and 3 and associated text: +2.2% in ZDB564; +2.4% in ZDB706; +0.9% when competed at ~1% frequency inside the full 31,000-generation population sample; group-wise effects along the line of descent, which the paper lists backward in time from the earliest Cit+ isolate: +2.4% [+1.4%, +3.4%] for ZDB19 and ZDB13. “from as early as 29,000 generations”. ; +0.4% [−1.3%, +2.0%] for three clones from 25,000 to 27,000 generations; −5.4% [−7.2%, −3.6%] for. “an earlier, intermediate set of isolates composed of ZDB425, ZDB458, and ZDB464”. ; and +1.7% [0.0%, +3.3%] for the. “initial cluster that groups ZDB409 and ZDB429 with the REL606 ancestor.” Twenty-five backgrounds were tested in all: “23 additional clones”. plus REL606 and ZDB706. The paper is also explicit about why direct-selection experiments differ: “this difference in environmental conditions relative to the glucose-limited transfer regime of the LTEE, in which cells are in stationary phase for only ~16–18 hours each day, dramatically increases the fitness benefit of evolving the rudimentary Cit+ phenotype ... Activating citT expression would be universally beneficial in any genetic background in these types of experiments.” ↩ ↩
13.
Quandt, Erik M. et al. “Recursive genomewide recombination and sequencing reveals a key refinement step in the evolution of a metabolic innovation in Escherichia coli.” PNAS 111(6):2217–2222. 2014. A genetics study of the citrate-using bacteria by Jeffrey Barrick's team, and the article credits Barrick with the dctA result. The authors built a method called REGRES, short for recursive genomewide recombination and sequencing, to strip out the mutations that were not needed for citrate use, and found that one mutation raising the activity of the dctA transporter greatly strengthened the trait. The article uses it for that refinement step. ↩
1. 13a.
  Section Abstract: “To determine which of the other >70 mutations in early Cit+ clones were needed to take full advantage of citrate, we developed a recursive genomewide recombination and sequencing method (REGRES) and performed genetic backcrosses to purge mutations not required for Cit+ from an evolved strain. We discovered a mutation that increased expression of the dctA C4-dicarboxylate transporter greatly enhanced the Cit+ phenotype after it evolved.” ↩
2. 13b.
  Section Abstract: “Surprisingly, strains containing just the citT and dctA mutations fully use citrate, indicating that earlier mutations thought to have potentiated the initial evolution of Cit+ are not required for expression of the refined version of this trait.” ↩ ↩
3. 13c.
  Section Results: “We were surprised to find that the citT actualizing mutation and the *dctA** refinement mutation are sufficient for a strong Cit++ phenotype, demonstrating that potentiating alleles are not strictly required for expression of this trait.” ↩ ↩
4. 13d.
  “DctA allows import of succinate to be powered by the protonmotive force. Therefore, increased expression of DctA is expected to be beneficial for aerobic citrate utilization because it makes it possible to reuptake succinate or other C4-dicarboxylates that are exported in exchange for citrate import by the CitT antiporter. This activity completes a cycle that permits sustained citrate transport and utilization without an unbalanced loss of intracellular C4-dicarboxylate substrates due to efflux.” ↩ ↩ ↩
5. 13e.
  “Unlike most strains of E. coli K-12, the ancestral strain of E. coli B used in the LTEE cannot use C4-dicarboxylates. This difference has been attributed [to] an inactivating mutation in the gene for DcuS, the sensor kinase in the DcuSR two-component system that regulates genes involved in C4-dicarboxylate transport and utilization.” This is the finding Minnich & Hovde 2025 build their strain argument on. ↩
14.
Cooper, Vaughn S. et al. “Mechanisms Causing Rapid and Parallel Losses of Ribose Catabolism in Evolving Populations of Escherichia coli B.” Journal of Bacteriology 183(9):2834–2841. 2001. A research paper on why all twelve lines lost the ability to use D-ribose (a sugar) within 2,000 generations. The authors isolated seven independent mutants, measured their fitness against the ancestor, and traced the loss to deletions of the ribose operon (the gene cluster that lets a cell use ribose). The article uses it as the cleanest case of loss of function in the experiment. ↩
1. 14a.
  Page 2834, section Abstract: “Twelve populations of Escherichia coli B all lost D-ribose catabolic function during 2,000 generations of evolution in glucose minimal medium. ... These Rbs⁻ mutants were all about 1 to 2% more fit than the progenitor.” ↩
2. 14b.
  Page 2834, section Abstract: “A fluctuation test revealed an unusually high rate, about 5 × 10⁻⁵ per cell generation, of mutation from Rbs⁺ to Rbs⁻, which contributed to rapid fixation.” ↩ ↩
3. 14c.
  Page 2834, section Abstract: “The physical extent of the deletion varied between mutants, but each deletion was associated with an IS150 element located immediately upstream of the rbs operon. The deletions apparently involved transposition into various locations within the rbs operon; recombination between the new IS150 copy and the one upstream of the rbs operon then led to the deletion of the intervening sequence.” ↩
4. 14d.
  Page 2834, section Abstract: “we used P1 transduction to restore the functional rbs operon to two Rbs⁻ mutants, and we constructed another Rbs⁻ strain by gene replacement with a deletion not involving IS150. All three of these new constructs confirmed that Rbs⁻ mutants have a competitive advantage relative to their Rbs⁺ counterparts in glucose minimal medium.” ↩
15.
Cooper, Vaughn S., and Richard E. Lenski. “The population genetics of ecological specialization in evolving Escherichia coli populations.” Nature 407:736–739. 2000. A Nature letter. The authors tested 64 carbon sources on 12 lines of E. coli that had evolved on glucose for up to 20,000 generations, and asked whether the losses of unused abilities came from selection or from mutations piling up. The article uses it for the decline in the range of foods the bacteria could still use. ↩
1. 15a.
  Page 737: “Sixteen of the 64 informative substrates showed parallel decay in the first 10,000 generations. ... In fact, nine substrates showed parallel decay after only 2,000 generations, at which time none of the populations had yet evolved a mutator phenotype. Average total catabolic function declined by 32% (from 1 to 0.68) in the first 10,000 generations, but declined only 15% (from 0.68 to 0.58) between 10,000 and 20,000 generations.” ↩ ↩
2. 15b.
  The paper's logic: antagonistic pleiotropy (trade-offs) predicts parallel losses concentrated early, when adaptation is fastest; mutation accumulation predicts scattered, non-parallel losses accumulating steadily. The observed pattern matches the former. Note that the argument is about which process dominates, not about whether the losses are real. ↩
16.
Tenaillon, Olivier et al. “Tempo and mode of genome evolution in a 50,000-generation experiment.” Nature 536(7615):165–170. 2016. A genome study of 264 complete genomes from the 12 E. coli populations, sampled at intervals up to 50,000 generations. It counted every kind of mutation, and found 14,572 point mutations and six populations that became hypermutable and carried 96.5% of them. The article uses it for its ledger of mutations, which counts deletions and genome shrinkage, and for the hypermutable populations. ↩
1. 16a.
  “In total, we found 14,572 point mutations; 500 insertions of IS (insertion sequence) elements; 726 deletions and 1132 insertions each ≤50 bp (small indels); and 267 deletions and 45 duplications each >50 bp (large indels).” ↩
2. 16b.
  “After 50,000 generations, average genome length declined by 63 kbp (~1.4%) relative to the ancestor.” ↩ ↩
3. 16c.
  “six populations (Ara–1, Ara–2, Ara–3, Ara–4, Ara+3 and Ara+6) had 96.5% of the point mutations, having evolved hypermutable phenotypes caused by mutations that affect DNA repair or removal of oxidized nucleotides.” In the discussion: “six populations evolved hypermutable phenotypes that increased point-mutation rates ~100-fold, and another evolved hypermutability caused by a transposable element.” ↩ ↩
4. 16d.
  “In population Ara+1, 31.8% of all mutations through 50,000 generations were IS150 insertions, compared with 12.3% for the other populations that never evolved elevated point-mutation rates.” ↩
5. 16e.
  “Hypermutability tended to decline over time as the load of deleterious mutations favoured antimutator alleles. All four populations that were hypermutable at 10,000 generations accumulated synonymous substitutions ... between generations 40,000 and 50,000 at much lower rates than from 10,000 to 20,000 generations.” ↩ ↩
6. 16f.
  “insertions and deletions ... comprise almost half the mutations in nonmutator clones at 50,000 generations.” ↩
7. 16g.
  “Fifty-seven genes had two or more mutations; these genes had 50.1% of the nonsynonymous mutations but constituted only 2.1% of the coding genome. (Only one gene had multiple synonymous changes.)” On the mode of adaptation: “Nonsynonymous mutations accumulated ~17.1 times faster than synonymous ones during the first 500 generations and ~3.4 times faster over 50,000 generations.” ↩
17.
Behe, Michael J. “Experimental Evolution, Loss-of-Function Mutations, and 'The First Rule of Adaptive Evolution'.” The Quarterly Review of Biology 85(4):419–445. Dec 2010. A review of evolution experiments on microbes over four decades, and the paper that introduced Behe's scheme for sorting adaptive mutations into gains, losses and modifications of molecular function. Its abstract says that by far the most common adaptive changes in those experiments come from the loss or modification of a pre-existing molecular function. The article uses it for Behe's scheme, his test for what would count as a gain, and his saturation argument. ↩
1. 17a.
  “As seen in Table 5, several laboratory studies have achieved thousand- to million-fold saturations of their test organisms with point mutations, and most of the studies reviewed here have at least single-fold saturation. Thus, one would expect to have observed simple gain-of-FCT adaptive mutations that had sufficient selective value to outcompete more numerous loss-of-FCT or modification-of-function mutations in most experimental evolutionary studies, if they had indeed been available.” This is the strongest argument in the paper and the one this article's earlier draft did not answer. ↩
2. 17b.
  On artificiality: “It is true that the laboratory is an artificial environment ... Nonetheless, results arguably similar to those that have been seen in laboratory evolution studies to date have also been seen in nature, such as the loss of many genes by Yersinia pestis ... and the loss-of-FCT mutations that have spread in human populations in response to selective pressure from malaria.” On timescale: “Although it is true that new complex gain-of-FCT mutations are not expected to occur on short time-scales, the importance of experimental studies to our understanding of adaptation lies elsewhere. ... If such adaptive mutations also arrive first in the wild, as they of course would be expected to, then those will also be the kinds of mutations that are first available to selection in nature.” ↩
3. 17c.
  Page 439: Behe's own framing of the rule, which critics of it rarely quote: “It is called a 'rule' in the sense of being a rule of thumb. It is a heuristic, useful generalization, rather than a strict law; other circumstances being equal, this is what is usually to be expected.” He says the same in Darwin Devolves. The article criticises Behe for claiming more than the data support, and it records where he does not. ↩
4. 17d.
  Page 429: Written before Blount et al. 2012 identified the mechanism: “If the phenotype of the Lenski Cit+ strain is caused by the loss of the activity of a normal genetic regulatory element, such as a repressor binding site or other FCT, it will, of course, be a loss-of-FCT mutation, despite its highly adaptive effects in the presence of citrate. If the phenotype is due to one or more mutations that result in, for example, the addition of a novel genetic regulatory element, gene-duplication with sequence divergence, or the gain of a new binding site, then it will be a noteworthy gain-of-FCT mutation.” ↩
5. 17e.
  Page 421: The category definition: “A 'gain-of-FCT' adaptive mutation is a mutation that produces a specific, new, functional coded element while adapting an organism to its environment. The construction by mutation of a new promoter, intron/exon splice site, or protein processing site are gain-of-FCT mutations. Also included in this category is the divergence by mutation of the activity of a previously duplicated coded element.” Behe also classifies a single amino-acid substitution as gain-of-FCT — sickle haemoglobin, for the gained ability of the molecule to adhere to itself — so the scheme demonstrably admits gains, and admits them from point mutations. ↩
18.
Lenski, Richard E. “Is the LTEE breaking bad?” Telliamed Revisited. Feb 26, 2019. A blog post by Richard Lenski, who has run the experiment in his lab for over 30 years, and the third post in his series answering Behe's Darwin Devolves. It argues that Behe reads the experiment in a biased way, since Behe counts a newly evolved function as a gain only when a new gene appears. The article uses it for Lenski's own statements that the experiment was not built to test new functions and that its simple flasks make a poor case for evolution in general. ↩
1. 18a.
  “The LTEE was designed (intelligently, in my opinion!) to be extremely simple in order to address some basic questions about the dynamics and repeatability of evolution, while minimizing complications. It was not intended to mimic the complexities of nature, nor was it meant to be a test-bed for the evolution of new functions.” ↩ ↩
2. 18b.
  “Indeed, the LTEE environment is so extremely simple that one might reasonably expect the bacteria would evolve by breaking many existing functions. That is because the cells could, without consequence, lose their abilities to exploit resources not present in the flasks ... And just as one would expect, the bacteria have diminished or lost various abilities during the LTEE.” And: “The LTEE fits the bill for Behe because it's just about the best case possible to showcase his rule. But just as loss of sight in cave-dwelling organisms is a special case that won't tell us how eyes evolved, one must be careful when extrapolating from this experiment to evolution writ large.” ↩ ↩
3. 18c.
  “if one assumes the ancestral state of a gene is perfect, then there's no room for improvement in its function, and the only possible functional changes are degradative. ... that assumption breaks down when an organism encounters a new environment, where the optimal state of a protein might differ from what it was before.” ↩
19.
Couce, Alejandro et al. “Changing fitness effects of mutations through long-term bacterial evolution.” Science 383(6681):eadd1417. 2024. A research article that inserted DNA fragments into genes of the LTEE ancestor and of evolved clones, then measured the fitness effect of hundreds of thousands of insertions in bulk competitions over 50,000 generations. An insertion inside a gene usually breaks it, so it works as a knockout. It found that harmful mutations barely changed in share while helpful ones fell sharply, and that many helpful changes were knockouts; the article uses it for that count, and Richard Lenski, who directs the LTEE, is a co-author. ↩ ↩
1. 19a.
  “We quantified the effects of hundreds of thousands of insertion mutations in 12 populations of Escherichia coli through 50,000 generations of experimental evolution.” ↩
2. 19b.
  The article quotes this in its text. The classification rule is stated in the same paragraph: “we assumed that nonsense, frameshift, deletions, and insertions cause losses of function.” The authors note the same pattern in an independent system: “another study with Methylobacterium extorquens adapted to use methanol as the sole carbon source also found that most early beneficial mutations appear to disrupt functions.” They call it the. “coupon-collecting”. model: cheap, rough-and-ready knockouts first, refinements later. ↩ ↩
3. 19c.
  On population Ara+2: “the fraction of beneficial insertion mutations is substantially larger in the ancestor than in the evolved backgrounds (6.8% for ancestor [Anc] versus 4.3% and 3.2% for 2K and 15K, respectively; p < 0.044 both cases ...). In contrast ... the deleterious fraction is essentially constant across the three backgrounds (20.5% for Anc versus 21.0% and 19.6% for 2K and 15K ...).” Replicated in Ara−1, an independently evolved lineage. Section Abstract: “the fraction of deleterious mutations hardly changed over time, while the beneficial tail declined sharply, approaching an exponential distribution.” ↩
4. 19d.
  Section Conclusions and Discussion: “other types of mutations occur in the LTEE that might have more complex effects. For example, point mutations and structural rearrangements may be more likely to generate gains or changes of function, which could lead to more unpredictable outcomes, as seen with the evolution of citrate utilization in one of the 12 LTEE lines.” ↩
20.
Behe, Michael J. Darwin Devolves: The New Science About DNA That Challenges Evolution. HarperOne. 2019. A 2019 book whose author argues that the experiment's beneficial mutations very likely worked by degrading genes. Chapter 7 is the one on the experiment, and the article uses that chapter to test the Discovery Institute's case. ↩
1. 20a.
  Chapter 7: “Even before the critical mutation occurred, a different mutation in a gene for a protein that makes citrate in E. coli degraded the protein's ability to bind another metabolite, abbreviated NADH, which normally helps regulate its activity. Another, later, mutation to the same gene decreased its activity by about 90 percent. ... Further computer analysis by the authors suggested that the citrate mutant would be even more efficient if two other metabolic pathways that were normally turned off were both switched on. They searched and discovered that two regulatory proteins that usually suppress those pathways had been degraded by point mutations; the traffic lights were now stuck on green.” Behe's "about 90 percent" is his gloss on a roughly tenfold reduction in transcript level, which is fair. ↩
2. 20b.
  Chapter 7: Endnote 29: “The citrate mutation can be generated in weeks rather than decades by altering the bacterial growth regimen; D. J. Van Hofwegen, C. J. Hovde, and S. A. Minnich, 'Rapid Evolution of Citrate Utilization by Escherichia coli by Direct Selection Requires citT and dctA,' Journal of Bacteriology 198 (2016): 1022-34.” ↩
3. 20c.
  Chapter 7, section Objections to the Rule: “it's not so much the rarity of constructive mutations that undermines Darwinian evolution—it's the frequency of damaging but helpful ones. Degradative but adaptive loss-of-FCT or modification-of-function mutations appear quickly even on short time scales, even in small populations. ... Damaging yet beneficial mutations will rapidly be selected when nothing else is available and compete fiercely with any gain-of-FCT mutations that might eventually arrive on the scene.” ↩ ↩
4. 20d.
  Chapter 7: “After fifty thousand generations of the most detailed, definitive evolution experiment ever conducted, after so much improvement of the growth rate that descendant cells leave revived ancestors in the dust, after relentless mutation and selection, it's very likely that all of the identified beneficial mutations worked by degrading or outright breaking the respective ancestor genes.” ↩ ↩ ↩
5. 20e.
  Chapter 7: On the citrate mutation: “In my QRB classification scheme the mutation would be counted as modification-of-function—because no new functional coded element was gained or lost, just copied.” Two paragraphs later: “Interesting as it is, the ambiguous citrate mutation that started the hoopla is a sideshow. The overwhelmingly important and almost completely unnoticed lesson is that genes are being degraded left and right.” ↩ ↩ ↩
6. 20f.
  Page 609: Endnote 15 to chapter 7: “More recent investigation by Lenski's lab suggests that mutations in a small minority (10 of 57) of selected E. coli genes may not completely break them but rather, as they put it, 'fine-tune' them (probably by degrading their functions): 'In some cases, adaptive mutations appear to modify protein functions, rather than merely knocking them out'”. The endnote is followed by a full and accurate citation of Maddamsetti et al. Behe read the paper and reproduced its claim; the whole of his counter-argument is the four words in the parenthesis. ↩ ↩
7. 20g.
  Chapter 7: “Later the researchers noticed that in one of the mutator cell lines the mutation rate had decreased by half, to about 75 times higher than normal. When they tracked down the cause of the modest improvement, it turned out to be a damaging mutation to a different protein, one that normally helps the cell metabolize nucleotides. So the unhelpful effects of damaging one gene were partially offset by a mutation that damaged a different gene. That's what random mutation does.” Behe's "150 times greater than normal" is not an error and not in conflict with the ~100-fold reported elsewhere: his endnote cites Wielgoss et al. 2013, which measured that figure for the mutT allele in Ara–1, whereas Sniegowski 1997 and Tenaillon 2016 report ~100-fold for mismatch-repair defects in other lineages. Different genes, different populations, different numbers. He does swap the two gene descriptions: it is mutT that handles nucleotides and mutY that is the repair glycosylase, not the other way round. ↩
8. 20h.
  Chapter 7: “the lineage of bacteria at Michigan State has surpassed sixty-five thousand generations, which is equivalent to over a million years in the history of a large animal species like humans.” ↩
21.
Maddamsetti, Rohan et al. “Core Genes Evolve Rapidly in the Long-Term Evolution Experiment with Escherichia coli.” Genome Biology and Evolution 9(4):1072–1083. 2017. A research paper co-written by Lenski that compared about 2,000 core genes, shared by 60 natural and laboratory strains of E. coli, with the genes that changed fastest in the experiment. It found that those fast-changing genes are conserved in nature, and that some adaptive mutations seem to tune protein function rather than knock it out. Behe cites it in endnote 15 to chapter 7 of Darwin Devolves. The article uses it as counter-evidence to Behe's claim that the experiment's beneficial mutations break genes, because only five of the 57 fastest-changing genes have nonsense mutations. ↩
1. 21a.
  “The 16 genes with positive essentiality scores are less likely to have been impacted by possible knockout mutations than the 40 genes with negative scores (Fisher's exact test: one-tailed P = 0.0297). ... Of the genes with positive essentiality scores, only topA and mrdA had any possible knockout mutations in any of the sequenced nonmutator genomes. The candidate knockout in topA is a small indel found in only one genome ... this small indel in topA probably does not destroy the gene's function.” ↩ ↩
2. 21b.
  “only 5 of the 57 genes, all with negative essentiality scores (yijC, malT, yabB, ybaL, and yeeF), have nonsense mutations in any of the nonmutator genomes ... This line of evidence therefore supports the hypothesis that some of the beneficial mutations in the LTEE modulate protein function, rather than knocking it out.” ↩
3. 21c.
  “Ten of the 57 genes that showed gene-level parallel evolution also show parallelism at the amino-acid level.” The specific cases: infC R132C and rpsD D50G, which. “clearly maps to their interface”. ; nadR Y294C, on the alpha-helix face that interacts with NAD; pykF A301S, at. “the A/A' multimerization interface of the pykF tetramer ... which is implicated in allostery.” The authors are candid that pykF knockouts are also beneficial, so the gene supports both readings. ↩ ↩
22.
Quandt, Erik M. et al. “Fine-tuning citrate synthase flux potentiates and refines metabolic innovation in the Lenski evolution experiment.” eLife 4:e09696. 2015. A genetics study of the gltA gene, which makes the enzyme citrate synthase, in the one population that learned to use citrate. It found that a first gltA mutation raised the enzyme's activity and helped keep the early citrate trait from dying out, and that a second wave of mutations later lowered the activity below the original level. The article uses it for the steps that came before and after the citrate trait. ↩
1. 22a.
  Section Abstract: “The first gltA mutation, which increases CS activity by disrupting NADH-inhibition of this enzyme, is beneficial for growth on the acetate and contributed to preserving the rudimentary Cit+ trait from extinction when it first evolved. However, after Cit+ was refined by further mutations, this potentiating gltA mutation became deleterious to fitness. A second wave of beneficial gltA mutations then evolved that reduced CS activity to below the ancestral level.” ↩ ↩
2. 22b.
  “it seems likely that both the evolved iclR and arcB alleles are loss-of-function mutations that derepress expression of enzymes needed for acetate assimilation.” On their reach, the authors wrote: “arcB/arcA mutations have been found in 11/12 of the LTEE populations (Plucain et al., 2014).” ↩
3. 22c.
  “when citrate is the sole carbon source, FBA predicts optimal growth when there is no flux through CS. In fact, any CS activity is detrimental.” FBA is flux-balance analysis, a constraint-based model of the whole metabolic network. ↩
4. 22d.
  Section Results: “Most secondary citrate synthase mutations reduce enzyme activity.” On the mechanism of one of them: “addition of the gltA2-R mutation reduced gltA mRNA levels by approximately tenfold, whether or not the gltA1 mutation was also present.” Section eLife digest: The digest states the endpoint: “the bacteria acquired a second gltA mutation that lowered citrate synthase activity to a level below what it had been in the original bacteria before the first gltA mutation.” ↩ ↩
5. 22e.
  Figure 4: The figure legend says: “these competitions show that the citT mutation would be highly deleterious to fitness if it evolved in a genetic background without the gltA1 mutation. In contrast, the citT mutation is neutral or possibly slightly beneficial to fitness when the gltA1 mutation is present.” The text says: “Epistatic interactions between gltA1 and the citT mutation in this evolved genetic background prevented a massive fitness defect that would have almost certainly led to the rapid extinction of any newly evolved Cit+ cells before this rudimentary trait could be refined to the advantageous Cit++ phenotype by further mutations.” ↩
23.
Lenski, Richard E. et al. “Sustained fitness gains and variability in fitness trajectories in the long-term evolution experiment with Escherichia coli.” Proceedings of the Royal Society B 282:20152292. 2015. A research article that ran more than 1,100 new competitive fitness assays on LTEE populations out to 60,000 generations, to test whether a power-law model of fitness gain predicts later fitness better than a hyperbolic model, which has a hard upper limit. Lenski, who leads the LTEE, is its first author. The article uses it as a prediction made in advance and tested against new data. ↩
1. 23a.
  Section Methods: “Our analyses used whole-population samples from nine of the 12 LTEE populations taken at three time points: 40 000, 50 000 and 60 000 generations ... We excluded three populations (Ara–2, Ara–3 and Ara+6) that no longer make colonies that can be reliably counted in the standard fitness assays or that evolved the ability to consume the citrate in the culture medium, which also precludes using the standard assays. Of the nine populations used in our study, three evolved hypermutable phenotypes: Ara–1, Ara–4 and Ara+3.” So the strongest fitness claim available is nine populations through 60,000 generations, not twelve through 82,000. The exclusion of Ara–3 is for a methodological reason, since Cit+ cells break the marker-based assay, rather than because its fitness fell. ↩ ↩
2. 23b.
  Page 3: “The probability that all nine populations would, by chance, yield monotonically increasing point estimates across three time points is …” The paper's figure is (1/3 × 1/2) to the ninth power, about 10⁻⁷. It goes on: “Six populations showed significant fitness gains over both component intervals, while three had significant gains in only one interval”. ↩
3. 23c.
  “The best fit of the hyperbolic model gave w = 1 + 0.7007 t/(t + 4431) ... This equation predicts grand-mean fitness values relative to the ancestor of 1.6308, 1.6437 and 1.6525 at 40 000, 50 000 and 60 000 generations, respectively. For the power-law model, the best fit yielded w = (0.00515 t + 1)^0.0950, which predicts values of 1.6597, 1.6951 and 1.7246 at the same time points.” ↩ ↩
4. 23d.
  Section Abstract: “Here, we examine whether the previously estimated power-law model predicts the fitness trajectory for an additional 10 000 generations. To that end, we conducted more than 1100 new competitive fitness assays. Consistent with the previous study, the power-law model fits the new data better than the hyperbolic model. ... our results imply that both adaptation and divergence can continue indefinitely—or at least for a long time—even in a constant environment.” ↩ ↩
24.
Sanford, John C. Genetic Entropy. FMS Publications. 4th edition. 2014. A book by a plant geneticist formerly at Cornell, arguing that harmful mutations too slight for selection to see build up in every genome, so that lineages decline over time. This edition adds a chapter on new evidence with a section on the LTEE. The article uses that section as the young-earth position on the experiment in its author's own words. ↩
1. 24a.
  Chapter 11, section 2b, Reexamination of the famous Lenski experiment: “Over a period of a few decades, a strain of E. coli bacteria became slightly better-adapted to an artificial nutrient medium. This is not unexpected or remarkable in any way. … The Lenski experiment needs to run many centuries before we can expect to see extreme manifestations of genetic entropy”. The next paragraph: “we see in every case, that the adaptation was accomplished by means of either loss-of-function or loss-of-regulation mutations (and hence loss of information) … the silencing of these temporarily "expendable" genes and the elimination of all non-relevant functions will continue indefinitely until the bacterial genome is stripped down to its minimal functionality. … Over hundreds of years, Lenski's "evolving" strains of bacteria must logically go extinct (if they could be maintained in the artificial environment that long). Nearly-neutral deleterious mutations will inevitably cause the bacterial genome to "rust out"”. ↩ ↩ ↩ ↩ ↩
25.
Behe, Michael J. The Edge of Evolution: The Search for the Limits of Darwinism. Free Press. 2007. A book by the biochemist who later wrote Darwin Devolves, arguing that mutation and selection readily build features reachable by single beneficial steps but rarely build ones that need several unselected changes at once. It estimates that limit from the records of malaria, HIV and Lenski's E. coli, and chapter 7 uses the LTEE, at about 10¹³ cells, as one of its four data points. The article uses it for what Behe said about the experiment before the citrate result. ↩
1. 25a.
  Chapter 7, section Lab Star: “Breaking some genes and turning others off, however, won't make much of anything. After a while, beneficial changes from the experiment petered out.” His endnote 26 gives the source, de Visser and Lenski 2002, and quotes it: “Their rate of improvement, as measured in competitions with the ancestor in that environment, has declined substantially over this period”. The source reports a slowing, not a stop. ↩ ↩
26.
Berean Archive. In-house calculation. n.d. A calculation written for this article, not a published source, of what Sanford's nearly-neutral mechanism can do at the LTEE's population size by drift alone. In a haploid population a mutation behaves as neutral when its selection coefficient is smaller than about 1/Ne; at Ne ≈ 10⁷ that is 10⁻⁷. At 6.64 generations a day, a thousand years is about 2.42 × 10⁶ generations. If every new mutation were harmful and sat just under that line, a non-mutator genome at 4.12 × 10⁻⁴ mutations per generation would gain about 1,000 of them per thousand years, costing at most 1,000 × 10⁻⁷ = 10⁻⁴ of fitness, or 0.01%. A 150-fold mutT mutator, at about 0.062 per generation, would lose at most about 1.5%. Mutations with larger effects are seen by selection and removed, except where linkage carries them along with a beneficial sweep, which this calculation leaves out and which is the forecast's real support. Sanford's own published runs use small populations and high mutation counts per generation; see the genetic entropy article for how the result depends on those two numbers. ↩ ↩
27.
uz-Zaman, Md Hassan et al. “Promoter recruitment drives the emergence of proto-genes in a long-term evolution experiment with Escherichia coli.” PLoS Biology 22(5):e3002418. 2024. A research article that searched the experiment's RNA and ribosome-profiling data for DNA that was silent in the ancestor but has started to be transcribed, which the authors call proto-genes. Of 63 cases of new transcription at 50,000 generations, the authors narrowed the field to nine proto-genes, and most proto-genes came from mutations that recruit an existing promoter. The article uses it for the nine proto-genes, the estimated rate of about one per 60,000 generations, and the authors' own admission that none has been shown to affect fitness. ↩
1. 27a.
  Section Abstract: “Proto-genes formed downstream of new mutations result either from insertion element activity or chromosomal translocations that fused preexisting regulatory sequences to regions that were not expressed in the LTEE ancestor.” From 63 statistically significant cases of new transcription at 50,000 generations the authors narrowed to 19 unambiguous candidates and then to nine proto-genes. Section Results: “For 5 of the 9 proto-genes, new transcriptional activity could be attributed to the activity of insertion sequence IS150 ... a 1,443-bp transposable element which carries an outward-facing promoter that can trigger downstream transcription.” ↩
2. 27b.
  “The longest ORF (117 aa) is homologous to hypothetical proteins in E. coli and Shigella sonnei, indicating that it has gene-like features as recognized by standard prokaryotic annotation ... Since none of the other proto-genes were associated with Ribo-seq reads, this ORF is the only example of a novel ribosome-associated transcript that we identified in the present study.” ↩
3. 27c.
  “we estimate the rate of proto-gene emergence to be about 1 event per 60,000 generations, with some appearing as early as 4,000 generations.” ↩
4. 27d.
  “Future experimental studies will be required to establish whether the novel transcripts and peptides we detected affect E. coli fitness or are byproducts of mutations that are beneficial because of their effects on nearby, existing genes.” Nothing in the paper claims a function for any of the nine. ↩ ↩
28.
Roth, John R., and Sophie Maisnier-Patin. “Reinterpreting Long-Term Evolution Experiments: Is Delayed Adaptation an Example of Historical Contingency or a Consequence of Intermittent Selection?” Journal of Bacteriology 198(7):1009–1012. 2016. An invited commentary published alongside Van Hofwegen et al. in the same issue, not a new experiment. Its authors are John R. Roth, a Salmonella geneticist at UC Davis with no connection to the design movement, and Sophie Maisnier-Patin. They argue that the 15-year delay in Lenski's experiment may come from short selection periods that rarely let mutants survive the transfers, rather than from a rare earlier mutation that had to come first. The article uses it because the agreement comes from a researcher with no connection to the design movement. ↩
1. 28a.
  Page 1009: “Thus, the delay may not reflect the necessity of a nonselective potentiation event (the 'historical contingency') but may rather be due to selective events with a low probability of serial transfer.” ↩
2. 28b.
  Page 1011: “The transitory nature of simple amplifications may explain why the posited simple duplications were not apparent in sequences of cells from early tubes from the LTEE—they may have been lost during nonselective growth prior to sequencing.” And on how one event can look like the other: “Amplification of this region would also provide multiple substrates for a deletion event that fuses an active promoter to the 'silent' citT gene. This rare event could also reflect a de novo duplication event within one of the initial repeats.” ↩
3. 28c.
  Page 1009: “More importantly, the authors report that the Cit+ variants appearing rapidly in batch conditions are generated by essentially the same series of events as those described by Blount et al. (potentiation, actualization, and refinement).” ↩
4. 28d.
  Page 1010: “However, dense cell cultures are likely to have low oxygen levels. This makes it likely that the citrate utilization genes may be expressed at least slightly in stationary phase and that citrate usage could be improved by increasing the dosage of the entire cit operon, including the critical citT gene.” This is offered as a hypothesis; it is not measured in either paper. ↩
29.
Minnich, Scott A., and Carolyn J. Hovde. “Evolutionary poker lacks a full deck when modelling the LTEE Cit+ phenotype.” The Journal of Physiology 603(5):1325–1328. 2025. A letter to the editor, the authors' reply to a 2024 agent-based model (a computer simulation that follows individual cells) of Lenski's experiment by Root-Bernstein and Bernstein. Its authors are Scott A. Minnich and Carolyn J. Hovde of the University of Idaho, whose 2016 lab work the model cites; they report bacteria able to use citrate (Cit+) arising after 12 to 100 generations, and they multiply two estimated rates to get a combined probability of about 1 in 10¹⁴ for the two changes Lenski's strain needed. The article uses it for that range, the phage transfer result and the restated numbers argument. ↩
1. 29a.
  Page 1325: “Finally, using phage transduction, we show that the Cit+ mutation can be transferred into wild-type E. coli K12 without prior citrate exposure, genetic proof that the aerobic Cit+ phenotypes do not require contingent single nucleotide polymorphism accumulation.” On the same page they reject Leon et al.'s framing directly: “Lenski and others argue that our results using both a direct selection in citrate medium and delayed transference of cultures in low glucose + citrate medium are irrelevant regarding acquisition of the Cit+ phenotype in the LTEE (e.g. see discussion of Leon et al., 2018, pp. 12-13). We argue otherwise.” ↩
2. 29b.
  Page 1325: “We isolated multiple independent Cit+ mutants that arose between 12 and 100 generations using a direct selection with citrate as the sole carbon source ... This is because extended time in stationary phase with citrate allows cells with citT amplifications to increase in number and thus increase the probability of a promoter capture.” The range and the mechanism are given together; the compression to the floor value happens downstream. ↩
3. 29c.
  Page 1326: The article quotes this passage in full in its body. Their internal citation is. “Pennesi, E. (2013)”. a misspelling of Elizabeth Pennisi's Science news feature "The man who bottled evolution," which is the source of the ~10¹⁴ cell figure. ↩ ↩ ↩
4. 29d.
  Figure 1: The legend says: “in E. coli B DcuS is defective and cannot phosphorylate DcuR ... preventing expression of DctA that is required for C4-dicarboxylate uptake. Overcoming the lack of the DcuSR two-component system requires a point mutation of the dctA promoter. This took 15 years (33,000 generations) to occur in LTEE and only happened in one of the 12 populations.” ↩
5. 29e.
  Page 1326: From the closing paragraph: “Had he chosen E. coli K12, rather than E. coli B (Lenski's starting strain which lacks a functional key gene in its genetic deck), it is likely that all 12 flasks would rapidly have become Cit+ and we would not be arguing about contingency and the requirement of rare and necessary non-repeatable antecedent mutations.” ↩
30.
Berean Archive. In-house argument. n.d. A joint probability obtained by multiplying 10⁻⁵ by 10⁻⁹ is the probability of both events occurring in one cell without selection between them. If the first event carries a lasting selective advantage, its carriers multiply before the second is needed and the correct calculation is sequential, which gives a very much larger number. Van Hofwegen et al. are consistent on their own terms because they argue the LTEE's daily dilution made the first step effectively unselectable there; Leon and Barrick agree that the same step would be. “universally beneficial in any genetic background”. under direct selection. The disagreement between the two sides is about which regime the LTEE resembles, not about arithmetic. An argument written for this article, not a published source, about two claims that are often quoted at each other. The article uses it to show that both sides have changed position on how potentiation works. ↩
31.
NCBI, GenBank. “Escherichia coli B str. REL606, complete sequence.” NCBI nucleotide database. n.d. The complete genome of REL606, the ancestral strain of the experiment, held under GenBank accession NC_012967.1. It is 4,629,812 bp long and was sequenced by Jeong et al. 2009. Wielgoss et al. 2011 name it as the ancestral reference genome, and the article uses its length in its table of figures for the experiment. ↩
32.
Lenski, Richard E. “Revisiting the Design of the Long-Term Evolution Experiment with Escherichia coli.” Journal of Molecular Evolution 91:241–253. 2023. “began in 1988”. “recently reached 75,000 generations”. A 2023 journal article by Richard Lenski that revisits the design of the experiment. The article uses it for the experiment's 1988 start and its count of 75,000 generations. ↩
33.
Lenski, Richard E. “Time to restart the LTEE, this virus be damned.” Telliamed Revisited. Apr 1, 2020. “each population produces around half a billion new cells each day.” A blog post by Lenski, an April Fools' piece. The post also records that the LTEE lines were frozen on 9 March 2020 for the pandemic, the only significant interruption in the experiment's history. It states one design figure in passing, which can be derived independently from Lenski et al. 1991, and the article uses it only for that daily cell count. ↩
34.
Good, Benjamin H. et al. “The Dynamics of Molecular Evolution Over 60,000 Generations.” Nature 551(7678):45–50. 2017. Whole-population metagenomic sequencing of the 12 E. coli populations at 500-generation intervals through 60,000 generations. It found several beneficial mutations competing inside each population at once, and nine of the twelve populations have clades that coexist for more than 10,000 generations. Lenski is a co-author. The article uses it for two points: beneficial mutations compete rather than combine in a population with no sex, and selection, not drift, must have raised the mutations it tracked. ↩
1. 34a.
  “With an effective population size of Ne ~ 10⁷, new mutations would require Δt ~ 0.1 Ne ~ 10⁶ generations to reach the 10% detection threshold by genetic drift alone. Thus, the mutations in Fig. 1 must have reached observable frequencies through the direct or indirect action of natural selection.” The same section gives the non-mutator accumulation rate: “decreasing from ~20 mutations in the first 10,000 generations to ~10 mutations in the last 10,000.” ↩ ↩
2. 34b.
  Section Abstract: “molecular evolution continues to be characterized by signatures of rapid adaptation, with multiple beneficial variants simultaneously competing for dominance in each population.” ↩
3. 34c.
  “9 of the 12 populations have clades that coexist for >10,000 generations, often persisting through generation 60,000. ... long-term coexistence is likely maintained by negative frequency-dependent selection, as has been demonstrated in Ara–2.” ↩
35.
Izutsu, Minako et al. “Effects of periodic bottlenecks on the dynamics of adaptive evolution in microbial populations.” Microbiology 170:001494. 2024. Section Abstract: “Adaptation began earlier and fitness gains were greater with 100- and 1000-fold dilutions than with 8-fold dilutions, consistent with the simulations when beneficial mutations are common.” The impact statement draws the conclusion the article uses: Section Impact statement: “imply a high rate of beneficial mutation that generates intense competition between mutant lineages.” A computer simulation study paired with an experiment in which 48 populations of E. coli were propagated for 150 days, with daily dilutions of 2-, 8-, 100- or 1000-fold. The authors ask which dilution speeds adaptation most, and find that the answer depends on how common beneficial mutations are. The article uses it as the LTEE group's own test of the transfer regime, and Richard Lenski, who leads the LTEE, is a co-author. ↩
36.
Behe, Michael J. “Rose-Colored Glasses: Lenski, Citrate, and BioLogos.” Evolution News & Views. Nov 13, 2012. A blog essay by the author of The Edge of Evolution, written after Blount's 2012 genome paper, replying to the biologist Dennis Venema, who had argued at BioLogos that the citrate result crossed Behe's proposed limit. Behe sets out how his limit counts mutations, counts the citrate steps himself, and classes them all as modification of function. The article uses it as Behe's own statement of what the citrate result does and does not test. ↩
1. 36a.
  Paragraph 12: “In arriving at the limits to Darwinism, I emphasized that that mechanism would certainly work if gradually-increasing, serial, beneficial mutations could do the job. Thus such mutations do not count in estimating the limit. Only required deleterious and neutral mutations count against the limit to Darwinian evolution.” ↩
2. 36b.
  Paragraph 12: “Venema counts the number of mutations needed to get a fully functioning citrate-importing function in Lenski's work, and arrives at roughly a half-dozen.” Venema's own count is cited here as Behe reports it. ↩
3. 36c.
  Paragraph 13: “When counting the mutations contributing to the edge of evolution for building a feature, only the ones directly involved in the feature are counted, not ones that indirectly contribute to a receptive genetic background (which are legion). Thus, unlike Venema, I count perhaps three to four mutations — the original duplication placing the oxygen-tolerant promoter near the citT gene, plus several rounds of duplication of that region.” ↩
4. 36d.
  Paragraph 14: “If, before Lenski's work was done, someone had sketched for me a cartoon of the original duplication that produced the metabolic change, I would have assumed that would be sufficient — that a single step could achieve it. The fact that it was considerably more difficult than that goes to show that even skeptics like myself overestimate the power of the Darwinian mechanism.” ↩
37.
Batten, Don. “Bacteria 'evolving in the lab'? 'A poke in the eye for anti-evolutionists'?” Creation Ministries International. Jun 13, 2008. A young-earth reply to Blount 2008 by a Creation Ministries International scientist, with an addendum written in March 2016 after Blount's 2012 genome paper and a note on Van Hofwegen 2016. In 2008 it guessed that the trait came from broken regulation of the existing transporter; the addendum scores that guess against the mechanism. The article uses it because Carter's widely read summary cites it for the "loss of control" description. ↩
1. 37a.
  Section Addendum (prepared March 2016): “This is interesting research, but there is nothing in it that supports microbes-to-man evolution; it falls far short. As I have said, there is nothing here that is beyond the 'edge of evolution' that Michael Behe described in his book on the topic.” ↩
2. 37b.
  Section Addendum (prepared March 2016): “While the existing controls and the citrate transporter gene were not actually broken, the transporter gene was duplicated (copied) to a different location away from the existing control sequence such that the production of the transporter is no longer suppressed in the presence of oxygen. … So the ability of the cell to control the production of the citrate transporter was indeed broken (the cell is no longer able to turn off the production of the transporter).” The same addendum scores its 2008 guess: “it turned out to be pretty close to what actually happened, which shows that creation thinking makes for good scientific predictions.” ↩
38.
Purdom, Georgia. “A Poke in the Eye?” Answers in Depth 3. Answers in Genesis. Jun 30, 2008. A reply by a molecular geneticist at Answers in Genesis to Blount 2008 and to Jerry Coyne's comment on it, written before the citrate mutation was identified. It argues that the trait is adaptation within a designed system, most likely by loss of regulation or specificity in an existing transporter. The article uses it as the young-earth prediction made before the mechanism was known. ↩
1. 38a.
  Section A Creationist Perspective: “Since E. coli already possess the ability to transport and utilize citrate under certain conditions, it is conceivable that they could adapt and gain the ability to utilize citrate under broader conditions. This does not require the addition of new genetic information or functional systems (there are no known "additive" mechanisms). Instead degenerative events are likely to have occurred resulting in the loss of regulation and/or specificity.” ↩ ↩
39.
Wile, Jay L. “Another Creationist Prediction Confirmed.” Goodreads author blog. May 16, 2016. A blog post by a young-earth creationist writer that reads Van Hofwegen, Hovde and Minnich's 2016 paper as confirming Purdom's 2008 article. The article uses it as the creationist claim that the citrate result was predicted. ↩
1. 39a.
  After the quotation from Purdom: “If Dr. Purdom is correct, it shouldn't be at all difficult to produce citrate-using Escherichia coli in an oxygen environment. Adaptive mutations are part of a designed mechanism that produces predictable results in the presence of environmental stress.” ↩
40.
Carter, Robert W. “Can mutations create new information?” Journal of Creation 25(2):92–98. Aug 2011 (edition of 2025). An article by a Creation Ministries International geneticist, first printed in the Journal of Creation in 2011 and updated online in 2025, arguing that mutations can copy, shuffle and degrade existing information but cannot build radically new functions. The article uses it as the young-earth summary that files aerobic citrate use under degraded information. ↩
1. 40a.
  Section Gene duplication: “Even if an example of a new function arising through gene duplication is discovered, the function of the new must necessarily be related to the function of the old, such as a new but similar catalysis end product of an enzyme.” ↩
2. 40b.
  Section Degraded information: “Examples include sickle cell anemia (which confers a resistance to the malaria parasite by producing deformed hemoglobin molecules), aerobic citrate digestion by bacteria (which involves the loss of control of the normal anaerobic citrate digestion), and nylon digestion by bacteria”. His source for the citrate example is Batten's 2008 article. ↩
41.
Purdom, Georgia. “Bacteria Evolve "Key Innovation" or Not?” Answers in Genesis blog. Nov 8, 2012. The same author's follow-up after Blount's 2012 genome paper identified the duplication. It describes the change as a loss of regulation and stresses how many mutations it took. The article uses it for the young-earth reading of the mechanism once it was known. ↩
1. 41a.
  Paragraph 4: “It's also not an example of the gain of a new function because the bacteria could already utilize citrate under certain conditions. There was a definite loss of regulation as the original citT gene was only expressed under low oxygen levels.” ↩
2. 41b.
  Paragraph 5: “I was amazed at just how many changes were necessary to allow the bacteria to perform a pre-existing function (using citrate) under different conditions (normal oxygen levels). Imagine how many changes would be necessary to generate truly novel proteins with truly novel functions necessary for molecules-to-man evolution. Impossible!” ↩
42.
Woods, Robert et al. “Tests of parallel molecular evolution in a long-term experiment with Escherichia coli.” PNAS 103(24):9107–9112. 2006. A sequencing study of the twelve populations at 20,000 generations. The authors sequenced four candidate genes, each with a mutation of unknown effect found in one population, and compared them with 36 randomly chosen genes. Two candidate genes, pykF and nadR, had substitutions in all eleven other populations. The article cites it as the source Lenski names for parallel changes that were not obviously contingent on prior mutations. ↩
43.
Leiby, Nicholas, and Christopher J. Marx. “Metabolic Erosion Primarily Through Mutation Accumulation, and Not Tradeoffs, Drives Limited Evolution of Substrate Specificity in Escherichia coli.” PLoS Biology 12(2):e1001789. 2014. A growth study of the experiment's bacteria after 50,000 generations, which measured directly how fast they grew on a range of food sources and found that the losses were much less widespread than earlier respiration assays had suggested. The strains with higher mutation rates lost the most, which points to accumulated broken genes rather than trade-offs. It reaches the opposite conclusion to Cooper & Lenski 2000 about the cause of the decay across a wider set of food sources; neither paper disputes that the decay occurred, and the article includes it because the mechanism is contested. ↩
44.
Sniegowski, Paul D. et al. “Evolution of high mutation rates in experimental populations of E. coli.” Nature 387:703–705. 1997. Page 703: Of the twelve populations, Ara–2, Ara–4 and Ara+3 at 10,000 generations were traced to defects in mutS, uvrD and mutL. The paper says they. “displayed mutation rates that were between one and two orders of magnitude higher than those in the ancestor”. A Nature letter reporting that some lineages in the experiment had become mutators, strains that copy DNA with more errors because a repair gene is broken. The paper is often cited for a flat hundredfold rise in mutation rate, which is the top of its range; "~100-fold" in Tenaillon et al. 2016 is a different measurement, over six populations and 50,000 generations. The article uses it for the three populations that became hypermutators by 10,000 generations. ↩
1. 44a.
  Page 703: The paper's mechanism: a mutator allele is itself neutral or slightly deleterious, but rises by hitch-hiking with the beneficial mutations it generates. Its phrase for this is. “chance associations with adaptive mutations”. ↩
45.
Wielgoss, Sébastien et al. “Mutation rate dynamics in a bacterial population reflect tension between adaptation and genetic load.” PNAS 110(1):222–227. 2013. Section Abstract: “we dated the emergence of these mutations and inferred that the mutT mutation increased the point-mutation rate by ~150-fold, whereas the mutY mutations reduced the rate by ~40-60%, with a corresponding decrease in the genetic load. Thus, the long-term fate of the hypermutators was governed by the selective advantage arising from a reduced mutation rate as the potential for further adaptation declined.” A genome study of one population from the long-term experiment. The authors sequenced 22 evolved clones and measured mutation rates with fluctuation tests, which count resistant mutants to estimate how often mutations arise. They found that a mutT defect raised the point-mutation rate about 150-fold, and that two later lineages with mutY mutations lowered it again. The article uses it for population Ara–1, the one case whose rise and fall in mutation rate has been worked out. ↩
1. 45a.
  Section Results and Discussion: “Loss-of-function mutations in mutY, which encodes a DNA repair glycosylase that excises mis-paired bases from DNA helices, also lead to elevated mutation rates on their own. However, mutY mutations have an antimutator effect in the context of a MutT defect because MutY mis-repairs 8-oxoG:A" pairings.” So the brake is itself a broken gene, and it works only because the first gene is already broken. ↩
2. 45b.
  Section Results and Discussion: The authors say the mutY alleles may have reached. “high frequencies because they conferred an advantage. However, it is also possible that these mutY mutations were not adaptive, especially given the facts that (i) the mutT lineage had a greatly elevated mutation rate”. The second reason they give is that the extra mutations could have carried them along. Neither Behe nor an earlier draft of this article reported the hedge. ↩ ↩
46.
Couce, Alejandro et al. “Mutator genomes decay, despite sustained fitness gains, in a long-term experiment with bacteria.” PNAS 114:E9026–E9035. 2017. Section Abstract: “Our results show that, despite sustained adaptive evolution in the long-term experiment, the signature of selection is much weaker than that of mutational biases in mutator genomes. This finding suggests that relatively brief periods of hypermutability can play an outsized role in shaping extant bacterial genomes.” A genome study that asks whether natural selection or mutational bias better predicts which sites change in the mutator populations of the experiment. It compares those genomes with mutation-accumulation lines, where selection is extremely weak, and with natural isolates, where selection is strong. Lenski is a co-author, along with Tenaillon and Schneider. The article uses it as the strongest evidence from the experiment that mutator genomes decay, because mutational bias outweighs selection even when only essential genes are counted. ↩ ↩
1. 46a.
  Figure 5: “The selection-related DCA score strongly influences genomic evolution in nature, whereas mutational biases have greater influences in the MAE and LTEE populations, even when the analysis is restricted to essential genes.” DCA is direct-coupling analysis, an epistatic model that predicts the fitness effect of an amino-acid substitution from co-evolutionary constraints across homologous sequences. ↩
2. 46b.
  Section Abstract: “Overall, these results highlight the importance of genomic draft, in which strong linkage limits the ability of selection to purge deleterious mutations. These insights are also relevant to other biological systems evolving under strong linkage and high mutation rates, including viruses and cancer cells.” ↩ ↩
47.
Lents, Nathan H. et al. “The end of evolution?” Science 363:590. Feb 7, 2019. The review's paragraph on the experiment: “There are indeed many examples of loss-of-function mutations that are advantageous, but Behe is selective in his examples. He dedicates the better part of chapter 7 to discussing a 65,000-generation Escherichia coli experiment, emphasizing the many mutations that arose that degraded function—an expected mode of adaptation to a simple laboratory environment, by the way—while dismissing improved functions and deriding one new one as a 'sideshow.'” A book review in Science of Michael Behe's Darwin Devolves, written by three scientists, one of them Richard Lenski, whose lab published the findings the review discusses. The review says Behe picks his examples selectively, and it discloses Lenski's role in those findings. The article uses it as a short statement, by a co-author of the experiment, of the complaint that Behe passes over the experiment's improved functions. ↩
48.
Berean Archive. In-house calculation. n.d. Cells born per population per generation = 4.95 × 10⁸ ÷ 6.6439 = 7.45 × 10⁷. At Blount's potentiated rate of 6.6 × 10⁻¹³ per cell-generation, twelve populations over 31,500 generations expect 18.6 independent Cit+ originations; at the ancestral 95% upper bound of 3.6 × 10⁻¹³, 10.1; at the low end of Blount's confidence interval (7.9 × 10⁻¹⁴), 2.2. For the eleven populations that stayed Cit− from generation 33,127 to 82,000: 4.0 × 10¹³ further cells and 26.4 expected originations at the potentiated rate. The caveat that matters: Blount measured the rate on minimal-citrate agar over 45 days of incubation, conditions that give a multi-step process far more time per "generation" than a daily transfer cycle does, so these figures should be read as an upper bound on what the LTEE regime itself would produce. A calculation made for this article, not a published study, of how many independent Cit+ originations the experiment's cells would be expected to produce at Blount's measured rates. The article uses it for the expected number of originations under Blount's rate. ↩ ↩
49.
Two literatures disagree on the direction of the ancestral dcuS lesion. n.d. A comparison of three published papers rather than a single source. Van Hofwegen et al. 2016 and Minnich & Hovde 2025 both call the ancestral lesion a five-base-pair deletion, whereas Turner et al. 2023, who characterised the repair, calls it a five-base-pair insertion that creates a three-copy tandem array. The article follows Turner's account because it explains the repair mechanism. Either way the gene is out of frame and DctA is not expressed, which is all either argument needs. ↩
50.
Hottes, Alison K. et al. “Bacterial adaptation through loss of function.” PLoS Genetics 9(7):e1003617. 2013. Behe quotes the sentence: “Overall, at least one beneficial null mutation was identified in all but five of the 144 conditions considered.” The same paragraph notes that. “single loss-of-function mutations are sufficient to increase the growth rate up to twofold, demonstrating the suboptimal utilization of existing capacities by the wild-type strain.” A meta-analysis of genome-wide fitness data from transposon-insertion and in-frame deletion mutants across 144 conditions, drawn from seven studies of two E. coli strains. It found that at least one beneficial null mutation (a mutation that removes a gene's function) appeared in all but five of the 144 conditions, and the authors suggest that null mutations play an underappreciated role in the early adaptation of bacteria. The article uses it because Behe's general claim rests on this meta-analysis, not on the LTEE alone. ↩
51.
Behe, Michael J. “Citrate Death Spiral.” Discovery Institute. 2020. Its central claim: “the only result from the decades-long, 50,000-plus generation E. coli evolution experiment that even seemed at first blush like it had a bit of potential to yield a novel pathway in the bacterium has resulted instead in spectacular devolution.” A Discovery Institute post by Michael Behe, the author of Darwin Devolves, responding to the 2020 eLife paper by Blount et al. on the citrate-using bacteria. The article uses it because the post draws on real data from that paper, and it states the reading that the article then qualifies. ↩
52.
Blount, Zachary D. et al. “Genomic and phenotypic evolution of Escherichia coli in a novel citrate-only resource environment.” eLife 9:e55414. 2020. A laboratory evolution study that founded populations of Cit+ E. coli (cells able to use citrate) and evolved them for 2,500 generations in citrate-only medium or in medium with glucose and citrate. It found many parallel mutations, some declines in fitness, and substantial cell death in Cit+ clones. The article uses it to check Behe's claim that the Cit+ population is a sick puppy against the measured death counts. ↩
1. 52a.
  “On average, 10.7% of the LTEE ancestral cells grown in DM25 were scored as dead in stationary phase. By contrast, when grown in the same DM25 medium, 29.6% and 39.9% of cells were scored as dead for the Cit+ clones isolated from LTEE population Ara–3 at 33,000 (CZB151) and 50,000 generations (REL11364), respectively. We observed similarly high proportions of dead cells for both clones in DM0 as well (33.1% and 44.2% for CZB151 and REL11364, respectively).” ↩
2. 52b.
  Section Discussion: “we observed extensive cell death in Cit+ clones, and the level of mortality varied considerably even between replicate assays for reasons that we do not understand.” The same discussion reports that. “most evolved clones show no significant increase in fitness, and some are less fit than their ancestor, even in the environment where they evolved”. a point in Behe's favour that his article does not use. ↩