Antibiotic Resistance

"Resistance can emerge relatively quickly in the case of some mutations in target genes … however, resistance to most antibiotics occurs through the aegis of extremely efficient enzymes, efflux proteins and other transport systems that often are highly specialized towards specific antibiotic molecules." — Kirandeep Bhullar and colleagues, PLoS ONE, 2012¹

Summary

Antibiotic resistance is the most-cited example of evolution caught in the act.  Sorting the cases by mechanism changes the picture, but it does not make the example go away.

1. Resistance reaches a bacterium by four routes. A gene arrives from another bacterium; a resistant cell was already in the population; something in the cell breaks or is turned down; or an existing protein is changed. This article scores each route with the four classes in Measuring Biological Information: transfer, loss, gain and modification.
2. Most clinical resistance is transfer or damage, and the creationist literature is right about that. mecA in MRSA came from another Staphylococcus species. vanA matches genes in soil Paenibacillus. Colistin resistance moved onto a plasmid called mcr-1. The commonest extended-spectrum beta-lactamases in hospitals today, the CTX-M enzymes, are a chromosomal gene of the gut and soil bacterium Kluyvera that was moved onto mobile DNA at least nine times, and that already attacked cefotaxime before it moved.^{2 2} Where resistance arises by mutation, the commonest mechanisms are a lost porin, a broken repressor, or a target protein that no longer binds the drug.
3. Transfer is vastly cheaper than mutation, and the numbers show why. Conjugation moved the mcr-1 plasmid into between one in ten and one in a thousand recipient cells.³ Max Demerec measured the penicillin-resistance mutation rate in Staphylococcus aureus at about one in 10⁸ cells per generation, in 1948.⁴ Transfer beats mutation by five to seven orders of magnitude per cell.
4. Resistance genes existed before anyone made an antibiotic. The best evidence needs no deep time at all. NCTC 1 is a dysentery bacterium taken from a British soldier in 1915, thirteen years before Alexander Fleming saw his first mould. It is resistant to penicillin and erythromycin.⁵
5. But high-level resistance was in the soil, not in the pathogens. The Murray Collection holds 683 strains of gut bacteria gathered between 1917 and 1954.⁶ Its strains carried resident resistance genes on their own chromosomes, as NCTC 1 did.⁶ Naomi Datta's team found transferable plasmids in them that carried no resistance.⁷ Tara Houndt and Howard Ochman tested the Murray E. coli against five drugs: none had high-level resistance except one probable contaminant, against about 20% of modern strains.⁸ Both halves are true and the creationist articles quote only the first.
6. Some antibiotic resistance is a real gain, and this site says so. Five point mutations turn TEM-1 beta-lactamase into an enzyme about 46,000 times more resistant to cefotaxime.⁹ Every one of the five changes an existing sequence. None of them knocks anything out, though the key substitution makes the enzyme about ten times less efficient on ampicillin and less stable, and a second substitution repays the stability.^{10 10} The word gain is used here in its plain sense: the cell can do something it could not do before, and lost nothing to get it. In the scoring of Measuring Biological Information the class is modification, because an existing enzyme was changed and no new coded element was added. Miriam Barlow and Barry Hall then evolved the same substitutions in a flask.¹¹ Calling that a loss requires a definition that carries the conclusion.
7. The fitness cost is real, large enough to matter, and often erased. A meta-analysis of 179 measured resistance mutations found mean relative fitness 0.880, a cost of about 12%.¹² With no drug and no further mutation, that cost would cut a resistant strain from half of a population to 1% in about 55 generations.¹³ Some mutations cost nothing. Trimethoprim resistance mutations averaged better than the ancestor with no drug present. Dan Andersson and Diarmaid Hughes conclude that resistance is not expected to go away if prescribing falls.¹⁴
8. The ledger, where it can be run, is small. One step per 10⁸ cells buys one substitution. Two steps at once is one in 10¹⁶, which Demerec priced at one double mutant in ten litres of packed bacteria.⁴ The five-step beta-lactamase was never found that way by anybody; it was assembled step by step in the clinic over decades of selection, and the laboratory reached it with libraries of 1.3 million clones per round.¹¹

The verdict.  Antibiotic resistance is real evolution.  Most of it is not construction.  A small, well-documented part of it is, and the site concedes that part rather than defending a universal claim that its own sources refute.

Part 1: Four routes to a resistant bacterium

Michael Behe, writing for general readers in 2019, sorted antibiotic resistance into three kinds, and his list is a good place to start because it comes from the design side:

"(1) the bug acquires a brand-new gene that allows it to neutralize the antibiotic, which would be categorized as a molecular gain-of-FCT; (2) a control region that normally turned off a bacterial gene that could already inactivate the drug to a small extent breaks, allowing the gene to be active all the time, which would actually be a molecular loss-of-FCT (because the control region no longer works); or (3) a single amino-acid residue of a bacterial protein that interacts with the drug, leading to the death of the bug, is substituted by another kind of amino acid that blocks the drug from working; that would be classified as a modification-of-function."¹⁵

Measuring Biological Information makes one change to that scheme.  Behe scores an acquired gene as a gain, and for the bacterium it is.  But the series is measuring how much working sequence mutation and selection originate, and a gene that arrived from somewhere else was originated somewhere else.  So transfer is a fourth class of its own.  A fifth thing also has to be separated out, because it is not a mutation at all:  sometimes the resistant cell was simply already there.

Route one: the gene arrives from another bacterium

Jose Munita and Cesar Arias give the three ways this happens:  "transformation (incorporation of naked DNA), ii) transduction (phage mediated) and, iii) conjugation (bacterial “sex”)."¹⁶  The genes travel on plasmids, transposons and integrons, and they cross species lines.

The clinical headliners are all transfers.  Methicillin resistance in Staphylococcus aureus comes from mecA, which encodes a penicillin-binding protein with low affinity for beta-lactams, carried on a cassette called SCCmec, and which probably came from Staphylococcus sciuri.¹⁶  Vancomycin resistance in enterococci comes from the vanHAX operon, and "Genes nearly identical to those of the vanA gene cluster (the most prevalent in clinical enterococcal strains) have been found in soil organisms such as Paenibacillus thiaminoluticus".¹⁶  The CTX-M extended-spectrum beta-lactamases trace back to a chromosomal gene in Kluyvera ascorbata, a harmless soil and water bacterium.¹⁷

The most recent of the famous cases is the clearest.  Colistin is a last-resort drug for bacteria that resist everything else.  Until 2015 the only known resistance to it was chromosomal mutation.  Then Yi-Yun Liu's team, running routine surveillance on E. coli from Chinese pigs, found a plasmid gene they named mcr-1.  "Until now, polymyxin resistance has involved chromosomal mutations but has never been reported via horizontal gene transfer."³  They found it in 21% of 804 animals sampled.³

Score:  transfer.  New sequence originated in the receiving lineage:  zero.  The creationist objection here is exactly right, and both sides say the same thing.  Georgia Purdom:  "No new DNA is generated (a requirement for molecules-to-man evolution), it is just moved around."¹⁸  Kevin Anderson:  "Horizontal transfer, though, does not account for the origin of resistance genes, only their spread among bacteria."¹⁹  Nothing in the mainstream literature disagrees.  Munita and Arias call the environment "a prolific source for the acquisition of antibiotic resistance genes" and that is a statement about where the genes come from, not about where they came from originally.¹⁶

Route two: the resistant cell was already in the population

In 1943 Salvador Luria and Max Delbrück showed that bacterial resistance to a virus arises before the virus is applied, by measuring how the number of resistant colonies varies between parallel cultures.²⁰  Five years later Max Demerec did the same thing for antibiotics.  His conclusion:

"resistance to certain concentrations of penicillin or streptomycin originates through mutation, and that resistant bacteria may be found in any large population, the proportion depending on the mutation rate"⁴

That sentence contains both sides of the argument.  The resistant cells are already there when the drug arrives, which is what Carl Wieland means by "the information to resist the antibiotic was already there in the bacterial population—it did not arise by itself, or in response to the antibiotic."²¹  And they are there because mutation put them there, at a measurable rate, in the generations before the drug arrived.

Score:  not a separate class.  Standing variation is a statement about timing, not about mechanism.  The mutation that made the resistant cell still has to be scored, as a loss, a gain or a modification.  Treating "it was already there" as an answer to the origin question just moves the question back a few generations.

Route three: something breaks, or is turned down

This is the largest class by mechanism, and the creationist literature has documented it more carefully than the popular pro-evolution accounts.

Kevin Anderson's 2005 paper in the Creation Research Society Quarterly is the best version.  His Table I lists fourteen antibiotics with the mechanism of resistance to each, and every single entry is a loss or a reduction:  loss of enzyme activity for actinonin, reduced formation of a porin for chloramphenicol, imipenem and tetracycline, loss of a regulatory protein for azithromycin and nalidixic acid, loss of affinity to RNA polymerase for rifampicin, loss of proton motive force for zwittermicin A.¹⁹

The mainstream reviews agree on the mechanisms.  Munita and Arias give the porin case:  "One classic example of porin-mediated resistance is the aberrant production of OprD in P. aeruginosa, which is normally used for the uptake of basic amino acids and antibiotics (i.e., imipenem, a potent anti-pseudomonal antibiotic from the carbapenem class)."¹⁶  Losing the door keeps the drug out.  Don Batten adds the case of fosfomycin, which rides into the cell on a nutrient transporter:  "Mutations in the genes that specify how the cell makes the transporters, or in regulator genes that stimulate their production, can result in no transporters, or less effective transporters."²²

A second family of losses works by removing a brake.  Efflux pumps push toxins out of the cell, and their production is held down by repressor proteins.  Break the repressor and the pumps run flat out.  Anderson describes the mar system in E. coli and draws the conclusion:  "a mutation that causes loss of regulatory control (in this case the repressor protein, MarR) does not offer a genetic mechanism that can account for the origin of this regulatory control."¹⁹

A third family removes the enzyme that turns a prodrug into a drug.  Metronidazole does nothing until a bacterial nitroreductase activates it.  Knock out rdxA and the drug stays inert.  Anderson:  "All these mutations result in loss of the enzyme activity necessary for the drug to be effective in the cell, hence the cell becomes resistant to metronidazole.  But, loss of enzymatic activity does not provide a genetic example of how that enzyme originally “evolved.”"¹⁹  The same logic covers outright losses of katG in tuberculosis.  The commonest isoniazid-resistance allele is different:  it keeps most of the enzyme's activity and lowers only the side reaction that activates the drug.  Mycobacterium tuberculosis gives the measurements.

Score:  loss.  New sequence originated:  negative.  And this is not a creationist reading of a mainstream result.  It is what the experiments say.  When Michael Baym's team sequenced 251 E. coli isolates that had evolved trimethoprim resistance on a giant agar plate, they found "three genes not classically associated with trimethoprim resistance were repeatedly mutated, often with a probable loss of function (frameshift or nonsense): a phosphate transporter (pitA), shikimate kinase I (aroK), and a negative regulator of the PhoQP system (mgrB).  Knockout of these genes in the ancestral strain confirmed their resistance phenotypes".²³  They built the knockouts and the knockouts were resistant.  That is as direct a demonstration as the loss argument could ask for.

Route four: an existing protein is changed

Here the picture splits, and the split is the substance of this article.

Most of the cases in this class are a target protein losing its grip on the drug.  Rifampicin binds a pocket in the beta subunit of RNA polymerase.  A point mutation in rpoB changes the pocket.  Munita and Arias:  "while these mutations result in decreased affinity of the drug for its target, they usually spare the catalytic activity of the polymerase, permitting transcription to continue".¹⁶  Anderson, from his own bench, says the same thing and draws the conclusion:

"In my laboratory, we routinely obtain mutant strains with a resistance level that is orders of magnitude greater than that of the wild-type strain.  When rifampin is present, this mutation provides a decided advantage for survival compared with those cells lacking these specific mutations.  But, each of these mutations eliminates binding affinity of RNA polymerase for the rifampin."¹⁹

Fluoroquinolone resistance works the same way through gyrA, streptomycin resistance through the ribosome and rpsL.  In each case a binding site gets worse at binding.  Score:  modification, down.  New sequence originated:  the substitution is one nucleotide, and the direction is downward, so the nucleotide conversion rule of the taxonomy article gives zero.

But one large group of cases in this class runs the other way, and Part 3 is about it.

What nobody has measured

A reader will want to know what share of clinical resistance each route accounts for.  No source read for this article gives that number, and this article does not invent one.  The share would depend on the organism, the drug, the country and the year.  What can be said is narrower and still useful:  for the drugs whose resistance is carried on mobile elements, such as methicillin, vancomycin, the extended-spectrum cephalosporins and colistin, transfer dominates, because transfer is cheap.  For rifampicin and the fluoroquinolones, mutation dominates, because those targets are chromosomal and a single substitution suffices.

The nearest thing to a ranking comes from the largest count of resistance deaths, the 2019 Global Research on Antimicrobial Resistance study.  It estimated 1.27 million deaths caused by resistant bacteria that year, and named the seven deadliest pairings of germ and drug:  "meticillin-resistant S aureus, caused more than 100 000 deaths attributable to AMR in 2019, while six more each caused 50 000–100 000 deaths".²⁴  The study does not sort them by mechanism.  Sorting them with the mechanism reviews gives four carried mainly by acquired genes (MRSA, the two cephalosporin-resistant pairings, and carbapenem-resistant Klebsiella), one probably so (carbapenem-resistant Acinetobacter), and two by chromosomal mutation (multidrug-resistant tuberculosis and fluoroquinolone-resistant E. coli).²⁵  That is a ranking, not a measured share, but it points the same way as the cheapness of transfer.

Part 2: Resistance genes are older than antibiotics

This is the strongest argument the creationist side makes on this topic, and it is correct.  It needs to be stated carefully, because two different claims are usually run together.

The claim that is established

Resistance genes existed before humans made or prescribed antibiotics.  Four independent lines support it.

A dysentery bacterium from 1915.  NCTC 1 is the first isolate ever accessioned into the United Kingdom's National Collection of Type Cultures:  a Shigella flexneri taken from a British soldier with dysentery on the Western Front.  Kate Baker's team sequenced it in 2014 and tested it against modern drugs.  "The NCTC1 accession card reported resistance to erythromycin and penicillin in tests done in the 1950s (figure 3).  These phenotypes were confirmed as part of this study, with NCTC1 having MICs of 16 mg/L and 24 mg/L for these compounds respectively."⁵  The genes are chromosomal:  an ampC beta-lactamase "widely conserved throughout many Enterobacteriacae" and a three-part efflux pump.⁵  The conclusion is stated in the paper:  "We show that this 1915 isolate, from before the discovery of penicillin, has intrinsic resistance to antimicrobials."⁵  This case rests on a museum accession record.  No dating method is involved and no reader has to accept anything about deep time.

MRSA on hedgehogs.  Jesper Larsen's team sequenced about a thousand mecC-carrying MRSA genomes and found that several lineages "probably originated in the early-to-late 1800s, long before the first β-lactam—penicillin G—became widely available as a therapeutic option in the 1940s."²⁶  They also found the reason.  European hedgehogs carry a skin fungus, Trichophyton erinacei, which "produces two β-lactam antibiotics that provide a natural selective environment in which methicillin-resistant S. aureus isolates have an advantage over susceptible isolates."²⁶  The date comes from a molecular clock, so it is an inference, but the clock runs over two centuries, not two billion years.

A sealed cave.  Kirandeep Bhullar's team cultured 93 bacterial strains from three deep sites in Lechuguilla Cave in New Mexico and tested them against 26 antimicrobials.  "In the collection of 33 Gram-positive strains, on average, 70% of the strains were resistant to 3–4 different antibiotic classes.  Three strains were resistant to 14 antibiotics, all of which were Streptomyces spp."¹  A follow-up study took one isolate, Paenibacillus sp. LC231, sequenced it, and mapped 18 chromosomal resistance elements, five with no known relative.²⁷

Permafrost DNA.  Vanessa D'Costa's team pulled DNA from frozen Yukon sediment and recovered fragments of a beta-lactamase, a tetracycline resistance gene, and a complete vancomycin resistance operon.  They synthesised four of the ancient vanA reading frames, expressed two, and found the enzymes worked:  "These results clearly show that the vanHAX genes identified in the ancient samples encode enzymes capable of genuine antibiotic resistance."²⁸

How each age claim was established, and where the article stops

The Lechuguilla and permafrost ages depend on methods a young-earth reader rejects, so they have to be reported honestly and then set aside.

D'Costa's 30,000 years is stratigraphic, not direct.  The cores came from 50 cm below a volcanic ash layer, and "The tephra has been dated at several sites in the area to about 25,300 radiocarbon (14C) years BP, or about 30,000 calendar years".²⁸  The sediment itself was not radiocarbon dated.  The claim that it never thawed is cryostratigraphic:  ice wedges are preserved above and below.²⁸

Bhullar's 4 to 7 million years is geological and comes from other papers, not from measurements in that one:  the cave formed by rising sulfuric acid "in the last 7 million years", the deep parts were cut off as the aquifer dropped, and an impermeable caprock "limits vadose water from entering the system".¹

Neither age matters to the argument.  What the argument needs is that the deep cave and the frozen sediment had no contact with pharmaceutical antibiotics, and both papers establish that by ordinary means:  the cave's caprock and its 20-year access closure, the permafrost's intact ice, the GFP-tagged contamination controls, over 300 control reactions, and the absence of any of the sequences from GenBank.²⁸  Say "these genes are not a product of the clinic" and stop.  Do not say "resistance is 30,000 years old" on a site that does not accept the date.

The claim that is not established, and the trap in it

"Resistance genes predate antibiotics" answers the claim that resistance genes appeared after 1940.  It does not answer the question of where the genes came from.

Antibiotics are not human inventions.  They are chemical weapons that soil bacteria and fungi make against each other.  Julie Perry, Nicholas Waglechner and Gerard Wright put it plainly:  "The origin of many modern resistance genes in pathogens is likely environmental bacteria, including antibiotic producing organisms that have existed for millennia."²⁹  They report that "most Streptomyces are resistant to an average of seven to eight antibiotics, including newly developed and clinically important therapeutics."²⁹  A bacterium that makes an antibiotic has to survive the antibiotic it makes.

So the finding that resistance genes are old is what both models predict.  It refutes the popular claim that resistance evolved from nothing since 1940.  It does not show that the genes were designed rather than evolved, and it does not show that they were evolved rather than designed.  On this question the evidence is silent because nobody has watched an enzyme or pump of the kind found in soil originate.  The one laboratory case of resistance arising from random sequence, discussed in Part 3, works by a different and much simpler route.

And the counterweight the creationist articles leave out

In 1983 Naomi Datta and V. M. Hughes went through the Murray Collection, strains of gut bacteria that E. G. D. Murray gathered between 1917 and 1954, before and during the introduction of antibiotics.  About a quarter of the strains carried machinery for moving DNA, and the plasmids they moved carried no resistance:

"E.D.G. Murray collected strains of Enterobacteriaceae from 1917 to 1954; we now report that 24% of these encode information for the transfer of DNA from one bacterium to another.  From at least 19% of the strains, conjugative plasmids carrying no antibiotic resistance were transferred to Escherichia coli K-12."⁷

That abstract does not say that no Murray strain was resistant, and they were not all susceptible.  Kate Baker's team sequenced 370 of the collection's 683 strains in 2015.⁶  They note that "AMR has been reported in the pre-antibiotic era Murray Collection strains, including tetracycline resistance in Proteus sp., ampicillin resistance in the Klebsiella and both ampicillin and kanamycin resistance in Escherichia sp.", and their genome search found "many resistance genes (often known to be chromosomally encoded) that were present across all members of a genus".⁶  These are the same kind of resident, low-level genes as NCTC 1's ampC.

What the old strains lacked was high-level resistance.  Tara Houndt and Howard Ochman measured five drugs against Murray E. coli, which they describe as taken mostly from human infections, and against two modern collections.  High-level here means resisting roughly 50 µg/ml or more.  "All of the pre-antibiotic era strains were susceptible to high levels of these antibiotics, whereas 20% of strains from contemporary populations of Escherichia coli and Salmonella enterica displayed high-level resistance to at least one of the antibiotics."⁸  The one exception they found, Murray strain 125, "was probably a contaminant".⁸

D'Costa's own abstract names this as the standard view being challenged:  "this view is reinforced by the fact that collections of microbes that predate the antibiotic era are highly susceptible to antibiotics".²⁸

Put these together and the correct statement is precise.  The pathogens carried their own chromosomal genes, which give low-level resistance.  The genes for high-level, transferable resistance were in the soil.  The delivery system, conjugative plasmids, was already in the pathogens.  What antibiotics did was load the third with the second.  Munita and Arias describe exactly this:  mutation and transfer are the two routes, and the environmental resistome is where the transferred genes come from.^{16 16}  An article that quotes only the pre-1940 evidence and not the Murray Collection is quoting half a result.

A claim that should not be repeated

Three creationist articles say resistant bacteria were recovered from the frozen bodies of polar explorers.  Carl Wieland:  "Bacteria revived from the frozen intestines of explorers who died in polar expeditions carried resistance to several modern antibiotics, which had not been invented when the explorers died."²¹  Don Batten and Georgia Purdom repeat it.

The claim traces to a real piece of work:  K. Kowalewska-Grochowska and colleagues at the University of Alberta presented "Isolation and sensitivities of century-old bacteria from the Franklin expedition" at a conference in Los Angeles on 24 October 1988.  It was never published as a paper, and the conference abstract could not be obtained.  Wieland's own footnote is not to it but to a trade newspaper story about it.  Purdom's footnote points to an Apologetics Press article that, when retrieved from the Internet Archive and read in full, does not mention the Arctic at all.  The details are in quarantine.md.

Nothing is lost by dropping it.  NCTC 1 makes the same point with a published genome, a museum accession record, measured MICs and named genes.

Part 3: The case that is a real gain

The claim this article concedes is narrow and specific.  Here is the evidence for it.

What Weinreich's team did

TEM-1 is the commonest beta-lactamase in Gram-negative bacteria.  It hydrolyses penicillins well and third-generation cephalosporins such as cefotaxime very badly.  Five changes turn it into an enzyme that handles cefotaxime:  four amino-acid substitutions (A42G, E104K, M182T, G238S) and one change in the control region upstream of the gene (g4205a), which raises expression about 2.5-fold.

Daniel Weinreich, Nigel Delaney, Mark DePristo and Daniel Hartl built all 32 combinations of those five changes and measured each one's cefotaxime resistance in E. coli.³⁰  Their Table 1 gives the numbers.  TEM-1 is inhibited at 0.088 µg/ml.  The quintuple mutant is inhibited at 4.1 × 10³ µg/ml.³⁰  That is a factor of about 46,000.⁹

The paper's headline result is about paths, not about the gain.  Of the 120 possible orders in which five mutations could accumulate, "102 of the 120 mutational trajectories from TEM wt to TEM* are selectively inaccessible," because four of the five mutations make things worse on some backgrounds.³⁰  Half of all realisations follow only four paths, or two under a more realistic model.³⁰

Why this is a gain and not a loss

Four features of the case, each checkable against the paper.

Nothing was knocked out.  G238S changes a residue in the substrate pocket.  M182T stabilises the protein.  The other three tune the enzyme and its expression.  None of them is a frameshift, a nonsense mutation, a deleted porin or a broken repressor.

The landscape has one peak and the first step is uphill.  "TEM* exhibits the highest resistance and, because at least one mutation increases resistance in all other alleles, the fitness landscape is single-peaked".³⁰  G238S alone raises the MIC from 0.088 to 1.4 µg/ml, sixteen-fold, so the path needs no neutral first step.³¹

The same mutations happen in the wild, and in a flask.  Miriam Barlow and Barry Hall mutated TEM-1 at about two mutations per copy, built nine libraries averaging 1.3 million clones, and selected for extended-spectrum resistance.¹¹  The result:  "Nine mutations have arisen independently multiple times in nature: L12F, E104K, R164S, R164H, M182T, A237T, G238S, E240K, and T265M.  Seven of those, E104K, R164S, R164H, M182T, G238S, E240K, and T265M, were recovered during our in vitro evolution experiments".¹¹  Their conclusion:  "in vitro evolution with the Mutazyme polymerase recovers substantially the same mutations that are selected in nature."¹¹  They predicted that resistance to cefepime, not then seen in nature, was likely to evolve.¹¹

The trade-off is a trade-off, not a loss.  G238S does cost something:  it increases aggregation and lowers thermodynamic stability, and M182T is needed to pay that back.³⁰  The old activity pays too, and the amount is measured.  Timothy Palzkill, a Baylor College of Medicine biochemist who has published TEM mutant studies himself, summarises the kinetics:  "While the G238S substitution increases the rate of cefotaxime and ceftazidime hydrolysis, it significantly decreases the rate of hydrolysis of penicillins.  For example, kcat/KM for ampicillin hydrolysis is reduced 10-fold compared to wild-type TEM-1 while kcat is decreased ~50-fold".¹⁰  E104K, by contrast, "has little effect on the hydrolysis of penicillins".¹⁰  M182T is a repair:  it "compensates for TEM ESBL substitutions that trade off protein stability for improved oxyimino-cephalosporin hydrolysis".¹⁰  So of the five changes, one trades away some penicillin activity and some stability, one pays the stability back, and one raises expression.

That is the shape Kevin Anderson and Georgia Purdom predicted in 2008, when they called resistance mutations "a very clear example of antagonistic pleiotropy".³²  The design side is right that the new activity was bought with a measurable loss.  It is wrong if it calls the result a loss overall.  The key substitution costs the enzyme about tenfold in efficiency on ampicillin, which it still hydrolyses, while the five changes together raise the cell's cefotaxime resistance about 46,000-fold.  The two numbers measure different things, an enzyme rate and a whole-cell dose, but the difference in scale is not in doubt.  That is what the taxonomy article calls a modification, shifted when the old activity is lost, and a gain when a new activity appears that was not there before.  TEM* keeps its old activity in reduced form and adds a new one.

Where this article agrees and disagrees with the taxonomy article

Measuring Biological Information scores this case as modification, shifted, with five new nucleotides.  This article agrees with that score and adds one qualification the taxonomy article could not make, because its writer had only the abstract.

The qualification is that the five changes are not five substitutions in a protein.  Four are;  the fifth, g4205a, is in a non-coding control region and works by raising expression.³⁰  So the nucleotide count of four coding changes plus one regulatory change is the honest version, and an article that prints "five substitutions in an enzyme" is slightly wrong.

The taxonomy article also scores CTX-M on plasmids as transfer, with zero new sequence in the receiver.  This article agrees, and the case histories above support it:  the Kluyvera origin is inferred from sequence identity of 95 to 100%.¹⁷

This matters more than the taxonomy article says, because CTX-M is now the ESBL that counts.  Rafael Cantón, José María González-Alba and Juan Carlos Galán:  "Phylogenetic analyses suggest that CTX-Ms were not originated by mutations from previous plasmid meditated enzymes but through mobilization of chromosomal bla genes from Kluyvera spp. … These original mobilized blaCTX-Ms genes affected cefotaxime to a higher degree than ceftazidime."²  The extended-spectrum activity was there before the gene moved.  TEM and SHV variants "were predominant in the ESBL landscape over the 1980s and 1990s", but since then "the CTX-M enzymes have nearly displaced other ESBLs enzymes in Enterobacteriaceae, including TEM and SHV ESBL variants".^{2 2}  Munita and Arias agree:  "CTX-M enzymes have become the most prevalent ESBL worldwide and are responsible for a large proportion of cephalosporin resistance in E. coli and K. pneumoniae."¹⁶  So the beta-lactamase gain this article concedes is real, but in today's hospitals the commonest extended-spectrum enzyme is a moved gene, not a built one.  That is a fair point for the design side.

It has a limit, and the same paper states it.  CTX-M enzymes were then tuned toward a second drug, ceftazidime, by point mutation.  CTX-M-15, now found worldwide, differs from its relative CTX-M-3 "in only one amino acid position (D240G)".²  Cantón's own laboratory selected ceftazidime-resistant variants carrying "A77V, P167S, and D240G", and "These changes have been described in the nature".²  So the dominant family was moved, then improved by the same kind of single-step change as TEM, one substitution at a time.  Each such step is a modification, shifted.

The strongest creationist statement, and why it fails

Kevin Anderson's summary paragraph makes a universal claim:

"Spontaneous mutation does provide a potential genetic mechanism for the origin of these genes, but such an origin has never been demonstrated.  Instead, all known examples of antibiotic resistance via mutation are inconsistent with the genetic requirements of evolution."¹⁹

The second sentence is false, and it was false when it was written.  The words "extended-spectrum", "ESBL", "TEM" and "penicillinase" do not appear anywhere in Anderson's paper.  He cites Barlow and Hall's 2002 paper in the Journal of Molecular Evolution and quotes its page 314, so he had been reading that pair of authors.  He does not cite their other 2002 paper, in Genetics, which is the one that evolved TEM-1 into extended-spectrum enzymes in a flask.  Weinreich's measurements came out the following year.

Purdom's version of the claim fails for the same reason:  "Mutation and natural selection, thought to be the driving forces of evolution, only lead to a loss of functional systems."¹⁸

Neither author stayed at the universal version.  In 2008 Anderson and Purdom wrote a joint paper for the International Conference on Creationism that opens by criticising their own side:  "creationists have tended to offer an inconsistent or incomplete perspective of “beneficial mutations” within a creation framework.  This includes the frequent denial that mutations can ever provide a beneficial phenotype".³²  On resistance their wording is now "typically", not "all":  "a significant number also result from point mutations. … these mutations typically reduce or inactivate cellular systems or functions."³²  That later statement is the one a reader should hold them to, and the TEM evidence above does not refute it.

The same paper gives the creationist answer to an enzyme that takes on a new substrate.  Discussing nylon-eating bacteria, it says:  "Nonetheless, reduced specificity of a pre-existing enzyme is biochemically degenerative to the enzyme, even if it provides a presumed phenotypic benefit."³²  Applied to TEM, that answer is half right.  The ESBL substitutions do cost penicillin activity and stability, as Part 3 measured.  But calling a 46,000-fold gain in resistance "degenerative" because the enzyme now works on more substrates is a choice of definition, the same move as Batten's in 2001 below, and it is not a measurement.  Wieland's too:  "In no case is there any evidence of an information-adding, 'uphill' change."²¹

The creationist statement that survives

Don Batten's 2017 article states the claim narrowly, and the narrow version stands:

"It might appear that a bacterium faced with an antibiotic and then acquiring such an ability would be a great example of evolution—a new enzyme; a new gene.  However, in the instances known, the machinery to manufacture the antibiotic-destroying enzyme has never been observed to arise by mutation."²²

Nothing in this article's sources contradicts that.  No beta-lactamase gene has been watched arising from DNA that was not a gene.

The nearest challenge is a 2019 experiment from Dan Andersson's laboratory, and a skeptic should know it.  Michael Knopp's team expressed random open reading frames in E. coli, libraries with "a total diversity of 5.82 × 10⁸", and selected for survival on aminoglycoside antibiotics.³³  "Selections on antibiotic-containing agar plates resulted in the identification of three peptides that increased aminoglycoside resistance up to 48-fold."³³  That is resistance from sequence that was not a gene, and it should not be waved away.

But the mechanism is the damaging kind.  The peptides lodge in the membrane and partly discharge its voltage, so less drug is pulled in.  The authors selected ordinary resistant mutants on the same plates for comparison:  "Whole-genome sequencing of six clones showed that all had mutations in genes associated with electron transport and generation of membrane potential", and adding the peptide Arp1 to those mutants raised their resistance no further, "indicating that the chromosomal mutations and peptides act by a similar mechanism (i.e., reduction in membrane potential)".³³  So random sequence reached exactly the level that breaking the respiratory chain reaches, by doing the same thing to the membrane.  It did not destroy the antibiotic, pump it out or reshape its target.  The peptides were also made from an inducible laboratory promoter on a plasmid, not found in a genome.  Batten's narrow claim survives the experiment.  A broad claim that random sequence can never confer resistance does not.  TEM* is an enzyme that already hydrolysed beta-lactams, retuned to a new substrate.  The distinction between "a new gene" and "an old gene doing something new" is the whole argument, and Batten draws it correctly.

Batten's 2001 article goes further and predicts the shape of the result:

"An enzyme is highly specialized to break down one specific type of chemical very well, and hardly affect other chemicals.  A mutation could reduce its specificity, i.e. it no longer does its main job so well, and affects other chemicals to some extent too.  Normally, a biological system with such a mutation would not function as well, and reduced specificity is reduced information by definition."³⁴

That describes the TEM case accurately, up to the last clause.  "Reduced specificity is reduced information by definition" is a definitional move, not a measurement.  TEM* is 46,000 times better at cefotaxime.  Whether that counts as information depends entirely on which definition is adopted before the evidence is looked at, which is the problem Measuring Biological Information exists to solve.

Two more cases worth naming

The megaplate.  Michael Baym's team built a 120 × 60 cm tray of agar in bands of rising antibiotic concentration and filmed E. coli crossing it.²³  Resistance rose 10,000-fold for trimethoprim in ten days and 100,000-fold for ciprofloxacin in twelve.²³  The video is the most widely shared demonstration of evolution ever made.

The sequencing is less advertised and more interesting.  The most-mutated gene was the drug's own target, folA.²³  Three other repeatedly-hit genes were knockouts confirmed by construction.²³  A mutator phenotype "emerged at least six times independently", every time by a mutation in dnaQ, the proofreading subunit of DNA polymerase III.²³  Breaking the proofreader raises the mutation rate, which helps under strong selection and is damage.  Nothing new was built on that plate.  It is a very good film of routes three and four.

A result that cuts against the ledger's tidiness.  Baym's team also found that "Mutations that increased resistance often came with a cost of reduced growth, which was subsequently restored by additional, compensatory, mutations".²³  The cost appeared and then was paid off, inside a twelve-day experiment.

Part 4: What resistance costs

The fitness-cost argument is the second pillar of the creationist case, and it is the one that has worn worst.

What the measurements say

Anita Melnyk, Alex Wong and Rees Kassen assembled every published measurement of the fitness effect of a chromosomal resistance mutation that used a competition assay rather than a growth-rate comparison.  That gave 179 measurements, 121 unique mutations, 8 species, 16 antibiotics.¹²  The result:

"We found a significant fitness cost of resistance mutations (mean fitness = 0.880, z = 129.3, P < 0.0001)."¹²

A mean relative fitness of 0.880 is a cost of about 12%.  Melnyk's team measure fitness as doublings:  the resistant strain doubles 0.88 times for each doubling of its sensitive rival.¹²  That is not a small cost.  With no drug and no further mutation, it would shrink a resistant strain from half of a population to 1% in about 55 generations.¹³  In population-genetics terms that is strong selection, and the popular creationist argument is right to notice it.  Where that argument goes wrong is in what follows from it, which the next section shows.  The authors are careful about the variation around the mean:

"The most parsimonious interpretation is that resistance mutations are often genuinely costly.  It seems unlikely that this result represents a publication bias because, if anything, the observation of no-cost mutations is the more novel result.  Notably, there is also substantial variation in costs of resistance with at least some mutations exhibiting little or no cost."¹²

And one class of resistance mutations is not costly at all.  "the mean fitness of mutations conferring resistance to the dihydrofolate reductase inhibitor trimethoprim was >1 (Fig. 2), indicating that these resistance mutations are beneficial in the absence of drug".¹²

Plasmid-borne resistance costs less than chromosomal mutation.  Tom Vogwill and Craig MacLean's meta-analysis of 49 plasmids across 16 studies found mean relative fitness about 0.91.³⁵

Compensation

Dan Andersson and Diarmaid Hughes wrote the standard review of the cost, and their conclusions are not the ones the creationist articles cite.  Costs depend on where the measurement is taken:  "Some resistance mutations that show no cost in bacteria grown in laboratory medium have high costs in laboratory mice and, conversely, some mutations that show no cost in mice can have substantial costs in vitro."¹⁴  Some mutations are genuinely free:  a substitution in ribosomal protein S12 "is a genuine cost-free resistance mutation", and the clinic selects for the cheap ones, because "no-cost (or low-cost) mutants are preferentially selected in humans."^{14 14}

Where a cost exists, a second mutation usually pays it off without losing the resistance.  "Several in vitro and in vivo studies have shown that in most cases it is possible to detect compensation when resistant bacteria are serially passaged".¹⁴  Melnyk's team report that "compensatory mutations can arise within a few generations following the emergence of resistance."¹²

The practical conclusion is the opposite of "superbugs are superwimps":

"The main finding is that reversibility in clinical settings is expected to be slow or non-existent.  The main reasons for this are that the intrinsic dynamics of reversal are expected to be slow (even if a fitness cost is present), compensatory evolution and cost-free resistances can reduce the cost and thereby reduce the driving force for reversibility"¹⁴

What this does to the creationist argument

Carl Wieland's 1997 article is where the popular version comes from.  His summary point:  "'Supergerms' are actually not 'super' at all.  They are generally less hardy, and less fit to survive outside of the special conditions in hospitals."²¹  Purdom:  "the change has come at a cost.  The altered protein is less efficient in performing its normal function, making the bacteria less fit in an environment without antibiotics."¹⁸

On the measurements, that is true on average and by about 12%, a cost large enough to matter, and it is false for some mutations, and it usually stops being true after a few more generations.  A 12% mean cost, routinely compensated, is not evidence that nothing was built.  It is evidence that building costs something, which every model expects.

Kevin Anderson states the argument more carefully than the popular articles, and his version should be the one engaged with.  He concedes the compensation point outright:  resistance "may also impart some decrease of “relative fitness” (severe in a few cases), although for many mutants this is compensated by reversion."  Then he moves the argument:  "The real biological cost, though, is loss of pre-existing systems and activities."¹⁹  That is a much better argument, because it does not depend on the size of the cost at all.  It depends on the claim that every case is a loss, which Part 3 shows is not true.

Part 5: The ledger

The series asks how many cells had to be searched to find a given change, and what the change bought.  For antibiotic resistance two of the three numbers are available and the third is not, so the ledger runs partway and then stops.

One step

Demerec measured the rate in 1948 and priced its consequences himself:

"It may be estimated from survival curves, however, that the mutation rate is low, in the neighborhood of 1 × 10⁻⁸."⁴

At 0.12 units of penicillin per millilitre, "about 4 per 10⁸ bacteria lived."⁴  So one resistance step costs a search of roughly 10⁸ cells.  A saturated overnight culture holds about 10⁹ cells per millilitre, so one millilitre of broth contains the search.  This is why single-step resistance to rifampicin appears so readily, and why Anderson can say he gets highly resistant mutants routinely.

Two steps at once

"The chance of two simultaneous mutations is 10⁻⁸ × 10⁻⁸ = 10⁻¹⁶, and of three simultaneous mutations, 10⁻⁸ × 10⁻⁸ × 10⁻⁸ = 10⁻²⁴.  Since the volume of an S. aureus cell is about one cubic micron, it would be expected that one double mutant, on the average, would be found in ten liters of bacteria, and one triple mutant in one million cubic meters."⁴

That is the whole benchmarking argument, written in 1948 by a mainstream geneticist.  Ten litres of packed bacteria for two simultaneous steps.  A million cubic metres for three.  The arithmetic checks:  a cubic micron is 10⁻¹⁵ litres, so 10¹⁶ cells occupy 10 litres and 10²⁴ cells occupy 10⁹ litres, which is 10⁶ cubic metres.³⁶

Why that is not the cost of the beta-lactamase

The five-step TEM path did not require five simultaneous mutations, and this is the point where the benchmarking argument has to be applied honestly.  Weinreich's landscape is single-peaked and the first step is beneficial on its own.^{30 31}  Each step can therefore fix before the next is needed, and the cost is five sequential searches of about 10⁸ cells each rather than one search of 10⁴⁰.

What the path does cost is order.  102 of the 120 orders are closed, and half the probability sits on two to four of the remaining eighteen.^{30 30}  Weinreich's own conclusion is about repeatability, not difficulty:  "It now appears that intramolecular interactions render many mutational trajectories selectively inaccessible, which implies that replaying the protein tape of life (28) might be surprisingly repetitive."³⁰

What the search bought

This is the number the series wants, and here it is small.  Under the nucleotide conversion rule of the taxonomy article, the TEM case originated five nucleotides:  four coding substitutions and one regulatory change.  No new gene, no new fold, no new coded element that did not exist.  An enzyme that hydrolysed one substrate now hydrolyses another.

Scale that against what the series is testing.  The number of bacterial cells on earth is often put near 10³⁰, and antibiotics have been in mass production since the 1940s, at "100,000–200,000 tonnes per annum worldwide" by 2002 and "well over 1 million tonnes since the 1940s."¹⁴  Eighty years of the most intense chemical selection pressure ever applied to any group of organisms, across a population larger than any other on earth, has produced:  a great deal of gene shuffling, a long list of broken porins and repressors, a long list of retuned targets, and a handful of enzymes with shifted substrate preferences.

That is the argument, and it is an argument about extrapolation, not about whether change happens.  It is also, honestly stated, a lower bound.  Nobody has catalogued every adaptive change in every bacterial lineage over eighty years, and nobody could.

Where the ledger stops

Three inputs the series normally uses cannot be supplied here, and the article says so rather than guessing.

• The denominator. "Total bacterial cells exposed to antibiotics since 1940" is not a measured quantity. The 10³⁰ figure for cells on earth is an estimate for all bacteria, not for those under drug selection, and using it would be inventing a denominator.
• The full numerator. There is no catalogue of every adaptive change in every bacterial species over the antibiotic era. The gains listed above are the ones that were studied, not a sample.
• The share by route. No source read gives a measured split between transfer, loss, modification and gain across clinical resistance.

Objections and responses

"Antibiotic resistance is the textbook case of evolution in action.  You are explaining away the best evidence there is."  It is evolution, and this article says so four times.  The question the series asks is different:  how much new working sequence did the process originate, and how many cells did it search to get it?  For the best-documented case the answers are five nucleotides and about five sequential searches of 10⁸ cells.  Those are the real numbers and neither side gets to round them.

"Your own sources say resistance genes are 30,000 years old and beta-lactamases are two billion years old.  You cannot use the papers and reject the dates."  The dates are used for nothing.  The claim taken from D'Costa 2011 and Bhullar 2012 is that those genes were never exposed to a manufactured antibiotic, which each paper establishes by contamination controls, cryostratigraphy and cave geology rather than by the date.  The 1915 Shigella and the 1800s hedgehog MRSA make the same point without any deep-time assumption, and they are cited first for that reason.  The molecular-clock ages are reported as what the mainstream method yields and are not relied on.

"If resistance genes already existed, then resistance is not evolution at all."  Two things are being run together.  A bacterium acquiring a plasmid is not originating anything, and the creationist sources are right about that.  But the Murray Collection shows the plasmids of 1917 to 1954 carried no resistance, so something changed between then and now.⁷  What changed is which genes are on which plasmids in which bacteria, and that is a real evolutionary change, even though it originated no sequence.

"Every resistance mutation is a loss.  Anderson's Table I proves it."  Table I is fourteen antibiotics and every entry is a loss or reduction, and every entry is correct.  It is also a selected list.  It does not include the TEM and SHV extended-spectrum beta-lactamases, the best-documented class of resistance built by point mutation, where the mutations improve an enzyme's action on a new substrate.  A table of fourteen cases cannot establish a claim about all cases when the omitted class is the counterexample.  The creationist side can fairly add two things.  Anderson himself narrowed the claim to "typically" in 2008.³²  And the TEM and SHV enzymes were the commonest ESBLs only in the 1980s and 1990s;  today the transferred CTX-M enzymes have nearly displaced them, so the counterexample is real but no longer the main clinical story.²

"The TEM mutations trade one activity for another.  That is not a gain, it is a shift."  That is the strongest version of the objection and it is close to right.  The taxonomy article scores the case as modification, shifted, and this article agrees with that score.  The reason it still counts against the universal claims of Anderson in 2005 and Purdom in 2007 is that a shift is not a loss.  Purdom's 2007 claim is that resistance mutations "only lead to a loss of functional systems," and a 46,000-fold rise in cefotaxime resistance, built by five changes that knock nothing out, is not that.  The objection does land a real point:  the key substitution costs the enzyme about tenfold in efficiency on ampicillin and some stability.¹⁰  So TEM* is a trade, not a free gain, and the two authors' 2008 wording, that resistance mutations are "a very clear example of antagonistic pleiotropy", describes it fairly.³²

"Resistant bacteria are less fit.  Remove the drug and they disappear."  The measured mean cost is 12%, several classes show no cost, trimethoprim resistance is beneficial without the drug, and compensatory mutations appear within a few generations.^{12 12 12}  The review that established the cost concludes that reversal in the clinic is "slow or non-existent."¹⁴  The creationist version of this argument was reasonable in 1997 and has not been updated.

"You are just running Behe's argument."  Behe's First Rule is a rate claim:  damaging mutations that help are expected to appear "100–1000 times" more often than constructive ones, so they arrive first.³⁷  He calls it "a rule of thumb.  It is a heuristic, useful generalization, rather than a strict law."³⁷  This article's evidence fits that rule and does not fit the stronger claim that constructive mutations never happen.  Behe himself scores an acquired resistance gene as a gain of function, which the popular creationist articles deny.³⁷

"If a gene can be tuned to a new substrate in five steps, why can it not be built from scratch in fifty?"  That is the real question and this article does not answer it.  Nothing here settles whether a beta-lactamase could be assembled from non-coding DNA, because nobody has watched that happen.  What the series can say is that eighty years of the strongest selection ever applied have not produced one, which is evidence about rates and not a proof of impossibility.  The one case of resistance from random sequence, Knopp's peptides, worked by discharging the membrane rather than by attacking the drug (Part 3).  The rarity arguments that get aimed at this question are discussed, with their weaknesses, in How Rare Is a Working Protein?.

Issues and unanswered questions

1. No measured share by route. No source read gives a split across clinical resistance as a whole. The 2019 global burden study ranks the deadliest pairings of germ and drug, and with the mechanism reviews that ranking shows transfer carrying most of the top seven, but the study itself does not classify by mechanism and the sorting is this article's.²⁴ A surveillance dataset that classified resistance determinants by mechanism, organism and year would give a real share.
2. The Franklin expedition claim is quarantined, not refuted. The 1988 conference abstract exists and was not obtained. The printed Program and Abstracts of the 28th ICAAC (American Society for Microbiology, 1988) would settle what was actually found.
3. Hughes and Datta 1983 was read as an abstract only. Its main point is now backed by measurement: Houndt and Ochman found no high-level resistance in Murray E. coli apart from one probable contaminant, and Baker's team found chromosomal resistance genes across the sequenced strains.^{8 6} The per-strain gene table is in Baker's Additional file 8, which was not read, so the article gives no count of Murray strains carrying each gene.
4. Liu 2016 was read as an abstract only. The conjugation frequency of 10⁻¹ to 10⁻³ per recipient cell is load-bearing in the ledger and comes from the abstract's Findings section.
5. No creationist paper addresses the TEM extended-spectrum beta-lactamases by name. The general answer exists: Anderson and Purdom's 2008 paper calls a broadened enzyme "biochemically degenerative", and this article engages that answer in Part 3.³² A full-text search of creation.com, the Answers Research Journal and the CRSQ index for "extended-spectrum" and "beta-lactamase" has still not been run. If a specific treatment exists it should be quoted and engaged.
6. The cost-free ribosomal substitution is numbered differently by two sources (K42R in Andersson and Hughes, K424R in Melnyk). The primary, Sander et al. 2002, was not read, so no residue number is printed here.
7. Tuberculosis and malaria are deliberately excluded. katG, rpoB and the tuberculosis ledger belong to Mycobacterium tuberculosis. Chloroquine, pyrimethamine and artemisinin belong to Plasmodium falciparum. The rpoB and gyrA mechanisms are used here as examples only.
8. The five-nucleotide figure for the TEM gain is a count of changed sites, not a measure of information. Whether that is the right unit is the open question of the whole series, and it is argued in Measuring Biological Information, not here.

Verdict

Antibiotic resistance is observed evolution and the creationist literature should stop implying otherwise.  What it is not is a demonstration that mutation and selection build new machinery.  Most clinical resistance is a gene moved from one bacterium to another, and the moving originates nothing.  Most of the rest is a porin lost, a repressor broken, a prodrug-activating enzyme knocked out, or a target protein that no longer grips the drug.  Kevin Anderson's fourteen-row table is right about every row in it.

The exception matters more than its size suggests.  Five point mutations, four in the protein and one in its control region, make a bacterium carrying TEM-1 beta-lactamase about 46,000 times more resistant to cefotaxime, and the same substitutions have arisen repeatedly in hospitals and been recovered in a flask.  Nothing was knocked out.  The key substitution did cost the enzyme about tenfold in efficiency on ampicillin and some stability, which a second substitution repaid, so the design side is right that the gain was a trade.  Calling the whole a loss requires a definition chosen in advance.  So the honest score is five new nucleotides, bought by five sequential searches of roughly 10⁸ cells each, out of eighty years of the heaviest chemical selection ever applied to any organisms on earth.  That is a real gain and a very small one, and both halves of that sentence are the point.  It is also a smaller part of today's clinical picture than it was in the 1990s:  the commonest extended-spectrum enzymes now are the CTX-M family, a Kluyvera gene that already attacked cefotaxime and was moved, though it too has since been tuned one substitution at a time.

The fitness-cost argument should be retired in its popular form.  A 12% mean cost is large enough to matter, but it is absent in some classes, reversed in others, and compensated within a few generations, so it does not show that resistance is degradation.  The argument that survives is Anderson's narrower one and Batten's:  no antibiotic-destroying enzyme has ever been watched arising from DNA that was not already a gene.  The nearest challenge, random peptides that raise aminoglycoside resistance, works by discharging the cell membrane, the same route as a broken respiratory chain.  That claim is true, it is checkable, and it is the one worth defending.

Sources