Mycobacterium tuberculosis

"In Mtb, all drug resistances are the result of chromosomal mutations and depend on the bacterium's capacity for mutation during the course of infection." — Christopher Ford and colleagues, Nature Genetics, 2011¹

"Two independent dating approaches suggest a most recent common ancestor for the M. tuberculosis complex less than 6,000 years ago." — Kirsten Bos and colleagues, Nature, 2014²

Summary

The malaria parasite and HIV are fast.  They mutate at high rates and turn over in days or hours, and a skeptic can fairly ask whether the ledgers this series publishes for them were built by picking the organisms that suit the argument.  Mycobacterium tuberculosis is the test of that objection.  It divides about once a day at best, its mutation rate per base is among the lowest measured for any bacterium, and it does not take up genes from other bacteria.  Every adaptive change in it had to arise by mutation inside its own lineage, which makes its ledger the cleanest among the bacteria.  Each change below is scored with the definitions in Measuring Biological Information.  The general argument about antibiotic resistance is made in Antibiotic Resistance and is not repeated here.

The main findings:

1. The search was large but shallower than malaria's. About 10¹⁹ bacilli were born in the roughly 80 years of the drug era, and they produced about 10¹⁶ new single-letter mutations.³ Every possible single-letter change to the genome arose about a billion times. For comparison, each possible change to the malaria genome arose about a hundred billion times in the same span.⁴
2. The mutation rate is measured, in two independent ways that agree. Whole-genome sequencing of bacteria recovered from infected macaques gives about 2 to 3 × 10⁻¹⁰ per base pair per generation at the laboratory generation time of 20 hours.¹ Sequencing patients months and years apart gives about 0.5 substitutions per genome per year.⁵ These are different quantities and they agree when converted.⁶
3. Resistance was found to every drug ever deployed against it, and not one case is a new gene. Seven drugs are scored below. Two work by having the bacterium switch the drug on, and both are defeated by wrecking the enzyme that does the switching. Four are defeated by a substitution in the target that lowers the drug's grip while the target keeps working. One new drug is defeated mostly by knocking out a repressor so an existing pump runs harder.
4. The clearest single case cuts against a simple "resistance is breakage" story. The commonest isoniazid-resistance allele, katG S315T, keeps 88% of its catalase activity and 83% of its peroxidase activity while losing the side reaction that turns the drug on.⁷ The enzyme is not broken. What it lost is the ability to activate a drug that did not exist before 1952.
5. Fitness costs are real, small for the common alleles, and routinely repaired. Laboratory rpoB mutants all lost ground to their ancestors, with relative fitness from 0.58 to 0.91.⁸ The commonest clinical allele had the smallest cost, and clinical strains carrying it averaged above 1.0.⁸ More than a quarter of multidrug-resistant strains worldwide carry a second mutation elsewhere in the same enzyme that restores fitness.⁹
6. Combination therapy is a working measurement of how expensive two required changes are. Isoniazid resistance arises at 2.56 × 10⁻⁸ per bacterium per generation and rifampicin resistance at 2.25 × 10⁻¹⁰.¹⁰ Both at once is one bacterium in 1.7 × 10¹⁷, and a patient carries about 10⁸.¹¹ That is why two drugs work where one fails, and it is the same arithmetic Michael Behe applied to chloroquine.
7. Over the long run this group's genomes have shrunk. The branches of the M. tuberculosis complex are defined by deletions, and its relative M. leprae has lost more than half of its ancestral genes.^{12 13} Nothing comparable was gained. These are comparisons between genomes, so the losses are inferred and not watched.
8. Ancient DNA dates the whole complex to under 6,000 years, and that result is now the mainstream one. Radiocarbon-dated Peruvian genomes give a common ancestor 4,064 to 4,449 years ago.² Dated European genomes agree.^{14 15} The older figure of 70,000 years was obtained by assuming the bacterium spread with humans out of Africa, because its authors could find no clock signal in modern strains alone.¹⁶ What this dates is one pathogen's common ancestor, not the earth; skeletal evidence of tuberculosis carries older conventional dates, and Part 6 gives them.
9. The ledger is a lower bound, and that weakens the argument this series makes. A quarter of the world carries latent infection, mutation continues during latency, and nobody can count those bacilli.^{1 17} Only changes someone was watching for are on the list.

Part 1: Why this organism

The objection this article answers is fair.  If the only organisms with published ledgers are RNA viruses and a protozoan with a two-day life cycle, the ledgers describe fast mutators and nothing else.  The tuberculosis bacillus sits at the other end of every dial.

It is slow.  Stewart Cole and the team that sequenced the genome in 1998 put it plainly:  "The generation time of M. tuberculosis, in synthetic medium or infected animals, is typically ∼24 hours."¹⁸  Nobody knows the generation time inside a human lung, and Ford and colleagues say so:  "the generation time of Mtb in humans or non-human primates is unknown."¹  Their estimates were therefore computed across generation times from 18 to 240 hours.  Escherichia coli divides in twenty minutes in rich medium;  the tubercle bacillus takes about seventy times longer.

It mutates slowly.  Ford's group infected nine cynomolgus macaques and sequenced 33 isolates recovered from their lesions, along with the strain used to infect them.  Fourteen unique single-letter differences turned up, and no insertions or deletions.¹  Converted to a rate at the laboratory generation time of twenty hours, that is 2.01 × 10⁻¹⁰ per base pair per generation in active disease, 2.71 × 10⁻¹⁰ in latent infection and 3.03 × 10⁻¹⁰ in reactivated disease.¹  A separate laboratory measurement by the fluctuation method gave 2.1 × 10⁻¹⁰.¹  The malaria parasite's measured rate is about the same per base pair, but it completes a generation every two days rather than every one to ten.

It does not swap genes with other bacteria.  This is what makes the ledger clean.  Antibiotic resistance in most hospital bacteria arrives on a plasmid from another organism, so the receiving lineage did no work.  Here there is nothing to receive.  The genome paper found the base composition constant throughout, "indicating that horizontally transferred pathogenicity islands of atypical base composition are probably absent," and concluded that whatever transfer happened, happened "before the tubercle bacillus adopted its specialized intracellular niche."^{18 18}  Ford's group state the consequence in one line:  "In Mtb, all drug resistances are the result of chromosomal mutations."¹

The genome is well mapped.  4,411,529 base pairs, about 4,000 genes and 3,924 open reading frames.^{18 18}

One warning about rates.  Two different numbers are called the mutation rate in this literature and they are not the same thing.  A rate per generation counts errors made when the genome is copied.  A rate per year, measured by sequencing a lineage at two times, counts differences that accumulated and survived.  Ford's 2011 paper measured the first.  Walker's 2013 study of 390 isolates from 254 patients in the English Midlands measured the second and got "0·5 single nucleotide polymorphisms (SNPs) per genome per year (95% CI 0·3–0·7)."⁵  The two agree once converted, because the per-generation rate rises as the assumed generation time lengthens, in just the way that keeps the product constant.⁶  That is Ford's central finding:  "the bacterial populations acquire mutations at the same rate over time, regardless of the number of bacterial replications that have occurred."¹  Eight of the ten mutations found in latent and reactivated animals were of the kind oxygen damage produces, which is the proposed reason.

One complication limits how precise any ledger can be.  The rate itself differs between lineages.  Ford's 2013 study found that "every strain from Lineage 2 acquired resistance to rifampicin (2μg/mL) at a significantly higher rate than every Lineage 4 strain, with a nearly 10-fold difference between the means of the two groups," and about threefold differences for two other drugs, and the difference survived correction for the number of ways each drug can be resisted.^{19 19}  Any single rate used below is an average over strains that differ by up to tenfold.

Part 2: The ledger

An organism's ledger has three entries:  how many individuals lived, how often each mutated, and what the mutations produced.  This part fills in the first two.

How many bacilli

This is the weakest input in the article.  Nobody has counted the bacilli in a human lung;  what has been measured is the concentration in sputum.  Moises Palaci and colleagues cultured sputum from 244 newly diagnosed adults in Vitória, Brazil.  Patients with a lung cavity averaged 5.2 log10 colony-forming units per millilitre, with a range from 1.1 to 7.3;  patients without a cavity averaged 4.0.²⁰  Jacques Grosset, describing where the bacilli that matter actually sit, calls the population in the cavity wall one "ranging well into the millions of organisms."²¹

So the standing population in one adult with active cavitary disease is somewhere between 10⁷ and 10⁹.  This article uses 10⁸ and carries the range through every figure.  Between 40% and 87% of adults have a cavity at diagnosis.²⁰

The World Health Organization estimated 10.7 million new cases in 2024, with an uncertainty interval of 9.9 to 11.5 million.²²  Streptomycin was the first antibiotic used successfully against tuberculosis, and isoniazid was introduced in 1952.^{23 23}  This article takes the drug era as 80 years;  counted from isoniazid alone it is 74, which changes no figure below by more than 8%.  At 10 million cases a year, 80 years is about 8 × 10⁸ cases.  A case lasts about three years untreated, from onset to cure or death, which Edine Tiemersma and colleagues established by reviewing studies from before chemotherapy.²⁴  Most cases in the drug era were treated sooner, so this article assumes one year of active disease per case and notes that three years would raise every total threefold.

How many mutations

The arithmetic for every row is in the note.³  Two independent routes give the same total.  Multiplying bacilli born by the per-generation rate gives 2.6 × 10¹⁶.  Multiplying the standing world population by the measured 0.4 mutations per bacillus per year and by 80 years gives 3.4 × 10¹⁶.  They agree because the per-generation rate and the per-year rate are the same measurement in different clothes.

How thoroughly the genome was searched

The genome offers 1.32 × 10⁷ possible single-letter changes.  At 10¹⁶ mutations, each one arose about 7.6 × 10⁸ times during the drug era, or about ten million times a year.²⁵  Any change that needs one specific mutation is therefore on offer somewhere in the world every day, and in most individual patients:  a patient carrying 10⁸ bacilli that divide daily produces about 7 new mutations of every possible kind per year.

A change that needs two specific mutations in the same cell at once is a different matter.  That happens about once in 2.2 × 10²⁰ bacilli.²⁵  Only about 10¹⁹ bacilli have been born in the whole drug era, so a specific pair arising in one step has probably not happened anywhere, ever, since streptomycin was introduced.  This is not a theoretical claim.  It is the reason tuberculosis is treated with several drugs at once, and Part 4 works it out.

For comparison, the Plasmodium falciparum article counts 1.5 to 7.5 × 10²¹ parasites and about 10¹⁹ mutations in 75 years, so each possible change arose there about 10¹¹ times.  The HIV article counts about 6 × 10²² virus particles and 10²² mutations in about 100 years.  The tuberculosis search is about a hundred times shallower per site than malaria's and far shallower than HIV's.⁴  The question this article exists to answer is whether the output falls in proportion.

Part 3: Drug by drug

Each drug below is a separate experiment.  The bacterium was put under pressure strong enough to kill nearly every cell, and the changes that let some survive were sequenced.

Rifampicin

Rifampicin binds the beta subunit of RNA polymerase and blocks the making of messenger RNA.  Resistance is almost always a substitution in rpoB, the gene for that subunit.  Juan Carlos Palomino and Anandi Martin summarise:  "In about 96% of M. tuberculosis isolates resistant to rifampicin, there are mutations in the so-called 'hot-spot region' of 81-bp spanning codons 507–533 of the rpoB gene."²³  The substitution changes the shape of the pocket so the drug binds less well.  The polymerase goes on transcribing, which is why the bacterium survives at all.  Modification, down.  Zero new nucleotides.

Isoniazid

Isoniazid is a prodrug.  It does nothing until the bacterium's own catalase-peroxidase, KatG, turns it into the compound that blocks mycolic acid synthesis.²³  The bacterium can therefore escape by losing the enzyme, and some strains do exactly that.  But the commonest resistant allele by far does something more interesting, and it is the hardest case in this article for the young-earth side.

Alexander Pym, Brigitte Saint-Joanis and Stewart Cole built isogenic strains carrying single katG alleles and measured the enzyme.  The S315T allele, which "at least 40% of all INH-resistant clinical isolates" carry, showed "only a small reduction in the catalase and peroxidase activities (12 and 17%, respectively)" against the wild-type control, and the strain stayed virulent in mice.^{7 7}  A different allele, T275P, lost the enzyme, and that strain was crippled.  Their summary is worth quoting in full because it states the cost-benefit exactly:

"it is unfortunate for the treatment and control of MDR-TB that M. tuberculosis can so successfully balance the competing demands of resistance to host killing and resistance to INH."⁷

The enzyme is not broken.  It has lost one side reaction on a molecule that did not exist before 1952, while keeping the work it does for the cell.  Scored on what the molecule does, the class is modification, down, because a reaction the enzyme used to catalyse is now catalysed less.  Scored on what it costs the organism, it is nearly free.  Zero new nucleotides.

The second route to isoniazid resistance is a change in the promoter of inhA, the gene for the enzyme the activated drug attacks.²³  More enzyme is made, so the same amount of drug blocks a smaller share of it.  That is an existing promoter made stronger.  Modification, up.  One nucleotide, by the conversion rule in Measuring Biological Information.

Pyrazinamide

Pyrazinamide is the second prodrug in the regimen.  It has to be converted to pyrazinoic acid by the bacterium's own pyrazinamidase, encoded by pncA.²³  Here the bacterium does simply wreck the enzyme, and the pattern of mutations shows it.  Palomino and Martin:  "Mutations in the gene pncA remain as the most common finding in pyrazinamide resistant strains.  These mutations, however, are scattered throughout the gene but most occur in a 561-bp region in the open reading frame or in an 82-bp region of its putative promoter."²³

Scattered is the diagnostic word.  When a function has to be preserved, the changes that survive selection cluster in a few positions, as they do in rpoB.  When a function has to be destroyed, almost any change will do, and they spread across the whole gene.  Loss.  A negative count.

Ethambutol, streptomycin and the fluoroquinolones

Three more targets, three more substitutions in a protein or RNA the bacterium already had.

Ethambutol blocks the building of the cell wall sugar arabinogalactan.  Resistance is linked to embB, "with mutations at position embB306 as the most prevalent in most of the studies performed."²³  The same paper notes that embB306 changes also turn up in susceptible isolates, so the association is not clean.  Modification, direction unmeasured.  Count left empty.

Streptomycin jams the start of protein synthesis at the small ribosomal subunit.  Resistance comes from rpsL, the gene for ribosomal protein S12, and from rrs, the 16S ribosomal RNA gene, and "account for 60%–70% of the resistance found."²³  The commonest change is a single substitution at codon 43 of rpsL.  The ribosome still translates.  Modification, down.  Zero.

The fluoroquinolones block DNA gyrase.  Resistance is "by chromosomal mutations in the quinolone resistance-determining region of gyrA or gyrB," most often at position 90 or 94 of gyrA.²³  The gyrase still unwinds DNA.  Modification, down.  Zero.

Bedaquiline, the newest drug

Bedaquiline was approved in 2012, the first genuinely new tuberculosis drug in forty years.  It blocks ATP synthase.  Resistance to it is the most instructive case in the article, because the bacterium was watched meeting a molecule nothing had ever met before.

Koen Andries and colleagues found two routes.  Substitutions in atpE, the gene for the drug's target, were the obvious one, but in an earlier fluctuation study "target-based mutations were identified in only 30% (15/53) of the isolates."²⁶  The rest had knocked out Rv0678, a repressor:  "non-target based resistance to BDQ, and cross-resistance to clofazimine (CFZ), is due to mutations in Rv0678, a transcriptional repressor of the genes encoding the MmpS5-MmpL5 efflux pump."²⁶  Breaking the repressor leaves the pump running, and the pump throws the drug out.  Andries found no fitness cost for these changes in broth or in mice.²⁶

So the commonest answer to a brand-new drug was to break an off-switch on a transporter the bacterium already had.  atpE:  modification, down, zero.  Rv0678:  loss, with an existing pump left switched on.  A negative count.

One further result belongs here, because it is the strongest point the creationist literature could make about this drug and nobody on that side has made it.  Cristina Villellas and colleagues sequenced Rv0678 in baseline isolates taken before any patient received bedaquiline.  Variants in that gene were present in 23 of 347 multidrug-resistant isolates, 6.3%, against 6 of 852, 0.7%, in a general population sample.²⁷  None of those patients had been given bedaquiline, and their own conclusion is that "the origin of these 'naive' Rv0678 RAVs is therefore elusive at this point in time."²⁷  The variation preceded the drug.  It must be said in the same breath that most of those isolates were not resistant:  of the 23, eight had raised drug concentrations needed to stop growth, eleven were normal and four were low.²⁷  Standing variation in a repressor is not the same thing as standing resistance.

Compensation, which is the real second act

The interesting evolution in this organism is not the resistance mutation.  It is what happens afterwards.

Sebastien Gagneux and colleagues competed rifampicin-resistant strains against their own susceptible ancestors.  "All of the rifampicin-resistant mutants had a statistically significant fitness cost compared with the susceptible ancestor, which is defined to have a competitive fitness of 1.0," and the values ran "from 0.58 to 0.91."^{8 8}  Then they looked at what patients actually carry.  The allele with the smallest laboratory cost, S531L, "is the most prevalent rifampicin resistance-conferring mutation, overall accounting for 54% of rifampicin-resistant isolates," while the costliest allele "has never been observed in clinical settings."⁸  Four of five clinical strains carrying S531L had no fitness cost at all.⁸

Iñaki Comas and colleagues then sequenced paired resistant and susceptible isolates from the same patients and found why.  Second mutations turn up in rpoA and rpoC, two other genes of the same RNA polymerase, in resistant strains and in no susceptible ones:  "89/329 (27.1%) of all MDR strains harboured a non-synonymous mutation in rpoA or rpoC.  ...  By contrast, none of the 171 rifampicin-susceptible control strains harboured any of these mutations."⁹  Three of four multidrug-resistant strains with no measured fitness defect carried one;  none of the six with a significant defect did.⁹  In the countries with the worst multidrug-resistant epidemics the share rises above 30%.⁹

How to score compensation is a genuinely open question for this series.  The second mutation raises the fitness of the strain that carries it, which by the conversion rule makes it a modification, up, worth one nucleotide.  But nothing new is being done.  The second change repairs damage the first one caused, and the compensated strain is still worse at its job than the bacterium was before either mutation.  This article counts these as modifications and records the judgement call in the issues below.

The ledger in one table

Ten scored changes.  Two are losses.  Eight are modifications, of which two are upward.  None is a gain.  No new gene, no new protein, no new binding between two proteins, and no case in which the bacterium can now do something no member of its species could do before, except survive a molecule invented in the twentieth century.

Part 4: What combination therapy measures

In 1970 Hugo David, at the Center for Disease Control in Atlanta, ran fluctuation tests on M. tuberculosis for four drugs.  His abstract gives the result:

"The fluctuation test shows that Mycobacterium tuberculosis mutates to resistance to isoniazid, streptomycin, ethambutol and rifampin spontaneously and at random.  The average mutation rates for the drugs, in the same order, were calculated to be 2.56 × 10⁻⁸, 2.95 × 10⁻⁸, 10⁻⁷, and 2.25 × 10⁻¹⁰ mutation per bacterium per generation."¹⁰

These are per-drug rates, not per-site rates.  They add up every mutation anywhere in the genome that confers resistance to that drug, which is why ethambutol's is the highest, at 10⁻⁷:  many changes will do.  Rifampicin's is the lowest, at 2.25 × 10⁻¹⁰, because the drug binds one small pocket and only a few substitutions in it work.

Now multiply.  A bacterium resistant to isoniazid and rifampicin at once, by two independent mutations in one cell, arises at 2.56 × 10⁻⁸ × 2.25 × 10⁻¹⁰ = 5.8 × 10⁻¹⁸, or one bacterium in 1.7 × 10¹⁷.¹¹  A patient with cavitary disease carries about 10⁸.  The doubly resistant mutant is therefore about a billion times rarer than the population that would have to contain it.

That is the whole basis of combination therapy, and it has worked for seventy years.  It is also a direct measurement of something this series argues about constantly:  what it costs to get two specific changes when neither one alone is enough.  Michael Behe made the same argument about chloroquine resistance in the malaria parasite in 2007, and the Plasmodium falciparum article concludes that he was right about the price and about the reason for it.  Tuberculosis clinicians reached the same arithmetic thirty-seven years earlier, for a different purpose, and built a treatment regimen on it.

The other half of the lesson is the half that cuts against the design argument, and it has to be stated with the same force.  Multidrug-resistant tuberculosis exists.  The World Health Organization estimated 390,000 new cases of rifampicin-resistant or multidrug-resistant disease in 2024.²⁸  Those strains carry both mutations.  They got them the cheap way:  one at a time, each under its own drug, each selected to near fixation inside a patient before the next arrived, and then passed to the next patient already resistant.  Two changes that are impossible together are easy in sequence when each one pays on its own.

That is the same contrast the Plasmodium falciparum article draws between chloroquine, where the first step is harmful and the price was about 10²⁰ parasites, and pyrimethamine, where every step pays and resistance appeared inside single patients.  The design argument is not that two changes cannot happen.  It is that two changes paying nothing until both are present are priced out at these population sizes, and the bacterium's own record is the measurement.

Part 5: The long run, which was mostly downhill

Everything above was watched.  This part is not.  What follows comes from comparing genomes and reading the differences as history, which assumes common descent inside the group.  A young-earth reader can grant that much within a bacterial genus without granting anything else.

The branches of the complex are defined by deletions.  Roland Brosch and colleagues screened 20 variable regions across 100 strains of five named species and found that the presence or absence of each region marks a branch rather than turning up independently:  "the majority of these polymorphisms did not occur independently in the different strains of the M. tuberculosis complex but, rather, resulted from ancient, irreversible genetic events in common progenitor strains."¹²  Fourteen of the regions, from 2 to 12.7 kilobases, are missing from the BCG vaccine strain.¹²  The cattle pathogen M. bovis is the end of a chain of such losses, and the paper is blunt that nothing came the other way:  its near-complete genome "revealed no new gene clusters that were confined specifically to M. bovis."¹²  That killed the old idea that human tuberculosis came from cattle.  Their own reading of what the losses did is worth quoting, because it is not what a creationist would predict either:  "Successive loss of DNA may have contributed to clonal expansion and the appearance of more successful pathogens in certain new hosts."¹²  Losing genes made it better at its job.

The nearest relative outside the complex has lost half its genome to decay.  Cole and the same group sequenced M. leprae, the leprosy bacillus, and called it "an extreme case of reductive evolution":  3.27 million base pairs against tuberculosis's 4.41 million, and "Less than half of the genome contains functional genes but pseudogenes, with intact counterparts in M. tuberculosis, abound."^{29 29}  Whole systems are gone:  "siderophore production, part of the oxidative and most of the microaerophilic and anaerobic respiratory chains, and numerous catabolic systems and their regulatory circuits."²⁹  Laura Gómez-Valero and colleagues reconstructed the common ancestor of the two and put numbers on it:  "During the reductive evolution of M. leprae, 1537 of 2977 ancestral genes were lost, among which we found 177 previously unnoticed pseudogenes."¹³  Most of that DNA is still sitting in the genome, decaying:  about 89% of the nucleotide content of the lost genes remains.¹³

Two cautions belong with those figures.  The 1,537 counts ancestral genes no longer working, not deleted DNA;  most are damaged copies still present.  And the ancestral inventory of 2,977 is itself a reconstruction, blind to genes both lineages lost.

What was gained over the same span?  Nothing of comparable size has been reported.  The published history of this group, read from its genomes, is a nested series of deletions running from a broader ancestor down to obligate pathogens that cannot live outside a host, with M. leprae unable to be grown in the laboratory at all.²⁹  That is what the creationist account of pathogens predicts.  Andrew Fabich, a microbiologist writing for Answers in Genesis, states the prediction in a testable form:  "pathogens (i.e., disease-causing germs) have only developed recently because of the Fall," and "Due to Adam's sin, all of creation was cursed to varying degrees; the genomic deterioration of microbes began at that time."^{30 30}  On the genomic half of that claim, the tuberculosis and leprosy data fit.

The honest qualification is that the fit proves less than it looks.  Reductive evolution in host-restricted bacteria is standard mainstream biology, predicted by population genetics rather than by theology:  a small population locked in a constant environment loses what it stops using.  The observation is agreed by both sides.  What is disputed is the timescale, which is Part 6.

Part 6: How old is the complex

Two published dates for the common ancestor of the M. tuberculosis complex differ by a factor of ten.  The difference is entirely a difference of method, both sides say so in print, and the younger date is the one obtained by measuring rather than assuming.

The old date, 70,000 years, was calibrated on human prehistory.  Iñaki Comas and colleagues used 259 modern genomes and no ancient DNA.  They wanted to use sampling dates and could not, and say why:  "when performing a tip-to-date analysis of those strains (N = 49), we found that in contrast to several other bacterial species, no significant correlation between isolation time and phylogenetic divergence exists in MTBC (correlation coefficient = 0.047)."¹⁶  So they calibrated internal nodes instead:  "we used as initial calibration points several key dates in human evolution," and the model they preferred tied the split of the two most basal lineages to the human mitochondrial L3 group at 70,000 years.¹⁶  The output was an age of about 70,000 years and a rate of 2.58 × 10⁻⁹ substitutions per site per year.¹⁶  That rate is roughly twenty times slower than the rates measured in outbreaks, and their explanation is that measured rates decay as the interval lengthens.¹⁶

The young date, under 6,000 years, was calibrated on radiocarbon.  Kirsten Bos and colleagues recovered three genomes from Peruvian skeletons about a thousand years old and used their radiocarbon dates as calibration points.  The measured rate was 4.6 × 10⁻⁸ substitutions per site per year (95% highest posterior density 3 to 6.2 × 10⁻⁸), about 0.20 per genome per year, which they note "agrees well with estimates of MTBC evolution in modern epidemiological contexts, and is more than tenfold faster than those using human dispersals out of Africa as calibration."²  The age that follows is "4,449 years before present (yr BP) (2,990–6,062 yr BP 95% HPD)" under a constant-size model and "4,064 yr BP (2,951–5,339 yr BP 95% HPD)" under an expanding one.²  Their own comment on the result:  "Our MRCA, which is at least an order of magnitude younger than previous estimates, presented us with a challenge."²

Two further datasets agree, and they do not depend on the Peruvian strains.  Gemma Kay and colleagues sequenced 14 genomes from a crypt in Vác, Hungary, with documented dates of death, and got a rate for Lineage 4 of 5.00 × 10⁻⁸ substitutions per nucleotide per year.  Their conclusion is careful and states the counter-evidence in the same sentence:  "This mutation rate is consistent with the hypothesis that the most recent common ancestor of the M. tuberculosis complex existed <6,000 years ago, but is inconsistent with the recovery of amplification products indicative of sub-lineages within the complex from Neolithic samples."¹⁴  Susanna Sabin and colleagues added a genome from a calcified lung nodule in a seventeenth-century Swedish bishop and reported "a most recent common ancestor date for the MTBC of between 2190 and 4501 before present."¹⁵  They also name the weakness in Bos's calibration rather than hiding it:  the Peruvian strains belong to the seal-adapted lineage, and "identical substitution rates between the pinniped lineage and human-adapted lineages of the MTBC cannot be assumed."¹⁵

The systematic study sides with the young date.  Fabrizio Menardo, Sebastián Duchêne, Daniela Brites and Sebastien Gagneux applied one set of methods to 6,285 genomes.  Gagneux was also the last author on Comas 2013, so this is the group revisiting its own result.  They obtained "a clock rate between 1x10^-8 and 5x10^-7 nucleotide changes per-site-per-year (0.04–2.2 SNPs per-genome-per-year), with substantial differences between clades," and an age for the complex "between 2,828 and 5,758 years old."^{31 31}  Their verdict:  the results "support the results of Bos et al. 2014 that found the MRCA of the MTB complex to be relatively recent, and not compatible with the Out-of-Africa hypothesis."³¹

What this does and does not show

A measured clock, calibrated on radiocarbon-dated samples, puts the common ancestor of every sampled member of this complex inside the last six thousand years, and inside the last four thousand five hundred on the best estimates.  A young-earth reader should notice that.  It is also narrower than it first appears, in four ways.

It dates one common ancestor, not the disease.  Menardo's group make the point against their own conclusion:  "it is also possible that extinct lineages were circulating and causing diseases much earlier."³¹

Older skeletal and biomarker evidence exists, and it is dated by radiocarbon, not by a clock.  Israel Hershkovitz and colleagues examined the bones of a woman and infant with tuberculosis lesions from Atlit-Yam, a village now submerged off Haifa, where "Calibrated radiocarbon dates range from 9250-8160 years BP."  They recovered DNA at five tuberculosis loci and confirmed it independently by detecting the bacterium's own mycolic acid lipids by chromatography.^{32 32}  Older claims exist:  Menardo's review notes markers reported "in a bison's bone about 17,500 years old."³¹  The same review records the standing objection, that some markers are not specific and some samples may be contaminated by environmental mycobacteria.³¹  This is a real conflict inside the mainstream literature and it is not resolved.  It should be reported as such rather than settled in whichever direction suits an argument.

The authors do not draw a young-earth conclusion and would reject one.  Sabin's paper calls its own result a "Neolithic emergence," which is what 2,190 to 4,501 years before present means on the conventional chronology of agriculture.  The finding constrains one bacterium and says nothing about the age of the earth.

The rate and the age move together.  If the clock really slows over long intervals, as Comas and colleagues argue, the young dates are too young.  Their evidence is a correlation across studies of different pathogens over different intervals.¹⁶  Against that, three independent ancient-DNA calibrations spanning 300 to 1,000 years all give rates near 5 × 10⁻⁸ and ages near 4,000 years.  The weight of measurement is on the young side and the weight of assumption on the old side.

What the creationist literature has said

Very little, and not about this.  A full-text search of creation.com's index for "tuberculosis" and "Mycobacterium" returns one article on the subject:  Tas Walker's response to a claim of tuberculosis in a 500,000-year-old skull.  His answer is about dating, not genetics:  "This fossil is not a theological problem but a dating one.  The researchers did not measure its age," and the skull "most likely, is post-Flood ... it would be less than 4,500 years old."^{33 33}  It was published in 2007, seven years before Bos.  Answers in Genesis has a blog post by Ken Ham on the BCG vaccine strain, correctly noting that the strain "lost something: this is not molecules-to-man evolution," which is what Brosch's deletion map shows.³⁴  Fabich's article on pathogens names tuberculosis among the ancient diseases to be explained but supports the claim with a footnote to a commentary on drug-resistance genetics that says nothing about antiquity.³⁰

No creationist source found for this article discusses Bos 2014, Kay 2015, Sabin 2020 or Menardo 2019.  That is a missed opportunity:  a measured clock giving a common ancestor about four thousand years old is the best result in this literature for a short chronology, and it came from people answering a question about pre-Columbian America.

Part 7: What cannot be priced, and what the ledger buys

Latency.  About a quarter of the world's population has been infected, and 5 to 10% of those infected ever develop disease.¹⁷  Ford's macaque study found that bacteria from latently infected animals "acquire mutations at a similar rate to rapidly replicating bacteria in vitro."¹  Nobody knows how many bacilli a latently infected person carries.  At a thousand each, two billion carriers hold 10¹² bacilli, negligible beside the 10¹⁵ in active disease.  At a million each they hold 10¹⁵, which doubles the ledger.  The cell is left empty on purpose.

Everything nobody looked for.  Drug resistance is found because patients stop getting better.  A change that let the bacterium survive one more day in a macrophage, or transmit a little better, would leave no such signal.  One such difference is already known:  Lineage 2 mutates about ten times faster than Lineage 4 for one drug, and that is itself an evolved property nobody has priced.¹⁹  The ledger counts what was watched, so it is a lower bound.  That cuts against the argument this series makes and cannot be fixed with the data that exist.

Unique functional sequence created.  The series measures nucleotides that had to be originated rather than copied, per organism born.  For this bacterium the numerator is the inhA promoter change, worth one nucleotide, and the compensatory substitutions in rpoA and rpoC, worth one each if they are counted at all.  Comas and colleagues list 28 distinct compensatory changes beyond those already known, so a generous numerator is about 30 nucleotides and a strict one is 1.⁹  Divided by 10¹⁹ bacilli, the rate is between 10⁻¹⁹ and 3 × 10⁻¹⁸ nucleotides per bacillus.³⁵

Set that beside the siblings.  The malaria parasite's figure is 2 × 10⁻²¹ to 7 × 10⁻²⁰ per parasite, and the HIV figure computed the same way is about 8 × 10⁻²⁰ per virus particle.  The slow organism's rate per individual is not lower.  It is higher, by one to two orders of magnitude, and that is not a point in the argument's favour.  The reason is arithmetic:  the numerators are all small handfuls of nucleotides, and tuberculosis simply has the smallest denominator.

That is the methodological result this article was written to test.  The benchmark is not an artefact of picking fast mutators, but not because the slow organism produced proportionally less.  It is because the numerator hardly moves.  Across an RNA virus with the highest known mutation rate, a protozoan and a slow clonal bacterium, the observed output of adaptive evolution under the strongest selection humans can apply is the same short list every time:  a few substitutions in existing proteins, some deletions, an occasional promoter turned up, and no new gene.

Objections and responses

Eighty years is a tiny slice of time.  Evolution is not supposed to build a new gene in a human lifetime.  Correct, and the argument does not claim otherwise.  The ledger provides a measured rate, and the extrapolation is the point:  10¹⁹ organisms under the strongest selection available produced no new coded element.  A skeptic who grants that and asks for more organisms and more time is making the right reply, and it is answered by the size of the numbers, not by the length of the experiment.

Drug resistance is a narrow target chosen after the fact.  This succeeds in part, and Part 7 concedes it:  only what someone measured is on the list.  It does not rescue the individual cases.  A drug that kills nearly every bacillus is about the strongest selection a population can face, the whole genome was free to answer it any way it could, and what came back was substitutions in existing targets and a wrecked prodrug-activating enzyme.

Loss can be adaptive, so calling these changes losses is a rhetorical trick.  Agreed, and this article says so twice.  Brosch's paper concludes that successive loss of DNA produced "more successful pathogens."¹²  Breaking pncA is a good move for a bacterium facing pyrazinamide.  The class describes the molecule, and the effect on the organism is reported beside it, never instead of it.

The katG case shows the scheme bending to fit.  This is the strongest objection in the article.  S315T keeps 88% of its catalase activity and is scored a loss of one specific reaction.  A critic can fairly say that a protein which resists a drug at a cost of 12% of its activity is a protein that has been improved for its environment, and that "modification, down" hides this.  Two answers.  First, the direction is recorded honestly:  the reaction the enzyme performs less well is named, and the cost to the organism is given as measured.  Second, the class does not depend on how small the cost is.  Nothing new is encoded.  If a reader prefers to score it a modification with no direction, the ledger's headline does not change:  no new coded element appeared.

Compensatory mutations are real innovation.  Something is improved that was not improved before.  They are the best candidate in this article, which is why they get their own section and an open question.  What weighs against them is that the improvement is relative to a damaged starting point.  A compensated resistant strain still does worse than the susceptible ancestor at everything except surviving rifampicin.  Counting a repair as origination would inflate the ledger in a way that would be unfair in the other direction.

The 6,000-year date is a clock result, and this site attacks clock results everywhere else.  A fair challenge.  The answer is the difference between a clock calibrated on assumed history and a clock calibrated on dated samples.  Comas 2013 fixed a node at 70,000 years because it supposed the bacterium travelled with humans, then read the rate off that assumption.  Bos 2014, Kay 2015 and Sabin 2020 dated the samples first and measured the rate from them.  Where they conflict the measured route should win, and that is the same principle this site applies against clock-derived dates elsewhere.  Note also what is being accepted here:  the radiocarbon dates of the calibrating samples, which are all inside the last few thousand years and inside the range where radiocarbon is least disputed.

Then this article accepts a 9,000-year radiocarbon date for the Atlit-Yam skeletons.  It reports it, which is not the same thing, and reports the objection to the molecular identification too.  A chronology placing the Flood around 2350 BC has a real problem with a 9,250-year-old settlement, and pretending otherwise would be dishonest.  How far back radiocarbon dates stay reliable is argued at Radiocarbon, and this article does not settle it.

Bacteria are not animals.  A bacterium with one copy of each gene cannot hide a harmful first step, so of course two-step changes are rare in it.  This is the strongest scientific objection to the extrapolation and it is correct as far as it goes.  Animals carry two copies of each gene and can hide a harmful mutation behind a working one, which should make some two-step changes easier.  Against that, animals live in populations of thousands to millions rather than 10¹⁹, and a rare first step in one species cannot meet a rare second step in another.  The net has not been computed for a case like this and is listed below as an open question.

Issues and unanswered questions

1. The number of bacilli per patient is an estimate, not a measurement. Palaci 2007 measured a concentration in sputum and Grosset 2003 says "millions" in the cavity. The step from those to 10⁸ per patient is this article's, and a hundredfold error in it moves every total in Part 2 by the same factor. Canetti's classic lesion counts from the 1950s were not obtained.
2. Latent infection cannot be priced. The number of bacilli in a latent infection is unknown, and mutation continues there.
3. The score for compensatory mutations is unsettled. They raise fitness, which the conversion rule counts, but they repair rather than originate. This is a question for Measuring Biological Information.
4. Whether two copies of each gene make two-step changes easier by enough to matter has not been computed here. The inputs would be population size, dominance, and the fitness of the intermediate. The literature is reviewed in Haldane's Dilemma.
5. Ethambutol's direction of effect was not measured and its cell is empty.
6. The conflict between ancient-DNA clock dates and Neolithic marker evidence is unresolved in the mainstream literature, and this article reports rather than resolves it.
7. Second-line drugs were not researched. Amikacin, kanamycin, ethionamide, linezolid, pretomanid and delamanid are missing from the table, and some are prodrugs whose activation can be lost, which would probably add more losses.
8. The 1970 rates were measured in one laboratory on unspecified strains and predate the discovery that lineages differ in mutation rate by up to tenfold.

Verdict

Mycobacterium tuberculosis is the slow end of the range, and it gives the same answer as the fast end.  About 10¹⁹ bacilli were born in the drug era and produced about 10¹⁶ mutations, enough for every possible single-letter change to arise a billion times.  What that search found, under the strongest selection medicine can apply, is ten scored changes:  two losses, eight modifications, no gains, and not one new gene, protein or regulatory element.  The best case against a simple breakage story is katG S315T, an enzyme that resists the drug while keeping most of its own work, and it is a change of two nucleotides at most in a gene that already existed.

The benchmark therefore survives the objection it was built to test.  The pattern is not an artefact of choosing fast mutators.  But the honest form of the result is narrower than the argument would like:  the rate of new functional sequence per organism is not lower in the slow bacterium.  It is higher, because the output stayed the same size while the population shrank.  What is constant across HIV, the malaria parasite and this bacterium is the numerator, and the numerator is always a handful of nucleotides.

The dating result is the article's second finding and may be its more important one.  Three independent calibrations on radiocarbon-dated ancient genomes agree that every sampled member of the complex descends from an ancestor about four thousand years old, and the systematic reanalysis by the group that published the older date now agrees.  That is one pathogen's family tree, dated by measurement rather than assumption.  It is not the age of the earth, and older skeletal evidence carries older conventional dates that remain unexplained.  Anyone building on it should say exactly that much and no more.

Sources