Phage Lambda and the OmpF Switch

Summary

A virus called lambda normally gets into Escherichia coli through one protein on the cell surface, LamB.  In 2012 Justin Meyer, then a graduate student in Richard Lenski's laboratory at Michigan State University, and five colleagues reported that lambda evolved to get in through a second protein, OmpF, and kept the first route as well.  It needed four changes in one gene to do it.  The change happened in one of six flasks in the first experiment, by day 8, and in 24 of 96 flasks in a larger one, at a median of 12 days.¹

This article answers one question:  did the four changes have to arrive together, or were they found one at a time because each one paid?  The answer is settled, and it is the second.  The four are jointly required for the new receptor.  Phage carrying three of the four cannot use OmpF at all.¹  But the earlier changes were not waiting for each other.  They spread because they made the virus better at binding the receptor it already used.  Meyer's paper argued this from the pattern of mutations in 2012.  Alita Burmeister measured it in 2016:  all six intermediate phage she tested were fitter than their ancestor, and all six adsorbed faster to LamB.²  Katherine Petrie's group found in 2018 that even the last of the four improved binding to the old receptor.³

Six findings follow.

1. The catalogue description for this article said the four mutations were "coordinated." That is wrong and this article does not say it. The correct phrase is jointly required, individually selected. The distinction is the whole substance of the case.
2. The search that found the switch was small. A successful flask made roughly 10⁸ to 10¹¹ phage before the switch appeared, depending on which published density is used. That is enough for every single point mutation in the 48,502-base genome to arise a few times to a few thousand times over, and nowhere near enough for four specified changes to arrive in one replication. The arithmetic is in Part 3.
3. This case does not test Michael Behe's limit, and Behe says so himself. His 2019 reply calls the lambda function "neither irreducibly complex nor beyond the edge of evolution."⁴ The Edge of Evolution treats paths whose every step helps as ordinary evolution.⁵ Richard Lenski's "strike two" against the book drops that condition, and to that extent overstates the case.⁶
4. It does contradict two narrower claims in that book. One is the prediction that a virus which learned to enter cells by binding a different surface protein would out-compete its siblings, and "That hasn’t happened."⁵ In the laboratory it happened 25 times. The other is the premise that three or four of the amino-acid changes needed for a new binding site "must occur together."⁵ Here they did not. Behe's own endnote had already conceded that a virus binding a cell-surface protein is an easier target than a binding site inside a cell.⁵
5. Scored with the series taxonomy, the switch is a gain, and a hard case. A binding activity now works that did not work before, which is question 3 of Measuring Biological Information Behe scores it a modification, and his 2010 rule genuinely says so, on one condition: the binding has to occur "at the same site on the protein."⁷ Nobody has measured where on the J protein OmpF binds. Behe's confidence rose from "strongly suggests" in 2012 to "precisely because" in 2019 with no new structural data in between.^{8 4}
6. The case cuts against the young-earth and design reading in one specific way, and that is stated plainly in Part 6. Real adaptive paths sometimes do have selectable intermediates, and when they do, a change needing four specified substitutions is cheap. The honest reply is not that intermediates never pay; it is that nobody on either side has a census of how often they do, and the one measured example had a new target that was the closest structural match to the old one in the whole cell.¹

Two host facts belong in the summary because the phage could not have done it alone.  The bacteria had to break their own malT gene first, which cut LamB production and made the phage desperate for another way in.¹  And bacteria that broke manY or manZ instead blocked the innovation completely:  in Animesh Gupta's 2022 replays, populations started on an already-resistant host produced the switch 0 times in 12, against 3 times in 12 on the normal host.⁹

Part 1: What was done

Lambda is the most studied virus that infects bacteria.  The strain Meyer used, cI26, cannot hide in the host chromosome;  it can only kill.  Its tail ends in a protein called J, 1,132 amino acids long, and J recognises LamB.¹⁰  LamB is a porin, a channel in the outer membrane that lets the sugars maltose and maltotriose through.  When E. coli grows on glucose it has little use for LamB and makes little of it.

Six flasks, each holding 10 ml of glucose minimal medium, were started with about 100 phage and about 1,000 bacteria.  Every day, 100 µl went into 9.9 ml of fresh medium, a 1 in 100 dilution.  The first experiment ran 28 days;  the larger one ran 20 days in 96 flasks.¹¹

The bacteria took the first step, and it was a loss.  They evolved resistance by mutating malT, the regulator that switches on lamB.  Those mutants fixed within 8 days.  They also pay for it:  a broken malT cannot grow on maltose.  The supplement notes that malT mutations are useful in glucose anyway, even with no phage, "probably because they reduce the basal expression of unnecessary gene products," which is why the bacteria took that route rather than mutating lamB itself.¹¹

The phage did not die out.  It held at about 10⁶ to 10⁷ per ml, roughly one phage per 100 to 1,000 cells, living on the few cells that happened to switch LamB on.¹  That is the condition Meyer set up on purpose:  a virus that could reach the other 99-odd percent of cells would win big.

In one of the six flasks it did.  By day 8 about 0.01% of the phage could infect cells with no LamB at all, and by day 15 most of them could.¹  Knockout strains identified the new receptor:  the evolved phage, EvoC, was stopped only by a host missing both OmpF and LamB, so OmpF was the new door.¹  It could still use LamB.¹

EvoC's genome carried five mutations, all five in J.¹  The 96-flask experiment then gave 24 more independent origins, all of them through OmpF, at a median of 12 days and a range of 9 to 17.¹

Part 2: Which mutations, and what makes them "all-or-none"

Meyer sequenced J from the 24 phage that gained the new receptor and from 24 that did not, matched day for day.  There were 241 single-nucleotide differences across the 48 alleles, no insertions or deletions, and only 40 distinct mutations, because the same ones kept recurring.  Every one changed an amino acid.  Phage that could use OmpF averaged 6.63 mutations in J;  phage that could not averaged 3.42.¹

Four features were shared by all 24 winners:  A3034G, G3319A, a change at either 3320 or 3321 (the same codon as 3319, residue 1107), and at least one change somewhere between positions 2969 and 2999.¹  Each of the four was also present in at least one loser, but no loser had all four.¹  Three losers had three of the four and still could not touch a lamB mutant.¹  Two assays confirmed it:  only the isolate D7, with all four and nothing else, adsorbed to lamB-negative cells at a measurable rate and grew on them.¹¹  Meyer called this "all-or-none" epistasis.¹

Rohan Maddamsetti's group tested the requirement directly in 2018 by building the combinations with genome engineering rather than waiting for them.  Every engineered phage that used OmpF had the change at residue 1012 and both changes at residue 1107, which together turn an aspartic acid into a lysine.  Those three alone were not enough:  a phage carrying only them was alive but could not use OmpF.  A fourth mutation was needed, and many different ones would serve.^{12 12}

So the target is slightly larger than "four specified mutations."  It is three specified nucleotide changes, two of them in one codon, plus any one of several others.  That makes the switch easier to reach than the 2012 paper implied, not harder.  It is the honest version of the number and this article uses it.

Note what all-or-none epistasis does and does not mean.  It means no subset of the four confers any OmpF use.  It does not mean the subsets are useless.  Those are different claims, and the confusion between them is the reason this case gets cited by both sides.

Part 3: The ledger

How much searching bought the switch?  The inputs are all published, and every one of them is a range rather than a number, so the answer is a range.

With a 1% daily transfer, each day's growth replaces the whole flask, so births per flask per day are about one flask-full.  Twelve days of that gives the figures below.¹⁵

Two things fall out of the table.

The first is that single mutations were free.  Every one of the roughly 10,000 possible point mutations in J arose hundreds or hundreds of thousands of times across the experiment.  A search that samples every one-step neighbour many times over is not searching;  it is enumerating.

The second is that four specified changes at once were impossible, and nobody claims otherwise.  The expected number of phage carrying a specified quadruple by simultaneous mutation, across the whole 96-flask experiment, is about 10^-20.  This is exactly the sort of number Behe's argument is built from, and it is why the stepwise question matters so much.  If the intermediates had not paid, the switch would not have appeared, and 24 out of 96 would have been 0 out of 96.

Unique functional sequence created.  Four nucleotides were changed, and they are not copies of anything;  no gene was duplicated.³  Against 1.2 × 10⁸ to 7.6 × 10¹⁰ phage-generations per successful flask, that is 3 × 10^-8 to 5 × 10^-11 new functional nucleotides per organism-generation.  Both bounds are for the successful flasks only.  Counting the 72 flasks that produced nothing lowers the figure about fourfold.  The denominator is a real measured population, which makes this one of the few entries in the series where the cell can be filled at all.

What the ledger cannot price:  the improvements in LamB binding that came first.  Those are real adaptive changes, they were measured, and they are not counted above because their contribution cannot be converted into new nucleotides without double-counting the four.  The figure is therefore a lower bound on what this system produced, which cuts against the young-earth side of the argument.

Part 4: Jointly required, individually selected

This is the part of the case that matters, so here is every line of evidence, in order of publication.

2012, from the pattern.  Meyer's argument was indirect and he said so.  Three facts pointed to selection on J before the new function existed.  All 248 mutations across 51 sequenced alleles changed an amino acid, where chance alone predicts about 3.19 silent-to-changed.  The mutations concentrated in the C-terminal quarter of J, the part known to contact LamB:  over 97% of them.  And the same mutations recurred across independent flasks far more than chance allows;  pairs of OmpF-using phage shared 61% of their mutations.  From those, Meyer concluded that selection acted on J before the new capacity existed, and that it "presumably improved" the grip on LamB.¹  The word "presumably" is doing real work.  In 2012 the fitness of the intermediates had not been measured.

The same paper is explicit that the new function cannot explain the first three steps:  "the all-or-none epistasis among the mutations means that selection for that new capacity per se was not responsible for the rise of the three prior mutations."¹

2016, measured.  Burmeister pulled one phage isolate out of each of six successful flasks, sampled exactly four days before OmpF use appeared, and checked that none of them could use OmpF.²  Each carried one or two of the four required changes plus other J mutations.  All six adsorbed faster to LamB than the ancestor did.  In direct competition, four of the six beat the ancestor on the original host, and all six beat it on the resistant malT host.²  Burmeister also states the negative claim plainly:  the four mutations are known not to have arisen simultaneously, because alleles carrying subsets of them were sampled at earlier time points.²

2018, mechanism.  Petrie's group asked how a J mutation could improve the old grip and open a new one at the same time.  Their answer:  the mutations make the phage particle less stable, and an unstable J can fold two ways.  A stock of genetically identical phage contains stable particles that use LamB and unstable ones that use OmpF.  Stability fell as mutations accumulated, and later specialising mutations restored it.  The same paper measured the last of the four required mutations and found that it, too, improved binding to LamB.³  So all four steps paid on the old receptor, not three.

2022, order.  Gupta's group built 671 engineered combinations of ten J mutations and measured the fitness of 580 of them against the ancestral host and 131 against the malT host.  The landscape changes shape when the host changes.  Reconstructing one flask's history, they found the first mutation on the winning line was favoured only against the ancestral host and the second only against the malT host.⁹  Then the replay:  12 populations started on a resistant host gave no OmpF users at all;  12 started normally gave three.⁹

Put together:  four changes, jointly required for the new receptor, each individually favoured, but favoured in different host backgrounds and in a particular order.  The path existed because the environment moved, and moved in the right sequence.  That is a stronger and stranger result than either side's slogan.

Part 5: Behe's test, his reply, and the answers

The test, from The Edge of Evolution (2007).  Behe takes the frequency of chloroquine resistance in the malaria parasite, about 1 in 10²⁰, and names a mutational problem of that difficulty a "chloroquine-complexity cluster," a CCC.⁵  Two independent CCCs multiply, so a "double CCC" is 1 in 10⁴⁰;  since fewer than about 10⁴⁰ cells have ever lived, he puts a "tentative line marking the edge of evolution" there.⁵  Chapter 7 then estimates from antibody shape-space work that a new protein-protein binding site needs five or six amino-acid changes, of which the neutral third can arrive separately, leaving "three or four amino acid changes … For the Darwinian step in question, they must occur together."⁵  The conclusion is stated in italics:  one new binding site is about as hard as chloroquine resistance, and two are beyond reach.⁵

Note what Behe concedes in the same book.  Where the steps are each an improvement, he treats the change as ordinary Darwinian evolution:  hemoglobin C-Harlem, two mutations, "happened exactly the way Darwin envisioned … each of which in turn was beneficial."⁵  His difficulty is specifically with steps that pay nothing until the last one lands.

Behe's reply in 2012, at full strength.  Three days after Meyer's paper appeared, Behe answered it.  The gist:  the paper is solid and oversold.  Viruses had been shown before to adapt to hosts with structurally similar surface proteins, and in nearly every case using the same protein and the same mutated site.  The retained LamB binding "strongly suggests" the same site on J is being used, and LamB and OmpF have similar three-dimensional structures, "so that strengthening the binding to one fortuitously led to binding to the other."  On his own 2010 scheme that is a modification of function, not a gain.  And the bacteria paid twice in broken genes:  malT to escape the original phage, then mannose permease genes to escape the evolved one.  His verdict:  "It seems Darwinian evolution took a little step sideways and two big steps backwards."^{8 8}

Two points in that reply are worth stating sharply because they are correct.  Behe never claimed the four mutations arrived together;  from the start he read them as strengthening LamB binding first.⁸  And the bacterial side of the ledger is loss.  malT and manYZ are broken genes, and any honest accounting of the experiment records them.

Behe's reply in 2019.  Darwin Devolves does not mention the experiment.  A search of the book for "lambda", "OmpF", "phage" and "bacteriophage" returns nothing;  Lenski says the same.⁶  Behe's 2019 answer is a blog post responding to Lenski's, and it is more direct than the 2012 one.  He grants that "the mutational pathway simply increased the initial strength of binding to the original protein until it also bound the structurally similar one", argues that a gradual path in one case does not make all paths gradual, and concludes that "the new lambda function is neither irreducibly complex nor beyond the edge of evolution."⁴  On the classification he admits the 2012 wording was inexact, then states the rule with a test:  modification, "precisely because it is using the very same site on the J protein that it uses to bind to LamB";  a different area of the protein, or a different protein, would make it a gain.⁴

What the critics answered.  Lenski's post is the fullest reply.  On Darwin's Black Box:  J needed several interacting residues, and removing any one stops the function, so an irreducibly complex system evolved.  On The Edge of Evolution:  "Lambda required four constructive changes to gain the ability to use OmpF, yet dozens of populations in tiny flasks managed to do this in just a few weeks," because the intermediates were strongly beneficial.  On Darwin Devolves:  nothing is broken in the phage, which still uses LamB and is better at it than its ancestor.^{6 6}

Where each side overreaches.  Lenski's strike two states the book as saying evolution "has a hard time making multiple constructive changes."  That drops the condition Behe actually relies on.  Behe's difficulty is with changes whose intermediates are not selectable, and Lenski's own explanation is that these intermediates were selectable.  The argument therefore does not land where he aims it.  "Dozens of populations" is also loose:  24 of 96, plus 1 of 6.

Behe's reply has a matching weakness.  He never addresses two things his own book says.  Page 139 predicts that a virus which learned to enter cells through a different surface protein would out-compete its siblings, and asserts that this has not happened.⁵  It then happened, 25 times, in flasks.  And chapter 7's number depends on the premise that the three or four changes "must occur together";  here a new receptor interaction requiring three specified amino-acid changes was assembled one step at a time.  The most his 2019 post offers against this is that a gradual path in one case does not make all paths gradual, which is true and is not an answer to the specific claim.

The fair summary is that this case is a defeat for two sentences in The Edge of Evolution and not a defeat for its central argument, which is about paths without selectable intermediates.  Both sides have written as if it were more.

Part 6: Gain or modification, and what the case shows

The score.  Measuring Biological Information asks four questions in order.  The sequence did not come from another lineage, so not a transfer.  Nothing in the phage stopped working;  it still uses LamB, and better than before.  Does a binding activity now work that did not work before?  Yes.  So:  gain, observed, four new nucleotides, and marked a hard case.

Behe's score, and why it is not unreasonable.  His 2010 definition puts under modification any mutation "shifting its activity somewhat (such as allowing a protein to bind a structurally-related ligand at the same site as its normal substrate)."⁷  In the same paper he sets out the rule for host shifts:  they are modification "if the binding occurs at the same site on the protein", and on that reading the coded element is the binding site, which is altered rather than created.⁷  Meyer's paper supports the "structurally-related" half:  OmpF has the most similar known crystal structure to LamB of any E. coli protein, and the two are not the closest pair by sequence, which the authors read as showing that overall shape matters more than particular residues for J binding.¹

Where his score is unsupported.  The "same site" half has never been measured.  There is no structure of J bound to either receptor.  What is known is that over 97% of the mutations lie in the C-terminal region that contacts LamB, and that the region is a specificity hotspot in wild lambdoid phage as well.¹²  Same region is not the same thing as same site.  Petrie's finding sharpens the problem:  the evolved J takes two conformations, and the OmpF-competent particles are the unstable ones.³  A second fold could present a surface the ancestral protein never presented.  Nobody has looked.

So the honest statement is:  by the site's own taxonomy, gain.  By Behe's stated rule, modification if and only if a structural fact holds that has not been tested.  This article records the gain, records his rule, and records that the test is missing.  It does not pretend his rule is arbitrary, and it does not pretend it has been satisfied.

What the case shows about extrapolation.  Something on the order of 10⁸ to 10¹¹ virus generations, in 10 ml of sugar water, bought four substitutions in one protein and one new binding partner.  Every step paid on the old function, and the new target was the closest structural analogue of the old target available anywhere in the host cell.  The switch is real, it is a gain, and it is small:  four nucleotides, no new gene, no new protein, no new part.

The extrapolation question is what this licenses.  It licenses the claim that multi-step changes are cheap when the steps pay.  It does not license the claim that multi-step changes are cheap in general, because the case contains no measurement of how often steps pay.  On the contrary, it documents the opposite in the same system:  72 of 96 populations failed inside 20 days, host manYZ mutations closed the path entirely, and starting the same experiment on a resistant host produced nothing in 12 tries.⁹  The path was narrow and the environment had to move through it in order.

The strongest objection to the design reading, stated and answered.  A skeptic will say the design argument assumes intermediates are usually neutral or harmful, that this case shows they can be beneficial, and that the assumption is therefore unwarranted.  The first two clauses are correct.  The third is not quite:  the assumption is unproven, not disproven, and the same is true of its opposite.  Neither side has a census.  What exists are individual cases:  lambda, where the intermediates paid;  and chloroquine resistance, where the first step confers no chloroquine transport and two specific changes are needed before any appears.¹⁶  Behe's own 2019 writing names that case as the exception where no gene was available to be usefully degraded.  The argument turns on the frequency, and the frequency is unmeasured.  Saying so is not a dodge;  it is the state of the evidence, and it is the reason the Benchmarking Evolution series counts what was searched rather than arguing from principle.

Objections and responses

"The four mutations were coordinated, which is what Behe said cannot happen."  They were not coordinated.  They are jointly required for the new receptor and were acquired one at a time under selection for the old one.  Four sources establish this, two of them by direct fitness measurement.  Any presentation of this case as a multi-mutation jump is wrong, whichever side makes it.

"Nothing new was created;  the phage already bound a porin and was tweaked to bind a similar porin."  This is Behe's reply and it is the strongest one available.  It is half right.  The two receptors are the closest structural pair in the host, and the mutations sit in the region that already did the binding.  But the phage now does something it could not do:  infect a cell with no LamB at all.  Calling that a tweak requires the claim that the binding happens at the same site on J, and that claim rests on inference from retained LamB binding, not on a structure.

"The bacteria only got worse, so this is net loss."  Partly true and worth keeping.  The host broke malT and, in many flasks, manY or manZ.  Those are losses, and they were adaptive.  But the malT mutation was beneficial in glucose even without phage, so scoring it purely as a cost to the bacterium is wrong.¹¹  And "the community's ledger is net negative" is a different claim from "the phage gained nothing."  The phage gained something.  The series scores molecules, then reports the effect on the organism beside it, precisely so that these two questions stay separate.

"Only 24 of 96 succeeded, so the innovation is rare."  It is contingent rather than rare.  The supplement notes that no population found it early, which is what a four-step requirement predicts, and that the rate of new successes fell late in the run as bacteria evolved the blocking mutations.¹¹  A quarter of replicate flasks finding the same solution in under three weeks is high repeatability, not rarity.  High repeatability also tells against a large target:  the same four changes, in the same order of magnitude of time, in every winner.

"This is a laboratory artefact;  the conditions were rigged."  They were designed, which is different.  Meyer chose glucose because it suppresses LamB, and predicted that reduced LamB would favour a phage that could use something else.  Two facts argue against dismissing the result.  The receptor region of J is the most variable and fastest-evolving part of the protein among wild lambdoid phage, and in-frame insertions and deletions cluster at the same two critical residues, which suggests host-range shifting is what this region does in nature.¹²  Against that, the specific substitutions found in the laboratory are not common in the natural homologues, so the experiment predicted where change would happen and not what change.¹²  That half is comparative inference, not observation.

"A virus is not a good model for anything larger."  True and it works against the extrapolation in both directions.  Lambda has 48,502 bases, a mutation rate per base about 140 times that of E. coli, and a population in one flask larger than the number of mammals in a large country.¹³  A change this cheap for lambda is not this cheap for a mammal, which is the series thesis.  It is equally true that a virus tail fibre binding an outer-membrane porin is a softer problem than a protein complex inside a cell, which is what Behe's chapter 7 estimate was about.  The case belongs in the ledger as a measured data point, not as a refutation of anyone's number.

"Lenski called it irreducibly complex, so Darwin's Black Box is refuted."  Lenski's point is fair on the definition's words:  remove one of the four and the OmpF function stops.  But the demonstration shows how the term fails as an argument, not that lambda is the machine Behe had in mind.  Four residues in one protein is not a flagellum, and the interesting question was never whether all-or-none function exists;  it is whether such functions can be reached, and that depends on the intermediates.

Issues and unanswered questions

1. Where does OmpF bind J? No structure exists for J bound to either receptor. This single measurement would settle whether the change is a gain or a modification under Behe's own published rule. It is the most valuable missing experiment in the case.
2. The published Science pagination of Meyer 2012 was not obtained. The article's locators are pages of the PubMed Central author manuscript, which carries identical text. Every footnote says so.
3. Phage densities in the 96-flask experiment were never tabulated. The ledger uses the main text's 10⁶ to 10⁷ per ml, measured in the six-flask experiment, and Gupta's round 6.3 × 10⁹ per population. A per-flask census would narrow the ledger by three orders of magnitude.
4. Meyer 2012 gives the range of emergence as 9 to 17 days; Burmeister 2016 gives 9 to 18. The article follows the primary.
5. No young-earth source appears to have written about this experiment. Searches of creation.com, Answers in Genesis and ICR found nothing on lambda and OmpF. The design-side primary literature here is Behe's two posts.
6. The unique-functional-sequence figure counts only the four substitutions. The earlier LamB improvements are real gains that cannot be converted to a nucleotide count without double counting, so the figure is a lower bound.

Verdict

The OmpF switch is a genuine gain of function, observed from start to finish, worth four nucleotides.  It required four changes in one gene that do nothing for the new receptor until all four are present, and it still evolved in twelve days in a quarter of the flasks, because every one of those changes improved the grip on the receptor the virus already used.  Two of the papers that establish this measured the intermediates directly rather than inferring them.  The catalogue entry that called the four mutations "coordinated" was wrong, and the corrected reading is the interesting one:  jointly required, individually selected, and dependent on the host changing in the right order.

For Behe's argument the case is a narrow loss and not a general one.  It falsifies his prediction that viruses do not learn to enter through new surface proteins, and it falsifies the premise that the three or four changes needed for a new binding interaction must arrive together, in the one binding problem anybody has watched from beginning to end.  It does not touch the CCC or the double CCC, because those concern paths without selectable intermediates, and Behe says himself that this path is not one of those.  His classification of the change as a modification follows from a rule he published two years before the experiment, and hangs on a structural fact nobody has measured.

For the series thesis the case is a fair and slightly uncomfortable data point.  On the order of 10⁸ to 10¹¹ virus generations, in sugar water, bought one new binding partner in one protein, with a target that was the closest structural match in the whole host cell and with every step paid for along the way.  That is cheap.  It is cheap for reasons the experiment documents and does not generalise:  a huge population, a fast mutation rate, a near target, and a moving environment that happened to open the steps in order.  Whether adaptive paths usually have paying intermediates is the question the whole argument turns on, and nobody on either side has measured it.

Closely related work on enzymes rebuilt or redirected under laboratory selection, including Barry Hall's ebg experiments, belongs to Directed Evolution and Protein Engineering.  The later fate of the generalist phage, which split into LamB and OmpF specialists and in some flasks became reproductively isolated, belongs to Observed Speciation.¹⁷

Sources