EXPERIMENTAL EVOLUTION, LOSS-OF-FUNCTION MUTATIONS, … · experimental evolution, adaptation, mutation, loss of function, malaria, Yersinia pestis abstract Adaptive evolution can

EXPERIMENTAL EVOLUTION, LOSS-OF-FUNCTION MUTATIONS,AND “THE FIRST RULE OF ADAPTIVE EVOLUTION”

Michael J. BeheDepartment of Biological Sciences, Lehigh University, Bethlehem, Pennsylvania 18015 USA

e-mail: [email protected]

keywordsexperimental evolution, adaptation, mutation, loss of function, malaria,

Yersinia pestis

abstractAdaptive evolution can cause a species to gain, lose, or modify a function; therefore, it is of

basic interest to determine whether any of these modes dominates the evolutionary process underparticular circumstances. Because mutation occurs at the molecular level, it is necessary toexamine the molecular changes produced by the underlying mutation in order to assess whethera given adaptation is best considered as a gain, loss, or modification of function. Although thatwas once impossible, the advance of molecular biology in the past half century has made itfeasible. In this paper, I review molecular changes underlying some adaptations, with a partic-ular emphasis on evolutionary experiments with microbes conducted over the past four decades.I show that by far the most common adaptive changes seen in those examples are due to the lossor modification of a pre-existing molecular function, and I discuss the possible reasons for theprominence of such mutations.

Adaptation by Gain, Loss, orModification of Function

IN THE ORIGIN OF SPECIES, Darwin(1859) emphasized the relentlessness of

natural selection:

[N]atural selection is daily and hourly scru-tinising, throughout the world, every varia-tion, even the slightest; rejecting thatwhich is bad, preserving and adding up allthat is good; silently and insensibly work-ing, whenever and wherever opportunityoffers, at the improvement of each organicbeing in relation to its organic and inor-ganic conditions of life. (p. 84)

Yet he realized that the changes that wereselected to adapt an organism to its envi-ronment did not have to be ones that con-ferred upon it a new ability, such as sight orflight. For instance, while observing somebarnacles, Darwin discovered unexpectedcases of the gross simplification of an or-ganism (Stott 2003):

The male is as transparent as glass . . . Inthe lower part we have an eye, & great testis& vesicula seminalis: in the capitulum wehave nothing but a tremendously long pe-nis coiled up & which can be exserted.There is no mouth no stomach no cirri, no

The Quarterly Review of Biology, December 2010, Vol. 85, No. 4

Copyright © 2010 by The University of Chicago Press. All rights reserved.

0033-5770/2010/8504-0002$15.00

Volume 85, No. 4 December 2010THE QUARTERLY REVIEW OF BIOLOGY

419

proper thorax! The whole animal is re-duced to an envelope . . . containing thetestes, vesicula, and penis. (p. 213)

Adaptive evolution can just as easily lead tothe loss of a functional feature in a specieslineage as to the gain of one. Over thecourse of evolutionary history, snakes havelost legs, cavefish have lost vision, and theparasitic bacterium Mycoplasma genitaliumhas lost its ability to live independently inthe wild, all in an effort to become betteradapted to their environments. Whatevervariation that sufficiently aids a particularspecies at a particular moment in a partic-ular environment—be it the gain or loss ofa feature, or the simple modification ofone—may be selected. Since species canevolve to gain, lose, or modify functionalfeatures, it is of basic interest to determinewhether any of these tends to dominateadaptations whose underlying molecularbases are ascertainable. Here, I survey theresults of evolutionary laboratory experi-ments on microbes that have been con-ducted over the past four decades. Suchexperiments exercise the greatest controlover environmental variables, and theyyield our most extensively characterized re-sults at the molecular level.

distinctions among adaptivemutations

Adaptation is often viewed from two dis-tinct aspects: the phenotypic aspect andthe molecular aspect. In order to avoidconfusion concerning how adaptive evolu-tion proceeds, these two aspects must bekept separate. A phenotypic adaptationmight have any number of underlying mo-lecular bases, resulting from either thegain, loss, or modification of a molecularfeature. As an example, consider a bacte-rium that evolves virulence in a mamma-lian host. If the adaptive phenotype isdeemed to be the gain of pathogenicity,then that could arise in several ways. Forinstance: 1) a gene for a bacterial proteinthat is the target of the mammalian im-mune response might be deleted, whichwould be the loss of a functional molecularfeature of the bacterium; 2) a gene for a

protein that is the target of the mammalianimmune response might sustain a missensemutation, causing an amino acid substitu-tion at a recognized epitope of the protein,thereby decreasing or eliminating the im-mune response while preserving proteinfunction, which could be considered a sim-ple modification of a pre-existing mo-lecular element; 3) the bacterium mightacquire a gene or gene cluster by horizon-tal transfer that leads to expression of anew surface feature that hides the epitopefrom the mammalian immune system,which would be a gain of a functional mo-lecular feature. The focus of this review ison the gain, loss, or modification of func-tional molecular features underlying adap-tation, no matter whether the phenotypicmanifestation strikes us as a loss or gain ofa property.

There are, of course, myriad functionalmolecular features of a cell, including thoseof all the major biomolecular categories: nu-cleic acids, proteins, polysaccharides, lipids,small metabolites, and so forth. However, thefunctional molecular features of only two cat-egories—nucleic acids and proteins—are di-rectly coded by the genome, and thus onlythese features can be directly affected by mu-tation; all other biological features are onlyaffected indirectly by mutations, throughtheir effects on nucleic acids and proteins. Inthis review, I focus on adaptive evolution bygain, loss, or modification of what I termFunctional Coded elemenTs (FCTs). An FCT is adiscrete but not necessarily contiguous re-gion of a gene that, by means of its nucleo-tide sequence, influences the production,processing, or biological activity of a particu-lar nucleic acid or protein, or its specificbinding to another molecule. Examples ofFCTs are: promoters; enhancers; insulators;Shine-Dalgarno sequences; tRNA genes;miRNA genes; protein coding sequences;organellar targeting- or localization-signals; intron/extron splice sites; codonsspecifying the binding site of a protein foranother molecule (such as its substrate,another protein, or a small allosteric regu-lator); codons specifying a processing siteof a protein (such as a cleavage, myristoyl-ation, or phosphorylation site); polyade-

420 Volume 85THE QUARTERLY REVIEW OF BIOLOGY

nylation signals; and transcription andtranslation termination signals. Examplesof general nucleic acid features that arenot FCTs include base composition, strandasymmetry, distance between coded ele-ments, and the like. They may enter intoconsideration indirectly only if they causean FCT to be lost or gained—for example,if altering the base composition also elim-inates the activity of an enhancer.

With these considerations in mind, Iclassify adaptive mutations as belonging toone of three modes:

1) A “loss-of-FCT” adaptive mutation is amutation that leads to the effective loss ofthe function of a specific, pre-existing,coded element, while adapting an organ-ism to its environment. The loss of theability of a frame-shifted gene to produce afunctional product, of an altered promoterto bind a transcription factor, or of a mu-tated protein to bind its former ligand, areexamples of loss-of-FCT mutations.

2) A “gain-of-FCT” adaptive mutation isa mutation that produces a specific, new,functional coded element while adaptingan organism to its environment. The con-struction by mutation of a new promoter,intron/exon splice site, or protein process-ing site are gain-of-FCT mutations. Also in-cluded in this category is the divergence bymutation of the activity of a previously du-plicated coded element.

3) A “modification-of-function” adap-tive mutation is a mutation whose definingproperty is negative—while adapting an or-ganism to an environment, it does not leadto the loss or gain of a specific FCT. Thisdefinition is intended to be broad enoughto act as a “catch-all” for anything that fallsoutside the first two modes. It includespoint mutations as well as other mutationsthat have a quantitative effect on a pre-existing FCT, increasing or decreasing itsstrength, for instance, or shifting its activitysomewhat (such as allowing a protein tobind a structurally-related ligand at thesame site as its normal substrate), butwithout effectively eliminating it. Thecategory “modification-of-function” alsoincludes the simple duplication of fea-tures, without divergence of activity, such as

a gene or regulatory region, since this pro-cess does not by itself produce a new func-tional element, although it may, like otherevents classified here as “modification-of-function,” be a step to future gain-of-FCTevents. It also includes mutations that mayact by more amorphous means, such as rear-ranging gene order, or changing the dis-tance or orientation between two interactingelements (e.g., the deletion of a stretch ofDNA that brings two interacting transcrip-tion factors closer, or that affects their orien-tation with respect to each other). Thiscategory is dubbed “modification-of-function”rather than “modification-of-FCT ” in orderto emphasize that the mutation may not nec-essarily exert its influence through alterationof a discrete coded element, as in the aboveamorphous examples.

In order to fall under any of the abovethree categories, a mutation must be adap-tive. Therefore, if, for instance, a mutationcauses a transcription factor binding site tobe formed or lost, but the mutation is ei-ther neutral or deleterious, it is excludedfrom consideration. Although assignmentof an adaptive mutation to one of themodes above may occasionally be ambigu-ous, the assignment is straightforward inthe great majority of cases in which molec-ular changes are elucidated, as we shall seebelow.

illustrations of modes of adaptationTo better our understanding of the

kinds of mutations that fall into these threemodes, it is useful to look at illustrationsfrom outside the realm of the experimen-tal evolution of microbes. Consider, for in-stance, human mutations that have beenselected in the past 10,000 years under thepressure of the malarial parasite Plasmo-dium falciparum. The reason for using suchexamples as these is that, because of itsthreat to human health, the interaction ofmalaria with humans has been very well-studied at the molecular level.

If a person of typical northern Europeandescent visited a malarious region of theworld and contracted the disease, shewould likely view the resistance to malariaamong many of the native peoples as an

December 2010 421EXPERIMENTAL EVOLUTION

unquestionably positive trait: the ability towithstand a debilitating disease. However,because of the detailed knowledge of themolecular bases of malaria resistance thathas been accumulated over past decades,that is not the way I would categorize ithere. Many persons indigenous to malari-ous regions are resistant, but they have ac-quired their resistance through differentmolecular mechanisms (Table 1). Al-though the mutations are all adaptive, dis-tinctions may be made among them. Themost well-known population of resistantpersons is heterozygous for sickle hemo-globin. In one of the two genes that theyinherited from their parents for the �chain of hemoglobin, the sixth codon des-ignates valine rather than the glutamicacid that the other � chain gene codes.When the red blood cell is invaded by themalarial parasite, the mutant, hydrophobicvaline residue allows hemoglobin mole-cules to aggregate with each other intolong microtubular structures. This poly-merization is associated with a measure ofresistance to malaria (Friedman 1978), al-though the exact mechanism for resistancein vivo is still a matter of debate (Verra etal. 2007). This is an example of an adaptivegain-of-FCT mutation because a codonhelping to specify a new, albeit weak—with

Kd � 1 mM (Behe and Englander 1979)—protein binding site has appeared.

Another population indigenous to somemalarial regions has a different point mu-tation in their hemoglobin. In this in-stance, the sixth codon of the � chain hasmutated from glutamic acid to lysine. Al-though the altered hemoglobin (HbC)does not aggregate as sickle hemoglobindoes, it confers resistance to malaria forreasons that are unclear. Because appar-ently no new, discrete, coded molecularfeature has been developed, this is catego-rized as a “modification-of-function” adap-tive mutation; no FCT has been lost orgained.

A third population of natives in somemalarial regions is comparatively healthydue to a different kind of mutation. Theyhave a condition called thalassemia, which,like sickle hemoglobin and HbC, also con-fers a measure of resistance to malaria. In athalassemic person, however, one of the �or � chain hemoglobin genes that is inher-ited from a parent is nonfunctional, forany one of a variety of reasons. In somethalassemic individuals, a whole gene hasbeen deleted, whereas, in others, a frame-shift mutation has caused the gene toproduce a nonfunctional protein. In stillothers, the genetic control region near

TABLE 1Human mutations selected for resistance to malaria

Clinical phenotype Underlying mutationMode of

adaptation Reference

Hb S �6 glu val G Cavalli-Sforza et al. (1994)Hb C �6 glu lys M Modiano et al. (2001)Hb E �26 glu lys M Weatherall et al. (2002)Hereditary persistence

of fetal hemoglobinDeletion/point mutations in control regions of

hemoglobin gamma chain geneL Forget (1998)

Thalassemia Deletion/point mutations in either � or �

hemoglobin genesL Flint et al. (1986)

G6PD deficiency Point mutations, deletions of G6PD gene L Ruwende et al. (1995)Loss of Duffy antigen in

red blood cellsPoint mutation switching off production of Duffy

antigen specifically in red blood cellsL Pogo and Chaudhuri (2000)

Band 3 proteindeficiency

Deletion L Kennedy (2002)

Abbreviations:G - adaptive gain of functional coded elementL - adaptive loss of functional coded elementM - adaptive modification of function


one of the globin genes has suffered achange that makes it nonfunctional, and soit produces no protein. Any of these wouldbe an example of a loss-of-FCT mutation.

Other populations are also malaria-resistant due to loss-of-FCT mutations,but in genes other than the globin genes.Take, for instance, the gene that codesfor glucose-6-phosphate dehydrogenase(G6PD). Any one of hundreds of separateknown mutations in the coding and con-trol regions of the gene has eliminated thefunctional enzyme, or has significantly de-creased its catalytic activity. Other loss-of-FCT mutations include the deletion of agene for Band 3 protein, and the loss of apromoter site for expression of Duffy anti-gen in red blood cells.

In summary, human genetic lines in ma-larious regions of the world have adaptedto an environment that includes the malar-ial parasite, thereby giving rise to a numberof molecular changes (Table 1). They havedone so by different modes, which can beclassified as adaptive mutations involvingloss- or gain-of-FCT, or modification-of-function. As we shall see below, microbesin laboratory evolution experiments alsoadapt in these ways.

Evolution Experiments with BacteriaRather than performing an exhaus-

tive review of laboratory evolution exper-iments, I will use several recent, influentialreviews as jumping-off points for my surveyof the relevant literature, starting with“Evolution experiments with microorgan-isms: the dynamics and genetic bases ofadaptation” (Elena and Lenski 2003). Intheir review, Elena and Lenski (2003) wereparticularly interested in laboratory studiesthat were “open-ended and long-term,”such as they themselves had conducted.They eschewed reviewing experiments that“targeted specific, often new, metabolicfunctions, even though this work is of greatinterest” (Elena and Lenski 2003:458). In-stead, for that topic, they referred readersto two previous reviews. Interestingly, eventhough their own review was quite up-to-date, the two reviews on specific metabolicfunctions that they cited—“Evolution of

new metabolic functions in laboratory or-ganisms” (Hall 1983) and Microorganisms asModel Systems for Studying Evolution (Mort-lock 1984a)—were both about twenty yearsold at the time, thus indicating a lack ofrecent work in that subfield.

metabolism of unusual compoundsI will begin by examining the early

work reported in Microorganisms as ModelSystems for Studying Evolution, edited byMortlock (1984a) (Table 2). Four of thefirst five chapters concern several speciesof bacteria in which cultures are evolvedto metabolize an unusual chemical foodsource that the unevolved species cannotgrow upon. In general, in each of theinvestigations, the first mutation selectedis either one that inactivates a repressorgene, thus turning the gene it regulatesinto a constitutively-expressed gene, or,less often, one that increases the copynumber of a particular metabolic gene.(The identities of some mutations in thisearly work were inferred by the authors,not demonstrated.) The amplified or de-regulated gene is one that typically codesfor an enzyme that metabolizes a relatedcompound and that has a limited abilityto metabolize the novel substrate. Theincrease in the concentration of enzymebrought about by deregulation or geneamplification allows it to metabolizeenough of the novel material to permitthe mutated bacterium to grow, albeitsometimes slowly. Similar results were re-ported by Clarke (1984) in Chapter 7,which focuses on mutations in the meta-bolic pathway of Pseudomonas that metab-olizes amides. During the course of manyexperiments, she obtained a variety ofpoint mutations in the regulatory proteinfor the amidase gene and in the pro-moter region for the amidase gene, aswell as in the gene itself.

Constitutive mutations are loss-of-FCT mu-tations because the organism has lost thefunction of a coded feature that controls syn-thesis of the enzyme. Once the bacteriumhas mutated, allowing it to utilize a novelsubstrate sufficiently enough to grow, fur-ther selection can often improve the growth


TABLE 2Work reported in Microorganisms as Model Systems for Studying Evolution (Mortlock 1984)

Chapter Laboratory phenotype Underlying mutationMode of


1 Xyltitol catabolism by Klebsiella Inactivation of a ribitoldehydrogenase repressor gene

L Mortlock(1984c)

l-arabitol catabolism by Klebsiella Inactivation of a ribitoldehydrogenase repressor gene

L

Increased catabolism of xylitol Apparent point mutations of ribitoldehydrogenase

M

Increased catabolism of xylitol Apparent alteration of ribitoldehydrogenase gene promoter

M

Increased catabolism of xylitol Apparent increase in copy numberof ribitol dehydrogenase gene

M

Xylitol transport acrossmembrane

Deregulation of d-arabitol operon L

2 Increased xylitol catabolism byKlebsiella

Apparent increase in copy numberof ribitol dehydrogenase gene

M Hartley(1984)

Increased xylitol catabolism byKlebsiella

Point mutation in ribitoldehydrogenase protein

M

Increased xylitol catabolism byE. coli

Transfer of Klebsiella rdh operon toE. coli plus gene amplification

M

Increased xylitol catabolism byE. coli

Transfer of Klebsiella rdh operon toE. coli plus point mutation

M

4 d-lyxose isomerase activity byKlebsiella

Uncertain; perhaps due toconstitutive d-mannoseisomerase

L Mortlock(1984b)

d-lyxose isomerase activity in E.coli

Uncertain; perhaps due toconstitutive d-mannoseisomerase

L

d-arabinose isomerase activity byKlebsiella

Constitutive l-fucose isomerase L

d-arabinose isomerase activity inE. coli

Point mutation-induced l-fucosepathway

M

5 Growth on l-1,2-propanediol ofE. coli

Possible alteration ofoxidoreductase promoter

M Lin and Wu(1984)

Growth on d-arabinose of E. coli Either constitutive expression offucose genes, or dual control ofregulator protein by l-fucoseand d-arabinose

LG

Growth on d-arabitol of E. coli Constitutive production ofoxidoreductase in already-mutated strain 3

L

Growth on xylitol of E. coli Constitutive production of d-xylosepathway

L

Growth on ethylene glycol of E.coli

Constitutive use of l-1,2-propanediol pathway

L

6 Growth of E. coli on lactose(after deletion of lacZ)

Point mutation in cryptic �-galactosidase ebgo; also renderedconstitutive by frameshifts andIS-disruption of ebg repressor

ML

Hall (1984)

Growth on lactulose Second point mutation in ebgo MIncreased growth on lactulose Point mutation in ebg repressor

allowing lactulose as inducerM

Growth on lactobionate Third point mutation in ebgo M

continued


rate through point mutations in the gene forthe overproduced enzyme. These additionalmutations are modification-of-function mu-tations, because no FCT is gained or lost.Increased-copy-number mutants are alsomodification-of-function mutants, becausethere is no evidence that any of the extracopies have diversified, which would thenplace it in the class of gain-of-FCT muta-tions.

An interesting variation on this patternis reported in Chapter 5 by Lin and Wu(1984). Mutants of E. coli and S. typhi-murium (S. enterica var Typhimurium) gainthe ability to metabolize the unusual sub-strate d-arabinose by altering the specificityof a regulatory protein. Normally, the en-zyme fucose isomerase is induced in thesebacteria when some fucose enters the celland binds to a positive regulatory protein,which then turns on the gene for the isom-erase. The regulatory protein of some mu-tants, however, responds to both fucoseand d-arabinose. It turns out that the un-usual substrate d-arabinose can be metab-olized by enzymes of the fucose pathway,and, because the protein has apparentlygained an additional binding site for thenovel substrate, the mutation is classed asgain-of-FCT.

recruitment of cryptic proteinsThe evolution of the ability of the bacte-

ria described above to metabolize novelcompounds depends on mutations inknown proteins of known metabolic path-ways. Several chapters in Microorganisms asModel Systems for Studying Evolution (Mort-lock 1984a), however, describe exampleswhere previously unsuspected, cryptic pro-teins were recruited to take over the func-tion of a known protein whose gene hadbeen deliberately deleted.

In Chapter 9, Kemper (1984) describesthe ability of the product of a gene of Sal-monella typhimurium, which he called newD,to replace a protein coded by the leuD geneof the leucine biosynthesis pathway. NewDordinarily binds tightly to a larger proteinsubunit that Kemper dubbed SupQ, just asLeuD binds to a larger subunit, LeuC. Inorder to allow NewD to restore leucinebiosynthetic function when leuD is deliber-ately deleted, the gene for SupQ also has tobe deliberately knocked out. Kemper ar-gues that supQ/newD was not simply a geneduplication of leuC/leuD (the genes of theleucine pathway) (Stover et al. 1988). How-ever, at that time, sequence data was moredifficult to come by than it is today. It isnow known that the S. typhimurium genomecontains a homolog of LeuD with 38%

TABLE 2Continued

Chapter Laboratory phenotype Underlying mutationMode of


7 Growth in presence ofbutyramide of P. aeruginosa

Possible point mutation inregulator protein gene amiR

M Clarke(1984)

Growth in presence ofsuccinate/lactamide

Possible point mutation inregulator protein gene amiR pluscatabolite repression mutant

M

Increased production ofamidase

Mutation in promoter M

Growth on butyramide,valeramide, acetanilide

Point mutations in amidase M

8 Resistance of yeast to allylalcohol

Point mutations in alcoholdehydrogenase

M Wills (1984)

9 Rescue of leuD deletion Availability of cryptic leuD-like gene M Kemper(1984)

Abbreviations:L - adaptive loss of functional coded elementM - adaptive modification of function


amino acid sequence identity. It seemslikely that the protein that Kemper discov-ered that could substitute for LeuD wasthis homolog. The S. typhimurium genomealso contains a homolog of LeuC with 52%amino acid sequence identity; it seemslikely that SupQ was this homolog. No suchLeuC or LeuD homologs exist in E. coli.Because newD is a homolog of LeuD thatalready binds a homolog of LeuC, this is amodification-of-function event.

In Chapter 6, Hall (1984) discusses hiswork over the previous decade with an en-zyme he dubbed Ebg, which stands forevolved �-galactosidase. He had shown thatE. coli that had had its lacZ gene (whichcodes for a �-galactosidase) intentionallydeleted could be mutated so that the al-tered Ebg protein metabolized the sugarlactose in its stead. In subsequent years,however, Hall (1995) showed unexpect-edly that Ebg was an unrecognized homo-log of LacZ, with 38% sequence identity.The Ebg active site was very similar to thatof LacZ, and acquired its ability to metab-olize lactose only when the Ebg active sitemutated to one identical to that of an ac-tive LacZ. Because the unevolved enzymecould already metabolize lactose at a lowrate, the adaptive point mutations thatincreased its activity are categorized asmodification-of-function ones. Mutationsin the ebg repressor that allowed the ebg�-galactosidase protein to be constitutivelyexpressed are loss-of-FCT.

In further experiments, both Kemper(1984) and Hall (2003) deleted the homol-ogous recruited protein genes ebg andnewD, but were unable to recruit any of thethousands of other cellular proteins to re-place them. Recently, Patrick et al. (2007)employed functional genomics tools to sur-vey 107 single-gene knockout strains of E.coli that could not grow on M9-glucose me-dium, transfecting each one with a plasmidlibrary that contained every E. coli open-reading frame. They discovered that about20% of the knockout strains could be res-cued by overexpression of at least oneother E. coli gene. Patrick et al. (2007)found that the missing gene product andits suppressor were generally not homo-

logs, unlike the previous cases of LacZ/Ebg and LeuD/NewD.

The work of Patrick et al. (2007) is not initself an experimental evolution study, asall of the manipulations were performedby the investigators and no spontaneousmutations arose. Nonetheless, it does dem-onstrate much potential redundancy in theE. coli genome that may contribute to itsevolvability, as the authors suggest. It maybe instructive to consider how newer workwould be classified if the events thatPatrick et al. (2007) investigated were tooccur in the wild. If, to adapt an organismin nature, a gene were initially lost by mu-tation, that of course would be a loss-of-FCT event. If subsequent rescue in naturerequired the overexpression of a poten-tially compensating gene, as in Patrick etal.’s (2007) experimental cases, then thatcould occur in at least several ways: 1) arepressor element of the compensatinggene could be deleted, rendering the geneconstitutive, which would be a loss-of-FCTmutation; 2) an existing promoter of thecompensating gene could have its se-quence altered, binding a transcriptionfactor more strongly and thus leading tooverexpression, which would be classed asa modification-of-function mutation, as anexisting FCT was altered, not gained orlost; 3) a new promoter could be con-structed by mutation to increase the tran-scription rate of the gene, which would bea gain-of-FCT mutation; 4) the compensat-ing gene could increase in copy number,which by itself would be classed as a mod-ification-of-function mutation; or 5) a copyof the compensating gene could diverge insequence to more efficiently assume theactivity of the deleted gene, which wouldbe classified as a gain-of-FCT mutation. Fi-nally, the effect of the deleted gene may becompensated in the wild not by increasingthe expression of its potential comple-ment, but instead by deleting or reducingthe expression of one or more other genes,which would be an adaptive loss-of-FCTmutation (this possibility was not investi-gated in the excellent work of Patrick et al.[2007]). As these examples make clear,any of multiple molecular evolutionary


pathways could underlie the rescuedphenotype. To understand whether anadaptation represents a gain, loss, ormodification of a function, the molecularevents underlying the adaptation mustfirst be understood.

the work of lenski and colleaguesIn the 1990s, investigators began to per-

form the “open-ended and long-term” evo-lution experiments that are the main focusof a review by Elena and Lenski (2003).Certainly, the most extensive and longestrunning investigation has been under-taken by Lenski himself, who has been con-tinuously growing cultures of E. coli in hislaboratory since the late 1980s and moni-toring the facets of their evolution (Lenski2004). Diluting a portion of the previousday’s culture a hundred-fold, each day canpotentially see 6 to 7 new generations ofbacteria (26.6 � 100). At intervals, Lenskifreezes a portion of a bacterial generationand stores it, so that, later, the descendantgenerations can be straightforwardly ana-lyzed and compared head-to-head withtheir ancestors. Over the years, the num-ber of generations has approached 50,000.With a cumulative population size of about1014 cells, Lenski’s investigation is largeenough and long enough to give solid, re-liable answers to many questions aboutevolution.

The results of a number of Lenski’s pa-pers over the term of the experiment arevery briefly summarized in Table 3, includ-ing his early work (not discussed here) inwhich the underlying molecular bases ofadaptive phenotypes had not yet beenidentified. Schneider et al. (2000) exam-ined multiple insertion sequence (IS) mu-tations that became fixed in the bacterialpopulations within ten thousand genera-tions. The mutations were of the kindcaused by IS elements, including insertionsand recombinations, that led to geneticinversions and deletions. By examining theDNA sequence of the E. coli in the neigh-borhood surrounding the IS elements, theinvestigators saw that several genes involved incentral metabolism were knocked out, as wellas some cell wall synthesis genes and sev-

eral others. In subsequent work, Cooper etal. (2001) discovered that twelve of twelvecell lines showed adaptive IS-mediated de-letions of their rbs operon, which is in-volved in making the sugar ribose. Thus,the adaptive mutations that were initiallytracked down all involved loss-of-FCT.

Several years later, when the cultureshad surpassed their 20,000th generation,Lenski’s group at Michigan State broughtmore advanced techniques to bear on theproblem of identifying the molecularchanges underlying the adaptation of theE. coli cultures. Using DNA expression pro-files, they were able to reliably track downchanges in the expression of 1300 genes ofthe bacterium, and determined that 59genes had changed their expression levelsfrom the ancestor, 47 of which were ex-pressed at lower levels (Cooper et al 2003).The authors stated that “The expressionlevels of many of these 59 genes are knownto be regulated by specific effectors includ-ing guanosine tetraphosphate (ppGpp)and cAMP-cAMP receptor protein (CRP)”(Cooper et al 2003:1074). They also notedthat the cellular concentration of ppGpp iscontrolled by several genes including spoT.After sequencing, they discovered a non-synonymous point mutation in the spoTgene. When the researchers examined tenother populations that had evolved underthe same conditions for 20,000 genera-tions, they found that seven others also hadfixed nonsynonymous point mutations inspoT, but with different substitutions thanthe first one that had been identified, thussuggesting that the mutations were de-creasing the protein’s activity.

The group then decided to concentrateon candidate genes suggested by the phys-iological adaptations that the cells hadmade over 20,000 generations. One suchadaptation was a change in supercoilingdensity; therefore, genes affecting DNA to-pology were investigated (Crozat et al.2005). Two of these genes, topA and fis, hadsustained point mutations. In the case oftopA, the mutation coded an amino acidsubstitution, whereas, with fis, a transver-sion had occurred at the fourth nucleotidebefore the starting ATG codon. The topA


mutation decreased the activity of the en-zyme, while the fis mutation decreased theamount of fis gene product produced.Moving the mutations into the ancestorimproved its fitness in minimal glucose me-dia.

The genes that had earlier been ob-

served to be disrupted by IS elements,thereby leading to greater fitness in thegrowth medium, were then sequenced intwelve culture lines that had evolved for20,000 generations (Woods et al. 2006). Alllines were observed to have point muta-tions in or near one or more of the genes;

TABLE 3Selected results of Richard Lenski’s long-term E. coli evolution project

Laboratory phenotype Underlying mutationMode of


Mean fitness of evolved strains improves by37% on minimal glucose medium over2000 generations; most improvementcomes early.

unknown — Lenski et al. (1994)

Over 10,000 generations, fitness and cellvolume increase rapidly, then slow down;fitness increases in discrete steps; authorsliken this to “macroevolution.”

unknown — Lenski and Travisano(1994)

Over 10,000 generations, mutator strainsevolve in 3 of 12 populations; probablyhitchhike with beneficial mutations.

Probable defect in methyl-directedmismatch repair pathway

—* Sniegowski et al.(1997)

After 6000 generations, two phenotypes, Land S, coexist in balanced polymorphism.Fitness is frequency dependent.

unknown — Rozen and Lenski(2000)

Study of nine IS mutants that are fixed after10,000 generations. IS had inserted intoseveral genes (pykF, nadR, pbpA-rodA, hokB/sokB) or caused rearrangements.

IS mediated insertions, deletions,rearrangements

L,LL,L

Schneider et al.(2000)

12 of 12 lines show deletion of rbs operonmediated by IS150. Selective value ofdeletion is 1%-2%.

Repeated deletion of rbs operon L Cooper et al. (2001)

DNA expression arrays showed 59 genechanges in parallel in two strains grownfor 20,000 generations. Many genescontrolled by CRP and ppGpp.

Eight separate point mutations inspoT, which controls cellularconcentration of ppGpp

M Cooper et al. (2003)

Sequence random genes from 10,000 and20,000 generations

No mutations in nonmutatorstrains. Some point mutationsin mutator strains, but noadaptive effect

—* Lenski et al. (2003b)

Mutations affecting DNA topology: topAmutant decreases protein activity; fismutant decreases amount of protein made

topA H33Y; fis A-C transversionfour nucleotides before startATG

MM

Crozat et al. (2005)

Four genes (pykF, nadR, pbpA-rodA,hokB/sokB) first identified by IS insertionsare examined in other populations.

Same four genes repeatedly sufferselectable mutations at differentsites

M,MM,M

Woods et al. (2006)

Comparison of protein profiles of evolved E.coli vs. ancestor

malT locus suffers multipledeletions, substitutions

L Pelosi et al. (2006)

Citrate utilization under aerobic conditions unknown — Blount et al. (2008)

*Identified or inferred mutations are not categorized because they are neutral, not adaptive.Abbreviations:

G - adaptive gain of functional coded elementL - adaptive loss of functional coded elementM - adaptive modification of function


however, the genes were mutated at a vari-ety of points. In other words, there wasstrong parallelism at the level of whichgene it is adaptive to mutate, but little orno parallelism for a particular mutation ina particular gene. The fact that multiplepoint mutations in each gene could servean adaptive role—and that disruption by ISinsertion was beneficial—suggests that thepoint mutations were decreasing or elimi-nating the protein’s function.

In an investigation of global protein pro-files of the evolved E. coli, Lenski’s groupdiscovered that the MalT protein of themaltose operon had suffered mutations in8 out of 12 strains (Pelosi et al. 2006).Several mutations were small deletionswhile others were point mutations, thussuggesting that decreasing the activity ofthe MalT protein was adaptive in minimalglucose media. This was tested by movingthe mutation into the ancestral strain,which subsequently gained in fitness.

Recently, Lenski’s group reported theisolation of a mutant E. coli that hadevolved a Cit� phenotype. That is, thestrain could grow under aerobic condi-tions in a culture of citrate (Blount et al.2008). Wild E. coli cannot grow under suchconditions, as it lacks a citrate permease toimport the metabolite under oxic condi-tions. (It should be noted that, once insidethe cell, however, E. coli has the enzymaticcapacity to metabolize citrate.) The pheno-type, whose underlying molecular changeshave not yet been reported, conferred anenormous growth advantage because theculture media contained excess citrate butonly limited glucose, which the ancestralbacteria metabolized. Blount et al. (2008)marshaled evidence to show that multiplemutations were needed in the populationbefore the final mutation conferred theability to import citrate; the activating mu-tation did not appear until after the30,000th generation.

As Blount et al. (2008) discussed, severalother laboratories had, in the past, alsoidentified mutant E. coli strains with such aphenotype. In one such case, the underly-ing mutation was not identified (Hall1982); however, in another case, high-level

constitutive expression on a multicopyplasmid of a citrate transporter gene, citT,which normally transports citrate in theabsence of oxygen, was responsible for elic-iting the phenotype (Pos et al. 1998). If thephenotype of the Lenski Cit� strain iscaused by the loss of the activity of a nor-mal genetic regulatory element, such as arepressor binding site or other FCT, it will,of course, be a loss-of-FCT mutation, de-spite its highly adaptive effects in the pres-ence of citrate. If the phenotype is due toone or more mutations that result in, forexample, the addition of a novel geneticregulatory element, gene-duplication withsequence divergence, or the gain of a newbinding site, then it will be a noteworthygain-of-FCT mutation.

The results of future work aside, so far,during the course of the longest, mostopen-ended, and most extensive laboratoryinvestigation of bacterial evolution, a num-ber of adaptive mutations have been iden-tified that endow the bacterial strain withgreater fitness compared to that of the an-cestral strain in the particular growth me-dium. The goal of Lenski’s research was notto analyze adaptive mutations in terms ofgain or loss of function, as is the focus here,but rather to address other longstanding evo-lutionary questions. Nonetheless, all of themutations identified to date can readily beclassified as either modification-of-functionor loss-of-FCT.

an ambiguous caseA fascinating case of concurrent gain-

and loss-of FCT can be seen in the work ofZinser et al. (2003). The authors examineda mutant of E. coli that thrived when theculture was starved, but that was outcom-peted under conditions of growth. Theydemonstrated that the result depended ontwo mutational events: 1) an initial inser-tion of an IS element between the tran-scriptional promoter of the cstA gene and anearby CRP box that regulated it; and2) inversion of the sequence flanked by thenew IS element and another IS element�60 kb distant and upstream of the ybeJ-gltJKL-ybeK operon. The inversion broughtthe ybeJ gene of the mutant, which encodes a


portion of an amino acid transporter, underthe influence of the CRP box that previouslyregulated the cstA gene and that encodes astarvation-inducible oligopeptide permease.In the new arrangement, the ybeJ gene wasactively transcribed, while the cstA geneno longer was. Interestingly, the ybeJ geneproduct helped the bacterium grow underconditions of starvation as expected, butthe authors demonstrated that expressionof the cstA gene itself would also be bene-ficial under the same conditions. Nonethe-less, the inverted arrangement was selectedbecause it possessed the greatest net fitness.

This example points to unavoidable am-biguity in the classification of some adap-tive evolutionary events. One can view theadaptation investigated by Zinser et al.(2003) in several ways: 1) as a modification-of-function event (i.e., the insertion of aduplicate IS element) followed, as a resultof the inversion, by the loss of a functionalcoded element controlling the cstA gene(the CRP box) and the gain of a functionalcoded element controlling the ybeJ gene(the same CRP box); or 2) since no FCTwas gained or lost in the inversion butrather was simply rearranged, the inversionsums to a second modification-of-functionmutation, albeit one in which there is nowa potential for a recombination switch un-der conditions of starvation vs. growth.

Evolution Experiments with VirusesBecause of their ability to rapidly re-

produce to enormous population sizes,viruses, like bacteria, are well-suited toexperimental evolutionary studies, andsuch studies have been well-reviewed inrecent years (Elena and Lenski 2003; Elenaand Sanjuan 2007; Bull and Molineux2008; Elena et al. 2008). Several generalfeatures distinguish viruses from bacteriaand alter evolutionary expectations. 1) Viralgenomes are usually much smaller thanbacterial genomes, often by a thousand-fold, and one consequence of this is thatmutations are much easier to identify bysequencing for viruses than for bacteria.Another consequence, however, is that vi-ruses do not have deep reserves of homol-ogous genes or other genetic material to

recruit as bacteria do, except that whichcomes from their cellular hosts. 2) Becauseof their small genome size, most of theirgenome is required for their life cycle; rel-atively few viral genes are dispensable.3) For RNA viruses, the mutation rate isgreatly increased, as much as a million-foldover that of cells. This allows various mu-tation “solutions” to a selective challengeto present themselves much more rapidlythan for cells.

Table 4 summarizes evolution experi-ments with viruses discussed in the reviewsabove or in related papers. Most of theexperiments discussed below disturbed vi-rus growth in various ways and followedtheir evolutionary reaction to the distur-bance. The papers are categorized by thegeneral selective pressure that was broughtto bear on the viruses.

recovery from multiple bottleneck-induced deleterious mutations

Two laboratories (Burch and Chao 1999;Bull et al. 2003) subjected two differentviruses—an RNA virus (�6) and a DNAvirus (T7)—to repeated population bottle-necks (i.e., severe reductions in populationsize that cause the loss of genetic variation)of just one or a few viruses. Because of thehigh mutation rate of RNA viruses, eachvirus has a high probability of having amutation, most likely deleterious, and,through many bottlenecks, the researcherswere able to accumulate many mutationsin the virus. Because of the lower rate ofmutation in DNA viruses, the workers us-ing phage T7 added a mutagen to thegrowth medium in order to increase itsmutation rate.

Burch and Chao (1999) showed that �6suffered a ten-fold loss of fitness after 20bottleneck generations of repeatedly pick-ing individual plaques grown on a plate ofbacterial host, and then using the plaqueto seed a new plate. When subsequentlyallowed to grow at larger populationsizes, the phage was able to recover fit-ness by compensatory mutations; how-ever, because the molecular natures ofthe mutations were not determined, theycannot be classified here.


Bull et al. (2003) were able to accumu-late hundreds of mutations in bacterio-phage T7, and to reduce its fitness to about0.0001% of the wild type. During the recov-ery phase of the experiment, two separatepopulations of phage gained a factor ofabout 100-fold and 10,000-fold in fitness dur-ing 115 and 82 passages, respectively. Bothpopulations were continuing to recoverfitness when the experiment was halted.Among the mutations accumulated duringthe bottleneck phase of the experiment wereseveral small and large deletions in nones-sential genes, one small insertion, and about400 point mutations, almost equally dividedbetween missense mutations and silent mu-tations. There were also several nonsensemutations in nonessential genes.

In the recovery phase, there were also sev-eral large and small deletions, as well as a fewdozen point mutations, most of which weremissense mutations that were likely subjectto positive selection. Recovery-phase muta-tions involving large or frame-shifting dele-tions in genes can, with reasonable confi-dence, be classified as loss-of-FCT. Selectablepoint mutations are more difficult to classifydefinitively. None were reported to haveformed identifiable, functional coded ele-ments, such as new promoter sites or proteinprocessing sites. However, the possibility re-mains that some new functional coded ele-ment that is not readily identifiable from itsnucleic acid sequence may have arisen inone or more cases. To rigorously test sucha possibility, careful biochemical analysiswould be required. In the absence of ev-idence for gain-of-FCT, selectable pointmutations are tentatively classified as“modification-of-function” in Table 4.

adapting to a new environmentA second category of selective pressure is

that of viruses placed in new environments.Cuevas et al. (2002) described the compe-tition of two neutrally-marked strains ofthe single-stranded RNA vesicular stomati-tis virus (VSV) when grown on the samemedium and the same cells, but at varyingpopulation sizes. There was remarkableparallelism in the exact nucleotide changesfor the two strains over multiple experi-

ments, thus suggesting that the mutationscompensate for growth in the particularenvironment. All mutations were nucleo-tide substitutions; no deletions or inser-tions were seen. Twelve substitutions weresynonymous, fifteen were nonsynonymous,and three were in intergenic regions.

Bull et al. (1997) have studied the adapta-tion of replicate lineages of the single-strandedDNA virus �X174 to high temperature(43.5°C) on several hosts. In one study of 119total substitutions at 68 sites in the virus, overhalf were identical with substitutions in otherlineages, and no deletions or insertions werereported. Adaptation improved fitness in pri-mary lines by about 4000-fold with S. typhi-murium as host, and by about 100,000-fold withE. coli as host. In another study, Wichman et al.(1999) grew �X174 on S. typhimurium at ele-vated temperature (43.5°C). Half of the muta-tions in one line also appeared in a second linegrown under identical conditions. Mostchanges were amino acid substitutions, butone common mutation was an intergenic de-letion of 27 nucleotides. Since no discrete mo-lecular features were attributed to the deletedintergenic region, that deletion can be consid-ered a modification-of-function—no func-tional coded element was lost. No discrete newmolecular features were identified within mu-tated coding regions, but further biochemicalanalysis would be required to determine if afunctional coded feature—such as a new pro-tein binding site—were gained. In the absenceof such evidence, selectable point mutationsare tentatively classified as “modification-of-function” in Table 4.

adapting to a new hostAnother category of selective pressure is

that of viruses forced to adapt to a novelhost organism in the laboratory. The adap-tation of a virus to a new host in the wildcan, of course, be a major—even cata-strophic—epidemiological event, as the ex-amples of HIV and, potentially, H1N1 show.Nonetheless, as emphasized throughout thisreview, to understand how adaptation pro-ceeds, the phenotypic and molecular as-pects must be kept separate. The abilityof a virus to grow on an alternate hostis a phenotype that may have a variety of


TABLE 4Summary of selected viral evolution experiments

Investigator action Underlying mutationMode of


Viruses subject to driftBacteriophage �6 subject to drift in

bottleneck populationsRecovery by growth in larger populations

showed many small steps; no sequencedata

— Burch and Chao (1999)

Bacteriophage T7 repeatedly passedthrough single-virus populationbottlenecks

Several dozen point mutations; severallarge and small deletions in recoveryphase

L,M Bull et al. (2003)

Viruses adapting to new environmentsTwo neutrally-marked strains of

VSV competed at differentpopulation ratios

Extensive parallel convergence at thenucleotide and amino acid level; someco-variations noted

M Cuevas et al. (2002)

Bacteriophage �X174 adapted tohigh temperature growth

Multiple convergent substitutionmutations across several lineages

M Bull et al. (1997)

Two populations of bacteriophage�X174 adapted to hightemperature and novel host

Multiple convergent mutations, mostlysubstitutions, in two populationsaccumulated in different orders

M Wichman et al. (1999)

Viruses adapting to new hosts�X174 adapted to growth on

Eschericia and Salmonella hostsAmino acid substitutions at 5 sites in

major capsid attachment protein affecthost preference

M Crill et al. (2000)

�X174 adapted to growth on E. colistrains differing inlipopolysaccharide attachmentsite

Amino acid substitutions at 10 sites inmajor capsid attachment protein affecthost preference

M Pepin et al. (2008)

�6 adapted to growth on 15Pseudomonas syringae strains

Amino acid substitutions at 9 sites inspike attachment protein affect hostrange

M Duffy et al. (2006)

Broad-specificity �6 adapted togrowth on single Pseudomonassyringae strain

Single substitution in spike attachmentprotein gene narrows host range

M Duffy et al. (2007)

Viruses manipulated to be defectiveDeletion of 19 intercistronic

nucleotides from RNA virus MS2containing Shine-Dalgarnosequence and two hairpins

One revertant deleted 6 nucleotides;another duplicated an adjoining 14-nucleotide sequence; missingfunctional coded elementssubstantially restored

G,G Olsthoorn and vanDuin (1996)

4 nucleotide deletion in lysis geneof MS2

Reading frame restored by deletions,insertions

G,G Licis and van Duin(2006)

Randomized operator sequence ofMS2

Modification-of-function mutationsappeared, but operator hairpin notrestored

M Licis et al. (2000)

Deletion of T7 viral ligase gene Loss-of-FCT and modification-of-functionmutations in several genes; new ligaseactivity not obtained from host

L,M Rokyta et al. (2002)

Bacteriophage T7 forced toreplicate using T3 RNApolymerase

9 of 16 promoters altered sequence;several missense modification-of-function changes

M Bull et al. (2007)

T7 RNA polymerase gene movedtoward end of viral genome

Loss of one E. coli polymerasetermination site; facilitatedrearrangement of RNAP gene closerto the beginning of the viral genome

L,M Springman et al. (2005)

continued


possible underlying molecular bases. Asreviewed below and as observed in exper-imental evolutionary studies, viruses havepresumably adapted to new hosts by mod-ification-of-function mutations rather thanby gain- or loss-of-FCT ones.

Many experimental studies have beenperformed investigating the adaptation ofviruses to new hosts, and several recentpapers illustrate the topic of gain- or loss-of-FCT that this review is concerned with.Commonly, although not exclusively (seeYin and Lomax 1983; Subbarao et al. 1993;Agudelo-Romero et al. 2008), mutations incoat proteins underlie the phenotypic vari-ation. Crill et al. (2000) studied the adap-tation of �X174 to Eschericia and Salmonellahosts. Over 11 days of selection on a host,up to 28 substitutions, as well as a 2-baseinsertion and 27-base deletion, were ob-served in all genes except one. The authorsdetermined that nonsynonymous nucleo-tide substitutions at 5 sites in the majorcapsid protein gene F affected host prefer-ence. The altered amino acid residues lieon the virion surface and correlate stronglywith attachment efficiency but, intrigu-ingly, are at sites on the protein other thanthe putative carbohydrate binding siteidentified by X-ray crystallographic studies(McKenna et al. 1994). Pepin et al. (2008)revisited host adaptation of �X174, but in amore specific context. They examined theadaptation of the virus to three strains of E.coli that differed from each other only by asingle sugar group in the lipopolysaccha-ride site of attachment. Of 21 mutations, all

were nucleotide substitutions; seven of thosewere synonymous, and, of the remaining 14,10 occurred in the major capsid proteingene F. Since the substitutions apparentlydid not form a new FCT or cause the loss ofone, but simply modified the strength of at-tachment, they are modification-of-functionmutations.

Duffy et al. (2006) examined the evo-lution of the RNA bacteriophage �6 on15 Pseudomonas syringae strains. Nine mu-tations were isolated that affected hostrange, and all were amino acid substitu-tions in the host-attachment spike-proteinP3. One virus strain carrying a substitutionof P3 that broadened the �6 host range wasused for further studies (Duffy et al. 2007).When the mutant strain was grown exclu-sively on an alternate permissive host, itacquired a further single nucleotide substitu-tion in the gene for the spike protein. Thesubstitution conferred increased growth onthe alternate host and re-narrowed the hostrange.

The work reviewed above presents an op-portunity to further clarify the classificationcategories introduced in this paper. Host ad-aptations could potentially be classified inseveral ways. Some investigators might cate-gorize the adaptations reported by Duffy etal. (2007), for instance, as gains and losses ofthe function of binding very specific hostligands. As I discussed earlier in this paper,however, I classify shifts in ligand preferenceas “modification-of-function” events if thebinding occurs at the same site on the pro-tein, as in the examples discussed here. From

TABLE 4Continued

Investigator action Underlying mutationMode of


Viruses forced to be interdependentSeparate viruses, f1 and IKe,

engineered to carry distinctantibiotic resistance markers

In media containing both antibiotics,phages co-packaged into f1 proteincoats; two-thirds of IKe genomedeleted, second antibiotic genecaptured by f1

L,MG

Sachs and Bull (2005)

Abbreviations:G - adaptive gain of functional coded elementL - adaptive loss of functional coded elementM - adaptive modification of function


this view, the functional coded element is theprotein binding site, which can be modifiedeither by amino acid substitution directly atthe site or elsewhere in the protein. Consid-ered in this way, a functional site is notgained or lost as host preference is altered,but is instead modified.

recovery from introduced defects:the rna bacteriophage ms2

Another category of selective pressure isthe intentional introduction by the investiga-tor of defined defects in a virus. Olsthoornand van Duin (1996) deleted a 19-nucleotideintercistronic segment of RNA phage MS2between the stop codon for the maturationprotein gene and the start codon for the coatprotein gene; this segment contains theShine-Dalgarno sequence and usually formstwo hairpins. The synthesis of the coat pro-tein is especially sensitive to the presence ofa hairpin containing its initiator codon, andthe investigators reported that the titer of thealtered phage dropped ten orders of magni-tude (Olsthoorn and van Duin 1996). Twoseparate kinds of revertants were isolated—one that had deleted 6 nucleotides thatcoded the final two residues of the matura-tion protein, and another that contained aduplication of an adjoining 14-nucleotidesequence. Remarkably, the deletion rever-tant substantially restored the coat pro-tein initiator hairpin and the Shine-Dalgarno sequence and, subsequently,approximated them more closely by ac-cumulating several point mutations. Theduplication revertant partially restored bothnative hairpins and the Shine-Dalgarnosequence, and subsequently accumulatedpoint mutations to more closely approxi-mate the starting phage. The titers of thereconstructed phages were very similar to thestarting phage.

Both the 6-nucleotide deletion and the14-nucleotide duplication are gain-of-FCT mutations since they both producednew coded molecular features in the vi-rus that did not exist in the immediateprecursor; that is, the virus that sustainedthe deliberately-deleted 19 nucleotides.The 6-nucleotide deletion did not inacti-vate the maturation protein, so it is not a

loss-of-FCT mutation. The 14-nucleotideduplication led to the formation of newfunctional coded elements (it did notsimply repeat pre-existing elements), soit is not just a modification-of-functionmutation. The subsequent point muta-tions that improve activity are modifica-tion-of-function.

Another paper from van Duin’s labora-tory investigated the evolution of phageMS2 in which a 4-nucleotide sequence wasdeleted from a 38-nucleotide intercistronicregion between the coat protein gene andthe replicase gene (Licis and van Duin2006). The gene for the lysis protein spansthe intercistronic region and overlaps intothe coat protein and replicase genes, sothat the deletion introduces a frameshiftinto it. Furthermore, the deletion destabi-lizes the operator loop, which is needed tobind a dimer of coat protein, initiating cap-sule formation, as well as regulating ex-pression of the replicase gene. The mutantphage decreased in fitness by a factor of106.

Most initial revertants at relatively low ti-ters first repaired the frameshift by insertinga nucleotide in this region. One deleted twomore nucleotides, which also restored thecorrect reading frame. Several further rever-tants were obtained by increasing the initialpopulation size of mutant viruses and passag-ing the virus. In particular, insertion rever-tants were obtained, one of which (revIN4)had inserted four nucleotides exactly at thesite where they initially had been deleted.These are all gain-of-FCT mutations. Afterfurther passages, most revertants acquiredpoint mutations that repaired damaged fea-tures, including the operator loop. Only onerevertant achieved a fitness equal to the wildtype phage after multiple passages, and thiswas a derivative of revIN4 that went on torevert completely to the wild type.

A third paper from van Duin’s lab con-cerning MS2 examined the evolutionaryconsequences of partially randomizing thesequence of the operator hairpin (Licis etal. 2000). The fitness of the initial mutantsdecreased anywhere from 103 to 107. Mostrevertants failed to reconstruct the opera-tor. The two best revertants, with 2% and


20% of the fitness of the wild type, respec-tively, did not restore the operator hairpin.Thus, the mutations in this study aremodification-of-function ones.

recovery from introduced defects:the dna bacteriophage t7

Bull, Wichman, Molineux, and colleagues(Bull et al. 1997, 2003, 2007; Bull and Mo-lineux 2008) introduced large changesinto the double stranded DNA bacterio-phage T7 to observe how it might recover.The DNA polymerase of T7 is much moreaccurate than the RNA polymerase of MS2,the system used by van Duin. Nonetheless,they were able to show considerable evolu-tion of the DNA phage.

In one experiment, the gene for DNAligase from phage T7 was intentionally de-leted and the phage’s evolutionary recov-ery was observed (Rokyta et al. 2002). Theauthors noted that the deletion exerts littleselective pressure if the phage is grown ona host that contains an active cellular li-gase. They grew the phage in a cell linewhose ligase had been previously mutatedto decrease its activity. Initially, the mutantT7 had very low fitness, but it eventuallyrecovered to approximately 20% of wildtype. Of the ten mutations recovered, fivewere determined to compensate for theloss of ligase activity. Three mutations werein genes, like ligase, involved in DNA me-tabolism: single amino acid substitutionsin phage DNA polymerase and primase-helicase, and formation of a stop codon atposition 26 of phage endonuclease. Onemutation was a deletion of 18 nucleotidesin gene 1.5—a gene of unknown function.The final mutation was the insertion of anucleotide into gene 2.8, causing a frame-shift. The authors determined, however,that the likely beneficial effect was to in-crease the transcription of neighboringgene 3. These can be classified as loss-of-FCT and modification-of-function muta-tions. Rokyta et al. (2002) remarked thatthey initially expected the phage to acquirea new ligase activity, either by recombina-tion or by gene duplication and divergenceby point mutation. If that had happened, itwould have been a notable gain-of-FCT

mutation; however, ligase activity was notrecovered.

In another study, the gene for RNA poly-merase (RNAP) was deleted from bacterio-phage T7, and the modified phage wasused to infect cells that harbored a plasmidcarrying a gene for bacteriophage T3RNAP (Bull et al. 2007). The bacterio-phages T3 and T7 are closely related, andthe two polymerases differ in sequence byonly 18%. T3 RNAP transcribes from T7promoters only at low levels, but a singlenucleotide change in the T7 consensuspromoter sequence allows significant activ-ity by the T3 polymerase. Bull et al. (2007)used this system to follow the expectedevolution in T7 promoters to accommo-date T3 polymerase. The initial fitness ofthe phage was a factor of about 1010 lessthan wild type, but, after adaptation, thefitness approached that of wild type.Sequencing revealed that 9 of 16 T7 pro-moters had acquired mutations—mostlynucleotide substitutions and one deletion.Seven missense mutations accumulated inproteins that do not bind RNAP, and thesewere determined to be generally beneficialand not strictly compensatory for T3RNAP. Several missense mutations also oc-curred in some genes to compensate forT3 RNAP. All of the mutations can be cat-egorized as modification-of-function.

In a third study, the order of genes inbacteriophage T7 was rearranged, with theT7 RNAP gene, which normally enters thecell early, placed at an ectopic positionnear the opposite end of its genome, thusenabling it to enter the cell late. Ordi-narily, the phage polymerase would assistin pulling the phage DNA into the cell byits transcriptional activity. By placing thegene toward the further end of the phage,that activity was precluded, and the fitnessof the phage dropped nine orders of mag-nitude. Springman et al. (2005) studiedthree revertants. Two revertants recoveredonly a comparatively small portion of thefitness they lost, increasing by about 1000-fold. Both of these revertants abolished anE. coli polymerase termination site, therebyallowing the cellular polymerase, whichusually only transcribes early genes of the


phage, to continue down to the ectopic siteof the phage polymerase gene and tran-scribe it. These are loss-of-FCT mutants, asthey have lost a coded regulatory site. Athird revertant recovered a much largeramount of fitness, increasing by about amillion-fold. The originally constructedmutant that gave rise to this revertant hadhad some flanking DNA sequence of thephage RNA polymerase gene purposely leftbehind in its original position, in order toallow for subsequent recombination of theectopic gene at its original site. After aboutfive passages, the expected recombinationoccurred with a concomitant large increasein fitness. This recombination can be catego-rized as a modification-of-function mutation,because coded features were rearranged, notgained or lost.

similar viruses constrained tooccupy the same cell

Sachs and Bull (2005) examined the evo-lution of two filamentous single-strandedDNA bacteriophages of E. coli, f1 and IKe.The two phages share 55% nucleotideidentity and contain ten similar genes. Theinvestigators inserted different antibioticresistance genes into each phage, so thatcells growing in a culture containing bothantibiotics would need to be infected byboth phages in order to survive. Similarly,one kind of phage could only survive in acell if the other kind of phage were therealso. Over the course of 50 passages, thefitness of the phages grew as they increas-ingly came to be packaged together intosingle f1 coats. During the evolutionaryprocess, f1 acquired eight point mutationsand IKe acquired nine point mutations, aswell as two large deletions. A deletion of206 nucleotides of noncoding DNAaround passage 15 gave the largest in-crease in fitness, and allowed virtually allIKe to be packaged into an f1 coat. Subse-quently, around passage 40, IKe suffered amassive deletion, losing two-thirds of itsgenome and retaining only the antibioticresistance gene, plus two genes controllingcopy number and co-interference. This re-sulted in a small, additional increase inphage fitness.

One can view these adaptations from sev-eral points of view. When considering themutations individually, the large deletionin IKe is a loss-of-FCT mutation, and thesmaller deletion in IKe and the point mu-tations in both phages are likely simplemodification-of-function. When consider-ing the phages as competing organisms,however, IKe was essentially destroyed inthe evolutionary process, and f1acquiredthe use of its critical antibiotic resistancegene. This could be considered a pseudo-horizontal-gene-transfer event from IKe to f1and, therefore, a gain-of FCT event for f1.

Mutational SaturationThe longer an evolution experiment is

run, and the larger the population of mi-crobes it harbors, the greater the chance formutations to appear that are rare and partic-ularly beneficial. Lenski’s (2004) long-termexperiment with E. coli, which is approaching50,000 generations and a cumulative popu-lation size of about 1014 organisms, is theclear leader in such projects.

There are several other such experi-ments to note, which involve particularlylarge numbers of microbes or generations.Wichman et al. (2005) described the adap-tive evolution of the bacteriophage �X174over the course of 13,000 generations,which, for the rapidly reproducing mi-crobe, required only 180 days. Although itwas not reported, the total population sizeof phage over the course of the experi-ment likely reached 1013-1014. Couñago etal. (2006) replaced the essential gene for ade-nylate kinase in Geobacillus stearothermophilus—amoderate thermophile—with that of Bacillussubtilis—a mesophile—which they then grewin a turbidostat at increasing temperatures.Over the course of 1500 generations, they iso-lated six thermostable mutants of the en-zyme—one single point mutant and five dou-ble point mutants derived from the singlemutant. The cumulative number of bacteriaover the term of the experiment was 1013-1014.

Table 5 shows the calculated saturationof genomes with mutations in these large-scale experiments, which can be estimatedfor E. coli, G. stearothermophilus, and �X174by using the value for the mutation rate of


Drake et al. (1998): approximately 0.003per genome per generation for DNA-basedmicrobes. On average in the populationduring the course of long-term experi-ments, each nucleotide in each genome isexpected to be substituted from 7�103 to6�107 times; of course, the exact rate ofsubstitution could vary considerably fromnucleotide to nucleotide. Thus these exper-iments plumb the depths of what adaptivemutation can accomplish in these systemsin a single step, or in a series of single,related steps—such as Couñago et al.’s(2006) double mutants and Blount et al.’s(2008) Cit� phenotype.

Implicationscircumstances of gain- and

loss-of-fct mutationsTables 2 through 4 summarize results

from the past four decades of evolutionaryexperiments with microbes, categorizingthe adaptive mutations as loss- or gain-of-FCT, or modification-of-function. As canbe seen, only one of the adaptive muta-tions from bacteria (Tables 2 and 3) isgain-of-FCT, yet several adaptive mutationsfrom experiments with viruses (Table 4)belong to that class as well. Why the differ-ence? One reason may be that, except forthe capture of an antibiotic resistance geneby phage f1, the viral gain-of-FCT muta-tions all reconstruct functional coded ele-ments that had been deliberately removedfrom—or rendered inactive in—the ances-tral virus, thus restoring pieces of a once-integrated system. That is, they began at apoint that was known to be able to benefitfrom a gain-of-FCT mutation. The lysisgene of MS2 was rescued from a frame-shift deletion mutation by adding or delet-ing additional nucleotides to restore the

correct reading frame. The mutant thatachieved the greatest fitness was the onethat reverted completely to the wild typesequence. Similarly, the very deleteriouseffects of a 19-nucleotide deletion of MS2containing important functional codedelements (several hairpins and the Shine-Dalgarno sequence) was overcome by gain-of-FCT mutations that restored those same ele-ments to a greater or lesser degree.

Considering the time- and population-scale constraints of the experiments, it isnot surprising that, when large experimen-tal deletions were constructed that re-moved the coding sequences of wholegenes (rather than just frame shift muta-tions or short control elements), the de-leted genes were not restored. However, itwas surprising that more modest adaptivegain-of-FCT mutations were not seen ei-ther. The removal of T7 ligase resulted inpoint mutations and deletions in othergenes involved in DNA metabolism,which are loss-of-FCT and modification-of-function mutations. Intentional dele-tion of the gene for T7 RNA polymeraseand infection of a cell harboring a T3 poly-merase gene yielded mutations that appar-ently strengthened weak T3 promoters,which are modification-of-functionchanges. Rearrangement of the order ofbacteriophage T7 genes, thereby decreas-ing its fitness, did not provoke the evolu-tionary construction of new coded controlelements. Rather, one existing element waslost (an E. coli polymerase termination site)and the gene order reverted, guided byflanking DNA that Springman et al. (2005)intentionally left in the viral DNA se-quence.

A second reason why several adaptivegain-of-FCT viral mutations but few bacte-

TABLE 5Saturation coverage of mutations in three microbes

Microbe GenerationsCumulative

population size NEstimated mutationrate per nucleotide

Average fold-saturationwith mutations

E. coli 50,000 1013–1014 7 � 10�10 7 � 103–7 � 104

G. stearothermophilus 1,500 1013–1014 5 � 10�10 5 � 103–5 � 104

�X174 13,000 1013–1014 6 � 10�7 6 � 106–6 � 107


rial ones were identified might be thatRNA viruses have increased mutation ratescompared to those of bacteria. This factormay also partially explain the occurrenceof gain-of-FCT mutations in experimentswith the RNA virus MS2, but not in exper-iments with the DNA virus T7, which has amuch lower mutation rate.

Tables 2–4 also show that adaptive loss-of-FCT mutations are less common forsmall viruses than for bacteria. No suchmutations were reported for the small bac-teriophages �6, �X174, VSV, or MS2 in thereviewed laboratory experiments. The smallDNA virus IKe suffered a massive loss whentwo-thirds of its genome was deleted. How-ever, it had been co-infected into cells withthe similar small DNA virus f1, which likelycould have supplied all the functions thatIKe lost. No adaptive loss-of-FCT mutationswere sustained by f1 itself. The larger DNAvirus T7 did acquire several loss-of-FCTmutations in various experiments. It seemslikely that the general explanation for thispattern is that the smaller the virus, thesmaller the percentage of its genome thatis dispensable for its basic life cycle. Thelarger a viral genome, the greater the per-centage that is dispensable. For larger viralgenomes, and for cells, more coded ele-ments are available for adaptation by loss-of-FCT mutations.

the first rule of adaptive evolutionAs seen in Tables 2 through 4, the large

majority of experimental adaptive mutationsare loss-of-FCT or modification-of-functionmutations. In fact, leaving out those experi-ments with viruses in which specific geneticelements were intentionally deleted and thenrestored by subsequent evolution, only twogain-of-FCT events have been reported: the de-velopment of the ability of a fucose regulatoryprotein to respond to d-arabinose (Lin andWu 1984), and the antibiotic gene capture byf1 (Sachs and Bull 2005). Why is this the case?One important factor is undoubtedly that therate of appearance of loss-of-FCT mutations ismuch greater than the rate of construction ofnew functional coded elements. Suppose anadaptive effect could be secured by diminish-ing or removing the activity of a certain pro-

tein. If the gene for the protein were, forinstance, 1000 nucleotides in length, thenthere would be numerous targets of opportu-nity for a loss-of-FCT mutation. The deletion ofany single nucleotide in the coding sequencewould alter the reading frame and likely de-stroy or greatly diminish protein activity. Theinsertion of a nucleotide anywhere in the cod-ing sequence would do the same. Longer in-sertions or deletions would commonly havethe same effect, as would alteration of a codonfrom sense to nonsense. All these would fallinto the category of loss-of-FCT mutations.

Nucleotide substitutions resulting in mis-sense mutations, although not likely to com-pletely eliminate protein activity, are very likelyto diminish activity to a greater or lesser extent,as, in multiple experiments, the majority ofamino acid substitutions have been found todecrease a protein’s activity (Reidhaar-Olsonand Sauer 1988; Bowie and Sauer 1989; Limand Sauer 1989; Bowie et al. 1990; Reidhaar-Olson and Sauer 1990; Axe et al. 1996; Huanget al. 1996; Sauer et al. 1996; Suckow et al.1996). Although, if residual protein activityremained, these would be categorized as mod-ification-of-function mutations under the ac-counting system used here—that is, the partialdiminishment of the function provides theadaptive effect. (A caveat: in the particular caseof microbes that were first allowed to accumu-late deleterious mutations and then recover,such as in Bull et al. (2003), it is likely thatmany adaptive point mutations are compensa-tory for initial deleterious mutations and there-fore increase (mutated) protein function.Biochemical analysis would be needed in orderto prove that protein function increased ordecreased and was the basis of the adaptiveeffect.) If the basic point mutation rate pernucleotide per generation were 10�9, then therate of appearance of an adaptive loss-of-FCTmutation would likely be on the order of10�6, because of the many ways possible to de-crease the activity of a protein. Indeed, anadaptive mutation rate of �10�5 was recentlymeasured in E. coli (Perfeito et al. 2007).

Contrast this with a situation in which aparticular nucleotide in a gene for a cer-tain protein has to be mutated in order togain an adaptive effect. (The particularmutation can be thought to help code for


a new binding site in the protein or toconstruct a new genetic control elementfrom a sequence that was already a near-match, in addition to other possibilities.These would be gain-of-FCT mutations.) Ifthe basic point mutation rate per nucleotidewere 10�9, then that would also be the rate ofappearance of the beneficial mutation. Evenif there were several possible pathways bywhich to construct a gain-of-FCT mutation,or several possible kinds of adaptive gain-of-FCT features, the rate of appearance of anadaptive mutation that would arise from thediminishment or elimination of the activityof a protein is expected to be 100–1000times the rate of appearance of an adaptivemutation that requires specific changes to agene.

This reasoning can be concisely stated aswhat I call “The First Rule of Adaptive Evo-lution”:

Break or blunt any functional coded elementwhose loss would yield a net fitness gain.

It is called a “rule” in the sense of being arule of thumb. It is a heuristic, useful gen-eralization, rather than a strict law; othercircumstances being equal, this is what isusually to be expected in adaptive evolu-tion. Since the rule depends on very gen-eral features of genetic systems (that is, themutation rate and the probability of a loss-of-FCT versus a gain-of-FCT mutation), it isexpected to hold for organisms as diverseas viruses, prokaryotes, and multicellulareukaryotes. It is called the “first” rule be-cause the rate of mutations that diminishthe function of a feature is expected to bemuch higher than the rate of appearanceof a new feature, so adaptive loss-of-FCT ormodification-of-function mutations thatdecrease activity are expected to appearfirst, by far, in a population under selec-tive pressure.

illustrations of the first ruleThe first rule, gleaned from laboratory

evolution experiments, can be used tointerpret data from evolution in nature,including human genetic mutations in re-sponse to selective pressure by malaria (Ta-ble 1). Hundreds of distinct mutations are

known that diminish the activities of G6PDor the �- or �- chains of hemoglobin, leadingto thalassemia. Yet it is estimated that thegain-of-FCT mutation leading to sickle he-moglobin has arisen independently only afew times, or perhaps just once, within thepast 10,000 years (Cavalli-Sforza et al.1994). Thus, loss-of-FCT adaptive muta-tions in this situation appeared several or-ders of magnitude more frequently thandid a gain-of-FCT mutation. Nonetheless,the sickle hemoglobin mutation did ariseand spread in a regional population.Therefore, if a gain-of-FCT mutation suchas the sickle gene has a sufficiently largeselection coefficient, then, even thoughadaptive loss-of-FCT mutations arrive morerapidly and in greater numbers, it is possi-ble for the gain-of-FCT mutation to out-compete them.

Another illustration from nature of thefirst rule can be seen in the adaptive evo-lution of the plague bacterium Yersinia pes-tis over the past 1,500–20,000 years (Wren2003). A likely evolutionary scenario for itsgreat virulence is that the plague bacte-rium serially acquired several plasmids thatconferred upon it the ability to be trans-ferred by flea bite (Carniel 2003). The 101kb pFra plasmid carries the yplD gene,which codes for a phospholipase D that isnecessary for the survival of the bacteriumin the flea proventriculus. The 9.6 kb pPlaplasmid codes for a plasminogen activator,which allows the bacterium to move in itshost unhindered by blood clotting. Theacquisitions of these genes are gain-of-FCTevents. Y. pestis has also subsequently lost alarge number of chromosomal genes—ageneral estimate is that 150 genes havebeen lost (Carniel 2003; Chain et al. 2006).A number of discarded genes have activi-ties in other Yersinia species that allowpathogen-host adhesion. The plague bac-terium has also acquired hundreds ofmissense mutations (Carniel 2003). Thediscarded genes are of course loss-of-FCTmutations, and many missense mutations arelikely to diminish activity. Thus, the organismadapted relatively quickly to its new lifestyle—first made possible by several gain-of-FCTevents—through much more numerous loss-


of-FCT and modification-of-function muta-tions.

Two recent laboratory studies also illus-trate the first rule. Ferenci (2008) discussesthe adaptive value of mutations to the genefor a specialized E. coli RNA polymerase �factor that contributes to the general stressresponse. The author observed that “rpoSmutations occurred, and indeed spread atrapid rates within a few generations ofestablishing glucose-limited chemostats”,and also that “The majority of rpoS muta-tions accumulating in glucose-limited cul-tures are loss-of-function mutations withlittle or no residual RpoS protein. . . . Themutations include stop codons, deletions,insertions as well as point mutations” (Fe-renci 2008:447). A second study showedthat the loss of mating genes during asex-ual growth in Saccharomyces cerevisiae pro-vided a 2% per-generation growth-rateadvantage (Lang et al. 2009). The authorsnoted that “in bacteria, gratuitous gene ex-pression reduces growth rate. . . . We sus-pect that the cost of gene expression is notspecific to bacterial enzymes or genes inthe yeast mating pathway, but rather re-flects a universal cost of gene expressionand that this cost must be borne in allenvironments where the gene is ex-pressed” (Lang et al. 2009:5758). Thus inany environment in which a gene becomessuperfluous or, more generally, in any en-vironment where its loss would yield a netfitness gain, the frequent mutations occur-ring in the population that tend to elimi-nate the functional coded element willturn adaptive.

how frequently are loss-of-fct andgain-of-fct mutations adaptive?

Although the rates of appearance ofloss-of-FCT and modification-of-functionmutations that degrade protein activityare always expected to be much greaterthan the rate of appearance of gain-of-FCTmutations, a separate question concernswhat fractions of the mutations in thosecategories are adaptive. That is, althoughloss-of-FCT mutations might appear rap-idly, if they do not yield a selective bene-fit—i.e., if they are not adaptive—then they

will not usually spread in a population. Inthe same vein, a gain-of-FCT mutation mayeventually appear that builds some new ge-netic feature such as a transcription factorbinding site, for instance; yet if the featureis not adaptive within the organism’s ge-netic context, it will not be selected (Stoneand Wray 2001).

These are empirical questions that aredifficult to answer conclusively. However,the data and experiments discussed inthis review offer some insights. In the mostopen-ended laboratory evolution experi-ment (Lenski 2004), in which no specificselection pressure was intentionally broughtto bear, all of the adaptive mutations thathave been so far identified have either beenloss-of-FCT or modification-of-function mu-tations, and there is strong reason to believethat most of the modification-of-functionmutations diminished protein activity. Ex-cept in cases where specific genetic featureswere first removed, as well as in the case ofantibiotic gene capture by f1, all adaptivemutations in laboratory evolution experi-ments with viruses seem to be loss-of-FCT ormodification-of-function mutations. Thus, ingeneral laboratory evolutionary situations(that is, where a microorganism was under ageneral selective pressure rather than aspecific one), adaptive loss-of-FCT ormodification-of-function mutations werealways available. This cannot be said forgain-of-FCT mutations.

One objection might be that the aboveexamples are artificial. They concern labo-ratory evolution, and it may be that dimin-ished expression of some pre-existing,commonly-held genes gives an organisman advantage over its conspecifics in such aconstant environment, but not in the var-ied and changing environments of nature.It is true that the laboratory is an artificialenvironment, and the opportunities forsome events that occur at irregular inter-vals in the wild—such as lateral gene trans-fer—are essentially nonexistent. This isclearly an area that needs to be addressedin more detail. Further laboratory evolu-tion studies with more complex environ-ments or cultures of mixed species wouldserve to shed light on the extent to which


such factors affect opportunities for adap-tation by gain- or loss-of-FCT mutations.Nonetheless, results arguably similar tothose that have been seen in laboratoryevolution studies to date have also beenseen in nature, such as the loss of manygenes by Yersinia pestis (after, of course, theacquisition of new genetic material in theform of several plasmids), and the loss-of-FCT mutations that have spread in humanpopulations in response to selective pres-sure from malaria. A tentative conclusionsuggested by these results is that the com-plex genetic systems that are cells will oftenbe able to adapt to selective pressure byeffectively removing or diminishing one ormore of their many functional coded ele-ments.

A second possible objection is that manyof the reviewed experiments were conductedon comparatively small populations of mi-crobes for relatively short periods of time, sothat although loss-of-FCT and modification-of-function mutations might be expected tooccur, there simply was not much opportu-nity to observe gain-of-FCT mutations. Afterall, one certainly would not expect newgenes with complex new properties to ariseon such short time-scales. Although it is truethat new complex gain-of-FCT mutations arenot expected to occur on short time-scales, the importance of experimentalstudies to our understanding of adapta-tion lies elsewhere. Leaving aside gain-of-FCT for the moment, the work reviewedhere shows that organisms do indeedadapt quickly in the laboratory—by loss-of-FCT and modification-of-function muta-tions. If such adaptive mutations also arrivefirst in the wild, as they of course would beexpected to, then those will also be the kindsof mutations that are first available to selec-tion in nature. This is a significant additionto our understanding of adaptation. Thatknowledge is also a necessary prerequisite forelucidating the nature of long-term adapta-tion, as consideration of how long-term ad-aptation proceeds must take into accounthow organisms adapt in the short-term.

Furthermore, although complex gain-of-FCT mutations likely would occur only onlong time-scales unavailable to laboratory

studies, simple gain-of-FCT mutations neednot take nearly as long. As seen in Table 1,a gain-of-FCT mutation in sickle hemoglo-bin is triggered by a simple point mutation,which helps code for a new protein bind-ing site. It has been estimated that newtranscription-factor binding sites in highereukaryotes can be formed relatively quicklyby single point mutations in DNA se-quences that are already near matches(Stone and Wray 2001). In general, if asequence of genomic DNA is initially onlyone nucleotide removed from coding foran adaptive functional element, then a sin-gle simple point mutation could yield again-of-FCT. As seen in Table 5, severallaboratory studies have achieved thousand-to million-fold saturations of their test or-ganisms with point mutations, and most ofthe studies reviewed here have at leastsingle-fold saturation. Thus, one would ex-pect to have observed simple gain-of-FCTadaptive mutations that had sufficient selectivevalue to outcompete more numerous loss-of-FCT or modification-of-function mutations inmost experimental evolutionary studies, if theyhad indeed been available.

A third objection could be that the timeand population scales of even the mostambitious laboratory evolution experi-ments are dwarfed when compared tothose of nature. It is certainly true that,over the long course of history, many crit-ical gain-of-FCT events occurred. However,that does not lessen our understanding,based upon work by many laboratoriesover the course of decades, of how evolu-tion works in the short term, or of how theincessant background of loss-of-FCT muta-tions may influence adaptation.

ConclusionAdaptive evolution can cause a species to

gain, lose, or modify a function. Therefore, itis of basic interest to determine whether anyof these modes dominates the evolutionaryprocess under particular circumstances. Theresults of decades of experi-mental labora-tory evolution studies strongly suggest that,at the molecular level, loss-of-FCT and di-minishing modification-of-function adaptivemutations predominate. In retrospect, this


conclusion is readily understandable fromour knowledge of the structure of geneticsystems, and is concisely summarized by the

first rule of adaptive evolution. Evolution hasmyriad facets, and this one is worthy of somenotice.

REFERENCES

Agudelo-Romero P., Carbonell P., Perez-AmadorM. A., Elena S. F. 2008. Virus adaptation by ma-nipulation of host’s gene expression. PLoS ONE3:e2397.

Axe D. D., Foster N. W., Fersht A. R. 1996. Activebarnase variants with completely random hydro-phobic cores. Proceedings of the National Academy ofSciences USA 93:5590–5594.

Behe M. J., Englander S. W. 1979. Mixed gelationtheory. Kinetics, equilibrium and gel incorpora-tion in sickle hemoglobin mixtures. Journal of Mo-lecular Biology 133:137–160.

Blount Z. D., Borland C. Z., Lenski R. E. 2008. His-torical contingency and the evolution of a keyinnovation in an experimental population of Esch-erichia coli. Proceedings of the National Academy ofSciences USA 105:7899–7906.

Bowie J. U., Sauer R. T. 1989. Identifying determi-nants of folding and activity for a protein of un-known structure. Proceedings of the National Academyof Sciences USA 86:2152–2156.

Bowie J. U., Reidhaar-Olson J. F., Lim W. A., SauerR. T. 1990. Deciphering the message in proteinsequences: tolerance to amino acid substitutions.Science 247:1306–1310.

Bull J. J., Badgett M. R., Rokyta D., Molineux I. J.2003. Experimental evolution yields hundreds ofmutations in a functional viral genome. Journal ofMolecular Evolution 57:241–248.

Bull J. J., Badgett M. R., Wichman H. A., HuelsenbeckJ. P., Hillis D. M., Gulati A., Ho C., Molineux I. J.1997. Exceptional convergent evolution in a virus.Genetics 147:1497–1507.

Bull J. J., Molineux I. J. 2008. Predicting evolutionfrom genomics: experimental evolution of bacte-riophage T7. Heredity 100:453–463.

Bull J. J., Springman R., Molineux I. J. 2007. Com-pensatory evolution in response to a novel RNApolymerase: orthologous replacement of a cen-tral network gene. Molecular Biology and Evolution24:900 –908.

Burch C. L., Chao L. 1999. Evolution by small stepsand rugged landscapes in the RNA virus �6. Ge-netics 151:921–927.

Carniel E. 2003. Evolution of pathogenic Yersinia:some lights in the dark. Pages 3–12 in The GenusYersinia: Entering the Functional Genomic Era, editedby M. Skurnik et al. New York: Kluwer Academic/Plenum.

Cavalli-Sforza L. L., Menozzi P., Piazza A. 1994. The

History and Geography of Human Genes. Princeton(NJ): Princeton University Press.

Chain P. S., Hu P., Malfatti S. A., Radnedge L., La-rimer F., Vergez L. M., Worsham P., Chu M. C.,Andersen G. L. 2006. Complete genome sequenceof Yersinia pestis strains Antiqua and Nepal516:evidence of gene reduction in an emerging patho-gen. Journal of Bacteriology 188:4453–4463.

Clarke P. H. 1984. Amidases of Pseudomonas aerugi-nosa. Pages 187–232 in Microorganisms as ModelSystems for Studying Evolution, edited by R. P. Mort-lock. New York: Plenum Press.

Cooper T. F., Rozen D. E., Lenski R. E. 2003. Parallelchanges in gene expression after 20,000 generationsof evolution in Escherichia coli. Proceedings of the Na-tional Academy of Sciences USA 100:1072–1077.

Cooper V. S., Schneider D., Blot M., Lenski R. E.2001. Mechanisms causing rapid and parallellosses of ribose catabolism in evolving populationsof Escherichia coli B. Journal of Bacteriology183:2834–2841.

Couñago R., Chen S., Shamoo Y. 2006. In vivo molec-ular evolution reveals biophysical origins of organ-ismal fitness. Molecular Cell 22:441–449.

Crill W. D., Wichman H. A., Bull J. J. 2000. Evolution-ary reversals during viral adaptation to alternatinghosts. Genetics 154:27–37.

Crozat E., Philippe N., Lenski R. E., Geiselmann J.,Schneider D. 2005. Long-term experimental evo-lution in Escherichia coli. XII. DNA topology as akey target of selection. Genetics 169:523–532.

Cuevas J. M., Elena S. F., Moya A. 2002. Molecularbasis of adaptive convergence in experimentalpopulations of RNA viruses. Genetics 162:533–542.

Darwin C. 1859. On the Origin of Species by Means ofNatural Selection, or the Preservation of Favoured Racesin the Struggle for Life. London (UK): John Murray.

Drake J. W., Charlesworth B., Charlesworth D., CrowJ. F. 1998. Rates of spontaneous mutation. Genetics148:1667–1686.

Duffy S., Burch C. L., Turner P. E. 2007. Evolution ofhost specificity drives reproductive isolation amongRNA viruses. Evolution 61:2614–2622.

Duffy S., Turner P. E., Burch C. L. 2006. Pleiotropiccosts of niche expansion in the RNA bacterio-phage �6. Genetics 172:751–757.

Elena S. F., Agudelo-Romero P., Carrasco P., CodonerF. M., Martin S., Torres-Barcelo C., Sanjuan R.2008. Experimental evolution of plant RNA vi-ruses. Heredity 100:478–483.


Elena S. F., Lenski R. E. 2003. Evolution experimentswith microorganisms: the dynamics and geneticbases of adaptation. Nature Reviews Genetics 4:457–469.

Elena S. F., Sanjuan R. 2007. Virus evolution: Insightsfrom an experimental approach. Annual Review ofEcology, Evolution, and Systematics 38:27–52.

Ferenci T. 2008. The spread of a beneficial mutationin experimental bacterial populations: the influ-ence of the environment and genotype on thefixation of rpoS mutations. Heredity 100:446–452.

Flint J., Hill A. V., Bowden D. K., Oppenheimer S. J.,Sill P. R., Serjeantson S. W., Bana-Koiri J., BhatiaK., Alpers M. P., Boyce A. J. 1986. High frequen-cies of �-thalassaemia are the result of naturalselection by malaria. Nature 321:744–750.

Forget B. G. 1998. Molecular basis of hereditary per-sistence of fetal hemoglobin. Annals of the New YorkAcademy of Sciences 850:38–44.

Friedman M. J. 1978. Erythrocytic mechanism ofsickle cell resistance to malaria. Proceedings of theNational Academy of Sciences USA 75:1994–1997.

Hall B. G. 1982. Chromosomal mutation for citrateutilization by Escherichia coli K-12. Journal of Bacte-riology 151:269–273.

Hall B. G. 1983. Evolution of new metabolic functionsin laboratory organisms. Pages 234–257 in Evolu-tion of Genes and Proteins, edited by M. Nei and R. K.Koehn. Sunderland (MA): Sinauer Associates.

Hall B. G. 1984. The evolved �-galactosidase system ofEscherichia coli. Pages 165–186 in Microorganisms asModel Systems for Studying Evolution, edited by R. P.Mortlock. New York: Plenum Press.

Hall B. G. 1995. Evolutionary potential of the ebgAgene. Molecular Biology and Evolution 12:514–517.

Hall B. G. 2003. The EBG system of E. coli: origin andevolution of a novel �-galactosidase for the metab-olism of lactose. Genetica 118:143–156.

Hartley B. S. 1984. Experimental evolution of ribitoldehydrogenase. Pages 23–54 in Microorganisms asModel Systems for Studying Evolution, edited by R. P.Mortlock. New York: Plenum Press.

Huang W., Petrosino J., Hirsch M., Shenkin P. S.,Palzkill T. 1996. Amino acid sequence determi-nants of �-lactamase structure and activity. Journalof Molecular Biology 258:688–703.

Kemper J. 1984. Gene recruitment for a subunit ofisopropylmalate isomerase. Pages 255–284 in Mi-croorganisms as Model Systems for Studying Evolution,edited by R. P. Mortlock. New York: Plenum Press.

Kennedy J. R. 2002. Modulation of sickle cell crisis bynaturally occurring band 3 specific antibodies—amalaria link. Medical Science Monitor 8:HY10–HY13.

Lang G. I., Murray A. W., Botstein D. 2009. The costof gene expression underlies a fitness trade-off inyeast. Proceedings of the National Academy of SciencesUSA 106:5755–5760.

Lenski R. E. 2004. Phenotypic and genomic evolution dur-ing a 20,000-generation experiment with the bacte-rium Escherichia coli. Plant Breeding Reviews 24:225–265.

Lenski R. E., Simpson S. C., Nguyen T. T. 1994. Geneticanalysis of a plasmid-encoded, host genotype-specificenhancement of bacterial fitness. Journal of Bacteriology176:3140–3147.

Lenski R. E., Travisano M. 1994. Dynamics of adapta-tion and diversification: a 10,000-generation ex-periment with bacterial populations. Proceedings ofthe National Academy of Sciences USA 91:6808–6814.

Lenski R. E., Winkworth C. L., Riley M. A. 2003. Ratesof DNA sequence evolution in experimental pop-ulations of Escherichia coli during 20,000 genera-tions. Journal of Molecular Evolution 56:498–508.

Licis N., Balklava Z., van Duin J. 2000. Forced retro-evolution of an RNA bacteriophage. Virology 271:298–306.

Licis N., van Duin J. 2006. Structural constraints andmutational bias in the evolutionary restoration ofa severe deletion in RNA phage MS2. Journal ofMolecular Evolution 63:314–329.

Lim W. A., Sauer R. T. 1989. Alternative packingarrangements in the hydrophobic core of lambdarepressor. Nature 339:31–36.

Lin E. C. C., Wu T. T. 1984. Functional divergence ofthe l-fucose system in mutants of Escherichia coli.Pages 135–164 in Microorganisms as Model Systemsfor Studying Evolution, edited by R. P. Mortlock.New York: Plenum Press.

McKenna R., Ilag L. L., Rossmann M. G. 1994. Anal-ysis of the single-stranded DNA bacteriophage�X174, refined at a resolution of 3.0 Å. Journal ofMolecular Biology 237:517–543.

Modiano D., Luoni G., Sirima B. S., Simpore J., VerraF., Konate A., Rastrelli E., Olivieri A., Calissano C.,Paganotti G. M., D’Urbano L., Sanou I., SawadogoA., Modiano G., Coluzzi M. 2001. Haemoglobin Cprotects against clinical Plasmodium falciparum ma-laria. Nature 414:305–308.

Mortlock R. P. 1984a. Microorganisms as Model Systemsfor Studying Evolution. New York: Plenum Press.

Mortlock R. P. 1984b. The development of catabolicpathways for the uncommon aldopentoses. Pages109–134 in Microorganisms as Model Systems forStudying Evolution, edited by R. P. Mortlock. NewYork: Plenum Press.

Mortlock R. P. 1984c. The utilization of pentitols instudies of the evolution of enzyme pathways. Pages1–21 in Microorganisms as Model Systems for StudyingEvolution, edited by R. P. Mortlock. New York:Plenum Press.

Olsthoorn R. C., van Duin J. 1996. Evolutionary re-construction of a hairpin deleted from the ge-nome of an RNA virus. Proceedings of the NationalAcademy of Sciences USA 93:12256–12261.

Patrick W. M., Quandt E. M., Swartzlander D. B.,


Matsumura I. 2007. Multicopy suppression under-pins metabolic evolvability. Molecular Biology andEvolution 24:2716–2722.

Pelosi L., Kuhn L., Guetta D., Garin J., Geiselmann J.,Lenski R. E., Schneider D. 2006. Parallel changesin global protein profiles during long-term exper-imental evolution in Escherichia coli. Genetics 173:1851–1869.

Pepin K. M., Domsic J., McKenna R. 2008. Genomicevolution in a virus under specific selection forhost recognition. Infection, Genetics and Evolution8:825–834.

Perfeito L., Fernandes L., Mota C., Gordo I. 2007.Adaptive mutations in bacteria: high rate andsmall effects. Science 317:813–815.

Pogo A. O., Chaudhuri A. 2000. The Duffy protein: amalarial and chemokine receptor. Seminars in He-matology 37:122–129.

Pos K. M., Dimroth P., Bott M. 1998. The Escherichia colicitrate carrier CitT: A member of a novel eubacterialtransporter family related to the 2-oxoglutarate-malate translocator from spinach chloroplasts. Jour-nal of Bacteriology 180:4160–4165.

Reidhaar-Olson J. F., Sauer R. T. 1988. Combinatorialcassette mutagenesis as a probe of the informa-tional content of protein sequences. Science 241:53–57.

Reidhaar-Olson J. F., Sauer R. T. 1990. Functionallyacceptable substitutions in two �-helical regions oflambda repressor. Proteins 7:306–316.

Rokyta D., Badgett M. R., Molineux I. J., Bull J. J.2002. Experimental genomic evolution: extensivecompensation for loss of DNA ligase activity in avirus. Molecular Biology and Evolution 19:230–238.

Rozen D. E., Lenski R. E. 2000. Long-term experimentalevolution in Escherichia coli. VIII. Dynamics of a bal-anced polymorphism. American Naturalist 155:24–35.

Ruwende C., Khoo S. C., Snow R. W., Yates S. N., Kwi-atkowski D., Gupta S., Warn P., Allsopp C. E., GilbertS. C., Peschu N. 1995. Natural selection of hemi- andheterozygotes for G6PD deficiency in Africa by resis-tance to severe malaria. Nature 376:246–249.

Sachs J. L., Bull J. J. 2005. Experimental evolutionof conflict mediation between genomes. Proceed-ings of the National Academy of Sciences USA 102:390 –395.

Sauer R. T., Milla M. E., Waldburger C. D., BrownB. M., Schildbach J. F. 1996. Sequence determi-nants of folding and stability for the P22 Arc re-pressor dimer. FASEB Journal 10:42–48.

Schneider D., Duperchy E., Coursange E., LenskiR. E., Blot M. 2000. Long-term experimental evo-lution in Escherichia coli. IX. Characterization ofinsertion sequence-mediated mutations and rear-rangements. Genetics 156:477–488.

Sniegowski P. D., Gerrish P. J., Lenski R. E. 1997.Evolution of high mutation rates in experimentalpopulations of E. coli. Nature 387:703–705.

Springman R., Badgett M. R., Molineux I. J., Bull J. J.2005. Gene order constrains adaptation in bacte-riophage T7. Virology 341:141–152.

Stone J. R., Wray G. A. 2001. Rapid evolution ofcis-regulatory sequences via local point mutations.Molecular Biology and Evolution 18:1764–1770.

Stott R. 2003. Darwin and the Barnacle. New York:W. W. Norton.

Stover C. K., Kemper J., Marsh R. C. 1988. Molec-ular cloning and characterization of supQ/newD,a gene substitution system for the leuD gene ofSalmonella typhimurium. Journal of Bacteriology170:3115–3124.

Subbarao E. K., London W., Murphy B. R. 1993. Asingle amino acid in the PB2 gene of influenza Avirus is a determinant of host range. Journal ofVirology 67:1761–1764.

Suckow J., Markiewicz P., Kleina L. G., Miller J.,Kisters-Woike B., Muller-Hill B. 1996. Geneticstudies of the Lac repressor. XV: 4000 singleamino acid substitutions and analysis of the result-ing phenotypes on the basis of the protein struc-ture. Journal of Molecular Biology 261:509–523.

Verra F., Simpore J., Warimwe G. M., Tetteh K. K.,Howard T., Osier F. H., Bancone G., Avellino P.,Blot I., Fegan G., Bull P. C., Williams T. N., Con-way D. J., Marsh K., and Modiano D. 2007. Hae-moglobin C and S role in acquired immunityagainst Plasmodium falciparum malaria. PLoS ONE2:e978.

Weatherall D. J., Miller L. H., Baruch D. I., MarshK., Doumbo O. K., Casals-Pascual C., RobertsD. J. 2002. Malaria and the red cell. Hematology:American Society of Hematology Education Program2002:35–57.

Wichman H. A., Badgett M. R., Scott L. A., BoulianneC. M., Bull J. J. 1999. Different trajectories ofparallel evolution during viral adaptation. Science285:422–424.

Wichman H. A., Millstein J., Bull J. J. 2005. Adaptivemolecular evolution for 13,000 phage genera-tions: a possible arms race. Genetics 170:19–31.

Wills C. 1984. Structural evolution of yeast alcoholdehydrogenase in the laboratory. Pages 233–254in Microorganisms as Model Systems for Studying Evo-lution, edited by R. P. Mortlock. New York: PlenumPress.

Woods R., Schneider D., Winkworth C. L., Riley M. A.,Lenski R. E. 2006. Tests of parallel molecular evo-lution in a long-term experiment with Escherichiacoli. Proceedings of the National Academy of SciencesUSA 103:9107–9112.

Wren B. W. 2003. The Yersiniae–a model genus to


study the rapid evolution of bacterial pathogens.Nature Reviews Microbiology 1:55–64.

Yin F. H., Lomax N. B. 1983. Host range mu-tants of human rhinovirus in which nonstructuralproteins are altered. Journal of Virology 48:410–418.

Zinser E. R., Schneider D., Blot M., Kolter R. 2003.Bacterial evolution through the selective loss ofbeneficial genes. Trade-offs in expression involv-ing two loci. Genetics 164:1271–1277.

Handling Editor: Daniel E. Dykhuizen


Documents

EXPERIMENTAL EVOLUTION, LOSS-OF-FUNCTION MUTATIONS, … · experimental evolution, adaptation, mutation, loss of function, malaria, Yersinia pestis abstract Adaptive evolution can