R E V I E W S

58 Bullough,P.A. et al. (1994) Nature 371, 37-43 66 White,J., Matlin,K. and Helenius,A. (1981)J. Cell Biol. 89,
59 Hughson,F.M. (1995) Curr. Opin. Struct. Biol. 5, 507-513 674-679
60 Klenk,H-D. et al. (1975} Virology' 68,426-439 67 Puri,A. et al. {1990)J. Virol. 64, 3824-3832
61 White,J.M. (1992) Science 258, 917-924 68 Leikina,E., Onaran, H.O. and Zimmerberg,J. (1992) FEBS Lett.
62 Carr, C.M. and Kim,P.S. (1993) Cell 73, 823-832 304, 221-224
63 Weber,T. et al. (1994)J. Biol. Chem. 269, 18353-18358 69 Zimmerberg,J. et al. (1987) Proc. Natl. Acad. Sci. U. S. A. 84,
64 Wharton,S.A.et al. (1995) EMBO J. 14, 240-246 1585-1589
65 Chen,J. et al. (1995) Proc. Natl. Acad. Sci. U. S. A. 92, 70 Breckenridge,L.J. and Almers,W. (1987) Proc. Natl. Acad. Sci.
12205-12209 U. S. A. 84, 1945-1949

Selfish operons and speciation

by gene transfer
Jeffrey G. Lawrence

ne distinctive feature Bacterial genes providing for single arrangement to allow cotran-

O of bacterial genomes is
the operon, a cluster of
cotranscribed genes that typi-
metabolic functions are found in operons, scription; there is no selection
possibly because this organization allows
efficient horizontal transfer among
for physical proximity without
cotranscription. This rare event
cally provides for a single meta- organisms. Transferred genes can confer must be strongly selected so
bolic function. Models for the novel metabolic phenotypes on their new that when it occurs it is not lost
origins of gene clusters can be hosts and allow rapid, effective by genetic drift. Moreover, such
divided into four classes (see exploitation of new environmental niches. rare events must occur repeat-
Box 1). The 'natal' model Moreover, the mobility of selfish operons edly to place each gene into every
asserts that some gene clusters may facilitate bacterial speciation. operon. Genomic rearrange-
arose by duplication and di- ments can perform such feats3;
vergence, and it does not apply J.G. Lawrence is in the Dept of Biological Sciences, however, three caveats should
to typical bacterial operons. University of Pinsburgh, Pittsburgh, PA 15260, USA. be noted. First, two genes are
tel: +1 412 624 4204, fax: +1 412 624 4759,
Bacterial operons have evolved e-mail: jlawrenc@vms, cis.pitt.edu juxtaposed most readily by the
by the assembly of previously deletion of the intervening
unlinked ancestral genes: the DNA: this is not possible when
remaining three models describe this process. As dis- the intervening DNA encodes useful functions. Second,
cussed below, the 'selfish operon' modeP is distinct inversions and translocations that bring some genes
from the other models in several ways: (1) it provides closer together also serve to disrupt existing gene clus-
a plausible mechanism for the gradual assembly of ters. Third, the necessity for genomic rearrangements
genes into operons, (2) it provides a selection mecha- can be alleviated if operators providing for co-regulation
nism both for the assembly of gene clusters and for evolve at unlinked genes I (e.g. the Escherichia coli arg
their maintenance over evolutionary time, (3) it is genes).
consistent with the observation that genes providing Both the co-regulation model and the 'Fisher' model
for nonessential functions are found in operons and also have difficulty explaining the composition of typi-
(4) it does not postulate that gene clusters initially cal bacterial operons. The co-regulation model pre-
provided any selective benefit to host organisms. dicts that genes whose co-regulation would be most
beneficial should be found in operons, but inspection
Can operons assemble in situ?. of the E. coli genome reveals that the genes providing
Since the discovery of the operon more than 35 years for virtually all of the central metabolic processes are
ago 2, the regulatory benefit of cotranscription has been not found in operons. With notable exceptions ~, most
assumed to select for operon assembly (see the 'co- operons provide for nonessential functions [e.g. amino
regulation' model in Box 1). Although co-regulation acid biosynthesis (trp, his, leu) and carbon source uti-
is an important consequence of operon assembly, and lization (rnel, lac, pdu)]. Extensions of the Fisher model
certainly plays a role in the maintenance of operon or- predict that genes encoding coadapted proteins may
ganization, it is difficult to explain operon formation be clustered to prevent detrimental recombination be-
by selection for cotranscription. For co-regulation to tween coadapted alleles4'5. Although this is plausible
select for operon assembly, previously unlinked genes when considering genes whose products physically
must be precisely juxtaposed in a single genomic re- interact, it is difficult both to reconcile this model with
Copyright © 1997 Elsevier Science Ltd. All rights reserved. 0966 842X/97/$17.00 PII: S0966-842X(97)01110-4

TRENDS IN MICROBIOLOGY 355 VOL 5 NO. 9 SEPTE2VlBER 1 9 9 7

R E V I E W S

genetic information is also frequently transmitted hori-

Box 1. Four models for mechanisms of gene clustering zontally among isolates of unrelated taxa 6. Horizontal
The natal model transfer in prokaryotes is facilitated by common pro-
Gene clusters originate in situ by gene duplication and divergence; cesses 7 involving conjugative plasmids, transducing
therefore, the proximity of genes is an historical property and pro- bacteriophages and the acquisition and incorporation
vides no direct benefit to the individual. This model predicts that of naked DNA. Horizontal transfer has played a role
clustered genes should encode proteins belonging to the same in the introduction of many classes of genes within bac-
family (e.g. mammalian globin gene clusters). Bacterial operons
terial chromosomes, including single genes (e.g. the en-
contain unrelated genes that belong to distinct, well-characterized
families 24 [e.g. the Escherichia coil mtlA, mUD and mtlR genes en-
teric bacterial gap gene 8) and operons (e.g. the E. coli
code a PTS (phosphoenolpyruvate-dependent phosphotransferase lac operon 9) that provide metabolic benefits to the cell.
system) enzyme li, a dehydrogenase and a DNA-binding repressor, Horizontal transfer can facilitate the formation of
respectively] and do not conform to the natal model. operons from previously unclustered genes. Consider a
weakly selected function (Wsf) that requires the prod-
The Fisher model
Gene clusters result from coadaptation: frequent recombination ucts of more than one gene (wsfA and wsfB). If one of
disrupts combinations of dispersed coadapted genes. Gene clus- the wsfgenes was horizontally transferred into a taxon
ters would offer a selective benefit to individuals in a genetically lacking the encoded function (Wsf-), this gene alone
variable, freely recombining population: maladapted recombinants could not confer a selectable phenotype (Fig. la): with-
will occur less frequently if the coadapted genes are clustered. out selection, the transferred gene cannot be maintained
This model extends Fisher's theories regarding linkage disequilib- in its new genome. A selectable phenotype can only be
rium in diploid populations4's. conferred when both wsfgenes are transferred (Fig. 1 b).
The co-regulation model As known gene transfer mechanisms preferentially mo-
Gene clusters facilitate coordinated expression and regulation: co- bilize small segments of DNA, wsfgenes that are fortui-
ordinated regulation provides a selective benefit to the individual. tously located within a small chromosomal segment
This idea gained popularity following the elucidation of operon or- have an enhanced probability of comobilization into
ganization:, Co-regulation can provide selection for the maintenance, a new taxon.
but not the origin, of gene clusters 1. In this way, horizontal transfer allows the gradual,
The selfish operon model stepwise formation of gene clusters and their subse-
Gene clusters allow dissemination of functionally related genes quent integration into operons. Following transfer of a
via horizontal transfer. The physical proximity of genes provides no chromosomal segment into a new host, any genes not
selective benefit to the individual organism but does enhance the under selection are subject to deletion 3 (Fig. 2). As these
fitness of the gene cluster itself, as clusters can be efficiently in- intervening genes are of no importance to the cell (they
herited horizontally as well as vertically1. are recent arrivals), such deletions would not be detri-
mental: it is this property of the horizontally transferred
DNA that allows gradual and efficient juxtaposition
clustered genes that encode noninteracting enzymes of genes providing for single metabolic functions via
and to explain why physical proximity almost always stepwise chromosomal deletions. This process allows
includes cotranscription. the genes providing for the weakly selected phenotype
to become progressively more tightly linked and in-
Selfish operons assemble in new genomes creases the likelihood that they will be subjected to
All of the obstacles preventing the formation of gene subsequent horizontal transfer. Eventually, the tran-
clusters can be circumvented if we postulate that operon scription of the clustered genes from a single promoter
formation occurs following horizontal gene transfer. will allow effective expression of all genes required for
Although bacterial reproduction entails the vertical the selectable function. The single promoter needed
transmission of parental DNA to two daughter cells, for expression of the selfish operon may be provided
at the site of integration, thereby allowing expression of
the selfish operon in bacteria that maintain very differ-
Glossary of gene functions ent transcription machinery. As cotranscription arose
as a result of gene clustering and may have served to
mtlA: Mannitol-specific Enzyme II of the phosphoenolpyruvate-
dependent phosphotransferase system (PTS). allow subsequent efficient horizontal transfer, we can
mtlD: Mannitol-l-phosphate dehydrogenase. consider the operon organization to be a selfish prop-
mtlR: Repressor of the mannitol operon. erty of the constituent genes 1.
arE,: Arginine biosynthetic genes. The selfish operon model contends that genes as-
trp: Tryptophan biosynthetic genes. semble into operons following horizontal transfer into
his: Histidine biosynthetic genes. naive genomes (Box 1). Therefore, this model predicts
leu: Leucine biosynthetic genes. that genes providing for central metabolic functions
meh Melibiose degradation genes. are least likely to be found in operons, as genomes naive
lac: Lactose degradation genes. to these functions are rare. Instead, operons are likely
phoA: Alkaline phosphatase. to comprise genes that provide for unusual functions,
spa: Genes required for surface presentation of antigens.
cob: Cobalamin biosynthetic genes.
which can effectively invade many naive genomes. In
pdu: Propanediol utilization genes. addition, selfish operons may provide for weakly se-
pocR: Regulation of cob and pdu operons. lected functions: if this function is under sufficiently
weak selection [average selective coefficients less than

TRE~'.'DS IN MlC:ROBIOI.OC;Y 356 voL 5 NO. 9 SEPTEMBER 1997

 R E V I E W S

(a) (b)

wsfA
G ~
wsfB
,
\
\\
\ 1 1
/I ~

Horizontal transfer ~ ~ ~ _ _ /
wsfA
,wsfB

Unclustered genes
cannot be mobilized
Mutation and loss Mutation and loss
by genetic drift by genetic drift

(9 Q
Fig. 1. The propagation of selfish clusters. (a) Unclustered genes for weakly selected function (wsf) can be lost from genomes: these
genes cannot be mobilized to new genomes. (b) Clustered wsfgenes can also be lost from genomes. However, if clustered, wsf
genes are transferred to a new genome. Following transfer, any essential genes (aeg) between the wsfloci are now useless genes
(hug) that can be deleted. The solid circles represent donor genomes, and the open circles represent recipient genomes.

1/2Ne (Ne represents the effective population size) are gressed selfish operons must be compared with the
effectively neutral1°], mutations in these genes will ac- potential for diversification generated by mutation
cumulate in the population and the function will be and adaptation. To compare these values, the rate of
lost from the species by genetic drift (Fig. 1). If selection horizontal transfer among extant genomes must be
is sufficiently weak, the gene for such a weakly selected elucidated.
function may be lost by genetic drift, as the mutants are To assess the rate of horizontal transfer in enteric
not counterselected (e.g. Salmonella enterica has lost bacteria, Lawrence and Ochman analyzed E. coli and
the phoA geneH). Species that have lost a weakly se- S. enterica chromosomal DNA (Ref. 16): three critical
lected function may regain this capability by horizontal parameters were determined (see Box 2). The amount
transfer of the necessary genes (e.g. the Salmonella cob of horizontally transferred DNA present in the E. coli
operonl2). chromosome was estimated to be 620 kb (-15% of
the chromosome), which is similar to previous esti-
Selfish operons move frequently among genomes mates 13-15.All of these estimates have taken advantage
Several studies have concluded that many bacterial of the remarkable similarity in base composition, codon
genes have been introduced recently into several species usage bias and dinucleotide frequencies between native
by horizontal transfer 13-16.As predicted by the selfish genes within bacterial chromosomes. These features
operon model, many of these genes comprise operons differ markedly in horizontally transferred genes. Al-
providing for useful, but nonessential, metabolic func- though horizontally transferred genes do not resemble
tions. Among the closely related
enteric bacteria E. coli and S.
wsfA wsfB wsfAB
enterica, horizontally transferred wsfA
genes provide for the biosynthesis
of polysaccharides, pilins, surface
antigens and coenzyme B12, for
the transport of iron, citrate and
phosphate, and for the degrada- Deletion of Fusion of
tion of lactose, propanediol and intervening genes into
phosphonates 1. genes an operon
To determine if horizontally in- Fig. 2. The creation of selfish operons. Any essential genes (aeg) located between genes for the
herited selfish operons can have a weakly selected function (wsf) in the donor taxon are now useless genes (nug) in the recipient
taxon. These nuggenes are subject to rapid deletion in bacterial chromosomes 3. Gradual, stepwise
substantial impact on the evolu- deletion of unselected DNA can lead to juxtaposition of the wsfAB genes in a single transcription
tionary history of bacterial taxa, unit. Open circles represent recipient genomes, and the solid parts of the circle represent horizon-
the potential for diversification tally transferred DNA.
imparted by the gain of intro-

TRENDS IN MICROBIOLOGY 357 VOE. 5 NO. 9 SEPTEMBER 1997

R E V I E W S

their new host genome upon integration, the direc-

Box 2. Calculating rates of horizontal transfer in tional mutation pressures T M of the recipient genome
Escherichia coil will cause them to ameliorate and more closely resem-
Which genes have been horizontally transferred? ble native genes. Amelioration analysis 16, which mod-
Certain features of DNA, such as the G+C content of codon pos- els how base composition changes over time following
itions and patterns of codon usage bias, are uniform between horizontal transfer, predicts that these 620 kb of intro-
genes within bacterial chromosomes 13-16. Horizontally transferred
genes differ from these consistent patterns. Analyses of the Escher-
gressed sequences have been present in the gen- E.coli
ome for an average of 25 million years.
ichiacoligenome predict that 618 kb (more than 15% of extant To estimate a rate of horizontal transfer, it must be
genes) of extant protein-coding DNA has been introduced by hori-
assumed that bacterial genomes are not constantly
zontal transfer since the divergence of this species from the Sal-
monellalineage1G.Assuming that (1) an equal amount of DNA has growing larger. Correcting for these losses, the com-
been deleted from the E. colichromosome and (2) that all hori- bination of these two parameters yields a rate of hori-
zontally transferred DNA, and 50% of the existing DNA, is equally zontal transfer of 31 kb every million years. This rate
susceptible to deletion, we estimate that at least 803 kb of DNA predicts that the E. coli
genome has gained and lost
has been introduced by horizontal transfer 16. nearly 3000 kb of protein-coding DNA since its diver-
How long have they been there? gence from the Salmonella
lineage (Fig. 3). The func-
Immediately following horizontal transfer, introgressed genes pos- tions provided by some of these introgressed genes
sess DNA sequence features that are reminiscent of their donor would allow E.coli
to explore novel ecological niches
genome (the nucleotide composition of a DNA sequence typically in a rapid and effective manner.
reflects an equilibrium between selection and directional mutation
pressures17.~8). Over time, these features begin to resemble their Selfish operons facilitate speciation
recipient genome, as all genes in a chromosome are subject to the In most current models, organisms w i t h i n a eukaryotic
same directional mutation pressures. Genes undergoing amelio- species exchange genes with each other more often than
ration within their new genome can be identified by their highly un- with organisms of other species 19. Therefore, speciation
usual sequence features, which do not resemble any donor or re-
events generate groups of reproductively isolated organ-
cipient pattern~L Assuming standard rates of evolution, the length
of time that these genes have been ameliorating, and consequently isms from a population of interbreeding individuals.
the time at which they were introduced, can be determined. Hori- As bacteria are primarily asexual organisms 2°, bacterial
zontally transferred genes in E. colihave been ameliorating for speciation is more difficult to define in these terms.
25.3 million years zS. Despite their reproductive habits, bacterial strains can
often be sorted unambiguously into nonoverlapping
What is the rate of horizontal transfer?
The rate of horizontal transfer (31 kb every million years) is calcu- groups or species (e.g.E. coil Salmonella
and each com-
lated by dividing the amount of DNA inferred to be novel to the E. coli prises distinct groups of strains that have been well de-
genome (803 kb) by its average time of introduction (25.3 million fined by DNA sequence comparisons21,22). However,
years). This rate predicts that an amount of DNA totaling 60% of the some bacterial strains experience high rates of recom-
extant genome (3100 kb over 100 million years) has been intro- bination that may obscure species boundaries 2°. The
duced into, and much of it lost from, the E. coligenome since its features that discriminate closely related bacterial
divergence from Salmonella. taxa probably reflect sets of selective pressures inher-
ent in their individual lifestyles: each bacterial species

C h r o m o s o m e of 1 O0 million years of evolution C h r o m o s o m e of

ancestral species derived species

old EFG
6 2 0 k b of old genes newKLM
23 '~'~wNOP~ ~
oldXYZ~
vHJkKU newUVW
RSJ

...... of
transferred D N A transferred D N A
not maintained
Fig. 3. Speciation of Escherichia coli by horizontal transfer. Since divergence from the Salmonella lineage, the E. coli chromosome
has gained ~620 kb of novel DNA (newgenes) from a pool of ~3000 kb of horizontally transferred DNA (Ref. 16): most of the hori-
zontally transferred DNA has been lost (see Box 2 for calculations). A corresponding amount of ancestral DNA (oldgenes) has been
lost. Open circles represent recipient genomes, and solid parts of the circle and solid arrows represent horizontallytransferred DNA.

TRENDS IN M|CROBIOLOGY 358 voL 5 NO. 9 SEPTEMBER 1997

 REVIEWS

occupies a distinct ecological niche that provides

selection for essential, niche-specific functions. Questions for future research
Selfish operons may provide the novel functions • To what extent has horizontal transfer played a role in bacterial
necessary to invade novel ecological niches 23. Horizon- speciation in different bacterial lineages? Can phenotypes that
tally transferred DNA includes selfish operons, which distinguish between bacterial species continue to be attributed
can provide novel functions immediately upon intro- to the gain or loss of DNA, such as pathogenicity islands25?
duction (e.g. the E. coli lac operon and the Salmonella • Can the evolution of novel metabolic functions by mutation and
adaptation be attributed to any bacterial speciation events?
spa, p d u and cob operonsl). In contrast, nucleotide sub-
• Are selfish operons found in eukaryotic taxa, where mechanisms
stitutions produce 22 kb of variant DNA into the E. coil of horizontal transfer are less frequent?
genome every million years~6: such modest, incremen- • Do genomic DNA sequences show consistent clustering of genes
tal changes in the encoded products would only rarely providing for nonessential functions?
confer novel functions. Consequently, selfish operons • Do genomic DNA sequences show evidence of large-scale hori-
enable rapid, effective and competitive exploration of zontal transfer of operons between unrelated taxa?
new niches: novel functions arising by mutational pro-
cesses allow only slow, initially inefficient exploitation
of new resources. To date, no phenotype distinguishing eds), pp. 2256-2276, ASM Press
E. coli and Salmonella can be attributed to the differ- 4 Fisher, R.A. (1930) The Genetical Theory of Natural Selection,
entiation of ancestral genes by point mutation: rather, Oxford UniversityPress
all described phenotypic differences (lactose utilization, 5 Casjens,S. et al. (1992) Virology 3, 383-397
6 Smith, M.W., Feng, D-W. and Doolittle, R.F. (1992) Trends
citrate utilization, indole production and propanediol Biochem. Sci. 17, 489-493
utilization) can be attributed to the gain or loss of genes. 7 Syvanen,M. (1994) Annu. Rev. Genet. 28,237-261
Although genes providing for distinct functions must 8 Doolittle, R.F. et al. (1990)J. Mol. Evol. 31,383-388
ultimately evolve through duplication and divergence, 9 Buvinger,W.E. et al. (1984)J. Bacteriol. 159, 618-623
this slow and inefficient process is unlikely to allow the 10 Kimura, M. (1983) The Neutral Allele Theory of Molecular
competitive exploitation of novel resources required Evolution, CambridgeUniversityPress
for bacterial speciation: such a process would require 11 DuBose,R.F. and Hartl, D.L. (1990) Mol. Biol. Evol. 7, 547-577
the absence of strong competition. Instead, novel func- 12 Lawrence,J.G. and Roth, J.R. (1996) Genetics 142, 11-24
tions probably evolve when selection is not intense. 13 Whittam, T.S. and Ake, S. (1992) in Mechanisms of Molecular
Evolution (Takahata, N. and Clark, A.G., eds), pp. 223-246,
For example, a novel function might evolve to benefit
Japan ScientificSocietyPress
an organism in an existing ecological niche where selec- 14 M6digue,C. et al. (1991}]. Mol. Biol. 222, 851-856
tion for this function is not critical: the novel function 15 Ochman, H. and Lawrence,J.G. (1996) in Escherichia coil and
would not be employed to invade a novel ecological Salmonella: Cellular and Molecular Biology (2nd edn)
niche, where selection for function would be intense. (Neidhardt, F.C. et al., eds), pp. 2627-2637, ASM Press
In this manner, functions can evolve under weak selec- 16 Lawrence,J.G. and Ochman, H. (1997)J. Mol. Evol. 44,
tion (in a donor species) and allow niche exploitation 383-397
under strong selection only following horizontal trans- 17 Sue&a, N. (1988) Proc. Natl. Acad. Sci. U. S. A. 85, 2653-2657
fer. Therefore, horizontal transfer serves to decouple 18 Sueoka,N. (1992)J. Mol. Evol. 34, 95-114
speciation from metabolic differentiation by mutation 19 Vrba, E.S. (1985) Species and Speciation, Transvaal Museum
20 Maynard Smith, J. et al. (1993) Proc. Natl. Acad. Sci. U. S. A.
and adaptation. 90, 4384-4388
21 Selander,R.K. et al. (1996) in Escherichia coil and Salmonella:
Conclusions Cellular and Molecular Biology (2nd edn) (Neidhardt, F.C. et al.,
The selfish operon model predicts that bacterial genes eds), pp. 2691-2707, ASM Press
providing for weakly selected functions tend to be found 22 Whittam, T.S. (1996) in Escherichia coli and Salmonella: Cellular
in operons, as that organization allows efficient hori- and Molecular Biology (2nd edn) (Neidhardt, F.C. et al., eds),
zontal transfer among genomes of genes otherwise sus- pp. 2708-2720, ASM Press
ceptible to loss by genetic drift. Upon introgression, a 23 Lawrence,J.G. and Roth, J.R. in Horizontal Transfer
selfish operon provides a novel metabolic function, al- (SD'anen, M. and Kado, C., eds), Chapman & Hall (in press)
lowing its new host to exploit a novel ecological niche 24 Orengo, C.A. et al. (1993) Protein Eng. 6,485-500
25 Barinaga, M. (1996) Science 272, 1261-1263
effectively. In this manner, the transfer of operons be-
tween taxa may provide sufficient ecological differen-
tiation for speciation.

Acknowledgements
The selfish operon model was developed in collaboration with John Letters to the Editor
Roth. I thank RogerMilkmanfor helpfuldiscussionsand CarolineAsh, Trends in Microbiologywelcomes correspondence. Letters (not more
Dan Dykhuizen and Roger Milkman for critical comments on the than 500 words) may relate to topics raised in earlier issues of the
manuscript. journal or to other matters of general interest to microbiologists.
Please send letters to:
References Caroline Ash, Editor, Trends in Microbiology,
1 Lawrence,J.G. and Roth, J.R. (1996) Genetics 143, 1843-1860 Elsevier Trends Journals, 68 Hills Road,
2 Jacob, F. et al. (1960) C R. Acad. Sci. 250, 1727-1729 Cambridge, UK CB2 1LA.
3 Roth, J.R. et al. (1996) in Escberichia coil and Salmonella: fax: +44 1223 444630, e-mail: tim@elsevier.co.uk
Cellular and Molecular Biology (2nd edn) (Neidhardt, F.C. et al.,

TRENDS IN MICR.OBIOI.OGY 359 VOL. N O . 9 SEP "BER 1 9 9 V