AND L. HARTL · chemostatorturbidostat, in mostapplications is basically avesselforgrowinglarge, densebatch cultures. Here wefocus onexperimental workinvolv-ing chemostats because

MICROBIOLOGY REVIEWS, June 1983, P. 150-168 Vol. 47, No. 20146-0749/83/020150-19$02.00/0Copyright C 1983, American Society for Microbiology

Selection in ChemostatsDANIEL E. DYKHUIZEN* AND DANIEL L. HARTL

Department of Genetics, Washington University School of Medicine, St. Louis, Missouri 63110

INTRODUCTION......................................... 150Chemostat Theory ......................................... 151Periodic Selecdon ......................................... 153Wail Growth ......................................... 156

MUTANTS......................................... 157ACCESSORY DNA ELEMENTS: GENERAL ASPECTS.............................. 160

Plasmids......................................... 160Temperate Phage and Transposons......................................... 162

NATURALLY OCCURRING GENETIC VARIANTS................................. 164ENGNIEERED MUTATIONS .............. ........................... 165CONCLUSIONS ......................................... 165ACKNOWLEDGMENT ......................................... 165LITERATURE CITED .......... ............................... 165

INTRODUCTION

In this paper we discuss some new uses for anold technique. The technique is the continuousculture of microorganisms, which in one form oranother is familiar to every microbiologist. Theparticular continuous-culture device we willconsider is the chemostat. A chemostat is basi-cally a culture vessel having an input aperturefor the influx of sterile nutrient medium from areservoir and an overflow aperture for the effluxof exhausted medium, living cells, and cellulardebris. The device (and the term "chemostat")was invented by Novick and Szilard (103); the"bactogene" is a virtually identical device de-veloped independently and simultaneously byMonod (95). In practice, a chemostat apparatusis complicated by various attachments for aera-tion of the medium and for the prevention ofcontamination, and the overflow is often con-trolled by means of a siphon. The rate of celldivision within the chemostat can be varied by afactor of 10 by appropriate adjustment of therate of inflow, but an equilibrium will eventuallybe reached in which the number of new cellscreated by division will be exactly balanced bythe number of cells siphoned off in the overflow.This equilibrium cell density is determined al-most exclusively by the concentration of limitingnutrient in the reservoir. Thus, chemostats pro-vide an environment in which cell division iscontinuous but population size is held constant,and the two parameters can be independentlymanipulated.Although chemostats maintain a continuous

culture by means of a constant influx of nutri-ents, other continuous-culture apparatus main-tain a constant value of different parameters.The best known of these devices is a turbidostat,

which maintains a constant turbidity of an expo-nentially growing culture by means of a photo-cell that regulates nutrient inflow by negativefeedback (13). Less well-known devices includethe pH auxostat, which maintains a constant pH(82), as well as other apparatus based on specificelectrodes that maintain constant oxygen, ni-trate, or ammonium (14). Brown and Oliver (12)have developed an ingenious feedback system inwhich the addition of ethanol to a continuousculture of Saccharomyces uvarum is used tocontrol growth rate. The growth rate is moni-tored by the output of CO2 which controls therate of inflow of ethanol, maintaining a constantselection pressure for ethanol tolerance; theyobtained mutants with an ethanol tolerance up toa level of 12% (wt/vol). These devices, andchemostats as well, are all different from oneanother and different from the classical fermen-tor, which although it can be used as either achemostat or turbidostat, in most applications isbasically a vessel for growing large, dense batchcultures.Here we focus on experimental work involv-

ing chemostats because the apparatus is by farthe simplest and the literature is the most exten-sive. We will restrict ourselves to the use ofchemostats to study differential growth ratesassociated with induced, naturally occurring, orgenetically engineered mutations, as well as withthe growth rate effects of plasmids, temperatephage, and transposons. Other areas of theliterature include rich and diverse studies of thedynamics of growth in chemostats (e.g., 22);inquiries into the nature of single-nutrient anddual-nutrient limitation (e.g., 25, 30); experi-ments on the physiology and metabolism of cellsgrowing under specified chemostat conditions(e.g., 115, 116, 126); and studies of microbial

150

on May 7, 2021 by guest

http://mm

br.asm.org/

Dow

nloaded from

http://mmbr.asm.org/

VOL. 47, 1983

ecology involving single or mixed species inchemostats with single or mixed resources (e.g.,36, 45, 54, 68, 114, 117). An excellent summaryof various aspects of chemostats, including theiruse in the study of mutation rates, has beenpublished by Kubitschek (71). More recent re-views focusing on selection are by Harder et al.(57) and Calcott (16).

Chemostat TheoryA theoretical basis for understanding the

dynamics of competition for a limiting resourcein a chemostat has been developed by Monod(94, 95) and studied by Herbert et al. (64),Powell (108), Moser (96, 97), Tempest (113), andHsu et al. (66). The equations are based on twofundamental assumptions, the first being that thespecific growth rate (,u) of a strain relative to itstheoretical maximum (ILmax) is described by aconventional Michaelis saturation equation in-volving the external substrate concentration ofthe form

IJ/Anmax = S(t)/[Ks + S(t)]where S(t) is external substrate concentrationwithin the chemostat culture at time t and Ks isthe saturation constant or "assimilation parame-ter" corresponding to the substrate concentra-tion giving a half-maximal specific growth rate.The second fundamental assumption is that sub-strate is converted into cell density (x) accordingto a constant yield coefficient (y) so that

dx(t)/dt = -y dS(t)/dtFor a monoculture chemostat the Monod equa-tions are, therefore,

dSt)[SR S(t)]D IJ.maxX(t)S(t)dt y[Ks + S(t)]

dx(t) 1L..x(t)S(t)dt Ks + S(t) - Dx(t)dt Ks + S(t)

(1)

(2)

where D is the dilution rate of the chemostat. SRrepresents the substrate concentration in thechemostat reservoir.At equilibrium, dx(t)/dt = 0, so, from equation

2,

S(t) -S = KsD/(Im,,, D) (3)

and this equilibrium (9) value of S(t) is attainedat an exponential rate (39, 66).With the value of 9 given in equation 3, the

specific growth rate of cells in the chemostat issimply D, so that the doubling time in thechemostat is given by (ln 2)/D. The quantity (ln2)/D is usually referred to as the generation timein the chemostat. At equilibrium, dS(t)/dt = 0also, and the density of cells is, from equation 1,

SELECTION IN CHEMOSTATS 151

X = Y(SR - S)However, S is typically very much less than SR.In one experiment involving glucose limitation,for example, SR was set at 500 Fxg/ml and maxand Ks were found to be 0.83/h and 7 iLg/ml,respectively (33). With a doubling time of 2 h, D= (ln 2)/2 = 0.35, and therefore S = 5.1 ,ug/ml,which is smaller than SR by a factor of 100.Consequently, the equilibrium cell density islittle affected by S and therefore little affected byD, which is the basis of the assertion madeearlier that chemostats permit virtually indepen-dent control of generation time and cell density.For two competing strains the Monod equa-

tions are

dS()[SR - S(1_ Imax(1)Xi(t)S(t)

dt y[Ks(1) + S(t)]

U,fa=(2)X2(t)S(t)Y2[Ks(2) + S(t)]

dxl(t) >La(lk()XIS(t Dlt

dt Ks(l) + S(t) Dit

dX2(t) Lmax(2)X2(t)S(t) Dx2(t)dt KS(2) + S(t)

(4)

(5)

(6)

where the symbols and convenient units forlimiting organic carbon are as follows:

t = time in hoursS(t), SR = concentration of the limit-

ing substrate in the chemo-stat at time t and in thechemostat reservoir, re-spectively, in units of mi-crograms per milliliter

ILmax(O, Kf(i), y = maximum specific growthrate, saturation constant,and yield constant of strain

(i = 1, 2). Units of P'maxare hour-1; Ks, micro-grams per milliliter; and y,cells per microgram of sub-strate

x,{t) = density of strain i (in cellsper milliliter) at time t (i =1, 2)

D = fraction of the chemostatvolume replaced per hourby inflow from the reser-voir.

As is the case with monoculture chemostats,S(t) decreases from SR at an exponential rate.After this initial rapid decrease, S(:) changesslowly and has a value approximately what it


http://mm

br.asm.org/

Dow

nloaded from


152 DYKHUIZEN AND HARTL

would be were only the better of the two strainspresent in the chemostat. The value that S(t)approximates is given by equation 3 with theappropriate values of IL. and KS.At this quai-steady state in which 5(t)

the most importaAi implication of the Monodequations relative tio competition is that

d Ift[xj(t)1x2(t)] pf(1)S 7LAax(2)9dt h:s(l + Ks(2)-

which foliows from equations 5 and 6. All quan-tities. on the right-hand side of equation 7 areconstaMs; thus a pot of ln(xl/x2) against time isexpected to yield a straight line with a slopeequal to the right-hand side of equation 7. Thisslope provides a convenient measure of therelative competitive ability of the two strains,and methods of estimation of the slope and itsvarious components are discussed by Dykhuizenand Hart (38).A second principal model for the dynamics of

chemostat grwth has been developed by Droop(29) and studied further by Droop (30, 31) andGoldman and McCarthy (53). As distinguishedfrom the Monod model, whih focuses on exter-nal substrate concentration S(t), the Droop mod-el focuses on internal substrate concentration,symbolized as Q(t) and measured in units ofmass per ¢eIl. In the fundamental Droop equa-tion the specific growth rate of a strain is givenby

where- maximal specific growth rate corre-

sponding to Q(t) -X 0K= minimum concentration of limiting nu-

trient per cell required before growthcan.proceed (note that this K has a yerydifferent meaning than the saturationcon0tant Ks in the Monod model.)

Q(t) concentration of limiting nutrient percell ("cell quota"), which equals thereciprocal of the yield when excretorylosses can be neglected

For two strains competing for a limiting re-source the Droop analogs'of equations 5 and 6become

dxl[Ql g.a,(1)IQi(t) - Kllxl(t)dt Qi(t)

- Dx1(t) (8)

dX2(t) ILmax(2)[Q2(t) - K2Jx2(t)

dt Q2(t) Dx2(t) (9)

where the subscripts 1 and 2 denote strains 1 and

2. As a chemostat is progressing toward a steadystate, Qj(t) -. Q1 and Q2(t) - 2, where thecircumflexes denote equilibrium values for thetwo strains. Thus, at the steady state

d ln[x1(t)/x2(t)J ILmax(1)[Qi- K1]dt 1

nmx(2)[02- K2]02

(10)

which again indicates expected linearity in a plotof ln(x1lxj) against time. It is to be noted that theMonod model (equation 7) contns four strain-specific parameters, whereas the Droop model(equation 10) contains six. For modeling thegrowth dynamics of single strains in chemostats,the model seems to make a difference. TheMonod model seems to be most valid when thelimiting nutrient is a source ofcarbon or energy,whereas the Droop model seems best suited forlimiting nitrogen, phosphorus, or vitamins (31,53). However, when interest is focused on thechange in relative proportion of two competingstrains, as it is in studies of differential growthrates, then choice of model is no longer criticalbecause both models imply that

ln[xj(t)/x2(t)] = ln[xl(O)/x2(0)J + st, (11)where xl(t) and x2(t) represent the relative pro-portion (or number) ofthe two competing strainsat time t (conveniently measured in hours) and sis a measure of differential growth rate per unittime. The strain designated by xl is favored,neutral, or disfavored relative to the strain desig-nated by x2 as s > 0, s = 0, or s < 0. Anappropriate statistical method for analyzing datarelative to chemostat competition is to estimatexl(t) and x2(t) for various t and then estimate theslope (s) by linear regression of ln[xj(t)/x2(t)]against t, significance tests being cared out bymeans of analysis of variance. If one of thestrains is resistant to a phage or antibiotic,estimates can conveniently be carried out byplating appropriately diluted chemostat sampleson nonselective and selective plates; the lattercan be used to estimate the density of theresistant strain directly, and the density of theother strain can then be estimated by the differ-ence between the total. (nonselective) and selec-tive plates.Of course, the simple dynamics of selection

outlined above require that the competing organ-isms be competing for -the same limiting sub-strate. If other phenomena are occuring, suchas onie strain teniding to clump or to stick to thechemostat walls and so avoid washing out of thechemostat at the theoretical rate, then the

MICROBIOL. REV.


http://mm

br.asm.org/

Dow

nloaded from



dynamics of the interactions become more com-plex. The theory also breaks down when thecompeting strains interact in any way other thanby competition for substrate, such as when oneof them produces substances that promote orinhibit growth of the other.The selective difference between two compet-

ing strains is usually reported in one of twoways, either as the selection per hour (s from thelinear regression in equation 11) or as the selec-tion per generation [calculated as (ln 2)s/D].Preference for selection per generation is some-times based on the view that this quantity will berelatively independent of generation time (i.e.,dilution rate), but this is by no means the case(33). There is also the view that selection pergeneration is the more natural way to measureselection, as this is the common manner ofquantifying selection in discrete-generationmodels in population genetics (58). However, inchemostat populations, time is continuous andthe generations are overlapping, and in this casethe natural measure of selection is selection perunit time (27). In population genetics, such con-tinuous-time measures of selection are known asMalthusian parameters (27), and the relationshipbetween these measures and more conventionalmeasures of selection are nontrivial unless theamount of selection is small. Consequently, inthis review we report selection coefficients perunit time wherever possible, but in each casewill also note the relevant dilution rate.Although the Monod model and the Droop

model of chemostat growth have been the mostwidely used and investigated, other theoreticalapproaches are possible. Among these are mod-els discussed by Williams (119, 120), Dabes et al.(28), Button (15), Hellingwerf et al. (63), and avariety of models discussed by Fredrickson etal. (46) and Chiu et al. (22).

Periodic SelectionOne important aspect of a chemostat is that

the organisms within it represent an evolving orat least potentially evolving system. This point isillustrated in Fig. 1, which is taken from Kubi-tschek (72, after 101). Theoretically, a mutationat low frequency should increase linearly infrequency because of recurrent mutation, pro-vided that neither the nonmutant nor the mutantorganism has a selective advantage (104). Figure1 shows this initial linear increase in the numberof bacteriophage T5-resistant (tonA) cells intryptophan-limited chemostats in Escherichiacoli strain B/r/1 trp at a generation time of 2.8 h.(tonA, at 3 min on the standard E. coli geneticmap [8], is involved in iron uptake.) Although T5resistance has no detectable effect on growthrate in chemostats limited for tryptophan, methi-onine, succinate, glucose, or phosphate (72),

*1500 -

LU.a.

1000 000

V500 0

200 400 600GENERATION

FIG. 1. Change in frequency of phage T5-resistantmutants in a tryptophan-limited chemostat ofE. coli B/r/1 trp at a generation time of 2.8 h. Overall cell densitywas ca. 2.5 x 108 cells per ml. (From 72, after 101,with permission of Cambridge University Press.)

there are growth rate effects of T5 resistancewhen chemostats are limited for nitrogen (97). Inthis latter case, the initial increase in T5 resist-ance is nonlinear; this sort of selection, whichcan be attributed to a pleiotropic effect ongrowth rate of the particular mutation understudy, has been termed "specific selection" byAtwood et al. (6, 7) because it is specific to theknown mutation.There is, however, another sort of selection

that is of importance in chemostats. The initiallinear increase in T5 resistance does not contin-ue indefinitely (Fig. 1). Sooner or later a selec-tively favored mutation arises at another locus,almost always within the much more abundantT5-sensitive population, and as the clone bear-ing the favored mutation increases in frequencyat the expense of all others, the frequency of theT5 resistance marker drops precipitously. Whenthe favored clone reaches a sufficiently highfrequency, mutations to T5 resistance begin tooccur in this genotype, and this mutation pres-sure causes another linear increase in T5 resist-ance. At some later time a different favorablemutation occurs, again most likely within themore abundant T5-sensitive population, and thewhole process regarding TS resistance repeatsitself. Repeated episodes of this sort lead to thejagged, erratic T5 resistance curve shown in Fig.1.The sort of selection described above, which

is not associated with the mutation under study,has been termed "nonspecific selection" byAtwood et al. (6, 7). Nonspecific selectionbrings about a hitchhiking effect for all genesinitially present in the favored clone, as exempli-

VOL. 47, 1983


http://mm

br.asm.org/

Dow

nloaded from



fied for the T5-sensitive gene in Fig. 1. Atwoodet al. also refer to nonspecific selection as "peri-odic selection," and this is the term most widelyused for the phenomenon. This term is perhapsunfortunate in two respects. First, periodic se-lection suggests a special type of selection,although the selection itself is entirely conven-tional; it is distinguished from other types ofselection only in that the genes responsible for itare unknown. Second, periodic selection sug-gests a mathematical periodicity or regular re-currence at predictable intervals; periodic selec-tion is, in fact, distinguished partly by itsunpredictability of time of occurrence.At least three aspects of periodic selection are

to be considered. First, it is a potentially impor-tant phenomenon in microbial evolution, ashitchhiking effects can be quite pronounced; thisaspect of the situation has been consideredtheoretically by Kubitschek (72), Koch (70), andLevin (77). Second, periodic selection can beconsidered as a phenomenon occurring withcontinuous culture that is worthy of investiga-tion in its own right. And third, periodic selec-tion can be considered as a nuisance factor thatcan confuse the results and interpretation ofother types of chemostat studies. These lattertwo points of view warrant a brief discussion.

Surprisingly little is known about the geneticor physiological basis of the favorable mutationsinvolved in periodic selection, although theysurely differ according to the limiting nutrient.An early study of such a mutation arising in atryptophan-limited culture was carried out byNovick and Szilard (104), who determined thatthe favored strain had a decrease in its Ks (recallthat Ks is the parameter in the Monod modelcorresponding to the substrate concentration atwhich the growth is half its maximum), without achange in maximum specific growth rate. Hartland Dykhuizen (39, 59) sampled glucose-limitedchemostats every 100 h for a total of 500 h andcharted the course of increase in competitiveability by competing each sample against a com-parable sample isolated 100 h previously. Ex-cept for the first 100 h, in which changes inmaximum specific growth rate were significant,most of the evolution involved reduction in thevalue of Ks, which is in agreement with Novickand Szilard's (104) result. However, whereassuch experiments reveal the basis of selection ina general sense, they leave its molecular andphysiological mechanisms unresolved. Paquinand Adams (106) have studied evolution in hap-loid and diploid yeast chemostats limited forglucose and observed a significant change in cellshape toward a more elongated longitudinal sec-tion; over a period of 250 generations, the geom-etry of the cells changes so as to give an overall13% increase in the ratio of surface area to

volume. Again, the genetic basis of suchchanges is unknown.

In a few cases the genetic basis of periodicselection has been resolved. For example, E.coli chemostats limited for lactose regularly andpredictably undergo a takeover by lactose-con-stitutive cells within the first 100 h (37, 65, 102,112). Interestingly, the selective advantage ofthe constitutive mutation is markedly reduced inthe presence of the gratuitous inducer isopropyl-P-D-thiogalactoside, and it is eliminated alto-gether at isopropyl-p-D-thiogalactoside concen-trations above 6 ,uM (36). Such selection forconstitutive mutations is by no means unusualand has, indeed, been used to select D-serinedeaminase constitutives in E. coli (11) and man-delate constitutives (61) and mandelamidaseconstitutives (75) in Pseudomonas putida.The reproducibility of selection for lactose

constitutives suggests that independent occur-rences of the initial periodic selection event mayinvolve the same loci. McDonald (89) studiedtwo E. coli strains that had arisen and beenselected during the course of serial subculture inglucose medium (not chemostats). The geneinvolved in the selection was found to be allelicin both strains and to be linked to rpsL (strepto-mycin resistance, ribosomal protein L12; mapposition, 72 min). Thus, evolution in indepen-dent serial cultures had led to the fixation ofgenotypically and phenotypically similar mu-tants.Another example of recurrence of mutations

responsible for periodic selection events isfound in the work of Collins et al. (23) involvingthe serine-glycine-alanine uptake system, whichis associated with the cotransport of protonsdescribed as proton symport. Unselected E. colistrains exhibit a slow increase in pH of theextracellular medium associated with alaninetransport that is consistent with a proton sym-port mechanism having a stoichiometry of 1proton/alanine. A strain evolved in an alanine-limited chemostat for 2 months was found tohave a stoichiometry of 4 protons/alanine. In anindependent chemostat sampled frequently, theevolving strain was found to change from astoichiometry of 1 proton/alanine to 2 protons/alanine after about 4 days, and from 2 protons/alanine to 4 protons/alanine after about 7 days.This latter value was maintained thereafter untilthe end of the experiment (about 1 month). Suchchanges in proton symport are energeticallyinefficient but allow the cells to maintain a muchgreater concentration gradient than otherwise,which is particularly important when the sub-strate limits growth as it does in chemostats. Ascompared with unselected cells, the mutant withthe stoichiometry of 4 protons/alanine had de-creased by 20-fold the concentration of alanine

MICROBIOL. REV.


http://mm

br.asm.org/

Dow

nloaded from



at which its growth rate was half-maximal. Thisexperiment not only illustrates the repeatabilityof periodic selection under these conditions, butit also serves to emphasize the importance oftransport mechanisms of substrates that are lim-iting to growth. In addition, the experimentillustrates very well the usefulness of chemo-stats in selecting mutants for which no obvioustraditional approach to mutant selection exists.

Supplementing the direct evidence for thereproducibility of early periodic selection eventsis a great deal of circumstantial evidence. In thelong-term study ofDykhuizen and Hartl (39, 59),for example, little divergence in competitiveability was found among strains that had beenevolving for the same length of time in replicatechemostats, even though the overall amount ofincrease in competitive ability was enormous.Similarly, Chao and Cox (20), studying periodicselection in chemostats containing the mutatorallele mutT (map position, 3 min) and wild-typeE. coli, observed that the amount of selectionassociated with the mutant favored by periodicselection varied little from chemostat to chemo-stat; they suggest that the higher-fitness mutantsthat arise in these chemostats are phenotypicallysimilar, if not identical, which leads to thespeculation that they may be genetically verysimilar. Whatever the genetic basis of suchadaptive mutations, they seem to be specific tothe particular growth conditions, as exemplifiedby the finding of Helling et al. (62) that favoredmutations that arise in glucose-limited chemo-stats provide no selective advantage in fructose-limited chemostats.

Little research has been carried out on thelong-term statistical characteristics of periodicselection. Interesting but unresolved issues per-tain to the interval between successive periodicselection events, such as whether the time inter-vaJ between successive periodic selection eventsincreases, decreases, or remains the same overthe course ofa long-term chemostat experiment.One experiment of this sort was carried out byNovick and Szilard and analyzed by Kubitschek(72); in this experiment the frequency of T5-resistant mutants in tryptophan-limited chemo-stats was monitored for 1,500 hours (600 genera-tions under these conditions). During the courseof the experiment, six periodic selection eventswere observed; the first took place after about170 h, and the others occurred at intervals ofabout 240 h with a slight tendency for theintervals between successive events to increase.The selective advantage of each new favoredmutation was also estimated, and it was substan-tially smaller in the last three events than in thefirst three. A similar result was obtained byHartl and Dykhuizen (59) who, in a shorterexperiment involving only 500 h, found much

smaller adaptive changes in the last 300 h than inthe first 200 h. However, the statistical charac-teristics of periodic selection are very poorlyknown and deserve further attention.A systematic study of the long-term evolution

of Saccharomyces cerevisiae due to periodicselection in phosphate-limited chemostats hasbeen carried out by Hansche and collaborators(43, 44, 55, 56). Strain S288C was grown for1,000 generations with limitation by P-glycero-phosphate at pH 6.0. The pH optimum of thenormal acid phosphatase, coded by the phoElocus on the right arm of chromosome II, is 4.2,and a pH of 6.0 reduces its activity by about701%. At about 180 generations an increase in celldensity was noted. This was found to be associ-ated with a single mutation, leading to a 30%oincrease in the efficiency of utilization of theorthophosphate liberated by the acid phospha-tase. By about 400 generations a second percep-tible increase in cell density had occurred in thechemostat. The strain associated with this in-crease was designated M2, and it was found tohave a mutation in or near the phoE structuralgene, leading to a 60%o increase in activity at pH6.0, an increased activity over a wide range ofpH, and a change in pH optimum from 4.2 to 4.8.A third major change in the chemostat wasdetected after about 800 generations. This wasdue to a mutation that led to cell clumping,which reduced washout of cells through theoverflow and thus led to the equivalent of anincreased survival time in the chemostat (43).The mutational changes in phosphate-limited

yeast chemostats are to some extent reproduc-ible. In a separate experiment limiting for 3-glycerophosphate, Francis and Hansche (44) ob-served an increase in cell density after 290generations. The mutant strain, designated M4,had a fourfold-increased activity at pH 6.0 and achange in p1l optimum of from 4.2 to 5.2 withlittle or no change in Km. Crosses between M2and M4 gave asci that segregated 2:2 for pHoptimum, suggesting that the mutations involvedwere allelic. In this experiment a clumping muta-tion also occurred, this time its effects beingnoted at generation 650.Further evolution of chemostat-derived

strains leads to other reproducible geneticchanges. In one case, Hansche (55) maintained astrain for 1,000 generations on limiting uridine5'-monophosphate at pH 6.0. The strain hadpreviously been maintained for 1,000 genera-tions on limiting f-glycerophosphate. (Uridine5'-monophosphate is hydrolyzed at about halfthe rate of ,B-glycerophosphate.) At the end ofthis second 1,000 generations, the evolved strainexhibited an overall fourfold greater activity onuridine 5'-phosphate than its parent. About halfof this increase was due to the additive effects of

VOL. 47, 1983


http://mm

br.asm.org/

Dow

nloaded from



a1

C14~~~~~~~s-4C

CI)

-6-

100

2 2 20

za

-16-

-18

-2

40 80 120 160 200 240 280 320

HOURS

FIG. 2. Selection against strain E. coli DD323 inmaltose-limited chemostats (solid lines). Note theabrupt change in slope at 100 to 140 h. Transfer ofculture to a fresh chemostat markedly increases theselection (dashed line), indicating that the previousdecrease in slope is due to wall growth.

two unlinked mutations. One, called abs, inhib-its abscission of daughter cells from the mothercell and leads to formation of spherical clumpsof 40 to 60 cells; Hansche suggests that thismutation increases the capacity of the cell wallto accommodate acid phosphatase molecules.The second mutation, called wal, in the absenceof the abs mutation, leads to cells With little or

no wall, which under usual culture conditionsresults in cells having little geometrical integrity.The other half of the overall activity increasewas due to the appearance of a new acid phos-phatase gene unlinked to the phoE locus. Theenzyme produced by the new gene was identicalin activity and kinetic properties to that in theparental strain, and Hansche (55) proposed its

origin as a duplication.Hansche et al. (56) studied five acid phosphate

duplications that had arisen in chemostat cul-ture. Two of these arose spontaneously andthree were induced by UV light. All five duplica-tion events involved the transposition of the pholocus and all distal loci on chromosome II to theterminus of a nonhomologous chromosome. Notandem duplications were observed. Among thefive duplications oiily one was mitotically stable;the others were lost at the rate of about 2% per

mitotic division. However, selection for stabilityin one of the unstable clones resulted in a 1,000-fold increase in stability.

In addition to the pH optimum mutants, theclumping mutants, abs, wal, and the duplica-tions, in at least one case there arose a strainproducing acid phosphatase constitutively. Thisconstitutivity was found to be due to a mutationin pho-80, which codes for the acid phosphataserepressor that normally represses phoE in thepresence of inorganic phosphate (73). Clearly,evolution in chemostats has led to significantchanges in the cells themselves (abs, wal) and toextensive evolution of acid phosphatase, itsgene number, and its regulation.As for periodic selection as a nuisance factor,

the most important implication is that curves formutation or selection cannot be expected toremain linear for more than the first 100 to 200 hof chemostat culture. Eventually periodic selec-tion will occur and thereby obscure the phenom-enon under study. Moreover, if strains are to bepreadapted to chemostat conditions before beingplaced in competition with one another, it isimportant that they be preadapted under asnearly identical conditions as achievable and forthe same length of time. Even so, the possibilityof evolutionary divergence among replicate che-mostats requires that the final experiments beaccompanied by the appropriate controls.

Wail GrowthThere is another mechanism whereby a selec-

tion curve may not follow its expected course,and this is wholly distinct from periodic selec-tion. The mechanism is wall growth, the attach-ment and growth of organisms on the surface ofcontinuous-culture vessels. The various causesof wall growth are rarely distinguished. One typeinvolves "sticky" mutations that sometimesarise during the course of a chemostat experi-ment; such wall growth usually results in a thick,visible film, and its importance depends on theparticular nature of the experiment. The type ofwall growth in question here involves the innatetendency of certain organisms to attach to glasssurfaces (24). This attachment occurs immedi-ately upon initiation of the culture and results ina virtually invisible film. In tryptophan-limitedE. coli chemostats maintained at a density of 108cells per ml, this wall population accounts for<1% of the total population (98). In any event,these attached cells form a relatively permanentpopulation that experiences a different selectiveregime than occurs in the liquid. Indeed, cellsattached to the wall tend to preserve their initialfrequencies even though the frequencies in theliquid medium may change by many orders ofmagnitude. As cell division occurs, the wallpopulation serves as a constant source of migra-

MICROBIOL. REV.


http://mm

br.asm.org/

Dow

nloaded from



tion into the liquid medium. If a strain is disfa-vored by selection, the initial selection willappear quite conventional because at first themigrants constitute a miniscule proportion of thedisfavored type. However, the disfavored straincan never be reduced to a frequency smallerthan that provided by the wall population. In-deed, as the disfavored strain decreases in fre-quency, the migrants constitute an increasinglylarger proportion of its total numbers, so that themigration begins to counteract the selection. Atsome point the selection curve undergoes aperceptible change in slope, which superficiallyappears as a pronounced decrease in the rate ofselection. Examples of these dynamics are illus-trated in the solid curves in Fig. 2, which showselection involving DD320 (Strs Lac-) andDD323 (Strr Lac') in maltose-limited chemos-tats at a doubling time of 2.3 h (Dykhuizen,unpublished data).

Normally, wall growth is not a serious prob-lem in short-term chemostats (i.e., up to about100 h), and it happens rarely and apparentlyrandomly (e.g., 98). Moreover, wall growth is ofless concern at intermediate or high populationdensities or at low dilution rates because asmaller proportion of the cells in the liquid willhave been derived from the wall population.Considerable variation in wall attachment char-acteristics occurs among species. E. coli andSerratia marcescens are quite high in their wallattachment propensity, Aerobacter aerogenes isintermediate, and Bacillus cereus and B. subtilisare very low (74). Fortunately, when there isdoubt about the cause of an apparent change inthe rate of selection, the hypothesis of wallgrowth is very easy to test. One simply initiatesa new chemostat from the liquid medium in theold one. If the rate of selection increases dramat-ically, wall growth is very likely involved. Oth-erwise genetic or physiological changes in thecells are to be suspected. This test is illustratedby a chemostat in Fig. 2 (dashed curve), whichwas initiated at 230 h with half of the liquidvolume in the chemostat indicated by the soliddots; the dramatic increase in rate of selection inthe newly initiated chemostat is evident. A sys-tematic study of various aspects of wall growthin E. coli chemostats has recently been carriedout by Goguen (Ph.D. thesis, University ofMassachusetts, Amherst, 1980).

MUTANTSChemostat studies are particularly useful for

determining quantitatively the functional effectsof known mutations. In such studies the phe-nomena of periodic selection and wall growthare rarely significant problems because the che-mostats are initiated with known frequencies oftwo appropriately marked competing strains and

are maintained for a relatively short period (e.g.,100 h) to estimate the initial change in relativefrequency of the strains. The method is suffi-ciently sensitive to detect very small differencesin growth rate. Using the sampling procedure ofDykhuizen and Hartl (38), for example, differ-ences in growth rate as small as 0.005 per h canbe detected with relative ease. Because of thissensitivity, the degree to which the competingstrains are isogenic becomes a critical issue, andvarious types of control experiments are neces-sary. We (38, 60) have distinguished three typesof controls. A positive control involves a pair ofstrains that undergo competition in a chemostatunder conditions in which the phenomenon ofinterest can readily be observed. For example,in studying the functional effects of a leakymutation, an appropriate positive control wouldinvolve a deletion mutation in competition withwild type in chemostats limited for the substrateof the enzyme in question; the role of thepositive control is to demonstrate that selectionin the anticipated direction can be detectedunder the chosen conditions. That having beendemonstrated. the negative control comes intoplay. The negative control involves the samestrains as the positive control, but now theconditions are so chosen that the selection dis-appears. In dealing with a specific enzyme, forexample, the ideal negative control involves achemostat limited for the product of the reactionor for a substrate whose metabolism does notinvolve the enzyme in question. In the negativecontrol the selection observed in the positivecontrol should disappear, which leads to theinference that the gene in question is responsiblefor the observed selection. The role of thenegative control is basically to detect effects thatmay be due to unrecognized genetic differencesbeteween the strains. Yet a third type of controlis a strain control, which is necessary whenevera genetic marker other than the mutation ofinterest is used to monitor the frequency of thecompeting strains, and its purpose is to verifythat the marker itself is not selected under theexperimental conditions. Useful genetic markersfor chemostats in E. coli include tonA (72) andrpsL (Dykhuizen, unpublished data; but see 41),which have negligible effects of their own inglucose-limited chemostats, and those in S. cer-evisiae include canavanine resistance (1).Although diploid and haploid strains of S.

cerevisiae can scarcely be considered "mu-tants" of one another, it suits our organizationto discuss them under this heading. Adams andHansche (1) have studied virtually isogenic hap-loid and diploid strains in chemostats limited forglucose, inorganic phosphate, or organic phos-phate. (So far as is known the strains weregenetically identical except for the mating type

VOL. 47, 1983


http://mm

br.asm.org/

Dow

nloaded from



locus and the canavanine resistance locus.) Withlimitation by glucose or inorganic phosphate, thehaploid and diploid strains were found to becompetitively equal; with limitation by organicphosphate the haploid strain was found to btcompetitively superior to the diploid. Adamsand Hansche (1) relate these results to the natureof the limitation and to cellular geometry asmeasured by the ratio of surface area to volume.With glucose limitation the surface area/volumeratio of diploids and haploids is virtually identi-cal, so little selection is expected. The surfacearea/volume ratio in diploids is reduced withlimitation by inorganic or organic phosphate.However, Adams and Hansche (1) argue thatgrowth rate under limitation by inorganic phos-phate is deternined by efficiency of substrateutilization as opposed to uptake. In this case thesurface area/volume difference would have noconsequence. With limitation by organic phos-phate, they argue, growth rate is limited byuptake, and in this case the haploids displace thediploids by virtue of their greater surface area/volume ratio. (For further discussion see-.118).One of the earliest chemostat studies of the

functional effects of laboratory mutants was thatof Zamenhof and Eichhorn (125). Prompted bythe hypothesis of Lwoff (81) that auxotrophicmutants would have a selective advantage overisogenic prototrophs in nonselective media be-cause of energetic economy due to fewer biosyn-thetic steps, they examined the effects of His-and Trp- auxotrophs in B. subtilis. When grownin the presence of histidine, the His- mutantwas favored at the rate of about s = 0.05 per h (D= 0.17), and in the presence of tryptophan, theTrp- mutant was favored at about the same rate.This is a remarkably high rate of selection inas-much as, under these conditions, the trypto-phan-synthesizing genes have only the residualactivity ofa repressed operon. Nevertheless, thestrains involved in the competition were isogen-ic (the prototroph was selected as a spontaneousrevertant of the auxotroph in each case), and thebasic result for Trp- has been confirmed byDykhuizen (33) and extended to tyrosine auxo-trophs by Mason and Slater (86). Zamenhof andEichhorn also competed an anthranilate-requir-ing strain against a tryptophan-requiring strain inthe presence of tryptophan. Selection favoringthe anthranilate-requiring strain occurred at therate of 0.288 per h, favoring their view thatearlier metabolic blocks are favored over laterones, presumably because of energetic econo-my; but in this case the strains were evidentlynot isogenic so the interpretation of the result isthrown in doubt.Whereas the results of competition between

auxotrophs and prototrophs in nonselective me-dia are clear and reproducible, the interpretation

based on energetic economy has not been con-firmed. The experiments lack a negative control,for example. Presumably, if energetic economyis involved, the selection should disappear inchemostats limited for something other than asource ofenergy, such as nitrogen or phosphate.In addition, several predictions of the energy-economy hypothesis have not been confirmed(33). First, reasonable calculations of theamount of energy saved under repressed condi-tions by not making the tryptophan-synthesiingenzymes and producing excess tryptophan sug-gest that the savings are on the order of0.01% ofthe total energy budget; this is 500 times smallerthan the observed amount of selection and is,indeed, below the limit of resolution of presenttechniques. Second, a polar nonsense mutationis not favored over a missense mutation, con-trary to expectation. Third, the selection for aTrp- mutant is the same in chemostats suppliedwith indole as it is with anthranilate, althoughenergetic considerations imply that there shouldbe a difference. Dykhuizen suggests that theselection for auxotrophs may be a consequenceof some general metabolic effect of tryptophanintermediates in the wild-type strain. In short,although selection for auxotrophs is dramaticand most likely an important phenomenon inmicrobial evolution, the metabolic basis for theselection can be disputed. The work of Collins etal. (23) cited earlier in connection with protonsymport in alanine transport provides a counter-example to metabolic economy inasmuch as themutants selected in the chemostat have evolvedan energetically less efficient transport mecha-nism. Although energetic economy must be im-portant in some overall, general sense, examplessuch as this one imply that it cannot be uncriti-cally invoked.The importance of metabolic economy has

also been tested explicitly by Andrews andHegeman (5), using a set of five isogenic strainscarrying various F'lacO' (lactose-constitutive)episomes grown in competition in turbidostats,which provides a comparison of growth ratesnear Um.a, The amount of lac operon proteinsproduced by these constitutives ranged from 0.6to 3.6% of total celiular protein, but no simpledirect relationship between the level of lac oper-on expression and selective disadvantage wasfound. Cultures were grown in 0.2% glycerol at20°C, corresponding to j,u,, = 0.10 h-1. Allstrains differed from the F'lac+ strain by theamount 0.02 selection per generation or more,which was attributed to possible genetic diver-gence between the strains arising from 12 gener-ations of preadaptation in monoculture turbido-stats. Three strains producing <0.91% lacprotein had a selective disadvantage of 0.021 or0.022 per generation, and two strains producing

MICROBIOL. REV.


http://mm

br.asm.org/

Dow

nloaded from



1.6 and 3.6% lac protein had a selective disad-vantage of about 0.026 per generation. Clearlythere is no linear relationship, and the correla-tion between selection rate and percentage ofwasted protein is not statistically significant. Ifanything, there is a sort of threshold effect inwhich strains producing a small amount of un-needed protein are selected against mildly if atall, whereas strains producing much more thanseveral percent of unneeded protein are at adetectable disadvantage. Andrews and Hege-man (5) suggest this explanation and provideadditional evidence that the amount of proteinsynthesis is not limiting to growth under theirconditions. Dykhuizen and Davies (36) havesuggested that the selection against high-levellac operon expression can be explained in partby indirect metabolic effects.Another argument for selection based on ener-

gy economy has been put forward in the case ofa petite mutant (rhoo no. 7) of S. cerevisiaerelative to its parental strain (C86-2-15B a) inanaerobic chemostats. This petite mutant has<5% of the normal complement of mitochondri-al DNA as assessed by density gradient centrifu-gation and fluorescent staining. James (67) hasshown that the petite mutant is selectively fa-vored over the wild type in anaerobic chemo-stats limited for glucose at a variety of genera-tion times. Two curious aspects of the selectionwere noted. First, there is a lag that occursbefore the onset of selection, which is particular-ly pronounced at a short generation time. Sec-ond, the displacement of the wild type does notgo to completion. Nevertheless, the petite strainis favored at a rate s = 0.03 (D =0.04) to s = 0.07(D = 0.16) per h depending on the growthconditions, and James (67) speculates that thismight be due to a smaller size of petite cellsleading to a smaller energy expenditure in main-tenance and reproduction. On the other hand,smaller cells also have a more favorable surfacearea/volume ratio, and this might account for atleast some of the selection.Gibson et al. (51) stimulated an interesting

series of experiments by their finding that strainscarrying the mutTI mutator gene were favoredover their otherwise isogenic counterparts inglucose-limited chemostats. In these experi-ments, the selective advantage of mutTl strainsranged from s = 0.004 (D = 0.46) to 0.032 (D =0.28), with an average of 0.014. However, it wasoriginally unclear whether the selective advan-tage was due to some unknown metabolic effectof mutTi or to mutTI strains having a greaterchance of undergoing a favorable mutation. Thisproblem was investigated further by Nestmannand Hill (99), who found a similar effect ofmutHi (map position, 61 min). The selectionwas as high as s = 0.07 (D = 0.10) in a few cases,

but interpretation is complicated in this casebecause the Gal- marker used in the strains wasnot itself neutral. In any event, if the selectiveadvantage is due to new favorable mutations,then the selection should be frequently depen-dent in the sense that the initial inoculum ofmutator cells should be able to be made smallenough that the mutational advantage of themutator could be overcome by the numericalpreponderance of the normal strain. Under theseconditions the mutator strain should be eliminat-ed from the chemostat. Nestmann and Hill (99)did observe this expected frequency dependencein two cases out of four, but the expected lagpreceding the onset of selection caused by afavorable mutation becoming fixed in the non-mutator strain was not observed.Further evidence on the mechanism of selec-

tion for mutators was provided by Cox andGibson (26). They studied mutator strains thathad survived competition in chemostats andfound that they were competitively superior in atleast three ways. First, they had an enhancedtendency to stick to glass powder, suggesting agreater ability to stick to the walls of the chemo-stat vessel. Second, they could utilize citrate, achelating agent in the chemostat medium, as anenergy source. And third, they were more resis-tant to starvation for glucose than were wild-type clones taken from the same chemostat.These results strongly support the mutationalorigin of the selective advantage of mutTI.Chao and Cox (20) have recently reinvestigat-

ed the kinetics of mutTi selection in chemostats,and they find the characteristics expected fromthe mutation hypothesis. There is indeed a lagbefore the onset of selection, averaging about 70generations and evidently due to the time re-quired for a favored mutation to become nearlyfixed in the mutTI population. Remarkably, thedegree of favorability of the favored mutant wasquite consistent from chemostat to chemostat,which was interpreted as meaning that the muta-tions could be genetically similar if not identical.Moreover, the kinetics of selection were consis-tent with the view that the favored mutationswere present in the original inoculum. Thisconclusion that early periodic selection eventsinvolve preexisting mutants was also reached byHelling et al. (62) relative to nonmutator strains.In any case, the mutator strain is not alone inhaving the favored mutation. The normal strainhas it, too, but in lower frequency. This resultsin the normal strain not being completely re-placed by reaching a plateau in its frequency,which persists until still more favorable muta-tions occur.Chao and Cox (20) also observed the expected

frequency dependence of selection. Where theinitial ratio of mutTllmut+ is 7 x 10-5 or great-

VOL. 47, 1983


http://mm

br.asm.org/

Dow

nloaded from



er, the mutT) strain increases in frequency afterthe lag, but when the ratio is less than 7 x 10-5,the mutT) strain decreases in frequency after thelag. This is tolerably in agreement with whatwould be expected based on the ratio of muta-tion rates in the competing strains, and indeedthe entire course of selection is in good agree-ment with the theoretical predictions put for-ward by Painter (105).

ACCESSORY DNA ELEMENTS: GENERALASPECTS

Campbell (18) has discussed the evolutionarysignificance of accessory DNA elements inmicroorganisms. Accessory DNA includes suchelements as plasmids, viruses, transposons, andinsertion sequences and can be distinguishedfrom typical chromosomal DNA by two proper-ties. The first is that the elements generallycontain no genes that are unconditionally re-quired for growth and reproduction of the orga-nism in its usual environment. Second, theseelements are able to replicate autonomously andin some cases increase their copy number rela-tive to the chromosomal DNA. Campbell arguesthat these elements should show coadaptationwith the host and that they might indeed benefitthe host in some manner under particular envi-ronmental conditions. If this is correct, then itwill be important to determine the conditionsunder which these elements provide a growthadvantage. These conditions will be relevant tothe mechanisms leading to the persistence ofsuch elements in populations. The relevance ofchemostats to these issues has been argued byWoods and Foster (121): "When the host iscomplex, and has differentiated tissues, it willhave some form of circulatory system. Thebacterium will therefore be in an environmentwhich is constantly renewed in the sense thatthere will be, due to the activities of the host,both a continual replenishment ofgrowth metab-olites and a continual removal of waste productsof metabolism. The situation is essentially simi-lar for organisms living in an intestinal tract. Thesystem is therefore analogous in many ways tocontinuous culture and less like the batch cul-tures that are the source of so much of thebacterial material used for metabolic studies invitro." See Mason and Richardson (84) for ageneral discussion of presumed conditions in themammalian intestinal tract and Mason and Rich-ardson (85) for the use of chemostats to simulatethe human gut. Freter and collaborators (47, 48,50) have recently developed an anaerobic con-tinuous-culture system containing a set of 95strictly anaerobic strains isolated from mousececal contents. This synthetic indigenous micro-flora can simulate many of the bacterial interac-tions observed in the mouse large intestine.

Plasmids

A substantial literature has developed con-cerning the behavior of plasmids in continuousculture, motivated in part by the as yet largelyunexploited potential of these techniques in in-dustrial application. The sort ofproblem that hasto be overcome is exemplified in the case ofplasmid pSC101 trp.I15.14 in E. coli W3110 ashost. The host in this case has a deletion of thetrp operon and of the tryptophanase gene, and ithas an inactive trp repressor; the plasmid carriestetracycline resistance and a trp operon contain-ing a mutant trpE gene that is insensitive tofeedback inhibition. In batch culture, this strainproduces 0.23 g of tryptophan per liter per h,which would warrant industrial production of L-tryptophan by fermentation (3). However, incontinuous culture, the plasmid is unstable inthat deletions and other mutational alterations ofthe trp operon are selected (32). Clearly, anunderstanding of the factors promoting plasmidstability would be an important achievement.Many general considerations of plasmids are

based on the supposition that the maintenance ofa plasmid represents a significant burden on theeconomy of the cell. In some cases this isdemonstrable. Many, but by no means all, plas-mids increase host generation time. Zund andLebek (127) have studied 101 R factors from thispoint of view. About one-quarter of these in-creased host generation time by more than 15%.Among R factors of >80 kilobases in size, themajority increased generation time. Beyond thisgeneral observation there appears to be no con-sistent relationship between plasmid size andeffect on growth rate under nonlimiting condi-tions.

In many cases a plasmid-free host readilyoutcompetes its plasmid-bearing competitor inchemostats. This is true of plasmid RP1, aconjugative R factor carrying resistances to am-picillin, carbenicillin, tetracycline, and kanamy-cin/neomycin, in E. coli host W3110 (91). Inter-estingly, the plasmid-free cells take over only inphosphate-limited chemostats in this case; underglucose or magnesium limitation the plasmid-free cells are gradually lost from the chemostats.Even in phosphate-limited chemostats when theplasmid-free strain takes over, takeover is notcomplete and the plasmid-bearing cells are sta-bly maintained at a low frequency. This persis-tence of plasmid-bearing cells has been noted inseveral other cases (52, 69). This could be atrivial artifact of wall growth of the plasmid-bearing strain, or it might represent some unrec-ognized mechanism of maintaining plasmids atlow frequency that would be highly relevant tothe persistence of plasmids in natural popula-tions. Godwin and Slater (52) have pointed out

MICROBIOL. REV.


http://mm

br.asm.org/

Dow

nloaded from


VOL. 47, 1983

that the selection against plasmid-bearingstrains, when it occurs, is sometimes muchgreater than would be expected from energeticeconomy. In a related context Helling et al. (62)have defined plasmid interference as those meta-bolic effects resulting from a plasmid that aredistinct from plasmid maintenance effects.

Helling et al. (62) have also studied thedynamics of competition between plasmid-bear-ing and plasmid-free strains. The host in thiscase was E. coli RH204 and the plasmid was oneof several non-colicin-producing derivatives ofRSF2124, which also carries transposon Tn3(amp). When mixed chemostats are initiated at asufficiently high frequency of the plasmid-bear-ing strain, its frequency at first declines, thenincreases again, and then decreases. These os-cillations are due to periodic selection. They aremediated by mutations of chromosomal genesbecause the favorable mutations do not cotrans-form with plasmid DNA. The nature of theoscillations is quantitatively in agreement withthe expectations of an extension of the Monodmodel to periodic selection. In a related study,Adams et al. (2) examined selection involvingthe original ColEl-producing plasmid RSF2124,which is 11.3 kilobases in size and present in 10to 30 copies per cell. In this case there is apronounced frequency-dependent effect: whenthe colicin-producing cells are at a sufficientlyhigh frequency initially, they produce enoughcolicin to inhibit growth of their competitors andincrease in frequency to near fixation; when theinitial frequency of the colicin producers is low,the inhibition of competitors is insufficient tocounteract the intrinsic growth rate disadvan-tage of the plasmid-bearing strain, and the plas-mid-bearing strain decreases in frequency. Inthe longer term, after 100 cell generations ormore, plasmid-free but colicin-resistant cells be-gin to take over the chemostat. However, instructured habitats (soft agar), the conditions aremore favorable for selection for colicinogenicbacteria (21).

Competition between a toxin-producing strainof S. cerevisiae and a sensitive strain in chemo-stats has been studied by Young and Philliskirk(124). Strains NCYC738 and NCYC235 are killeryeasts producing a glycoprotein toxin associatedwith a cytoplasmically inherited double-strand-ed RNA. These strains and their killer-curedderivatives all displace a sensitive nonisogeniccompetitor (NCNY1006) in glucose-limited che-mostats, but displacement by the killer strainsoccurs more rapidly, particularly under condi-.tions of temperature, pH, dilution rate, andother factors that promote production of thekiller substance. This phenomenon is relevant inthe brewing industry where, in one case ofcontinuous fermentation, a killer contaminant


displaced the resident and dominated the fer-mentor (88).The fate of a monoculture chemostat contain-

ing a plasmid-bearing strain depends on theplasmid, the host, the limiting nutrient, and thedilution rate. A key factor is segregational lossof the plasmid, theoretical aspects of which havebeen studied for F factors by Anderson andLustbader (4). Jones et al. (69) found that plas-mids RP1, pDS401 (a ColK derivative), andpDS1109 (a ColEl derivative) were maintainedat least for 120 generations in glucose- or phos-phate-limited chemostats. Interestingly, thecopy number of pDS1109 fell by fivefold during80 generations but was restored to its originallevel after a single cycle of batch growth. Incontrast, plasmids pMB9 and pBR322 generateplasmid-free segregants after approximately 30generations. Jones et al. (69) suggest that fidelityof segregation at low copy number determinesstability and argue that the par locus of pBR322(90) might well be involved. Similarly, in studiesof RP1 (91), the plasmid was stably maintainedin chemostats limited for carbon, magnesium, orphosphate and at all dilution rates between 0.05and 1.0 h-1. A similar stability of R6 has beenreported by Wouters and van Andel (123).

Factors influencing plasmid stability havebeen studied by Roth et al. (109), Noack et al.(100), and Roth and Noack (110). The importanceof the plasmid and conditions is exemplified bythe difference between pBR322 and pBR325.Under the glucose- and nitrogen-limiting condi-tions of Noack et al. (100), pBR322 is stablymaintained, whereas pBR325 is lost under limit-ing glucose at low dilution rates (0.15 h-1).Temperature effects are apparent in the work ofWouters et al. (122), who found conditions un-der which pBR322 was lost from chemostats at37, 40, and 42°C but stably maintained at 30°C.Host genotype is also important, as pBR322 ismaintained in strains E. coli GY2354 and GM31but not in an E. coli K-12 recA recB sbcB host(100).The fate of a plasmid in chemostats is fre-

quently not simply a matter of retention or loss.In certain cases when pBR325 is lost, for exam-ple, the tetracycline resistance phenotype is lostvery rapidly, whereas the ampicillin and chlor-amphenicol resistances cosegregate and are lostmore slowly (100). Perhaps unexpectedly, thetetracycline-sensitive derivative of the plasmidexhibits no detectable change in molecularweight. An even more complex situation regard-ing marker loss from TP120 (=R46), which car-ries resistances to tetracycline, ampicillin, strep-tomycin, and sulfonamide, has been studied byGodwin and Slater (52). In carbon-limited che-mostats, only the tetracycline resistance is lost.By contrast, in phosphate-limited chemostats,


http://mm

br.asm.org/

Dow

nloaded from



the tetracycline or the ampicillin resistance, orboth, may be lost. The resistances lost and theirpattern of loss vary from experiment to experi-ment, and various mutant plasmids may befound in the same chemostat. The molecularbasis of these changes is unknown.The transmissibility ofconjugative plasmids in

chemostats has also been studied with an eye todeveloping satisfactory models for the mainte-nance of plasmids in nature (76). Levin et al. (79)have estimated the transfer rates in glucose-limited chemostats of three plasmids: F'lac pro;Rl (kanamycin, chloramphenicol, ampicillin), anaturally occurring plasmid; and Rldrd-19, aderivative of Rl derepressed for transfer func-tions. Rates of transfer for F'Iac pro and Rlwere 3.26 x 10-12 and 2.55 x 10-l ml/cell perh, respectively, whereas that for Rldrd-19 was1.83 x 10-11 ml/cell per h. The rate of transfer inchemostats is thus low but by no means negligi-ble and is markedly influenced by derepressionof transfer functions. Moreover, rates of transferin chemostats are smaller than in exponentialcultures by toughly 2 orders of magnitude (79).

Freter et al. (49) have also studied the transferof plasmids Rl and Rldrd-19 among E. colistrains maintained in anaerobic continuous cul-tures along with 95 strictly anaerobic strainsconstituting a synthetic large intestinal microflo-ra which can simulate bacterial interactions ob-served in the mouse gut. The density of E. colimaintained in this complex community is muchless than in cultures containing no competingmicroflora, and since the plasmid-bearing donorstrain is added to the culture after the culture isequilibrated and cannot establish a permanentpopulation of its own, plasmid transfer occurs intwo distinct ways: first, via direct transfer fromdonor to recipient while the donor is passingthrough the culture vessel; and second, viatransfer from transconjugants (plasmid-infectedrecipients) to uninfected recipients within thestable population. Remarkably, the donor-to-recipient transfer rate constant for plasmid Rl inthis complex community was 2.3 x 10-14 ml/cellper h (average of 12 experiments), which isnearly identical to the estimate of Levin et al.(79); transfer of Rldrd-19 seems to be somewhatimpaired by the presence of indigenous micro-flora, its donor-to-recipient transfer rate being3.1 x 10-i ml/cell per h (average of 5 experi-ments). However, the rate constants for trans-conjugant-to-recipient transfers were 5 to 6 or-ders of magnitude higher than the donor-to-recipient rates, being 3.9 x 1i-8 (average of 12experiments) in the case of Rl and 1.5 x 10-7(average of 5 experiments) in the case ofRldrd-19. This large difference occurs in part becauseof derepression of RI transfer functions in thetransconjugants but also because the donor

strains are near stationary phase whereas thetransconjugants are growing, and stationary cul-tures have lower transfer rates (79).Long-term continuous cultures studied by

Freter et al. (49) revealed that plasmid transferoften does not go to completion. Rather, thepopulations of transconjugants and potential re-cipients achieve and maintain constant levels.This phenomenon was analyzed by means of aseries of increasingly complex mathematicalmodels incorporating such features as repres-sion oftransfer functions (relevant to plasmid Rlonly), segregational loss of the plasmid, reduc-tion in growth rate of plasmid-bearing bacterialhosts, competition for nutrients, and bacterialattachment to t4e wall of the gut or culturevessel. The long-term results were best approxi-mated by a model assuming an inferior growthrate and a small wall-attached portion of thepopulation of the plasmid-bearing bacteria. Plas-mid transfer efficiencies were also studied ingnotobiotic mice carrying a synthetic indigenousmicroflora resembling in their functions the nor-mal indigenous microflora of the mouse largeintestine. On the whole, plasmid transfer in vivodid not differ from that occurring in vitro in thepresence of gut-simulating anaerobic strains,and any peculiarities noted in vivo could bereferred to the population density of E. coli inthe gut.Levin and Rice (78) have also studied the

transfer rates of a nonconjugative plasmid(pCR1) when mobilized by an F'lac pro mobiliz-ing factor. In glucose-limited chemostats theyestimated an overall transfer rate of 6.4 x 0-12ml/cell per h, with about 26% of the transfersinvolving only the nonconjugative plasmid, 68%involving only the conjugative plasmid, and 6%involving both. Again, the rate of transfer inchemostats was about 2 orders of magnitudesmaller than in exponential cultures. Such "tri-parental matings" have been studied theoretical-ly by Freter et al. (49), who find in computersimulations that high-efficiency transfer of theconjugative plasmid is actually detrimental tothe transfer of the nonconjugative plasmid, thereason being that the conjugative plasmid rapid-ly sweeps through the population of potentialrecipients and so renders this population im-mune to subsequent cotransfers of the nonconju-gative plasmid. They suggest that the low levelof fertility of wild-type E. coli strains may actu-ally be optimal for maintaining an ecologicalbalance among diverse types of plasmids.

Temperate Phage and TransposonsChemostats have also been used to examine

the selective forces impinging on lysogens andon transposon-bearing strains. Theoretical con-siderations relating to this point have been orga-

MICROBIOL. REV.


http://mm

br.asm.org/

Dow

nloaded from


VOL. 47, 1983

nized and cogently discussed by Campbell (17,18) and Campbell et al. (19).Paynter and Bungay (107) have studied the

temperature induction of bacteriophage A instrain E. coli 159T- in continuous culture. Aculture initially inoculated with the lysogen andmaintained at 30°C rapidly achieves a steady-state cell density of lysogens. Shocking theculture briefly at 42°C leads to a rapid 100-folddecline in cell density due to induction of theprophage. The density then gradually increasesfor about 24 h and achieves another steady-statelevel slightly greater than the original, presum-ably because there are fewer sensitive, nonlyso-genized cells in this population. Another tem-perature shock leads to repetition of this cycle ofinduction and recovery.More directly relevant to evolutionary consid-

erations are studies on the relative growth ratesof phage lysogens and isogenic nonlysogens inchemostats. Edlin et al. (42) and Lin et al. (80)have shown that lysogens (X cI857 ind- susJ)are favored at an approximate rate of s = 0.03per h in glucose-limited chemostats independentof the initial frequency of the lysogen. Thisphenomenon occurs with limitation by glucose,glycerol, lactose, or acetate, but it occurs only inaerobic chemostats; under anaerobic conditionsthe lysogen is disfavored. Such selective differ-ences are not observed in batch cultures, andthey have been attributed to a generally highermetabolic rate of lysogens under carbon sourcelimitation as judged by a tetrazolium assay ofNADH level in aerobically grown cells (80).

In any event, similar selective effects havebeen reported for Iysogens of bacteriophage PI(Plcm), P2 (P2-186p), and Mu (Mu cts) (41).Edlin (40) has studied certain outer membraneproteins of lysogens and nonlysogens by means

of sodium dodecyl sulfate-polyacrylamide gelelectrophoresis. He emphasizes two abundantproteins, the relative proportions of which arealtered in lysogens or in nonlysogens grownanaerobically. The gels show other differencesas well, and it is not necessarily the case that anyof the differences are associated with the selec-tion observed in the chemostat. Conceivably,there could be many differences between lyso-gens and nonlysogens that are real but notinvolved in the observed selection.

Lin et al. (80) have attributed the selectiveeffect of A to the A rex function because a

presumptive rex- lysogen was selectively neu-tral or nearly neutral. Since the mutation inquestion was obtained as a mutant lysogen sen-

sitive to phage T4 rll and not characterizedfurther, it could have been mutated in rexA orrexB (87). Indeed, the mutant lysogen couldhave involved phage genes in addition to rex or

might even have been a bacterial mutation over-


riding the T4 rIl exclusion function of rex+ andtherefore not rex- at all. The situation has beenclarified somewhat by Dykhuizen et al. (35). Inaddition to confirming the selective effect of Alysogens, they observed an initial slight decreasein the frequency of the lysogen before the favor-able selection commenced. However, studieswith rex5a, which is in the rexB cistron (87),revealed the same favorable selection as occurswith wild-type X. The rexB function does notseem to be involved in the selective effect,although there is a curious and unexplainedinteraction with malT such that malT X rex5alysogens are favored in chemostats whereasmalT X rex+ lysogens are disfavored. On theother hand, the rex am301 (= rexQ) mutationwas associated with a much reduced amount ofselection. The rex am301 mutation is in the rexAcistron near the carboxyl end of the codingsequence (87), and the small amount of selectionobserved could perhaps be attributed to residualactivity of this truncated polypeptide. In anyevent, if rex is involved in the X lysogen growthrate advantage, it would seem to be the rexAfunction. The effect of X, whatever its cause,seems to be quite general among lysogens but,unfortunately, its detailed mechanism is un-known.The same features that make chemostats suit-

able for studies of the evolution of lysogenymake them suitable for studying the evolution oftransposable elements. A novel in vivo effect ofthe transposable element TnS (kanamycin/neo-mycin resistance) has been described by Bieland Hartl (9, 10). Strains carrying TnS are fa-vored over their otherwise isogenic counterpartsin chemostats limited for glucose, lactose, glyc-erol, or proline at various densities and dilutionrates. The effect is independent offrequency andis substantial, ranging from 2 to 8% per h (D =0.35) in various experiments, but it is not ob-served in all E. coli strains. Derivatives ofCSH12 consistently fail to exhibit a detectableeffect ofTnS in chemostats. In other strains suchan effect is observed, and it is independent ofsite of insertion, having been observed withrandom insertions, with lac: :TnS, ilvD: :TnS, andeven with an F'lacP::TnS. The selection doesnot involve actual transposition of the element,as strains selected in chemostats retain TnS in itsoriginal position without change in copy num-ber. However, whereas the selection does notinvolve actual transposition, it may involve atransposase-related function because the selec-tion is not observed with TnS-112, a transposi-tionless deletion mutation missing the transpo-sase-coding region of the right-hand insertionsequence flanking the element. A similar growthrate advantage, less thoroughly studied, seemsto occur with TnJO (10). This TnJO effect has


http://mm

br.asm.org/

Dow

nloaded from



been confirmed and studied further by Chao etal. (L. Chao, C. Vargas, B. B. Spear, and E. C.Cox, submitted for publication), who provideevidence that the effect might be due to newfavorable mutations induced by transposition ofone or both flanking insertion sequences withouttransposition of the intact element. Altogether,these observations demonstrate a subtle physio-logical effect of transposons that may be signifi-cant in the fluctuating environmental conditionsthat bacteria experience in nature, and theysuggest a model for the proliferation and mainte-nance of insertion sequences and transposableelements in the absence of other identifiableselection pressures.The selective advantage associated with TnS

appears to be a temporary physiological effectoccurring immediately upon carbon or energylimitation or both. This is indicated in part by thecourse of selection. The TnS-bearing strain nev-er becomes completely fixed. Selection mayincrease its frequency from 50 to 95 or 99%, butthe competing strains are maintained at approxi-mately constant frequencies thereafter. Thisconstant frequency is maintained even after theculture has been transferred to a new chemostat,eliminating the possibility that this constancy isdue to wall growth (Dykhuizen, unpublisheddata). This effect can also be observed when theTnS-bearing strain and its isogenic counterpartare maintained in monoculture chemostats be-fore they are mixed in the experimental chemo-stat. With such preadaptation the selective ad-vantage of TnS declines and by 72 h ofpreadaptation has disappeared almost complete-ly (10). This effect of preadaptation is the re-verse of that reported for A lysogens. When a Alysogen and its isogenic nonlysogen are pre-adapted to limiting glucose, selection in a mixedchemostat in favor of the lysogen is even greaterthan that observed otherwise (42).

NATURALLY OCCURRING GENETICVARIANTS

A very recent use of chemostats is in the studyof functional effects of genetic variants that arefound in natural populations. Natural popula-tions of E. coli And probably many other micro-organisms contain an astonishing amount ofgenetic variation. Extensive genetic variationassociated with electrophoretically distinct en-zyme variants (allozymes) among natural iso-lates of E. coli has been described by Milkman(92, 93) and Selander and Levin (111). The use ofchemostats has permitted a study of the func-tional effects of these variants. Two extremeviews regarding the functional effects of allo-zyme-associated alleles have come to be knownas the selectionist hypothesis and the neutralisthypothesis. The selectionist view holds that

such genetic variants are adaptively significantand that they are maintained in natural popula-tions by virtue of their functional effects on theorganism. The neutralist view holds that suchalleles have so little functional effect that theyare not acted upon directly by natural selection,their frequencies being determined largely by abalance between mutation and random geneticdrift. In spite of their large actual populationsize, bacterial populations can be susceptible torandom drift because of frequent extinction andrecolonization of local populations (83). Ofcourse, both extreme views of genetic variationare to some extent caricatures, and many inter-mediate positions are possible.

In one series of experiments, Dykhuizen andHartl (38, 60) studied the functional effects of sixalleles of gnd (map position, 44 min) coding forelectrophoretic variants of 6-phosphogluconatedehydrogenase that had been found in wildisolates of E. coli. By a series of Pl-mediatedtransductions, these alleles were transferred intoan isogenic genetic background of E. coli K-12.Appropriately marked strains were then placedin pairwise competition in chemostats limited forgluconate, glucose, or a mixture of ribose andsuccinate. With gluconate limitation, a gnd mis-sense mutation is selected against at the rate s =0.09 (D = 0.35) per h; this experiment serves asa positive control. This same missense-bearingstrain shows no selective effect in chemostatslimited for a mixture of nbose and succinate;this medium establishes a negative control.Computer simulations ofthe sampling prociedureindicated that a rate of selection as small as s =0.005 per h could be detected in the proce-dures.The situation regarding selection ofgnd alleles

proved to be unexpectedly complex. Four of thealleles had no detectable selective effect (i.e., s< 0.005 h-1) in gluconate-limited chemostats;these are evidently neutral or nearly neutralunder these conditions. One allele (S1) wasfound to be detrimental (s = 0.03 h-1) in gluco-nate-limited chemostats but was neutral in glu-cose-limited chemostats. The remaining allele(S8) had an interaction with the tonA (T5r)neutral marker such that the tonA gnd+(S8)combination was selected against (s = 0.02 h-1).These results imply a great deal of functionaldiversity among allozyme-associated alleles.Many are neutral or nearly neutral under normalconditions but could have a potential for selec-tion that could be significant under the appropri-ate conditions of environment or genetic back-ground. Indeed, two alleles that otherwiseseemed neutral could be shown to be selectivelyimportant in a genetic background containing anedd mutation that cuts off the alternative meta-bolic route of gluconate metabolism, and the

MICROBIOL. REV.


http://mm

br.asm.org/

Dow

nloaded from



selection could be related to the Km of theallozymes (38).

Similar experiments with naturally occurringalleles of pgi (map position, 91 min), whichcodes for phosphoglucose isomerase, have alsobeen carried out (34; D. E. Dykhuizen and D. L.Hartl, submitted for publication). Five allo-zyme-associated alleles were studied in thiscase. In chemostats limited for glucose, selec-tive differences, if they occur at all, are belowthe limit of resolution of the technique. Four ofthese alleles are also neutral or nearly neutral inchemostats limited for fructose, which is also asubstrate for phosphoglucose isomerase. One ofthe alleles does exhibit a small but consistentselective effect (s = 0.002 h-i; D = 0.35) infructose-limited chemostats, illustrating again apotential for selection among naturally occurringallozymes that becomes important under theappropriate conditions.

ENGINEERED MUTATIONSRecombinant DNA techniques have made

possible the in vitro creation of mutations withdefined nucleotide sequence by means of site-directed mutagenesis. For the study of genestructure-function relationships, particularly asregards growth rate, the methods of site-directedmutagenesis represent a quantum advance over

more traditional techniques because the site-directed mutations can be produced and studiedindependently of any a priori assumptions con-

cerning their effects on phenotype. Traditionalmethods of mutagenesis require detectable phe-notypic effects for identification of the newmutants. These methods yield a biased sampleof all possible mutations that can occur becausethose having minimal phenotypic effects go un-

detected. This is the underlying reason why suchmethods typically yield mutations having largeeffects on, for example, enzyme activity. Stud-ies of naturally occurring genetic variants alsoinvolve a biased sample of mutations, but herethe bias is in the opposite direction. Naturallyoccurring genetic variants will include onlythose that have survived the screen of naturalselection. Near-neutral mutations will be amongthis group, as will those mutations that areselectively favored in particular environments or

genetic backgrounds. Although such studies are

important for an understanding of microbial pop-

ulation structure and evolution, they do notinvolve a random sample of mutations.

Site-directed mutagenesis can in principle beused to generate a random set of mutationsindependent of their phenotypic effects. Severalmethods are currently available for specific invitro mutagenesis and have already providedimportant information on which parts of genes

are essential for function. On the other hand, thefunctional assays currently in use are admittedlyrather crude, typically involving the ability ofappropriately transformed cells to give rise tocolonies on a plate or on their maximum growthrate. Chemostats would provide a useful tool forfiner-scale functional assays and could be usedto focus on the "fine tuning" of gene function orregulation. Chemostats are rapid, convenient,and reproducible and have a fine limit of resolu-tion of differences in growth rate. Many muta-tions that have growth rate effects that are easilydetectable in chemostats have no discernibleeffects on such gross characteristics as maximalspecific growth rate. When more than first-orderinformation about the efficacy of gene functionis desired, some continuous-culture techniquesuch as chemostats would seem to be the meth-od of choice.

CONCLUSIONS

Studies of selection in chemostats have re-vealed a number of important phenomena, and itmay be well to summarize them here. First, theselective events referred to as periodic selectionare not as haphazard as commonly assumed. Atleast in some cases (e.g., lac constitutives), themutants selected in replicate chemostats arehighly predictable and have a defined geneticand physiological basis. This opens the prospectof controlling or perhaps eliminating undesirablegenetic changes that commercially importantstrains sometimes undergo in prolonged continu-ous culture. Second, the widely invoked appealto energy economy does not account for allcases of selection in chemostats. Even the pro-totype case of selection in favor of amino acidauxotrophs is not proven critically and warrantsfurther investigation. Third, the selection ob-served in favor of lysogens or transposon-bear-ing strains is an important phenomenon insearch of a molecular mechanism, as it undoubt-edly plays a role in the natural history of theseelements. Fourth, the behavior of cloning vec-tors in chemostats is a subject of great interestbecause of the potential uses of continuousculture for the study of the functional effects ofdefined, engineered mutations and for potentialapplications in industrial microbiology.

ACKNOWLEDGMENT

This work was supported by Public Health Service grantGM30201 from the National Institutes of Health.

LITERATURE CITED1. Adams, J., and P. E. Hansche. 1974. Population studies

in microorganisms I. Evolution of diploidy in Saccharo-myces cerevisiae. Genetics 76:327-338.

2. Adams, J., T. Kinney, S. Thompson, L. Rubin, and R. B.Helling. 1979. Frequency-dependent selection for plas-mid-containing cells of Escherichia coli. Genetics

VOL. 47, 1983


http://mm

br.asm.org/

Dow

nloaded from



91:627-637.3. Mba, S., T. I_umala, ad H. Taekawa. 1980. Enhance-

ment of trypotophan production by Escherichia coli asan application of genetic engineering. Biotechnol. Lett.2:525-530.

4. Aner, T. F., and E. LUatbaer. 1975. Inheritability ofplasmids and population dynamics of cultured cells.Proc. Natl. Acad. Sci. U.S.A. 72:4085-4089.

5. Andre, K. J., and G. D. Heg_m . 1976. Selectivedisadvantage of noifunctional protein synthesis in Esch-erichia coli. J. Mol. Evol. 8:317-328.

6. Atwood, K. C., L. K. Schneide, and F. J. Ryan. 1951.Periodic selection in Escherichia coli. Proc. Nati. Acad.Sci. U.S.A. 37:146-155.

7. Atwood, K. C., L. K. Schdelder, and F. J. Ryan. 1951.Selective mechanisdss in bacteria. Cold Spring HarborSymp. Quant. Biol. 16:345-355.

8. Badn_, L J., and K. B. Low. 1980. Linkage map ofEscherichia coli K-12, edition 6. Microbiol. Rev. 44:1-56.

9. 1d, S. W., an D. L. Hrtl. 1981. Beneficial effects ofTn5 are independent of transposition. Genetics 97:s11.

10. Bid, S. W., and D. L. Harti. 1983. Evolution of transpo-sons: natural selection for Tn5 in Escherichia coli K12.Genetics 103:581-592.

11. Blom, F. R., and E. McFall. 1975. Isolation and charac-terization of -serine deaminase constitutive mutants byutilization of D-serine as sole carbon or nitrogen source.J. Bacteriol. 121:1078-1084.

12. Brown, S. W., and S. G. Olver. 1982. Isolation ofethanol-tolerant mutants of yeast by continuous selec-tion. Eur. J. Appi. Microbiol. Biotechnol. 16:119-122.

13. Bry.., V., and W. Szybalki. 1952. Microbial selection.Science 114:45-51.

14. Buonpy, H. R., L. S Cbeel, ad N. A. Andr_. 1981.Auto-selection of very rapidly growing microorganisms.Adv. Biotechnol. 3:235-241.

15. Button, D. K. 1978. On the theory of control of microbialgrowth kinetics by limiting nutrient concentrations. DeepSea Res. 25:1163-1177.

16. Calct, P. H. 1981. Genetic studies using continuousculture, p. 127-140. In P. H. Calcott (ed.), Continuousculture of cells. CRC Press, Boca Raton, nla.

17. Cmpbel, A. 1961. Conditions for the existence ofbacteriophage. Evolution 15:153-165.

18. Campbl, A. 1981. Evolutionary significance ofaccesso-ry DNA elements in bacteria. Annu. Rev. Microbiol.35:55-83.

19. Campbell, A., M. BDeedik, ad L. Hefenan. 1979.Viruses and inserfting elements in chromosomal evolu-tion, p. 51-59. In A. S. Dion (ed.), Concepts of thestructure and function of DNA, chromatin and chromo-somes. Year Book Medical Publishers, Chicago.

20. Chao, L., nd E.C. Cox. 1983. Competition betweenhigh and low mutating strains of Escherichia coli. Evolu-tion 37:125-134.

21. Chao, L., and B. R. Levin. 1981. Structured habitats andthe evolution of anticompetitor toxins in bacteria. Proc.Nad. Acad. Sci. U.S.A. 78:6324-6328.

22. Chlu, S. Y., L. E. Eickson, L. T. Fan, and 1. C. Kao.1972. Kinetic model identificaton in mixed populationsusing continuous culture data. Biotechnol. Bioeng.14:207-231.

23. Collns, S. H., A. W. Jarvis, R. J. Lindsy, ad W. A.Hamilton. 1976. Proton movements coupled to lactateand alanine transport in Escherichia coil: isoltion ofmutants with altered stoichiometry in alanine transport.J. Bacteriol. 126:1232-1244.

24. Contols, D. E. 1959. Kinetics of bacterial growth: rela-tionship between population density and specific growthrate of continuous cultures. J. Gen. Microbiol. 21:40-50.

25. Cooney, C. L., and D. I. C. Wag. 1976. Transientresponse of Enterobacter aerogenes under a dual nutri-ent limitation in a chemostat. Biotechnol. Bioeng. 16:189-198.

26. Cox, E. C., md T. C. GIbUs. 1974. Selection for highmutation rates in chemostats. Genetics 77:169-184.

27. Crow, J. F., and M. Kimura. 1970. An introduction topopulation genetics theory. Harper & Row, New York.

28. Dab., L. N., R. K. Finn, and C. R. WUie. 1973.Equations of substrate-limited growth: the case forBlackman kinetics. Biotechnol. Bioeng. 15:1159-1177.

29. Droop, M. R. 1968. Vitamin B12 and mane ecol. 4.The kinetics of uptake, growth and inhibition in Mono-chrysis lutheri. J. Mar. Biol. Assoc. U.K. 48b689-733.

30. Droop, M. R. 1973. Some thoughts on nutrient limitationin algae. J. Phycol. 9:264-272.

31. Droop, M. R. 1974. The nutrient status of algl cells incontinuous culture. J. Mar. Biol. Assoc. U.K. 54:82S-855.

32. Dwlvedl, C. P., T. I_aak, and S. Alba. 1982. Instabilityof plasmid-harboring strain of E. coli in continuousculture. Biotechnol. Bioeng. 24:1465-468.

33. D en, D. 1978. Selection for tryptophan auxotrophsof Escherichia coli in glucose-limited chemostats as atest of the energy conservation hypothesis of evoution.Evolution 32:125-150.

34. D , D. E. 1982. Chemostat studies of selectioninvolving electrophoretic variants of phosphoglucoseisomerase in Escherichia col. Genetics lgst2l.

35. Dykbulzen, D., J. H. Campbell, and B. G. Roffe. 1978.The influence of a A prophage on the growth rate ofEschenchia colt. Microbios 23:99-113.

36. Dykhulen, D., and M. Davies. 1980. An expermentalmodel: bacterial specialists and generalists competing inchemostats. Ecology 61:1213-1227.

37. Dkh n, D., and D. Hard. 1978. Transport by thelactose pennease of Escherichia coil as the basis oflactose killing. J. Bacteriol. 135:8764882.

38. Dykhnle, D., ad D. L. Had. 1980. Selective neutrali-ty of 6PGD allozymes in E. coli and the effects of geneticbackground. Genetics 96:801-817.

39. Dykhzen, D., and D. Hard. 1981. Evolution of competi-tive ability in Escherichia col. Evolution 35:581-594.

40. _n, G. 1978. Alteration of Escherichia coli outermembrae proteins by prophages. A model for benevo-lent virus-cell interaction, p. 1-14. In J. G. Stevens, G. J.Todaro, and C. F. Fox (ed.), Persistent viruses. Academ-ic Press, Inc., New York.

41. Eu., G., L. Li, and R. BIe. 1977. Reproductivefitness of P1, P2, and Mu lysogens. J. Virol. 21:560-564.

42. Ed, G., L. Li, and R. Kudne. 1975. A lysoens of E.coli reproduce more rapidly than non-lysogens. Nature(London) 255:735-737.

43. Frads, J. C., ad P. E. Hanche. 1972. Directed evolu-tion of metabolic pathways in microbial popubtions. I.Modification of the acid phosphatas pH optimum in S.cerevisiae. Genetics 70:59-73.

44. Frauns, J. C., and P. E. Hande. 1973. Directed evolu-tion of metabolic pathways in microbial p6pulatins. II.A repeatable adaptation in Saccharomyces cerevisiae.Genetics 74:259-265.

45. Fredr , A. G. 1977. Behavior of mixed cultures ofmicroorganisms. Annu. Rev. Mierobiol. 31:63-87.

46. Frodrldkion, A. G., R.D. MAloe 1ff,ad H. M. T yac .1970. Mathematical models for fermentation processes.Adv. Appl. Microbiol. 13:419-465.

47. Froer, R, H.Brck.er, M. Bewey, D. Cleve, and A.Aranki. 1983. Mechanisms that control bacterial popula-tions in continuous-fow culture models of mouse largeintestinal flora. Infect. Immun. 39676-6 .

48. Freter, R., H. DIlkner, J. F_ke, M. M. Vkerman, adK. E. Carey. 1983. Survival and implantation of Esche-rchia coli in the intestinal tract. Infect. Immun. 39.686-703.

49. Freter, R., R. R. Frder,_d H. Brkkner. 1983. Experi-mental and mathematical models of Escherichia colfplasmid transfer in vitro and in vivo. Infect. Immun.39:60-84.

50. Freter, R., E. Stauffer, D. Cleven, L. V.Hdem, and

MICROBIOL. REV.


http://mm

br.asm.org/

Dow

nloaded from


VOL. 47, 1983

W. E. C. Moore. 1983. Continuous-flow cultures as invitro models of the ecology of large intestinal flora.Infect. Immun. 39:666-675.

51. Gibson, T. C., M. L. Scheppe, and E. C. Cox. 1970.Fitness of an Escherichia coli mutator gene. Science169:686-688.

52. Godwin, D., and J. H. Slater. 1979. The influence of thegrowth environment on the stability of a drug resistantplasmid in Escherichia coli K12. J. Gen. Microbiol.111:201-210.

53. Goldman, J. C., and J. J. McCarthy. 1978. Steady stategrowth and ammonium uptake of a fast-growing marinediatom. Limnol. Oceanogr. 23:695-703.

54. Gottchal, J. C., and J. G. Kuenen. 1980. Mixotrophicgrowth of Thiobacillus A2 on acetate and thiosulfate asgrowth limiting substrates in the chemostat. Arch. Mi-crobiol. 126:33-42.

55. Hansche, P. E. 1975. Gene duplication as a mechanism ofgenetic adaptation in Saccharomyces cerevisiae. Genet-ics 79:661-674.

56. Hansche, P. E., V. Beres, and P. Lange. 1978. Geneduplication in Saccharomyces cerevisiae. Genetics88:673-687.

57. Harder, W., J. G. Kuenen, and A. Matin. 1977. Microbialselection in continuous culture. J. Appl. Bacteriol. 43:1-24.

58. Hartl, D. L. 1980. Principles of population genetics.Sinauer Associates, Sunderland, Mass.

59. Hartl, D. L., and D. E. Dykhuizen. 1979. A selectivelydriven molecular clock. Nature (London) 281:230-231.

60. Hartl, D. L., and D. E. Dykhuizen. 1981. Potential forselection among nearly neutral allozymes of 6-phospho-gluconate dedydrogenase in Escherichia coli. Proc. NatI.Acad. Sci. U.S.A. 78:6344-6348.

61. Hegeman, G. D. 1966. Synthesis of the enzymes of themandelate pathway by Pseudomonas putida. III. Isola-tion and properties of constitutive mutants. J. Bacteriol.91:1161-1166.

62. Helling, R. B., T. Kinney, and J. Adams. 1981. Themaintenance of plasmid-containing organisms in popula-tions ofEscherichia coli. J. Gen. Microbiol. 123:129-141.

63. Hellingwerf, K. J., J. S. Lolkema, R. Otto, 0. M. Neijssel,A. H. Stouthamer, W. Harder, K. van Dam, and H. V.Westerfoff. 1982. Energetics of microbial growth: ananalysis of the relationship between growth and itsmechanistic basis by mosaic nonequilibrium thermody-namics. FEMS Microbiol. Lett. 15:7-17.

64. Herbert, D., R. Elsworth, and R. C. Telling. 1956. Thecontinuous culture of bacteria; a theoretical and experi-mental study. J. Gen. Microbiol. 14:601-622.

65. Horiuchi, T., J.-I. Tomizawa, and A. Novick. 1962.Isolation and properties of bacteria capable of high ratesof,-galactosidase synthesis. Biochim. Biophys. Acta55:152-163.

66. Hsu, S. B., S. Hubbell, and P. Waltman. 1977. A mathe-matical theory for single nutrient competition in continu-ous cultures of micro-organisms. SIAM (Soc. Ind. Appl.Math.) J. AppI. Math. 32:366-383.

67. James, T. W. 1978. Selection and evolution of yeast cellsin a chemostat, p. 127-137. In E. R. Dirksen, D. M.Prescott, and C. F. Fox (ed.), Cell reproduction. Aca-demic Press, Inc., New York.

68. Jannasch, H. W., and R. I. Mateles. 1974. Experimentalbacterial ecology studied in continuous culture. Adv.Microb. Physiol. 11:165-212.

69. Jones, I. M., S. B. Primrose, A. Robinson, and C. C.Ellwood. 1980. Maintenance of some ColEl-type plas-mids in continuous culture. Mol. Gen. Genet. 180:579-584.

70. Koch, A. 1974. The pertinence of the periodic selectionphenomenon to prokaryote evolution. Genetics 77:127-142.

71. Kubitschek, H. E. 1970. Introduction to research withcontinuous cultures. Prentice-Hall, Inc., EnglewoodCliffs, N.J.


72. Kubitschek, H. E. 1974. Operation of selection pressureon microbial populations. Symp. Soc. Gen. Microbiol.24:105-130.

73. Lange, P., and P. E. Hansche. 1980. Mapping of acentromere-linked gene responsible for constitutive acidphosphatase synthesis in yeast. Mol. Gen. Genet.180:605-607.

74. Larson, D. H., and R. L. Dimmick. 1964. Attachment andgrowth of bacteria on surfaces of continuous-culturevessels. J. Bacteriol. 88:1380-1387.

75. Laverack, P. D., and P. H. Clarke. 1979. Selection ofmandelamidase constitutive mutants in continuous cul-ture of Pseudomonas putida. Biotechnol. Lett. 1:353-358.

76. Levin, B. R. 1980. Conditions for the existence of Rplasmids in bacterial populations, p. 197-202. In S.Mitsuhasi, L. Rosival, and V. Kreemery (ed.), Antibioticresistance: transposition and other mechanisms. Spring-er-Verlag, New York.

77. Levin, B. R. 1981. Periodic selection, infectious geneexchange and the genetic structure of E. coli popula-tions. Genetics 99:1-23.

78. Levin, B. R., and V. A. Rice. 1980. The kinetics oftransfer of nonconjugative plasmids by mobilizing conju-gative factors. Genet. Res. 35:241-259.

79. Levin, B. R., F. M. Stewart, and V. A. Rice. 1979. Thekinetics of conjugative plasmid transmission: fit of asimple mass action model. Plasmid 2:247-260.

80. Lin, L., R. Bitner, and G. EdlIn. 1977. Increased repro-ductive fitness of Escherichia coli lambda lysogens. J.Virol. 21:554-559.

81. Lwoff, A. 1944. L'Evolution physiologique, etude despertes de fonctions chez les microorganismes. Hermanand Co., Paris.

82. Martin, G. A., and W. P. Hemplling. 1976. A method forthe regulation of microbial population density duringcontinuous culture at high growth rates. Arch. Micro-biol. 107:41-47.

83. Maruyama, T., and M. Kimura. 1980. Genetic variabilityand effective population size when local extinction andrecolonization of subpopulations are frequent. Proc.NatI. Acad. Sci. U.S.A. 77:6710-6714.

84. Mason, T. G., and G. Richardson. 1981. A review:Escherichia coli and the human gut: some ecologicalconsiderations. J. Appl. Bacteriol. 1:1-16.

85. Mason, T. G., and G. Richardson. 1982. Observations onthe in vivo and in vitro competition between strains ofEscherichia coli isolated from the human gut. J. AppI.Bacteriol. 53:19-27.

86. Mason, T. G., and J. H. Slater. 1979. Competitionbetween an Escherichia coli tyrosine auxotroph and aprotrophic revertant in glucose- and tyrosine-limitedchemostats. Antonie van Leeuwenhoek J. Microbiol.Serol. 45:253-263.

87. Matz, K., M. Schmandt, and G. N. Gussin. 1982. The rexgene of bacteriophage A is really two genes. Genetics102:319-327.

88. Maule, A. P., and P. D. Thomas. 1973. Strains of yeastlethal to brewery yeasts. J. Inst. Brew. London 79:137-141.

89. McDonald, D. J. 1955. Segregation of the selectiveadvantage obtained through orthoselection in Escherich-ia coli. Genetics 40:937-950.

90. Meacock, P. A., and S. N. Cohen. 1980. Partitioning ofbacterial plasmids during cell division: a cis-acting locusthat accomplishes stable plasmid inheritance. Cell20:529-543.

91. Melling, J., D. C. Eliwood, and A. Robinson. 1977.Survival of R-factor carrying Escherichia coli in mixedcultures in the chemostat. FEMS Microbiol. Lett. 2:87-89.

92. Milkman, R. 1973. Electrophoretic variation in E. colifrom natural sources. Science 182:1024-1026.

93. Milkman, R. 1975. Allozyme variation in E. coli ofdiverse natural origins, p. 273-285. In C. L. Markert


http://mm

br.asm.org/

Dow

nloaded from



(ed.), Isozymes, vol. 4. Academic Press, Inc., NewYork.

94. Mood, J. 1942. Recherches sur la croissance des cul-tures bacteriennes. Herman and Co., Paris.

95. Mosd, J. 1950. La technique de culture continue.Th6orie et applications. Ann. Inst. Pasteur Paris 79:390-410.

96. Mos, H. 1957. Structure and dynamics of bacterialpopulations maintained in the chemostat. Cold SpringHarbor Symp. Quant. Biol. 22:121-137.

97. 'Mos, H. 1958. The dynamics of bacterial populationsmaintained in the chemostat. Carnegie Inst. WashingtonPubI. 614.

98. Monso, R. J., and B. A. Brdges. 1964. 'Take-over'-anunilsual selection process in steady-state cultures ofEscherichia coil. J. Gen. Microbiol. 37:411-418.

99. Nstmnm, E. R., and R. F. Hil. 1973. Populationchanges in continuously growing mutator cultures ofEscherichia coli. Genetics 73:41-44.

100. Noack, D., M. Roth, R. Geudhr, G. Miller, K. Undisz,C. H anr,d S. Gdpfr. 1982. Maintenance andgenetic stability of vector plasmids pBR322 and pBR32Sin Escherichia coil K12 strains grown in a chemostat.Mol. Gen. Genet. 134:121-124.

101. Novck, A. 1958. Some chemical bases for evolution ofmicro-organisms, p. 533-546. In A. A. Buzzati-Traverso(ed.), Perspectives in marine biology. University ofCalifornia Press, Berkeley.

102. Novick, A., and T. Holueli. 1961. Hyper-production of3-galactosidase by Escherichia coli bacteria. Cold SpringHarbor Symp. Quant. Biol. 26:239-245.

103. Novlck, A., asd L. Szaird. 1950. Description of thechemostat. Science 112:715-716.

104. Novlck, A., and L. Szlard. 1950. Experiments with thechemostat on spontaneous mutations of bacteria. Proc.Natl. Acad. Sci. U.S.A. 36:708-719.

105. PaIter, P. R. 1973. Mutator genes and selection for themutation rate in bacteria. Genetics 79:649-660.

106. Paquln, C., ad J. Ads. 1981. Specification of adap-tive changes in yeast popultions. Genetics 97:s83-84.

107. Paynter, M. J. B., and H. R. Bwuaay Im. 1970. Responsesin continous culturs of lysogeni Escherichia colifollowing induction. Biotechnol. Bioeng. 12:347-351.

108. PowelR, E. 0. 1958. Critera for the growth of contami-nants and mutaats in continuous culture. J. Gen. Micro-biol. 18:259-268.

109. Roth, M., G. Muller; and D. Noack. 1980. Loss of themulti-copy resistance plasmid pBR325 from Escherichiacoli GY2354 pBR32S during continuous cultivation, p.143-146. In S. Mitihasi, L. Rosival, and V. Kredmdry(ed.), Antibiotic resistance: transposition and othermechaaisms. Springer-Verlag, New York.

110. Rodh, M., and D. Noack. 1982. Genetic stability ofdifferentiated functions in Streptomyces hygroscopicusin relation to conditions of continuous culture. J. Gen.Microbiol. 128:107-114.

111. Selad_, R. K., and B. R. Le-An. 1980. Genetic diversityand structuit of Escherichia colt populations. Science

210:545-547.112. SHr, R. S., ad R. I. Matle. 1969. Control of mixed-

substrate utilization in continuous cultures of Escherich-ia coli. J. Bacteriol. 97:535-543.

113. Tempest, D. W. 1969. The continuous cultivation ofmicroorganisms. I. Theory of the chemostat, p. 260-276.In J. R. Norris and D. W. Ribbons (ed.), Methods inmicrobiology, vol. 2. Academic Press, London.

114. Tempest, D. W., and 0. M. NeUdel. 1978. Eco-physio-logical aspects of microbial growth in aerobic nutrient-limited environments. Adv. Microb. Ecol. 2:105-153.

115. Temped, D. W., and J. T. M. Woaers. 1981. Propertiesand performance of microorgnisms in chemostat cul-ture. Enzyme Microb. Technol. 3:283-290.

116. Veldamp, H. 1976. Continuous culture in microbialphysiology and ecology. Meadowfield Press, Durham,England.

117. Vedkamp, H. 1977. Ecological studies with the chemo-stat. Adv. Microb. Ecol. 1:59-94.

118. Wes, R. L., J. R. Kukora, and J. Ada. 1975. Therelationship between enzyme activity, ceN geometry,and fitness in Saccharomyces cerevislac. Proc. Natl.Acad. Sci. U.S.A. 72:794-798.

119. Wila, F. M. 1971. Dynamics of microbial popula-tions, p. 197-267. In B. Patten (ed.), Systems analysisand simulation in ecology. Academic Press, Inc., NewYork.

120. Wilams, F. M. 1973. Mathematics of microbial popula-tions, with emphasis on open systems. Conn. Acad. ArtsSci. Trans. 44:397-426.

121. Weods, D. D., ad M. A. Foster. 1964. Metabolic consid-erations relating to the life of bacteria in vivo, p. 30-43.In H. Smith and J. Taylor (ed.), Microbial behaviour, 'invivo' and 'in vitro.' Cambridge University Press, Cam-bridge.

122. Woters, J. T. M., F. L. Drieuls, P. J. Pacask,M.-L. H. A. v Oppe ay, ad J. G. van Ade. 1980.Persistence of the pBR322 plasmid in Escherichia coliK12 grown in chemostat cultures. Antonic van Leeu-wenhoek J. Microbiol. Serol. 46:353-362.

123. W ters, J. T. M., ad J. G. va Andel. 1979. R-plasmidpersistence in Escherichia coli grown in chomostat cul-tures. Antonie van Leeuwenhoek J. Microbiol. Serol.45:317-318.

124. Youg, T. W., and G. Pb.irk. 1977. The productionofa yeast killer factor in the chemostat and the effects ofkiller yeasts in mixed continuous culture with a sensitivestrain. J. AppI. Bacteriol. 43:425-436.

125. Za , S., and H. H. Eklbben. 1967. Study ofmicrobial evolution through loss of biosynthetic func-tions: establishment of "defective" mutants. Nature(London) 216.456-458.

126. Zevenboom, W. 1980. Growth and nutrient uptake kinet-ics of Oscillatoria agardhil. Krips Repro Meppel, Am-sterdam.

127. Zud, P., and G. Lebek. 1980. Generation time-prolong-ing R plasmids: correlation between increases in thegeneration time ofEscherichia coli caused by R plasmidsand their molecular size. Plasmid 3:65-69.

MICROBIOL. REV.


http://mm

br.asm.org/

Dow

nloaded from


Documents

AND L. HARTL · chemostatorturbidostat, in mostapplications is basically avesselforgrowinglarge, densebatch cultures. Here wefocus onexperimental workinvolv-ing chemostats because