Gene Replacement in Mycobacteria by Using Incompatible Plasmids

A simple and efficient delivery system was developed for makingtargeted gene knockouts in Mycobacterium smegmatis. This deliverysystem relies on the use of a pair of replicating plasmids,which are incompatible. Incompatible plasmids share elementsof the same replication machinery and so compete with each otherduring both replication and partitioning into daughter cells.Such plasmids can be maintained together in the presence ofantibiotics; however, removal of selection leads to the lossof one or both plasmids. For mutagenesis, two replicating plasmidsbased on pAL5000 are introduced; one of these plasmids carriesa mutated allele of the targeted gene. Homologous recombinationis allowed to take place, and either one or both of the vectorsare lost through the pressure of incompatibility, allowing thephenotypic effects of the mutant to be studied. Several differentplasmid combinations were tested to optimize loss in the absenceof antibiotic selection. pAL5000 carries two replication genes(repA and repB), which act in trans, and the use of vectorsthat each lack one rep gene and complement each other resultedin the loss of both plasmids in M. smegmatis and Mycobacteriumbovis BCG. The rate of loss was increased by the incorporationof an additional incompatibility region in one of the plasmids.To facilitate cloning when the system was used, we constructedplasmid vector pairs that allow simple addition of selectionand screening genes on flexible gene cassettes. Using this system,we demonstrated that M. smegmatis pyrF mutants could be isolatedat high frequency. This method should also be useful in otherspecies in which pAL5000 replicates, including Mycobacteriumtuberculosis.

INTRODUCTION

Top

Introduction

Production of mutations in mycobacteria is a fundamental approachfor discovering the function of genes in order to obtain knowledgeconcerning biology and pathogenicity. This knowledge may contributeto vaccine development by identifying virulence determinantsfor rational attenuation, and it may contribute to drug discoveryby identifying possible targets. Mutagenesis may be achievedin a random or targeted manner. Both allelic exchange and transposonmutagenesis are rare genetic events in mycobacteria, so an efficientdelivery system is required, in which the disrupted gene ortransposon is introduced on a vector, homologous recombinationor transposition takes place, and the vector is then lost. Thishas been accomplished in mycobacteria by the use of nonreplicating(suicide) plasmids (10, 14, 16), a temperature-sensitive plasmid(6), and a temperature-sensitive phage (1). In this study wetook a different approach and used plasmid incompatibility.This technique has the advantage of using replicating plasmids,which results in a prolonged time in the cell for homologousrecombination or transposition to occur but avoids the use ofa temperature-sensitive replicon which can be difficult to losein slowly growing species with narrow temperature ranges, suchas Mycobacterium tuberculosis.

Plasmid incompatibility is defined as the inability of a pairof coresident plasmids to be stably maintained in the absenceof external selection. This is due to sharing of one or morecomponents of plasmid partitioning or replication systems. Plasmidloss due to incompatibility is commonly due to interferencewith the ability of the plasmid to correct stochastic fluctuationsin its copy number (13). By imposing different kinds of selectionon pairs of incompatible plasmids, copy number can be manipulated.We have previously mapped the replication functions of pAL5000(23, 24) (Fig. 1). Two replication genes, repA and repB, encodinga putative primase and a DNA-binding protein, respectively,are essential for replication and can be supplied in trans.The incompatibility functions (inc) of pAL5000 were mapped toa cis-acting 120-bp fragment upstream of repA, which includesthe origin of replication (oriM). The oriM/inc region is capableof conferring incompatibility to otherwise unrelated replicons.

View larger version (11K):
[in this window]
[in a new window]
FIG. 1. Diagram of the pAL5000 replication genes. The repA and repB genes are both essential for replication. The origin and incompatibility functions map upstream of these genes, within the 1-kb SmaI fragment.

We show here that maintenance or loss of plasmids can be controlledin both Mycobacterium smegmatis mc²155 and Mycobacterium bovisBCG by using pairs of pAL5000-derived plasmids, and we developedan efficient delivery system for targeted gene replacement usingincompatibility as a pressure to eliminate plasmids.

MATERIALS AND METHODS

Top

Materials and Methods

Bacterial strains and plasmids.
Plasmids used in this work are listed in Table 1. M. smegmatisstrain mc²155 (22) was cultured in Lemco medium (10 g of peptoneper liter, 5 g of Lemco powder [Oxoid] per liter, 5 g of NaClper liter) containing 0.05% (wt/vol) Tween 80 or on Lemco agarplates (Lemco medium containing 15 g of agar per liter). M.bovis BCG strain NCTC 5692 was grown in Middlebrook 7H9 liquidmedium (Difco) supplemented with 10% (wt/vol) OADC (oleic acid,albumin, dextrose, catalase; Becton Dickinson) and 0.05% (wt/vol)Tween 80 or on Middlebrook 7H10 agar (Difco) supplemented with10% (wt/vol) OADC. Antibiotics were added at the following concentrationswhen appropriate: kanamycin, 20 µg ml^-1; hygromycin, 100µg ml^-1; and gentamicin, 10 µg ml^-1. When gentamicinand kanamycin were used in combination, the concentration ofkanamycin was increased to 50 µg ml^-1 as there is a degreeof cross-resistance from gentamicin to kanamycin.

View this table:
[in this window]
[in a new window]
TABLE 1. Plasmids used in this study

Electroporation.
Competent M. smegmatis and M. bovis BCG cells were preparedas described by Parish and Stoker (15). Transformations wereperformed by using 400-µl aliquots of cells with 1 to2 µg of DNA for M. smegmatis and 1 to 5 µg of DNAfor M. bovis BCG. Cotransformants (cells transformed with twodifferent plasmids) were selected on plates supplemented withantibiotics selecting for both plasmids.

Plasmid stability assay.
A single M. smegmatis or M. bovis BCG cotransformant colonywas used to inoculate 5 to 10 ml of medium (see above) withantibiotics selecting for both plasmids and incubated at 37°Cwith shaking until the A₆₀₀ was approximately 0.5 (3 to 4 daysfor M. smegmatis and 3 weeks for M. bovis BCG). One hundredmicroliters of each culture was plated onto a 190-mm nonselectiveagar plate (medium without antibiotics) and incubated at 37°Cto obtain a lawn of growth. Serial dilutions of the culturewere plated onto selective and nonselective plates to obtainsingle cells in order to determine the proportion of antibiotic-resistant(plasmid-carrying) cells. Cells were scraped from the lawn ona nonselective plate into 1 ml of medium. Dilutions were platedonto selective and nonselective plates. Each plating was referredto as a round. This procedure was repeated through rounds ofnonselective plating until no colonies could be counted on theantibiotic plates (i.e., when undiluted cells did not give riseto any colonies); no more than three rounds were necessary.Cells were always scraped from a nonselective plate having alawn of growth. The relative stabilities of the two plasmidswere monitored by calculating the reduction in plasmid numbersat each round from the colony counts and plotting the reductionsin plasmid number against the rounds of plating.

The loss factor (LF) (12) was calculated by using the equationLF = 1 - (V₁/V₂)^1/g, where V₂ is the total number of cells (cellsgrowing on nonselective medium) after g generations and V₁ isthe number of cells that retained the plasmid (cells growingon selective medium) after g generations. We estimated thateach round was about 23 generations (the time necessary forone cell to grow into a colony of 10⁷ cells). The average lossfactor was calculated from three separate experiments and wasexpressed as the mean ± standard deviation.

Isolation of mutants.
The method used to isolate mutants was essentially the sameas the method used to assess plasmid stability. However, followinggrowth of a cotransformant in antibiotic-containing medium,10⁹ cells were initially plated onto an agar plate supplementedwith gentamicin to obtain sufficient numbers of plasmid-carryingcolonies for mutant isolation. Cells were then scraped and subjectedto nonselective rounds of plating by plating dilutions ontoplates containing gentamicin and X-Gal (5-bromo-4-chloro-3-indolyl-ß-D-galactopyranoside)in order to isolate white gentamicin-resistant colonies (i.e.,colonies containing the mutant pyrF allele but lacking the plasmid-bornelacZ gene).

Analysis of M. smegmatis mutants.
White colonies were initially tested for uracil auxotrophy andresistance to 1 mg of 5'-fluoroorotic acid (5'-FOA) (Sigma)per ml on minimal medium (M9 minimal salts [Difco], 20% [wt/vol]glucose, 2 ml of glycerol per liter, 0.2 mM uracil). DNA wasprepared from 5-ml liquid cultures (2, 20) and used to confirmmutant generation by Southern blotting and PCR analysis.

To assess the frequency of mutant generation, serial dilutionswere plated four times, and three independent experiments wereperformed. Phenotypically, pyrF mutants should have producedwhite colonies, been uracil auxotrophs, been resistant to 5'-FOA,and grown on plates containing 10% (wt/vol) sucrose (sacB wascarried on pINC52).

RESULTS AND DISCUSSION

Top

Results and Discussion

Incompatibility between plasmids can provide the basis for a delivery system.
The essence of any delivery system is that a DNA molecule carryinga transposon or altered gene is introduced and then lost. Efficientloss is essential, as the events being sought are rare and itis necessary to select for them. Thus, if a researcher is lookingfor gene replacement with an allele carrying an antibiotic resistancemarker, the vector carrying this marker must be lost. This ismost straightforwardly done by using a nonreplicating vector.The use of replicating vectors should theoretically increasethe frequency of mutant isolation compared with that obtainedwith suicide vectors since (i) replicating vectors have an increasedtime for homologous recombination events to occur and (ii) DNAreplication and recombination occur simultaneously in the cell,but replicating vectors are then more difficult to remove. Conditionallyreplicating vectors have been used for gene replacement witha temperature-sensitive origin of replication (4, 17). Temperature-sensitivevectors are, however, weakly thermosensitive in slowly growingmembers of the M. tuberculosis complex because of their narrowtemperature range (17). These vectors therefore have to be combinedwith a counterselectable marker, such as sacB, to eliminateclones that still contain vector DNA. Growth at the permissivetemperature is very slow, while colonies grown at the nonpermissivetemperature also take longer to grow and have been reportedto be microscopic (unpublished results).

Having shown previously that incompatibility could in principlebe used (23), we carried out experiments using different plasmidpairs in order to optimize this system.

Model A: use of codependent plasmids.
We previously showed that when M. smegmatis is transformed withtwo pAL5000-based plasmids carrying different markers, selectionfor one plasmid causes the second plasmid to be lost due toincompatibility pressure (23). This is not an ideal deliverysystem, in part because it requires a functional plasmid tointegrate into the chromosome, which appears to be lethal forthe pAL5000 replicon in M. tuberculosis (7).

We therefore used the fact that repA and repB act in trans (23)to develop a system in which we used a pair of codependent plasmids,each containing the pAL5000 minimal origin of replication (SmaIfragment in Fig. 1) but lacking one replication gene. This minimalorigin contains regions involved in incompatibility and is referredto below as oriM/inc. As neither plasmid can replicate independentlyin mycobacteria, selection for one plasmid automatically selectsfor both, while growth in the absence of selection should resultin the loss of both plasmids and thus the plasmid carrying themutated target gene should be lost.

We constructed an Hyg^r plasmid carrying repA only (pNUT2) anda Kan^r plasmid that carried only repB (pNUT5). We called thispairing model A (Fig. 2A). The plasmids were cotransformed intoM. smegmatis mc²155 and selected on plates containing both hygromycinand kanamycin. Plasmid loss was determined during serial cultureon Lemco agar in the absence of antibiotics. Plasmids were lostat a steady rate of 10^-1 per round with a loss factor of 0.095± 0.016 (Fig. 2B). When model A cotransformants wereobtained, the colony size was also small but recovered afterrestreaking onto fresh plates, possibly because this alloweda pool of the Rep proteins to be established, although we cannotrule out the possibility that there were compensatory mutations.

View larger version (13K):
[in this window]
[in a new window]
FIG. 2. Loss kinetics with model A plasmids. (A) Plasmid constructs. kan, kanamycin resistance gene; hyg, hygromycin resistance gene; repA and repB, replication genes from pAL5000; open circle, E. coli ColE1 oriE; cross-hatched circle, pAL5000 oriM. (B) Plasmid loss kinetics in M. smegmatis. Symbols: ${blacksquare}$ , Hyg^r colonies (pNUT2); ${triangleup}$ , Kan^r colonies (pNUT5). The dotted line indicates that the limit of sensitivity was reached.

Effects of different antibiotic resistance genes.
A combination of antibiotic resistance genes cloned into thetwo incompatible plasmids was shown to be important in termsof loss rates. We have previously shown that plasmids carryingthe kan gene are less stable than plasmids carrying hyg (23).In order to extend these observations, we compared the stabilityof pAL5000-based plasmids carrying gm with the stability ofthe same plasmids that had gm replaced with kan or hyg at thesame site and in the same orientation; hence, the plasmids wereidentical except for the resistance gene. No difference in stabilitybetween plasmids carrying gm and plasmids carrying hyg was observed.However, plasmids carrying kan were less stable than both ofthese. Therefore, it may be beneficial to include kan on oneplasmid. A reason proposed for this is that the hyg gene fromStreptomyces hygroscopicus is GC rich and so may be more compatiblewith mycobacteria than the transposon-derived kan genes havinglower G+C contents (23).

Model B: addition of a second oriM/inc region.
In order to increase the rate of loss, we used the observationmade in previous work that a second copy of the replicationorigin/incompatibility region increased incompatibility if itwas cloned close to the other copy in the same orientation (23).We therefore adapted model A plasmids by adding a second oriM/incregion to pNUT5, producing pINC9. We called the pairing of pINC9(repB kan oriM/inc x2) with pNUT2 (repA hyg oriM/inc) modelB (Fig. 3A). Colonies were obtained at a frequency that wasabout 1% of the frequency obtained with a single replicatingplasmid. The colonies were very small in the initial transformationbut recovered after restreaking.

View larger version (12K):
[in this window]
[in a new window]
FIG. 3. Loss kinetics with model B plasmids. (A) Plasmid constructs. kan, kanamycin resistance gene; hyg, hygromycin resistance gene; repA and repB, replication genes from pAL5000; open circle, E. coli ColE1 oriE; cross-hatched circle, pAL5000 oriM. (B) Plasmid loss kinetics in M. smegmatis and in M. bovis BCG. Symbols: ${blacksquare}$ , Hyg^r M. smegmatis(pATB12); ${square}$ , Hyg^r M. bovis BCG. The dotted line indicates that the limit of sensitivity was reached.

Additionally, we found using other models a significant degreeof plasmid rearrangement in the ColE1 origin regions (data notshown). We therefore used different Escherichia coli origins(ColE1 and P15A) for the remaining experiments, and this removedthe problem.

Plasmid loss was measured in the absence of selection. Bothplasmids were simultaneously lost at a rate of approximately10^-3 per round of plating, corresponding to a loss factor of0.213 ± 0.068 (Fig. 3B). This rate was higher than therate seen with model A. Again, we looked for interplasmid recombinationbut did not observe any (data not shown).

The same experiment was repeated with M. bovis BCG (Fig. 3B).The plasmids were lost rapidly (approximately 10^-2 per roundof plating), although the loss factor (0.178 ± 0.066)was slightly lower than that seen in M. smegmatis (Fig. 3B).We concluded that we could introduce pairs of plasmids and maintainthem by antibiotic selection but we could cure the plasmidsefficiently by removing the selection. This provided the basisfor an efficient delivery system. The most promising systemwas model B, and we carried out mutagenesis experiments usingthis model.

Targeted mutagenesis of M. smegmatis pyrF by using an incompatibility-based delivery system.
Having shown that model B complementing plasmids were lost efficientlywhen selection was removed, we tested them as a delivery systemfor targeted mutagenesis. The M. smegmatis pyrF gene encodesorotidine monophosphate decarboxylase, which allows growth inmedia deficient in uracil but is lethal to cells grown in thepresence of 5'-FOA. Mutants lacking this gene have been generatedby other methods, making it a suitable test system.

A 3.2-kb SphI fragment containing the pyrF gene interruptedby gm was cloned into pNUT12, producing pYRANA3 (Fig. 4A). pYRANA3and pINC52 were cotransformed into M. smegmatis mc²155, withselection for gentamicin and kanamycin. The cells were grownin selective liquid media (containing gentamicin and kanamycin)before they were plated on agar containing gentamicin. Cellswere then scraped and plated onto antibiotic-free plates andtaken through rounds of plating. As we were looking for relativelyrare events, such as homologous recombination (11), the numberof cells used to seed each plate had to be high, and we used10⁹ cells. The rationale of the screening procedure was thatthe selection with gentamicin would select for cells carryingthe mutant pyrF gene, either on the plasmid or integrated intothe chromosome. To eliminate the former possibility, we platedthe cells in the presence of X-Gal, which distinguished betweencolonies carrying the pINC52 (lacZ⁺) plasmid (blue) and plasmid-freecolonies (white). Thus, cells were plated onto plates containinggentamicin and X-Gal at each round to identify mutants, andwhite colonies were picked as potential mutants for furtheranalysis. Note that as pINC52 carries the sacB gene (which providesan additional screen), plasmid-carrying cells are sucrose sensitive,while mutants lacking the plasmid are also sucrose resistant.

View larger version (14K):
[in this window]
[in a new window]
FIG. 4. Frequency of pyrF mutant generation in M. smegmatis. (A) Plasmid constructs. kan, kanamycin resistance gene; cm, chloramphenicol resistance gene; gm, gentamicin resistance gene; amp, ampicillin resistance gene; repA and repB, replication genes from pAL5000; open circle, E. coli ColE1 oriE; cross-hatched circle, E. coli P15A oriE; solid circle, pAL5000 oriM. (B) Plasmid loss and mutant isolation. Symbols: ${blacksquare}$ , total Gm^r colonies (plasmid loss); ${square}$ , white Gm^r colonies (pyrF mutants). The error bars indicate standard deviations from three separate experiments.

After the first round of plating, white colonies were observedat an average frequency of 2.6% ± 0.9%. During the secondround, the frequency of mutants increased to 22% ± 29%(one experiment yielded 64% mutants), but the frequency of mutantsfell during the third round to 5.5% ± 0.50% (Fig. 4B).All of the white colonies (n = 95) were tested phenotypically;90 of these had the expected phenotype of a pyrF mutant (uracilauxotrophs and those resistant to 5'-FOA and sucrose). Fivecolonies were not auxotrophic and still retained plasmids; restrictionanalysis showed that they had undergone rearrangements in thelacZ gene (data not shown). Thirty-one randomly selected auxotrophiccolonies were also tested genotypically by PCR and Southernblot analysis and were confirmed to be double-crossover mutants.

Fitness of pyrF mutants.
One surprise was that there was not a gradual accumulation ofmutants when organisms were plated on gentamicin. Gm^r bacteriawere seen either because they had both plasmids or because oneplasmid had integrated into the chromosome as a single or doublecrossover. As the former was constantly decreasing and the lattershould have been stable and increasing, the number of whitecolonies should have increased with every round. In fact, anincrease was seen initially, but then there was a decrease inthe subsequent round; this pattern was observed in nine separateexperiments. The fact that there was not an accumulation waslikely because of the greater fitness of the wild-type cellsduring competitive growth (21, 25). We confirmed this by growingmutant and wild-type cells in a mixed culture (data not shown).This procedure could be manipulated by making plasmid carriagemore stressful. For example, we could use kan as the selectablemarker in the target gene as it increased loss; if it was placedon the other plasmid, the competitive advantage of the wild-typecells might have been decreased. However, it is likely thatmost mutations result in some disadvantage, and this suggeststhat taking the bacteria through more than two rounds of nonselectivegrowth is unlikely to provide great benefit.

Improved flexibility of plasmid pairs.
Having shown that the plasmids could be used to efficientlyisolate mutants, we addressed the lack of cloning sites in thevectors, which severely restricted their utility. Derivativesof pNUT12 and pINC52 were constructed which had an increasednumber of convenient restriction sites (Fig. 5). The pINC seriesof plasmids all contained repB, two oriM/inc regions that wereadjacent and in the same orientation, and an amp gene, and oriEwas derived from ColE1. The complementing pNUT series containedrepA, a single oriM/inc region, and a cm gene, and oriE wasderived from P15A. The two-plasmid series contained differentunique restriction enzymes for cloning (Fig. 5). As two plasmidsare needed for this delivery system, the mutated gene can becloned into either plasmid depending on which plasmid carriesthe most convenient sites for the target gene.

View larger version (24K):
[in this window]
[in a new window]
FIG. 5. Flexible vectors for cloning genes for mutagenesis. The target gene is cloned into one of these plasmids (either the pNUT or pINC series depending on which series has the most convenient restriction sites for cloning of the target gene) and mutated (or a mutated gene can be cloned in directly). If the gene is cloned in pNUT20 or pNUT21, the preparation should be cotransformed with a pINC-based plasmid (i.e., pINC12, which contains lacZ for screening and repB); if it is cloned into pINC11 or pINC13, the preparation should be cotransformed with a pNUT-based plasmid (i.e., pNUT22, which contains lacZ for screening and repA). A cotransformant can be grown in selective media and then plated onto nonselective media (to allow plasmid loss by incompatibility). Mutants can be identified by plating preparations onto plates containing X-Gal plus the relevant antibiotic (mutants produce white colonies). cm, chloramphenicol resistance gene; amp, ampicillin resistance gene; repA and repB, replication genes from pAL5000; open circle, E. coli ColE1 oriE; cross-hatched circle, E. coli P15A oriE; solid circle, pAL5000 oriM.

In addition, a PacI site was incorporated into both plasmidsso that PacI marker cassettes in the pGOAL series of plasmids(16) could be easily cloned onto either vector. For example,the pGOAL19 cassette contains hyg, P_Ag85lacZ, and P_hsp60sacB.We also constructed plasmids that contained the markers kanand P_Ag85lacZ/kan for both plasmids of the incompatible pairsince these cassettes are not available as PacI fragments inthe pGOAL series.

In order to confirm that these plasmids were suitable deliveryplasmids, combinations were tested by using pyrF as the targetgene as described above. pyrF::gm was cloned into pINC11, andthis plasmid was cotransformed with pNUT22 (Table 1). pyrF mutantswere again isolated after one round of plating on nonselectiveagar, and the frequency was 4 to 16% of the colonies on a plate.

Conclusions.
In this work a two-plasmid incompatibility system was developedas a delivery system for mycobacteria. Initially, the loss kineticsof different pairs of incompatible plasmids that had distinctproperties, including deletions in the rep genes, differentselectable markers, and additional copies of the oriM/inc region,were monitored. Both models examined showed that incompatibilitycould be used to efficiently eliminate plasmids from cells undernonselective conditions and therefore had the potential foruse in a delivery system.

Model B was tested by using the pyrF gene of M. smegmatis, forwhich a high mutation frequency (10^-1 to 10^-2) was obtained.This system is not only efficient and simple but rapid, sincemutants were isolated during the first round of plating. Thus,M. smegmatis mutants could be obtained in as little as 9 daysby using model B once plasmids had been constructed. Other groupshave used pyrF in gene replacement experiments with nonreplicatingplasmids and have obtained uracil auxotrophs at efficienciesof 40% (18), 1 to 10% (19), and 5% (8). Our results are comparableto these results; our average frequency was 10%, and the frequencywas 64% on one occasion.

Although only used here for allelic replacement, this deliverysystem should also be suitable for transposon mutagenesis. Plasmidincompatibility has been used to deliver a transposon on a replicatingvector in Erwinia herbicola. The incompatibility pressure converteda stable transposon-carrying vector into a suicide vector afterintroduction of an incompatible plasmid (26). In Pasteurellahaemolytica, gene replacement was forced through extended growthof cells when a pair of incompatible plasmids was made to coexist(3). A negative selection strategy had to be used in the laststage to exclude plasmids, as transformants mainly containedtwo plasmids. Strong incompatibility between plasmids was alsoused to deliver a stable replicative plasmid into the chromosomeof Corynebacterium glutamicum via homologous recombination,although the plasmid remained in the chromosome (9).

The approach taken in this work not only should contribute tothe tools used for mycobacterial genetics since it providesa simple and efficient delivery system but also should be usefulfor other organisms for which plasmids have been characterizedand transformation techniques have been described. As more informationconcerning other mycobacterial plasmids becomes available, thisprinciple could be used in other mycobacterial species in whichpAL5000 is unable to replicate, such as the members of the Mycobacteriumavium-Mycobacterium intracellulare-Mycobacterium scrofulaceumcomplex.

ACKNOWLEDGMENTS

C.A.P. and T.P. were supported by the GlaxoSmithKline ActionTB initiative.

We thank Pelle Stolt for donation of plasmid vectors.

FOOTNOTES

* Corresponding author. Present address: Department of Pathology and Infectious Diseases, Royal Veterinary College, Royal College St., London NW1 0TU, United Kingdom. Phone: 44 (0) 20 7468 5272. Fax: 44 (0) 20 7468 5306. E-mail: nstoker{at}rvc.ac.uk.

Present address: Department of Medical Microbiology, Barts andthe London, Queen Mary's School of Medicine and Dentistry, LondonE1 2AA, United Kingdom.

REFERENCES

Top