letters to nature

22. Wassermann, E. M. Risk and safety of repetitive transcranial magnetic stimulation: Report and improvement (CSI; sequential random mutagenesis and screening),
suggested guidelines from the international workshop on the safety of repetitive transcranial magnetic
stimulation. June 5±7, 1996. Electroencephalogr. Clin. Neurophysiol. 108, 1±16 (1998).
which is the leading method for the development of commercial
23. Pascual-Leone, A., Wassermann, E. M., Grafman, J. & Hallett, M. The role of the dorsolateral microorganisms4±6. During each cycle of CSI a population of
prefrontal cortex in implicit procedural learning. Exp. Brain Res. 107, 479±485 (1996). improved mutants is identi®ed from which the single best perform-
er is taken forwards. As the population of improved mutants (as
Acknowledgements opposed to the single best) from each a cycle comprises a pool of
We thank our subjects for their cooperation, J.-P. Ndayisaba for technical assistance, and useful genetic diversity, the shuf¯ing of this population should
D. G. Schoenberg for editing. W.M. was supported by the Max Kade Foundation. produce a new combination of mutants of superior performance,
just as it does for subgenomic fragments7,8 (Fig. 1). Two primary
Competing interests statement
questions were addressed by this study: can a population of bacterial
The authors declare that they have no competing ®nancial interests.
cells be ef®ciently shuf¯ed such that signi®cant progeny contain
Correspondence and requests for materials should be addressed to M.H. genetic information from more than two parents; and does such a
(e-mail: hallettm@ninds.nih.gov). population contain individuals having marked improvements in a
desired phenotype? We chose to shuf¯e Streptomyces, as they are the
primary natural producers of commercial antibiotics, have a long
history of CSI, and useful recombination formats are described9,10.
................................................................. Although there are various means of affecting recombination
between bacterial cells, protoplast fusion9,10 is very ef®cient between
Genome shuf¯ing leads to rapid Streptomyces, resulting in recombination ef®ciencies greater than
0.20 (ref. 11). The ef®ciency of recombination between fused
phenotypic improvement in bacteria protoplasts of four multiply auxotrophic strains of Streptomyces
coelicolor have been measured previously12. The simple two-marker
Ying-Xin Zhang*, Kim Perry*, Victor A. Vinci², Keith Powell*, progeny made up 3% of the population, while the complex three-
Willem P. C. Stemmer* & Stephen B. del CardayreÂ* and four-marker progeny were present at only 0.04% and 0.00005%
of the population, respectively. As ef®cient production of complex
* Maxygen, 515 Galveston Drive, Redwood City, California 94063, USA progeny increases both the sequence and functional diversity of a
² Eli Lilly and Company, Lilly Corporate Center, Indianapolis, Indiana 46285, population, improving the representation of complex progeny
USA within the recombined population was necessary.
.............................................................................................................................................. We reasoned that improvement in the distribution of complex
For millennia, selective breeding, on the basis of biparental progeny could be achieved by the recursive fusion of a mixed
mating, has led to the successful improvement of plants and protoplast population (see Supplementary Information). Each
animals to meet societal needs1. At a molecular level, DNA pooled fusion would mimic each thermal cycle of a DNA shuf¯ing
shuf¯ing mimics, yet accelerates, evolutionary processes, and reaction, and recursive fusions should result in the ef®cient shuf-
allows the breeding and improvement of individual genes and ¯ing of the population. To test this idea, we shuf¯ed the same set of
subgenomic DNA fragments. We describe here whole-genome S. coelicolor strains by recursive protoplast fusion (see Methods).
shuf¯ing; a process that combines the advantage of multi-parental Protoplasts from the four strains were mixed, fused, and regener-
crossing allowed by DNA shuf¯ing with the recombination of ated under non-selective conditions. Spores from the regenerated
entire genomes normally associated with conventional breeding. protoplasts were pooled, resulting in the ®rst fusion library (F1). F1
We show that recursive genomic recombination within a popula-
tion of bacteria can ef®ciently generate combinatorial libraries of
new strains. When applied to a population of phenotypically Asexual Sexual
selected bacteria, many of these new strains show marked recursive mutagenesis recursive recombination
improvements in the selected phenotype. We demonstrate the
use of this approach through the rapid improvement of tylosin
production from Streptomyces fradiae. This approach has the
potential to facilitate cell and metabolic engineering and provide
Fitness

a non-recombinant alternative to the rapid production of Shuffle

improved organisms. Mutate Mutate
Mutate
Evolution is a continuous process of genetic variation and Select one,
phenotypic selection2. Recombination within a selected population discard
others
ampli®es the genetic diversity of the population by creating new
mutant combinations, and can thereby improve the performance of
individuals within the population. We are interested in the applica-
tion of recombination formats for the rapid improvement of Parent
Select and shuffle population
biological systems, and here describe its application to the improve- of improved progeny
ment of whole-microbial genomes. Although classical breeding Progeny
addresses entire genomes, it allows for recombination between
Cycles (evolutionary time)
only two parents per generation. In contrast, DNA shuf¯ing
addresses DNA fragments and allows for recombination between Figure 1 Asexual versus sexual evolution. Asexual evolution is the sequential process of
multiple parents at each generation3. The production of the result- accumulating individual mutations. Selection of the ®ttest results in the capture of only a
ing multi-parent `complex progeny' is important for the marked single mutant. It is slow, as individuals within a population evolve alone as opposed to
acceleration of directed evolution realized through DNA shuf¯ing. sharing information and evolving as a group. Genetic diversity is lost and deleterious
A practical combination of classical breeding and DNA shuf¯ing mutations that are dif®cult to lose accumulate. Sex allows the information within a
should thus provide a means rapidly to breed populations of population to be shared. Mating within a selected population consolidates genetic
organisms to produce combinatorial libraries of complex progeny. information by providing a mechanism for the combination of useful mutations and the
The directed evolution of microorganisms has been accom- loss of deleterious mutations. Sexual evolution thus produces populations containing
plished traditionally though the asexual process of classical strain individuals that have a far greater ®tness than their parents.

644 © 2002 Macmillan Magazines Ltd NATURE | VOL 415 | 7 FEBRUARY 2002 | www.nature.com
 letters to nature
was grown as a population and used to prepare protoplasts, which Lilly (Fig. 2a). We aimed to use genome shuf¯ing to facilitate the
were similarly fused and regenerated. This process was repeated four creation of new, high tylosin-producing strains from SF1. To
times resulting in the F4 population. The distributions of two-, generate a diverse population for genome shuf¯ing, SF1 was
three- and four-marker progeny from F1 and F4 were determined subjected to one round of CSI using nitrosoguanidine (NTG) as
and are shown in Table 1. Consistent with ref. 12, a single fusion mutagen14. A total of 22,000 individual mutants were grown in a 96-
resulted in the ef®cient production of two-marker progeny (10%) well format and screened for tylosin production (see Methods), and
and the poor production of three-marker (0.4%) and four-marker 11 strains were selected as a population for further evolution based
(0.00007%) progeny. However, recursive fusion ef®ciently pro- on improved tylosin titres relative to SFI (Fig. 2b). Protoplasts from
duced the two- (60%), three- (17%) and four-marker (2.5%) each of the 11 strains were prepared, mixed in equal proportions,
progeny, a 40±105-fold increase in complex progeny (the distribu- and recursively fused. We screened 1,000 progeny from the ®rst
tion of phenotypes within each population is shown in Table 2 of cycle of shuf¯ing for further tylosin titre improvements. Seven
Supplementary Information). Thus recursive protoplast fusion was strains were identi®ed that produced more tylosin than the NTG-
successful in the shuf¯ing of the four S. coelicolor strains and could mutated parents, and some produced tylosin at levels comparable to
be used for the shuf¯ing of a phenotypically selected population for SF21 in our high-throughput screen (Fig. 2b). These 7 strains were
the purposes of directed evolution. subjected to an additional cycle of shuf¯ing, and 1,000 progeny were
Streptomyces fradiae is used for the commercial production of screened. Another 7 strains with further-improved tylosin titres
tylosin, a complex polyketide antibiotic13. Strain SF1 is a stable clone were identi®ed, each producing more tylosin than SF21 under high-
of the S. fradiae natural isolate, whereas SF21 produces tylosin at a throughput conditions (Fig. 2b). The performance of the best two
high titre and is derived from SF1 through 20 cycles of CSI at Eli shuf¯ed strains, GS1 and GS2, was compared to SF1 and SF21 in
250-ml shake ¯asks. GS1 and GS2 produce up to ninefold more
tylosin than SF1 and are statistically indistinguishable in tylosin titre
a
SF21 GS1, GS2
for SF21 (Fig. 2c).
NTG Cellular phenotypes are by nature complex, and result from the
NTG
c dynamic interaction of genes and gene products distributed
NTG
NTG GS Strain Titre (rel. g l –1) throughout a cell's genome. The mutations leading from SF1 to
NTG
NTG
SF21, GS1 and GS2 probably lie throughout the S. fradiae genome
NTG SF1 1.0 ± 0.1 and are not simply associated with the tylosin biosynthetic cluster.
NTG
NTG Classical Genome SF21 6.2 ± 2.4
In an attempt to differentiate at the molecular level strains SF1,
NTG
UV
strain shuffling GS SF21, GS1 and GS2, a regulatory gene within the tylosin biosyn-
UV
improvement
GS1 8.1 ± 1.2 thetic pathway believed to be involved in increased tylosin titre15 was
UV
(20 years (1 year ampli®ed by the polymerase chain reaction and sequenced. SF21,
NS GS2 6.2 ± 1.2
HNO2
HNO2
106 assays) 24,000 assays) GS1 and GS2 all had a single-nucleotide change in this gene relative
HNO2 NTG to SF1; however, the mutation in GS1 and GS2 was at a different
UV
HNO2 position in the gene than that in SF21. Thus, SF21, GS1 and GS2
HNO2 had converged on at least one functionally similar mode of
SF1 achieving improved titre, but had done so in a structurally distinct
manner. At the physiological level, GS1 and GS2 are clearly distinct
from SF21, as can be seen by the strain's differing morphologies.
Whereas SF21 forms a small rough colony, GS1 and GS2 form
b Screened: 22,000 1,000 1,000 round, smooth, raised colonies that resemble SF1. This difference
4 may re¯ect the large mutational load SF21 carries relative to the
Tylosin titre (relative; g l –1)

shuf¯ed strains.
In summary, two rounds of genome shuf¯ing were suf®cient to
3
achieve results that had previously required 20 rounds of CSI.
Theoretically, this is the difference between the asexual process of
2 CSI and the hypersexual process of genome shuf¯ing. Practically,
this is the difference between 24,000 assays and approximately 1 year
1 of effort, and roughly 1,000,000 assays and 20 years of effort (Vinci,
V. A., personal communication). Recombination has been discussed
previously for the purposes of strain improvement16±20. In contrast,
0
we demonstrate that recursive recombination within selected popu-
SF1
SF21

NTG Genome Genome

shuffling 1 shuffling 2 lations markedly increases the rate of strain evolution. By expanding
the reach of shuf¯ing technology from subgenomic fragments to
Figure 2 Genome shuf¯ing versus classical strain improvement. a, Comparison of whole cells and organisms, genome shuf¯ing provides a new tool for
classical strain improvement to genome shuf¯ing for production of tylosin from S. fradiae. cell and metabolic engineering that requires no sequence informa-
SF21 was generated through 20 rounds of mutagenesis and screening from SF1. GS1 tion or sophisticated genetic tools.
and GS2 were generated from two rounds of genome shuf¯ing of a population of
classically improved strains derived from SF1. The classical approach required twenty
Table 1 Distribution of phenotypes in a four-strain cross of S. coelicolor
years and approximately a million screens, whereas genome shuf¯ing required only a year
and 24,000 screens. HNO2, nitric acid; UV, ultraviolet irradiation; NTG, nitrosoguanidine; Phenotype Single fusion* Recursive fusion²
.............................................................................................................................................................................
NS, natural selection; GS, genome shuf¯ing. b, Relative production of tylosin in 300-ml, Two markers 8.4% 60%
96-well fermentations of SF1, SF21 (dotted line marks the production level of SF21); Three markers 0.73% 17%
Four markers 0.000045% 2.5%
NTG, 11 NTG-improved mutants derived from 22,000 screened; genome shuf¯ing 1, 7 .............................................................................................................................................................................
The distribution of phenotypes from each fusion is reported in Supplementary Information.
strains identi®ed from the shuf¯ing of the NTG population, 1,000 were screened; and * Phenotypes were determined from colony counts on de®ned medium containing 16 combinations
genome shuf¯ing 2, the 7 strains identi®ed from the shuf¯ing of genome shuf¯ing 1, of the four supplements (See Methods). Each phenotype is corrected for dilution and the presence
of prototrophic markers, and divided by the total colonies growing on completely supplemented
1,000 were screened. c, Relative production of tylosin from SF1, SF21, GS1 and GS2 medium. The value shown represents the sum of the frequencies from each phenotypic class.
from 250-ml shake ¯ask fermentations. ² The distribution of 483 individual colonies characterized for marker phenotype.

NATURE | VOL 415 | 7 FEBRUARY 2002 | www.nature.com © 2002 Macmillan Magazines Ltd 645
letters to nature
The application of natural evolutionary principles is extremely
powerful in the manipulation of complex biological systems from
single genes and proteins to whole organisms. The ability to rapidly
evolve microbial cells provides a means to accelerate the commer-
cialization of a variety of microbial products from pharmaceuticals
to commodity chemicals, as well as create new cell lines useful for
fundamental research. Although we describe recursive protoplast
fusion as a useful method for the genome shuf¯ing of Streptomyces,
other mechanisms of genetic exchange are also useful for the
genome shuf¯ing of bacteria, and lower and higher eukaryotes21.
As most of these formats rely on natural homologous recombina-
tion, organisms modi®ed by these means are not considered
`genetically modi®ed'. Genome shuf¯ing thus represents a practical
method for the rapid manipulation of the complex phenotypes of
whole cells and organisms. M 21. Del Cardayre, S. B. et al. Evolution of whole cells and organisms by recursive sequence recombination.
Methods
Recursive protoplast fusion
Protoplasts from S. coelicolor M124 (proA1 argA1 cysD18), 2684 (proA1 argA1 uraA1),
2685 (proA1 cysD18 uraA1) and 2686 (argA1 cysD18 uraA1) were prepared, fused and
regenerated as described previously22. Spores from the regenerated cells were pooled and
used to inoculate a culture for the production of protoplasts. Protoplasts from this mixed
culture were fused and regenerated. This process was repeated two additional times. In all
cases rich medium supplemented with cystine, proline, arginine and uracil to minimize
differential growth of individual auxotrophs was used for growth and regeneration.
Fusions typically resulted in less than 1% background and greater than 10,000 regenerated
colonies. The shuf¯ing of S. fradiae was achieved in the same manner, except protoplast
formation, fusion and regeneration were performed according to ref. 9. We carried out
prototroph and tylosin analysis on random samples of pooled progeny22 (spore pre-
parations for S. coelicolor or sonicated mycelia for S. fradiae).
HTP fermentation and tylosin assay
Clonal mycelial fragments were plated on R2YE23 medium and grown at 30 8C for ®ve days.
Colonies were picked into 300 ml of vegetative medium (Eli Lilly) in 96-well microtitre
plates. These were shaken at 30 8C and 85% relative humidity for 4 days, and 20-ml samples
were transferred to 300 ml of 50% fermentation medium (Eli Lilly) in 96-deep-well plates.
The plates were similarly incubated for 7 days, at which time 0.9 ml of methanol was added
to each well, and the plates were centrifuged. A 20-ml sample of each supernatant was
sampled and diluted into 0.1 M phosphate buffer, pH 7.0, and the tylosin content was
measured spectrophotometrically as described previously24. Tylosin A titre from the top
5±10% tylosin producers from each plate was more accurately determined by reverse-
phase high-performance liquid chematography25. Those having improved performance
were further con®rmed in quadruplicate by a second fermentation starting from the
original vegetative culture plate.
Shake ¯ask analysis
Three 250-ml ¯asks containing 50 ml of vegetative medium were inoculated from
frozen stocks. These were incubated as described above for three days, and 5-ml
samples from each were used to inoculate three separate ¯asks of 100% fermentation
medium. The 9 ¯asks from each strain were then incubated for 8 days and the average
tylosin A content of the methanolic supernatant from each culture was determined as
described above.
Received 12 September; accepted 11 December 2001.
1. Burbank, L., Whitson, J., John, R., Williams, H. S. & Society, L. B. Luther Burbank, his Methods and
Discoveries and their Practical Application (Luther Burbank, New York, 1914).
2. Darwin, C. On the Origin of Species by Natural Selection, or the Preservation of Favored Races in the
Struggle for Life (John Murray, London, 1859).
3. Ness, J. E. et al. DNA shuf¯ing of subgenomic sequences of subtilisin. Nature Biotechnol. 17, 893±896
(1999).
4. Aharonowitz, Y. & Cohen, G. The Microbiological production of pharmaceuticals. Sci. Am. 245, 141±
152 (1981).
5. Demain, A. L. & Solomon, N. A. (eds) Manual of Industrial Microbiology and Biotechnology (ASM,
Washington, 1986).
6. Vinci, V. A. & Byng, G. in Manual of Industrial Microbiology and Biotechnology (eds Demain, A. L. &
Davie, J. E.) 103±113 (ASM, Washington, 1999).
7. Stemmer, W. P. DNA shuf¯ing by random fragmentation and reassembly: in vitro recombination for
molecular evolution. Proc. Natl Acad. Sci. USA 91, 10747±10751 (1994).
8. Christians, F. C., Scapozza, L., Crameri, A., Folkers, G. & Stemmer, W. P. Directed evolution of
thymidine kinase for AZT phosphorylation using DNA family shuf¯ing. Nature Biotechnol. 17, 259±
264 (1999).
9. Baltz, R. H. Genetic recombination in Streptomyces fradiae by protoplast fusion and cell regeneration.
J. Gen. Microbiol. 107, 93±102 (1978).
10. Hopwood, D. A., Wright, H. M., Bibb, M. J. & Cohen, S. N. Genetic recombination through protoplast
fusion in Streptomyces. Nature 268, 171±174 (1977).
11. Hopwood, D. A. & Wright, H. M. Factors affecting recombinant frequency in protoplast fusions of
Streptomyces coelicolor. J. Gen. Microbiol. 111, 137±143 (1979).
646 © 2002 Macmillan Magazines Ltd
12. Hopwood, D. A. & Wright, H. M. Bacterial protoplast fusion: recombination in fused protoplasts of
Streptomyces coelicolor. Mol. Gen. Genet. 162, 307±317 (1978).
13. Baltz, R. H. & Seno, E. T. Genetics of Streptomyces fradiae and tylosin biosynthesis. Annu. Rev.
Microbiol. 42, 547±574 (1988).
14. Baltz, R. H. in Manual of Industrial Microbiology and Biotechnology (Am. Soc. Microbiol., Washington,
1986).
15. Stratigopoulos, G. & Cundliffe, E. Expression analysis of the tylosin-biosynthetic gene cluster: pivotal
regulatory role of the tylQ product. Chem. Biol. 1, (in the press).
16. Baltz, R. H. in Biotechnology of Antibiotics (ed. Strohl, W. R.) 49±62 (Marcel Dekker, New York,
1997).
17. Hamlyn, P. F. & Ball, C. in Genetics of Industrial Microorganisms 185±191 (American Society for
Microbiology, Washington, 1979).
18. Hopwood, D. A. in Genetics of Industrial Microorganisms 1±9 (American Society for Microbiology,
Washington, 1979).
19. Hopwood, D. A. & Chater, K. F. Fresh approaches to antibiotic production. Phil. Trans. R. Soc. Lond. B
290, 313±328 (1980).
20. Ikeda, H., Inoue, M. & Omura, S. Improvement of macrolide antibiotic-producing streptomycete
strains by the regeneration of protoplasts. J. Antibiot. (Tokyo) 36, 283±288 (1983).
International Patent Application No. WO 00/04190. (2000).
22. Hopwood, D. A. & Wright, H. M. Bacterial protoplast fusion. Mol. Gen. Genet. 162, 307±317
(1978).
23. Keiser, T., Bibb, M. J., Buttner, M. J., Chater, K. F. & Hopwood, D. A. Practical Streptomyces Genetics
(John Innes Foundation, Norwich, 2000).
24. Baltz, R. H. & Seno, E. T. Properties of Streptomyces fradiae mutants blocked in biosynthesis of the
macrolide antibiotic tylosin. Antimicrob. Agents Chemother. 20, 214±225 (1981).
25. Saliwanchik, R. in Manual of Industrial Microbiology and Biotechnology (eds Demain, A. L. &
Solomon, N. A.) 410±435 (Am. Soc. Microbiol. Washington, 1986).
Supplementary Information accompanies the paper on Nature's website
(http://www.nature.com).
Acknowledgements
The authors would like to thank D. Hopwood for the S. coelicolor strains, E. Lilly for
S. fradiae SF1 and SF21, R. Baltz, E. Cundliffe, G. Huisman, V. Gavrilovic, R. Patnaik,
M. Lassner, T. Cox and S. Louie for technical discussion and assistance, and the National
Institute of Standards and Technology Advanced Technology Program for ®nancial
assistance.
Correspondence and requests for materials should be addressed to S.B.d.C.
(e-mail: stephen.delcardayre@maxygen.com).
.................................................................
Vesicular restriction of synaptobrevin
suggests a role for calcium in
membrane fusion
Kuang Hu, Joe Carroll, Sergei Fedorovich, Colin Rickman,
Andrei Sukhodub & Bazbek Davletov
MRC Laboratory of Molecular Biology, Hills Road, Cambridge CB2 2QH, UK
..............................................................................................................................................
Release of neurotransmitter occurs when synaptic vesicles fuse
with the plasma membrane. This neuronal exocytosis is triggered
by calcium and requires three SNARE (soluble-N-ethylmaleimide-
sensitive factor attachment protein receptors) proteins: synapto-
brevin (also known as VAMP) on the synaptic vesicle, and
syntaxin and SNAP-25 on the plasma membrane1±4. Neuronal
SNARE proteins form a parallel four-helix bundle that is thought
to drive the fusion of opposing membranes5,6. As formation of this
SNARE complex in solution does not require calcium, it is not
clear what function calcium has in triggering SNARE-mediated
membrane fusion. We now demonstrate that whereas syntaxin
and SNAP-25 in target membranes are freely available for SNARE
complex formation, availability of synaptobrevin on synaptic
vesicles is very limited. Calcium at micromolar concentrations
triggers SNARE complex formation and fusion between synaptic
vesicles and reconstituted target membranes. Although calcium
does promote interaction of SNARE proteins between opposing
membranes, it does not act by releasing synaptobrevin from
NATURE | VOL 415 | 7 FEBRUARY 2002 | www.nature.com