Refactoring bacteriophage T7
Leon Y Chan
1,3
, Sriram Kosuri
2,3
and Drew Endy
2,
*
1
Department of Biology, Massachusetts Institute of Technology, Cambridge, MA, USA and
2
Division of Biological Engineering, Massachusetts Institute of Technology,
Cambridge, MA, USA
3
These authors contributed equally to this work
* Corresponding author. Division of Biological Engineering, Massachusetts Institute of Technology, 68-580, 77 Massachusetts Avenue, Cambridge, MA 02139, USA.
Tel.: þ 1 617 258 5152; Fax: þ 1 617 253 5865; E-mail: endy@mit.edu
Received 15.7.05; accepted 23.7.05
Natural biological systems are selected by evolution to continue to exist and evolve. Evolution likely
gives rise to complicated systems that are difficult to understand andmanipulate. Here, we redesign
the genome of a natural biological system, bacteriophage T7, in order to specify an engineered
surrogate that, if viable, would be easier to study and extend. Our initial design goals were to
physically separate and enable unique manipulation of primary genetic elements. Implicit in our
design are the hypotheses that overlapping genetic elements are, in aggregate, nonessential for T7
viability and that our models for the functions encoded by elements are sufficient. To test our initial
design, we replaced the left 11 515 base pairs (bp) of the 39 937bpwild-type genomewith 12179bp of
engineered DNA. The resulting chimeric genome encodes a viable bacteriophage that appears to
maintain key features of the original while being simpler to model and easier to manipulate. The
viability of our initial design suggests that the genomes encoding natural biological systems can be
systematically redesigned and built anew in service of scientific understanding or human intention.
Molecular Systems Biology 13 September 2005; doi:10.1038/msb4100025
Subject Categories: synthetic biology
Keywords: bacteriophage T7; synthetic biology; refactor
Introduction
In nature, the success of an individual organism depends
directly on its ability to continue to exist and replicate. Not
surprisingly, natural biological systems appear to have
evolved, and continue to evolve, to meet these requirements
(e.g., Block et al, 1982; Aho et al, 1988). However, should we
also expect that the ‘design’ of an evolved organism would be
further optimized for the purposes of human understanding
and interaction? Evidence drawn from fields outside biology
suggests that the answer is no.
For example, consider two different approaches to program-
ming computers and electronics: ‘genetic programming’ and
‘structured design.’ In genetic programming, evolutionary
algorithms are used to evolve computer software or electrical
hardware for a particular task (Koza et al, 2003). The absolute
performance of evolved systems often meets, and sometimes
exceeds, that produced by human-directed designs (Spector
et al, 1999). However, so-evolved systems lack human
readable descriptions and are difficult to understand, fix, and
modify for new applications. By contrast, a structured design
process produces systems that, in addition to functioning, are
designed to be easy to understand and extend (Abelson et al,
1996). Not surprisingly, an artifact produced via structured
design may not be optimal when evaluated only in terms of
absolute algorithmic or physical performance. However, a
structured design process can bypass two limitations, direct-
descent and replication-with-error, which constrain the
designs of evolved systems. Thus, we might paradoxically
expect that a structured design process will, when practical,
produce artifacts whose designs can ‘evolve’ more quickly.
Here, we converted the genome of a natural biological
system, bacteriophage T7, to a more structured design. Our
work was initially motivated by past failures in modeling T7
development (below) and by a desire to better understand how
the parts that comprise bacteriophage T7 work together to
encode a functioning whole (Kirschner, 2005). The approach
we used was inspired by the practice of ‘refactoring,’ a process
that is typically used to improve the design of legacy computer
software (Fowler et al, 1999). In general terms, the goal of
refactoring is to improve the internal structure of an existing
system for future use, while simultaneously maintaining
external system function.
T7 is an obligate lytic phage that infects Escherichia coli
(Dunn and Studier, 1983; Studier and Dunn, 1983). T7 was
twice isolated fromWardMacNeal’s ‘standard anti-coli-phage’
mixture (Demerec and Fano, 1945; Delbru¨ck, 1946). Mac-
Neal’s ‘mixture’ may have been cultured in series—T7was the
only identifiable isolate (Studier, 1979). One of the two original
T7 isolates was reportedly chosen for future use and master
cultures of ‘wild-type’ T7 have beenmaintained since (Supple-
mentary information). Genetics, and then biochemistry,
enabled the discovery and characterization of some of the
individual elements that participate in T7 development
(Molineux, 2005). Sequencing of the T7 genome revealed
additional elements (Dunn and Studier, 1983), not all of which
& 2005 EMBO and Nature Publishing Group Molecular Systems Biology 2005 1
Molecular Systems Biology (2005) doi:10.1038/msb4100025
& 2005 EMBO and Nature Publishing Group All rights reserved 1744-4292/05
www.molecularsystemsbiology.com
Article number: 2005.0018

have obvious functions (Molineux, 2005). As a result, T7 now
provides a goodmodel system for studyingwhat fraction of the
functional information encoded on the genome of a natural
biological system has been described, and how much of what
might still be understood is likely to matter (Davis, 1946).
For example, the T7 protein coding domains were first
characterized by the isolation and analysis of randomly
generated amber mutants. A total of 19 genes were identified
by mapping mutants that disrupt T7 DNA synthesis, particle
maturation, and lysis (Hausmann and Gomez, 1967; Studier,
1969; Hausmann and LaRue, 1969). Two additional genes, T7
DNA ligase and protein kinase, were isolated via loss of
function and deletion, respectively (Masamune et al, 1971);
the genetic analysis of ligase and kinase mutants was carried
out using mutant host strains that do not support the growth
of ligase- or kinase-defective phage (Studier, 1969). Up to 30
T7 proteins were observed by pulsing phage-infected cells
with radioactive amino acids (Studier and Maizel, 1969;
Studier, 1973a, b). Further experiments, such as electrophore-
tic mobility shifts of amber mutants, provided evidence for up
to 38 T7 proteins (Studier, 1981). Sequencing of the genome
confirmed the previously constructed genetic maps (Dunn and
Studier, 1983). But, analysis of the complete genome sequence
also revealed that the set of protein coding domains found via
mutagenesis, screening, andmapping was not exhaustive, and
that additional unidentified open reading frames occupied
most of the remainder of the genome. Some of these
unidentified open reading frames can be labeled as putative
protein coding domains based on the inferred strengths of
adjacent upstream ribosome binding sites (RBSs). In all, up
to 57 genes encoding 60 potential proteins have been found
or postulated (Molineux, 2005). However, only 35 of these
60 proteins have at least one known function. And, of the
25 nonessential proteins, only 12 are conserved across the
family of T7-like phage (Molineux, 2005). Can we safely
ignore these uncharacterized protein coding domains in our
models of phage infection? Should we edit the genome to
remove them?
As a second example, the E. coli RNA polymerase promoters
on the T7 genome (A0, A1–3, B, C, and E)were first mapped by
in vitro transcription studies (Davis andHyman, 1970;Minkley
and Pribnow, 1973; Golomb and Chamberlin, 1974a, b; Niles
and Condit, 1975; McAllister and McCarron, 1977; Stahl and
Chamberlin, 1977; Kassavetis and Chamberlin, 1979; Panayo-
tatos and Wells, 1979) and subsequently confirmed by
sequencing (Oakley and Coleman, 1977; Boothroyd and
Hayward, 1979; Rosa, 1979; Carter and McAllister, 1981;
Osterman and Coleman, 1981; Dunn and Studier, 1983).
Results of in vitro transcription reactions using T7 genomic
DNA as template agreed with the available in vivo transcrip-
tion data (Studier, 1973a, b; Summers et al, 1973; McAllister
and Wu, 1978; McAllister et al, 1981). However, the cloning of
random sections of the T7 genome into a plasmid that selected
for transcription activity from the cloned fragment identified
other possible promoters (Studier and Rosenberg, 1981).
Sequence analysis of the cloned sections identified B10
regions with homology to known promoters; footprinting
assays identified two additional promoters (Dunn and Studier,
1983). But, any contribution of these putative promoters to
wild-type T7 infection is not now defined. As with some of the
T7 genes, should we ignore these promoters in our models?
Should we delete these elements from the T7 genome? Is there
other information encoded on the wild-type T7 genome that
we should include in our models, ignore, or actively remove?
One practical test of the understanding encoded in a model
of a system is to use themodel to help predict what will happen
when either the system or its environment is changed. In the
case of T7, the experimentalists who originally discovered
much of how the phage works developed the best descriptive,
system-level models for T7 infection. Their models were made
by integrating knowledge of the individual parts and mechan-
isms that act during infection, from genome entry to phage
particle formation (Studier and Dunn, 1983). Two features
specific to T7 biology made the construction of system-level
models easier. First, compared to other phage, T7 is relatively
independent of complex host physiology. For example, the
optical density of T7-infected cultures stops increasing at
the time of infection, T7 encodes phage-specific RNA and
DNA polymerases, and E. coli mRNA and protein synthesis
is inhibited within the first B6min of T7 infection. Second,
RNA polymerase pulls most of the T7 genome into the
newly infected cell (Zavriev and Shemyakin, 1982; Garcia and
Molineux, 1995). Polymerase-mediated genome entry is a
relatively slow process that results in the direct physical
coupling of gene expression dynamics to gene position. For
example, a gene cannot be expressed until its coding domain
enters the newly infected cell.
Building on this work, others and we developed computa-
tional, quantitative models of T7 infection in order to explore
questions related to the organization of genetic elements on
the T7 genome, and the timing and control of gene expression
across uncertain physical environments (Endy et al, 1997;
Endy et al, 2000; You and Yin, 2002; You et al, 2002). Initially,
our models were used to test the hypothesis that the results
of 60 years of research on bacteriophage T7, conducted by
many researchers across many labs, could be integrated to
produce a T7-like computer simulation. The resulting model
and simulation recapitulated the apparent molecular details
and dynamics of T7 development quite well. Unfortunately,
the model itself was of little interest, as it turned out to be
overfitted to limited experimental data—changes to the model
led to predictions that were unbelievable (D Endy, unpub-
lished). A subsequent revision, inwhich carewas taken to only
include known facts andmechanisms, produced amodel of T7
that matched the available system-wide data less well, but that
was more useful as a tool for exploring how changes to the
phage genome and the host cell environment impact phage
development (Endy et al, 1997). However, in using these
computational models, some predictions did not agree with
experiments (Endy and Brent, 2001). For example, a mutant
phage expected to grow faster than the wild type grew slower
(Endy et al, 2000).
Upon inspection, disagreements between model-based
prediction and experiment could have arisen for at least three
reasons. First, our models could not meaningfully include
unknown functions. For example, disruption of an unchar-
acterized nonessential gene, 1.7, appeared to impact DNA
replication (Endy et al, 2000). While differences between
expectation and observation can suggest follow-on science,
a lack of complete component-level understanding debased
Refactoring bacteriophage T7
LY Chan et al
2 Molecular Systems Biology 2005 & 2005 EMBO and Nature Publishing Group

our system-level analyses. Second, the boundaries of genetic
elements on the T7 genome aremore complex than ourmodels
of the genome. For example, genes 2.8 and 3 are most easily
modeled as separable genetic elements even though the actual
genes 2.8 and 3 overlap (Figure 1A and Supplementary Figure
S3). Element overlap may also encode uncharacterized
function(s) having to do with the regulation of protein
synthesis or the coupling of selective pressures during
evolution. For example, bioinformatic analysis of microbial
genomes suggests that gene overlaps are conserved across
evolutionary distance (e.g., Johnson and Chisholm, 2004).
Element overlap also prohibits independent element manip-
ulation. For example, on the wild-type genome, we cannot
change the gene 3 RBS without at least changing the codon
usage of gene 2.8. Third, a computer model built with
separable parts that encode independent functions can be
overmanipulated relative to the actual physical system. For
example, while we could simulate the expected behavior of
large sets of permuted genomes, we could not easily move
a single open reading frame to another arbitrary position on
the actual T7 genome (Endy et al, 2000).
Wild-type T7 is a superb organism for discovering the
primary components of a natural biological system (Studier,
1972). However, is the original T7 isolate also best suited for
understanding how all parts of the phage are organized to
encode a functioning whole? Given our experiences, we
decided to attempt to engineer a surrogate genome, which
we designated T7.1.
Results
Design goals
Six goals drove our design of the T7.1 genome. First, we wanted
to define a set of components that function during T7
development and, for each element, choose an exact DNA
sequence thatwe could use to encode element function. Second,
we wanted the DNA sequence encoding the function of any one
element to not overlap with the DNA sequence encoding any
other element. Third, we wanted the DNA sequence of each
element to encode only the function assigned to that element
and not any other functions. Fourth, we wanted to enable the
precise and independent manipulation of each element. Fifth,
we needed to be able to construct the T7.1 genome. Sixth, we
needed the T7.1 genome to encode viable bacteriophage; at the
start of this work, we were uncertain how many simultaneous
changes the wild-type genome could tolerate.
Design process
Ageneral algorithm describing our genome refactoring process
is given in Supplementary Figure S1. Briefly, we began design
of the T7.1 genome by reannotating the genome of wild-type
T7. The wild-type T7 genome is a 39 937 base pair (bp) linear
double-stranded DNA molecule (Dunn and Studier, 1983). We
annotated the genome by specifying the boundaries of the
following functional genetic elements: 57 open reading frames
with 57 putative RBSs encoding 60 proteins, and 51 regulatory
elements controlling phage gene expression, DNA replication,
and genome packaging.
To specify the architecture of T7.1, we organized the
functional genetic elements into 73 ‘parts.’ Each part contains
one or more elements. While the DNA sequence of elements
within parts may overlap, there is no overlap across part
boundaries (Figure 1B). Next, we organized contiguous parts
into ‘sections’ with section boundaries defined by restriction
endonuclease sites found only once in the sequence of the
wild-type genome. Six sections, alpha through zeˆta, make up
the T7.1 genome (Figure 2A and Supplementary Figure S2).
Sections were used to compartmentalize changes across the
genome. In addition, sections can be built, tested, and
manipulated independently.
To specify the DNA sequence of T7.1, we eliminated
sequence overlap across part boundaries. Overlaps were
eliminated by exact duplication of the wild-type DNA
sequence; subsequent sequence editing produced a single
instance of any duplicated element (Figure 1B and Supple-
mentary Figure S3). All sequence edits within open reading
frames were silent and maintained the wild-type tRNA
specification or, when necessary, specified a higher abundance
tRNA (Ikemura, 1981). We also added bracketing restriction
endonuclease sites to insulate and enable the independent
manipulation of each part (Figure 2C and E and Supplemen-
tary Figure S2). Bracketing sites are not used elsewhere in the
sequence of any one section but are reused across sections.
The DNA sequence of T7.1 changes or adds 1424 bp to the
wild-type genome (Supplementary Figure S3).
Construction and testing
The sections that comprise the T7.1 genome can be built and
tested independently. We constructed the first two sections,
alpha and beta (Materials and methods). Alpha and beta
contain the first 32 of 73 parts of the T7.1 genome, replacing
the left 11 515 bp of the wild-type genome with 12179 bp of
redesigned DNA, and encoding the entire T7 early region, the
Figure 1 Element decompression and part design. (A) The coding regions of genes 2.8 and 3 overlap in the wild-type T7 genome. The RBS of gene 3 (underlined) is
encoded within gene 2.8.(B) Distinct genetic parts make up the T7.1 genome. The natural RBS and start codon (green) for gene 3 are disrupted by point mutations
(capitals); mutations do not change the amino-acid sequence of the 2.8 protein. Parts 28 and 29 are separated by bracketing restriction sites, BamHI (blue) and EagI
(orange). Supplementary Figure S3 lists all changes in the DNA sequence of T7.1 relative to wild-type T7.
Refactoring bacteriophage T7
LY Chan et al
& 2005 EMBO and Nature Publishing Group Molecular Systems Biology 2005 3

study or extend the original. Here, we decided to redesign the
genome of a natural biological system, bacteriophage T7, in
order to specify an engineered biological system that is easier
to study and manipulate. The new genome, T7.1, is based on
our incomplete understanding of the information encoded in
the wild-type genome and our desire to insulate and
independently manipulate known primary genetic elements.
We constructed the first two sections of T7.1, making over 600
simultaneous changes or additions to the wild-type DNA, and
observed that the resulting chimeric phage are viable.
Phage viability demonstrates the following for sections
alpha and beta. First, our parts as chosen can be separated
by exogenous DNA sequence. Second, any functions encoded
by genetic element overlap are, in aggregate, nonessential
under standard laboratory conditions. Third, our current
understanding of T7 is not insufficient to specify a viable
bacteriophage. Viability does not demonstrate sufficiency
because (i) if the chimeric phage had not been viable, then
our current understanding would have been demonstrably
insufficient, and (ii) while T7.1 is based on our current
understanding, we do not have an exact understanding of all
functions encoded in the T7.1 genome (e.g., genes of unknown
function). Finally, viability, combined with the observed
similarities in lysis times, suggests that T7.1 preserves
polymerase-mediated genome entry and remains relatively
independent of host cell physiology.
The T7.1 genome is easier to model and study. For example,
by removing genetic element overlap, the T7.1 genome better
matches the understanding of T7 biology encoded in our
models, relative to the wild-type phage. However, more work
is needed to demonstrate that the dynamic behavior of the
system encoded by the T7.1 genome is easier to predict. Such
work will benefit from the fact that the parts of T7.1 can be
independently manipulated.
Our design of T7.1 was constrained by fears of producing
a nonviable DNA fragment that would have been difficult to
analyze and rescue. Given our initial success with T7.1, we
have decided to revisit and extend our original design goals.
For example, the design of our next phage, T7.2, will include (i)
reduced gene sets that eliminate nonessential and nonconserved
protein coding domains, (ii) codon shuffling of protein coding
domains in order to disrupt secondary and cryptic regulatory
elements, and putative mRNA secondary structure, and (iii)
standard regulatory elements and regulatory element spacing.
By actively removing all of the uncharacterized elements thatwe
know about, as well as taking steps to disrupt any uncharacter-
ized elements as yet unknown, we will be able to better study
how the parts of T7 work to encode a functioning whole.
We constructed sections alpha and beta manually. Con-
current advances in de novo DNA synthesis technology have
recently enabled the rapid automatic synthesis of DNA
fragments the size of the T7.1 genome sections (Stemmer
2
L
o
g

l
a
d
d
e
r
2
L
o
g

l
a
d
d
e
r
P
2
.
S
p
h
I
P
4
.
H
i
n
d
I
I
I
P
5
.
B
s
s
H
I
I
P
6
.
S
e
x
A
I
U
1
.
S
a
c
I
P
7
.
M
l
u
I
U
2
.
N
h
e
I
P
8
.
B
s
i
W
I
P
9
.
R
s
r
I
I
P
1
0
.
S
a
c
I
I
P
1
1
.
E
a
g
I
P
1
3
.
E
c
o
R
I
P
1
4
.
P
f
o
I
U
3
.
A
p
a
L
I
P
1
6
.
X
m
a
I
P
1
8
.
N
c
o
I
P
2
1
.
A
a
t
I
I
P
2
3
.
A
g
e
I
P
2
3
.
A
g
e
I
P
2
4
.
B
s
t
E
I
I
P
2
5
.
B
s
i
W
I
P
2
6
.
E
c
o
R
I
P
2
7
.
X
m
a
I
P
2
8
.
B
a
m
H
I
P
2
9
.
E
a
g
I
P
3
0
.
S
a
c
I
I
P
3
1
.
P
c
i
I
P
3
2
.
S
a
l
I
u
n
d
i
g
e
s
t
e
d
u
n
d
i
g
e
s
t
e
d
Figure 3 Cutting parts from T7.1. (A) Restriction enzymes specific to the sites that bracket parts (P#.Enzyme) and added unique restriction sites (U#.Enzyme) were
used to cut section alpha (Supplementary information). A subset of the digests is shown. As built, part 1 cannot be removed. (B) Restriction digests cutting out all parts
in section beta. As built, part 28 cannot be removed.
Refactoring bacteriophage T7
LY Chan et al
& 2005 EMBO and Nature Publishing Group Molecular Systems Biology 2005 5

et al, 1979; Dunn and Studier, 1980). A second terminator specific to
T7 RNA polymerase was suggested by in vitro transcription studies on
digested T7 DNA (Golomb and Chamberlin, 1974a, b; Niles and
Condit, 1975). The terminator, named ‘Tø,’ was shown to function
in situ (Dunn and Studier, 1980) and on plasmids (McAllister et al,
1981). Both TE and Tø have stem loop structures that are thought to set
termination efficiency (Dunn and Studier, 1973). The stem loop and
flanking sequence, which includes a poly-uridine tract, were taken
together to define the element we used here. While other terminators
have been postulated, their precise location and function, if any,
during wild-type infection are tenuous (Dunn and Studier, 1983), and
thus we did not include them in our annotation.
RNaseIII recognition sites
The definition of an RNaseIII recognition site that we used here is a
contiguous stretch of DNA that, when transcribed, produces a region
of mRNA that is recognized and cleaved (at some efficiency) by
RNaseIII. Sites for specific cleavage of T7 RNA by RNaseIII were first
shown in vitro and then correlated to in vivo data (Dunn and Studier,
1973). In time, 10 RNaseIII sites were mapped and their sites of
cleavage identified (Dunn and Studier, 1983). The sites are thought to
stabilize the 3
0
end of T7 transcripts by providing a stem loop that
prevents the binding of scanning single-stranded RNases. A down-
stream gene often immediately follows an RNaseIII site. Thus, we kept
the RNaseIII recognition site elements as short as possible—with a
minimum boundary set by the probable stem loop structures (Dunn
and Studier, 1983).
DNA replication origins
The definition of a DNA replication origin that we used here is a stretch
of DNA that is used to initiate the copying of phage DNA during T7
infection. The primary replication origin was mapped to the dual
promoter region downstream of ø1.1A and ø1.1B by analysis of
replication bubbles in electron micrographs (Dressler et al, 1972;
Wolfson et al, 1972) and subsequently sequenced (Saito et al, 1980).
The secondary origin at øOL was identified using mutants that lacked
the primary origin (Tamanoi et al, 1980; Studier and Rosenberg, 1981).
Finally, plasmids containing cloned fragments of T7 DNAwere used to
screen for regions that act as replication origins during T7 infection;
these experiments revealed that øOR and ø13 have origin activity
(Dunn and Studier, 1983). While the precise boundaries of the repli-
cation origins are unknown, each appears to be linked to a functioning
RNA polymerase promoter (Zhang and Studier, 2004). Here, we only
annotate and define an element for the primary origin.Whilewe do not
include other replication origins as elements, we do preserve the RNA
polymerase promoters that are associatedwith these secondary origins
as elements, and thus possibly the secondary origins as well.
Terminal repeats and short repeats
The definition of a terminal repeat that we used here is a contiguous
stretch of DNA present at both ends of the T7 genome, and a short
repeat is a series of direct repeats of DNA near the end of the genome.
Both the left and right ends of the T7 genome contain exact 160 bp
direct repeats (Ritchie et al, 1967). Also, adjacent to the direct repeats
on both ends of the genome are regions of DNA that contain 12
regularly arranged and highly conserved 7 bp sequences termed the
short repeats left, SRL, and right, SRR (Dunn and Studier, 1981). The
terminal repeats and SRL/R are thought to be involved in concatemer
formation, DNA packaging, and particle maturation (Kelly and
Thomas, 1969). However, the mechanisms by which the direct repeats
and the SRL/R act are unclear. Thus, we treated each end’s direct
repeat and SRL/R as a monolithic element (the design of T7.1 does not
make any changes to the DNA sequence of these elements).
Design of T7.1 genome
Overview
The design of T7.1 genome uses six sections, alpha through zeˆta. Each
section contains parts that are amalgamations of one or more
functional genetic elements (Supplementary Figure S2). In our design,
the modification of parts on the full T7.1 genome is a two-stage
process. First, we can manipulate parts to construct a section. Second,
we can combine sections to assemble a full genome. We improved
upon the design of sections beta through zeˆta based on our experience
constructing section alpha.
Definitions
#-Cutter—a restriction enzyme that cuts a particular DNA sequence #
times;
Functional genetic element—a promoter, protein coding domain, RBS,
etc., defined during our reannotation of the T7 genome;
Part—a piece of DNA that encodes one or more functional genetic
elements and is bracketed by a pair of identical restriction sites;
Construct—any amalgamation of functional genetic elements or parts;
Section—a segment of the T7.1 genome the boundaries of which are
1-cutters on the wild-type T7 genome.
T7.1 genome sections
We used sections to limit the number of simultaneous changes to the
wild-type T7 sequence and to make the construction process more
manageable. Two practical considerations drove our choice of section
boundaries. First and foremost, the boundaries of the sections had to
be compatible with the sparse distribution of 1-cutter sites across the
wild-type genome. (The use of 1-cutter sites for section boundaries
allows refactored sections to be easily combined with other sections or
with wild-type DNA.) Second, the number of parts per section was
limited by the number of ‘useful’ 0-cutters across the DNA sequence of
each wild-type section. Useful 0-cutters are specific, free or smaller
recognition sites, dam/dcm insensitive, and leave sticky-end overhangs.
From functional genetic elements to T7.1 parts
Parts are made up of one or more functional genetic elements. Parts
were sometimes defined to have more than one element in order to
maintain the natural proximity of elements known, or likely, to be
physically or functionally coupled. For example, we grouped most
RBSs and downstream protein coding domains into two-element
parts. Also, some functional genetic elements overlap so severely as
to prevent efficient separation (e.g., the genes 4A, 4B, 4.1, and 4.2).
Finally, some functional genetic elements were very short (o150bp)
such that variants containing deletions or separations of the individual
elements could be easily constructed (e.g., the E. coli promoter C and
RNaseIII site R1). In total, we combined the elements that make up
T7.1 into 73 parts. We numbered parts, 1–73, starting from the genetic
left end.
The arrangements of parts on the wild-type T7 DNA sequence
sometimes resulted in the overlapping of the DNA sequence specifying
parts. To remove part–part overlap, we duplicated the DNA sequence
of the overlap, providing both parts with an independent copy of the
previously overlapping sequence. If, as a result of sequence duplica-
tion, either of the parts encoded a function specific to an element in the
other part, we mutated the sequence to eliminate the duplicate
function. All mutations to protein coding domains were silent and
result in either no change in the tRNA or, when necessary, specify a
higher abundance tRNA (Ikemura, 1981). Parts separation is detailed
in Supplementary Figure S2.
We surrounded each part with a restriction site pair that is not
contained elsewhere in that part’s section. Typically, we added
bracketing restriction sites to the DNA sequence of each part, but,
when appropriate, we integrated the sites into the natural DNA
sequence. Also, to help reduce the length of T7.1, where possible, we
chose adjacent restriction sites to have overlapping sequence with one
another.
One of themost significant differences between the design of section
alpha and the other sections was in our choice of bracketing restriction
sites. In section alpha, we picked restriction enzymes that did not cut
within section alpha only. However, as the construction of alpha
proceeded, and cloning directly into the phage became useful, we
adjusted our design strategy to use restriction enzymes that did not cut
within the entire genome wherever possible.
Refactoring bacteriophage T7
LY Chan et al
& 2005 EMBO and Nature Publishing Group Molecular Systems Biology 2005 7

Design features
Deletion and insertion: The design of the T7.1 genome allows for the
simple deletions of parts. Generally, we isolate the section containing
the part by digesting with the bracketing restriction enzyme. We ligate
the fragments to reform the section minus the deleted part, and then
join the section to the rest of the genome. Insertion of a newpart can be
more involved. Most simply, if there is a pre-existing restriction site
due to a deletion operation, thenwe can insert a newpart in its place. If
no such site exists, another method involves using two restriction
enzymes, NgoMIV and BspEI, that are 0-cutters across both the wild-
type T7 and all refactored sections. NgoMIV and BspEI have different
recognition sequences but produce the same overhang upon digestion.
This allows for ligation of a product into these sites, while simul-
taneously preventing the restriction sites from being reformed. Thus,
we can replace a part adjacent to the desired insertion site with the
same part that has anNgoMIV site appended to it. Then,we amplify the
part to be inserted with bracketing BspEI sites and insert the part into
the NgoMIV site. Since neither restriction site is reformed upon
insertion, this method is idempotent.
Unstuffing hooks: Since we did not know how a phage made of
separated parts would function (e.g., would it form plaques?), we
thought that it would be prudent to be able to easily revert to the wild-
type T7 sequence for purposes of comparison and debugging. Thus,
we used silent mutations to add additional 1-cutter restriction sites to
section alpha. These new restriction sites, labeled U1–4, are useful if
we desired to replace refactored regions with wild-type sequence. In
sections beta through zeˆta, such extra sites were superfluous because
we used 0-cutters to bracket parts; 0-cutters can also be used to revert
refactored regions to wild-type sequence.
Scaffolds: We used scaffolds to build sections alpha and beta.A
scaffold is essentially the sequence that remains when all parts are
removed from the section. As such, the scaffold contains all the
restriction sites required to assemble the parts to form the section. In
addition, if a fully refactored phage was not viable, we could use the
scaffold to incrementally revert the sequence back to wild type in an
attempt to restore function.
Construction of section alpha
The design of the scaffold for section alpha included all functional
genetic elements from the left end of T7 through R0.3,R0.5, parts 17,
19, 21, plus the restriction sites required to add all remaining parts. The
section alpha scaffold does not contain any known protein coding
domains. We sent the scaffold sequence (1334 bp) to Blue Heron
Biotechnology for synthesis (http://www.blueheronbio.com/). Blue
Heron could not assemble the scaffold using the standard cloning
plasmids then in use (we have sinceworkedwith Blue Heron to fix this
problem—below). Blue Heron agreed to ship the section alpha scaffold
as four fragments with point mutations in each fragment. The point
mutations were
Fragment 1: single-base changes at 89(G-T), 168(A-T), 169(C-A),
245(G-A), and 249(C-A) as well as single-base dele-
tions at 138 and 159;
Fragment 2: a single-base deletion in the 35 box of the A1
promoter;
Fragment 3: a four-base deletion between the 35 and 10 boxes
of the A3 promoter;
Fragment 4: a single-base deletion in the loop of TE.
We decided to discard Fragment 1 but to correct and make use of
Fragments 2, 3, and 4. We built a new vector, pREB, to facilitate the
assembly of section alpha. pREB (for rebuild) started as a chimera of
the inducible copy control system of pSCANS-5 and the insulated
multicloning site (MCS) of pSB2K3-1 (below). We completed pREB by
adding a smaller MCS containing PstI, BstBI, and BclI restriction
endonuclease sites and by removing 19 other restriction sites from the
plasmid backbone.
To build section alpha, we first cloned parts 5, 6, 7, 8, 12, 13, 14, 15,
16, 18, 20, 22, and 24 into pSB104. We cloned part 11 into pSB2K3. We
cloned each part with its part-specific bracketing restriction sites
surrounded by additional BioBrick restriction sites (Knight, 2002). We
used site-directed mutagenesis on parts 6, 7, 14, and 20 to introduce
the sites U1, U2, U3, and U4, respectively. Our site-directed
mutagenesis of part 20 failed.
We used site-directed mutagenesis to remove a single Eco0109I
restriction site from the vector pUB119BHB carrying the scaffold
Fragment 4. We cloned part 15 into this modified vector. We then
cloned scaffold Fragment 4 into pREB and used serial cloning to add
the following parts: 7, 8, 12, 13, 14, 16, 18, 20, 22, and 23. We digested
the now-populated scaffold Fragment 4 withNheI and BclI and purified
the resulting DNA.
Next, we cloned parts 5 and 6 into pUB119BHB carrying scaffold
Fragment 3. We used the resulting DNA for in vitro assembly of a
construct spanning from the left end of T7 to part 7. To do this, we cut
wild-type T7 genomic DNA with AseI, isolated the 388 bp left-end
fragment, and ligated this DNA to scaffold Fragment 2. We selected the
correct ligation product by PCR. We fixed the mutation in part 3 (A1)
via a two-step process. First, PCR primers with the corrected sequence
for part 3 were used to amplify the two halves of the construct to the
left and right ends of part 3. Second, a PCR ligation joined the two
constructs. We added scaffold Fragment 3 to the above left-end
construct once again by PCR ligation as described above. We repaired
themutation in part 4 (A2, A3, and R0.3) following the same procedure
as with part 3. We used a right-end primer containing an MluI site
to amplify the entire construct, and used the MluI site to add part 7.
We used PCR to select the ligation product, digested the product with
NheI, and purified the resulting DNA.
We isolated the right arm of a BclI digestion of wild-type T7 genomic
DNA and used ligation to add the populated left-end construct and the
populated Scaffold Fragment 4. We transfected the three-way ligation
product into IJ1127. We purified DNA from liquid culture lysates
inoculated from single plaques. We used restriction enzymes to digest
the DNA and isolate the correct clones.
Next, we added part 11 via three-way ligation and transfection.
Because the restriction sites that bracket part 9 (RsrII) also cut wild-
type T7 DNA, we needed to use in vitro assembly to add this part to a
subsection of section alpha. To do this, we used PCR to amplify the
region spanning parts 5–12 from the refactored genome. We cut the
PCR product with RsrII and ligated part 9. We used PCR to select
the correct ligation product; this PCR reaction also added a SacII site
to the fragment. We digested the PCR product with SacI and SacII
and cloned onto the otherwise wild-type phage. Lastly, we used the
SacII site to clone part 10 onto the phage.
Construction of section beta
We constructed section beta using a process similar to that used with
alpha. A scaffold with all restriction sites as well as part 26 was made
by Klenow extension of overlapping primers. We digested this DNA
with BstBI and cloned it onto pREB.We then added the following parts:
23, 24, 27, 28, 30, 31, and 32. We had to clone part 32 (containing gene
3.8) as a truncation since we were unable to clone the full-length
part, probably due to the apparent toxicity of gene 3.8 product. The
truncated version of part 32 still included the BglII site to allow for
assembly of section beta onto phage. We added parts 25 and 29, also
previously reported to be toxic, in vitro. To insert part 25, we amplified
a region spanning parts 23–27 by PCR. We cut this fragment with
BsiWI. Part 25 was then ligated to each of these fragments separately
and selected for by PCR. We cut both PCR products with DraIII, a
restriction site internal to part 25, ligated and then selected for full-
length part 25 by PCR.We cut part 25 with BclI and MluI, purified, and
ligated it to wild-type fragments. We used a similar approach to insert
part 29 (using the EcoO109I site internal to this part). Lastly, we cut
both phage genomes with MluI; we ligated the left fragment of the
genome containing the refactored region spanning parts 23–27 to the
right fragment of a genome containing the refactored region spanning
parts 27–32.
Synthesis and construction errors
Differences between the designed and constructed sections alpha and
beta are detailed in Supplementary Tables S1 and S2.
Refactoring bacteriophage T7
LY Chan et al
8 Molecular Systems Biology 2005 & 2005 EMBO and Nature Publishing Group

Protocol, strains, and media
Detailed laboratory protocols, strain information, and media recipes
are provided as Supplementary information.
Supplementary information
Supplementary information is available at Molecular Systems Biology
website (www.nature.com/msb).
Accession numbers
The DNA sequence encoding the as-built sections alpha and beta are
available via GenBank (DQ100054, DQ100055). The entire T7.1 design
and our reannotation of the wild-type T7 genome are also available
online:
http://web.mit.edu/endy/www/ncbi/
Competing interests
The authors have declared that no competing interests exist.
Author contributions
DE conceived the project. LYC, SK, and DE designed the experiments.
LYC and SK designed the T7.1 genome, developed all software, and
performed the experiments. LYC, SK, and DE wrote the paper.
Acknowledgements
We thank Ian Molineux, Priscilla Kemp, and Heather Keller for
discussions and advice throughout thework.We thank John Dunn and
Barbara Lade for the pSCANS-5 vector. We thank Roger Brent, Eric
Eisenstadt, Tom Knight, andmembers of the Endy group for additional
discussions and sustained encouragement.We thank Jorge Borges and
Adolfo Casares for ‘On Exactitude in Science’ (Davis, 1946). We thank
Austin Che, Heather Keller, Alex Mallet, Kathleen McGinness,
Samantha Sutton, Ty Thomson, Elizabeth Vesilind, and Rebecca Ward
for comments on the manuscript. We thank Felice Frankel for plaque
photography and encouragement. This work was funded by grants
to DE from the US Office of Naval Research, DARPA, and NIH. SK
was supported by an NIH MIT BPEC training fellowship. Additional
support was provided by MIT.
References
AbelsonH, SussmanGJ, Sussman J (1996) Structure and Interpretation
of Computer Programs, 2nd edn. Cambridge, MA, USA: MIT Press
Aho A-C, Donner K, Hyden C, Larsen LO, Reuter T (1988) Low retinal
noise in animals with low body temperature allows high visual
sensitivity. Nature 334: 348–350
Block SM, Segall JE, Berg HC (1982) Impulse responses in bacterial
chemotaxis. Cell 31: 215–226
Boothroyd JC, Hayward RS (1979) New genes and promoters
suggested by the DNA sequence near the end of the coliphage T7
early operon. Nucleic Acids Res 7: 1931–1943
Carlson R (2003) The pace and proliferation of biological technologies.
Biosecur Bioterror 1: 203–214
Carter AD, McAllister WT (1981) Sequences of three class II promoters
for the bacteriophage T7 RNA polymerase. J Mol Biol 153: 825–830
Davis BL (1946) Del rigor en la ciencia. Los Anales de Buenos
Aires an˜o1
Davis RW, Hyman RW (1970) Physical locations of the in vitro RNA
initiation site and termination sites of T7 DNA. Cold Spring Harb
Symp Quant Biol 35: 269–282
Delbru¨ck M (1946) Bacterial viruses or bacteriophages. Biol Rev Camb
Philos Soc 21: 30–40
Demerec M, Fano U (1945) Bacteriophage-resistant mutants in
Escherichia coli. Genetics 30: 119–136
Dressler D, Wolfson J, Magazin M (1972) Initiation and reinitiation of
DNA synthesis during replication of bacteriophage T7. Proc Natl
Acad Sci USA 69: 998–1002
Dunn JJ, Studier FW (1973) T7 early RNAs are generated by site-
specific cleavages. Proc Natl Acad Sci USA 70: 1559–1563
Dunn JJ, Studier FW (1980) The transcription termination site at the
end of the early region of bacteriophage T7 DNA. Nucleic Acids Res
8: 2119–2132
Dunn JJ, Studier FW (1981) Nucleotide sequence from the genetic left
end of bacteriophage T7 DNA to the beginning of gene 4. J Mol Biol
148: 303–330
Dunn JJ, Studier FW (1983) Complete nucleotide sequence of
bacteriophage T7 DNA and the locations of T7 genetic elements.
J Mol Biol 166: 477–535
Endy D, Brent R (2001) Modelling Cellular Behavior. Nature 409:
391–395
Endy D, Kong D, Yin J (1997) Intracellular kinetics of a growing
virus: a genetically structured simulation for bacteriophage T7.
Biotechnol Bioeng 55: 375–389
Endy D, You L, Yin J, Molineux IJ (2000) Computation, prediction, and
experimental tests of fitness for bacteriophage T7 mutants with
permuted genomes. Proc Natl Acad Sci USA 97: 5375–5380
Fowler M, Beck K, Brant J, Opdyke W, Roberts D (1999) Refactoring:
Improving the Design of Existing Code. Boston, MA, USA: Addison-
Wesley Professional
Garcia LR, Molineux IJ (1995) Rate of translocation of bacteriophage
T7 DNA across the membranes of Escherichia coli. J Bacteriol 177:
4066–4076
Golomb M, Chamberlin M (1974a) A preliminary map of the
major transcription units read by T7 RNA polymerase on the T7
and T3 bacteriophage chromosomes. Proc Natl Acad Sci USA 71:
760–764
Golomb M, Chamberlin M (1974b) Characterization of T7-specific
ribonucleic acid polymerase. IV. Resolution of the major in vitro
transcripts by gel electrophoresis. J Biol Chem 249: 2858–2863
Hausmann R, Gomez B (1967) Amber mutants of bacteriophages T3
and T7 defective in phage-directed deoxyribonucleic acid synthesis.
J Virol 1: 779–792
Hausmann R, LaRue K (1969) Variations in sedimentation patterns
among deoxyribonucleic acids synthesized after infection of
Escherichia coli by different amber mutants of bacteriophage T7.
J Virol 3: 278–281
Ikemura T (1981) Correlation between the abundance of Escherichia
coli transfer RNAs and the occurrence of the respective codons
in its protein genes: a proposal for a synonymous codon choice
that is optimal for the E. coli translational system. J Mol Biol 151:
389–409
Johnson ZI, Chisholm SW (2004) Properties of overlapping genes are
conserved across microbial genomes. Genome Res 14: 2268–2272
Kassavetis GA, Chamberlin MJ (1979) Mapping of class II promoter
sites utilized in vitro by T7-specific RNA polymerase on bacterio-
phage T7 DNA. J Virol 29: 196–208
Kelly Jr TJ, Thomas Jr CA (1969) An intermediate in the replication of
bacteriophage T7 DNA molecules. J Mol Biol 44: 459–475
Kirschner MW (2005) The meaning of systems biology. Cell 20:
503–504
Knight T (2002) Idempotent vector design for standard assembly of
biobricks. MIT Synthetic Biology Working Group Technical Report 0
[http://web.mit.edu/synbio/release/docs/biobricks.pdf]
Kodumal SJ, Patel KG, Reid R, Menzella HG,Welch M, Santi DV (2003)
Total synthesis of long DNA sequences: synthesis of a contiguous
32-kb polyketide synthase gene cluster. Proc Natl Acad Sci USA 101:
15573–15578
Koza JR, Keane MA, Streeter MJ, Mydlowec W, Yu J, Lanza G (2003)
Genetic Programming IV: Routine Human-Competitive Machine
Intelligence. Dordrecht, Netherlands: Kluwer Academic Publishers
Masamune Y, Frenkel GD, Richardson CC (1971) A mutant of
bacteriophage T7 deficient in polynucleotide ligase. J Biol Chem
246: 6874–6879
Refactoring bacteriophage T7
LY Chan et al
& 2005 EMBO and Nature Publishing Group Molecular Systems Biology 2005 9

McAllister WT, McCarron RJ (1977) Hybridization of the in vitro
products of bacteriophage T7 RNA polymerase to restriction
fragments of T7 DNA. Virology 82: 288–298
McAllister WT, Morris C, Rosenberg AH, Studier FW (1981) Utilization
of bacteriophage T7 late promoters in recombinant plasmids during
infection. J Mol Biol 153: 527–544
McAllister WT, Wu HL (1978) Regulation of transcription of the late
genes of bacteriophage T7. Proc Natl Acad Sci USA 75: 804–808
Minkley EG, Pribnow D (1973) Transcription of the early region of
bacteriophage T7: selective initiation with dinucleotides. J Mol Biol
77: 255–277
Molineux IJ (2005) The T7 Group. In The Bacteriophages, Calendar RL
(ed) Chapter 20. Oxford: Oxford University Press
Niles EG, Condit RC (1975) TranslationalMapping of Bacteriophage T7
RNAs synthesized in vitro by purified T7 RNA polymerase. J Mol
Biol 98: 57–67
Oakley JL, Coleman JE (1977) Structure of a promoter for T7 RNA
polymerase. Proc Natl Acad Sci USA 74: 4266–4270
Olson ER, Flamm EL, Friedman DI (1982) Analysis of nutR: a region of
phage lambda required for antitermination of transcription. Cell 31:
61–70
Osterman HL, Coleman JE (1981) T7 ribonucleic acid polymerase-
promoter interactions. Biochemistry 20: 4884–4892
Panayotatos N,Wells RD (1979) Recognition and initiation site for four
late promoters of phage T7 is a 22-base pair DNA sequence. Nature
280: 35–39
Ritchie DA, Thomas Jr CA,MacHattie LA,Wensink PC (1967) Terminal
repetition in non-permuted T3 and T7 bacteriophage DNA
molecules. J Mol Biol 23: 365–376
Rosa MD (1979) Four T7 RNA polymerase promoters contain an
identical 23 bp sequence. Cell 16: 815–825
Saito H, Tabor S, Tamanoi F, Richardson CC (1980) Nucleotide
sequence of the primary origin of bacteriophage T7 DNA replica-
tion: relationship to adjacent genes and regulatory elements. Proc
Natl Acad Sci USA 77: 3917–3921
Smith HO, Hutchison III CA, Pfannkoch C, Venter JC (2003) Generating
a synthetic genome by whole genome assembly: phiX174
bacteriophage from synthetic oligonucleotides. Proc Natl Acad Sci
USA 100: 15440–15445
Spector L, Barnum H, Bernstein HJ, Swamy N (1999) Finding a better-
than-classical quantum AND/OR algorithm using genetic
programming. In IEEE Proceedings of the 1999 Congress on
Evolutionary Computation, pp 2239–2246. Piscataway, NJ:
IEEE Press. For additional examples, see http://www.genetic-
programming.com/humancompetitive.html
Stahl SJ, Chamberlin MJ (1977) An expanded transcriptional map of
T7 bacteriophage. Reading of minor T7 promoter sites in vitro by
Escherichia coli RNA polymerase. J Mol Biol 112: 577–601
Stemmer WP, Crameri A, Ha KD, Brennan TM, Heyneker HL (1995)
Single-step assembly of a gene and entire plasmid from large
numbers of oligodeoxyribonucleotides. Gene 164: 49–53
Studier FW (1969) The genetics and physiology of bacteriophage T7.
Virology 39: 562–574
Studier FW (1972) Bacteriophage T7. Science 176: 367–376
Studier FW (1973a) Genetic analysis of non-essential bacteriophage T7
genes. J Mol Biol 79: 227–236
Studier FW (1973b) Analysis of bacteriophage T7 early RNAs and
proteins on slab gels. J Mol Biol 79: 237–248
Studier FW (1979) Relationships among different strains of T7 and
among T7-related bacteriophages. Virology 95: 70–84
Studier FW (1981) Identification and mapping of five new genes in
bacteriophage T7. J Mol Biol 153: 493–502
Studier FW, Dunn JJ (1983) Organization and expression of
bacteriophage T7 DNA. Cold Spring Harb Symp Quant Biol 47
(Part 2): 999–1007
Studier FW, Maizel JV (1969) T7-directed protein synthesis. Virology
39: 575–586
Studier FW, Rosenberg AH (1981) Genetic and physical mapping of the
late region of bacteriophage T7 DNA by use of cloned fragments of
T7 DNA. J Mol Biol 153: 503–525
Studier FW, Rosenberg AH, Simon MN, Dunn JJ (1979) Genetic and
physical mapping in the early region of bacteriophage T7 DNA.
J Mol Biol 135: 917–937
Summers WC, Brunovskis I, Hyman RW (1973) The process of
infection with coliphage T7. VII. Characterization and mapping of
the major in vivo transcription products of the early region. J Mol
Biol 74: 291–300
Tamanoi F, Saito H, Richardson CC (1980) Physical mapping
of primary and secondary origins of bacteriophage T7 DNA
replication. Proc Natl Acad Sci USA 77: 2656–2660
Tian J, Gong H, Sheng N, Zhou X, Gulari E, Gao X, Church G (2004)
Accurate multiplex gene synthesis from programmable DNA
microchips. Nature 432: 1050–1054
Wolfson J, Dressler D, Magazin M (1972) Bacteriophage T7 DNA
replication: a linear replicating intermediate (gradient centri-
fugation-electron microscopy-E. coli-DNA partial denaturation).
Proc Natl Acad Sci USA 69: 499–504
You L, Suthers PF, Yin J (2002) Effects of Escherichia coli physiology on
growth of phage T7 in vivo and in silico. J Bacteriol 184: 1888–1894
You L, Yin J (2002) Dependence of epistasis on environment and
mutation severity as revealed by in silico mutagenesis of phage T7.
Genetics 160: 1273–1281
Yount B, Curtis KM, Baric RS (2000) Strategy for systematic assembly
of large RNA and DNA genomes: transmissible gastroenteritis virus
model. J Virol 74: 10600–10611
Zavriev SK, Shemyakin MF (1982) RNA polymerase-dependent
mechanism for the stepwise T7 phage DNA transport from the
virion into E. coli. Nucleic Acids Res 10: 1635–1652
Zhang X, Studier FW (2004) Multiple roles of T7 RNA polymerase and
T7 lysozyme during bacteriophage T7 infection. J Mol Biol 340:
707–730
Refactoring bacteriophage T7
LY Chan et al
10 Molecular Systems Biology 2005 & 2005 EMBO and Nature Publishing Group