Targeted Destruction of Intracellular DNA Using a CRISPR-Based Genetic Device
that can be Carried Indefinitely in the Host Genome
ARGIVES
By
MASSACHUSETTS INSTITUTE
OF TECHNOLOLGY
Brian James Caliando
JUN 3 0 2015
M.S., Stanford University (2008)
B.A., Stanford University (2006)
LIBRARIES
Submitted to the Department of Biological Engineering
in Partial Fulfillment of the Requirements for the Degree of
Doctor of Philosophy in Biological Engineering
At the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
June 2015
@ 2015 Massachusetts Institute of Technology. All Rights Reserved.
Signature redacted
A u th o r............................................................................................
Brian J. Caliando
Department of Biological Engineering
February 23, 2015
Certified by.............................................................................
.............
Signature redacted
Christopher A. Voigt
Professor of Biological Engineering
Thesis Advisor
Signature redacted
Approved by...........................................................................
Forest M. White
Associate Professor of Biological Engineering
Graduate Program Chair, Department of Biological Engineering
This doctoral thesis has been examined by a committee of the Department of Biological
Engineering as follows:
Certified by
Signature redacted
.................................................. .. 9Fe .ihn
Feang
Assistant Professor of Biological Engineering
W.M. Keck Career Development Assistant Professor of Brain and Cognitive Sciences
TWe-sis Committee Chair
Signature redacted
Certified by
............................................................................................................................................................
Christopher A. Voigt
Professor of Biological Engineering
Thesis Advisor
Signature redacted
Certified by
...........................................................................
Kristala L. J. Prather
Associate Professor of Chemical Engineering
Committee Member
Targeted Destruction of Intracellular DNA Using a CRISPR-Based Genetic
Device that can be Carried Indefinitely in the Host Genome
by
Brian James Caliando
Submitted to the Department of Biological Engineering on February 23th, 2015
in Partial Fulfillment of the Requirements for the Degree of Doctor of
Philosophy in Biological Engineering
Abstract
Environmental release of synthetic DNA resulting from the disposal of spent microbial biocatalyst potentially
represents an ecological risk to the environment or a financial risk to biotechnology firms, who might have their
intellectual property stolen as a consequence. Thus, a genetically-encoded device that is capable of degrading
DNA in a controlled manner would be a valuable and enabling tool. To that end, we have constructed a modular,
switchable, genetically-encoded E. coli device for the controlled destruction of user-specified DNA targets in vivo
that is based on the organism's native type-IE CRISPR (Clustered Regularly Interspaced Short Palindromic Repeats)
DNA interference (DNAi) pathway. The optimized DNAi device is comprised of two components: a chromosomallyintegrated actuator element, which encodes the minimal set of CRISPR-associated (cas) genes required for DNAi
activity, and a reprogrammable targeting plasmid, which encodes the CRISPR array specifying the DNA target(s).
The device is stable in the OFF state, with >98% of cells retaining a low-copy DNA target over the course of an 8-hr
Upon DNAi activation, the target plasmid is lost from all but ~1 in 108 cells and there is a
corresponding >10,000-fold decrease in the abundance of the target DNA sequence as recovered by PCR. When
the device is targeted to the host genome instead of a plasmid, activation also results in the self-destruction of the
experiment.
host, killing all but -1 in 10 8 of cells in the ON state but with no appreciable effect on cell viability in the OFF state.
Further characterization has also revealed that when DNAi activity is maintained in the OFF state, the overall
maintenance cost to the host strain is exceedingly low; the device remains functionally stable over hundreds of cell
generations in continuous culture, has little-to-no impact on host growth or plasmid stability, and doesn't interfere
with ectopic over-expression of other proteins. The DNAi device is therefore a powerful tool that can potentially
be combined with other genetically engineered systems to create safer and more secure forms of biotechnology.
Thesis Supervisor: Christopher A. Voigt
Title: Professor of Biological Engineering
5
6
Acknowledgements
First and foremost, I would like to thank my advisor, Chris Voigt, for his mentorship and support
and for his seemingly unshakable trust in my abilities. I am also very grateful to my committee members
Feng Zhang and Kristala Prather for their advice and guidance over these past years.
As a member of Voigt Lab, I have been unfathomably lucky to work with some of the most
intelligent, capable, and all-around excellent human beings every single day. We've shared scientific
ideas, lab memes, countless cups of coffee, and an embarrassingly large quantity of alcohol, and in doing
so we've become irreplaceable parts of each other's lives. As such, I am grateful to them all. In particular,
from my time at UCSF, I'd like to give special thanks Dan Widmaier and Ethan Mirsky, who both were
instrumental in showing me the ropes when I first came to Voigt Lab as a rotation student so long ago.
Without their help, I'd probably still be trying to use MEGAWHOP to build my constructs on Chris' advice.
From my time at MIT, I am especially grateful to Jenn Brophy, Miryoung Song, and Brynne Stanton. When
things looked bleakest, I could always trust them to help catch me before I slipped down the rabbit hole
into despair. In addition, I have been especially impressed by the undergraduates that I have encountered
since arriving at MIT, and my UROP, Rui Wang, represents the very best among them. Her intelligence,
curiosity, kindness, and dedication have made me proud to be her mentor.
I would also like to reserve a special thanks for the lab managers and administrators, both past
and present, who have kept Voigt Lab whirring like a well-oiled machine. At UCSF, our lab manager Rena
Hill not only knew how to keep our wayward bunch in line, but also was a pleasure to share a bay with.
Her keen sense of humor was rivaled only by her professionalism. At MIT, Barbara Karampalas and Terry
King have risen to meet every single challenge associated with a lab that has nearly doubled in size in just
a few short years. Their dedication, commitment, and understanding are truly unmatched, and the lab
would be completely adrift without their watchful guidance.
I would also like to thank John and Ellen Essigmann and the rest of the Simmons Hall GRTs and
staff for welcoming me into their dorm community as a member of the family. Serving as a GRT afforded
me the opportunity to give some of myself back to MIT, and doing so kept me more firmly grounded when
things seemed overwhelming. I wouldn't trade the experience,.or the friends I made there, for anything.
In addition, I owe an extra special debt of thanks to Ruby Yu. As my companion during what were
probably the most difficult periods of my time at MIT, she stood by me with more patience, kindness, and
caring than I probably deserved. She will forever be one of my closest friends.
To my parents, Catherine and Jim, I wish to express my deepest love and gratitude for all of their
emotional support, especially during this last home stretch of my scholastic journey. I am also equally
grateful to my brother Michael, his wife Kristina, and the rest of my extended family for their collective
belief in me and for their constant encouragement.
I would also like to acknowledge the numerous funding sources that made this work possible: the
NSF Synthetic Biology Engineering Resource Center (SynBERC), the National Institutes of Health (NIH), the
Defense Advanced Research Projects Agency (DARPA), the UCSF TETRAD Program, and the MIT
Bioengineering department.
7
A b stra ct .........................................................................................................................................................
Acknow ledgem ents.......................................................................................................................................
Table of Contents ..........................................................................................................................................
List of Figures ..............................................................................................................................................
List of Tables ...............................................................................................................................................
Perm issions .................................................................................................................................................
5
7
8
10
11
12
1.
Introduction..........................................................................................................................................
13
2.
M aterials and M ethods .......................................................................................................................
2.1. Strains, Plasm ids, and M edia.....................................................................................................
2.2. Strain Construction ......................................................................................................................
17
17
17
2.2.1. FLP-Mediated Integration ................................................................................................
2.2.2. IntS-M ediated Integration ................................................................................................
2.3. Plasm id Stability Assay ("Electroporation-based," Am picilin).................................................
2.4. Preparation of Phagem id Virion Stock Solutions......................................................................
2.5. Phagem id Virion Transduction Blocking Assay.........................................................................
2.6. Plasm id Knockout Assay ..........................................................................................................
2.6.1."Plate-based," Spectinom ycin ...........................................................................................
2.6.2. "Cytom etry-based," Red Fluorescent Protein .................................................................
2.7. Plasm id Knockout Kinetic Assay (PAM Experim ents)...............................................................
2.8. Survivor Analysis Assay (Plasm id Knockout Experim ent) ..........................................................
2.9. Isolation of Target Plasm id DNA ................................................................................................
2.10.
Isolation of Target Chrom osom al DNA .......................................................................
2.11.
Quantitative PCR ...............................................................................................................
2.12.
M easuring the Im pact on Cell Growth ..........................................................................
2.12.1. "Plate-based," CFU/m L ................................................................................................
2.12.2. "OD600-based", Doubling Tim e.....................................................................................
2.12.3. "Cuvette-based", Cell Density.......................................................................................
2.13.
Evolutionary Stability Experim ents................................................................................
2.14.
Cell Death Assays (DNAi-M ediated Killing)...................................................................
2.15.
Survivor Analysis Assay (Cell Death Experim ent)..........................................................
2.16.
Cell Death Assays (Integrase-M ediated Killing) ............................................................
17
18
3.
19
20
20
21
21
22
22
23
23
24
24
25
25
25
26
26
27
27
28
Construction and Optimization of a CRISPR DNA-interference device for Escherichia coli.............. 29
3.1. Device Design ...............................................................................................................................
29
3.2. Initial Characterization of Unstable DNAi Devices ...................................................................
34
3.3. Characterization of Stable DNAi Devices Harboring Chromosomal Actuators......................... 37
3.4. Param eterization and Optim ization of DNAi Device Kinetics................................................... 41
3.4.1. Introduction .........................................................................................................................
41
3.4.2. cRNA Expression Level.....................................................................................................
42
3.4.3. Spacer Sequence ..................................................................................................................
43
3.4.4. AraC Expression Levels.....................................................................................................
8
45
3.4.5. Proto-spacer Adjacent Motif (PAM ) ...............................................................................
3.4.6. Actuator Copy Number .....................................................................................................
3.5. Optimization of DNAi Device Knockout Efficiency ...................................................................
3 .5 .1 . Intro d u ctio n .........................................................................................................................
3.5.2. Design Parameters Affecting Plasmid Knockout Efficiency ............................................
3.5.3. Analysis of Plasmid Knockout Survivors...........................................................................
3.5.4. Design Principles for Optimizing Plasmid Knockout Efficiency........................................
47
51
53
53
54
55
57
4.
Characterization of the Optimized CRISPR DNAi Device .................................................................
4.1. Scope of DNAi Device Plasmid Knockout Functionality...........................................................
4.1.1. Effect of Target Copy Number on DNAi Knockout Efficiency and Kinetics......................
4.1.2. DNAi-Mediated Depletion of Target Plasmid DNA .........................................................
4.2. Impact of the DNAi Device on Host Growth and Genetic Stability .........................................
4.2.1. Effect of DNAi Activity on Host Growth..........................................................................
4.2.2. Long-term Stability of DNAi Device Function .................................................................
4.2.3. Long-term Stability of Host Metabolism while Carrying the DNAi Device .......................
4.3. Targeting of DNAi Activity to the Host Genome......................................................................
4.3.1. Using the DNAi Device to Effect Cell Death .....................................................................
4.3.2. Analysis of Cell Death Assay Survivors.............................................................................
4.3.3. DNAi-Mediated Depletion of Chromosomal DNA .........................................................
4.3.4. Orthogonal Cell Death Mechanisms Increase Killing Efficiency .......................................
59
59
59
60
61
61
62
63
64
64
67
69
70
5.
Conclusions and Discussion..................................................................................................................
73
A. Supporting Information........................................................................................................................83
A.1 Supplementary Figures.......................................................................................................
83
93
A.2 Supplementary Tables .........................................................................................................
B ib liog ra p h y ..............................................................................................................................................
9
1 01
List of Figures
Figure 1.1: Biochemical mechanism of CRISPR-mediated DNA interference (DNAi) .............................
Figure
Figure
Figure
Figure
Figure
16
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
3.1: Schem atic of the DNAi Device ............................................................................................
29
3.2: Effect of cas over-expression on cell growth ........................................................................
30
3.3: Genetic instability of the pWUR cas over-expression system ..............................................
31
3.4: Schematic of M13-based phagemid transduction blocking assay....................................... 35
3.5: Comparison of phagemid transduction efficiencies for DNAi devices with different actuator
co n fig u ratio n s ...............................................................................................................................
36
3.6: Plasmid knockout efficiencies with the DNAi actuator is carried on different plasmid
backbones and genom ic configurations ...................................................................................
39
3.7: Comparison of plasmid knockout kinetics in early genomic actuator prototypes ............... 40
3.8: Effects of crRNA expression level on target plasmid knockout kinetics .............................. 42
3.9: Effect of spacer sequence on target plasmid knockout kinetics .........................................
44
3.10: Optimization of araC expression levels with respect to target plasmid knockout kinetics ... 46
3.11: Effect of PAM sequence on phagemid transduction efficiency .........................................
48
3.12: Effect of PAM sequence on the kinetics of plasmid knockout ...........................................
49
3.13: Effect of actuator copy number on the kinetics of target plasmid knockout .................... 52
3.14: Systematic optimization of target plasmid knockout efficiency ........................................
54
3.15: Characterization of DNAi device functionality and escape mutants .................................
56
Figure
Figure
Figure
Figure
Figure
Figure
Figure
Figure
4.1: Plasmid knockout kinetics and target DNA depletion using the optimized DNAi device ........ 60
4.2: Growth impact of strains carrying the DNAi device ............................................................
61
4.3: Long term plasmid knockout performance of the DNAi device in continuous culture ........ 63
4.4: Stability of target plasmid protein expression during cell passaging experiments..............64
4.5: Cell killing dynamics upon targeting of Cas activity to the genome ....................................
65
4.6: DNAi-mediated cell killing in the presence and absence of antibiotic selection..................66
4.7: DNA recovery as a function of distance from the proto-spacer ..........................................
69
4.8: Combining DNAi-mediated and integrase-mediated cell killing methods in a single host ...... 71
Figure
Figure A.1: Actuator plasm id m aps.........................................................................................................
83
Figure A.2: CRISPR targeting plasmid and basic target plasmid maps...................................................84
Figure A.3: Strain construction plasm id m aps .........................................................................................
85
Figure A.4: Fluorescence m easurement plasm id maps.......................................................................... 86
Figure A.5: Schematic of FLP-mediated actuator genomic integration methodology ........................... 87
Figure A.6: Schematic of Int5-mediated actuator genomic integration methodology .......................... 88
Figure A.7: Schematic of three-plasmid, two-input cas/CRISPR over-expression system ..................... 89
Figure A.8: Characterization of PBAD input ... .....................................
.... .......................................... 90
Figure A.9: Maps of target plasmids used in PAM experiments............................................................ 91
Figure A.10: The dynamics of plasmid loss for each active PAM sequence ..........................................
92
10
List of Tables
Table 3.1: Sum mary of kinetic parameters for active PAM s .................................................................
50
Table 4.1: Characterization of suicide assay survivors...........................................................................
68
Table A.1: CRISPR element sequences...................................................................................................
93
Table A.2: Host chrom osomal modifications.........................................................................................
94
Table A.3: Genom ic modification primers .............................................................................................
95
Table A.4: Int5-specific recom bination sites...........................................................................................
96
Table A.5: qPCR primer sets........................................................................................................................
97
Table A.6: Constitutive promoter sequences ........................................................................................
98
Table A.7: R BS sequences ...........................................................................................................................
99
Table A.8: Classification of DNAi survivor LOF m utations via genetic com plementation ........................
11
100
Portions of the following document are reprinted and/or adapted with permission from Caliando,
B.J. & Voigt, C.A., Stable integration of CRISPR for targeted destruction of plasmid DNA. Nature
Communications (in press). Copyright 2015 Nature Publishing Group, Macmillan Publishers Ltd.
12
1. Introduction
The ability to program cells to eliminate engineered DNA at a defined time point or
change in environments would benefit many applications in biotechnology. For example, after
bio-manufactu ring a chemical, cells could be programmed to degrade their DNA at the end of the
process or when they are removed from a defined medium. This would aid the protection of
sequence information as a trade secret, make it easier to remove DNA contamination from a
product, reduce the cost of biomass disposal, and decrease the amount of DNA in the
environment after an accidental release. There are similar needs for "out of the bioreactor"
applications, such as using engineered cells as living therapeutics (e.g., in the gut microbiome) or
in forming associations with crop plants in a field1 . In these cases, it is impossible to recollect
cells for disposal, so they need to be programmed to degrade their own DNA when they leave a
defined environment or after a defined time period.
Various genetic switches have been developed that induce cell death2 . The actuation is
-
based on toxic proteins that are under tight regulatory control to avoid background expression3
4 and have been combined with synthetic genetic circuits that trigger cell death under a shift in
environmental conditions or change in cell state
. Inducing cell death does not address the
problem of the release of DNA, which persists after cells die or are killed. Indeed, waste streams
from fermenters are rich with recombinant DNA, even when the cells have been inactivated by
heat, pH, and antibiotics' 0 and, in fact, these methods of rapid cell death exacerbate the release
of extracellular DNA". The waste biomass of engineered microbes is often used as agricultural
.
fertilizer (e.g., NovoGro TM), and this has been shown to contain significant amounts of DNA1 0
Furthermore, DNA molecules are stable and, once introduced, plasmid and genomic DNA can be
recovered from environmental samples via PCR for 1-5 monthsl1 - 1 3 and is likely to be detectible
longer with advances in deep sequencing1 4 . In one study, 35% of plasmid DNA molecules that
were exposed to the extreme heat and pressures of atmospheric re-entry on the surface of a
rocket still retained their biological function's
The ease of recovering DNA from environmental samples poses a challenge for the
biotech industry.
First, the value and competitive advantage of a company is largely in its
engineered DNA constructs and strains and these involve large financial investments and
13
development time. Protecting this information as a trade secret is nearly impossible when
samples can be easily recovered from a waste stream (or as a contaminant in a consumer
product), sequenced, and then rebuilt using chemical DNA synthesis. Therefore, the copying of
the products of genetic engineering no longer requires the transfer or theft of intact DNA
molecules or living cells. In the future, it may be possible to access complete organisms through
the synthesis and transplantation of entire genomes16-2 2
Once DNA is stably introduced into cells, removing it is tedious and difficult to scale to an
industrial process. The traditional way to remove plasmids is to "cure" them by culturing on nonselecting media over multiple generations
2 3 24
, . Another method is to express non-specific DNA
nucleases, which efficiently cause cell death (1:105)8, but their non-specificity can severely inhibit
cell growth even when maintained in the uninduced state 2 . DNA can also be removed via
downstream processing of the fermentation product or waste streams. This is necessary for the
production of therapeutic proteins and DNA (for gene therapy and vaccines), where the US Food
.
and Drug Administration (FDA) enforces allowable limits for contaminant DNA (10 ng/dose) 25 - 27
This either requires high-performance purification methods and/or removal of DNA specifically
via ion exchange columns or membrane filtration, all of which are expensive and difficult to
.
implement 25 , 27,28
We have constructed a genetic device whose input is a transcriptional signal and whose
output is the targeted degradation of DNA. Defining a transcriptional input enables it to be
connected to the output of inducible systems, environmental sensors, or genetic circuits 29 . The
actuation is based on the RNA guide-directed DNAse machinery from Type-I CRISPR defense
systems [reviewed in
30,31]
(Figure 1.1). CRISPR is an "immune system" for prokaryotes that
prevents transduction by phage and conjugative plasmids by targeting their DNA for
degradation30-33. The targeted DNA sequences are specified by the CRISPR array, which is a series
of ~30-40 bp spacers separated by short palindromic repeats30-33. The array is transcribed as a
pre-crRNA and is processed into shorter crRNAs that associate with the Cas protein complex to
target complementary DNA sequences known as proto-spacers 34--46. These proto-spacer targets
must also have an additional neighboring sequence known as proto-spacer adjacent motif (PAM)
that is required for target recognition 4 5,4 7- 51. After binding, a Cas protein complex serves as a DNA
14
endonuclease to cut both strands at the target 4 4 ,4s and subsequent DNA degradation occurs via
exonuclease activity4 4 . The RNA-guiding capability of the Cas protein(s) has been widely applied
to problems in biotechnology, including genome editing and synthetic regulations2 -s.
We have built a device that stably maintains all of the necessary components for DNAimediated nuclease activity and, only when induced, functions to degrade target DNA without
affecting non-targeted DNA. This required engineering the device to retain a large dynamic range
when induced, yet have almost no basal activity when carried in the host genome in the
uninduced state. To be useful, the device must not impose a growth burden on the host, remain
stable and active over many generations, and not reduce target DNA prior to induction (e.g.,
lower plasmid copy number).
15
Pa3
PcosA
PCRSpR
Spacer
Repeat
(32bp)
(29bp)
CRISPR
cos genes
Cas/CRISPR
Pre-crRNA
CasABCDE
("Cascade")
Cas3
(nuclease)
CasE
2 Effector
Complex
4
?
Target
Recognition
-- --
crRNA guides
------
Target DNA
DNA Scan 3
(1-D)
Proto-spacer
PAM
Cas3-Mediated
Complementary
Strand
(neg. super-coiled)
5'-CCG
'CGcrRNA
Spacer
Proto spacer
-PAM
srpand
Target Recognition R-Loop
6
7
Cas3-Medtated
Resection
Target Cut
Recycled (?)
Figure 1.1 Biochemical mechanism of CRISPR-mediated DNA interference (DNAi). (1) Cas3, CasABCDE, and the precrRNA are expressed from the CRISPR locus (casi and cas2, which are immediately downstream from casE, are omitted
for clarity). CasABCDE associates to form a protein complex ('Cascade'), and CasE processes the pre-crRNA into singlespacer crRNA units. (2) Cascade binds a single crRNA to form an effector complex, which then binds non-specifically to
negatively supercoiled DNA. (3) The effector complex searches the DNA molecule for its cognate proto-spacer by means
of one-dimensional facilitated diffusion. (4) Upon encountering a matching proto-spacer with a valid PAM sequence, the
dsDNA is melted and the crRNA base pairs with the complementary DNA strand forming an R-loop. (See boxed inset for
detail.) (5) This recognition event transiently recruits Cas3, which makes a nick in the displaced strand via ssDNA
endonuclease activity. It then begins an initial resection of the exposed ssDNA ends by means of its 3'
exonuclease/helicase activity. (6) The crRNA-complimentary strand is eventually exposed, and Cas3 makes a second
endonucleolytic cut creating a double strand break in the target. It is unknown whether the effector complex is recycled
following target cleavage. (7) Resection of the target continues in part via Cas3-mediated exonuclease activity until the
target is repaired or degrades completely.
16
2.
Materials and Methods
2.1
Strains, plasmids, and media.
All strains were derived from a triple-knockout E. coli MG1655 A(araC-araBAD) APlacl:Lacl
A(cas3-CRISPR) parent (Table A.2). Standard plasmids were cloned and propagated using an E.
coli Mach1-T1R host (F-
p80(/acZ)AM15
A/acX74 hsdR(rK-mK+) ArecA1398 endAl tonA) (Life
Technologies), while plasmids containing an R6Ky origin of replication were cloned and
propagated using an E. coli TransforMax EC100D pir-116 host (F mcrA A(mrr-hsdRMS-mcrBC)
qp8OdlacZAM15 AacX74 recAl endAl araD139 A(ara, leu)7697 galU galK A- rpsL (StrR) nupG pir116(DHFR)) (Epicenter). Actuator plasmids were constructed using a pOR10 backbone60 and
.
encode cas components amplified from plasmids pWUR397 (cas3) and pWUR400 (casABCDE)34
All target plasmids and all CRISPR targeting plasmids were derived from a pUC119 backbone".
Cells were plated on LB (LB Miller Medium, Difco, #244620) supplemented with 1.5% (w/v) agar
(Bacto Agar, Difco, #214010), and liquid cultures were grown in 2YT media (2xYT (Yeast Extract
Tryptone) Medium, Difco #244020) for all experiments. Ampicillin (100 tg/ml; sodium ampicillin,
Gold Biotechnology #A-301-25), kanamycin (50 pg/ml; kanamycin sulfate, Gold Biotechnology
#K-120-10), spectinomycin (100 ig/ml; spectinomycin dihydrochloride, Gold Biotechnology #S140-5), and/or chloramphenicol (35 pg/ml; USB #23600-25G) were used as appropriate.
L-
arabinose (Ara; Sigma-Aldrich #A3256-100G) or glucose (Glc; BDH #BHD8005-500G) was used as
inducer or repressor, respectively.
2.2
2.2.1
Strain construction
FLP-mediated integration
A genomic knockout of the CRISPR-cas locus (Table A.2) was performed in a wild-type
MG1655 background via the XRED method using a linear kanamycin-resistance cassette amplified
from pKD136 1 (Table A.3). The FRT-flanked kanamycin marker was then removed via FLP
recombinase (pCP20) 9 1 leaving a single FRT site scar sequence. In the process, the temperature-
sensitive pCP20 was cured from the cell.
The resulting marker free strain was then re-
17
transformed with pCP20, FLP was expressed by brief thermal induction at 37*C, and the resulting
cells were prepared as electrocompetent and transformed with the ~11kb DNAi actuator
integrating plasmid pACT-A (Figure A.1). FLP-mediated recombination between the plasmidborne FRT site and the chromosomal FRT site then creates a chromosomally-integrated actuator
sequence within the donor strain (Figure A.5). Recombinants were selected by growth on LB
plates containing kanamycin at 37*C following a 3-hr recovery in 2YT at 37*C.
Successful
chromosomal integration at the targeted locus was confirmed with colony PCR, and loss of the
temperature-sensitive pCP20 expression plasmid was confirmed by Amps phenotype. Using this
modified strain as a transduction donor, the antibiotic marker-linked integration locus was then
transferred to the final triple-knockout recipient host (Table A.2) via PMir viral transductions 6 3 to
yield the 1xA DNAi device strain. The KanR marker and R6K origin could not be removed. A
schematic depicting the sequences of manipulations is given in Figure A.5.
2.2.2
Int5-mediated integration
Genomic knock-outs were performed via the XRED method using a linear kanamycin-
resistance cassette amplified from pKD13 62 (Table A.3). Knock-ins of the ~11kb DNAi actuator
&
plasmids (pACT-B, & pACT-C, Figure A.1) or their fluorescence measurement plasmids (pGFP-B
pGFP-C, Figure A.4) were achieved by adapting high-efficiency site-specific chromosomal
integrations 6 4 . First, the Streptomyces phage PhiK38-1 attB site65 (Figure A.3, Table A.4) was
inserted into the host chromosome via the ARED method. Next, the linked KanR marker was
removed via pCP20 62, and the resulting attB+ Kans host was transformed with a temperaturesensitive plasmid (plnt5ts, AmpR) expressing the cognate phage integrase Int5 under the control
of a PBAD promoter 65 (Figure A.3). Transformants carrying the integrase plasmid were grown in
2YT containing ampicillin and 1 mM arabinose in a shaking incubator at 30"C and 250 rpm to an
OD6oo "0.5, at which point electrocompetent cells were prepared. The resulting Int5-expressing
cells were transformed with an integrative actuator plasmid (pACT-B or pACT-C, KanR, nonreplicative R6Ky origin) containing the PhiK38-1 attP site via electroporation and subsequently
recovered in 2YT at 37"C for 3 hrs. Recombinants were selected for by growth on LB plates
containing kanamycin at 37"C. Successful chromosomal integration at the targeted locus was
18
confirmed with colony PCR, and loss of the temperature-sensitive Int5 expression plasmid was
confirmed by Amps phenotype. For each modified genomic locus, a corresponding donor strain
in a wild-type MG1655 background was created. These antibiotic marker-linked loci were then
serially transferred to the final recipient host via Pivir viral transductions". Finally, the
genomically-integrated FRT-flanked R6Ky origin and kanamycin resistance marker was removed
by expression of FLP recombinase following transformation with temperature-sensitive plasmid
pCP20 (AmpR, CmR) 6 2 . Loss of the FRT-flanked region and curing of pCP20 was subsequently
confirmed by colony PCR and the presence of Kans, Amps, and Cms phenotypes. The 1XB and lxc
DNAi devices contain either the pACT-B, or the pACT-C actuator chromosomally integrated at the
native CRISPR-cas locus, respectively. The
3
XB
and 3Xc DNAi device both contain the pACT-B
actuator integrated at both the araC-araBAD and the CRISPR-cas loci, but the
3
XB
device contains
the pACT-B actuator chromosomally integrated at the Lad locus, while the 3xc device has the
pACT-C actuator integrated there instead (Tables A.2 and A.3). The 3x GFP strain was constructed
identically, except with pGFP-B or pGFP-C chromosomally integrated in lieu of pACT-B or pACTC, respectively. Each copy of the actuator or GFP plasmid is associated with its own copy of the
araCgene, which is under control of either its native Pcpromoter (pACT-C) or a constitutive
PJ23117
promoter (pACT-A, and pACT-B). A schematic depicting the sequences of manipulations is given
in Figure A.6.
2.3
Plasmid stability assay ("electroporation-based," ampicillin).
Host cells containing a DNAi device (pWUR397/pWUR400, KanR, StrR) and an RFP' target
plasmid (AmpR) were prepared as electrocompetent aliquots and then frozen at -80C until
needed. As required, frozen aliquots were thawed and transformed with 100 ng CRISPR targeting
plasmid (CmR) via electroporation, and then recovered for 1 hr at 37*C without shaking. The
fraction of host cells retaining the target plasmid (Target') was calculated from the ratio of colony
forming units obtained from plating samples onto both target-selective (+ampicillin) and nonselective (-ampicillin) LB plates containing all other appropriate antibiotics to select for all
actuator and CRISPR targeting components.
19
2.4
Preparation of phagemid virion stock solutions.
The viral origin of replication (bp 5505-5811) was deleted from M13KO77 4 to create
M13Aori (KanR, p15A), a packaging-defective viral helper plasmid that enables the packaging and
secretion of phagemids supplied in trans but is incapable of infectious self-propagation. MG1655
(F-) cells were contransformed with M13Aori and either a pPAM-TC (RFP-, StrR) or a pPAM-NNN
(64 PAM library, RFP-, StrR) target plasmid. Single colonies were selected by growth on kanamycin
and spectinomycin, and were used to inoculate 2YT containing the appropriate antibiotics.
Samples were grown for 24 hr at 370 C at 250 rpm. The cultures were centrifuged at 4,000 rpm
at 4 0 C, and the virion-containing supernatant was washed with chloroform, centrifuged again at
4,000 rpm, and the virion-containing aqueous layer was extracted. High levels of RFP expression
from pPAM-NNN-RFP in combination with M13 viral protein expression are lethal to the host,
and so these plasmids could not be similarly packaged in this manner.
2.5
Phagemid virion transduction blocking assay.
Host receiver cells (MG1655 F') were co-transformed with a p15A-based actuator plasmid
(pACT-02, KanR) and either an on-target (pCR-X, CmR) or off-target (pCR-N2, CmR) CRISPR
targeting plasmid. Single colonies were selected by growth on kanamycin and chloramphenicol
plates and were used to inoculate 2YT containing the appropriate antibiotics. Samples were
grown overnight in a shaking incubator at 37'C and 250 rpm, back-diluted to OD 6oo ~0.05 into
fresh 2YT containing antibiotics, induced with 2mM Ara, and then grown for an additional ~3-4 h
at 37*C and 250 rpm until OD 600 0.3-0.7. Cultures were concentrated 50-fold and split into 10 gl
aliquots. To each induced aliquot, 10 pi of target plasmid phagemid (pPAM-NNN, StrR) stock
solution (10-fold dilution, Methods) was added. Infections were incubated at room temperature
for 5 min, diluted with 80 p] of additional 2YT, and recovered at 37*C for 1 hr without shaking.
The titer of target-positive transductants (in cfu/gl) was then measured by plating 10-fold serial
dilutions of each infection onto LB containing spectinomycin. The raw relative transduction
efficiency (RTERAW) for each PAM phagemid in the library was then calculated as (# transductants
X strain)/(# transductants
N2
strain). This raw value incorporates both CRISPR-dependent and
CRISPR-independent differences in transduction efficiency between the on-target (X) and off20
target (N 2 ) strains.
Therefore, in order to control for any possible CRISPR-independent
differences in transduction, infection with a control phagemid (pPAM-TC), which entirely lacked
proto-spacer X, was similarly quantified for both strains.
The final, corrected relative
transduction efficiency (RTE) was then calculated as RTE = K x RTERAW, where K is the CRISPRindependent correction factor equal to (# transductants N 2 strain)Tc/(# transductants X strain)Tc
for the pPAM-TC transductions.
Plasmid knockout assays
2.6
2.6.1
"Plate-based," spectinomycin
Host cells containing a DNAi device and an RFP+ target plasmid (StrR) were transformed
with a CRISPR targeting plasmid (CmR), plated onto LB containing all appropriate antibiotics in
addition to 0.5% glucose, and grown for 12 hr at 37"C. Single colonies were used to inoculate
2YT (1 ml) containing all appropriate antibiotics and 0.5% glucose, and the resulting liquid
cultures were grown in a shaking incubator for 3 hr at 37"C and 250 rpm until OD 6oo 0.25-0.75.
Cultures were spun down at 15,000 rpm, washed once with fresh 2YT (1 ml), and back-diluted to
OD600 0.01into 2YT (2 ml) containing appropriate antibiotics but without selection for the target
plasmid (-spectinomycin). The diluted culture was then split into two 1 ml samples, and DNAi
activity was either induced by adding 2 mM arabinose or repressed by adding 0.5% glucose.
Cultures were then grown in a shaking incubator for 8 hr at 37*C and 250 rpm and periodically
sampled. The fraction of host cells retaining the target plasmid (Target+) was calculated from
the
ratio of colony forming
units
obtained from
plating onto both
target-selective
(+spectinomycin) and non-selective (-spectinomycin) LB plates containing all other appropriate
antibiotics. In our hands, the plate-based assay was imprecise for measuring samples expected
to contain a high fraction (>50%) of target-positive host cells (i.e., DNAi OFF or off-target CRISPR
samples). A cytometry-based assay was employed for precise quantification of these samples
(see Plasmid knockout assay ("cytometry-based," red fluorescent protein) below).
21
2.6.2
"Cytometry-based," red fluorescent protein
Samples were prepared and induced identically to the plate-based knockout assay
described above, except that for sampling, aliquots were back-diluted 1:1000 into 2YT containing
0.5% Gic and all appropriate antibiotics except spectinomycin, and then outgrown in a shaking
incubator at 37C and 250 rpm for 8-12 hr. This served to arrest further target plasmid loss while
allowing for the dilution of accumulated intracellular RFP in the target-negative cells present.
Following outgrowth, aliquots were diluted into PBS containing 1 mg/ml kanamycin to arrest
further growth and protein synthesis and stored at 4'C.
Cells were analyzed using a BD
Biosciences LSRFortessa. The fraction of cells containing the target plasmid was calculated as
fraction of cells with red fluorescence values greater than a 1000 au cut-off. Data were analyzed
using FlowJo (TreeStar Inc., Ashland, OR), and populations were gated based on forward and side
scatter. For all samples, the gated population contained between 104 and 101 cells.
2.7
Plasmid knockout kinetic assay (PAM experiments).
Samples were prepared as for the plate-based knockout assay, except the sample
volumes were reduced to 500 pI and all cultures were grown in 96-well format in a shaking
incubator at 37*C and 990 rpm. Cultures were sampled and subsequently outgrown as described
for the cytometry-based knockout assay. For targets with strong PAM sequences (AAA, AAC,
AAG, AAT, AGG, ATG, GAG, or TAG), experiments were carried out for 4.5 h with sampling every
10 m beginning at 2 h post-induction. For targets with slow (CAG, GGG, GTG, TAA, or TAG) or
intermediate (ATA or TTG) PAM sequences, experiments were carried out for 8 h with sampling
every 15 m starting at 3 h post-induction. Following outgrowth, aliquots were diluted into PBS
containing 1 mg/mL kanamycin to arrest further growth and protein synthesis and stored at 4*C.
Cells were analyzed using a BD Biosciences LSRFortessa. Data were analyzed using FlowJo
(TreeStar Inc., Ashland, OR), and populations were gated based on forward and side scatter. For
all samples, the gated population contained between 104 and 106 cells.
22
2.8
Survivor analysis assay (plasmid knockout experiment).
Surviving StrR colonies taken from the +spectinomycin plates resulting from a plate-based
knockout assay (see 2.6.1 above) were selected to inoculate 5-ml cultures of 2YT containing 0.5%
glucose and appropriate antibiotics for target selection (+spectinomycin). Cultures were grown
at 37*C with shaking at 250 rpm until OD600 0.3-0.8, and then prepared as chemically competent
cells using a Mix & Go E. coli Transformation Kit (Zymogen) in accordance with the manufacturer's
instructions. The resulting 500 pl stock of competent cells from each surviving clone was split
into 5x100 pl aliquots in a 96-well format, and each aliquot was then transformed with either an
additional pCR targeting plasmid complement (pCR-X, -Y, -Z, or -N, pUC19, CmR) or an actuator
plasmid complement (pACT-02,
p15A, KanR) in accordance with the kit manufacturer's
instructions. The transformant cell mixtures were recovered in 1 ml 2YT without antibiotics for
1 hr at 37*C and 990 rpm and then back-diluted 1:200 into 1 ml fresh 2YT containing antibiotics
necessary for pCR/pACT-02 selection (+chloramphenicol or +kanamycin).
The resulting
polyclonal mixtures of transformants were grown overnight at 37 0 C and 990 rpm, back-diluted
1:1000 (OD -0.01), and subjected to an additional 8 hr of DNAi inducing conditions (2YT with
2mM Ara, -spectinomycin), at which point the fraction of target' cells was measured in
accordance with standard cytometry-based protocols (Methods). The particular loss-of-function
mutation class was then assigned based upon which complementation plasmids did or did not
restore a surviving clone's DNAi activity (90% target plasmid retention cutoff) using Table A.8.
2.9
Isolation of target plasmid DNA.
Following a plasmid knockout experiment 500 pL of the resulting culture was collected
and spun down at 15,000 rpm, and the supernatant was removed and discarded. Any resulting
pellets were stored at -20*C until isolation could be performed in parallel. For plasmid knockout
samples, plasmid DNA extraction was performed using a miniprep kit (Qiagen) and all samples
were normalized to a final volume of 100 pL of Buffer EB (Qiagen) for subsequent qPCR analysis.
Based on experiments in which comparable samples of target-negative cells were doped with
known amounts of gel-purified target plasmid (pTAR(S)) and then subjected to analogous DNA
extraction techniques, the efficiency of miniprep isolation was estimated at 35
23
18 %.
2.10
Isolation of target chromosomal DNA.
A 250 il aliquot of the culture was collected and spun down at 15,000 rpm, and the
supernatant was removed and discarded. Any resulting pellets were stored at -20C until
isolation could be performed in parallel. Total genomic DNA was extracted using a Wizard
Genomic DNA Purification Kit (Promega) in accordance with the manufacturer's protocol, and all
samples were resuspended in a final volume of 100 p.L of TE buffer (Promega) for subsequent
qPCR analysis. The efficiency of total genomic DNA isolation was calculated to be 14
3.0 % by
generating qPCR standard curves from known quantities of target DNA and assuming one copy
of the genome per cfu and 3 x 109 cfu per 250 il sample.
Instead, we were forced to use alternative methods to quantify DNAi activity in which at
least one of the three essential components, actuator, CRISPR array, or target DNA, were
genetically absent from
2.11 Quantitative PCR.
Quantitative PCR reactions (10 liL total volume) containing 1 pL of either sample or
standard DNA template were prepared from SsoFast EvaGreen Supermix with Low ROX (Bio-Rad)
in accordance with the manufacturer's specifications. The samples were amplified and measured
using a Mastercycler RealPlex2 (Eppendorf) real-time PCR thermocycler.
The thermocycling
protocol (40 cycles) was as follows: (1) Initial denaturation - 2:00 at 95*C; (2) Melting - 0:03 at
95*C; (3) Annealing/synthesis: 0:30 at 65*C.
Primers and amplicons are described in
Supplementary Table 5, and all primers amplified their targets with 100
5% efficiency. For each
amplicon, a series of 5-fold serial dilutions was prepared from a 1 ng/pi stock solution of gelpurified target DNA. Each of the eight dilutions spanning 5- to 5-8 ng/pL was then subjected to
measurement by qPCR in triplicate, and the resulting Ct values were used to prepare a calibration
curve of Ct vs. relative copy number (RCN), for which the most dilute solution (5-8 ng/pL) was
arbitrarily designated as RCN = 1. When absolute quantification was required, a sample's RCN
value was determined using the calibration curve and subsequently converted to an absolute
copy number (ACNsample; given in copies/pl sample) value using the formula ACN = RCN x (6.02 x
24
1023 copies/mol) x (5-8 ng/pW sample)/M, where M is the molecular weight of the specific dsDNA
amplicon (in ng/mol, Supplementary Table 5). The isolated DNA samples loaded into the qPCR
reactions were more concentrated relative to the initial cultures from which they were obtained,
and so ACNsample was then converted to ACNculture (given in copies/ml culture) according to the
formula
ACNculture =
ACNsample x (1000 pil/ml)/C, where C is equal to the initial volume of culture
sampled divided by the final volume of the isolated DNA sample. For miniprepped DNA (500l
culture into 100 pI EB) C = 5, and for isolated genomic DNA (250pl culture into OOpIl TE) C = 2.5.
2.12
2.12.1
Measuring the impact on cell growth
"Plate-based", cfu/mL
Strains containing variants of the DNAi devices were co-transformed with both a target
plasmid and a corresponding on-target CRISPR plasmid, plated onto LB containing antibiotics in
addition to 0.5% glucose, and grown for 12 hr at 37"C. Single colonies were used to inoculate
2YT (1 ml) containing all appropriate antibiotics and 0.5% Glc, and the resulting liquid cultures
were grown in a shaking incubator for 3 hr at 37 0 C and 250 rpm until OD6oo 0.25-0.75. Cultures
were spun down at 15,000 rpm, washed once with fresh 2YT (1 ml), diluted to OD6oo 0.01 into
2YT (3 ml) containing kanamycin and/or chloramphenicol as appropriate, and then split into 3x1
ml samples in a 96-well format. Each sample was induced with Ara (0 - 10mM) and then grown
in a shaking incubator for 8 hr at 37*C and 250 rpm. The viable cell titer of each sample was then
measured by plating serial dilutions onto non-selective (-spectinomycin) LB plates containing all
other appropriate antibiotics.
2.12.2
"OD 6oo-based", doubling time
Samples are prepared as above in "Measuring the impact on cell growth ('plate-based',
cfu/ml)", except that washed cultures are diluted to OD600 = 0.01 into non-selective (spectinomycin) 2YT (2.4 ml) containing choloramphenicol and then split into 12x200 gI samples
in 96-well format in an optically-clear plate. Each sample is induced with Ara (0 - 10mM) and
then grown in a BioTek Synergy H1 Hybrid Microplate Reader for 4 hr at 37*C and 560 rpm with
concomitant measurement of each sample's absorbance (600nm) every 3 min. The exponential-
25
phase growth rate, p (in min-'), is then determined by least-squares fit of measured OD 6oo values
corresponding to the interval between 60 and 90 min post-induction using the equation OD6oo =
Aoe-9, where t is time in minutes, and Ao is the OD6oo at t = 0. The doubling time, t (in min), is
then calculated as t = ln(2)/p.
2.12.3
"Cuvette-based", cell density
Samples of BL21-Al strains harboring either pWUR397* (KanR), pWUR400* (StrR) or no
plasmid were prepared as above in "Measuring the impact on cell growth ('plate-based', cfu/ml)",
except that washed cultures are diluted to OD600
=
0.02 into 2YT (60 ml) containing the
appropriate antibiotics for plasmid selection and then split into 2x30 ml samples in 250-ml baffled
shake flask format. Each sample was then either induced with IPTG (1 mM) or left uninduced
and then grown for 12 hr at 37*C and 250 rpm with concomitant measurement of each sample's
absorbance (600nm) every 60 min. Cell density values reflect 1-cm path length measurements
of OD60o absorbance using a Varian Cary Bio 50 UV-vis spectrometer.
2.13
Evolutionary stability experiments.
The strain containing the 3xc DNAi device was transformed with the on-target Y+Z dual
CRISPR plasmid (pCR-YZ, CmR) and a pSC101 target plasmid (pTAR(S), RFP+, StrR), plated onto LB
containing chloramphenicol and spectinomycin in addition to 0.5% Glc, and grown for 12 hr at
370 C.
A single colony was used to inoculate 2YT (1 ml) containing chloramphenicol,
spectinomycin, and 0.5% GIc ("cell passage media"), and the resulting liquid culture was grown
for 3 hr in a shaking incubator at 37*C and 250 rpm until OD600 = 0.25-0.75. These cells were
washed with 2YT (1 ml), and then back-diluted to OD60 0 = 10-5 into fresh cell passage media (1 ml)
to mark the start of the experiment ('day 1', t = 0). The culture was grown continuously in a
shaking incubator at 37 0C and 250 rpm with back-dilution to OD 600 = 10-5 (75,000- to 100,000fold) into fresh cell passage media (1 ml) occurring every 12
2 hr. In parallel with every fourth
such back-dilution (corresponding to a ~48-hr period) beginning with the first occurring at t = 0
measurements of DNAi device knockout efficiency and target plasmid stability were performed
on a sample removed from the passaged culture in accordance with aforementioned protocols
26
(see both "Plasmid knockout assay" sections above).
Continuous culture and periodic
measurement of the sample were performed in this manner for a period of 90 days.
2.14
Cell death assays (DNAi-mediated killing).
-
The strain containing the 3xc DNAi device was transformed with either an on-target (pCR-G1 or
G2, CmR) or an off-target CRISPR targeting plasmid (pCR-N, CmR), plated onto LB containing
chloramphenicol in addition to 0.5% GIc, and grown for 12 hr at 37*C. Single colonies were used
to inoculate 2YT (1 ml) containing chloramphenicol and 0.5% glucose, and the resulting liquid
cultures were grown in a shaking incubator for 3 hr at 37*C and 250 rpm until OD6oo 0.25-0.75.
Cultures were spun down at 15,000 rpm, washed once with fresh 2YT (1 ml), and back-diluted to
OD60 0 0.01 into 2YT (2 ml) containing chloramphenicol. The diluted culture was then split into
two 1 ml samples, and DNAi activity was either induced by adding 2 mM arabinose (DNAi ON) or
repressed by adding 0.5% glucose (DNAi OFF). Cultures were then grown in a shaking incubator
for 8 hr at 37*C and 250 rpm with periodic removal of samples for analysis. For each sample, the
titer of viable cells (in cfu/ml) was then determined by 10-fold serial dilution and subsequent
plating onto LB agar containing chloramphenicol. The fraction of viable cells was then calculated
as the ratio of the cell titer in the DNAi ON state to that in the DNAi OFF state for a given CRISPR
target.
2.15
Survivor analysis assay (cell death experiment).
Viable colonies surviving suicide assay conditions (see above) were selected at random
from the terminal quantification plates (t = 8 hr) and used to inoculate 2YT (500 pl) containing
chloramphenicol and 0.5% GIc in 96-well format. Cultures were grown in a shaking incubator for
3 h at 37*C and 990 rpm until OD60 0 0.25-0.75. Cultures were spun down at 15,000 rpm, washed
once with 2YT (500 ptl), and then back-diluted 1:500 into fresh 2YT (500 pl) containing
chloramphicol and inducer (2 mM arabinose) in 96-well format. Induced cultures were grown in
a shaking incubator for 8 h at 37*C and 990 rpm at which point they were diluted 10 6-fold into
fresh 2YT containing chloramphenicol. Diluted cultures were immediately spotted (20 pl) onto
LB agar containing chloramphenicol, and these plates were dried at room temperature for 1 hr
27
prior to overnight incubation at 37*C.
Spots producing colonies or dense cell growth were
designated as corresponding to a surviving clone that had lost DNAi activity (escape mutant)
while spots with zero cell growth at this dilution were designated as corresponding to a surviving
clone that had retained DNAi activity (transient escape). The escape mutants were subsequently
screened for residual actuator activity through complementation with the cis-targeting plasmid
pCisTAR(S)-X in the manner previously described (See "Survivor analysis assay (plasmid knockout
experiment)" above). The remaining unclassified clones that had retained actuator activity were
then subjected to colony PCR and sequencing in order to identify whether mutations in the
plasmid-based CRISPR array or in the chromosomal proto-spacer/PAM had caused the loss of
function.
2.16 Cell death assays (Integrase-mediated killing).
The strain containing the lXB DNAi device and attint5-flanked KanR-Spacer X chromosomal
loop-out cassette was transformed with either an on-target CRISPR plasmid (pCR-X, pUC19, CmR),
an IntS actuator plasmid (plnt5act, p15A, AmpR) or both, plated onto LB containing appropriate
antibiotic in addition to 0.5% Gic, and grown for 12 hr at 37*C. Single colonies were used to
inoculate 2YT (1 ml) containing appropriate antibiotics and 0.5% glucose, and the resulting liquid
cultures were grown in a shaking incubator for 3 hr at 37*C and 250 rpm until OD600 0.25-0.75.
Cultures were spun down at 15,000 rpm, washed once with fresh 2YT (1 ml), and back-diluted to
OD 600 0.01 into 2YT (2 ml) containing appropriate antibiotics for plasmid selection but without
selection for the loop-out cassette(-kanamycin). The diluted culture was then split into two 1 ml
samples, and DNAi activity was either induced by adding 2 mM arabinose (DNAi ON) or repressed
by adding 0.5% glucose (DNAi OFF). Cultures were then grown in a shaking incubator for 8 hr at
37*C and 250 rpm with periodic removal of samples for analysis. For each sample, the titer of
viable cells was then calculated as the ratio of the kanamycin-resistant cell titer in the DNAi ON
state to that in the DNAi OFF state for a given strain.
28
3. Construction and optimization of a CRISPR DNA-interference device
for Escherichia coli
3.1
Device Design
Our initial conception of the CRISPR DNA-interference (DNAi) device (Figure 3.1) was one
designed to convert a generic input of promoter activity into an output of CRISPR-mediated DNA
interference. In essence, the device would be composed of two essential genetic components.
The first component, the actuator (Figure A.1), would encode of all the catalytic elements
necessary and sufficient for producing DNA interference in E. coli, specifically, cas3 and
casABCDE. The second component, the CRISPR array (Figure A.2A), would serve as the device's
targeting information. These arrays would encode one or more repeat-flanked spacer elements
that, once transcribed and subsequently processed by casE, would become the crRNA guide
strands that complex with actuator components and direct their catalytic activity to the
appropriate DNA targets in a sequence-specific manner.
Input
larabinose]
Target DNA
DNAi Device
P
Cf
Spa er
acer
caEProto-s
CasE
V
RFP
crRNA guides
Repeat
ori
PAM'
CRISPR
colE1
Actuator
PBA2~
Plasmid or
E.
Icos
casABcDE cas3
-
...
.....
genes
Targe
cohi chromosome
Daed
Figure 3.1: Schematic of the DNAi device. A schematic of the device (dashed box) is shown. The DNAi
is a genetically encoded machine that converts a promoter input into an output of CRISPR-mediated
DNA-interference activity and is composed of two essential components: a chromosomally-integrated
actuator element, which encodes the minimal set of cas genes required for DNAi activity, and a
reprogrammable targeting plasmid, which encodes the CRISPR array specifying the DNA target(s).
Actuator activation with arabinose via PBAD input promoter results in the expression of Cas3, an ssDNA
nuclease/helicase, and the CasABCDE complex. The CRISPR array is transcribed from a constitutive
promoter (PJ23117). The target plasmid, which possesses a valid PAM and proto-spacer combination is
recognized by the DNAi machinery and degraded.
29
An input promoter would serve as the device's master ON/OFF switch, and we elected to
place only the actuator components under inducible control, while leaving transcription of the
CRISPR targeting components under the control of constitutive a70 promoters. An alternative
implementation, such as a dual-induction system in which each of these two components was
under the control of an independent inducible promoter, might have demonstrated tighter
control over the system's expression, but at the same time would also have made for a more
complicated experimental setup.
This was undesirable for rapid and reliable prototyping.
Furthermore, expression of the cas components rather than of the CRISPR array was deliberately
chosen as the point of control, because high levels of cas expression, particularly of cas3, were
observed to be significantly toxic to the host (Figure 3.2). Having the device's default OFF state
correlate with low levels of cas expression would therefore generate less selective pressure
against maintenance of the device's components.
Fiure 3.2: Effect of cas over-expression on cell
10
growth. Data on the growth of E. coli BL21-Al cultures
is shown for strains over-expressing different cas
components from modified pWUR plasmids in which
the native P17 promoter was replaced with a strong
10
IPTG-inducible promoter.
CD
0
/
II
10
a
A
ff
10-2
0
2
4
6
8
10
12
Time Post-Activation (hr)
OD600 measurements of
cultures over time is shown for strains harboring
pWUR397* (cas3 over-expression, green diamonds),
pWUR400* (casABCDE over-expression, blue circles),
no plasmid (orange squares) in both the induced (1
mM IPTG, dashed line) and uninduced (0 mM IPTG,
solid line) states. All data were obtained by cuvettebased cell growth assay (Methods). Plasmid maps for
pWUR397* and pWUR400* are provided in
Supplementary Figure A.6
A plasmid-based target was designed to test the device's ability to remove stablymaintained DNA sequences from the host, because in the absence of antibiotic selection, such a
target can be deleted from the host genotype without significantly disrupting cell growth or
viability. Our prototype target plasmid design (pTAR(S), Figure A.2B) utilized a low-copy pSC101
origin of replication (10-12 copies/genome) in order to make the experimental system more
sensitive to the device's leakage activity in the OFF state. Also encoded on the target plasmid
30
were constitutively-expressed antibiotic resistance and RFP cassettes, which in combination
provided a dual-phenotype reporter system for quantifying the presence or absence of the target
plasmid within the host cell population either by plating assay or by cytometry, respectively
(Figure 3.1, A.2B).
Additionally, we initially opted for all-plasmid DNAi device. This strategy hopefully would
not only facilitate rapid and modular construction and reprogramming of device prototypes, but
also allow the device's end-users to more easily integrate DNAi technology into their own
engineered host strains.
B
A
.
100 .m
1.30
pSC101
M13
pAmpTAR(S)
C C
AX
X
<A lROR
or
<
M
R+pCR
QOR
j
10-,
10-3
AmpR
blo(X*) b/aAX
IN2
X
N2
X
Actuator pWUR397 + pWUR400
34
of
Figure 3.3: Genetic instability of the pWUR cas over-expression system (A) Map for the pAmpTAR(S) pair
of
versions
different
encode
but
resistiance
ampicillin
confer
versions
Both
pSC101-based RFP' target plasmids.
blaAX
while
PAM,
5'-ATG
an
to
X
coupled
proto-spacer
encodes
bla
the AmpR (bla) allele (bottom). Wild-type
contains a series of synonymous mutations that have removed proto-spacer X without altering the PAM or the
(B)
resistance phenotype. Sequences for spacer X and the recoded AX segment of bla are provided in Table A.1.
to
subsequent
host
pWUR397/400
a
within
plasmid
target
pAmpTAR(S)
a
of
Data reflecting the stability
combination
in
X)
or
(N2
spacers
different
for
shown
is
(pCR)
plasmid
targeting
a
CRISPR
transformation with
with different proto-spacer sequences (X+ or AX). The fraction of host cells that were positive for transformation
with pCR (CmR) and yet retained the target plasmid (Target', Amp') is shown for when transformation restored
it
the full complement of components required for target-directed DNAi activity (pCR-X/bla(X*)) and for when
and
did not (pCR-N2/bla(X'), pCR-N2/blaAX, and pCR-X/blaAX). The E coli MG1655-derived parent is T7RNAPby
obtained
were
Data
system.
this
for
levels
leakage
minimum
the
reflects
pWUR397/400
so expression from
independent
three
of
average
the
represent
and
(Methods)
assay
stability
plate-based
electroporation
experiments performed on different days. Error bars show the standard deviation.
Brouns et al had previously constructed an inducible, T7 RNAP-controlled, three-plasmid
cas/CRISPR over-expression system ('pWUR' plasmids, Figure A.7) that efficiently blocked both
31
phage infection and plasmid transformation3 4
4
. We had intended to use this design as a starting
point for our own device, but unfortunately our preliminary experiments demonstrated that
these components were intrinsically too leaky to construct a genetically stable DNAi device in Ktype E. coli. We transformed an E. coli MG1655-derived host that was genetically null for T7
RNAP with the two actuator plasmids, pWUR397 and pWUR400, along with one of two pSC101based target plasmid variants (Figure 3.3A). Both target plasmids conferred ampicillin resistance,
but pAmpTAR(S)-X encodes the wild-type bla sequence, which contains a valid PAM/proto-spacer
combination ('X'), while pAmpTAR(S)-AX encodes a mutant version of bla with a series of
synonymous mutations that remove the proto-spacer X sequence. In this state, possessing both
target and actuator but lacking an appropriate CRISPR array to target the cas protein activity,
these strains are negative for DNAi activity 34 6 4 . We then proceeded to challenge both strains by
transforming
them with
an additional CmR-conferring
CRISPR
plasmid
encoding the
corresponding spacer X (pCR-X). If actuator leakage was sufficiently low in this state so as to
avoid DNAi-mediated target loss, then we would expect all CmR pCR-X* transformants to retain
the target plasmid and its associated AmpR phenotype as well, regardless of whether CRISPR X
properly targeted the corresponding plasmid's bla gene. This was not observed, however, and
based on plating experiments, only 0.4% of pAmpTAR(S)-X+ cells that were successfully
transformed with pCR-X (CmR phenotype) were able to retain the target, while >97% of
transformants from the pAmpTAR(S)-AX strain retained the AmpR phenotype (Figure 3.3B, red
bars).
In addition, this difference could not be attributed to DNAi-independent variability,
because when transformed with a negative control CRISPR plasmid that encodes spacers
targeting neither pAmpTAR(S) plasmid nor the host genome (pCR-N 2, CmR), ~100% Of CmR
transformants from both strains retain the AmpR phenotype as well (Figure 3.3B, pink bars).
Thus, the leakage from the pWUR actutator plasmids' T7 promoters, even in the total absence of
their cognate phage polymerase, was sufficient to destabilize the target plasmid, and so we
deemed it necessary to design an entirely new cas/CRISPR system with tighter control in the OFF
state.
32
Our revised actuator implementation encompassed three major alterations to the
cas/CRISPR system of Brouns et al. First, both because the pWUR T7 promoters were intrinsically
leaky and because T7 RNAP activity is generally toxic to an E. coli host, we discarded T7RNAP as
an intermediary in the actuator expression cascade. Second, even though both the native E. coli
CRISPR machinery and all previously developed cas/CRISPR over-expression systems had divided
the cas3 and casABCDE actuator components into two separate operons
34
,66,67 , we elected to
simplify expression by merging the two into one contiguous transcriptional unit encoded on a
single plasmid (Figure A.1). In doing so, we both reduced the probability of homology-dependent
recombination between re-used promoter elements and removed the need for an additional
plasmid and antibiotic selection. Third, we selected the arabinose-inducible PBAD promoter from
the wild-type E. coli araBAD operon as the input, because it demonstrates an exceptionally tight
OFF state and a large dynamic range (Figure A.8). With actuator expression coupled directly to
PBAD
activity, the DNAi device could be switched into the ON state through the addition of
arabinose, and into the OFF state through the addition of glucose to the growth media.
This systematic overhaul provided the impetus to make a number of other smaller
optimizations as well. For instance, although RSF and p15A origins of replication are generally
regarded as compatible, we observed some inconsistency in device performance stemming from
what was later determined to be interference between pWUR397 and other p15A plasmids (data
not shown).
More consistent results were obtained when we abandoned an RSF plasmid
backbone in favor of a colEl backbone (pBR322 variant; -50-100 copies per genome6 7) for
actuator expression (Figure A.1). In addition, over-expression of Lacl from individual cassettes
on each component plasmid was causing growth defects, and since our revised device's input
promoter was no longer designed to be IPTG-responsive, these cassettes were removed. Finally,
the pWUR47x-based CRISPR targeting plasmids contained both a strong RBS and the native
CRISPR leader sequence immediately upstream of the array.
The RBS element was toxic,
presumably due to over-expression of the short cryptic peptide ORF that was found downstream,
and the leader sequence itself was determined to be non-essential for DNAi activity. Both were
therefore removed from subsequent constructs. These CRISPR plasmid modifications (Figure
A.2) together decreased the doubling time of a MG1655 F' host strain containing the
33
PBAD-
inducible DNAi machinery from ~30 min to 25 min, which was comparable to that of the DNAiparent (25 min)
3.2
Initial characterization of unstable DNAi devices
Unfortunately, despite the noted improvements apparent in our first-generation device's
toxicity profile, our extensive cas/CRISPR restructuring was insufficient to reduce DNAi leakage
levels to the point where the actuator, CRISPR, and target plasmids could all be maintained stably
within a single host. As a result, while it was possible to transform and select for the requisite
triple-transformants on LB plates, these hosts were not amenable to liquid culture under full
antibiotic selection (data not shown). These difficulties prohibited us from performing a proper
target plasmid knockout experiment, in which DNAi activity is deliberately induced in the absence
of antibiotic selection for the target plasmid in order to observe its disappearance from the host
over time.
Instead, we were forced to use alternative methods to quantify DNAi activity in which at
least one of the three essential components, actuator, CRISPR array, or target DNA, were
genetically absent from cells up until the very point of challenge. Previous work had established
methods in which a strain's DNAi activity was measured either by its ability to block infection
with a virulent phage that encoded a valid PAM/proto-spacer target or by its ability to prevent
transformation of naked plasmid DNA encoding a missing DNAi component
34,44,5O,51,63.
While
these approaches were both clearly effective, we decided that rigorous phage DNA replication
and competing phage metabolism would create too high of a bar for target DNA elimination and
therefore would underestimate the leakiness of our system. Direct transformation of plasmid
DNA, however, was insufficiently high-throughput for rapid and precise testing. Ultimately, we
opted for a phagemid transduction blocking assay, which was hybrid of these two approaches
(Figure 3.4). First, we created a pSC101-based target plasmid (pPAM-ATG, StrR) which encoded
an M13 viral origin of replication (i.e. a phagemid). The viral origin allowed the target plasmid
DNA, which by itself was non-infectious and not readily transmissible between cells, to be
packaged into M13 virus-like particles that were capable of transducing FI E. co/i hosts (Methods).
A pre-prepared stock of these pPAM-ATG virions was then used to challenge two host strains
34
whose DNAi machinery had been pre-induced (2 mM arabinose) or pre-repressed (+1% glucose).
One strain encoded a valid, on-target CRISPR spacer (X) capable of effecting DNAi against the
incoming target DNA, while the other strain encoded invalid, off-target CRISPR spacers (N 2). The
strain's effective DNAi activity was then quantified as the relative transduction efficiency (RTE),
which was defined the ratio of the number transformants in the on-target case divided by that
of the off-target after adjustments for DNAi-independent differences in transduction (Methods).
Virion
Target Phagemid
Viral
Packaging
pPAM-ATG
r
pSC101
i
M13Aori4
DNAP
Infection
h-..osX..M13 ori
MG1655 F
Valid PAM
(Receiver)
.
X
Transduction
(StrR)
Invalid PAM
Transduction
Blocked
(Strs)
Figure 3.4: Schematic of M13-based phagemid transduction blocking assay. Stock solutions of M13 transducing
virions encapsulating the target phagemid (pPAM-ATG) are prepared and then used to infect host receiver cells
(MG1655 F*) containing both a DNAi actuator and either an on-target (pCR-X, CmR) or an off-target (pCR-N2, CmR)
CRISPR targeting plasmid, and whose DNAi activity has be pre-induced (2 mM arabinose) or pre-repressed (+1%
glucose). The titer of target-positive transductants (in cfu/gl) is then measured by plating assay and the sample's
relative transduction efficiency (RTE) is calculated as the ratio between the titer of transductants in the on-target
case to that in the off-target case. See Methods 2.5 for greater detail. Plasmid maps for the associated constructs
are available in Figures A.1, A.2, and A.9, respectively.
When our first-generation DNAi device utilizing a high-copy colEl-borne PBAD-controlled
actuator was tested in this manner (Figure 3.5, blue bars), the ON state (+2 mM arabinose)
demonstrated robust DNAi-mediated transduction-blocking activity, which reduced the strain's
RTE to less than 2.6x10-5 . Nevertheless, as suspected from our experiences with direct target
plasmid transformation, the DNAi OFF state (+1% glucose) was still unusably leaky, and the
4
strain's RTE was only ~20-fold higher (RTE 5.6x10 ) than the ON state. For our second-generation
device, we changed the actuator plasmid's origin of replication from colEl to p15A (20-30
copies/genome), thereby lowering the actuator copy number approximately 3- to 5-fold. This
modification also required a compensatory change of the CRISPR plasmid's origin to avoid
compatibility issues.
We wished to accomplish this exchange while maintaining a roughly
equivalent level of CRISPR array transcription, and so while we elected to raise the CRISPR
35
component's copy number ~20-fold through utilization of a high-copy pUC19-based colEl origin
(>500 copies/genome), we also simultaneously swapped out the array's strong constitutive
promoter PJ23119 for a weaker mutant variant PJ23117, which demonstrates roughly 15- to 20-fold
lower activity.
100.
r.
0
S- - - - - - - -
- - - -
- - - -
101.
0
.0
10-2,
M4
in
10-3
Z
10-4
I;
10-S
DNAi
Spacer
Actuator
OFF ON OFF ON OFF ON
X
pACT-01
X
pACT-02
(colEl)
(p1SA)
X
pACT-A
(1xA gen.)
Figure 3.5: Comparison of phagemid transduction
efficiencies for DNAi devices with different
actuator configurations. Data for the DNAimediated blocking of transduction with a target
phagemid virion (pPAM-ATG) are shown for
different spacers in combination with different
actuators (plasmid-borne p15A and BAC ori and 1x
genomic). The relative transduction efficiency
(RTE), which reflects the ratio of the number of
phagemid transformants (Str) generated upon
infection of an on-target strain (X) to that
generated upon infection of an off target strain (N2,
Methods), is shown for all strains in the OFF
(repressed by 1.0% glucose) and ON (2mM
arabinose) states. The dashed line corresponds to
RTE = 1. All data were obtained by plate-based
phagemid-blocking assay (Methods) and represent
the average of three independent experiments
performed on different days. Error bars reflect the
standard deviation.
This second-generation device was subjected to a similar set of target phagemid blocking
experiments (Figure 3.5, green bars). Interestingly, despite the copy number-related drop in
actuator expression in the ON state (+2mM arabinose), the strain's RTE (3.5x10-4 ) was a mere
1.4-fold higher than that of the strain harboring the colEl-based actuator. More importantly,
however, the stability of the corresponding OFF state was improved 500-fold (RTE = 0.28x10').
These non-linear relationships between phagemid blocking and copy number suggested that the
first-generation actuator was already operating in or near the saturated (i.e. zeroth-order) regime
for DNAi activity even in the OFF state.
This notable improvement in the DNAi device's OFF state represented substantial
progress towards our goal of a leak-free system, and we wished to determine whether further
decreases in actuator copy number might achieve an even tighter OFF state. Unfortunately,
because our target plasmid already utilized a pSC101 origin, and a BAC-based plasmid would be
incompatible with the F' episome required for M13 phagemid virion infection, we were left with
36
very few choices among the commonly-utilized, well-characterized E. coli expression vectors for
one with a copy number lower than that of p15A-based plasmids. We therefore decided at this
juncture to abandon temporarily our all-plasmid approach to device construction and instead to
attempt to integrate the actuator components directly into the host chromosome. This would
ensure both a lower, more precise distribution of copy numbers and greater genetic stability than
could be achieved with plasmid-based expression.
After a series of manipulations to the MG1655 F' host chromosome and modifications to
the actuator plasmid (pACT-A, Figure A.1), the complete actuator sequence was integrated into
the host chromosome via FLP-mediated loop-in mechanism (Figure A.4).
During the design
process, we deliberately decided that this integration would result in a net replacement of the
native cas/CRISPR locus (Table A.1) for two reasons. First, even though the native E. coli DNAi
machinery in a wild-type genetic background remains completely silent under all experimental
conditions tested thus far, its removal was desirable for purposes of experimental control and
scientific rigor.
Second, previous work had already established that the actuator's cas
components could be inducibly over-expressed from this location without significant issue.
With this chromosomally-actuated 1xA (where the subscript 'A' denotes the actuator
variant) DNAi strain in hand, we repeated the phagemid transduction-blocking experiment once
more (Figure 3.5, pink bars). While the strain's RTE in the DNAi ON state (1.1x10-2) was over 300fold less efficieny than when the p15A actuator was used, the corresponding DNAi OFF state was
effectively 100% silent within experimental error (RTE >1).
3.3
Characterization of stable DNAi devices harboring chromosomal actuators
While this phage blocking result reassured us that the DNAi OFF state of this chromosomal
actuator strain was indeed stable, we took the opportunity to make additional practical
modifications to the host's genetic background before proceeding any further in our
characterization. In addition to the existing Acas/CRISPR knockout that served as the means of
integrating our own actuator sequence, two additional marker-free deletions, AaraC-araBAD and
AP/ac:Iac/ were introduced via standard XRED-based mechanism (Methods, Table A.2). The araC-
araBAD locus encoding the native arabinose pathway was deleted to prevent the host from
37
metabolizing the input promoter's inducer molecule over the course of an experiment, and the
APlac,:IacI deletion was introduced to allow for more precise control over Lac dosage should an
IPTG-inducible promoter be re-introduced into our device design at a later point in time. Equally
importantly, however, these deletions effectively created two more potential integrations sites
for additional chromosomal copies of the actuator sequence (Table A.2).
Following these modifications to the host's genetic background, we proceeded to
perform a proper plasmid knockout experiment using the iXA chromosomal actuator, a CRISPR
plasmid (pCR, pUC19 ori, CmR) encoding either the X (on-target) or the N 2 (off-target) spacers,
and the pTAR(S) (pSC101 ori, StrR, RFP*) DNA target (Figure 3.6).
When properly targeted,
induction of DNAi activity removed the target plasmid from >99.96% of cells, yet in the DNAi OFF
state the target was retained at what was effectively 100% given the precision limits of our platebased assay (Methods). As expected, no appreciable loss of the target plasmid was observed in
the off-target case, and so we concluded that the observed on-target plasmid loss was indeed
DNAi-dependent. Both of our conditions for the dynamic range of DNAi activity had been met,
and so in this particular regard our device was a success.
Having determined that lowering the actuator's copy number was a successful strategy
for achieving our DNAi device performance goals, we temporarily back-tracked to construct and
test a BAC-encoded version of the actuator in the hopes that it could be applied to realize our
initial intentions of an all-plasmid DNAi device. Similar to the aforementioned experiments with
the chromosomal actuator, DNAi strains were constructed that contained either a BAC- or pl5A-
born actuator, an on-target (X) or off-target (N 2) pCR-borne CRISPR array, and a pSC101-based
target plasmid (pTAR(S), RFP*' StrR). These strains were subjected to standard plasmid knockout
conditions (Methods), and the fraction of host cells retaining the target plasmid was measured
(Figure 3.6).
While the on-target strain harboring the BAC actuator variant did indeed
demonstrate improved performance over the analogous p15A-based strain, the BAC-based
device, which removed the target from 99.6% of cells in the DNAi ON state and retained the
target in only 81% of cells in the DNAi OFF state, was still unable to meet either criteria for a
successful DNAi device. In addition, unlike actuator sequences carried on the genome, which
demonstrate lower toxicity, both the pl5A- and BAC-based actuators significantly reduced the
38
growth of the host (data not shown). We therefore elected to abandon plasmid-based actuators
once and for all.
Although our chromosomally-integrated version of the DNAi device met our criteria for
successful target plasmid knockout, it too still possessed a few undesirable properties that
merited refinement. First, because FRT-FRT recombination perfectly conserves the FRT site
sequence, the FLP-based method used to insert the actuator sequence could have inadvertently
allowed multiple, iterative integrations to occur at the single targeted locus. Thus, the precise
B
0
X
100
_t
)
CRISPR
-
A
2
Target
Plasmid
~1N
0
1
pSC1O1
DNAi
Spacer
Actuator
OFF ON
OFF
N2
ON
OFF ON OFF ON
N2
X
pACT-03
(BAC)
pACT-02
(p1SA)
X
OFF ON OFF ON
N2
1XA
(genome)
X
OFF
ON
OFF ON
X
2
1x8
(genome)
and
Figure 3.6: Plasmid knockout efficiencies when the DNAi actuator is carried on different plasmid backbones
a
non-coding
in
present
is
X
genomic configurations. (A) The designs for target spacer sequences are shown.
correspond to
region of the pSC101 target plasmid (pTAR(S)) with an ATG PAM, and neither of the dual N2 spacers
provided in
are
spacers
these
for
Sequences
genome.
a valid proto-spacer target in either the plasmid or the host
combination
in
spacers
different
for
shown
are
plasmid
Supplementary Table 1 (B). Data for the knockout of target
cells that
with different actuators (genomic 1XA and 1XB, and plasmid-borne p15A and BAC ori). The fraction of
and ON
glucose)
0.5%
by
(repressed
OFF
the
in
shown
is
(Target')
retain the pSC101 origin pTAR(S) target plasmid
the
and
assay
cytometry
the
using
obtained
were
data
ON
1x/N
the
(2mM arabinose) states. The OFF data and
three
of
average
the
represent
data
.
All
(Methods)
assay
plate-based
the
remainder were obtained by
independent experiments performed on different days and the error bars are the standard deviation.
number of chromosomal copies present was, at least initially, uncertain. Second, if multiple
copies were indeed present in the chromosome, then these repeated sequences were present at
the target locus in the form of a tandem array of head-to-tail repeats. The MG1655-based host
strain, designed to mimic an industrially relevant fermentation strain, was recA', and so such a
construction would be inherently prone to actuator elimination through homology-based
recombination68 . If all of these were required for the full level of DNAi activity observed in the
ON state, this could destabilize or otherwise break the device. Third, as constructed, there was
39
no mechanism to cure the Kan' marker used for selection, and since chromosomal integrants in
E. coli are typically stable in the absence of selection, it was suboptimal to waste this rather useful
genetic part in such a manner.
We therefore redesigned our chromosomal insertion strategy to remedy these problems
(Figure A.3, A.6).
recombinase
unidirectional
Time Post-Activation (hr)
Utilizing a phage-derived
(Int5) and
0
its
6
12
18
24
corresponding attB/attP sites in lieu of FLP and
its reversible FRT/FRT recombination reaction
a101
ensured that the actuator sequence could
I2
integrate precisely once at each chromosomal
target locus.
C
0
In addition, the integrase sites
10-
io-
were strategically repositioned within both the
10-4
revised actuator plasmid (pACT-B, Figure A.1)
and the host's chromosomal target locus in
10.
order to allow for the KanR cassette to be
Figure 3.7: Comparison of plasmid knockout kinetics in
early genomic actuator prototypes. A time course is
FLP-mediated
shown for the knockout of pSc101-based target plasmid
extracted
via
subsequent
FRT/FRT recombination. This ability to reclaim
pTAR(S) from DNAi strains programmed with spacer X
and harboring either a1xA (blue triangles) or a1xB (green
squares) chromosomal actuator. Measurements reflect
the selection marker ensured that if additional the fraction of cells that retain the pTAR(S) target
plasmid (Target') in the DNAi ON state (2mM arabinose)
copies of the actuator were required, they and were performed via plate-based assay (Methods).
data are the average of three independent
could be integrated elsewhere in the host All
measurements performed on different days and error
bars show the standard deviation.
chromosome, albeit one at a time.
Once we had constructed the equivalent marker-free, definitively single-copy (1XB)
version of the chromosomal actuator, we performed both time-course and end-point
measurements to assess the plasmid knockout performance of the corresponding 1XB DNAi
device and compared it to that of our original 1XA DNAi device (Methods). As hoped, when both
devices were programmed with spacer X to effect the knockout of pTAR(S), the overall plasmid
knockout efficiency (Figure 3.6) and the knockout kinetics (Figure 3.7) were virtually identical for
both versions of the device. Thus, in addition to the engineering success that these results reflect,
the matching kinetics also retroactively confirmed that the original IxAgenomic actuator strain
40
had indeed harbored just a single copy of the actuator sequence. Nevertheless, the lXB version
of the chromosomal actuator, with its genomic antibiotic resistance marker removed, was
superior to the original 1XA version, and so the latter was abandoned in favor of the former.
3.4
Parameterization and Optimization of DNAi Device Kinetics
3.4.1
Introduction
At the simplest level our DNAi device was designed to work as a molecular machine that
converts a simple promoter input signal into a quantifiable output of DNAi activity. Nevertheless,
given the layered and complex interplay of DNAi components and endogenous host components
that is required for proper device function, the true input-output relationship is not that simple,
and many design specifications other than input promoter activity can alter the device's
performance characteristics. It would therefore be useful to understand the influence that these
other variables have on this input-output relationship so that the device can be tuned to meet
the unique demands of the largest number of possible biotechnology applications.
Furthermore, when utilized for purposes of plasmid knockout, the DNAi device's output
can be quantified in terms of either its efficiency, i.e. the fraction of host cells retaining the target
following DNAi induction, or its kinetics, i.e. the rate at which the plasmid is lost from the pool of
host cells. In order to have full control over the device in an arbitrary context, it would be
necessary to understand the relationship each design specification has with both of these output
parameters.
With respect to their influence on the DNAi device's plasmid knockout rate, we selected
the following five design specifications for parameterization and/or optimization: crRNA
expression level, spacer sequence, araC dosage, PAM selection, and actuator copy number. For
each kinetic parameter examined, a strain containing the lxB genomic actuator and the standard
pTAR(S) target served as the base case unless otherwise stated. Host cells were subjected to
DNAi-inducing conditions (+2 mM arabinose) over the course of an 8-h plasmid knockout
experiment, and the maximal rate of target plasmid loss was determined by measuring the
fraction of target-positive cells at various time points using a plate-based assay (Methods).
41
3.4.2
crRNA expression level
The first parameter that we examined was the dose-dependence of the device's DNAi
activity on the level of crRNA expression. As per our initial design, the weak PJ23117 promoter
present on the pCR plasmid (pUC19 ori, CmR) was responsible for providing constitutive
expression of the essential crRNAs.
While certainly functional, it was unclear at this stage
whether this particular expression level, which had been chosen somewhat arbitrarily, was subsaturating or super-saturating with respect to DNAi activity. We therefore constructed a CRISPR
plasmid in which expression of spacer X
Time Post-Activation (hr)
0
was instead placed under control of an
aTc-inducible Ptet promoter (pCR-Xtet,
2
8
6
4
100 4
p15A ori, CmR) and then measured the
rate of DNAi-dependent target plasmid
4)
1010-1
-
loss under both Pterinduced (+100 ng/ml
aTc) and Ptet leakage (0 ng/ml aTc)
0
U
L.
U-
%
10-
--
--
-
conditions (Figure 3.8). When compared
to our base case in which spacer X was
expressed from the standard pCR-X
construct (blue triangles), the rate of
target plasmid loss was significantly faster
in the Ptet-induced (black circles, dashed
line) case while significantly slower in the
uninduced case (black
corresponding
circles,
solid
suggested
line).
that
These
pCR-X
was
findings
in
fact
delivering an intermediate dose of crRNA
relative to the levels of the DNAi cas
machinery provided under DNAi ON
conditions.
Additional
observations
10-5.
10-6.-
Figure 3.8: Effects of crRNA expression level on target
plasmid knockout kinetics. A time course is shown for the
knockout of pSC101-based target plasmid pTAR(S) from DNAi
strains harboring a 1XB chromosomal actuator and expressing
different amounts of the crRNA corresponding to spacer X.
The X spacer was expressed either constitutively from a
medium- (pCR-Xpi5a, p15a ori, green squares) or high-copy
(pCR-X, pUC19, blue triangles) plasmid or inducibly from a Ptet
promoter (pCRtet-X, p15A, black circles). In cases of Pter
mediated expression, host strains were examined in eitherthe
presence (100 ng/ml aTc, dotted line) or the absence (0 ng/ml
aTc, solid line) of inducer. Measurements reflect the fraction
of cells that retain the pTAR(S) target plasmid (Target') in the
DNAi ON state (2mM arabinose) and were performed via
plate-based assay (Methods). All data are the average of three
independent measurements performed on different days and
error bars show the standard deviation.
42
suggested that the
PJ23117
promoter is exceptionally weak within the context of the pCR plasmid,
and it is instead the construct's high copy number (>500 copies/cell) that allows it to provide a
reasonable amount of crRNA. When the DNAi kinetics of a lower-copy p15A version of pCR-X
were measured (Figure 3.8, green squares), the performance was lower than even that
corresponding to leakage levels of Ptet.
This facet of the DNAi device's design could have
significant consequences if an unmodified version is embedded within an engineered host that
uses strong
G70
underperform if
promoters as part of its other programming, because DNAi may fail or
PJ23117
has trouble competing for the host RNAP pool.
Plans for a comprehensive sweep of crRNA dosage in order to find the point at which DNAi
activity from the 1xB device becomes saturated were abandoned in favor of other efforts.
Nevertheless, the capacity to tune DNAi activity in either direction through manipulation of
crRNA dosage is a beneficial feature that may allow the device to interface with a greater number
of biotechnology platforms.
3.4.3
Spacer sequence
The power of CRISPR DNAi interference is largely derived from its ability to target specific
pieces of DNA for elimination based solely on short stretches of their sequence. These sequences
therefore serve as a form of molecular fingerprint for those specific targeted molecules. For
applications in which native rather than synthetic DNA is targeted, changing DNAi targets might
necessitate using a different fingerprint for identification, and in terms of CRISPR biochemistry,
this would translate into programming the DNAi with a different spacer sequence.
Unfortunately, previous work with the E. coli Type-IE cas/CRISPR system had failed to
address the question of whether the CRISPR spacer sequence can affect the kinetics of DNAi-
mediated target degradation, and so we performed a set of experiments to investigate this
variable as a possible kinetic tuning parameter.
For these experiments, we compared the
performance characteristics of spacer X to two additional spacers, Y and Z (Table A.1). Our
experiments (Figure 3.9) showed that DNAi activity utilizing CRISPRs X (blue triangles) and Z (black
circles) were comparably fast, yet the activity corresponding to spacer Y (green squares) was
43
significantly slower.
This difference in
Time Post-Activation (hr)
kinetics could not simply be attributed to
0
2
4
6
8
differences in the designated target's PAM,
100,.
as both the faster X and the slower Y protospacers possessed an 5'-ATG PAM. Thus,
102
-
spacer sequence can play a direct role in
10-1
A
the kinetics of the associated DNAi activity.
explanation for the considerably slower
0
-
Experiments to determine the mechanistic
U-
10-3.
10.
rate of spacer Y-mediated plasmid loss
were beyond the scope of this work. A list
of
possible
explanations
for
this
phenomenon includes, but is not limited to,
different crRNA lifetimes, different rates of
CasE-mediated
pre-crRNA
processing,
differences in affinity for the Cascade
complex, altered RNA-DNA base-pairing
thermodynamics during target recognition,
and sequence-dependent differences in the
10-6 j
Fiaure 3.9: Effect of spacer sequence on target plasmid
knockout kinetics. A time course is shown for the knockout
of pSC101-based target plasmid pTAR(S) from DNAi strains
harboring a 1xB chromosomal actuator and programmed
with different CRISPR spacer sequences. Spacer X (5'-ATG
PAM, blue triangles) targets a non-coding region within the
target, while spacers Y (5'-ATG PAM, green squares) and Z
(5'-AAG PAM, black circles) target regions within the target's
StrR coding sequence. The fraction of cells that retain the
pTAR(S) target plasmid (Target ) is shown in the DNAi ON
state (2mM arabinose). Measurements were performed via
plate-based assay (Methods). All data reflect the average of
three independent measurements performed on different
days and error bars show the standard deviation.
rate of Cas3-mediated cutting of the target.
Curious as to how one fast and one slow spacer might interact to affect DNAi
performance, we constructed and subsequently tested a fourth pCR variant, which encoded the
dual Y+Z spacer combination. The rate of plasmid loss with dual spacers was much closer to that
of the case with the single, faster spacer Z, (data not shown) strongly implying that competition
between the two spacers any the rate-limiting step of the DNAi biochemical pathway was
minimal. From these aggregate observations, we concluded that spacer sequence could indeed
significantly influence the rate of plasmid loss. Furthermore, the significantly slower kinetics
associated with spacer Y provided us with a tool to make more precise time-dependent
measurements as we proceeded with additional experiments.
44
3.4.4
AraC expression levels
The transactivating protein AraC is essential for sensing arabinose and activating our
device's
PBAD
input promoter, and so the third DNAi parameter we optimized for kinetic
performance was araC dosage. Somewhat serendipitously, while testing our DNAi device base
case
(1XB
actuator), pTAR(S)) with spacer Y (pCR-Y) within the context of an araC/araBAD'
genetic background (Figure 3.10, green squares) we noticed that the kinetics of target knockout
were significantly faster compared to those observed when testing our standard triple knockout
host strain (blue triangles), which contained a AaraC-araBAD deletion.
This was somewhat
paradoxical, because the araBAD' strain was capable of metabolizing the arabinose used to
activate the DNAi input promoter, and so we expected that this would reduce the effective
inducer concentration, lower the device's maximal DNAi activity, and thus ultimately slow the
kinetics of plasmid knockout. Indeed, we were able to confirm that arabinose metabolism was
responsible for the lower knockout efficiency observed in araBAD' strain following the standard
8-h
induction period (target' fraction
1.4x10- 3),
because increasing the initial inducer
concentration above 2 mM or re-inducing cells mid-experiment (at t = 4 h) both increased the
overall target knockout efficiency after 8 h (data not shown). Nevertheless, the discrepancy in
the strains' kinetics remained unexplained.
We hypothesized that differences in araC expression were in fact responsible for our
unusual observations. In the context of a wild-type araC+ E. coli background, expression of araC
is controlled by the Pc promoter, which is auto-regulated to provide a dynamic level of
transactivator that increases to meet the host's increased demands upon arabinose induction.
In contrast, in the context of our AaraC-araBAD DNAi strain, this auto-regulatory Pc-araC circuit
has been removed and has been replaced with a single copy of a constitutively-expressed araC
cassette (PJ2 3 117-araC), which was added to the chromosome as part of the pACT-B plasmid. Given
how weak we had determined promoter PJ23117 to be, however, it was reasonable to suspect that
this construct was unable to provide the host with sufficient AraC transactivator for optimal
PBAD
function, even though the cassette provided the ORF with a strong RBS (BBa_B0034, Table A.7).
45
To test our hypothesis, we constructed an araC complementation plasmid that encoded
the same PJ23117-araC cassette (pAraC-117, p15A, KanR) and that, based on its copy number, was
predicted to boost AraC levels within our DNAi host strains ~20-30-fold. As predicted, upon
complementation the knockout kinetics of the DNAi base case strain improved considerably
(Figure 3.10, black circles), exceeding even those observed in the alternative araC/araBAD+ host
background (Figure 3.10, green squares)
this
While
had
improvement
Time Post-Activation (hr)
validated our initial hypothesis, it was still
0
2
4
8
6
uncertain whether this araC boost was
1004
saturating the host's demands for AraC
protein, and so we built and tested a second,
S10-1
U
c 10-
analogous araC complementation plasmid
-
with a constitutive promoter variant (Pi2310s,
Table A.6) that was ~4-fold stronger (pAraC105, p15A, KanR).
knockout
kinetics
The target plasmid
associated
.104
with this
10-5-
.A
.Ix-
stronger complementation plasmid (Figure
diamonds)
red
those
matched
associated with the weaker araC construct
(black circles), thereby indicating that in
both cases the AraC levels were indeed
saturating, at least with respect to their
ability to affect DNAi activity. Nevertheless,
the need for an additional accessory plasmid
to achieve optimal DNAi device function was
undesirable from a design standpoint. We
therefore
chose
chromosomal
to
actuator
redesign
plasmid
to
the
re-
incorporate the native Pc-araC architecture
-ly
1x+ pAraC-117
4 pAraC-10s
Figure 3.10: Optimization of araC expression levels with
respect to target plasmid knockout kinetics. A time
course is shown for the knockout of pSC101-based target
plasmid pTAR(S) from DNAi strains harboring a 1xe or 1xc
chromosomal actuator and spacer Y (pCR-Y) that express
different levels of araC in either a wild-type
araC/araBAD' (dashed line) or a AaraC-araBAD (solid
line) genetic background. The 1xB actuator expresses araC
from a constitutive PJ23117-araC cassette. Both the 1xc
actuator and the wild-type araC locus express araC from
an auto-regulated Pc-araC cassette. The pAraC-117 and
105 complementation plasmids (p15A, KanR) encode an
additional (weak) Pn3117-araC and (stronger) PJ231os-araC
respectively.
cassette,
expression
constitutive
Measurements reflect the fraction of cells that retain the
pTAR(S) target plasmid (Target') in the DNAi ON state
(2mM arabinose) and were performed via plate-based
assay (Methods). All data are the average of three
independent measurements performed on different days
and error bars show the standard deviation.
-
3.10,
10-6
U.
roC./orAD')
46
(pACT-C),
and then integrated this plasmid into our triple-knockout host to create the
corresponding lxc actuator strain. As we had hoped, when tested under our base case conditions
(pCR-Y, pTAR(S)), this revised accessory-free DNAi device demonstrated an equivalent rate of
plasmid knockout to that previously observed with the accessory plasmid (Figure 3.10, purple
X's). We therefore decided that arac dosage had been successfully optimized both with respect
to DNAi kinetics and with respect to our additional practical design constraints.
3.4.5
Proto-spacer adjacent motif (PAM)
In addition to spacer/proto-spacer complementarity, CRISPR-mediated DNA interference
activity also requires a valid 3-bp 5' proto-spacer adjacent motif (PAM)
, 5, 1. Previous work by
4 4 50
other groups using an experimental system in which phage-targeted cas/CRISPR over-expression
blocked the propagation of M13 infection had empirically established a rather limited set of four
functional PAM sequences, 5'-AAG, -AGG, -ATG, and -GAG". For the DNAi device to be utilized
to its greatest extent, however, a more comprehensive and sensitive analysis of PAM function
and kinetics was required.
To this end, we constructed a comprehensive set of 64 target plasmids each containing
one of the possible 3-bp PAM sequences immediately adjacent to non-coding proto-spacer X
(Figure A.9). Similar to the pPAM-ATG target used for our initial phagemid blocking experiments,
this set of target plasmids conferred streptomycin resistance (RFP-) and included an M13 origin
of replication that allowed the plasmid to be packaged into a phage-like transducing particle in
the presence of a M13K07 helper element (Methods). The corresponding library of transducing
particles was prepared and used to infect samples of F' host cells harboring a p15A plasmid-based
actuator and either an on-target (X) or off-target (N) CRISPR plasmid that had been pre-induced
with arabinose (2 mM). If a particular library member contained a valid PAM for proto-spacer X,
the resulting properly-targeted DNAi activity would reduce its relative transduction efficiency
(RTE, Methods) (Figure 3.4).
Using this transduction-blocking assay we identified ten strong PAMs (Figure 3.11, dark
blue shading, -Log(RTE)
2.0), which included all four canonial sequences (black outline) and five
weak PAMs (light blue shading, 0.5 < -Log(RTE) < 2.0). In addition, the 5'CCG anti-PAM (gold
47
outline), which allows the cas/CRISPR machinery to discriminate self (i.e. host-encoded CRISPR
spacer) from non-self (i.e. valid proto-spacer target), demonstrated no appreciable phagemidblocking activity (-Log(RTE) = 0.1). This expanded PAM set will allow designers to target a
considerably broader range of native sequences while simultaneously enabling more
comprehensive prediction of possible off-target interactions.
- Loglo(RTE)
14 N
3
NI
.2 0.4 -0.1 0.1 0.4 0.2 0.0 0.1
-0.1
C
0.1 0.2 0.4 0.0
G
.3 0.1 0. 0.2 0.2 0.1
0.8 0.1 0.2 0.2
.
0.8 0.2 0.1 0.1
.1 0.1 0.1 0.2 0.1
0.1 1.0
0.1 0.5 0.8
0.1 0.1 -0.1 .1 0.1 -0.1 0.1 0.3 0.1 0.3 0.2 0.1 0.2 0.1 0.0
T
LN2
0.2
A
A
C
GTA
A
C IG T A
C
C
G
G
T
AC
T
Figure 3.11: Effect of PAM sequence on phagemid transduction efficiency.
Data reflect the negative log-scale RTE values for each PAM as determined via
the phagemid blocking assay described in Figure 3.4 and Methods. Ten strong
PAMs (dark blue; -Log(RTE) > 2.0) and 5 weak PAMs (light blue; 0.5 < -Log(RTE)
< 2.0) were identified in this manner. The previously identified canonical PAM
sequences are outlined in black 49 . The gold outline corresponds to 5'-CCG,
which matches the 3 nt of the crRNA immediately 5' to the spacer sequence
was used as a negative control. Data reflect the average of three independent
measurements performed on different days.
We subsequently performed a series of plasmid knockout experiments in order to
quantify more precisely the relative kinetic effects of the fifteen strong and weak PAMs on DNAi
activity. These experiments required a higher time-resolution than our previous analyses, and so
a cytometry based assay was developed to measure the fraction of target-positive cells in a given
sample based on RFP fluorescence (Methods).
Host strains containing a lxc chromosomal
actuator, an on-target (X) CRISPR targeting plasmid, and a pSC101 target plasmid (RFP', Figure
A.9) encoding spacer X coupled to one of the 15 active PAMs or a 5'-CCG negative control PAM
was subjected to standard plasmid knockout conditions (Figure 3.12). The kinetic profiles of
target knockout for each PAM were then characterized with respect to three parameters (Table
3.1): time to initiation of target loss (tiag), whether the target decay curve possessed one
(monophasic) or two (biphasic) distinct linear regimes, and half-time of target loss for a given
linear regime ('n).
48
B
A
0
Time Post-Activation (hr)
6
4
2
8
100
-0x
CRISPR
10-1
pSC101
0
RFP
Target
Plasmid
PAM
(5'-NNN)
x
Std
ILL
10-2
AAG AGG
ATG GAG
10-3
-
Strong PAM
Weak PAM
CCG
10-4
Figure 3.12: Effect of PAM sequence on the kinetics of plasmid knockout. (A) The X spacer is designed to be present
in a non-coding region of the pPAM-NNN-RFP target plasmid series (pSC101 origin, StrR), where NNN corresponds to
a different X-associated 3-bp PAM sequence of the form 5'-NNN-[Spacer X]. The X spacer sequence is presented in
Table A.1. (B) The dynamics of plasmid loss are shown for each of the active PAM sequences. Data are for the lxc
DNAi device, X spacer targeting plasmid, and a pSC101 target (RFP*, StrR) and reflect the fraction of cells in the ON
(Methods). Black
state (2 mM arabinose) that retain the target plasmid (Target') as determined by PAM kinetic assay
50
lines (AAG, AGG, ATG & GAG) correspond to the canonical PAM set identified by Westra et a . All canonical PAMs
are classifiably strong PAMs. Target plasmid decay curves are truncated at the point in time where the fraction of
Target' cells reaches its minimum value. Error bars have been omitted for clarity but are shown in Figure A.10.
49
Table 3.1. Summary of kinetic parameters for active PAMs
PAM"
tnit (h:m)b
kic
Ti (min)d
k 2 ce
T2 (min)d
-Logio(RTEY
AAA
2:50
2.56
16.2
6.07
2.98
4.0
AAC
2:40
3.50
11.9
5.66
3.19
3.5
AAG
2:40
19.9
2.09
AAT
2:50
1.76
23.6
AGG
2:50
8.63
4.82
3.6
ATA
3:30
7.11
5.85
2.2
ATG
2:50
11.1
3.75
2.9
CAG
4:00
0.961
43.3
GAG
2:50
7.84
5.31
GGG
5:00
0.114
363
0.402
103
0.5
GTG
4:00
1.40
29.8
2.78
15.0
0.8
TAA
4:00
1.63
25.6
2.88
14.4
0.8
TAG
2:50
9.65
4.31
TGG
4:00
1.58
26.3
TTG
3:30
7.74
5.37
3.8
6.33
1.95
2.85
21.3
3.2
0.8
3.4
4.0
2.77
15.0
1.0
2.1
a. Bold face corresponds to canonical PAMs
b. Time to initiation of target loss (toit) was measured as time required for
target retention to drop below 95%
c. The linear region(s) of log-scale decay curves from Fig. 3C were fit to
the equation y = A*exp(-k*t)
d. Half-time of decay in minutes, T,, was calculated as 60*ln(2)/kn
e. For kinked curves with two distinct linear regimes, ki corresponds to
the fit for the earlier segment, and k 2 for that of the later.
f. Values correspond to Fig. 3.B
50
Based on these three parameters, the fifteen strong and weak PAMs could be further
subdivided into five sub-classes. The first and fastest class contained the four canonical PAMs
(5'-AAG, -AGG, -ATG, and -GAG - Figure 3.12, black lines) in addition to the strong PAM 5'-TAG
and was characterized by a rapid-onset (2.5 h < tiag < 3.0 h), very fast (2.0 m <
c < 6.0 m),
monophasic decay of the target signal. The second set, containing the 5'-AAA, -AAC, and -AAT
strong PAMs, demonstrated a similar onset of target knockout (2.5 h < tag< 3.0 h), but the target
signal decay was biphasic with a slow initial phase (10 m < T,< 25 m) preceding a fast later phase
(2.5 m
< t2 <
3.5 m). The third intermediate group contained the strong PAMs 5'-ATA and -TTG
and demonstrated notably slower onset of knockout (tag = 3.5 h) and a slightly slowed (,r -5.5 m)
monophasic signal decay. The fourth set, comprised of the four slow PAMs 5'-CAG, -GTG, -TAA,
and -TGG, conferred a late onset of knockout (tag= 4.0 h) and a biphasic, slow decay (14m <t 2
<
22 m). The final class contained only a single PAM, 5'-GGG, which demonstrated the latest onset
of target knockout (tag = 5.0 h) and a biphasic signal decay with the longest half-times (1i=
m and
T2 =
363
103 m). It also should be noted that for all samples target knockout ceased no later
than 7 h post-induction, and thus knockout was significantly more efficient overall for the first
three faster PAM groups (~10-4) compared that of to the last two slower PAM groups
(>10-2).
This
extensive characterization enables a designer to select a PAM with a kinetic profile tailored to
the desired application.
3.4.6
Actuator copy number
The primary goal of our initial attempts to build a functional DNAi device was to suppress
the activity of the OFF state so that the construct would be sufficiently genetically stable to
perform the necessary characterization experiments in a reliable fashion. Unfortunately, as we
observed from phage blocking experiments (Figure 3.5), our efforts to bring the DNAi OFF state
under control also had the consequence of lowering the maximal activity in the DNAi ON state.
However, a device with a maximal dynamic range in DNAi activity would enable the largest
number of possible applications, and so it would be worthwhile to investigate whether the
device's maximal induced activity can be increased without causing the device to fail in the OFF
state (i.e. 8-h target retention <95%).
The easiest way to accomplish this would be to add
51
We therefore constructed two
additional chromosomal copies of the actuator sequence.
additional DNAi device variants in which either two copies ('2x') or three copies ('3x') of the
actuator sequence (pACT-B) had been integrated into the host chromosome. These addition
integrations were targeted to the AaraC-araBAD and/or the APiaci:/aC/ loci, which had been
altered during the original parent strain's construction (Table A.2).
The kinetics of pTAR(S) target plasmid loss were measured for host strains harboring
either the
lXB,
2
XB,
or 3 XB pACT-B-derived actuator and spacer Y (pCR-Y, CmR). In line with our
expectation, increased actuator copy number correlated with an increased rate of target plasmid
Time Post-Activation (hr)
These
disappearance (Figure 3.13).
2
initial multi-copy actuator strains had
however,
so
and
they
(U
We
0-
wished to confirm whether boosting
profiles even further, and so we
10-4
CO
0
araC levels would improve their kinetic
.4-j
10-6
LA_
10-8
augmented the 3x actuator with the
stronger
pAraC-105
(p15A,
KanR)
accessory plasmid and then re-assayed
its performance. As expected, the rate
of
target
considerably
increased
knockout
(Figure
3.13,
red
strain, we constructed the 3x strain to
include an
auto-regulated
Pc-araC
cassette by integrating pACT-C at the
2x,
--
-amnem
3x,
3
10-10
+ pAraC-105
FiRure 3.13: Effect of actuator copy number on the kinetics of
target plasmid knockout A time course is shown for the knockout
of pSC101-based target plasmid pTAR(S) from DNAi strains
programmed with either spacer Y (solid line) or dual spacer Y+Z
(dotted line). These strains contain either a 1xB, 2XB, 3xB or 3xc
chromosomal actuator. Both the pAraC-105 accessory plasmid
(p15A, KanR) and the pACT-05-derived actuator encode araC
diamonds), and so similar to our expression
reconstruction of the lxc actuator
8
10-2
possessed sub-optimal araC levels and
were likely under-performing.
6
100
been built with the original PJ2311-araC
cassette,
4
cassettes
(PJ231os-araC
and
Pc-araC
cassette,
respectively) that significantly boost araC expression within the
host. Measurements reflect the fraction of cells that retain the
pTAR(S) target plasmid (Target') in the DNAi ON state (2mM
arabinose) and were performed via plate-based assay (Methods).
All data are the average of three independent measurements
performed on different days and error bars show the standard
deviation.
52
APLacI:/aC/ locus, thereby yielding and 3xc actuator strain.
Combining what we had learned from our series of kinetic optimizations, we assembled
a DNAi strain harboring the 3xc (v1.0) actuator, dual Y+Z spacers (pCR-YZ), and measured the
kinetics of pTAR(S) knockout under DNAi ON conditions. As predicted, the rate of plasmid loss
was the fastest yet observed (Figure 3.13, purple x's), slightly exceeding that of the accessorysupplemented, all-pACTO4 3x strain (red diamonds). A strain with an additional fourth copy of
the actuator sequence (pACT-C), which was integrated at the site of a ArhaSR-rhaBADM deletion,
was constructed, but its kinetic characterization was abandoned in favor of other efforts. The 3xc
actuator was decidedly sufficient for our subsequent investigations.
The rather dramatic >10 4 -fold increase in maximal plasmid knockout efficiency that
accompanied these actuator modifications was the focus of the experiments discussed in greater
detail below.
3.5
Optimization of DNAi device knockout efficiency
3.5.1
Introduction
Contemporaneous with our efforts to elucidate and optimize the DNAi device's relevant
kinetic parameters, we were engaged in parallel efforts to maximize its efficiency with respect to
target plasmid knockout. Based on our phage blocking experiments, in which as few as ~1 in 10s
transduction events was successful (Fig 3.5), we hoped to be able to exceed our original 99.9%
target plasmid elimination threshold by at least a few additional orders of magnitude or perhaps
more.
After examining our initial kinetic data, we noticed that all of our DNAi strains harboring
an integrated copy of pACT-C and its associated Pc-araC cassette always reached their minimum
value for target plasmid retention well before then end of the 8-h DNAi induction time-course,
even when the slowest spacer (Y) was used (Figures 3.10 & 3.13). We therefore opted to focus
our efficiency optimization efforts on this class of actuator variants only. In doing so, we could
be confident that our measurements were not kinetically-limited.
53
3.5.2
Design parameters affecting plasmid knockout efficiency
Our standard plasmid knockout efficiency assay was similar to our kinetic measurements.
Strains harboring a lxc or 3xc chromosomal actuator variant and a pSC101-based target plasmid
-
(pTAR(S), RFP+, StrR) were programmed with one or more on-target CRISPR spacers (pCR-X, -Y,
Z, -YZ, pUC19, CmR). A lXc actuator strain programmed with spacer N (pCR-N), which lacked a
cognate proto-spacer in either the target plasmid or the host genome, served as an off-target
negative control.
Cultures were then subjected to conditions that either induced (+2mM
arabinose, DNAi ON) or repressed (+0.5% glucose, DNAi OFF) DNAi activity for 8 h in the absence
of antibiotic selection for the target (-spectinomycin), at which point the fraction of targetpositive cells was measured via plating assay (Methods).
100.
10-1.
10-9
DNAi
OFF ON
CRISPR
N
Actuator
1xc
X
1xc
ON
OFF ON
OFF ON
OFF ON
OFF ON
OFF ON
Y
Z
Y+Z
X
1xc
1xc
1xc
3xc
Y
3xc
Z
3xc
Y+Z
xc
OFF ON, OFF ON
OFF
Figure 3.14: Systematic optimization of target plasmid knockout efficiency. Data for the knockout of target
plasmids are'shown for different spacers and numbers of actuators carried in the chromosome (lxc and 3xc). The
fraction of cells that retain the pSC101 origin pTAR(S) target plasmid (Target+) is shown in the OFF (repressed by
0.5% glucose) and ON (2mM arabinose) states. The OFF data and the 1x/N ON data were obtained using the
cytometry assay and the remainder were obtained by the plate-based assay (Methods). All data are the average
of three independent measurements performed on different days, and error bars are the standard deviations
We examined nine different DNAi strains in which proto-spacer location, spacer
multiplicity, PAM sequence, and actuator copy number were independently varied in order to
find the set of design parameters that maximized the device's knockout efficiency (Figure 3.14).
From a strictly practical perspective, our optimization was highly successful. The fraction of
target-positive host cells following knockout was reduced more than 10 4 -fold from ~2x10-4 for
each of the 1x actuator cases to less than 4x10-9 for the best-performing 3 xc actuator/Y+Z device.
In absolute terms, this fraction corresponded to a target-positive host cell titer of only 3x101
54
cfu/ml. Furthermore, we were also reassured to observe that our systematic enhancement of
DNAi ON state kinetics had not simultaneously increased leakage activity to the point where it
destabilized the device in the DNAi OFF state. In all eight on-target strains tested, including the
one believed to have the highest specific activity of DNAi (3xc actuator, Z CRISPR), 97-99% of host
cells retained the target plasmids for the necessary 8-h interval under DNAi-repressing
conditions.
3.5.3
Analysis of plasmid knockout survivors
While finding a set of design parameters that improved function was a relatively
straightforward exercise in combinatorics, extracting the relevant design principles underlying
the success of the 3xc/Y+Z device was more complicated.
For example, it was initially unclear
why certain alterations, such as increasing the spacer multiplicity from one to two (i.e. Y or Z vs.
Y+Z), improved performance when implemented in the context of the 3x actuator but not in the
context of the 1x actuator. It was also not immediately obvious why a similar discrepancy applied
to devices with a single non-coding target (X) versus a single essential target (Y or Z).
To determine the causes of this and other effects, we conducted failure mode analysis in
which we characterized individual spectinomycin-resistant clones that had survived our plasmid
knockout conditions with their target DNA intact. Specifically, we wished to determine whether
these survivors still possessed intact DNAi machinery or whether they were loss-of-function
mutants. In addition, if they were indeed mutants, we sought to determine which of the three
essential DNAi components (i.e. actuator, spacer, or proto-spacer) had been inactivated.
Unfortunately, the DNAi strain variants from our efficiency optimization experiments
carried their CRISPR and target components as two separate plasmids, which we feared would
complicate our genotypic analysis. We therefore opted to perform our failure mode analysis on
a set of analogous DNAi strains that instead contained just a single pSC101-based cis-targeted
plasmid. These cis-targeted plasmids encode both a constitutively expressed CRISPR array and
the corresponding valid proto-spacer sequence(s) (pCisTAR(S), StrR, RFP+, Figure 3.15A).
55
A
PJ2310
AX
pSC101
C
pCisTAR(S)-M
Y Z3
.C
M13
on
Y
CRISPR
PJ 23 117
N..+
B
100
4C
.
106
DNAi
Spacer
Actuator
Targeting
C
OFF ON
N
1xc
OFF ON
X
1xC
cis
cis
*
OFF ON J OFF ON
Z
Y
1xC
1xc
Target mutation2.x0
No mutation
Loss-of-function mutants:
1.3x10-
1.3x10-5
5
OFF ON
Y+Z
1.3x10-5
xc
OFF ON
Y+Z
3xc
cis
cis+trans
43/43
0/48
3
cis
cis
cis
cis
5.6x10-6 8.8x106
Actuator mutation
CRISPR mutationn
N J OFF ON
OFF
X
Y+Z
3xc
1xc
6
7.3x10-
A7N9
W
W
44/44
48/48
WL
48/48
W
48/48
42/42
Figure 3.15 Characterization of DNAi device functionality and escape mutants. (A) The map for the pCisTAR(S)-M set
of cis-targeting plasmids, which encode both a constitutively-expressed spacer ('M') and its corresponding protospacer, is depicted. The set is otherwise isogenic with pTAR(S). Spacer combinations are shown below, and their
sequences are provided in Supplementary Table A.1. (B) Data for the knockout of cis-targeted plasmid are depicted for
different spacers in combination with different actuators (genomic 1x and 3x). The fraction of cells that retain the cistargeted plasmid (Target') is shown in the DNAi OFF (repressed by 0.5% glucose) and ON (2mM arabinose) states. The
OFF data and the 1x/N ON data were obtained using the cytometry assay and the remainder were obtained by the
plate-based assay (Methods). All data represent the average of three independent experiments performed on different
days and the error bars are the standard deviation. (C) DNAi device failure modes for each on-target strain subjected
to DNAi ON conditions in (B) are shown. The particular loss-of-function mutation class was assigned based upon which
of a series of complementation plasmids did or did not restore a surviving clone's DNAi activity (Methods) using the
framework described in Table A.8. The numbers labeling each pie sector indicate the predicted fraction of total cells
from the experiment in (B) that contain the specified mutation, and this fraction was calculated as [Fraction target+
DNAi ON cells from (B)]x[# Clones containing specified mutation]/[Total # clones analyzed]. Pie charts are vertically
aligned with the corresponding strain in (B).
56
We harvested surviving clones (StrR, target+) from cis-only targeted DNAi strains encoding
X, Y, Z, or Y+Z spacer combinations that had been subjected to 8 h DNAi-inducing conditions
(Methods). Subsequent mutation analysis revealed that 100% of the isolates from all cis-only
targeted strains were loss-of-function that could not be re-induced (Figure 3.15C).
We next examined the statistics of individual component failures for each strain, and thus
made the key observation that under these conditions, the device's knockout efficiency is
ultimately dictated by the single DNAi component that is most likely to fail. Changing a specific
DNAi design parameter can indeed improve knockout efficiency significantly, but only when that
change lowers the probability of the dominant component's failure to a sufficient extent that
mutations in the next most failure-prone component now come to dominate, thereby
establishing a new, higher efficiency ceiling.
3.5.4
Design principles for optimizing plasmid knockout efficiency
We were able to determine that there are two main avenues through which one can lower
the probability of a DNAi component's failure. First, redundant copies of any component (i.e.
actuator, spacer, or proto-spacer) means that all copies must break down before the component
fails entirely, and such coincidences are less likely. Second, choosing proto-spacers located
within essential ORFs limits the number of possible mutations that are capable of both disabling
DNAi and leaving the cell viable, thereby lowering the frequency of proto-spacer inactivation.
In light of this cogent theory explaining the workings of our DNAi device, the somewhat
peculiar trends from our trans-targeted DNAi efficiency measurements highlighted above appear
less obtuse. Spacer multiplicity only improves performance in the 3xc actuator case, because for
all lxc actuator strains, the actuator is the most failure-prone component whereas in the 3xc
strains it is the least. Increasing actuator copy number from lxc to 3xc has large positive effect
on efficiency when a single essential sequence is targeted (proto-spacer Y or Z) but has very little
effect when a single non-essential sequence is targeted (proto-spacer X), presumably because
proto-spacer mutations become dominant in both 3xc cases and the probability of incurring a
DNAi-inactivating mutation is much more like to occur in proto-spacer X than in proto-spacers Y
or Z in a viable cell.
57
These design principles also empowered us to make forward predictions regarding
improvements in our device's performance.
For instance, we observed that the cis-targeted
3xc/Y+Z DNAi strain was ~1000-fold less efficient in removing the target plasmid than the
corresponding trans-targeted strain. With the aid of our escape mutant characterization, we
determined this discrepancy was likely the result of lower redundancy of the CRISPR spacer in
the former case (10-12 copies vs. "500 copies).
We therefore hypothesized that adding an
additional ~500 redundant copies of the Y+Z spacer combination to the cis-targeted strain
through complementation with our trans-targeting CRISPR plasmid (pCR-YZ, pUC19, CmR) would
restore the strain to maximal efficiency. Indeed, we were correct. After the (cis+trans)-targeted
3x/Y+Z strain was subjected to standard plasmid knockout conditions, the fraction of host cells
retaining the target plasmid was reduced from 1.9x10-6 to 2.0x10-9 (Figure 3.15B) This case also
demonstrates that if our design strategy reduces the individual probabilities of failure each
component low enough, the system is no longer mutation limited. Instead, of the 48 surviving
(cis+trans)-targeted clones tested, all 100% were transient escapees whose DNAi activity could
be re-induced (Figure 3.15C). As expected, a similar analysis of the equivalent trans-only targeted
3xc/Y+Z DNAi ON survivors again revealed that 100% of clones (48 of 48) were transient escapees
as well.
58
4.
Characterization of the optimized CRISPR DNAi device
4.1
Scope of DNAi device plasmid knockout functionality
4.1.1
Effect of target copy number on DNAi knockout efficiency and kinetics
Having arrived at an optimal design, the 3xc actuator coupled with a Y+Z spacer, we
wished to further investigate the DNAi device's properties and capabilities.
We began by
screening its activity against an array of similar targets with varying copy numbers. These target
plasmids
were
constructed
by substituting
the
pSC101
origin
of
replication
(10-12
copies/genome) present in the original target plasmid (StrR, RFP+) with either a pBAC (1-2
copies/genome), a p15A (20-30 copies/genome), or a pUC19 (>500 copies/genome) origin. All
four target variants were then transformed into a host containing the optimized DNAi device (3x
actuator, Y+Z CRISPR), and the resulting strains were subjected to plasmid knockout conditions
(2 mM arabinose, -spectinomycin) for 8 h, during which time the fraction of host cells retaining
the target plasmid was measured periodically.
Under these conditions, certain elements of the performance of the optimized DNAi
device were largely independent of copy number (Figure 4.1A). The overall efficiencies of target
plasmid knockout after 8-h induction for pBAC- (8.9x10 1 0 ), pSC1O1- (2.9x10-9 ), pl5A- (4.3x10-9 ),
and pUC19-based (1.5x10-9) targets were all comparably high. In addition, the kinetic profiles of
DNAi-mediated knockout for each target were reasonably similar.
Once the
PBAD
promoter
activates after 2 h (Figure A.8), plasmid knockout was quite rapid, and the largest change in the
fraction of cells retaining the target occurred within the first hour following the onset of
observable plasmid loss. While the precise rate of this plasmid loss did appear to be somewhat
inversely related to target copy number, the effect was only prominent for the very-high-copy
pUC19-based target. It should be noted, however, that preliminary experiments with a much
less active DNAi device (1xB/Y CRISPR) suggested that target copy number exerts a much stronger
effect on DNAi kinetics in this slower regime (data not shown).
59
A
Time Post-Activation (hr)
8
4
6
2
0
B
102
pSC101 Target
pBAC Target
p15A Target
10-2
LPBAC
-
zl
Pci9
...
p15A1
p
I
UOFN
pUC19 Target
-2
ff
106
I
I03
I
101
DNAI
spacer
OFF ON
N
OFF
ON
Y+Z
OFF ONIOFF ON
N
Y+Z
OFF ON OFF ON
N
Y+Z
ON
OFF
N
OFF ON
Y+Z
Figure 4.1 Plasmid knockout kinetics and target DNA depletion using the optimized DNAi device. (A) A time
course is shown for the knockout of target plasmids (pTAR('X'), StrR, RFP ) containing different origins of
replication. The inducer (2 mM arabinose) is added at t = 0 hr and time aliquots are analyzed using the plating
assay to determine the fraction retaining the target plasmid (Methods). The vertical dashed line indicates t =
2.25 hr, the time at which the PBAD promoter activates to initiate the synthesis of DNAi actuator components
(Figure A.8). All of the data were gathered using the 3x DNAi device and the Y+Z dual spacers. When the target
contains the pUC19 origin, the plasmid containing the targeting array was changed to a compatible RSF origin
(pCR-YZ* and pCR-N*). (B) The recovery of target plasmid DNA sequences via PCR after the 8 hr induction of DNAi
as performed in (A) is quantified via qPCR and is given in terms of absolute copy number per milliliter of culture
(ACNculture; Methods). All data represent the average of three independent experiments performed on different
days and the error bars are the standard deviation.
4.1.2
DNAi-mediated depletion of target plasmid DNA
As we had originally envisioned using the DNAi device to prevent the escape or theft of
proprietary or dangerous genetic information, we next examined the ability of the DNAi to render
the target DNA sequence more difficult to recover via PCR-based methods. The optimized ontarget (Y+Z) and off-target (N) DNAi strains containing targets with varying copy number as
described above were subjected to DNAi activation (2 mM arabinose) or repression (0.5%
glucose) in the absence of selection for the target plasmid (-spectinomycin) for 8 hr. Total
plasmid DNA was then isolated from each sample, and qPCR was used to quantify the amount of
recoverable target DNA sequence in terms of absolute copy number (Figure 4.1B, Methods). As
expected, no appreciable reduction in the amount of PCR-recoverable DNA sequence was
observed upon device activation for all off-target variants (CRISPR N). In contrast, when the DNAi
was properly targeted (CRISPR Y+Z) device activation drastically diminished the amount of
recoverable DNA present in the sample relative to that present in the corresponding repressed
sample. The fold-reduction in copy number between the DNAi OFF state and the DNAi ON state
60
was 1.1x104 , 4.6x10 4, 6.8x10 2, and 1.2
x10
4
for the pBAC-, pSC101-, pl5A-, and pUC19-based
targeted, respectively. Thus, the DNAi not only enables the controlled and precise alteration of
host cell genotype and phenotype, but also represents a way of destroying the very DNA
sequence information underlying that genotype.
4.2
Impact of the DNAi device on host growth and genetic stability
4.2.1
Effect of DNAi activity on host growth
Given the properties of the DNAi actuator cas components to bind and cleave DNA, off-
target catalytic activity could potentially result in toxicity manifested as slower growth of the
We assessed the toxicity of the DNAi ON state by measuring cell density (in cfu/ml)
host.
following 8 h of continuous growth in the presence of varying concentrations of inducer (Figure
3
4.2A). The measurements were based on the xc actuator and contained the Y+Z dual spacers
B
A
24
2.5 -
-.
2.0
22
-
0
1.5 4
1.0
18
0.0
-
0
/
-
,
- , - 10-1
10-3
Ara (mM)
' -
,16
101
0
,
,
-- , ,10-1
10-3
Ara (mM)
-
0.5
101
Figure 4.2: Growth impact of strains carrying the DNAi device. (A) The cell density is shown as a function of
inducer (0-10 mM Ara) for cells containing either the 3x DNAi (black circles) device or no actuator (gray squares),
along with the Y+Z dual CRISPR targeting array (pCR-YZ, CmR), and a target plasmid (pTAR(S), RFP , StrR).
Measurements of viable cell titers in cfu/ml were made by plating serial dilutions of cultures onto solid media
without selection for the target plasmid (Methods). (B) Doubling time (in min) as a function of inducer (0-10
mM Ara) for host strains carrying a Y+X CRISPR plasmid (pCR-XY, CmR) and either the 3x DNAi device (black
circles) or no actuator (gray squares). Data reflect the least-squares fit of an exponential growth model prepared
from OD6 00 values measured at 3-min intervals between 60 and 90 min post-induction (Methods). Arrows
indicate 2 mM arabinose, the inducer concentration at which all plasmid knockout and cell killing experiments
are performed. The data are the average of three independent experiments performed on different days and
the error bars are the standard deviation.
61
and the pTAR(S) targeting plasmid, which was being knocked out during the time course. Activity
was well tolerated at the inducer concentrations used for the experiments in this manuscript (2
mM). Even upon maximal induction (10 mM), less than a 40% reduction in culture density was
observed. Furthermore, the rate of cell growth was minimally affected as well. When analogous
growth rate measurements were performed comparing the 3xc DNAi strain to the actuator-free
strain, the maximal difference in doubling time was <1.5 min (Figure 4.2B)
4.2.2
Long-term stability of DNAi device function
If the DNA targeted for DNAi knockout contains an essential gene, then even properly-
targeted DNAi activity creates a toxic growth effect and thus a selective pressure against its own
maintenance. Given enough time, even a slight pressure from the device's leakage activity might
result in the accumulation of loss-of-function mutations that inactivate the device entirely. The
long-term stability of DNAi performance in plasmid knockout experiments was therefore tested
over many generations in continuous liquid culture. Cells containing the optimized 3x/Y+Z DNAi
device and a pSC101 target were repeatedly passaged in the DNAi OFF state (0.5% glucose) with
selection for the target plasmid (+spectinomycin) every 12 h for 90 days. At 48-h intervals, a
sample of the culture was removed and subjected to a standard 8-h plasmid knockout
experiment in which the DNAi was either induced (2 mM arabinose, DNAi ON) or repressed (0.5%
glucose, DNAi OFF) under conditions allowing for the loss of the target plasmid (-spectinomycin).
Initially, the device performed as previously characterized; less than 1 in 108 of cells in the ON
state and >98% of cells in the OFF state retained the target (Figure 4.3). This optimal level of
performance was observed through day 21, which corresponded to approximately 350 doublings
and 101 07-fold total amplification of the initial sample based on OD600 measurements. After this
point, the plasmid knockout efficiency in the DNAi ON state did begin to decline, albeit
exceptionally slowly. Even towards the end of the 90-day experiment, the device was still largely
functional. During no single 14-day interval over the course of the entire experiment did the
average knockout efficiency in the DNAi ON state ever decline to a point where greater than 1 in
107 host cells retained the target plasmid.
Furthermore, the OFF state remained completely
stable, and the average fraction of cells retaining the target plasmid never dipped below 98%. By
62
the end of the full 90 days, which corresponds to a remarkable 1700 cell divisions, we had greatly
exceeded our original goal of stably maintaining DNAi device performance over 100 cell
generations. The fact that the continual passaging over this period also corresponds to a 10s20_
fold amplification of the initial inoculant made us more confident that a strain harboring the DNAi
device could be cultured to populate even the absolute largest of industrial fermenters with little
worry about compromising the strain's integrity.
Time (d)
0
I
.
.
I
.
I
70
60
50
40
30
20
10
.
a
.
90
80
a
10-2.
C 10-6.
I-U
LA.I
I01.Z
.
0-10-III
Figure 4.3: Long term plasmid knockout performance of the DNAi device in continuous culture. (A) The activity
of the DNAi device was characterized periodically for three months. A strain containing the 3x DNAi device and
the Y+Z dual spacer (pCR-YZ) and pSC101 target plasmid (pTAR(S)) was passaged every 12 hours under conditions
where the device is OFF (0.5% glucose) (Methods). Every 2 days, aliquots were taken and analyzed via the
These
cytometry assay to determine the fraction of cells containing the target plasmid (Target') (white circles).
was
plasmid
target
the
of
fraction
the
and
hours
8
for
samples were then induced with 2 mM arabinose
dashes).
(black
determined via plating assay (Methods)
4.2.3
Long-term stability of host metabolism while carrying the DNAi device
Ultimately, the DNAi device was intended to be used in combination with other
engineered biological systems. We therefore sought to determine whether the device might
Protein overdestabilize the performance of other genetic programs running in parallel.
expression is one hallmark of synthetic biochemical pathways, and so over the course of our 90day experiment, every time we assayed for DNAi performance, we also performed a parallel
cytometric measurement of host RFP levels, which were generated from the pTAR(S) target
plasmid's constitutive expression cassette, when the device was in its OFF state. As hoped, the
average expression was exceptionally stable. If one does not include the three outlying points,
63
which correspond to technical errors in measurement rather than aberrations in strain
performance, then the relative standard deviation of mean RFP expression at the end of the full
90 days was <1.2% (Figure 4.4A). In addition, as the corresponding RFP histograms show, this
remarkable consistency in the mean RFP value does not mask a significant shift in the populationwide distribution of expression levels (Figure 4.4B). When combined with the data regarding the
DNAi device's stable performance, these data strengthen the premise that the DNAi device in its
OFF state represents a very small metabolic burden to its host.
B
A
106
Day 1
00000 0
105
0000000000000000000000000 00000000000
-Ci
0
o
Day 31
0
UDay
U_ 103
L.
Day 11
ADay 21
41
Day 51
103
C102_____
EDay 61
101
Day 81
IDay 91
1001
0
10
20
30
40
50
60
70
80
90
100
101
102
10104
10s
106
RFP Fluorescence (A.U.)
Time (d)
Figure 4.4: Stability of target plasmid protein expression during cell passaging experiments. (A) Median RFP
expression level for host cells containing the 3x DNAi device, a Y+Z CRISPR (pCR-YZ, Cm ), and an RFP+ target
plasmid (pTAR(S), StrR) taken every 2 days as the sample is repeatedly passaged for 90 days in the DNAi OFF state
(+0.5% glucose) (Methods). Outlier data points were due to technical error. (B) An overlay of RFP fluorescence
histrograms corresponding to single cytometry measurements taken every 10 days during the experiment
described in (A). Each histogram comprises >10,000 cells.
4.3
Targeting of DNAi activity to the host genome
4.3.1
Using the DNAi device to effect cell death
In addition to its ability to cure plasmids, CRISPR-mediated DNAi activity can be directed
against the host genome to kill the cel1 59,70. In order to determine whether the DNAi could be
used to effect cell killing in a user-controlled manner, we constructed host strains containing the
3x genomic actuator along with CRISPR targeting plasmids encoding either a single genome-
64
targeted spacer (either G, or G 2) or an off-target spacer (N). The G 1 spacer was selected to target
only a single chromosomal sequence that is directly adjacent to the actuator at the ACRISPR-cas
locus. The G 2 spacer targets the attL sequence present in all three of the actuators, which are
C
B
A
Time Post-Activation (hr)
8
6
4
2
0
CRISPR
G2
GI
N
Chromosomal Targets
+1
*
oriC
G2
100
19.
c-
0
0
Z
1.)
10-4
101.
E colistr. K-12
MG1655
.U
Zi
14'
10-6
G2o '11
G2
103.107.
V-~
.2
101
1-8.
0
8
6
4
2
Time Post-Activation (hr)
Figure 4.5: Cell killing dynamics upon targeting of Cas activity to the genome. (A) Two on-target spacers (GI and
G 2 ) are shown along with the locations of their corresponding sequences in the E. coli K12 host chromosome. The
G 2 spacer targets a site within the chromosomally-inserted actuator sequences (starred) and so those proto-spacer
locations coincide with the three actuator locations . An off-target spacer (N) is used as a control as it does not
target a sequence in the genome. The spacing from the G, target for the qPCR assays in Figure 4.6 is shown. (B)
The kinetics of cell killing is shown, where the efficiency is presented as a ratio of the titer of viable cells between
the DNAi ON (+2mM arabinose) and OFF (+0.5% glucose) states. The data are shown for the 3x actuator and the
G, (pCR-G1) or G 2 (pCR-G2) targets (green diamonds and orange squares, respectively). The N control (pCR-N) is
shown as blue circles. (C) The data from (B) expressed in terms of viable cell titers (cfu/ml). Vertical dashed lines
indicate t = 2.25 hr, the time at which the PBAD promoter activates to initiate the synthesis of DNAi actuator
on
components (Figure A.8). All data represent the average of three independent experiments performed
deviation.
different days, and error bars indicate the standard
separated by chromosomal distances of 300 kbp and 2.8 Mbp, respectively. Both the G 1 and G 2
spacers are oriented such that their respective crRNAs basepair with the genomic (-) strand upon
target recognition. The relative chromosomal locations and multiplicities of the Gi and G2 protospacer targets are shown in (Figure 4.5A). With respect to the G2 proto-spacers, the genomic
distance between sites varies between 300 kbp and 2.8 Mbp, and as such each intervening
segment contains many essential host genes.
65
These on-target and off-target strains
Time Post-Activation (hr)
were subsequently induced with 2 mM arabinose
(
)
(DNAi ON) or repressed with 0.5% glucose (DNAi
1004
OFF), and over an 8 h interval the ratio of viable
cells for a given DNAi condition was periodically
measured as the cell titer of the genome-
-
4
6
8
LA_.
0
z
0
10-4
targeted sample (Gn) divided by that of the offtarget sample (N) (Figure 4.5B). For both the G 1
2
0
10-
and G 2 - targeted strains in the DNAi ON state,
host cell death was highly efficient, and the
terminal viable cell ratios for the G 1 and G 2
samples were 7x10-7 and 8x10~9 , respectively.
These values corresponded to exceptionally low
3
absolute viable cell titers of 7x10 cfu/mL for the
-
Gi-targeted strain and 9x10 1 cfu/mL for the G 2
targeted strain (Figure 4.5C). Similar to DNAimediated plasmid knockout, cell death was also
quite rapid following its onset, with the largest
change in the ratio of viable cells occurring
between 2 and 3 h post-induction. Furthermore,
after 8 hrs in the DNAi OFF state, the viable cell
ratios corresponding to both genome-targeted
samples were slightly greater than 1.
This
Fiaure 4.6: DNAi-mediated cell killing in the
presence and absence of antibiotic selection.
The kinetics of cell killing is shown, where the
efficiency is presented as a ratio of the titer of
viable cells between the DNAi ON (+2mM
arabinose) and OFF (+0.5% glucose) states. The
data are shown for the 3x actuator and the G 2
(pCR-G2, CmR) targets as determined by cell
death assays (Methods) performed by plating
without
solid media
with
and
onto
chloramphenicol (orange squares and black
triangles, respectively). The off-target negative
control (pCR-N, CmR) is shown as blue circles.
The vertical dashed line indicates t = 2.25 hr, the
time at which the
PBAD
promoter activates to
initiate the synthesis of DNAi actuator
components (Figure A.8). All data represent the
average of three independent experiments
performed on different days, and error bars
indicate the standard deviation.
suggested that genome-targeted DNAi leakage activity, which might have otherwise resulted in
on-target host cell death and slower net cell growth relative to the off-target control, was
negligible. Targeting DNAi activity to the genome thereby killing the host and curtailing its
biochemistry therefore enables yet another viable and efficient means of halting the
amplification of all DNA sequences harbored within the cell.
66
4.3.2
Analysis of cell death assay survivors
The aforementioned measurements quantifying the efficiency of DNAi-mediated killing
were performed in strict accordance with our standard cell death assay, in which both the liquid
media used during DNAi induction and the plates used to assay cell titer contain chloramphenicol
in order to select for the CRISPR targeting plasmid (pCR). As the use of antibiotics is often costprohibitive in large-scale fermentations, we were curious as to how the absence of this selection
would affect the DNAi device's performance.
The G2-targeted strain was assayed once more
under identical experimental conditions except that in addition to the normal plating with
chloramphenicol selection, plating without selection was performed in parallel. Somewhat
surprisingly, while the apparent efficiency of killing as measured with the +chloramphenicol
plates reached its usual value, the apparent efficiency of killing in the same samples as measured
with the non-selective plates was more than a factor of 104 lower (Figure 4.6). This ratio of
efficiencies implied that >99.99% of colonies on the non-selective plates had lost their
chloramphenicol resistance marker and the linked CRISPR array. This was confirmed by colony
PCR (data not shown). We concluded that when implementing a DNAi device under non-selective
conditions, the probability of plasmid loss is sufficiently high that it can become yet another
mechanism for incurring a DNAi efficiency-limiting mutation.
We next wished to determine whether the DNAi device's failure modes for cell killing
were similar to those observed in our previous plasmid knockout experiments. Operating on
the assumption that they were indeed similar, we formed a number of hypothesis about the
composition of the survivor populations resulting from the Gr and G2-targeted experiments
prior to beginning our investigations. First, because the efficiency of Gr-targeted DNAi activity
was almost 100-fold lower that of G 2 -targeted killing, we projected that 100% of the surviving
clones that we screened would be loss-of-function mutants. Furthermore, because of the
redundancies built into both the actuator (3 copies/cell) and the CRISPR spacer (pCR-G1, pUC19
ori, >500 copies/cell) that lowered their respective probabilities of failure, we also hypothesized
that 100% of the G 1 survivors would have mutations in the chromosomal proto-spacer target (1
copy/cell). Indeed, both of our hypotheses were correct (Table 4.1). Of the 108 clones
screened, none could be re-induced to effect cell suicide, and all 108 had a noticeable defects
67
at the G, locus that would be expected to render DNAi at that target inactive given what is
known about the stringency of crRNA-proto-spacer base-pairing requirements (Methods) 5 1. For
most of the clones (82 of 108), this defect was rather large; a ~450-bp amplicon roughly
centered on the Gi spacer failed to amplify by colony PCR suggesting that a large deletion of at
least a few hundred base pairs had occurred. For the remaining 26 of 108, colony PCR
succeeded, and so the G, locus was sequenced using these amplified products. As previously
implied, all 26 of the sequenced clones had noticeable defects in G 1. Some were point mutants,
a few possessed sizeable insertions disrupting G 1, but most were smaller deletions within or
around the proto-spacer. The overwhelming preponderance of deletions in the region is
consistent with, but not conclusive proof of, the notion that the host attempts to repair the
double-stranded break created at the proto-spacer subsequent to DNAi-mediated cutting. Of
course, if the repair successfully re-circularizes the genome before an essential gene is
damaged while simultaneously damaging the proto-spacer sequence in the process, then the
cell can survive indefinitely as an escape mutant.
Table 4.1: Characterization of suicide assay survivors"
Strain
DNAi Phenotype
Escape Mutations
Actuator
CRISPR
TClones
Transient
Mutatione
Actuatord
CRISPR'
Proto-spacerf
IndeterminateQ
3x DNAi
Gi
108
0 (0%)
108 (100%)
0
0
108
0
3x DNAi
G2
105
82 (78%)
23 (22%)
0
0
6
17
a.
b.
c.
Surviving colonies following 8-hr DNAi induction (see Methods "Suicide Assay")
Transient escape survivors are able to re-initiate cell killing activity upon a second 8-hr induction of DNAi activity
Escape mutation survivors are unable to re-initiate cell killing activity upon a second 8-hr induction of DNAi activity
d.
e.
Actuator escape mutants are unable to knock out CRISRP/target hybrid plasmid pSELF-X upon DNAi activation (Methods)
Escape mutants are identified via colony PCR of the relevant DNA followed by direct Sanger sequencing.
f.
g.
Includes inactivating mutations in PAM sequence
Location of escape mutation could not be determined due to confounding growth defects
Reassured by the success of our predictions for the Gi-targeted case, we made another
prediction for the G2-targeted sample. Given that a certain level of redundancy had been built
into all components (3 copies actuator; >500 copies spacer; 3 copies
G2
proto-spacer), we
hypothesized that all of our survivors would be re-inducible, similar to what we observed for
knockout of pTAR(S) with optimized DNAi device. Unfortunately, this time we were only partly
68
correct (Table 4.1). Of the 105 clones examined, when DNAi was re-induced under the usual
conditions, 82 were indeed transient escapees (>100-fold drop in titer relative to DNAi OFF state)
rather than loss-of-function mutants. Furthermore, for the remaining 23 clones, phenotypic
determination of the host's DNAi status it was difficult due to confounding growth defects, which
possibly could have resulted from DNAi-mediated chromosomal damage). Nevertheless, even
though 17 of these clones were ultimately unclassifiable, through colony PCR and sequencing it
was determined definitively that the remaining 6 contained DNAi-disabling mutations at all three
G 2 proto-spacer sites. This too is consistent with our theory of post DNAi repair and resultant
damage to chromosomal sequence, as the odds of three different sites along the genome
acquiring these mutations independently is certainly lower than the observed frequency of this
genotype. Of course, if DNA cutting and repair are ultimately shown to be part of the biochemical
forces shaping the performance of the DNAi device, this information must be factored into any
subsequent designs.
4.3.3
DNAi-mediated depletion of chromosomal DNA
One of the primary advantages of using a targeted endonuclease is that specific regions
of the genome can be targeted for degradation. This could be used to remove preferentially DNA
associated with a highly engineered region or insertions containing synthetic pathways. Because
the host's endogenous exonuclease activity combined with that specific to Cas3 presumably
100.
0-.
cc
E0
Z z
-<
-
10-_.
104.
Spacer N G, NI Gj NI G1 NI G 1 NI GI
1kbp I 10 kbp 100 kbp I Mbp
Dist. to G1 0 bp
Figure 4.7: DNA recovery as a function of distance from
proto-spacer. The ability to recover chromo-somal DNA
sequences via PCR after the induction of DNAi for 8 hours
is quantified as a function of distance from the single
each
For
target.
proto-spacer
chromosomal
is
recovery
of
efficiency
the
chromosomal locus,
measured as the ratio of the relative copy number (RCN)
of the DNAi ON state divided by that of the
corresponding DNAi OFF state (Methods). Chromosomal
distances are measured as the center-to-center spacing
between the PCR amplicon indicated and the singlyoccurring G1 chromosomal protospacer (NCBI
NC_000913.3 position 2,887,466). Sequences for primer
pairs are provided in Supplementary Table 5. All data
of three independent
represent the average
experiments performed on different days, and error bars
indicate the standard deviation.
69
chews the DNA target from both ends of the initial cut, we expected that, upon chromosomal
targeting of DNAi activity, the DNA sequence corresponding to the region near the targeted
protospacer would be significantly more difficult to recover. We applied a qPCR-based assay to
test this hypothesis using DNAi activity targeted with the G, spacer, which is centered on position
2,887,466 of the E. coli str. K-12 genome (NCBI numbering), and compared it to the control (N).
Both on- and off-target samples were induced, and qPCR was used to amplify five chromosomal
loci at increasing distances (0 bp, 1 kbp, 10 kbp, 100kbp, and 1 Mbp) from the G 1 cut site (Figure
4.7, Methods). When comparing the DNAi ON to the DNAi OFF state in the case of the G1 spacer,
the amount of DNA remaining from the region surrounding the target (0-1kb) is reduced >103fold. By 10kb, there is ~3.3-fold more DNA, but there is still significant loss. By 1MB, the amount
of DNA is equal to that which is recoverable from the N control, whose DNA recoverability was
not appreciably altered by the DNAi device's ON/OFF state at any distance from the G 1 protospacer. This demonstrates that the specificity of the Cas complex can be applied to selectively
target regions of the genome while leaving non-targeted regions intact.
Notably, all three copies of the DNAi actuator are entirely encoded within 10kb of a G 2
spacer. More specifically, the cas3 gene essential to DNAi function is within either 0.5 kbp (pACTA/B) or 1.5 kbp (pACT-C).
It is thus exceedingly likely that during the G2-targeted cell killing
experiments, the vast majority of host cells either irreparably damaged or completely degraded
all three copies of their own actuator sequences before submitting to cell death.
We also
previously demonstrated that cis-targeted plasmids (pCisTAR(S)-M), in which both a self-targeted
CRISPR spacer and its cognate proto-spacer are encoded on the same plasmid within 1 kbp of
one another, are knocked out in a DNAi-dependent manner with efficiencies equivalent to the
corresponding trans-targeted case. Taken together, these experiments demonstrate that the
cas/CRISPR system is capable of targeted degradation of its own genetics.
4.3.4
Orthogonal cell death mechanisms increase killing efficiency
Finally, given that DNAi-mediated cell killing efficiency was improved considerable when
we designed redundant proto-spacer targets to avoid inactivation mutations, we hypothesized
that the efficiency of cell killing could be similarly improved by layering DNAi-mediated
70
chromosomal degradation with another redundant method for cell killing. To test this we
designed and implemented a second site-specific integrase-mediated killing strategy that
65
employed the Int5 site-specific serine recombinase as its actuator . The rather atypical design
was as follows. Beginning with an older 1xB chromosomal actuator strain, we proceeded to
integrate an additional synthetic DNA sequence encoding both a kanamycin resistance cassette
and our standard proto-spacer X coupled to an 5'-ATG PAM into the genome at the AaraCaraBAD locus.
This "loop-out" cassette is
Time Post-Activation (hr)
0
flanked by the recombinase's cognate attBints
and attPints sites, oriented in parallel such that
2
6
4
8
100
initiation of the site-specific recombination
results in excision of the kanamycin resistance
C
gene and thus cell death in the presence of
kanamycin.
C
A dedicated plasmid-based actuator that
expressed the Int5 recombinase (plnt5act, p15A,
AmpR)
was
constructed
to
activate
10-2
10-4
cc
10-6.
10-8-
the
integrase-mediated cell death pathway, while the embedded lxB actuator could be programmed
with spacer X (pCR-X, pUC19, CmR) to effect
DNAi-mediated cell death.
For convenience,
both actuators were coupled to PBAD promoters
Figure 4.8: Combining DNAi-mediated and
intergase-mediated cell killing methods in a single
host. The kinetics of cell killing is shown for host
strains harboring both a 1x chromosomal actuator
(pACT-B-derived) and an Int5-targeted attBint 5-KanR-
so that both systems could be switched into
X-attPnW 5 chromosomal loop-out cassette along with
the addition of
cassettes for the expression of either spacer X (pCRX, CmR), Int5 (plnt05at,, Am pR) or both. Upon
their ON
states
through
arabinose (2 mM) and into their OFF states
through the addition of glucose (0.5%).
The aforementioned integrase-sensitive
base strain was then transformed with the
appropriate sets of plasmids to effect cell death
either through integrase-mediated killing alone
(plnt5act), through DNAi-mediated killing alone
71
induction, killing was initiated via chromo-sometargeted DNAi activity (blue triangles, pCR-X),
integrase mediated looping-out of an essential gene
(green
squares,
plnt5+),
or
both
methods
simultaneously (black circles, pCR-X+/plnt5act). All
data reflect the ratio of viable cells between the ON
(2 mM arabinose) and OFF (0.5% glucose) states
when plated on kanamycin. Measurements were
performed via plate-based assay (Methods). All
data reflect the average of three independent
measurements performed on different days and
error bars show the standard deviation.
(pCR-X), or through both pathways simultaneously (plnt5act and pCR-X). Over an 8-h period, the
fraction of viable, kanamycin-resistance cells was measured as the ratio of cell titers between the
devices' ON states (2 mM arabinose) and OFF states (0.5% glucose) in accordance with an
adapted version of our cell death assay (Methods). As expected, while both DNAi-mediated
killing alone (Fig 4.8, blue triangles) and integrase-mediated killing alone (green squares) were
indeed effective at leaving only a small fraction of the host cells viable (4.3x10-5 and 6.3x10-4
respectively), the combination of the two killing modes simultaneously improved overall
efficiency to where less than 3 in 108 cells survived induction (black circles). Thus, programming
the host with redundant mechanisms of cell killing does indeed increase the robustness of the
suicide response.
72
5.
Discussion and Conclusions
Having precise, dynamic control over the DNA contents of a living cell represents a
potentially powerful means of manipulating biological systems for a wide variety of applications.
To that end, we have exploited the CRISPR-mediated DNA interference phenomenon to design
and construct a genetically-encoded DNAi device that degrades a set of user-specified DNA
targets in a live E. coli host in response to a chemical input signal. The target DNA is completely
stable when DNAi is in the OFF state but is rapidly and efficiently degraded when the device is
switched into the ON state. Through an extensive set of parameterization and optimization
experiments, we have determined which particular design specifications are important for
maximizing both the kinetics and the degradative efficiency of CRISPR-mediated DNAi activity
while minimizing its impact on the host's underlying metabolism. As a result, the optimized
device is effective at knocking out plasmid targets with a wide range of copy numbers,
demonstrates very little toxic effect on the host's growth or ability to over-express proteins, and
is robust to continuous culture over hundreds of generations. Finally, we have shown that when
DNAi activity was targeted to sites on the host chromosome, host cell death occurs, and the
sequence surrounding the target site is rendered more difficult to recover.
We have deliberately focused much of our experimental inquiries and optimization
efforts on improving the DNAi device within the context of information security and
biocontainment applications given the growing needs therein. For many biotechnology firms,
engineered DNA constructs can be valuable assets that can confer a competitive advantage in
the marketplace, but protecting these assets by maintaining trade secrecy with respect to their
sequence information is a task fraught with unique challenges.
First and foremost, the
fundamental imperative of living entities to propagate their genetic material enables techniques
such as cell culture and PCR that allow researchers to amplify samples as small as a single cell or
nucleic acid fragment with high-fidelity into sufficient material to obtain reliable sequence
information. This particular facet of biology has long been recognized as a potential avenue to
procure genetically modified samples without the original developers' consent.
During the
1950s and 60s, at the height of phage research, investigators toyed with the idea of obtaining
73
samples of mutant phage variants from competing groups by swabbing the letters and envelopes
.
used for written correspondence between the two labs 71
The combination of next-generation sequencing and rising large-scale gene synthesis
technologies could undermine attempts at trade secrecy even further. For example, it may soon
prove unnecessary to possess even a single complete and/or viable physical sample in order to
replicate a piece of genetically-engineered biotechnology.
Instead, incomplete fragments of
DNA, potentially as short as tens of base-pairs and derived from a collection of disparate, nonfunctional sources, can be sequenced, re-assembled in silico, disseminated globally, and then
synthesized de novo and transformed into an appropriate host on site to recapitulate perfectly a
piece of proprietary technology. Under such conditions, the abstract sequence information itself
harbors the value independent of how it is encoded. One documented episode of informationbased theft with noteworthy economic repercussions occurred between 2007 and 2010, during
which time a corporate spy was selling trade secrets he had stolen from two of his employers,
American biotech firms Dow AgroSciences LLC and Cargill Inc., to competing Chinese corporate
and academic interests. Among the secrets stolen from Cargill was an electronic DNA sequence
for a proprietary enzyme, with the associated loss valued at $7 million (USD) 7 1.
One worrisome implication of these occurrences is that many entities exiting a research
laboratory, whether they are electronic communications, waste streams, or finished products,
can potentially harbor valuable sequence information would undermine trade secrecy and
therefore may represent significant economic liabilities. Thus, there appears to be real value in
technologies, such as our DNAi device, that can be used to "decontaminate" laboratory effluent
and remove this sequence information, thereby mitigating these risks.
In its current
manifestation, the DNAi device could be employed to protect sensitive sequence information in
a manner analogous to how a paper shredder destroys printed information prior to disposal.
Following fermentation, the user actively intervenes and supplies a chemical input signal, which
then switches the device into the ON state and results in sequence degradation within the
sample. The resulting cellular material, with much of the sensitive information removed, can
then be discarded or recycled as necessary with greater confidence that it will not be exploited
by competitors.
74
With a specific application in mind, the question then becomes whether or not the
performance of our optimized DNAi device is up to such a task. Our data suggest that the DNAi
device reduces the concentration of targeted DNA >300-fold out to 10 kbp from the site of the
initial DNAi-mediated cut, and a good read depth for a dedicated, single-lane MiSeq-type
microbial genome re-sequencing experiment is ~100-300 reads. Thus, even under the most
laboratory-like conditions in which the engineered host strain of interest is obtained "fresh from
fermentation," our device's action will likely cause considerable difficulty in the de novo
reconstruction of the targeted sequence compared to non-targeted regions. Nevertheless, we
anticipate that our device's action will have the greatest impact on samples obtained from the
greater environment, where
DNA degradation
from
environmental
stress and
high
concentrations of background DNA will further obscure the signal from the low-concentration
targeted regions. In fact, compared to a similar DNA degradation device constructed from a nonspecific nuclease that degrades all DNA equivalently, the targeted nature of DNAi-mediated
nuclease activity exploits this competitive inhibition effect inherent to "unbiased" sequencing
methods to a greater extent.
In more practical terms, eking out a minimum read depth of 1 across the sequence of
interest may be sufficient to extract all, or at least the vast majority, of the information encoded
therein. A ~300-fold depletion of DNA to counteract a read depth of 300 does suggest that
perhaps the optimized device's performance still teeters on the tipping point for usefulness. An
additional 10- or 100-fold improvement in the efficiency of chromosomal DNA elimination might
therefore be necessary to demonstrate truly robust function in real-world applications. Given
our extensive knowledge of the DNAi device's workings and limitations, however, we can
conceive of a number of design solutions that might confer the necessary enhancements. For
instance, tighter spacing of multiple proto-spacers blanketing a target sequence might create the
necessary redundancy to reduce the probability of overall failure at that locus.
In addition,
because it appears that DNAi-mediated degradation competes with the host's error-prone
attempts to repair the resulting double-stranded break (DSB) in its genome, targeting the rec
genes responsible for this repair might result in more thorough and persistent exonuclease
degradation".
Finally, deliberately flanking targeted regions with stretches of repeated
75
sequences and/or chi recombination hot-spots might incentivize this repair process to recombine
out the necessary sequences following DSB formation.
Of course, there is no way of knowing for certain whether or not the current performance
of the optimized 3x DNAi device or any reasonable derivative is sufficient to meet the needs of
our proposed DNA sequence "paper shredder" application without first empirically testing its
function under appropriate real-world environmental conditions.
Unfortunately, those
experiments were beyond the scope of this work, but they are certainly reasonable ones to
perform as part of any subsequent endeavors.
Ultimately, with respect to the protection of proprietary sequence information from
theft, it is likely that our original paper shredder analogy can be carried one step further. With
sufficient skill, patience, and diligence, any DNAi-treated sequence can presumably be
reassembled from partial samples much like paper shreds can be meticulously reconstructed into
a full document.
Whether partial DNAi-mediated degradation alone proves a significant
deterrent against illicit efforts to procure a particular genetically-engineered DNA sequence will
likely depend on the balance between the cost of sequence reconstruction and the economic
value of the reassembled sequence.
For these reasons, the relative imperfection of the DNAi device is perhaps even better
suited to biocontainment applications.
Unlike theft protection, where either one's security
measures succeed in preventing a competitor's acquisition of your information or they don't, the
criteria for success and failure are not quite as binary with regard to biocontainment problems.
Instead, a biocontainment solution that offers a reasonable amount of risk reduction or
environmental remediation might be better than none at all. In addition the DNAi device's ability
to effect the demise of both live cells and their DNA allows it to deal with a broader range of
potential contaminating threats. For instance, an industrial accident could release 1017-1019 cells
into the environment 3 . In such a scenario, a DNAi-mediated reduction of viable cells by a factor
of 10- is equivalent to reducing the number of cells present in a 1M liter fermenter to those
present in 10 ml of culture. This is well below the log-6 reduction in viable cells in the liquid/solid
waste stream required by the US Environmental Protection Agency (EPA) and meets the log-8
76
reduction required by the US National Institutes of Health (NIH) E. coli host-vector (EK2)
guidelines 7 4 for plasmid retention before disposal.
At the same time, the proliferation of antibiotic resistant pathogenic bacteria has
succeeded in raising more general concerns about the safety of releasing synthetic DNA
constructs into the environment in any form due to the perceived risk of inadvertent horizontal
gene transfer events involving naturally competent strains of microbes 75 . Many existing methods
for effecting programmed cell death only lyse the host and do little to remediate levels of cellular
DNA, which escapes into the environment 75 . When this extracellular DNA is released into the
environment, it does eventually decay due to the action of nucleases produced by the soil
microbiota", but with half-lives on the order of days. In fact, experiments with recombinant E.
coli showed that DNA in soil could be detected by PCR for 28-60 days after release.'
Applying
DNAi activity to reduce extracellular DNA levels by ~104 -fold and live cell titers by 10 8 -fold could
therefore potentially reduce the duration of the environmental exposure to this DNA by many
days or months.
Furthermore, if specific segments of a sample's DNA are thought to be
particularly dangerous or undesirable, the device's DNAi can be easily reprogrammed to target
those particular threats.
Another particularly useful and somewhat unique characteristic of the DNAi device that
we observed though our experiments is its capacity to degrade its own genetic information
without fully interfering with its overall function. This property of "self-elimination" could prove
extremely useful from both a biocontainment and secrecy protection standpoint. In terms of
biocontainment, the release of the DNA sequences that encode the DNAi device, which is of
synthetic origin itself, is potentially just as troublesome to future regulators as the synthetic DNA
that the device was originally implemented to prevent from escaping. Thus, this capacity for
"self-elimination" avoids a significant regulatory hurdle that might otherwise hinder attempts at
real-world applications for other biocontainment systems. In terms of protecting the information
contained within DNA from theft, self-elimination is helpful in that it removes valuable clues as
to a recovered DNA sample's true value. For instance, if a competitor detects the DNAi device
itself within one particular sample among a collection of others, then they are more likely to
deduce that the sample in question contains the valuable proprietary information they seek.
77
Whereas before resource limitations may have made it impossible for them to pursue every lead
and forced them to choose randomly, the additional information may incentivize them to
concentrate their remaining resources to pursue this preliminary hit.
In its current incarnation, the optimized DNAi device converts a chemical signal,
specifically a high concentration of arabinose, into an output of DNAi activity, and as we have
demonstrated, this arrangement works exceptionally well for measuring the device's properties
within the well-controlled laboratory environment. Outside of the lab, however, that property
could prove quite limiting, because more practical biotechnology applications may require the
DNAi device to respond to a vastly different set of chemical or environmental signals.
For
instance, even from the standpoint of the aforementioned theft prevention and biocontainment
applications it would be preferable to connect the DNAi input to sensors that activate when the
cells are removed from a defined media or exposed to particular conditions (e.g., light or oxygen).
Taken further, even more sophisticated controls can be achieved by connecting the device to the
output promoter of a transcriptional genetic circuit; for example, genetic logic to integrate signals
from multiple sensors that define a cocktail of chemicals specific to a media or environment. That
way, the device can act as a safeguard without the need for active user intervention. Fortunately,
the DNAi device was deliberately designed with this type of reprogramming in mind. In theory,
any genetic circuit could be connected to the DNAi actuator component so long as its output
spanned the required range of promoter activity, which we have explicitly characterized.
Despite the DNAi device's successes, potential strengths, and adaptability, it is highly
unlikely that the technology we have developed represents the singular, ultimate solution to any
problem involving the unintended proliferation of DNA sequence. Given the relative fluidity of
DNA's many usable and transmittable formats and the ease with which it can be amplified both
in vitro and in vivo, any viable laboratory-, commercial-, or industrial-scale solution that provides
comprehensive protection against unintended dissemination of the encoded information will
likely require a combination of layered, redundant, compatible strategies rather than a single
safeguard. Our preliminary experiments in which we combined DNAi- and integrase-mediated
killing within a single host (Figure 4.8) have demonstrated that redundant cell death mechanisms
can indeed be harnessed to increase the overall efficiency of killing. We therefore have every
78
reason to anticipate that DNAi will prove compatible with others' published systems for the
containment of genetically modified organisms, including devices that implement programmed
cell death, DNA deletion, and the recoding of the genome to require non-natural amino acids
(reviewed in
75,76)
in addition to having successfully constructed a highly functional and potentially useful
genetic device, our investigations have also reemphasized the importance of certain design
principles for successfully engineering synthetic biochemical pathways.
Chief among these
insights are strategies for reducing leaky gene expression. Specifically, chromosomal expression
appears to be less leaky compared to plasmid-based expression of comparable copy number.
Our optimized device with three chromosomal copies of the actuator sequence per cell
demonstrated greater retention of the target plasmid in the DNAi OFF state (>98%) than the
equivalent device harboring a BAC-based actuator (81%) that is maintained at 1-2 copies/cell.
It is also clear that there can be positive externalities to reducing the leakiness of a system
to its bare minimum, regardless of whether or not that extreme is required for the system's
proper function. Our optimized 3x DNAi strain carried >25 kbp of actuator DNA that was of no
direct benefit to the host in either the DNAi ON or the DNAi OFF state, and yet that sequence was
maintained without antibiotic selection and without any noticeable loss of function for over 360
cell generations. It was only after 1700 generations that even a reasonable loss of functionality
occurred. Thus, the rate of genetic drift within a population appears to be quite low, and from
the DNAi host strain's perspective, the cost of carrying extra base-pairs is exceedingly low as long
as those base pairs are not expressing anything.
When it comes to ensuring reliability of function of a genetically encoded device,
redundancy is perhaps most important. Even if the device is perfectly silent in its OFF state and
thus represents an effectively neutral bit of genetic information from the perspective of host
viability, genetic drift will ultimately result in the accumulation of loss-of-function mutations that
will appear only after the device is switched into the ON state. Thus, unless there is a means of
positively selecting in favor of a gene that is silent (a premise that seems intrinsically
contradictory), redundancy is the only remedy to lower the probability of a component's total
failure.
79
This finding is not novel; others who have built kills switches and suicide devices that have
employed other biochemical mechanisms for inducing cell death have similarly found that
mutation rates dominate failure rates under most conditions. Nevertheless, with respect to our
DNAi-based device, our redundancy strategy was sufficiently successful that we lowered the
probability loss-of-function mutation to the point where it no longer dominated. Instead, in the
case of the optimized 3x device, a cell was more likely to transiently deactivate its DNAi pathway
as a result of reaching stationary phase than it was to have undergone a DNAi-inactivating
mutation over the course of our 8-hr knockout and killing experiments. The specific cause of this
stationary phase deactivation was never determined, but considering the cells do not undergo
appreciable cell death at this stage, perhaps a re-design of the system's components can provide
a solution, in which case mutation may once more become limiting.
Our experiments in which we measured the efficiency of cell killing without selection for
a plasmid encoding essential DNAi component highlighted an inherent instability in plasmid
maintenance (Fig 4.6). This finding was exceptionally surprising considering that the plasmid in
question was high-copy, and so the notion that failure of DNA replication or segregation could
cause hundreds of copies to fail seemed improbable. Fortunately, the probability of a given cell
failing in this manner is low (~104 for a pUC19-based plasmid under our conditions), and so if the
particular application of a genetic program requires <99.9% reliable performance, this failure
mode will likely have little-to-no impact.
Nevertheless, for applications in which the robust
function of the maximal fraction of cells is desirable, as was the case with our DNAi device, this
is a significant limitation, especially in the context of large-scale fermentations in which the use
of antibiotics is cost-prohibitive.
In these cases, this flaw must be circumvented in another
manner, possibly through improvements in genetic design. One obvious and readily available
solution for our DNAi device is to do away with plasmids all together and instead to integrate the
CRISPR targeting arrays into the chromosome as we did with the actuator sequences. Other
design solutions, such as incorporating a toxin-antitoxin system or moving an essential gene onto
a plasmid, allow for negative selection against plasmid loss without the need to add antibiotics
or other expensive chemicals.
80
Along similar lines, when extremely robust function and genetic stability is required, it is
best to use natural selection to one's advantage and design genetic components such that their
failure is actively selected against. For instance, when choosing how to integrate multiple copies
of the actuator sequence, we could have opted to insert long head-to-tail tandem repeats into a
single genomic locus.
Unfortunately, long repeated sequences such as these are generally
unstable in recA+ strains, because recombination reaction will change the copy number in one
direction or another unless a selective pressure exists to the contrary 68. However, by choosing
instead to integrate the copies of our actuator sequence at great distances from one another and
to point them in the same direction, we guaranteed that any analogous recombination events
that occurred in this context would remove a large swath of the host genome and kill the cell.
Our preference for essential DNA sequences as proto-spacers similarly reflects this design
principle for two reasons. These DNAi elements are less likely to fail due to the accumulation of
random mutations, because most of the base pair changes that would otherwise render DNAi
inactive are also lethal to the host and thereby prune themselves from the population.
Furthermore, if DNAi mediated cutting is indeed actively counter-balanced by the host's
somewhat error-prone attempts to repair its broken DNA, then a very likely outcome of DNAi
activity at a specific site is a net alteration to the underlying sequence. Again, if this sequence is
essential, this alteration will be selected against for the same reasons as above.
The DNAi device we have constructed is a powerful, multipurpose tool that has been
highly engineered for superior function. Given its abilities to cure plasmids, kill its host, or both,
and to do so in a manner that places exceedingly little burden on the host's competing metabolic
process, it may find use in a wide array of biotechnologies in addition to our proposed application
of protecting sensitive DNA sequence information from theft or inadvertent release. Of course,
while our device has operated strictly in the context of K-type E. coli, the organism from which
its genetic components are derived, in the true spirit of synthetic biology, metagenomic
engineering, and the unifying principles of biology's Central Dogma, there is every reason to
believe that with the proper modifications our system could be adapted to function in diverse
microorganisms or higher-order species.
The extensive parameterization of the device's
components that we have performed will be particularly useful in this organism-specific tailoring
81
process. At the very least, however, the lessons learned during the construction and optimization
of the DNAi device will aid in designing the next generation of genetically encoded devices that
are both highly functional and highly robust to the demands placed upon them.
82
A.
Supporting Information
A.1
Supplementary Figures
B
A
PBAD
PBAD
PJ2317
PJ23117
cas genes
A' B C DEaraC
cas genes
araC
A 9 C D E
pACT-A
pACT-01
C7
PBAD
Kan"colEl
T
F
cas genes
A B C D E
cas genes
A 6C
D
araC
E
PBAD
KanRp15A
araC
pACT-B
pACT-02
1!7
6T
P123117
PMDo
P123117
Kan
attP
T
R6K
FIRT
Kan
PC PBAD
PJ2317
cas genes
cas genes
A. B 'C DE
4
arac
araC
A a C Q E
pACT-C
pACT-03
attP
KnBACT
:
C
R6K
FRT
KanR
PJ231XX
araC
pAraC-xx
pp5a
CKanR
the episomal
Supplementary Figure A.1: Actuator and araC accessory plasmid maps. (A) With the exception of their origins of replication,
the control of the
actuator plasmids pACT-01 through pACT-03 are isogenic with one another. Expression of the cas3ABCDE operon is under
araC gene is expressed
arabinose-inducible PBAD promoter. P8AD function requires the expression of the AraC transactivator protein, and the
of the cas3ABCDE
expression
pACT-C,
through
pACT-A
from a weak constitutive PJ23117 promoter. (B) For the integrating actuator plasmids,
and utilizes
pACT-B
and
pACT-A
both
in
constitutive
is
araC
of
Expression
promoter.
operon is under the control of the arabinose-inducible PBAD
present
site
FRT
single
The
araC.
of
expression
controls
Pc
promoter
auto-regulated
native
the
pACT-C
the same araC cassette as in (B), but in
into
integrate
to
plasmids
this
allowing
thereby
site
FRT
chromosomal
corresponding
a
with
recombination
in pACT-A enables FILP-mediated
Int5enabling
of
function
analogous
the
serves
pACT-C
and
pACT-B
in
present
the host chromosome via loop-in mechanism. The attP sites
pACT-C possess R6K origins in
mediated recombination with a corresponding chromosomal attB site. Integration constructs pACT-A through
initiate from these loci. The
not
does
DNA
chromosomal
of
replication
unregulated
MG1655,
coli
E.
as
such
hosts
order to ensure that in pirrecombination between
FRT site present in pACT-B and pACT-C allows the plasmid KanR marker and R6K origin to be excised via FLP-mediated
of pACT-A. (C) The araC
integration
following
excised
be
cannot
marker
The
site.
FRT
the plasmid-borne FRT site and an adjacent chromosomal
(P2 117), while pAraC-105 uses a
accessory plasmids are isogenic except that pAraC-117 expresses araC from weak constitutive promoter
stronger constitutive promoter (Po
2 31 05 )-
83
A
PJ23117
N
PJ 23 1 7
PJ 23117
PJ23117
Repeat
z
A
z
Y
Spacer -;
P
M13 ori
colEl
(pUC19)
M13 ori
colEl
(pUC19)
M13 ori
colEl
(pUC19)
p CR-YZ
pCR-Z
pCR-Y
pCR-N
EPJ23117
N2A
PJ2127
1
r+ N
Xmm
z
Y
M13 ori
colEl
(pUC19)
N28
NC-N
CMR
LMR
CMR
M13 ori
RSf
M13 ori
RSf
r* G, r4
I pCRG1 I
pCR-Xtt
CMR
CMR
M13 ori
p15A
M13 ori
colEl
(pUC19)
x
+
pCR-Xpi 5 a
CM1
M13 ori
PJ 231W Ptet
A x AtetR1
r+
pR-G2
M13 ori
colEl
(pUC19)
PJ23117
G
CMR
colEl
(pUC19)
CMR
M13 ori
colEl
(pUC19)
PJa3117
PJ2311,
CR-N2
pCRX
pCR-YZ*
pCR- N*
I
M13 ori
p15A
B
PAM: 5'-ATG
PJ2,3O
PAM:
P231O0
5'-ATG
|tx
x
ft
pTAR(B)
pTAR(S)
Y Z
Y Z
PJ23I00
IpTAR(P)
Y Z
PAM:
M13 ori
pSC101
p15A
M13 ori
5-ATG
TT|x
pTAR(C)
YZ
Sr
SOr
StraRr
BAC
PAM; 5'-ATG
Tix
ft
M13 ori
colEl
(pUC19)
M13 ori
Supplementary Figure A.2: CRISPR targeting plasmid and basic target plasmid maps. (A) All DNAi device
experiments require at least one repeat-flanked CRISPR spacer to direct DNAi activity. The transcription of all
CRISPR arrays was constitutive. Due to the high copy number of the pUC19-based colEl plasmid, a weak PJ23 117
promoter was used. When experimental conditions created a conflict between the CRISPR plasmid's origin of
replication and that of the target plasmid (Figs 1D and 1E), the CRISPR plasmid colEl origin was substituted with
an RSF origin (highlighted in red). For these RSF variants, a stronger constitutive promoter (PJ2311 - highlighted
in red) was used in order to offset the drop in crRNA expression due to lower CRISPR plasmid copy number. (B)
The RFP-expressing targets used for all plasmid knockout experiments are homologous and span more than two
orders of magnitude with respect to copy number (1-2 copies/genome for BAC up to >500 copies/genome for
pUC19). All copy number variants constitutively express high levels of RFP, confer resistance to streptomycin
(and spectinomycin), and possess a non-coding proto-spacer (X) coupled to an ATG PAM in addition to two StrRcoding proto-spacers, Y and Z, with ATG and AAG PAMs, respectively. Increasing target plasmid copy number
with a pUC19-based colEl origin caused RFP-mediated toxicity when expressed from the strong promoter PJ23100
An attenuated promoter (PJ231*) in which the -10 element had been mutated from TACAGT to TACTAT was
used instead.
84
PC PWa
attB FRT
FRT
KaR (rev)
araC
int5
(fwd)
pKD13-AttB
AMR
"
Am
E6K
pint~ts
R6K
7
psc101V
Supplementary Figure A.3: Strain construction plasmid maps. pKD13-attB
encodes a modified Kan' knockout cassette coupled to an Int5 attB site.
Upon amplification with appropriate primers ("fwd" and "rev") encoding 5'and 3'- homology (light and dark blue overhangs, respectively) to a target site
within the host genome, the resulting linear DNA can be transformed into a
XRED-expressing host to both knockout the targeted region and knock in a
chromosomal attB site. plnt5ts confers arabinose-inducible expression of
The pInt5t, plasmid possesses a
the Int5 site-specific recombinase.
temperature-sensitive pSC101 origin of replication that allows the plasmids
to be cured from the host via growth at 37*C following successful
recombination.
85
PBAD
PC
P23117
MWbo
araC
araC
pGFP-C
pGFP-B
R6K
attP
POAD RiboJ
R6K
attP
FRT
FRT
KanR
KanR
Supplementary Fixure A.4: Fluorescence measurement plasmid maps.
Fluorescence measurement plasmids pGFP-B and pGFP-C are isogenic with
actuator plasmids pACT-B and pACT-C (Figure A.1), except that a RiboJ-B0034GFP reporter cassette was substituted for the actuator cas operon. The were
integrated instead their corresponding pACT counterparts at the same three
chromosomal locations to create the 3x GFP reporter strain (Methods).
86
PCR Cassette (amplified from pKD13-attB)
qKan R
FRT
FRT
,4-
(!R.ED-
CRISPR
cas genes
~ .4.......IL.
[arabinose]
Integration locus
(e.g. native cas/CRISPR)
Deletion of host genes
FRT
FRT
FRT
4
Removal of marker (via pCP20/FLP)
4
Addition of pIntS1 , and pACT-01
PJ
23117cas
E
PRM
PI
PBAD
genes
D
C
B
cl
A
pACT-A
pCP20
R6KFRT
Ri
AM R
KanR
CMR
pSC101ts
FRT
Actuator integration
Phage clean-up (via Pivir transduction)
PBAD
pACT-A-derived actuator
Supplementary Figure A.5: Schematic of FLIP-mediated actuator genomic integration methodology. A FRTflanked KanR knockout cassette is amplified from pKD13 using primers containing 5'- and 3'-homology (light
and dark blue segments, respectively) to regions of the MG1655 chromosome targeted for actuator insertion
(See Table A.3 for primer sequences). The knockout cassette is integrated into an MG1655 donor strain via ?RED
1
recombinase9 , thereby simultaneously deleting a segment of the host chromosome. (See Table A.2 for deleted
segments) The KanR marker is removed via FLP recombinase (pCP20) , and the marker-free strain is retransformed with pCP20. FLP is expressed, and cells are transformed with integrating actuator plasmid pACT-A.
FLP-mediated recombination between the plasmid-borne FRT site and the chromosomal FRT site then creates a
chromosomally-integrated actuator sequence within the donor strain. The integrated actuator locus is then
moved from the donor strain to the final recipient host strain via Plvir phage transduction. The Kan Rmarker and
R6K origin cannot be removed from the recipient host. Integration leaves the FRT sequences unaltered, and so
repeated plasmid integrations resulting in a heat-to-tail array of actuator sequences. It is therefore possible to
achieve clones with actuator copy numbers greater than one from a single transformation.
87
PCR Cassette (amplified from pKD13-attB)
attB FRT
FRT
-
cas genes
,Y .D E~j
~~
**
(-[arabinose]
CRISPR
~..
........
Integration locus
(e.g. native cas/CRISPR)
Deletion of host genes
FRT
attB FRT
Removal of marker (via pCP20/FLP)
attB FRT
I
'PBAD
PC
Addition of pInt 5t and pACT-01
*pwP
jarabinose]+
aabns]
cas genes
p
p +1oE
a
intS
FRT
R
ac
pint5t
pACT-C
Am R
KanR
pSC101ts
attB FRT
Actuator integration
PC PRAD
Phage clean-up (via Pivir transduction)
Removal of marker and R6K ori (via pCP20/FLP)
PC P8AD
pACT-C-derived actuator
Supplementary Figure A.6: Schematic of IntS-mediated actuator genomic integration methodology. A Kan knockout cassette containing
to
the Int5 attB site is amplified from pKD13-attB using primers containing 5'- and 3'-homology (light and dark blue segments, respectively)
is integrated
regions of the MG1655 chromosome targeted for actuator insertion (See Table A.3 for primer sequences). The attB cassette
A.2 for
Table
(See
chromosome.
host
the
of
segment
a
deleting
simultaneously
,
thereby
into an MG1655 donor strain via XRED recombinase
with pInt5 . Int5
deleted segments) The Kan marker is removed via FLP recombinase (pCP20) 1, and the marker-free strain is transformed
Int5-mediated
shown).
as
pACT-C
(i.e.
plasmid
actuator
integrating
an
with
transformed
and
expression is induced with arabinose
actuator
recombination between the plasmid-borne attP site and the chromosomal attB site then creates a chromosomally-integrated
recipient host strain via
final
the
to
strain
donor
the
from
moved
then
is
locus
actuator
integrated
The
strain.
donor
the
within
sequence
9
host via FLP
P1, phage transduction. The Kan marker, R6K origin, and IntS attR site are subsequently removed from the recipient
was
procedure
analogous
an
recombinase (pCP20) to yield the final 1x actuator strain. For host strains containing the 3x genomic actuator,
actuator.
the
as
iterated two additional times using pACT-B
88
A
Input 1
[arabinosel -4
PC
PaAO
col,
oC-aroBAD locus
7RNAP tet
E.cE.
Input 2
-f
I
I
[IPTG1
®
-#
--
______
--
PT 7/LacO
-
P,/LacO
P,/LaCO CRISPR
C D E
A 8, genes
z
Plac
LPaipWUR397
RSF
Kan
Jac[
Leader 4
+cos
pWUR400
CDF
lacl
Str
pWUR47xI
C
1:
p15a
CmlacI
Jr
B
Input 1
(IPTG-
Cas3
CasABCDE
I
I
crRNA guides
AIk
I
-
P 23119/LacO
--
-1
P23119/LacO
cas genes
P
23 9/LaCOCRISPR
Leader
SpWUR397*
r*
,I
pWUR400*
pWUR47x*
acd
PWaci
p15a
CDF
RSF
Str
CMR
lacl
Supplementary Figure A.7: Schematic of three-plasmid, two-input cas/CRISPR over-expression system (Brouns et al, 2008)
(A) Upon addition of both inputs (arabinose and IPTG), T7 RNAP is expressed from its chromosomal cassette and the hybrid
PT/LacO promoters are de-repressed. T7 RNAP then initiates transcription from its cognate promoters present on all three
pWUR plasmids, resulting in expression of all three cas/CRISPR components essential to DNAi activity. The pWUR397
expresses Cas3, pWUR400 expresses CasABCDE ('Cascade'), and a pWUR47x variant expresses the pre-crRNA, which is
processed by CasE and then used to target DNAi activity. The pWUR47x family of plasmids also possesses a strong RBS and
the native K-type E. coli CRISPR leader sequence immediately upstream of the first CRISPR repeat. This RBS was found to
cause growth defects within its host, and the leader was found to be non-essential. Both were ultimately removed from
later-generation CRISPR plasmids. (B) Schematic of the simplified pWUR* overexpression system. All pWUR* plasmids are
isogenic with their parent pWUR counterparts, except that their respective PT/LacO hybrid promoters have been replaced
with P1 23 1 /LacO hybrid promoters. In the absence of the LacO operator, the wild-type PJ 23119 promoter is a strong
constitutive promoter. The net result of these substitutions is that expression of all cas/CRISPR components are now
independent of T7 RNAP. The need for arabinose induction of PBAD-T7 RNAP is therefore removed, and the two-input system
in (A) is reduced to a single input (IPTG).
89
B
A
101
104
100
101
102
,
,-
2
10
0
10010
0)
0)
0
010o-6
ie
TiePs-Atvto)(r
4
20-
0
otAtvain(r
C:D2
10'10
10-1
Time Post-Activation (hr)
Time Post-Activation (hr)
Supplementary Figure A.8: Characterization of PBAD Input. (A) PBAD input promoter expression levels over time
for the 3x GFP reporter strain (green squares, left-hand axis in a.u., Methods) are shown for cells grown under
standard DNAi induction conditions (2YT, 2mM Ara). The data from Figure 4.1 depicting the kinetics of pTAR(S)
plasmid knockout using the 3xc DNAi strain containing pCR-YZ (blue triangles, right-hand axis in fraction target+
cells) are overlaid for reference. Cultures of the 3x GFP strain containing pTAR(S) and pCR-YZ were prepared,
induced, cultured, and sampled in a manner identical to that used in Figure 4.1 (Methods). Cytometric analysis
of GFP fluorescence was also performed similarly, except that populations were gated based on forward and side
scatter and then based on RFP expression from pTAR(S) (1000 RFP au cut-off) to ensure that only viable cells
were counted. The vertical dashed line at t = 2.25 hr indicates the maximal lag time for PBAD-mediated synthesis
of the GFP reporter. This lag correlates exceptionally well with the observed lag in the induction of DNAi-
mediated plasmid knockout. (B)
PBAD
input promoter expression levels over time for the 3x GFP strain (green
squares, left-hand axis) and the kinetics of pTAR(S) plasmid knockout for the 3xc DNAi strain (blue triangles, righthand axis) are shown for cells grown under leakage conditions (2YT, OmM Ara, 0% Glc). Again, both strains
contained the pTAR(S) & pCR-YZ target/CRISPR plasmid pair. With the exception of the inducer concentrations,
1
experiments were performed identically to those depicted in (A). The horizontal dashed line at 5x10 a.u. reflects
the estimated minimum expression threshold for DNAi activity, defined as the expression level required to
observe a fraction of target-positive cells <95% in the corresponding plasmid knockout experiment, and is
approximately "3-fold higher than the expression level of PBAD in the glucose-repressed state (corresponding to
t = 0 hr, Methods). All measurements were background-corrected by subtracting the auto-fluorescence of a
triple-knockout E. coli MG1655 parent strain (No DNAi, Methods) containing target and CRISPR plasmids pTAR(S)
and pCR-YZ that was grown under identical conditions. Data points depicted as hollow green squares include
measurements whose value was at or below the detection limit for the instrumentation used, and were therefore
rounded up to the minimum measurable value (0.1 a.u.). All data represent the average of three independent
experiments performed on different days, and error bars indicate the standard deviation.
90
Pla
PaL"
,AM:_5'-NNN
PJ23100
PAM:
wt
pPAM-TC
i
pPAM-NNNr
StrR
pSC101
M13
ori
L
pSC101
*NN
pPAM-NNN-RFP
Y Z
Y Z
Y Z
TTLIx
5
StrR
StrR
M13
ori
pSC101
M13
ori
Supplementary Figure A.9: Maps of target plasmids used in PAM experiments. All PAM targets possess a
pSC101 origin of replication. The additional M13 origin of replication allows these plasmids to be packaged
as ssDNA into a viral capsid when the host co-expresses the full compliment of M13 viral proteins. The
resulting plasmid-containing phagemid virion particles are capable of directly transducing F+ E. coli strains.
91
0
.2
8
6
4
2
Time Post-Activation (hr)
2
0
8
6
100
4
2
0
101.
100
10-.
10'
10-1
10-'.
10-2
10-2
10-'
10-'
10*4
10-4.
10-s
10-'
0
t
2
6
2
8
100
10-'
10-1]
10-'
10-2
101
10-3
10-'
10-3
10-'
10-s
10-4
10-5
10-
10-5.
10-'
AGG
2
4
6
4
4
6
8
0
10-k
Time Post-Activation (hr)
2
0
8
6
100
4
6
0
8
10-1
10-2
10,3
10'.
10.
10'.
10-4
10-4.
10-4.
10-5.
100
0
100)
10-
10-'
10-'
10-.
10-2
10-1
10-2
102
10-'
10'1
10-'
10a'.
10-4
10-4
10-4.
10-d
10-5.
TAG
2
TTG
TGG
8
2
4
6
8
TAA
GTG
Time Post-Activation (hr)
2
0
8
6
4
100
0
6
1 001
10-2.
8
4
cAG
102
6
2
10-'
10.
4
8
10-s
10-
2
6
10-2
10-'.
10-'.
0
100,
4
AAT
10-1
GGG
2
10-'
AAG
100.
10-'
8
10-4.
100
GAG
6
10-3.
ATG
2
4
10-2.
ATA
)
8
2
100
icy'.
Time Post-Activation (hr)
2
0
8
6
4
100
10-5
.
4
AAC
100
0
0
10-i
10-1
AAA
8
6
4
4
6
8
10-. CcG
Supplementary Figure A.10: The dynamics of plasmid loss for each active PAM sequence. The overlay from
Figure 3.12 is separated into individual plots for increased visual clarity. PAM sequences are indicated in the
lower left of each graph, and colors correspond canoncial, strong, weak, and negative control PAMs as in Figure
3.11. Data are for the 1xc DNAi device, X spacer targeting plasmid (pCR-X, CmR), and a pSC101 target (pPAMNNN-RFP, RFP+, StrR) and reflect the fraction of cells containing the target plasmid as determined by PAM kinetic
assay (Methods). All data reflect the average of three independent experiments performed on different days,
and error bars indicate the standard deviation.
92
A.2
Supplementary Tables
Table A.1: CRISPR element sequences
Element
PAM
Target
Spacer G 1
CACCCGACGCCATTCTATACCTGATAATTCTT
AAG
N.C.a (genome)
Spacer G2
TTGTCAACGCTCCCAGGCCGTCCACTCCCTGA
AAG
attLb
Spacer Ni
TGGTCTTCAGCCTTCTAGAGATATGAAGACTT
n/a
none
Spacer N2AI
CTTTCGCAGACGCGCGGCGATACGCTCACGCA
n/a
none
Spacer
GACTCACCCCGAAAGAGATTGCCAGCCAGCTT
n/a
none
Spacer AX
ACTGTTCGGGAGCTTTGTTCAGCAGCGATAAC
ATG
bIaAX (plasmid)
Spacer X
ACAGTAAGAGAATTATGCAGTGCTGCCATAAC
ATG
N.C.a or bla (plasmid) d
Spacer Y
ACATTCTTGCAGGTATCTTCGAGCCAGCCACG
ATG
StrR
Spacer Z
CCACACAGTGATATTGATTTGCTGGTTACGGT
AAG
StrR
Repeat
GAGTTCCCCGCGCCAGCGGGGATAAACCG
N 2 .c
-
-
Sequence
a. N.C = Non-coding sequence
b. Att sites correspond to IntS recombinase used for actuator chromosomal integration
c. Spacers designated N2A and N 2. in this work correspond to spacers with designations N, and N, from pWUR477 from Brouns et al.
Science, 2008.
d. Spacer X targets a non-coding region in target plasmid pTAR(S) and the wild-type bla (Amp") allele in target plasmid pAmpTAR(S)-X
93
Table A.2: Host chromosomal modifications
Locus
A(CRISPR-cas)
A(araC-araBAD)
ALaci
Native Sequence
Deleteda
Actuatorb
(1x strainc)
Actuatorb-c
(2x strain)
Actuatorb
(3x strainc)
2,879,926 - 2,889,743
pACT-A, -B, or -C
pACT-B
pACT-B
65,854 - 71,265
none
pACT-B
pACT-B
366,428 - 367,586
none
none
pACT-B or -C
a. Numbering based on NCBI entryfor E. coli str. K-12 substr. MG1655 (NC_000913.3)
b. All actuators are oriented with respect to the chromosomal (-) -strand
c. For a given actuator copy number, different actuator variants are delineated by a subscript indicative of the pACT version integrated
at their variable position. For example, the 3x strain harboring 2x pACT-B plus Ix pACT-C is designated 3 xc
94
Table A.3: Genomic modification primers
Primer0
Sequenceb
A(CRISPR-cas3)-fwd
5'- TATCAGGTATAGAATGGCGTCGGGTGCTTGAGGCTGTCTGgtgtaggctggagctgcttc
A(CRISPR-cas3)-rev
5'-
A(araC-araBAD)-fwd
5'- GCTACTCCGTCAAGCCGTCAATTGTCTGATTCGTTACCAAgtgtaggctggagctgcttc
A(araC-araBAD)-rev
5' - AGCCTGGTTTCGTTTGATTGGCTGTGGTTTTATACAGTCAattccggggatccgtcgacc
ALacI-fwd
5'-- TCTGGTGGCCGGAAGGCGAAGCGGCATGCATTTACGTTGAgtgtaggctggagctgcttc
ALacd-rev
5' - TGCCTAATGAGTGAGCTAACTCACATTAATTGCGTTGCGCattccggggatccgtcgacc
AATGCTACCTCTGGTGAAGGAGTTGGCGAAGGCGTCTTGAattccggggatccgtcgacc
a. Forward and reverse designations are given with respect to actuator orientation following integration
b. Underlined sequence corresponds to homology (40 bp) with host chromosome. Lower case sequence corresponds to priming regions
present within pKD13-AttB plasmid for amplifying the knockout cassette
95
Table A.4: Int5-specific recombination sites
Site
Sequence0
attB
GAGCGCCGGATCAGGGAGTGGACGGCCTGGGAGCGCTACACGCTGTGGCTGCGGTCGGTGC
attP
CCCTAATACGCAAGTCGATAACTCTCCTGGGAGCGTTGACAACTTGCGCACCCTGATCTG
attL
GAGCGCCGGATCAGGGAGTGGACGGCCTGGGAGCGTTGACAACTTGCGCACCCTGATCTG
attR
CCCTAATACGCAAGTCGATAACTCTCCTGGGAGCGCTACACGCTGTGGCTGCGGTCGGTGC
a. All sequences given in the forward orientation. Underlined bases corresponds to recombination point.
96
Table A.5: qPCR primer sets
Set
Dist.
to Gil
Reverse primer
Forward primer
Amplicon
length
Mol. Wt.
(ng/mol)
A
0
CGAGCAGCATTCCTGATTTTA
CTCGAGGAAGCAGCTCCA
110 bp
B
1
GGCGTCGCTTGATGAACTGATA
CCGCACGCACCGTAAAGT
83 bp
5.11X10
C
10
GCGCATTTTGATTGGCATT
CCCAGCAAAATACCGCGAT
99 bp
6.11x10'
D
100
GCAATTTTTCCCGGCAAAATTACAA
CATGTTAAGAACCTCCCTATCCGTATATAA
100 bp
6.16x101
E
1000
GGTTGATCCAAGCTTCCTGACA
AAGAATGGCTGGGATCGTGGGTT
94 bp
5.80x1013
Stra
-
AAAACAAAGTTAAACATCATGAGGGAA
GATGACGCCAACTACCTCTGATA
78 bp
4.81X1013
6.80X1013
a. Center-to-center distances (in kbp) between the corresponding qPCR amplicon and the protospacer G, sequence (NCBI position
2,887,466) as measured along the host chromosome in the (+)-strand direction.
97
13
3
3
Table A.6: Constitutive promoter sequences
Promoter
TTGACGGCTAGCTCAGTCCTAGGTACAGTGCTAGc
PJ231OO
P
Sequence0
b23100-
TTGACGgctagctcagtcctaggTACTATGCTAGc
PJ23101
TTTACAgctagctcagtcctaggATAATAGCTAGc
P123105
TTTACGgctagctcagtcctaggTACTATGCTAGc
PJ23117
TTGACAgctagctcagtcctaggGATTGTGCTAGc
a. Italicized and underlined sequences correspond to promoter -35 and -10 elements,
respectively. Lowercase base corresponds to the promoter -1 position.
b. Mutations in -10 element relative to P23m are given in bold
98
Table A.7: RBS sequences
RBS Sequence0
Plasmid
ORF
pACT-01/02
cas3
pACT-01/02
casA
pACT-Olb
araC
pACT-02d
araC
pTAR
RFP
pINT5t5
lnt5
ATACCCGTTTTTTTGGGCTAGCAGGGACAGGATATGTGAGTAAACGTCGTTATCTTACCGGTAAAGAAGT
p1NT5tsb
araC
GTTTTTGTCATGGCTTTGGTCCCGCTTTGTTACAGAATGCTTT..
pGFP-01/02c-d
GFP
pJ23101-GFPe
GFP
a.
b.
c.
d.
e.
ATACCCGTTTTTTTGGGCTAGCAATAAGGAGATATACCAT
GCACTCGAGTGTAGAAATAATTTTGTTTAACTTTAATAAGGAGATATACCAT
GTTTTTGTCATGGCTTTGGTCCCGCTTTGTTACAGAATGCTTT... TTCTTCTCTGAATGGCGGGAGTATGAAAAGTAIG
TACTAGAGAAAGAGGAGAAATACTAGATM
TTTCCCCCGGAAACCAATAAAAGAAGGCCATCGTCAIQ
TCAGGCCA
TTCTTCTCTGAATGGCGGGAGTATGAAAAGTATI
CCTGTcACCGGATGTGCTTTCCGGTCTGATGAGTCCGTGAGGACGAAACACCTCTACAAATAATTTTGTTTAAAAAGAGGAGAAATACTAG&Tr
TTACTAGAGTCACACAGGAAAGTACTA&MI
RBS defined as the entire 5' UTR spanning the transcriptional +1 site (bold) through the start codon (underlined)
Native araC RBS sequence corresponding to E. coli str. K-12 substr. MG1655 (NC_000913.3 positions 70225-70389)
RBS includes RiboJ self-cleaving ribozyme. RNA auto-cleavage occurs 3' to the lowercase cytosine.
BBa_0034 RBS sequence is in italics
BBa_0032 RBS sequence is in italics
99
Table A.8: Classification of DNAi survivor LOF mutations via genetic complementation
Original
pCisTAR(S)
pCisTAR(S)-X
pCisTAR(S)-Y
pCisTAR(S)-Z
pCisTAR(S)-Y+Z
DNAi activity observed +pACT-03/pCR? (Yes/No)
+pACT-03
+pCR-N
+pCR-X
+pCR-Y
+pCR-Z
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
No
Yes
Yes
Yes
No
No
No
Yes
Yes
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
No
Yes
Yes
Yes
No
No
Yes
No
Yes
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
Yes
Yes
Yes
No
No
No
No
Yes
Yes
No
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
Yes
Yes
Yes
No
No
No
No
Yes
No
No
No
No
No
No
No
LOF assignment
No LOF
Actuator
CRISPR
Target
Multiple mutations
No LOF
Actuator
CRISPR
Target
Multiple mutations
No LOF
Actuator
CRISPR
Target
Multiple mutations
No LOF
Actuator
CRISPR
Targets (both)
Multiple mutations
a. While the chart does not cover all 32 Yes/No combinations for each pCisTAR(S) variant, none of the 27 other ambiguous cases were
observed
100
Bibliography
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
Arkin, A. Setting the standard in synthetic biology. Nat. Biotechnol. 26, 771-774 (2008).
Molin, S. et al. Suicidal Genetic Elements and their Use in Biological Containment of Bacteria.
Annu. Rev. Microbiol. 47, 139-166 (1993).
Isaacs, F. J. et al. Engineered riboregulators enable post-transcriptional control of gene
expression. Nat. Biotechnol. 22, 841-847 (2004).
Callura, J. M., Dwyer, D. J., Isaacs, F. J., Cantor, C. R. & Collins, J. J. Tracking, tuning, and
terminating microbial physiology using synthetic riboregulators. Proc. Natl. Acad. Sc. 107,
15898-15903 (2010).
Curtiss, R. et al. Salmonella enterica Serovar Typhimurium Strains with Regulated Delayed
Attenuation In Vivo. Infect. Immun. 77, 1071-1082 (2009).
Kong, W. et al. Regulated programmed lysis of recombinant Salmonella in host tissues to
release protective antigens and confer biological containment. Proc. Natl. Acad. Sc. 105, 93619366 (2008).
You, L., Cox, R. S., Weiss, R. & Arnold, F. H. Programmed population control by cell-cell
communication and regulated killing. Nature 428, 868-871 (2004).
Ahrenholtz, I., Lorenz, M. G. & Wackernagel, W. A conditional suicide system in Escherichia coli
based on the intracellular degradation of DNA. Appl. Environ. Microbiol. 60, 3746-3751 (1994).
Djordjevic, G. M., O'Sullivan, D. J., Walker, S. A., Conkling, M. A. & Klaenhammer, T. R. A
triggered-suicide system designed as a defense against bacteriophages.J. Bacteriol. 179, 67416748 (1997).
Andersen, J. T., Schafer, T., Jorgensen, P. L. & Moller, S. Using inactivated microbial biomass as
fertilizer: the fate of antibiotic resistance genes in the environment. Res. Microbiol. 152, 823833 (2001).
Nielsen, K. M., Johnsen, P. J., Bensasson, D. & Daffonchio, D. Release and persistence of
extracellular DNA in the environment. Environ. Biosafety Res. 6, 37-53 (2007).
Mao, D. et al. Persistence of Extracellular DNA in River Sediment Facilitates Antibiotic
Resistance Gene Propagation. Environ. Sci. Technol. 48, 71-78 (2014).
Recorbet, G., Picard, C., Normand, P. & Simonet, P. Kinetics of the persistence of chromosomal
DNA from genetically engineered Escherichia coli introduced into soil. Appl. Environ. Microbiol.
59, 4289-4294 (1993).
Daniel, R. The metagenomics of soil. Nat. Rev. Microbiol. 3, 470-478 (2005).
Thiel, C. S. et al. Functional Activity of Plasmid DNA after Entry into the Atmosphere of Earth
Investigated by a New Biomarker Stability Assay for Ballistic Spaceflight Experiments. PLoS
ONE 9, e112979 (2014).
Gibson, D. G. et al. Creation of a Bacterial Cell Controlled by a Chemically Synthesized Genome.
Science 329, 52-56 (2010).
Gibson, D. G. et al. Complete Chemical Synthesis, Assembly, and Cloning of a Mycoplasma
genitalium Genome. Science 319, 1215-1220 (2008).
Karas, B. J. et al. Assembly of eukaryotic algal chromosomes in yeast.j. Biol. Eng. 7, 1-12 (2013).
Gibson, D. G., Smith, H. 0., Iii, C. A. H., Venter, J. C. & Merryman, C. Chemical synthesis of the
mouse mitochondrial genome. Nat. Methods 7, 901-903 (2010).
Dymond, J. S. et al. Synthetic chromosome arms function in yeast and generate phenotypic
diversity by design. Nature 477, 471-476 (2011).
Lartigue, C. et al. Genome Transplantation in Bacteria: Changing One Species to Another. Science
317, 632-638 (2007).
Karas, B. J. et al. Transferring whole genomes from bacteria to yeast spheroplasts using entire
bacterial cells to reduce DNA shearing. Nat. Protoc. 9, 743-750 (2014).
101
23. Trevors, J. T. Plasmid curing in bacteria. FEMS Microbiol. Lett. 32, 149-157 (1986).
24. Armstrong, K. A. et al. A 37 x 103 molecular weight plasmid-encoded protein is required for
replication and copy number control in the plasmid pSC101 and its temperature-sensitive
derivative pHS 1.J. Mol. Biol. 175, 33 1-348 (1984).
25. Charlton, H. R., Relton, J. M. & Slater, N. K. H. Characterisation of a generic monoclonal antibody
harvesting system for adsorption of DNA by depth filters and various membranes.
Bioseparation8, 281-291 (1999).
26. Lokteff, M. et al. Residual DNA Quantification in Clinical Batches of BBG2Na, a Recombinant
Subunit Vaccine Against Human Respiratory Syncytial Virus. Biologicals29, 123-132 (2001).
27. Prazeres, D. M. F., Ferreira, G. N. M., Monteiro, G. A., Cooney, C. L. & Cabral, J. M. S. Large-scale
production of pharmaceutical-grade plasmid DNA for gene therapy: problems and bottlenecks.
Trends Biotechnol. 17, 169-174 (1999).
28. Ng, P. & Mitra, G. Removal of DNA contaminants from therapeutic protein preparations.J.
Chromatogr.A 658, 459-463 (1994).
29. Brophy, J. A. N. & Voigt, C. A. Principles of genetic circuit design. Nat. Methods 11, 508-520
(2014).
30. Makarova, K. S. et al. Evolution and classification of the CRISPR-Cas systems. Nat. Rev.
Microbiol. 9, 467-477 (2011).
31. Westra, E. R., Buckling, A. & Fineran, P. C. CRISPR-Cas systems: beyond adaptive immunity. Nat.
Rev. Microbiol. 12, 317-326 (2014).
32. Barrangou, R. et al. CRISPR Provides Acquired Resistance Against Viruses in Prokaryotes.
Science 315, 1709-1712 (2007).
33. Marraffini, L. A. & Sontheimer, E. J. CRISPR Interference Limits Horizontal Gene Transfer in
Staphylococci by Targeting DNA. Science 322, 1843-1845 (2008).
34. Brouns, S. J. J. et al. Small CRISPR RNAs Guide Antiviral Defense in Prokaryotes. Science 321,
960-964 (2008).
35. Haurwitz, R. E., Jinek, M., Wiedenheft, B., Zhou, K. & Doudna, J. A. Sequence- and StructureSpecific RNA Processing by a CRISPR Endonuclease. Science 329, 1355-1358 (2010).
36. Przybilski, R. et al. Csy4 is responsible for CRISPR RNA processing in Pectobacterium
atrosepticum. RNA Biol. 8, 517-528 (2011).
37. Nam, K. H. et al. Cas5d Protein Processes Pre-crRNA and Assembles into a Cascade-like
Interference Complex in Subtype I-C/Dvulg CRISPR-Cas System. Structure 20, 1574-1584
(2012).
38. Garside, E. L. et al. Cas5d processes pre-crRNA and is a member of a larger family of CRISPR
RNA endonucleases. RNA 18, 2020-2028 (2012).
39. Niewoehner, 0., Jinek, M. & Doudna, J. A. Evolution of CRISPR RNA recognition and processing
by Cas6 endonucleases. Nucleic Acids Res. 42, 1341-1353 (2014).
40. Deltcheva, E. et al. CRISPR RNA maturation by trans-encoded small RNA and host factor RNase
III. Nature 471, 602-607 (2011).
41. Hatoum-Aslan, A., Maniv, I. & Marraffini, L. A. Mature clustered, regularly interspaced, short
palindromic repeats RNA (crRNA) length is measured by a ruler mechanism anchored at the
precursor processing site. Proc. Natl. Acad. Sc. 108, 21218-21222 (2011).
42. Jore, M. M. et al. Structural basis for CRISPR RNA-guided DNA recognition by Cascade. Nat.
Struct. Mol. Biol. 18, 529-536 (2011).
43. Lintner, N. G. et al. Structural and Functional Characterization of an Archaeal Clustered
Regularly Interspaced Short Palindromic Repeat (CRISPR) -associated Complex for Antiviral
Defense (CASCADE).J. Biol. Chem. 286, 21643-21656 (2011).
44. Westra, E. R. et al. CRISPR Immunity Relies on the Consecutive Binding and Degradation of
Negatively Supercoiled Invader DNA by Cascade and Cas3. Mol. Cell 46, 595-605 (2012).
102
45. Jinek, M. et al. A Programmable Dual-RNA-Guided DNA Endonuclease in Adaptive Bacterial
Immunity. Science 337, 816-821 (2012).
46. Wiedenheft, B. et al. RNA-guided complex from a bacterial immune system enhances target
recognition through seed sequence interactions. Proc. Natl. Acad. Sci. 108, 10092-10097
(2011).
47. Deveau, H. et al. Phage Response to CRISPR-Encoded Resistance in Streptococcus thermophilus.
J. Bacteriol. 190, 1390-1400 (2008).
48. Mojica, F. J. M., Diez-Villasefior, C., Garcia-Martinez, J. & Almendros, C. Short motif sequences
determine the targets of the prokaryotic CRISPR defence system. Microbiology 155, 733-740
(2009).
49. Sashital, D. G., Wiedenheft, B. & Doudna, J. A. Mechanism of Foreign DNA Selection in a Bacterial
Adaptive Immune System. Mol. Cell 46, 606-615 (2012).
50. Westra, E. R. et al. Type I-E CRISPR-Cas Systems Discriminate Target from Non-Target DNA
through Base Pairing-Independent PAM Recognition. PLoS Genet 9, e1003742 (2013).
51. Semenova, E. et al. Interference by clustered regularly interspaced short palindromic repeat
(CRISPR) RNA is governed by a seed sequence. Proc. Nat. Acad. Sci. 108, 10098-10103 (2011).
52. Qi, L. S. et al. Repurposing CRISPR as an RNA-Guided Platform for Sequence-Specific Control of
Gene Expression. Cell 152, 1173-1183 (2013).
53. Jiang, W., Bikard, D., Cox, D., Zhang, F. & Marraffini, L. A. RNA-guided editing of bacterial
genomes using CRISPR-Cas systems. Nat. Biotechnol. 31, 233-239 (2013).
54. Cong, L. et al. Multiplex Genome Engineering Using CRISPR/Cas Systems. Science 339, 819-823
(2013).
55. Mali, P. et al. RNA-Guided Human Genome Engineering via Cas9. Science 339, 823-826 (2013).
56. Jinek, M. et al. RNA-programmed genome editing in human cells. eLife 2, e00471 (2013).
57. Bikard, D. et al. Programmable repression and activation of bacterial gene expression using an
engineered CRISPR-Cas system. Nucleic Acids Res. 41, 7429-7437 (2013).
58. Robert J. Citorik, Mark A. Mimee & Timothy K. Lu. RNA-Guided Nucleases as ProgrammableSpectrum Antimicrobials and Microbial Population Sculptors. Science
59. Gomaa, A. A. et al. Programmable Removal of Bacterial Strains by Use of Genome-Targeting
CRISPR-Cas Systems. mBio 5, e00928-13 (2014).
60. Tamsir, A., Tabor, J. J. & Voigt, C. A. Robust multicellular computing using genetically encoded
NOR gates and chemical /'wires/'. Nature 469, 212-215 (2011).
61. Vieira, J. & Messing, J. in Methods in Enzymology (ed. Ray Wu, L. G.) Volume 153, 3-11
(Academic Press, 1987).
62. Datsenko, K. A. & Wanner, B. L. One-step inactivation of chromosomal genes in Escherichia coli
K-12 using PCR products. Proc. Natl. Acad. Sci. 97, 6640-6645 (2000).
63. Thomason, L. C., Costantino, N. & Court, D. L. in CurrentProtocolsin MolecularBiology (John
Wiley & Sons, Inc., 2001). at
<http://onlinelibrary.wiley.com/doi/10.1002/0471142727.mb0117s79/abstract>
64. Hirano, N. et al. Site-specific recombination system based on actinophage TG1 integrase for
gene integration into bacterial genomes. Appl. Microbiol. Biotechnol. 89, 1877-1884 (2011).
65. Lei Yang, Conor McClune, Timothy K. Lu & Christopher A. Voigt. Design of orthogonal memory
circuits.
66. Pul, U. et al. Identification and characterization of E. coli CRISPR-cas promoters and their
silencing by H-NS. Mol. Microbiol. 75, 1495-1512 (2010).
67. Datsenko, K. A. et al. Molecular memory of prior infections activates the CRISPR/Cas adaptive
bacterial immunity system. Nat. Commun. 3, 945 (2012).
68. Tyo, K. E. J., Ajikumar, P. K. & Stephanopoulos, G. Stabilized gene duplication enables long-term
selection-free heterologous pathway expression. Nat. Biotechnol. 27, 760-765 (2009).
103
69. Lee, N. L., Gielow, W. 0. & Wallace, R. G. Mechanism of araC autoregulation and the domains of
two overlapping promoters, Pc and PBAD, in the L-arabinose regulatory region of Escherichia
coli. Proc. Nat. Acad. Sci. 78, 752-756 (1981).
70. Yosef, I., Goren, M. G., Kiro, R., Edgar, R. & Qimron, U. High-temperature protein G is essential for
activity of the Escherichia coli clustered regularly interspaced short palindromic repeats
(CRISPR)/Cas system. Proc. Nat. Acad. Sci. 108, 20136-20141 (2011).
71. Brenner, S. Bacteriophage tales. Curr. Biol. 7, R736-R737 (1997).
72. Gullo, K. Ex-Dow Scientist Who Stole Secrets Gets 7 years, 3 Months Prison. Bloomberg
Businessweek (2011).
73. Kuzminov, A. & Stahl, F. W. Stability of linear DNA in recA mutant Escherichia coli cells reflects
ongoing chromosomal DNA degradation.]. Bacteriol. 179, 880-888 (1997).
74. National Institutes of Health. Guidelines for Research Involving Recombinant or Synthetic
Nucleic Acid Molecules. (2013).
75. Moe-Behrens, G. H. G., Davis, R. & Haynes, K. A. Preparing synthetic biology for the world. Front.
Microbiol. 4, (2013).
76. Wright, 0., Stan, G.-B. & Ellis, T. Building-in biosafety for synthetic biology. Microbiology 159,
1221-1235 (2013).
104