Article
https://doi.org/10.1038/s41467-023-37788-z
Addressable and adaptable intercellular
communication via DNA messaging
John P. Marken
1
& Richard M. Murray
1
Engineered consortia are a major rese
arch focus for synthetic biologists
because they can implement sophisticat
ed behaviors inaccessible to single-
strain systems. However, this function
al capacity is constrained by their con-
stituent strains
’
ability to engage in complex co
mmunication. DNA messaging,
by enabling information-
rich channel-decoupled c
ommunication, is a pro-
mising candidate architec
ture for implementing complex communication. But
its major advantage, its messages
’
dynamic mutability, is still unexplored. We
develop a framework for addressabl
e and adaptable DNA messaging that
leverages all three of these advantages and implement it using plasmid con-
jugation in E. coli. Our system can bias
the transfer of messages to targeted
receiver strains by 100- to 1000-fold, and their recipient lists can be dynami-
cally updated in situ to control the
fl
ow of information through the population.
This work lays the foundation for future
developments that further utilize the
unique advantages of DNA messaging t
o engineer previou
sly-inaccessible
levels of complexity into biological systems.
A major current focus of synthetic biology research is to expand
beyond the
fi
eld
’
s original paradigm of engineering a single cell strain
for a particular application and to instead engineer consortia, which
are populations consisting of multiple distinct cell types
1
,
2
. By enabling
the division of labor among its constituent strains, a consortia-based
approach allows each strain to specialize itself to its assigned task
while minimizing the metabolic burden on itself
3
.Engineeredcon-
sortia are, therefore, able to achieve higher levels of functional
complexity
4
–
6
and evolutionary stability
7
,
8
than analogous single-strain
systems.
In order for an engineered consortium to function properly,
however, it is necessary that each of its constituent strains can stably
coexist and act in concert with each other. This coordinated activity is
maintained by intercellular communication systems that allow the
strains to dynamically instruct each other to perform programmed
functions, like modulating their growth rate or activating a target gene.
Theachievablecomplexityofaconsortium
’
s behavior is therefore
constrained by the capacity of its communication channels to transmit
complex messages
6
. Realizing this, the synthetic biology community
has placed much effort towards expanding the toolbox of intercellular
communication channels and enabling increasingly information-dense
communication between cells
9
–
15
.
These efforts have almost exclusively focused on a molecular
architecture that we will term small molecule actuated communication
(SMA communication), wherein a sender cell synthesizes a small
molecule that diffuses through the extracellular environment to enter
a receiver cell that contains the requisite machinery to initiate a pre-
programmed response to the signal. SMA communication channels
were originally implemented using molecular parts co-opted from
quorum sensing systems
9
, but in recent years the toolbox has expan-
ded to include metabolites
10
, hormones
13
, and antibiotics
16
,
17
as signal
vectors.
An alternative molecular architecture, DNA messaging, was pro-
posed in a pioneering report by Ortiz and Endy
14
. Here, horizontal gene
transfer mechanisms are co-opted into a communication channel that
transmits DNA-encoded messages between cells (Supplementary
Fig. 1). Because the actual content of the message is an arbitrary
genetic sequence within the mobile vector itself, Ortiz and Endy
coined the term
“
message-channel decoupling
”
to describe the fact
that a single DNA-based communication channel can send different
messages that contain different types of instructions to the recipient
cells
14
. In contrast, SMA communication channels exhibit message-
channel coupling because the nature of the encodable message is tied
to the molecular identity of the signaling molecule. A homoserine
Received: 26 November 2022
Accepted: 31 March 2023
Check for updates
1
Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA, USA.
e-mail:
jmarken@caltech.edu
Nature Communications
| (2023) 14:2358
1
1234567890():,;
1234567890():,;
lactone, for example, can only be used to encode the instruction to
activate its cognate transcription factor, and an antibiotic can only be
used to encode the instruction to kill its susceptible cell
strains (Fig.
1
a).
A second important advantage of DNA communication is that a
single DNA message can encode a large amount of information con-
tent, as many horizontal gene transfer systems can easily transfer
several kilobases of arbitrary sequence
18
–
20
. In contrast, SMA channels
can only modulate their activity via the concentration of their signal
vector, a single small molecule. This heavily constrains the information
density of the message to the point where in applications like digital
computation, where concentrations are interpreted binarily as either
OFF or ON, a single SMA channel can only transmit a single bit of
information
21
.
Together, these advantages suggest that DNA messaging is an
ideal communication architecture for engineering complex consortia
with sophisticated information processing requirements. But although
the ten years since the Ortiz
–
Endy report have seen an increased use of
horizontal gene transfer systems by synthetic biologists to engineer
environmental microbiomes in the gut or soil
22
–
24
, further studies of
such systems
’
ability to act speci
fi
cally as a communication framework
for engineered consortia have only been performed
computationally
25
–
27
. Thus, to our knowledge, the original Ortiz
–
Endy
report remains the only experimental usage of DNA-based commu-
nication to date.
Why is the case? One reason is that, though it was pioneering in its
foresight, the Ortiz
–
Endy implementation did not demonstrate a third
property of DNA communication that is critical in enabling the
implementation of qualitatively new functionalities
—
the dynamic
mutability of DNA messages. Unlike SMA channels, where the message
is encoded into the structure of an immutable signal molecule, cells
have the ability to express DNA editors that can make targeted changes
to the content of the message in situ (Fig.
1
b). This ability has only
expanded with the recent explosion in research on programmable
DNA editors like CRISPR
–
Cas systems, integrases, and base editors
28
,
29
.
Although theoretical reports have rightly identi
fi
ed mutability as a key
advantage of DNA messaging
25
, to date, this property has not been
experimentally demonstrated.
We, therefore, set out to develop a general and scalable archi-
tecture for DNA messaging that allows users to fully take advantage of
all three of its unique properties: message-channel decoupling, high
information density, and dynamic message mutability. In order to
ensure our framework
’
s compatibility with arbitrary messages trans-
ferred along arbitrary horizontal gene transfer systems, we used
channel-orthogonal molecular tools to implement a functionality that
is required in all communication systems
—
the ability to address the
message to a targeted set of recipients.
Our addressing framework uses CRISPR
–
Cas systems to internally
validate each message transfer event within the consortium, enabling
the targeted delivery of a given message to any subset of the strains in
a population. We additionally designed a framework for using inte-
grases to modularly update messages
’
recipient lists in situ, enabling
the control of information
fl
ow through a population. This work
establishes a universally applicable framework for effective DNA-based
communication that sets the stage for future efforts that expand its
ability to implement previously-inaccessible functionalities into engi-
neered consortia.
Results
Incorporating massage addressability into a plasmid
conjugation-based c
ommunication system
We
fi
rst describe the implementation of an addressability system for
our DNA messaging framework. Any such implementation requires a
means for the molecular recognition of speci
fi
c genetic sequences,
and we chose to use the CRISPR
–
Cas adaptive immunity system due to
its ability to programmatically target and cleave desired nucleotide
sequences on genetic vectors entering the cell
30
,
31
. Although multiple
different Cas systems have been demonstrated to cleave and degrade
DNA vectors within cells
32
,
33
,wespeci
fi
cally chose to use the
S. pyo-
genes
Cas9 endonuclease system because it contains the required
binding, unwinding, and cleaving activities within a single protein,
facilitating its use in many different host organisms
34
.Additionally,
well-developed procedures exist for generating large libraries of
orthogonal single-guide RNAs
(gRNAs) for the Cas9 system
15
,
35
,
36
,and
the small footprint of the gRNA binding site (23 bp) means that many
such sites can be incorporated onto a DNA message without sig-
ni
fi
cantly burdening any potential sequence length constraints from
the transfer system. Together, t
hese properties make the Cas9
–
gRNA
system an ideal candidate for implementing a scalable, modular, and
host-orthogonal addressing system for DNA messaging.
The design of our addressability framework is as follows. Each
receiver cell in the consortium expresses both Cas9 and a unique gRNA
that serves as a molecular signature encoding its strain identity. The
sender cells themselves require no additional molecular machinery,
but the DNA message must contain an array of gRNA binding sites that
Fig. 1 | Architectures for engineered intercellular communication. a
Small
molecule actuated (SMA) communication systems exhibit message-channel cou-
pling, meaning that the behavior they induce in the receiver cell is hard-coded into
the molecular identity of the signaling molecule itself. This molecular identity
cannot be changed without disrupting the functioning of the channel itself.
b
DNA
communication systems exhibit message-channel decoupling, meaning that a
given channel can transmit multiple types of messages to induce any genetically-
encodable response in the receiver cell. Furthermore, the cells themselves can
express molecular DNA editors to change the content of the messages in situ,
closing the loop to enable autonomous system recon
fi
guration.
Article
https://doi.org/10.1038/s41467-023-37788-z
Nature Communications
| (2023) 14:2358
2
correspond to the receiver strains that should not receive the message.
We will refer to this array as the address region because it acts as a
blocklist, encoding the recipient list of the message as the set of strains
whose gRNAs are not encoded in the address (Fig.
2
a). When the
message is transferred to a receiver cell, the Cas9
–
gRNA complex
checks the validity of the transfer
—
if the transfer is invalid, then the
complex will bind to the cognate site on the address region and cleave
the message, leading to its degradation. If the transfer is valid, then the
Cas9
–
gRNA complex is unable to interact with the message, and so the
message freely propagates within the receiver cell. This process is
schematized in Fig.
2
b.
An important property of this addressing framework is that the
transfer validation system interacts with the message itself rather than
the transfer machinery that carries the message. This means that a
single DNA channel can send messages that are addressed to different
recipients. When addressability is implemented via channel-intrinsic
properties, such as in the Ortiz
–
Endy system
’
s reliance on the M13
bacteriophage
’
snarrowinfectionhostrange
14
, every message that is
transmitted by a channel must go to the same recipient list regardless
of its content.
In demonstrating the incorporation of our message-addressing
framework into a DNA-based communication system, we chose to
deviate from Ortiz and Endy
’
s original choice of the
fi
lamentous bac-
teriophageM13andinsteadusedaplasmidconjugation-basedcom-
munication system. This is because the properties of plasmid
conjugation systems are better aligned with the advantages of DNA-
based communication as a whole
—
plasmids can encode larger mes-
sages, with conjugative plasmids regularly reaching lengths of hun-
dreds of kilobases
20
,
37
, and can transfer to taxonomically-diverse
recipients
38
,
39
, facilitating their use in multispecies consortia. We spe-
ci
fi
cally chose to use the F
HR
system developed by Dimitriu et al.
18
,
which is based on the
Escherichia coli
fertility factor F, the canonical
representative of conjugative plasmids
40
.
Cas9-mediated blocking of plasmid receipt is inducible and
orthogonal
In order to demonstrate that Cas9-mediated cleavage can indeed block
the receipt of a mobilized plasmid, we performed pairwise sender-
receiver experiments in
E. coli
consortia using the F
HR
-based commu-
nication system. Receiver cells containing a genomically integrated
spectinomycin resistance cassette were transformed with a plasmid
encoding OHC14-HSL-inducible expression of Cas9 and one of two
gRNAs (
“
A
”
or
“
B
”
), and sender cells containing a genomically inte-
grated apramycin resistance cassette were transformed with the F
HR
helper plasmid and a pSC101 message plasmid that constitutively
expresses a yellow
fl
uorescent protein and chloramphenicol resistance
gene. Two variants of this message plasmid were constructed, differing
in whether their address region contained a single A binding site or a
single B binding site (Fig.
3
a).
With this setup, selective plating could be used to individually
isolate the senders, receivers, and transconjugants from a mixed
population and calculate their densities. We performed mating
experiments on all four combinations of sender-receiver pairs in the
presence and absence of OHC14-
HSL induction and measured the
densities of each strain after 6 h of growth in a shaken LB coculture
(Fig.
3
b). The message plasmid was transferred ef
fi
ciently to the
receivers in this timeframe, with an average of 64% of receivers being
converted to transconjugants across all transfers to on-target reci-
pients (Supplementary Fig. 2).
We then quanti
fi
ed the effectiveness of Cas9-mediated plasmid
blocking by calculating the plasmid transfer rate in each experiment,
de
fi
ned as the transconjugant density divided by the product of the
total sender and total receiver den
sities. We observed that when the
Cas9 system was induced, the A-containing message plasmid had a 185-
fold higher transfer rate to its valid recipient (the B receiver) than to its
invalid recipient (the A receiver) (
p
= 0.03, paired
t
test), and that for
the B-containing message plasmid, the difference was 520-fold
(
p
=0.01, paired
t
test) (Fig.
3
c). When the Cas9 system was not
induced in the receiver cells, this biased transfer was not observed
(
p
= 0.28, 0.94, paired
t
test, for A and B message plasmids, respec-
tively) (Fig.
3
c).
Having demonstrated that our addressability system performed
successfully in a two-strain population, we next asked whether our
system could scale to multi-strain populations where a given address
region may need to encode several gRNA binding sites. We con-
structed three different receiver strains that, in addition to the spec-
tinomycin resistance gene, each express a distinct
fl
uorescent protein
(mScarlet-I
41
,sfYFP
42
, or TagBFP
43
) from a genomically integrated cas-
sette. In this way, all three receivers could be mixed together with the
sender strain in a four-strain coculture, and the colors could be used to
determine the density of each distinct receiver strain after selective
plating. In order to further assess the generality of our Cas9-mediated
blocking system, we used a set of orthogonal gRNAs developed by
Didovyk et al.
35
instead of reusing the A and B gRNAs from the previous
experiment. We transformed each of the colored receiver strains with a
plasmid encoding Cas9 and one of three of the Didovyk gRNAs (D1, D2,
or D3) and constructed sender strains containing one of eight message
plasmids addressed to every possible combination of the three recei-
ver strains (Fig.
4
a).
We found that even in the more complex setting of a four-strain
population, our system was able to preferentially deliver the message
Fig. 2 | Addressable DNA messaging. a
Schematic of an addressable DNA message.
The content of the message is an arbitrary genetic sequence, and the address
region uses gRNA binding sites to act as a blocklist that determines the message
’
s
recipient list by excluding transfer to all encoded strains. The origin of transfer
(
oriT
) allows the message to interact with the cognate horizontal gene transfer
machinery in the sender cell.
b
Schematic of transfer blocking. The DNA message is
initially transferred promiscuously to all receiver strains in the population. As the
message enters a receiver cell, the binding sites on the address region become
exposed to cleavage by the Cas9
–
gRNA complex expressed within the cell. This
cleavage only occurs if a binding site on the address matches the gRNA expressed in
the receiver cell, thus ensuring that the message only persists within its appropriate
recipients by eliminating the messages sent to invalid recipients.
Article
https://doi.org/10.1038/s41467-023-37788-z
Nature Communications
| (2023) 14:2358
3
to its appropriate recipients. Across all transfers to on-target reci-
pients, the average fraction of receiver cells converted to transconju-
gants was 60%, and the fold change in transfer rate between valid and
invalid recipients was often over 1000-fold (Fig.
4
b; Supplementary
Fig. 3). Although the three gRNAs used in the receivers were previously
reported to be of comparable effectiveness in a dCas9-mediated
transcriptional repression assay
35
, the D1 and D2 gRNAs were able to
block invalid transfers much more strongly than the D3 gRNA
—
the
geometric mean of the fold change in transfer rates between valid and
invalid recipients across all conditions where the invalid recipients
expressedtheD3gRNAwas79-fold,comparedto1256-foldand1577-
fold for the D1 and D2 gRNAs, respectively (Supplementary Fig. 3).
Cells can use integrases to edit DNA messages in situ and update
their recipient list
Having demonstrated that our Cas9-mediated blocking system can
successfully implement high-
fi
delity addressable communication
between cells, we next proceeded to incorporate adaptability into the
message transmission framework by enabling the programmable
in situ editing of a message
’
s recipient list. This can be accomplished by
applying molecular DNA editors to modify the gRNA binding sites on
the address region. Speci
fi
cally, a system for programmable address
editing should have the ability to both add a new binding site to the
array and remove (or invalidate) an existing binding site from the array.
Serine integrases are a class of proteins that are well-suited for this
task because of their ability to bind to speci
fi
c attachment sequences
and add, remove, or swap the regions between these sites depending
on their con
fi
guration and orientation along the DNA
44
,
45
.Their
Fig. 3 | Cas9-mediated cleavage of incoming plasmids can bias their transfer to
targeted recipients. a
Schematic of the experimental setup. Senders (
S
)and
Receivers (
R
) carrying one of two plasmid variants are grown together in a cocul-
ture, and selective plating is used to isolate them, as well as the transconjugants (T),
from the mixed culture. Note that transconjugants will appear on the receiver-
selecting plates, so
R
is the total density of receivers in the population (Methods).
b
Endpoint strain densities, measured in colony forming units (CFUs) per mL of
culture. (
c
) Transfer rates, calculated as
T
/(
S
∗
R
), of the message plasmid in each of
the conditions in (
b
). Dots show the values from three biological replicates mea-
sured on different days, and bars depict the geometric mean of these values. Km
kanamycin, Cm chloramphenicol, Ap apramycin, Sp spectinomycin, Cin = OHC14-
HSL. Source data are provided as a Source Data
fi
le.
Fig. 4 | Programmable delivery of message plasmids to arbitrary subsets of a
multi-strain population. a
Schematic representation of the intended recipient list
for each of the eight message plasmids. Dark squares indicate an invalid transfer,
and light squares indicate a valid transfer.
b
The observed geometric mean of the
transfer rates to each receiver type, calculated from three biological replicates
measured on different days. The color map is scaled logarithmically over four
orders of magnitude. Individual transfer rate values are shown in Supplementary
Fig. 3. Source data are provided as a Source Data
fi
le.
Article
https://doi.org/10.1038/s41467-023-37788-z
Nature Communications
| (2023) 14:2358
4