Article

Sub-1.4 cm3 capsule for detecting labile

inflammatory biomarkers in situ

https://doi.org/10.1038/s41586-023-06369-x M. E. Inda-Webb1,2,12, M. Jimenez3,4,5,12, Q. Liu6, N. V. Phan4, J. Ahn4, C. Steiger3,4,5,

A. Wentworth3,4,5, A. Riaz6, T. Zirtiloglu6, K. Wong5, K. Ishida3,4,5, N. Fabian3,4,7, J. Jenkins3,4,
Received: 10 May 2022
J. Kuosmanen3,4, W. Madani3,4,5, R. McNally5, Y. Lai1,2, A. Hayward3,4,5,7, M. Mimee8, P. Nadeau9,
Accepted: 26 June 2023 A. P. Chandrakasan10, G. Traverso3,4,5 ✉, R. T. Yazicigil6 ✉ & T. K. Lu1,2,11 ✉

Published online: 26 July 2023

Check for updates Transient molecules in the gastrointestinal tract such as nitric oxide and hydrogen
sulfide are key signals and mediators of inflammation. Owing to their highly reactive
nature and extremely short lifetime in the body, these molecules are difficult to detect.
Here we develop a miniaturized device that integrates genetically engineered probiotic
biosensors with a custom-designed photodetector and readout chip to track these
molecules in the gastrointestinal tract. Leveraging the molecular specificity of living
sensors1, we genetically encoded bacteria to respond to inflammation-associated
molecules by producing luminescence. Low-power electronic readout circuits2
integrated into the device convert the light emitted by the encapsulated bacteria to a
wireless signal. We demonstrate in vivo biosensor monitoring in the gastrointestinal
tract of small and large animal models and the integration of all components into a
sub-1.4 cm3 form factor that is compatible with ingestion and capable of supporting
wireless communication. With this device, diseases such as inflammatory bowel
disease could be diagnosed earlier than is currently possible, and disease progression
could be more accurately tracked. The wireless detection of short-lived, disease-
associated molecules with our device could also support timely communication
between patients and caregivers, as well as remote personalized care.

Our ability to diagnose and monitor inflammatory gastrointestinal dis- environment underlying the metabolic pathways of both the human
orders would be transformed if we could profile labile, oxidation-related host and its resident microorganisms. Developing non-invasive
biomarkers and their responses to dietary change and therapies in situ. technologies that can continuously monitor the gastrointestinal
Many microbiome-related conditions, notably inflammatory bowel dis- environment in situ would both expand our understanding of what
ease (IBD), are associated with chronic intestinal inflammation result- causes inflammation and indicate directions for improving therapies.
ing from dysregulated immune homeostasis—specifically, increased Furthermore, diagnosing multi-faceted diseases such as IBD, in which
oxidation3,4. Malnutrition5, antibiotic resistance6, antibiotic dysbiosis7,8, biomarker levels vary substantially among patients, would benefit from
neurodegenerative diseases9 and mitochondrial genetic disorders10 the simultaneous detection of a panel of oxidation-related biomarkers
are associated with redox imbalance in the gastrointestinal tract, and (for example, nitric oxide17 (NO), ROS18, thiosulfate19 and tetrathionate19).
oxidative stress may also underlie poor responses to chemotherapy11, Current methods of diagnosing gastrointestinal inflammation
and vaccines12, as well as aging13. include endoscopy20, which is invasive and should only be performed
Bacterial infections14 and antibiotics7 may substantially increase with limited frequency, and stool analysis, which may not accurately
concentrations of oxidants, such as reactive oxygen species (ROS) and reflect intestinal conditions owing to differential growth of certain
reactive nitrogen species (RNS). These molecules are labile, which can species21, ambient oxidation13 and loss of labile disease-mediating
make it difficult to detect their presence or accurately measure their molecules. Culture enrichment22 (for example, for analysing low-
concentration in the body. Devices reported to sense labile molecules abundance microorganisms in stool samples with omics techniques)
in the gastrointestinal tract (such as oxidizing gases or volatile organic may also distort the initial bacterial ratio. Because the faecal microbiota
compounds) are limited to off-the-shelf sensors that use non-specific only partially represents the autochthonous microbiota in direct con-
metal-oxide-sensing elements15,16. Thus, the current standard of clinical tact with the intestinal mucosa23, a biopsy may be required for complete
care is limited with respect to our capacity to evaluate the chemical analysis24.

1
Synthetic Biology Group, MIT Synthetic Biology Center, Department of Biological Engineering, Massachusetts Institute of Technology, Cambridge, MA, USA. 2Research Laboratory of Electronics,
Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, USA. 3Department of Mechanical Engineering, Massachusetts Institute of
Technology, Cambridge, MA, USA. 4The Koch Institute for Integrative Cancer Research, Massachusetts Institute of Technology, Cambridge, MA, USA. 5Division of Gastroenterology, Brigham and
Women’s Hospital, Harvard Medical School, Boston, MA, USA. 6Electrical and Computer Engineering Department, Boston University, Boston, MA, USA. 7Division of Comparative Medicine, MIT,
Cambridge, MA, USA. 8Department of Microbiology, Biological Sciences Division and Pritzker School of Molecular Engineering, The University of Chicago, Chicago, IL, USA. 9Analog Devices,
Boston, MA, USA. 10Department of Electrical Engineering and Computer Science, MIT, Cambridge, MA, USA. 11Senti Biosciences, South San Francisco, CA, USA. 12These authors contributed
equally: M. E. Inda-Webb, M. Jimenez. ✉e-mail: [email protected]; [email protected]; [email protected]

386 | Nature | Vol 620 | 10 August 2023

 a Biomarker

Recombinase

GFP
Recombinase-based switch
Model of chemical landscape of
c health and disease

BM3 BM3 BM3 Follow-up

Immature microbiome
BS1 BS1 BS1
BS2 BS2 BS2
BS3 BS3 Treatments
BS3
b
Treatments

Development
1. Molecular
detection
8.5 mm

3. Wireless Health Disease

2. Light signal

14.25 mm Mature microbiome

Microbiome state

Fig. 1 | General platform for developing a miniaturized capsule for real-time signal from the intestinal biomarkers detection has been tested in terminal
detection of labile mediators of disease in the gut. a, Probiotic bacteria are experiments in pigs). c, Top, the device could enable remote detection for
engineered to respond to an array of IBD biomarkers (BM). A recombinase- immediate follow-up after therapy, as well as monitoring the gut chemical
based genetic memory system is used to validate the bacterial biosensors (BS) environment for longer-term treatment with personalized therapies or
in animal models. Biosensing bacteria are then re-engineered to respond by dietary and lifestyle changes. Microbiome dynamics could be characterized
luminescing and packaged in a capsule that protects the cells during exposure as depicted in the simplified model in the bottom right. Starting from infancy,
to low pH (such as during passage through the stomach) and interfaces the microbiome gradually reaches one of several adult states characterized by
them with miniaturized electronics (the illustration shows the design and health or disease. Monitoring the gut chemical environment is essential for
dimensions of our fabricated device). b, While in transit through the intestines, timely treatments: perturbations that would lead to an unhealthy state can
the biosensing bacteria sense the metabolites as they are being produced inside be resisted (red blocked arrow) or an effective treatment plan can be used to
the body and the miniaturized capsule transmits the bacterial luminescence return to a healthy state 58 (green arrows). The biosensor image is reproduced
signal wirelessly to an external device, such as a mobile phone (the wireless from ref. 59.

Engineered living bacteria, due to their innate sensing ability and as a smartphone. We have tested this multi-diagnostic pill in live pigs,
robust functionality in the gut, have been proposed as a solution to demonstrating that a human-scale diagnostic device can be built to
detect analytes associated with intestinal inflammation in complex detect transient mediators of gastrointestinal inflammatory diseases
physiological environments1—such as in clinical samples25 (for example, (Fig. 1).
serum or urine) and inside the gut, detecting either biomolecules sup-
plemented in diet26–28 or by-products of inflammation19,29,30. However,
current applications of biosensors still depend on the complex analysis Design of biosensing genetic circuits
of bacterial gene expression, RNA or DNA in stool, rather than real-time Intestinal bacteria have natural sensors that continuously detect spe-
reporting from localized regions within the body. cific molecules in the gut. We chose probiotic Escherichia coli Nissle
Electronic devices that continuously collect, process and wirelessly 1917 as a chassis owing to its resilience and excellent safety profile
transmit information can also be used to analyse the gastrointestinal in the gastrointestinal tract36, and engineered it to detect the labile
tract. However, capsule endoscopy cameras currently approved by IBD-mediating molecules NO, H2O2, thiosulfate and tetrathionate
the US Food and Drug Administration31 do not directly measure the (Extended Data Fig. 1).
molecular mediators of disease such as ROS or RNS. Other ingestible, We first optimized the NO sensor by combining the bacterial NO
ultralow-power electronic devices currently under development biosensor NorR17,37,38 with a DNA recombinase core circuit39 to create
can be used to visually evaluate the gastrointestinal tract and meas- memory circuits that could record NO exposure of bacterial biosen-
ure gas concentrations, temperature and pH levels15,32 but require sors as they travelled through the gastrointestinal tract (Fig. 2a) in
functionalization with fragile transducers to convert biochemi- the dextran sodium sulfate (DSS) model of intestinal inflammation
cal information into electronic signals, which limits specificity and in mice. Because the recorded information is stored in the DNA and
robustness15,33. passed from generation to generation, it can be retrieved by meas-
To overcome these challenges and to leverage the promise of tran- uring GFP expression from the bacteria recovered from faeces.
sient biomarker panels, we combined natural protein-based sensing Using flow cytometry, the percentage of cells that were GFP ‘ON’18
elements for NO, hydrogen peroxide (H2O2), tetrathionate and thiosul- (the percentage of GFP-positive cells) was calculated at each concen-
fate with genetically encoded memory circuits, incorporated them into tration of diethylenetriamine–nitric oxide adduct (DETA–NO). After
probiotic bacteria, and validated their function in vivo in mouse and multiple rounds of optimization to improve the signal-to-noise ratio
pig models of inflammation. We then integrated these bacterial sensors (SNR) of the genetic circuits and characterization (Extended Data
with a custom-designed integrated photodiode array and readout chip. Fig. 2a–c), our NO recombinase-based memory system responded
This integrated system builds on our early prototype34, which was only to a concentration threshold of 30 μM NO with high specificity and
validated for the detection of blood in vivo but was considerably larger was not influenced by other RNS (NOx) present in the gut environ-
(more than 9 cm3) than any known safe ingestibles. The new pill has a ment (Fig. 2a). The NO sensor was tested under anaerobic condi-
volume of less than 1.4 cm3 and a pill form factor that conforms to the tions (Extended Data Fig. 2d) and over time (Extended Data Fig. 2e,f).
size of safe ingestible non-deformable dosage forms on the market35. Similarly, we built recombinase-based memory circuits to optimize
Our miniaturized wireless bio-electronic pill can safely process data the detection of H2O2, thiosulfate and tetrathionate (Extended Data
with low power consumption and transmit it to a portable device such Fig. 3a–d).

Nature | Vol 620 | 10 August 2023 | 387

Article
a b
100 Sensor + NO Bacterial
gavage
30 ***
Sensor + NOx
80 Control + NO

GFP+ cells (%)

GFP+ cells (%)

25
60 NO Day 0 0

Control
norR bxb1

DSS
40 PnorV
20
20
gfp
0 15
0.1 1 10 100 1,000 10,000 Control DSS
Stool collection
Nitric oxide concentration (μM)
c d Negative Healthy Ischemia-
Negative control 0.6 mM DETA-NO control reperfusion
0 mM DETA-NO 6 mM DETA-NO
NO sensor 1 NO sensor 2
100 100
** Mesentery artery ***
80 ** *** 80
GFP+ cells (%)

Intestinal arteries

GFP+ cells (%)

60 *** Buffer
60

40 Small intestine 40

20 20

0 0
NO NO NO H D K H D K
sensor 1 sensor 2 sensor 3

Fig. 2 | Validation in vitro and in vivo of probiotic bacteria engineered to ***P = 0.002. Two-tailed unpaired Student’s t-test. c, The NO sensor was
detect NO. a, Inset, in response to NO, the transcription factor NorR binds to validated in pigs. Right, experimental design: intestines were clamped to
the PnorV promoter (inset), activating the transcription of bxb1 recombinase, separate control and treated compartments. Bacterial NO sensors 1, 2 and
leading to inversion of the gfp expression cassette (located between attB and 3—with activation thresholds of 0.6 mM, 0.6 to 6 mM and 6 mM DETA–NO,
attP sites, triangles) and expression of GFP. Flow cytometry was used to respectively—were placed in the different compartments. n = 3 per group.
determine the percentage of GFP-positive cells at various NO concentrations. d, Ischaemia-reperfusion model of inflammation in pigs. Data are shown from
RNS other than NO (NOx) did not activate the system. n = 3 per group. b, Left, four independent experiments (four pigs: H, D, M and K) on different days,
experimental timeline for validation in mice. C57BL/6 male mice were treated with multiple compartments per animal. ***P = 0.0005. Two-way ANOVA for
for 7 days with 3% DSS in drinking water. DSS-treated mice and control were multiple comparisons. In c,d, bacteria were collected from the intestine after
gavaged every other day with engineered bacteria carrying the NO memory two hours of exposure to the analyte or the ischaemia-reperfusion. In b–d, the
system. Stool was collected 6 h after gavage. Right, the NO sensor was percentage of GFP-positive cells was measured by flow cytometry (n = 10,000
significantly activated on day six of DSS treatment. DSS: n = 10; control: events) from separately grown culture after retrieval. Data are mean ± s.e.m.
n = 10. Sensor 1: **P = 0.0015; sensor 2: **P = 0.0067, ***P = 0.0007; sensor 3:

We also created a disease-stage detector based on the concentra- GFP activation increased significantly at day nine after the start of
tion of NO detected, which would indicate whether inflammation DSS treatment (Extended Data Fig. 4a), correlating with the peak of
was mild, moderate or severe (Extended Data Fig. 3e). The incor- inducible nitric oxide synthase (iNOS) activation40. Tracking NO with
porated recombinase-based switch was also used to discretize the the disease-stage detector revealed an exacerbated inflammatory
biomarker input levels and create digital memories in the cells18. response following antibiotic treatment in a chronic inflammation
Tuning the sensitivity of the NO biosensor resulted in different model (Extended Data Fig. 4d,e). We also tested our bacterial sensors
activation thresholds, so that physiologically relevant concentra- for detection of ROS (H2O2), tetrathionate and thiosulfate; activation
tions of the biomarker (low, medium or high) could be measured was detected as inflammation progressed, with thiosulfate detection
in animal models. All of the sensors developed for in vivo valida- at day six after the start of DSS treatment and tetrathionate (a product
tion in mice measured GFP expression as a readout of the memory of thiosulfate oxidation) detection at day ten, overlapping with ROS
system activation; the sensors were activated within minutes detection (Extended Data Fig. 3b,c).
(Extended Data Fig. 2e), making them suitable for detection in the To validate the bacterial sensors in a disease model that is similar
intestine36. in physiology and scale to human anatomy41, we tested the bacte-
rial biosensors in living pigs using an ischaemia-reperfusion injury
model of intestinal inflammation42. We injected the bacterial bio-
Biosensors detect disease in animal models of IBD sensors directly into the intestinal lumen of a sedated pig, in either
To examine the in vivo functionality of our bacterial sensors, we evalu- inflamed segments or healthy segments with or without added bio­
ated whether they could detect chemically induced gastrointestinal markers. The NO-sensing bacteria detected different concentrations of
inflammation while passing through the gastrointestinal tract in NO in the control group (Fig. 2c) and emitted a positive signal in
a mouse model of colitis (Fig. 2b). After inflammation was induced some of the animals in the disease group (ischaemia-reperfusion;
with DSS, control and treated mice were orally gavaged with the NO Fig. 2d). Although previous direct NO measurements in inflamed
biosensors. Six hours later34, the percentage of GFP+ cells recovered intestines are limited and uncertain, based on the lowest detectable
from faecal samples was analysed by flow cytometry. NO biosensors signal from the NO biosensor bacteria 1 (around 8 μM; Extended
detected significantly more GFP+ cell activation in faecal pellets from Data Fig. 3e) our results agree with an estimated range of expected
treated mice than in those from healthy controls (Fig. 2b). Inflammation NO (1–10 μM) in humans and suggest that the biosensors are functio­
in the colitis model was measured over time (Extended Data Fig. 4a) nal in the complex, dynamic environment of an inflamed intestine43.
and independently validated (Extended Data Fig. 4b,c). Our biosen- Additionally, the ROS, thiosulfate and tetrathionate sensors also
sor thus detected the presence of NO as a marker of gastrointestinal detected significant quantities of analytes in control-treated pigs
inflammation in vivo. (Extended Data Fig. 4f).

388 | Nature | Vol 620 | 10 August 2023

 a b
Filter membrane

Adhesive film

8.5 mm
3D printed top
(chamber bodies)
14.25 mm

Clear backing film

Microelectronics PCB

3D printed bottom 5 mm

BS3
Ref
BS1
c d

BS2
ROS BS, exp 1 ROS BS, exp 3
BM ROS BS, exp 2 ROS BS, exp 4
BSP Luc 1 4
On Off On
2 3
150 400
TT
Relative photocurrent (fA)

Relative photocurrent (fA)

Buffer
300
100

200
50
100
0
0

–50
–100
0 30 60 90 120 0 6 12 18 24
Time (min) Time (h)

Fig. 3 | Design and in vitro characterization of a device for miniaturized used to detect bacterial sensor output; (2) battery; (3) microcontroller and
wireless sensing with cell-based biosensors. a, Components and dimensions radio chip; (4) antenna. Lines represent the mean and the shaded area indicates
of the device. b, Design of a miniaturized pill casing with a bacterial-electronic the the s.e.m. for three independent replicates, conducted with one induced
chamber interface. Top, side view of the device. Bottom, fully assembled pill device (tetrathionate) and one uninduced device (buffer). BM, biomolecule
with the permeable membrane attached. The bacterial chamber–casing that induces the circuit; BSP, biosensing protein; Luc, luciferase reporter
unibody design uses a thin clear backing film to place the bacteria close to operon; ref, reference chamber. d, In vitro characterization of the device for
the photosensitive electronic chip. A double-sided adhesive film enables a long-term use, measuring reversibility over time for 24 h. Probiotic bacteria
low-profile seal to the permeable outer filter membrane. c, Top left, the genetic engineered to detect ROS (ROS BS) were encapsulated in the device and
circuit and wireless signal over time from the tetrathionate sensor encapsulated immersed in bacterial growth media intermittently supplemented with 20 mM
in the device and immersed in bacterial growth media supplemented with H2O2 to test reversibility. Wireless signals were recorded over time. The dynamic
100 mM tetrathionate. Low-power CMOS-integrated photodiodes convert circuit for ROS performed with stringent control during the off phase and
bioluminescence emitted from the bacterial sensor to a photocurrent, which showed a perfect reactivation over the second on phase. Lines represent four
is converted to quantifiable digital data and transmitted wirelessly to the independent replicates (exp 1–4) conducted with two different chips on different
external device. Top right, Multiple biosensors (BS) can be used together days over multiple cycles induced with 20 mM H2O2, followed by a wash and
with this array to study a metabolic pathway (for example, H2O2, thiosulfate incubation in growth medium and induced again by supplementing the growth
and tetrathionate sensors; Extended Data Fig. 3c). (1) A threshold-based medium with 20 mM H2O2.
bioluminescence detector with a CMOS-integrated photodiode2 array was

(Extended Data Fig. 5b) were sealed with porous membranes (nominal
Validation of a bacterial-electronic sensor pill pore size, 0.4 µm), which did not interfere with the detection of the tar-
To advance our bacterial sensors towards clinical application, we get molecules when the chambers were placed in faeces (Extended Data
integrated them into a bacterial-electronic pill. Specifically, we Fig. 6a). Additionally, to protect the on-board bacteria during ingestion
designed a system with a size and form factor conforming to a proven and transit through the stomach, we developed free-standing enteric
non-deformable dosage form (Extended Data Fig. 5a) by developing polymer films that prevent low pH ingress through simulated ingestion
an optimized luminescent readout via a custom microelectronic bio- (Extended Data Figs. 6b–e and 7). To generate versions of the bacterial
luminescence detector2 and bacterial chambers built directly into sensors that can optimally couple to the photosensitive electronics,
the pill casing. we replaced the GFP readout with a self-contained bioluminescence
Our multi-diagnostic device requires nutrients and analytes to be readout, the luxCDABE operon from Photorhabdus luminescens34,
exchanged efficiently while retaining the engineered microorganisms and the memory system with a reversible analogue output. We con-
and simultaneously allowing the generated light signal to reach the structed this genetic circuit in E. coli Nissle 1917 and exposed the result-
electronic detectors. Thus, we developed a pill casing that incorporates ant strain to DETA–NO as a source of NO (Extended Data Fig. 8a–c).
a bacterial-electronic chamber interface in a unibody design (Fig. 3a,b) The NO-biosensing bacteria responded rapidly to DETA–NO exposure
in which the bacteria are precisely aligned in a tablet-shaped two-by-two (time to reach maximum signal (tmax) = 60 min) with a high luminescence
array with the microelectronic photodiodes and a hermetic seal is main- output and a SNR of 170 (Extended Data Fig. 8a). Production of the lumi-
tained around the electronic components. This design enabled sensing nescence machinery was also induced by DETA–NO in cultures grown
from an array of engineered microbial strains expressing luciferase under anaerobic conditions (Extended Data Fig. 8d) and the machinery
in response to several biomarkers (NO, ROS, tetrathionate and nega- was stable through simulated ingestion (Extended Data Fig. 7). ROS, thi-
tive reference). To retain the bacterial cells, the integrated chambers osulfate and tetrathionate sensors were also built in combination with

Nature | Vol 620 | 10 August 2023 | 389

Article
a b c
Mesentery artery Compartmentalized intestine
with the devices inside the
Intestinal arteries abdomen
37 ºC

Buffer

Small intestine

Uninduced Induced

d e
100 1.0
TT
Buffer
Relative photocurrent (fA)

0.8
50

Sensitivity
0.6
0
0.4

–50
0.2
TT sensor
Null sensor
–100 0
0 30 60 90 120 0 0.2 0.4 0.6 0.8 1.0
Time (min) 1 – specificity

Fig. 4 | Validation of the integrated device for miniaturized wireless at 37 °C. Wireless signals transmitted from inside the abdomen were detected
biosensing in live pigs. a, Schematic of the experiment. The bacterial by a commercial receiver, which can be connected to a laptop computer or
biosensors encapsulated in the device were tested in situ in a terminal procedure a mobile phone. d, Kinetics of the tetrathionate sensor embedded in the
in anaesthetized Yorkshire pigs weighing around 90 kg. After opening the abdominal cavity of a pig. The response of the device placed in the compartment
abdominal cavity, we placed the device through a small incision directly into with tetrathionate was clearly distinguishable from that of the device in the
the lumen of the small intestine, compartmentalized with clamps. b, Schematic compartment with the buffer control. Dark lines represent the mean of
of compartmentalization of a section of the intestine that was clamped for independent experiments in different pigs with up to two devices in separate
experimentation. Tetrathionate (100 mM) was injected with a syringe into the compartments (tetrathionate (TT) and buffer) and the shaded area represents
clamped intestinal compartment. Buffer was added as a control in a different the s.e.m. Tetrathionate: n = 5; buffer: n = 3. e, The receiver operating
compartment. c, Compartmentalized intestines were kept inside the abdomen characteristic of the device with sensing after 60 min.

the luxCDABE operon for real-time detection and characterized in vitro successfully interface with the pill to give an electronic readout in the
(Fig. 3c and Extended Data Fig. 8a–c); the tetrathionate sensor reached presence of the target molecules (Extended Data Fig. 9b,c) and even
the highest luminescence values in simulated intestinal fluid (Extended show reversibility (Fig. 3d; ROS sensor).
Data Fig. 8b) and was selected for initial optimization of an integrated To model the complex milieu of the gastrointestinal tract, the inte-
device. Finally, we designed a millimetre-scale capsule printed circuit grated device was also tested in vivo in pig small intestines. Upon induc-
board (PCB) housing a complementary metal-oxide-semiconductor tion, luminescence was detected by the custom-designed electronic
(CMOS) bioluminescence detector chip with an integrated photodiode readout circuits in the capsule; the information was wirelessly trans-
array achieving high sensitivity (Fig. 3c). The custom chip integrates mitted in real time from inside the body of the living pig to an external
a threshold-based bioluminescence detector, time-to-digital con- recording device (Fig. 4a–c). This design enabled remote monitoring
verter, voltage references, voltage regulators and four 1 mm × 1 mm of biomolecules in the gut for 4 h, with 70 fA relative photocurrent
CMOS-compatible photodiodes2 (Extended Data Fig. 9a). Besides the detected in the induced compartment, and a baseline of −50 fA from
custom chip, the miniaturized capsule includes a commercial micro- the uninduced control (Fig. 4c and Extended Data Fig. 10a–c). The
controller and wireless transmitter. Bioluminescence from activated receiver operating characteristic of the tetrathionate sensing showed
cells was detected by CMOS-integrated photodiodes located below a detection latency of 1 h and reached a sensitivity and specificity of
each chamber. Custom-designed electronics processed the lumi­ 100% at 120 min; thus, the miniaturized device can effectively detect
nescence data by periodically sampling the photon-generated charge the target analyte in the harsh intestinal environment (Fig. 4d and
(with a programmable integration time of approximately 26 s). The Extended Data Fig. 10d).
detected luminescence was converted to a digital code by the low-power
luminescence readout chip and transmitted wirelessly for calibration,
display and recording. Discussion
In vitro, with 1 µl of sensor bacteria culture per chamber, the inte- We have built a microbial biosensor that is compatible with ingestion
grated device detected the presence of tetrathionate. We recorded and can sense an array of biomarkers in situ as they are being produced.
125 fA of relative photocurrent produced by the induced tetrathionate This technology can potentially support remote disease management
bacteria sensor, with a baseline of around 0 fA from the uninduced by providing quantitative, real-time and multiplexed information link-
control (Fig. 3c). We next evaluated the performance of the other bacte- ing perturbations in the gastrointestinal tract microbiome to disease.
rial sensors in this optimized, integrated electronic pill. Although the To validate the cell-based biosensors in preclinical disease models,
thiosulfate sensor did not have a luminescence readout compatible we coupled engineered sensing bacteria with a recombinase-based
with the electronics (as expected from the plate reader-based char- memory system. Our memory system records information as soon as
acterization in Extended Data Fig. 8c), the NO and ROS sensors could metabolites are produced in the gut, activating switches within minutes

390 | Nature | Vol 620 | 10 August 2023

of exposure. The recombinase-based switch discretizes the magnitude acknowledgements, peer review information; details of author contri-
of a given biomarker; this quantitative response may align with disease butions and competing interests; and statements of data and code avail-
stage and therefore indicate the severity of inflammation. Although the ability are available at https://doi.org/10.1038/s41586-023-06369-x.
parameters measurable with this device may have no absolute ‘healthy’
range, measurements taken over time may reveal patterns predictive of 1. Inda, M. E. & Lu, T. K. Microbes as biosensors. Annu. Rev. Microbiol. 74, 337–359 (2020).
acute disease episodes (flares), making it possible to anticipate disease 2. Liu, Q. et al. A threshold-based bioluminescence detector with a CMOS-integrated
symptoms. Similarly, the switch acts as a peak detector for sensing and photodiode array in 65 nm for a multi-diagnostic ingestible capsule. IEEE J. Solid State
Circuits 58, 838–851 (2023).
recording maximum levels of intestinal biomarkers. 3. Bourgonje, A. R. et al. Oxidative stress and redox-modulating therapeutics in inflammatory
To create our miniaturized low-power electronic device, we inte- bowel disease. Trends Mol. Med. 26, 1034–1046 (2020).
grated CMOS-compatible photodiodes with discrete-time signal pro- 4. Lee, J.-Y., Tsolis, R. M. & Bäumler, A. J. The microbiome and gut homeostasis. Science 377,
eabp9960 (2022).
cessing circuits. The whole capsule, which is less than 1.4 cm3 in size, 5. Million, M. et al. Increased gut redox and depletion of anaerobic and methanogenic
can simultaneously detect multiple disease biomarkers on a stringent prokaryotes in severe acute malnutrition. Sci. Rep. 6, 26051 (2016).
power budget. With the integrated CMOS system and pill casing design, 6. Pribis, J. P. et al. Gamblers: an antibiotic-induced evolvable cell subpopulation differentiated
by reactive-oxygen-induced general stress response. Mol. Cell 74, 785–800.e7 (2019).
we could detect the bioluminescent signal from just 1 µl of bacterial 7. Reese, A. T. et al. Antibiotic-induced changes in the microbiota disrupt redox dynamics in
culture in the milieu of the intestinal lumen. Furthermore, a coin-cell the gut. eLife 7, e35987 (2018).
battery can power the miniaturized capsule for a month2, so this device 8. Rivera-Chávez, F. et al. Depletion of butyrate-producing clostridia from the gut microbiota
drives an aerobic luminal expansion of Salmonella. Cell Host Microbe 19, 443–454 (2016).
could also potentially be used as an implant44,45. 9. Dumitrescu, L. et al. Oxidative stress and the microbiota–gut–brain axis. Oxid. Med. Cell.
We also demonstrated that the use of solid, non-deformable devices Longev. 2018, 2406594 (2018).
and free-standing polymer films provides a means to protect both 10. Yardeni, T. et al. Host mitochondria influence gut microbiome diversity: A role for ROS.
Sci. Signal. 12, eaaw3159 (2019).
the viability and the function of the on-board biosensor bacteria 11. Jose, S., Bhalla, P. & Suraishkumar, G. K. Oxidative stress decreases the redox ratio and
through the harsh environment of the stomach. Beyond this initial folate content in the gut microbe, Enterococcus durans (MTCC 3031). Sci. Rep. 8, 12138
line of defence, the filter membrane also serves to contain the biosen- (2018).
12. Hagan, T. et al. Antibiotics-driven gut microbiome perturbation alters immunity to vaccines
sor cells while excluding host immune and microbial cells, preventing in humans. Cell 178, 1313–1328.e13 (2019).
unexpected off-target interactions and enhancing biocontainment 13. Million, M. & Raoult, D. Linking gut redox to human microbiome. Hum. Microbiome J. 10,
27–32 (2018).
while allowing free exchange of nutrients and other small molecules
14. Rivera-Chávez, F. & Bäumler, A. J. The pyromaniac inside you: Salmonella metabolism in
of interest for sensing46,47. Our device-based protection strategy is also the host gut. Annu. Rev. Microbiol. 69, 31–48 (2015).
orthogonal to and can be combined with direct cell encapsulation 15. Kalantar-Zadeh, K. et al. A human pilot trial of ingestible electronic capsules capable of
sensing different gases in the gut. Nat. Electron. 1, 79–87 (2018).
strategies based on secretion of capsular polysaccharides48, hydrogels44
16. Steiger, C. et al. Dynamic monitoring of systemic biomarkers with gastric sensors. Adv. Sci.
and synthetic coating of microbial surfaces (alginate49, chitosan49, 8, 2102861 (2021).
polydopamine50, lipids51 and nanoparticles52) that can provide down- 17. Archer, E. J., Robinson, A. B. & Süel, G. M. Engineered E. coli that detect and respond to
gut inflammation through nitric oxide sensing. ACS Synth. Biol. 1, 451–457 (2012).
stream protection against bile acids, antimicrobials and bacteriophages.
18. Rubens, J. R., Selvaggio, G. & Lu, T. K. Synthetic mixed-signal computation in living cells.
The diagnostic accuracy and specificity of our device are based on Nat. Commun. 7, 11658 (2016).
the simultaneous testing of multiple labile by-products of inflamma- 19. Daeffler, K. N. et al. Engineering bacterial thiosulfate and tetrathionate sensors for detecting
gut inflammation. Mol. Syst. Biol. 13, 923 (2017).
tion (such as NO and ROS), intestinal gases (such as H2S, measured as
20. Annese, V. et al. European evidence-based consensus: inflammatory bowel disease and
thiosulfate) and other important molecules for the microbiome (for malignancies. J. Crohns. Colitis 9, 945–965 (2015).
example, tetrathionate). As biomarker levels may vary greatly among 21. Amir, A. et al. Room-temperature, correcting for microbial blooms in fecal samples during
shipping. mSystems 2, e00199–16 (2017).
individuals, a panel of biomarkers would be required to accurately
22. Raymond, F. et al. Culture-enriched human gut microbiomes reveal core and accessory
diagnose IBD and other multi-faceted diseases. resistance genes. Microbiome 7, 56 (2019).
Our miniaturized device offers a route for non-invasively evaluat- 23. Zmora, N. et al. Personalized gut mucosal colonization resistance to empiric probiotics is
associated with unique host and microbiome features. Cell 174, 1388–1405.e21 (2018).
ing changes in the intestinal biochemical milieu and overcomes the 24. Jain, U. et al. Debaryomyces is enriched in Crohn’s disease intestinal tissue and impairs
limitations of microbiome characterization by 16S rRNA or metagen- healing in mice. Science 371, 1154–1159 (2021).
omic sequencing53,54, as well as those of existing ingestible biosen- 25. Courbet, A., Endy, D., Renard, E., Molina, F. & Bonnet, J. Detection of pathological
biomarkers in human clinical samples via amplifying genetic switches and logic gates.
sors used in animal models, which require the complex analysis of Sci. Transl. Med. 7, 289ra83 (2015).
microbial gene expression, RNA or DNA in stool19,26–29. The biosensors 26. Kotula, J. W. et al. Programmable bacteria detect and record an environmental signal in
described here also have the potential to expand the range of biomark- the mammalian gut. Proc. Natl Acad. Sci. USA 111, 4838–4843 (2014).
27. Lim, B., Zimmermann, M., Barry, N. A. & Goodman, A. L. Engineered regulatory systems
ers being targeted by other ingestible electronic systems55. The capsule modulate gene expression of human commensals in the gut. Cell 169, 547–558.e15
could be designed to report its location while in transit56 and perform (2017).
cell-based computation to further expand the multiplex capabilities 28. Mimee, M., Tucker, A. C., Voigt, C. A. & Lu, T. K. Programming a human commensal
bacterium, Bacteroides thetaiotaomicron, to sense and respond to stimuli in the murine
of the underlying electronic system. For example, AND gates57 could gut microbiota. Cell Syst. 1, 62–71 (2015).
be incorporated to determine the co-localization of biomarkers for 29. Riglar, D. T. et al. Engineered bacteria can function in the mammalian gut long-term as
understanding metabolic pathways or biomarker discovery in animal live diagnostics of inflammation. Nat. Biotechnol. 35, 653–658 (2017).
30. Pickard, J. M. et al. Rapid fucosylation of intestinal epithelium sustains host–commensal
models for many microbiome-linked diseases. symbiosis in sickness. Nature 514, 638–641 (2014).
We envision that with further clinical testing in humans, this device 31. Swain, P. Wireless capsule endoscopy. Gut 52 (suppl. 4), 48–50 (2003).
could be developed as a first-line at-home screen for non-invasive con- 32. van der Schaar, P. J. et al. A novel ingestible electronic drug delivery and monitoring
device. Gastrointest. Endosc. 78, 520–528 (2013).
tinuous monitoring of the chemical environment of the gastrointestinal 33. Din, M. O. et al. Interfacing gene circuits with microelectronics through engineered
tract and customized for numerous gastrointestinal disorders, with population dynamics. Sci. Adv. 6, eaaz8344 (2020).
the potential to be a safer and inexpensive point-of-care alternative to 34. Mimee, M. et al. An ingestible bacterial-electronic system to monitor gastrointestinal
health. Science 360, 915–918 (2018).
endoscopy. Tracking and quantitatively assessing multiple biomarkers 35. Bass, D. M., Prevo, M. & Waxman, D. S. Gastrointestinal safety of an extended-release,
would provide a potential framework for patients to assess the effects nondeformable, oral dosage form (OROS). Drug Saf. 25, 1021–1033 (2002).
of diet, lifestyle and other interventions to improve health outcomes. 36. Isabella, V. M. et al. Development of a synthetic live bacterial therapeutic for the human
metabolic disease phenylketonuria. Nat. Biotechnol. 36, 857–864 (2018).
37. Bush, M., Ghosh, T., Tucker, N., Zhang, X. & Dixon, R. Transcriptional regulation by the
dedicated nitric oxide sensor, NorR: a route towards NO detoxification. Biochem. Soc.
Online content Trans. 39, 289–293 (2011).
38. Chen, X. J., Wang, B., Thompson, I. P. & Huang, W. E. Rational design and characterization
Any methods, additional references, Nature Portfolio reporting sum- of nitric oxide biosensors in E. coli Nissle 1917 and mini SimCells. ACS Synth. Biol. 10,
maries, source data, extended data, supplementary information, 2566–2578 (2021).

Nature | Vol 620 | 10 August 2023 | 391

Article
39. Ceze, L., Nivala, J. & Strauss, K. Molecular digital data storage using DNA. Nat. Rev. Genet. 52. Hu, Q. et al. Engineering nanoparticle-coated bacteria as oral DNA vaccines for cancer
20, 456–466 (2019). immunotherapy. Nano Lett. 15, 2732–2739 (2015).
40. Beck, P. L. et al. Paradoxical roles of different nitric oxide synthase isoforms in colonic 53. Vujkovic-cvijin, I. et al. Host variables confound gut microbiota studies of human disease.
injury. Am. J. Physiol. 286, 137–147 (2004). Nature 587, 448–454 (2020).
41. Jiminez, J. A., Uwiera, T. C., Douglas Inglis, G. & Uwiera, R. R. E. Animal models to study 54. Chiu, C. Y. & Miller, S. A. Clinical metagenomics. Nat. Rev. Genet. 20, 341–355 (2019).
acute and chronic intestinal inflammation in mammals. Gut Pathog. 7, 29 (2015). 55. Steiger, C. et al. Ingestible electronics for diagnostics and therapy. Nat. Rev. Mater. 4,
42. Strand-Amundsen, R. J. et al. Ischemia/reperfusion injury in porcine intestine—viability 83–98 (2019).
assessment. World J. Gastroenterol. 24, 2009–2023 (2018). 56. Ma, Y., Selby, N. & Adib, F. Minding the billions: ultra-wideband localization for deployed
43. Lundberg, J. O. N., Lundberg, J. M., Alving, K. & Hellström, P. M. Greatly increased luminal RFID tags. In Proc. MobiCom ’17 https://doi.org/10.1145/3117811.3117833 (Association for
nitric oxide in ulcerative colitis. Lancet 344, 1673–1674 (1994). Computing Machinery, 2017).
44. Liu, X. et al. Magnetic living hydrogels for intestinal localization, retention, and diagnosis. 57. Siuti, P., Yazbek, J. & Lu, T. K. Synthetic circuits integrating logic and memory in living
Adv. Funct. Mater. 31, 2010918 (2021). cells. Nat. Biotechnol. 31, 448–452 (2013).
45. Song, M. et al. In 2020 IEEE International Solid- State Circuits Conference 474–476 58. Lloyd-Price, J., Abu-Ali, G. & Huttenhower, C. The healthy human microbiome. Genome
https://doi.org/10.1109/ISSCC19947.2020.9063083 (IEEE, 2020). Med. 8, 51 (2016).
46. Krawczyk, K. et al. Electrogenetic cellular insulin release for real-time glycemic control in 59. Inda, M. E., Mimee, M. & Lu, T. K. Cell-based biosensors for immunology, inflammation,
type 1 diabetic mice. Science. 368, 993–1001 (2020). and allergy. J. Allerg. Clin. Immunol. 144, 645–647 (2019).
47. Zhao, H. et al. Autonomous push button-controlled rapid insulin release from a
piezoelectrically activated subcutaneous cell implant. Sci. Adv. 8, 24 (2022). Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in
48. Harimoto, T. et al. A programmable encapsulation system improves delivery of therapeutic published maps and institutional affiliations.
bacteria in mice. Nat. Biotechnol. 40, 1259–1269 (2022).
49. Li, Z. et al. Biofilm-inspired encapsulation of probiotics for the treatment of complex
infections. Adv. Mater. 30, e1803925 (2018). Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this
50. Chen, W. et al. Bacteria-driven hypoxia targeting for combined biotherapy and photothermal article under a publishing agreement with the author(s) or other rightsholder(s); author
therapy. ACS Nano 12, 5995–6005 (2018). self-archiving of the accepted manuscript version of this article is solely governed by the
51. Cao, Z., Wang, X., Pang, Y., Cheng, S. & Liu, J. Biointerfacial self-assembly generates lipid terms of such publishing agreement and applicable law.
membrane coated bacteria for enhanced oral delivery and treatment. Nat. Commun. 10,
5783 (2019). © The Author(s), under exclusive licence to Springer Nature Limited 2023

392 | Nature | Vol 620 | 10 August 2023

Methods For in vitro kinetic studies, subcultured cells were mixed with an
inducer in a 96-well plate and immediately placed in the plate reader
Bacterial strains and culture conditions set at 37 °C without shaking. Luminescence and absorbance were read
Routine cloning and plasmid propagation were performed in E. coli at 3-min intervals.
E. cloni 10G (Lucigen). For all in vitro and in vivo experiments, the
probiotic strain E. coli Nissle 1917 was transformed with gene circuits Mouse experiments
built on plasmids and for the recombinase system, a bacterial artificial Approval for mouse experiments was obtained from the Committee
chromosome (BACs) was also used, as indicated in Supplementary on Animal Care at the Massachusetts Institute of Technology. Male
Tables 2 and 3. Cells were routinely cultured at 37 °C in Luria–Bertani C57BL/6J mice (8–10 weeks of age) were obtained from Jackson Labs
(LB) medium (Difco). Where appropriate, growth medium was sup- (stock no. 000664). Conventional conditions were used to house and
plemented with antibiotics at the following concentrations: 50 μg ml−1 handle the mice: 12/12 light cycle with 30–70% humidity and temp of
kanamycin, 100 μg ml−1 carbenicillin, 25 μg ml−1 chloramphenicol, 21 ± 1 °C; the mice were housed on ward wood shavings with reverse
100 μg ml−1 ampicillin and 100 μg ml−1 spectinomycin. osmosis drinking water. One week before the experiments began, the
mice were acclimatized to the animal facility.
Plasmid construction and circuit characterization Mice were randomly allocated to experimental groups. Researchers
All plasmids were constructed by combining PCR fragments gener- were not blinded to group assignments. Overnight cultures of
ated by Phusion High-Fidelity DNA Polymerase (NEB) using Gibson E. coli Nissle 1917 grown in Teknova Hi-Def Azure Media with appro-
assembly60, starting from DNA sources as referenced in Supplementary priate antibiotics and 0.2% glucose were centrifuged at 5,000g for
Table 1, from gBlocks manufactured by IDT or acquired from Addgene. 5 min and resuspended in an equal volume of 20% sucrose. Animals
Genetic parts and plasmids used in this study are listed in Supplemen- were inoculated with 200 μl of bacteria culture (approximately 108
tary Tables 1 and 3, and are available from Addgene. Assembly products colony-forming units (CFU)) by oral gavage. Faecal pellets were col-
were introduced by transformation into chemically competent E. coli lected 6 h post-gavage34, and homogenized in 1 ml PBS with a 5 mm
E. cloni 10G, and sequences were confirmed using Sanger sequencing. stainless steel bead using a TissueLyser II (Qiagen) at 25 Hz for 2 min.
To characterize the constructs built in conjunction with the memory Samples were centrifuged at 500g for 30 s to pellet large faecal debris.
system, appropriate antibiotics were added to Teknova Hi-Def Azure Supernatant was cultured in Teknova Hi-Def Azure Media with appro-
Media containing 0.2% glucose, and E. coli E. cloni 10G colonies were priate antibiotics and 0.2% glucose and incubated aerobically with
inoculated into this culture medium. Cultures were incubated aerobi- shaking for 16–18 h at 37 °C. Cells were then assayed on the flow cyto­
cally with shaking for 16–18 h at 37 °C, then diluted 2,500× into fresh meter. For Extended Data Fig. 4d, carbenicillin and chloramphenicol
Hi-Def Azure Media (also containing appropriate antibiotics and 0.2% were administered with the media for oral gavage. Where appropri-
glucose) and incubated aerobically with shaking for another 20 min ate, growth media was supplemented with antibiotics at the follow-
at 37 °C. Cultures (200 ml) were then transferred to a 96-well plate, ing concentrations: 50 μg ml−1 kanamycin, 100 μg ml−1 carbenicillin,
and the respective inducers, H2O2 (hydrogen peroxide Sigma–Aldrich 25 μg ml−1 chloramphenicol, 100 μg ml−1 ampicillin and 100 μg ml−1
H1009-100ML), DETA–NO (Sigma–Aldrich D185-50MG), potassium spectinomycin.
tetrathionate (Sigma–Aldrich P2926-100G) and sodium thiosulfate
(Sigma–Aldrich 217263-250G), were added at appropriate concentra- Pig experiments
tions via serial dilution. Plates were incubated either aerobically with Approval for pig experiments was obtained from the Committee on
shaking or anaerobically for 20 h at 37 °C for all experiments. For experi- Animal Care at the Massachusetts Institute of Technology. Female
ments performed in anaerobic conditions, cultures were grown and Yorkshire pigs (50–95 kg), received from Cummings Veterinary School
manipulated in a Coy anaerobic chamber with an atmosphere of 85% at Tufts University in Grafton, MA, were randomly selected for the
N2, 5% H2 and 10% CO2 at 37 °C. All media was pre-reduced overnight in experiments and housed under conventional conditions. Prior to the
anaerobic atmosphere before inoculation of cultures. After incubation, experiment, animals were given a clear liquid diet for 24 h. The day of
the optical densities of cultures were measured at 600 nm in a plate the experiment, the morning feed was withheld. Pigs were sedated with
reader. For flow cytometry, cells were diluted in cold 1× PBS and 2% Telazol (tiletamine plus zolazepam 5 mg kg−1), xylazine (2 mg kg−1) and
sucrose to reach an optical density of 0.02 (at 600 nm), then assayed atropine (0.04 mg kg−1) at the start of the experiment. The jejunum
on a BD LSRFortessa. A minimum of 10,000 gated events was recorded. was accessed via a midline laparotomy and the lumen sectioned into
BD FACSDiva 9.0 and FlowJo software v10 were used to export and several test compartments using Mayo–Robson intestinal clamps.
process flow cytometry software files and analyse data and statistics. Ischaemia-reperfusion injury was used as a model of intestinal inflam-
Forward scatter (FSC) and side scatter (SSC) were used to gate for cells mation and was caused by clamping the mesentery of the target intes-
excluding debris. The GFP emission was collected using the FITC detec- tinal segment with haemostatic clamps for 2 h and then releasing the
tor with a 530/30 filter. Side scatter pulse height (SSC-H) was used to clamps to allow reperfusion for at least 1 h. At the end of the proce-
threshold on cells. Doublet gating was done on a SSC-W vs SSC-H plot, dure, pigs were euthanized with Fatal Plus (sodium pentobarbital):
side scatter pulse width (SSC-W), as exemplified in Supplementary 115–120 mg kg−1 and heart rate was assessed to ensure the pig was eutha-
Fig. 1. Each experiment was performed with at least three biological nized. For bacteria-only experiments, overnight bacterial cultures
replicates. were diluted 1:10 in LB, cultured for 20 min, resuspended in 1 ml PBS
after centrifugation and were injected into the target intestinal section
Growth and induction via a syringe. Healthy intestinal sections were used as is or injected
For genetic circuit characterization, overnight cultures were diluted with 200 μl of the target analyte (DETA–NO, H2O2, tetrathionate or
1:100 in fresh LB and incubated with shaking at 37 °C for 2 h. Cultures thiosulfate) as described in the text. After the experiment, cells were
were removed from the incubator and 200 μl of culture was transferred retrieved by flushing the intestinal section with 10 ml of PBS injected
to a 96-well plate containing various concentrations of inducer. The and retrieved via a syringe. For device bacteria experiments, overnight
plate was returned to a shaking incubator at 37 °C. Following 2 h of bacterial cultures were diluted 1:10 in LB, cultured for 20 min and 1 μl of
incubation, luminescence was read using a BioTek Synergy H1 Hybrid fresh culture (concentrated 100× by centrifugation) was used to fill the
Reader with a 1 s integration time and a sensitivity of 150. Luminescence pill casing chambers and sealed as described in ‘Pill casing manufacture’.
values, measured in relative luminescence units (RLU), were normalized Devices were inserted into the intestinal lumen through a small incision,
by the optical density of the culture measured at 600 nm. manually passed into the target intestinal section and isolated from
Article
the incision site by luminal clamping as described above. Tetrathion-
ate (100 mM) was injected via a syringe into the clamped intestinal In vitro device viability and luminescence measurements after
compartment and data from the capsules was wirelessly acquired via simulated ingestion
a 915 MHz radio attached to a laptop (see Preparation of electronic Devices were loaded with 2 μl of bacteria as described above using a
components). Devices were manually retrieved from the jejunum. strain of E. coli Nissle 1917 carrying a plasmid with constitutive expres-
A total of three pigs were included in the experiments; three pigs on sion of the luciferase reporter operon (sMJD026). Enteric-film pro-
different days, two intestinal sections per animal, one for adminis- tected devices and unprotected devices were submerged in simulated
tering the inducer molecule and the other compartment as negative gastric fluid (USP, pH 1.2, no enzymes) and/or simulated intestinal fluid
control. (USP, pH 6.8, no enzymes) and incubated at 37 °C in a tube revolver
on oscillating mode (Thermo Scientific, 88881001) for the indicated
Preparation of electronic components time (Extended Data Fig. 7) to simulate ingestion and transit through
The electronics in the capsules consisted of four photodiodes the stomach. Following incubations, chambers were dried, and the
(Integrated CMOS P+/NWELL/PSUB photodiodes), a custom bio- luminescence signal was recorded (ChemiDoc, BioRad) and quanti-
luminescence detector chip fabricated in a CMOS 65 nm process2, fied using FIJI (ImageJ) as described in (Extended Data Fig. 7). Then the
a microcontroller (PIC12LF1840T39A, Microchip) and radio chip optically clear adhesive backing film was removed, 1-μl samples of each
(PIC12LF1840T39A, Microchip Technology Inc.), and a 915 MHz chip chamber were retrieved, serially diluted in PBS and spotted onto LB agar
antenna (0915AT43A0026, Johanson Technology). We used a com- plates with a pin replicator (V&P Scientific, VP 407A). Given a transfer
mercial receiver (CC1200, Texas Instruments). The upper side of the volume of ~1.5 μl, dilutions with countable colonies (between 1–20)
top PCB holds the fully quartz lid-packaged CMOS chip together were used to determine the CFU per ml in the initial chamber volume.
with an additional on-board LDO (ADP-166, Analog Devices). The
assembly was coated with 1 μm of Parylene C to act as a moisture Radio frequency experiments
barrier for the electronic components. Parylene C coating was per- In order to test wireless transmission in water, the device was immersed
formed using Specialty Coating System Labcoter 2 (PDS 2010) with in water in a 1 l beaker (deposited in the middle, diameter 10 cm) and
1 g of dichlorodi-p-xylene to reach a target layer thickness of 1 μm as we could measure a wireless power of −70 dBm, which is well above
described34. the sensitivity limit of −110 dBm as informed in the datasheet of the
receiver (CC1120, Texas Instruments). This confirms that the wireless
Pill casing manufacture signal can be transmitted at least 5 cm (half diameter) through water,
Pill casing top and bottom blanks were printed by selective laser sinter- with significant (40 dB) link margin left over.
ing in Grey resin on a Form 2 printer (Formlabs), post-processed accord- Additionally, when the same wireless transmitter (PIC12LF1840T39A,
ing to the manufacturer’s standard protocols and then flat outer faces Microchip) was placed fully inside the cavity of an anaesthetized pig,
were sanded to size. In the final design, Isopore membranes (0.4-μm a signal could be detected with the receiver (CC1120, Texas Instruments)
pore size, Millipore-Sigma) were cut to size with a punch and then 5 m from the animal. When the integrated testing device was placed in
attached to the casing body via a thin, laser cut double-sided adhesive the intestine inside of an anaesthetized pig, we could regularly measure
layer (3M VHB 5906). Unfilled pill casing tops were conditioned for 24 h a wireless power ranging between −90 dBm to −100 dBm (which is
in LB broth, and on the day of the experiment, the chambers were filled above the sensitivity limit) when the receiver was 1 m from the animal.
from the inside face with bacterial suspensions and subsequently sealed
with a thin, laser cut, optically clear adhesive backing film (GeneMate Calibration procedure for converting detector counts to
Polyolefin Films with Silicone Adhesive). Bacteria-filled casing tops estimated photocurrent
and empty bottoms were then pressed fit around the electronic system One-time optical calibration was performed to obtain the custom inte-
and the outer seam waterproofed with silicone (Elite Double 32, grated circuit (IC) performance and calibration parameters. During
Zhermack). Porous membrane types were initially screened as the optical calibration, the wireless capsule holding the custom lumi-
described in Extended Data Fig. 6a using a standard two-chamber Franz nescence detector IC, a standalone photodiode IC, and a green LED
cell. For experiments with enteric film-protected pill casings, the enteric (λ = 520 nm) were placed inside a metal box covered by a blackout cloth
films were solvent-cast from Eudragit L100-55 (Evonik Industries) to prevent ambient light from the environment. The capsule and the
plasticized with 50% w/w triethyl citrate and adhered to the outside of standalone photodiode IC were placed 1.5 cm adjacent to each other
the capsule with a double-sided adhesive layer (3M VHB 5906) before on one side of the metal box, while the LED was placed 30 cm opposite
loading the chambers with bacteria and without conditioning the both ICs on the other side of the metal box. We applied 5 different volt-
capsules in LB. age levels (0 V, 2.1 V, 2.14 V, 2.165 V, 2.185 V) across the LED to obtain
different optical power levels. The custom luminescence detector IC2
In vitro device measurements wirelessly transmitted sensor readout Ni for each sensing channel i at
LB culture medium was pre-warmed for at least 2 h prior to the start a given optical power. The standalone photodiode IC was exposed to
of experiments. For device-bacterial experiments, overnight cultures this optical power simultaneously, and the IC reported a photocurrent
were diluted 1:10 in LB, subcultured for 20 min, and concentrated level IPD through a sub-femtoamp SourceMeter (K6430, Keithley Instru-
100× by centrifugation. One microlitre of the concentrated culture ments). The resolution of a sensing channel i is defined as:
(~1010–1011 CFU ml−1) was then added to pill casing chambers. Wild-type
E. coli Nissle 1917 was added to the reference channel for all experi- Resolution = ∆IPD /∆Ni
ments. Once all four channels were loaded, the cell carrier was fastened
to the capsule and fully submerged in pre-warmed media. LB culture where ΔIPD is the difference between photocurrent values reported
media supplemented with inducer (100 mM tetrathionate). Cultures from the standalone photodiode IC for two LED bias voltages; and ΔNi
were wrapped several times in thick black fabric to block external light is the difference between luminescence detector IC output counts for
and placed in an incubator at 37 °C, and data was collected wirelessly for the same LED bias voltages. The resolution for a single channel was
2 h. At the end of the experiment, devices were disassembled and cell calculated using a linear regression model to fit the luminescence IC
carriers were discarded. Capsules were sterilized with 70% ethanol and output counts and photocurrents from standalone photodiode IC over
thoroughly washed with distilled water. Capsules were left to air-dry five LED bias voltages. The resolution of the luminescence detector
and re-used for future experiments. chip is 5.8–6.5 fA per count. The minimum detectable signal at one
LED bias voltage was calculated using the resolution of each sensing Acknowledgements This work was supported by Leona M. and Harry B. Helmsley Charitable
Trust (3239), Pew Charitable Trusts (to M.E.I.-W.; 00030623) and Catalyst Foundation (to R.T.Y.,
channel times one standard deviation of the IC output at this opti- Q.L. and M.E.I.-W.; Secure Bio-Engineered Sensors for Disease Management, SAP grant no.
cal power. The worst-case minimum detectable signal is 71 fA. The 55208844). M.J. was supported by the Translational Research Institute of Space Health through
estimated photocurrents for in vitro and in vivo measurements were Cooperative Agreement NNX16AO69A. G.T. was supported in part by the Department of
Mechanical Engineering, MIT and the Karl van Tassel (1925) Career Development Professorship,
calculated by taking the product of resolutions from one-time optical MIT. Part of this material is based on research sponsored by 711 Human Performance Wing
calibration and measured IC output counts in each measurement. We (HPW) and Defense Advanced Research Projects Agency (DARPA) under agreement number
used a moving average filter of 25 samples (~15 min moving average). FA8650-21-2-7120. The US Government is authorized to reproduce and distribute reprints for
governmental purposes notwithstanding any copyright notation thereon. The views and
conclusions contained herein are those of the authors and should not be interpreted as
Data analysis, statistics and computational methods necessarily representing the official policies or endorsements, either expressed or implied,
All data were analysed using GraphPad Prism version 9.1.2 (GraphPad of 711 Human Performance Wing (HPW) and Defense Advanced Research Projects Agency
(DARPA) or the US Government.
Software). Sequences were analysed using SnapGene version 5.1.2 (www.
snapgene.com). As noted, error bars represent the s.e.m. of at least three
Author contributions M.E.I.-W., M.J., Q.L., N.V.P., C.S., M.M., P.N., G.T., R.T.Y. and T.K.L. conceived
independent experiments carried out on different days. Significant dif- and designed the research. M.E.I.-W., M.J., Q.L., C.S., A.W., M.M., P.N., A.P.C., G.T., R.T.Y. and
ferences between groups was determined using an unpaired, two-tailed T.K.L. conceptualized the miniaturized pill form factor, including integration of the bacteria,
electronics and pill casings. M.E.I.-W. designed and performed in vitro biological experiments.
Student’s t-test assuming unequal variance and for curves over time,
Q.L., A.R. and T.Z. designed and built the integrated electronic circuits. M.E.I.-W. designed
two-way ANOVA for multiple comparisons. Fold change or SNR was and performed in vivo mouse experiments. M.J., J.A., A.W. and K.W. developed the pill casing
determined by dividing the normalized luminescence values (divided manufacturing process and validated the pill casing robustness in vitro, including membrane
attachment. M.J., A.W., K.W. and R.M. generated 3D-printed device components. M.J. developed
by optical density) of samples treated with the maximal inducer con- and validated the free-standing enteric films. M.E.I.-W., M.J., Q.L., N.V.P., C.S., A.H. and GT
centration with uninduced samples. The ROC was calculated based validated early prototypes. M.J., N.V.P., C.S., K.I., J.J., J.K., A.H. and G.T. conceptualized and
on multiple independent experiments (n = 5 for tetrathionate and validated the intestinal compartment animal model. K.I., N.F., J.J., J.K., A.H. and W.M. carried
out animal husbandry and anaesthesia of pigs. M.E.I.-W., M.J. and Q.L. experimentally tested
n = 3 for buffer compartments). Parts of the figures were drawn using the function of the integrated devices in vitro and in vivo. M.E.I.-W., M.J. and Q.L. performed
images (‘complete digestive apparatus’, ‘multi-well plate’, ‘pig’, ‘rodents’, formal analysis of the data. M.E.I.-W., M.J., Q.L., J.A., P.N. and R.T.Y. contributed to data analysis
‘Bacteriology_virology’ and ‘Blood_immunology’) from Servier Medical of in vitro and in vivo experiments of the integrated devices. M.E.I.-W., M.J., Q.L., N.V.P., J.A., A.W.,
P.N. and R.T.Y. contributed to visualization. M.E.I.-W., M.J., Q.L., M.M., P.N., A.P.C., G.T., R.T.Y. and
Art. Servier Medical Art by Servier is licensed under a Creative Com- T.K.L. wrote the manuscript. M.E.I.-W., M.J., Q.L., P.N., C.S., A.W., A.H., G.T. and R.T.Y. managed
mons Attribution 3.0 Unported License (https://creativecommons.org/ daily project progress and personnel. M.E.I.-W. and T.K.L. supervised and managed general
licenses/by/3.0/). We used Cadence IC 6.1.7 for circuit design, layout project administration. R.T.Y., G.T. supervised and managed funding of the project related to
the integrated electronics and integrated pill casings and swine husbandry, respectively.
design and simulations (https://www.cadence.com/en_US/home.html), M.E.I.-W., M.J., N.V.P., C.S., Y.L., M.M., P.N., A.P.C., R.T.Y., G.T. and T.K.L. contributed with
and Calibre 2016 for circuit verification (https://eda.sw.siemens.com/ funding acquisition.
en-US/ic/calibre-design/circuit-verification/).
Competing interests MIT and Boston University have filed a Patent Cooperation Treaty (PCT)
Reporting summary patent application (WO/2022/232188) regarding ingestible biosensors and methods of their use.
T.K.L. is a co-founder of Senti Biosciences, Synlogic, Engine Biosciences, Tango Therapeutics,
Further information on research design is available in the Nature Corvium, BiomX, Eligo Biosciences and Bota.Bio. T.K.L. also holds financial interests in nest.
Portfolio Reporting Summary linked to this article. bio, Ampliphi, IndieBio, MedicusTek, Quark Biosciences, Personal Genomics, Thryve, Lexent
Bio, MitoLab, Vulcan and Serotiny. A.P.C. is on the board of Analog Devices. M.J. consults for
VitaKey. C.S. is currently employed by Bayer AG (Germany). Complete details of all relationships
for profit and not for profit for G.T. can found at https://www.dropbox.com/sh/szi7vnr4a2ajb56/
Data availability AABs5N5i0q9AfT1IqIJAE-T5a?dl=0.
All data are provided in the paper and in the supplementary informa-
tion. Genetic sequences and plasmids have been deposited into the Additional information
Supplementary information The online version contains supplementary material available at
Addgene repository under Addgene identifiers 199782–199792. https://doi.org/10.1038/s41586-023-06369-x.
Correspondence and requests for materials should be addressed to G. Traverso, R. T. Yazicigil
60. Gibson, D. G. et al. Enzymatic assembly of DNA molecules up to several hundred kilobases. or T. K. Lu.
Nat. Methods 6, 343–345 (2009). Peer review information Nature thanks Jeff Hasty and the other, anonymous, reviewer(s) for
61. Reinders, C. I. et al. Rectal mucosal nitric oxide in differentiation of inflammatory bowel their contribution to the peer review of this work. Peer review reports are available.
disease and irritable bowel syndrome. Clin. Gastroenterol. Hepatol. 3, 777–783 (2005). Reprints and permissions information is available at http://www.nature.com/reprints.
Article

Extended Data Fig. 1 | Inflammatory bowel disease (IBD) is mediated by (TS) in the GI tract are mucin-derived cysteine and sulfate, which are metabolized
labile molecules that are not detectable with current technologies. to hydrogen sulfide (H2S)19. During ulceration, epithelial cells and red blood
Following an inflammatory insult, disproportionate mucosal immune responses cells enter the colon; these cells produce enzymes that convert H2S to TS. In the
via cytokine signaling lead to the release of redox-active molecules such as presence of ROS, TS is oxidized to tetrathionate (TT). Consumption of TT and
reactive oxygen species (ROS) and nitric oxide (NO). The resulting oxidative sulfate allows certain pathogens to establish a foot-hold for infection19,29, evoking
stress inhibits microbial growth in the gut lumen. However, chronic intestinal further immune responses. These mediators of disease are labile and cannot be
inflammation damages the epithelium and destroys the epithelial barrier, measured with existing technology. With only the limited information current
allowing intestinal microbes to invade the mucosa. The sources of thiosulfate approaches provide, breaking this positive feedback loop is challenging.
Extended Data Fig. 2 | See next page for caption.
Article
Extended Data Fig. 2 | Genetic circuit optimization and characterization did not affect growth of bacteria when ON vs. OFF states were compared. d, NO
of incorporated recombinase-based switch. a, Dose-response curves of detection in anaerobiosis. e, Time course of switch activation. The recombinase
NO-sensing genetic circuits in E. coli Nissle 1917. The translational initiation system triggered GFP expression within minutes (5 min for sensor NO Sensor 1
strength of the recombinase Bxb1 was varied by using different computationally and NO Sensor 2, 10 min for NO Sensor 3, and less than 1 min for the ROS sensor)
designed ribosome binding sites (RBS). Predicted RBS strengths are listed in of exposure to the target molecule. f, Correlation of number of cells, time and
the inset. Lower RBS strength led to a higher SNR. b, The memory circuit in the concentration to show the performance of the system, (n = 3 per group). Lines
three NO sensors was stable over multiple rounds of re-growth. Engineered represent the mean. Data are represented as mean ± SEM of three independent
bacteria collected in stool were cultured in a selective media to measure NO biological replicates derived from flow cytometry experiments (a–b, d–e), each
detection, and the memory system was validated to ascertain that it accurately of which involved n = 10,000 events.
reflected, over multiple rounds of culturing, the initial input. c, GFP expression
Extended Data Fig. 3 | See next page for caption.
Article
Extended Data Fig. 3 | Multiple specific disease biomarkers detected in of 0.1 mM, 1 mM and 10mM respectively and 1mM for the positive control with
vitro and in vivo. a–c, The bacterial sensors were validated for the ROS H2O2, DETA-NO). In (a and d) lines represent the mean, the errors (SEM) are derived
TS, and TT in vitro (a,d) and in vivo (b–c), following the protocol shown in Fig. 2. from flow cytometry experiments of three representative biological replicates,
In the presence of H2O2, the transcription factor OxyR is oxidized and activated each of which involved n = 10,000 events. In (b-c) individual points represent
in E. coli. To construct a ROS biosensor that detected H2O2, the recombinase independent biological replicates, n = 5 animals per group, and the bars (b),
gene bxb1 was placed under the control of the OxyR-regulated oxyS promoter, **p = 0.0091 (ROS, day 14), *p = 0.0243 (TT, day 10), ***p = 0.0006 (TS, day 6),
oxySp, on the same genetic circuit18 (panel a). To construct the TT and TS and lines (c) show the mean with SEM, **p = 0.0011 (TS, day 6), **p = 0.0029
sensors, we sought to overcome the oxygen repression that can affect Fumarate (ROS, day 10), *p = 0.0126 (ROS, day 14), **p = 0.0053 (TT, day 10), two-way
and Nitrate Reductase Regulator (FNR)-dependent sensors such as the ANOVA for multiple comparisons. e, The NO biosensor was evaluated for its use
previously reported two-component system TtrSR 29. Oxygen levels fluctuate as an NO disease stage detector. NorR, constitutively expressed from a library
in the gut, depending on the level of disruption of the mucosal epithelium. of ribosome-binding sites (RBS), exhibited different NO activation thresholds.
To avoid this cross-repression, we used two newly identified sensors to express Selected NO sensors (Sensor 1, 2, and 3) detected three concentrations (15, 30,
the recombinase system for detecting TS and TT: a TT sensor from Shewanella and 80uM), which could correspond respectively to mild, moderate, and severe
baltica, which does not depend on the FNR system, and the ThsRS sensor from states of inflammation61. Based on the recombinase system described in
Shewanella halifaxensis, the only genetically encoded TS sensor characterized Fig. 2a, flow cytometry was used to measure the percentage of GFP-positive
so far19. Both sensors distinguished their target molecules from other terminal cells at different concentrations of NO. For each point, the mean of three
electron acceptors in vitro19. d, The cross-reactivity of the NorR-engineered biological replicates, each with n = 10,000 flow cytometry events, is plotted.
bacteria was tested against ROS (H2O2), TT and TS in a series of dose-response Error bars are the standard error of the mean (SEM).
curves (series of two-fold dilutions of the inducer with an initial concentration
Extended Data Fig. 4 | See next page for caption.
Article
Extended Data Fig. 4 | In vivo validation of inflammatory biosensors. especially high for mouse #3. ‘DSS” samples, n = 5 and “Control” samples, n = 5.
a, Detection of NO by the NO biosensor as a marker of GI inflammation in vivo *p = 0.0408, **p = 0.0038, two-way ANOVA for multiple comparisons. e, Control
over time, n = 10 animals per group. Data are represented as mean ± SEM, for basal % of GFP+ over 4 days. After six rounds of re-growth of Sensor 1 and
**p = 0.0012 (day 13), ***p = 0.0005 (day 6), ***p = 0.0002 (day 11), ****p < 0.0001 Sensor 2, the background signal of non-induced cells does not continue
(day 9), two-way ANOVA for multiple comparisons. b, Independent validation expanding over time, which validates them for use in animal models. Engineered
of the presence of inflammation in the DSS colitis model by quantifying iNOS bacteria were cultured in a selective media to measure NO detection, and the
expression during DSS treatment, weight loss, and the lipocalin-2 (LCN-2) memory system was validated to ascertain that it accurately reflected, over
biomarker, n = 10 animals per group. Data are represented as mean ± SEM, multiple rounds of culturing (6 rounds, during 4 days) the initial input, n = 3 per
****p < 0.0001, two-tailed unpaired Student’s t test. c, Histological scores of group. Data are represented as mean ± SEM. f, Sensor validation in pigs.
inflammation and necrosis, indicating the validity of the DSS model. Other Experimental design: intestines were clamped to separate the different
indicators were observed but not quantified: bloody and loose stools, poor compartments (control vs. treated), and bacterial sensors were placed in
vigor, anal prolapse, and shortening of the colon upon dissection and gross the different compartments (left panel). All sensors registered significant
morphological examination. Lines represent the mean. Error bars represent activation in the presence of their respective inducers (300 uM H2O2, 30 mM
the SEM of independent biological replicates. d, Antibiotic-triggered redox TS, 3 mM TT, right panel). The bacteria were collected from the intestine after
imbalance measured by the NO sensor. NO Sensor 2 allowed us to detect an two hours of exposure to the analyte, and the percentage of GFP-positive cells
exacerbated inflammatory response after antibiotic treatment (carbenicillin was measured by flow cytometry. Lines represent the mean. The errors (SEM)
and chloramphenicol) in a chronic DSS inflammation model, which implies are derived from flow cytometry experiments of three representative
multiple rounds of DSS treatment. Our biosensor for NO shows an increase of biological replicates, each of which involved n = 10,000 events. Here, we show
NO expression after 4 and 20–30 days of antibiotic treatment in both healthy data of three independent experiments (three animals [M, U, T, K] on different
and DSS-treated mice, with a significant switch activation on day 9 in the days, multiple compartments per animal). ****p < 0.0001, two-way ANOVA for
DSS-treated mice and on day 29 in the chronic DSS inflammation model, multiple comparisons.
Extended Data Fig. 5 | Design and manufacture of a bacteria-electronic pill b, Pill casing manufacturing process. Casing blanks are 3D printed via
casing compatible with ingestion. a, Size comparison of ingestible electronics selective laser sintering (Formlabs) with supports only on the top and bottom
and solid dosage forms with established safety rates. The safety of ingestible face to preserve the thin wall features. The top and bottom faces are then
devices depends, in part, on ensuring that these devices will not damage, sanded to size. The filter membrane is cut to size with a punch and the double-
obstruct, or be retained in the GI tract. The current design was built to conform sided adhesive film is laser cut with through holes aligned to the chambers.
with the dimensions and form factors of solid dosage forms with known safety After pressing these outer layers together, the chambers are filled with
profiles and obstruction/retention rates (Procardia XL)55. Our system integration bacterial suspensions from the inside and sealed off with a thin clear adhesive
at the bacterial, electronics, and pill casing level allowed a significant reduction film to yield the fully sealed bacterial chamber/casing unibody. Scale bar = 5 mm.
in size compared to a previously reported prototype (>9 mL to <1.4 mL)34.
Article

Extended Data Fig. 6 | Validation of semipermeable membranes and of a pill casing protected from low pH ingress, compatible with ingestion. b,
coatings for sealing and protection of on-board bacterial sensors. a, Effect The assembled pill casing/chamber unibody can protect the sensor bacteria
of membrane on analyte diffusion. The porous membranes, when placed in against low pH during stomach transit by inclusion of a film made of an enteric
feces, did not interfere with detection of the target molecules. Fouling of the polymer (L100-55) attached via an adhesive layer (black). c, The enteric film
membrane by faecal matter may pose a problem, so we also screened several hardens after exposure to simulated gastric fluid (SGF), but dissolves away
membrane materials, e.g., polyethersulfone (PES), polycarbonate (PC), and after a brief exposure (< 1 hr) to neutral pH (PBS) allowing the bacteria to be
polyvinylidene fluoride (PVDF), to find which material allowed the highest exposed to the chemical environment of the small intestine. d, close-up view
diffusion of the analyte molecules across the membrane. Several porous (red, dashed square in panel b) of enteric protected pill casings showing no
membranes were tested in Franz Cells (inset on the right) in the presence membrane fouling after exposure to neutral pH (PBS). e, analysis of internal pH
of feces. All the membranes tested showed similar results, with a similar of fluid inside casing chambers by spotting contents onto pH paper. First two
percentage of detection from the NO bacterial sensors after the analyte rows are control spots of the indicated fluids. Last two rows are spots from
had passed through the membrane. The errors (SEM) are derived from flow three chambers each from a protected or un-protected pill casing exposed to
cytometry experiments of three representative biological replicates, each simulated gastric fluid (SGF) for 18 h. Bottom legend indicates corresponding
of which involved n = 10,000 events. b–e, Manufacturing and performance pH of the resulting color change.
Extended Data Fig. 7 | Enteric-protected pill casings preserve viability and of individual replicate chambers. N = 8 (4 chambers x 2 pill casings), Multiple
luminescence of on-board bacterial sensors through simulated ingestion. unpaired t-tests (two-tailed), *p = 0.0253 (unprotected), *p = 0.0195 (free cells),
a, Un-protected or enteric-protected pill casings were loaded with constitutively (ns) not significant. c, At end of exposure, the luminescence of the same chambers
luminescent bacterial cells and exposed to simulated intestinal fluid (SIF only, was also recorded (BioRad, ChemiDoc) from the inner side of the casings through
blue) or simulated ingestion with a 1-hour exposure to pH 1.2 simulated gastric the optically clear backing film. Total chamber luminescence was quantified
fluid (SGF+SIF) at 37 C. b, At end of exposure the internal chamber contents (FIJI, Image J) and divided by the total chamber colony forming units (CFU)
were extracted and viability was measured through serial dilution and colony from panel b. The limit of detection (LOD) was set at 2.5x the standard deviation
counting. As a control, an equivalent number of cells was directly resuspended of the background signal divided by the largest viability value observed. Values
in the exposure fluids (free cells) and pelleted between treatments. The viability below the LOD were set equal to the LOD and all values were normalized to set
of the common cell stock used to load the pill casings and for the free cell control the LOD = 1. Geometric mean and geometric 95% confidence intervals are
was kept refrigerated during the exposures and its viability was also measured. plotted on top of individual replicate chambers. N = 8 (4 chambers x 2 pill
Geometric mean and geometric 95% confidence intervals are plotted on top casings). d, Images used for luminescence quantification in panel c.
Article

Extended Data Fig. 8 | Kinetic response of the inflammatory biosensors values were measured thirty minutes to two hours post-exposure to the inducer
with the luciferase readout. a–b, E. coli Nissle biosensors were treated and normalized to the optical density of the culture. Lines represent the mean.
with their target analytes in LB (a) or in simulated intestinal fluid (b). The Error bars represent the SEM of three independent biological experiments.
luminescence response was measured in a plate reader every 3 min for 10 h. The d, Nitric oxide (NO) detection and luciferase expression in anaerobiosis,
signal-to-noise ratio (SNR) was calculated by dividing the OD600-normalized replicating the gut environment. Luminescence values were measured
luminescence values induced by the OD600-normalized luminescence values aerobically in a plate reader after overnight aerobic or anaerobic growth and
of uninduced samples. c, Response curve of the inflammatory biosensors with exposure to the inducer (DETA-NO 1mM) and normalized to the optical density
the luciferase readout. E. coli Nissle inflammatory sensor strains were treated of the culture. Lines represent the mean. Error bars denote the SEM for three
with various concentrations of their target analytes; maximal luminescence independent biological replicates.
Extended Data Fig. 9 | Electronic design and in vitro validation of the characterization of the device for miniaturized wireless sensing of NO and
integrated bacterial-electronic pill. a, Schematic of the miniaturized capsule ROS with cell-based biosensors. Wireless signal over time from the NO (b) and
PCB. The custom bioluminescence detector IC was fabricated in 65 nm CMOS ROS (c) sensors encapsulated in the device and immersed in bacterial growth
technology2. The top PCB holds the custom-designed multi-channel and media supplemented with 5 mM and 20 mM, respectively. Low-power CMOS-
time-multiplexed bioluminescence detector, a 6.8 mm x 2.1 mm coin-cell integrated photodiodes converted bioluminescence emitted from the
battery, two 220 µF decoupling capacitors, and an 8-position female connector. bacterial sensor into a photocurrent, which was converted into quantifiable
The bottom PCB holds a microcontroller with an integrated transmitter, digital data and transmitted wirelessly to the external device. Lines represent
a crystal oscillator, an 8-position male connector, an antenna and other the mean, and error bars denote the SEM for three independent replicates,
components for wireless data transmission. The components on the top conducted with one induced device (NO or ROS) and one uninduced device
and bottom PCB communicate through the two connectors. b–c, In vitro (Buffer); Rel. photocurrent, Relative Photocurrent.
Article

Extended Data Fig. 10 | In vivo validation of the integrated bacterial- c, Comparison of light detection between different chambers. E. coli Nissle
electronic pill. a,b, Individual replicates of TT sensing in the pig intestinal strains containing a functional biosensor circuit for TT detection (TT sensor),
environment. The devices with the TT sensor were deposited in the intestinal and E coli Nissle without the gene for luciferase (null sensor) were loaded into
compartments and TT (100 mM, blue) or buffer alone (black) were injected the device. Devices were deposited in the intestine compartments and after
after temperature stabilization (~15 min, 37 °C). Readings from the device temperature stabilization (~15 min), TT (100 mM, blue) or buffer alone (black)
were wirelessly collected for 120 min following device deposition. Dark trace were injected. Compartmentalized intestines were kept inside the abdomen, at
represents the mean of 3 replicates measurements (3 animals on different days, 37 °C, and wireless signals transmitted from inside the abdomen were collected
2 devices per pig, in two different compartments) and pale traces indicate for 120 min to analyze the kinetic response of the devices in the abdominal
the individual current values for a given device (two channels measured per cavity of the pig. Photocurrents provided relative to a one-time calibration
device). Photocurrents are provided relative to a one-time calibration value at value at t = 15 min. Non-induced sensor cells (black lines) decrease their
t = 15 min. Non-induced sensor cells (black lines) decrease their luminescence luminescence output throughout the experiment (as shown when tested in
output throughout the experiment (as shown in vitro, tested in simulated simulated intestinal fluid, Extended Data Fig. 8b), while induced cells express
intestinal fluid, Extended Data Fig. 8b), while induced cells express higher higher levels of luciferase compensating signal loss over time. The response of
levels of luciferase, compensating signal loss over time. For all the replicates, the device placed in the compartment with TT was clearly distinguishable from
the response of the device placed in the compartment with TT was clearly that of the device in the compartment with the buffer control. Null cells (light
distinguishable from that of the device in the compartment with the buffer blue and grey traces) maintain constant values throughout. Error bars denote
control. b, For clarity, individual photocurrent values corresponding to SEM for three experiments (3 animals on different days, 2 capsules per animal).
different time points (15, 30, 60 and 120 min) for n = 3 or n = 5 samples for d, Validation of the whole integrated device for miniaturized wireless
different set of conditions, including: TT sensor + TT, null sensor + TT, null biosensing in living pigs over time. The receiver operating characteristic (ROC)
sensor + buffer and TT sensor + buffer. Data are represented as mean ± SEM. of the device sensing TT reached a sensitivity and specificity of 100% at 120 min.