A wearable electrochemical biosensor for
the monitoring of metabolites and
nutrients
In the format provided by the
authors and unedited
Supplementary information
https://doi.org/10.1038/s41551-022-00916-z
C
ontent
s
Supplementary Note 1
|
Optimization and characterization of the LEG
-
MIP biosensors.
Supplementary Note 2
|
Iontophoresis
-
based localized sweat stimulation.
Supplementary Note 3
|
Refreshing time analysis and simulations.
Supplementary Note 4
|
Metabolic syndrome, BCAAs, and COVID
-
19.
Supplementary Fig. 1
|
Fabrication process of the multifunctional flexible wearable sensor patch.
Supplementary Fig. 2
|
Characterization of the LEG.
Supplementary Fig. 3
|
Integrated flexible Nut
riTrek prototype for personalized nutritional monitoring.
Supplementary Fig. 4
|
The detailed circuit diagram of the NutriTrek.
Supplementary Fig. 5
|
Fully integrated NutriTrek smartwatch for personalized nutritional monitoring.
Supplementary Fig. 6
|
Schema
tic of the preparation procedure and detection mechanism of the LEG
-
MIP
AA sensors.
Supplementary Fig. 7
|
Characterization and validation of the MIP sensor preparation.
Supplementary Fig. 8
|
Characterization of the electrochemical kinetics of the LEG
-
MIP e
lectrodes.
Supplementary Fig. 9
|
Theoretical and experimental optimization of the MIP composition.
Supplementary Fig. 10
|
Evaluation of the effect of incubation time and sample volume on the LEG
-
MIP
sensor performance.
Supplementary Fig. 11
|
Microscopic characterization of the RARs on the LEG.
Supplementary Fig. 12
|
Electrochemical characterization of the RARs on the LEG.
Supplementary Fig. 13
|
Characterizations of the LEG
-
MIP sensor regeneration.
Supplementary Fig. 14
|
Selectivity studies of
the LEG
-
MIP sensors for detecting two electroactive amino
acids: Trp and Tyr.
Supplementary Fig. 15
|
Schematic illustration of the sensitivity calculation of the LEG
-
RAR
-
MIP sensor.
Supplementary Fig. 16
|
LSV voltammograms of the LEG
-
PB
-
MIP sensors for ind
irect detection of all nine
essential amino acids.
Supplementary Fig. 17
|
LSV voltammograms of the LEG
-
PB
-
MIP sensors for indirect detection of multiple
vitamins, metabolites, and lipids.
Supplementary Fig. 18
|
LSV voltammograms and the corresponding calib
ration curves of the LEG
-
PB
-
MIP
sensor for indirect detection of cortisol
and mycophenolic acid
.
Supplementary Fig. 19
|
Characterization of the long
-
term storage stability of the LEG
-
MIP sensors.
Supplementary Fig. 20
|
Characterization of the reproducibili
ty of the LEG
-
MIP sensors from five different
batches.
Supplementary Fig. 21
|
Microscopic and electrochemical characterization of the surface of the LEG
-
MIP
electrodes.
Supplementary Fig. 22
|
Comparison of the performance of the MIP sensors based on differ
ent electrodes:
LEG, printed carbon electrode (PCE), Au electrode (AuE), and glassy carbon electrode (GCE).
Supplementary Fig. 23
|
Characterization of the performance of the LEG
-
AQCA
-
multi
-
MIP BCAA sensor.
Supplementary Fig. 24
|
Selectivity of the LEG
-
MIP
Trp sensors and the LEG
-
PB
-
MIP Leu sensors over
other major analytes in human sweat.
Supplementary Fig. 25
|
Evaluation of the LEG
-
MIP sensor for low concentration amino acid analysis.
Supplementary Fig. 26
|
GC
-
MS analyses of the Tyr, Try, Leu, Ile and Val
in standard analyte solution.
Supplementary Fig. 27
|
GC
-
MS analyses of human sweat and serum samples collected at the same time.
Supplementary Fig. 28
|
The performance of the LEG
-
MIP sensor under varied temperature and electrolyte
levels.
Supplementary Fig
. 29
|
In situ calibration strategies of the wearable LEG
-
MIP sensors involving a two
-
step
DPV
-
scan calibration and real
-
time temperature/electrolyte calibrations.
Supplementary Fig. 30
|
Investigation of the localized and surrounding sweat stimulation with
muscarinic
agents: carbachol and pilocarpine.
Supplementary Fig. 31
|
Effects of inlet numbers and inlet span of the microfluidics on target refreshing.
Supplementary Fig. 32
|
Effects of inlet
-
outlet orientations of the microfluidics on target refreshing.
S
upplementary Fig. 33
|
On
-
body evaluation of the microfluidic flexible sensor patches for carbagel
-
based
iontophoretic sweat stimulation and sampling at rest.
Supplementary Fig. 34
|
Characterization of the dynamics of the continuous microfluidic sensing.
Su
pplementary Fig. 35
|
Characterization of continuous microfluidic sensing performance under different
flow rates.
Supplementary Fig. 36
|
Continuous on
-
body Trp and Tyr analysis with real
-
time sensor calibrations using a
wearable sensor array on three subjec
ts during a constant
-
load cycling exercise.
Supplementary Fig. 37
|
Automated voltammogram analysis.
Supplementary Fig. 38
|
Dynamic monitoring of central fatigue using the Trp/BCAA sensor array patches.
Supplementary Fig. 39
|
Iontophoresis
-
based continuous on
-
body Trp and Tyr analysis using a wearable
sensor array with and without supplement intake (Subject 2).
Supplementary Fig. 40
|
Iontophoresis
-
based continuous on
-
body Trp and Tyr analysis using a wearable
sensor array with
and without supplement intake (Subject 3).
Supplementary Fig. 41
|
Iontophoresis
-
based continuous on
-
body Trp and Tyr analysis using a wearable
sensor array with and without supplement intake (Subject 4).
Supplementary Fig. 42
|
Tyr and Trp levels in cont
inuous on
-
body Trp and Tyr sensing using wearable
sensor arrays with and without supplement intake.
Supplementary Fig. 43
|
Sweat rate and the concentrations of amino acids (Trp here) and Na
+
on human
subjects.
Supplementary Table 1
|
Comparison of the
proposed approach with existing sensing mechanisms toward
wearable sweat sensing.
Supplementary Table 2
|
Sweat analytes and their concentrations used for the selectivity studies.
Supplementary Table 3
|
The sensitivities
(normalized by electrode area)
of t
he LEG
-
PB
-
MIP sensors for
small molecule quantification.
Supplementary Table 4
|
Numerically simulated time for average volumetric concentration to reach 90%
and 95% of the newly supplied concentration.
Supplementary Video 1
(separate f
ile)
|
NutriTrek: A wearable
bio
sensor
for real
-
time nutritional
monitoring.
Supplementary Video 2
(separate file)
|
Evaluation
of microfluidic sweat refreshing.
Supplementary Video 3
(separate file)
|
Iontophoresis
-
based microfluidic sweat sampling at rest.
Supplementary referenc
es
Supplementary
Note
1
|
O
ptimization and characterization of the LEG
-
M
IP biosensors
Characterization of the MIP sensor preparation.
The preparation of the
LEG
-
MI
P sensors was
characterized electrochemically with differential pulse voltammetry (DPV) in 0.1 M KC
l
solution containing
2.0 mM K
4
Fe(CN)
6
/K
3
Fe(CN)
6
(1:1) (
Supplementary Fig. 7a
). The LEG displayed a high oxidation peak
owing to its large electrochemically active surface area. The redox peak substantially decreased after the
MIP film deposition (co
-
polymerization of APBA and pyrrole in the presence of Trp her
e) due to the fact that
the less conducting polymer layer blocked the LEG from the redox reporter solution. The template molecules
were removed during template extraction step, leaving behind imprinted cavities that are complementary,
both chemically and s
terically to the template molecules. These cavities allow reporter ions to reach the
electrolyte/electrode interface, resulting in a rise of the redox peak current.
Raman spectrum was also used to study the surface roughness of the LEG
-
MIP sensor during p
reparation
process. Raman intensity is influenced by the scattering of the exciting light from the sample surface, and
thus decreases with the increase of surface roughness. As shown in
Supplementary Fig. 7b
, Raman
intensity of C=C backbone stretching incr
eased after polymerization (smooth surface), and then decreased
after template extraction (rough surface), indicating the residual cavities on the surface resulted from the
template extraction
1
. To further validate the successful preparation of the MIP layer, a
non
-
imprinted polymer
(NIP) film was prepared on the LEG as the control. The standard MIP template extraction procedure and
further incubation in 50 μM Trp did not lead to substantial
signal change of the LEG
-
NIP in the standard
redox solution (
Supplementary Fig. 7c
).
For the preparation of the
LEG
-
RAR
-
MI
P sensors, redox reporters such as Prussian blue (PB) and
anthraquinone
-
2
-
carboxylic acid (AQCA) were deposited between the MIP an
d graphene layers. For the PB
RAR, the preparation process was characterized electrochemically with linear sweep voltammetry (LSV) in
0.1 M KCl as illustrated in
Supplementary Fig. 7d
. The LEG
-
PB displayed a high reduction peak of PB
which decreased after
deposition of the polymer film (co
-
polymerization of APBA and pyrrole in the presence
of Leu here) due to the PB blockage by the polymer. The extraction of the template molecules (with CV
sweeping in 0.1 M HCl and 0.1 M KCl) leads to the target selective c
avities and increases the exposure of
the PB film to the electrolyte solution, resulting an increased redox signal.
The Raman spectra of the
LEG
-
PB
-
MI
P Leu showed similar behavior as the LEG
-
MIP Trp sensor: Raman
intensity of C=C backbone stretching incre
ased after polymerization on LEG
-
PB, and then decreased after
template extraction, the residual cavities were left on the surface resulted from the template extraction
(
Supplementary Fig. 7e
). To further validate the successful preparation of the MIP layer
on the LEG
-
PB, a
NIP film was prepared on the LEG
-
PB as the control. The standard MIP template extraction procedure and
further incubation in 50 μM Leu did not lead to substantial signal change of the LEG
-
PB
-
NIP
(
Supplementary Fig. 7f
).
Electrochemical
kinetics of the LEG
-
MIP electrodes.
The electrochemical kinetic process on the modified
electrode plays an important role in understanding whether the reaction process at the modified electrode is
controlled by adsorption and/or diffusion. Cyclic voltamme
try (CV) was used to study the effect of scan rate
on the peak current for both the direct and indirect detection LEG
-
MIP sensors (
Supplementary Fig. 8
).
Since the electroactive target (e.g., Trp) can be directly oxidized at a given voltage, the LEG
-
MIP Tr
p sensor
was evaluated in 0.01 M PBS containing 50
μ
M Trp (
Supplementary Fig. 8a,b
). A linear dependence was
obtained between the anodic peak current and scan rate, indicating that the oxidation of Trp on the direct
detection MIP sensor is controlled by ad
sorption processes. On the other side, the redox peak of the RAR
(e.g., PB) can be directly used to study the electrode kinetics in 0.1 M KCl (
Supplementary Fig. 8c,d
). In
this case, both anodic and cathodic peak currents showed proportional relationships
to square root of the
scan rate, suggesting that electrochemical redox reactions at LEG
-
PB
-
MIP Leu sensor were a diffusion
-
controlled process
2
. The relation between measured peak height current density
J
pa
(
μ
A mm
-
2
) and scan rate
v
(mV s
-
1
) for direct and indirect detection MIP sensors are as follows:
LEG
-
MIP Trp sensors:
J
pa
= 0.1718 + 0.006
v
LEG
-
PB MIP Leu sensors:
J
pa, anodic
=
-
8.338 + 3.5031
√
v
J
pa, cathodic
= 5.007
–
3.7458
√
v
The above results explain the reasons why the
current signal has a linear relationship with the concentration
of the target in direct detection, while it is log
-
linear with the target levels in indirect detection. To minimize
the influence of oxidation reactions of common sweat interferants, the reduc
tion peak of PB is chosen for
further analyzing of in direct detection.
Theoretical and experimental optimization of MIP composition.
MIPs can either rely on covalent or non
-
covalent interactions. In the case of a wearable sensor which should be capable of
regeneration for
continuous monitoring, weak reversible non
-
covalent interactions are ideal. There are multiple of monomers
which are capable of forming non
-
covalent bonds with amino acids (e.g., Tr
p
and Leu), however we
narrowed our search to electroacti
ve monomers since sensor fabrication with such monomers requires only
electropolymerization on the working electrode in the presence of the desired template molecules. In
addition, electroactive monomers efficiently transduce binding events, thus improving
detectability
3
.
Thus,
the formulations such as choices of monomers and monomer/template ratios have substantial influence on
the sensitivity and the selectivity of the MIP sensor.
Taking the Trp sensor design as an example, we utilized density functional theo
ry (DFT) calculations to
quantify the binding energy between Trp and six commonly used electroactive monomers:
aminophenylboronic acid (APBA), aniline, ethylenedioxythiophene (EDOT), phenylene, pyrrole, and
thiophene (
Supplementary Fig. 9a,b
). The calculat
ions were carried out using the ORCA software
4
. The
semiempirical Austin Model 1 (AM1) was used first to achieve a rough estimate of geometric optimal
configurations. The higher level B3LPY functional with a 6
-
31(d,p) basis set was then used to calculate final
geometric configurations and bindin
g energies. Binding energies were calculated with the typical formula:
∆
E= E
Monomer
-
Template
–
(E
Template
+E
Monomer
)
The DFT simulated bonding energies of the monomer
-
target complexes were demonstrated in
Supplementary Fig. 9b
. To maximize sensitivity of the MIP it is common to select the monomer which has
the highest binding affinity to Trp. Further, it has been previou
sly demonstrated that the co
-
polymerization of
a monomer with high affinity and a monomer with low affinity (crosslinker) to the template can produce highly
selective MIPs by mitigating non
-
selective binding
5
. APBA exhibits the highest interaction energy with the
Trp, indicating that APBA is an ideal crosslinker or co
-
monomer for Trp MIP. The choice of pyrrole (which
has lowest interaction energy) and APBA
as the monomer and crosslinker could leads to MIPs with both high
selectivity and high regeneration capability.
Our experimental data demonstrates that the choice of APBA/aniline also leads to high sensitivity (reflected
by the current peak height of the
LEG
-
MIP sensor in 50
μ
M Trp) compared to other individual monomers and
other monomer/crosslinker combinations (
Supplementary Fig. 9c
). The ratio of template, crosslinker, and
monomer is another key parameter MIP quality. Based on the experimental data illu
strated in
Supplementary Fig. 9d
, the ratio of 1:2.5:7.5 (template/crosslinker/monomer) led to the optimal sensitivity
for Trp detection.
Optimization of the LEG
-
MIP recognition
in vitro.
To obtain the optimal sensor performance for rapid
sample analysis,
the influences of sample incubation time and volume were evaluated experimentally. As
demonstrated in
Supplementary Fig. 10a,b
, the current density of the peak height of the LEG
-
MIP Trp
sensors increases rapidly with the increase of incubation time initia
lly, and then gradually stabilizes after 5
min (with an optimal incubation time of 7 min), indicating the saturated adsorption for Trp. Unlike the
incubation time, sample volumes (between 0.028
–
1.1
μ
L mm
-
2
) didn’t show substantial influence on the
sensor r
esponse as illustrated in
Supplementary Fig. 10c,d
.
Characterization of the RARs for indirect MIP detection.
The microstructure and element composition of
the LEG
-
RAR
s
were characterized in
Supplementary Fig. 11
. The LEG displayed an ultra
-
thin 3D flakes
with few
-
layer features while the C and N elements obtained by laser pyrolysis of PI were evenly distributed
in the flakes (
Supplementary Fig. 11a
). PB RAR nanoparticles with a diameter of about 100 nm were
successfully immobilized on the graphene surface
as illustrated in
Supplementary Fig. 11b,e
. The PB RAR
maintained its microstructure after 60 cycles of electrochemical cycling (
Supplementary Fig. 11c
). The
successful modification and electrochemical stability of AQCA were also confirmed by FTIR and
element
mapping (
Supplementary Fig. 11d,f
).
The high demand for electrochemical stability of wearable sensors in practical applications poses high
requirements for RARs. Therefore, the performance of four RARs including PB, AQCA, MB, and thionine
before a
nd after repetitive LSV or DPV scans was investigated (
Supplementary Fig. 12
). As summarized in
Supplementary Fig. 12e
, PB and AQCA displayed best stability among these four.
In situ
regeneration of the LEG
-
MIP sensors.
Sensor
regeneration is the major hu
rdle that limits the
applicability of current bioaffinity sensing technologies for wearable continuous monitoring. We have
investigated the possibility of using various controlled voltammetric techniques to realize the repeatable,
wash
-
free,
in situ
regene
ration of the LEG
-
MIP sensors.
For direct detection of the electroactive targets (e.g., Trp) using LEG
-
MIP sensors, as the molecule is
specifically fixed in the receptor site, the molecule can be oxidized once a proper redox voltage is applied to
the ele
ctrode; the oxidation product doesn’t have good binding affinity to the MIP receptor site, leading to
sensor regeneration (
Supplementary Fig. 13a
). After the detection DPV scan, applying a fixed redox
potential of 0.7 V for 12 s can reset the sensor respon
se to the background current with a ~100% recovery
ratio (
Supplementary Fig. 13b
). Considering that high temporal resolution (regeneration speed) and high
accuracy (recovery ratio) are both critical for continuous wearable sensing, we evaluated various
ele
ctrochemical voltammetric methods such as DPV, LSV, IT and CV for the LEG
-
MIP sensor regeneration
(
Supplementary Fig. 13c,d
): the results show that IT offers the best performance in recovery rate.
The regeneration for the LEG
-
RAR
-
MIP sensors for detection
of general targets, particularly non
-
electroactive targets such as BCAAs, is based on a different mechanism. After the extraction of the AA
template from the MIP layer, the complementary cavity of corresponding AA exists within a polymer
backbone or scaff
old
6
. The negatively charged targe
t AA will be specifically adsorbed into the shape
imprinted cavity. A negative electrical signal applied on the working electrode will repel the negatively
charged AA from the receptor site (
Supplementary Fig. 13e
). The LSV curve shows that the sensor can
be
restored to the same level as background signal under an applied negative voltage (
-
0.2 V) (
Supplementary
Fig. 13f
). The regeneration method and the regenerate time was optimized by using various electrochemical
voltammetric methods such as DPV, LSV, I
T and CV. The results in
Supplementary Fig. 13g,h
show that IT
method
under the potential of
-
0.2V for 50 s is optimal for regeneration of the LEG
-
PB
-
MIP sensor.
Evaluation of the batch
-
to
-
batch sensor reliability and reproducibility.
Traditional MIP was s
ynthesized
commonly using less controllable thermal or optical approaches in a bulk solution, which could have the
limitations such as the production of irregular particles
7
–
9
. In this work, the MIP layer
is electrochemically
deposited on the surface of our mass
-
producible, highly consistent laser
-
engraved graphene (LEG)
electrodes via cyclic voltammetry. The data in
Supplementary Fig. 20a
show that the electrochemical
behaviors of our LEG electrodes, LEG
-
M
IP electrodes before template (Trp) extraction, and the LEG
-
MIP
electrodes after template (Trp) extraction show very small variation, indicating the high reproducibility and
reliability of our MIP preparation process. To further verify the reliability of t
he LEG
-
MIP sensor, 5 different
sensor batches (3 sensors per batch) were prepared to evaluate the reproducibility in sensitivity and
selectivity for both electroactive (Trp) and non
-
electroactive (Leu) targets. As illustrated in
Supplementary
Fig.
20b,c
, the Trp and Leu LEG
-
MIP sensors from all batches exhibited very similar sensitivities with
relative standard deviations of 5.16% and 2.14%, respectively (n=15) in the presence of the same
concentration of target analytes. The Trp and Leu sensors obt
ained from all batches also showed high
selectivity for the target over other amino acids with similar structures (as shown in
Supplementary Fig.
20d,e
).
The characterization of the multi
-
MIP BCAA sensors.
To ensure the multi
-
MIP BCAA sensors have same
res
ponse to Leu, Ile, and Val, 5 mM of each target molecule was used to prepare the MIP film. Negative
DPV from 0
–
-
0.8 V was performed to characterize the LEG
-
AQCA
-
multi
-
MIP BCAA sensor in BCAA
solutions (1:1:1), and a log
-
linear relationship between the peak
height decrease and BCAA with sensitivity
of 939.2 nA
mm
-
2
per decade of concentration was observed (
Supplementary Fig. 23a,b
). The fabricated
indirect LEG
-
AQCA
-
multi
-
MIP BCAA sensor can be regenerated
in situ
upon constant potential applied to
the workin
g electrode in both PBS buffer (
Supplementary Fig. 23c
) and raw sweat (
Supplementary Fig.
23d
).
Supplementary
Note
2
|
Iontophoresis
-
based localized sweat stimulation
Iontophoresis is a common procedure that enables on
-
demand sweat induction by transderm
al delivery of a
muscarinic agonist that stimulates sweat gland to produce sweat. Despite its widespread use in cystic
fibrosis diagnosis, the choice of agonists is still mostly limited to pilocarpine and acetylcholine, which only
affect the sweat glands w
here the agonist is delivered. Here we use carbachol, a muscarinic agonist that has
nicotinic effects, which enable the sudomotor axon reflex sweating (SAR) and sweat glands neighboring the
dosed area also produce sweat for sampling (
Fig. 3f
)
10
. Carbachol is a cholinomimetic ester more r
esistant
to acetylcholinesterase hydrolysis than acetylcholine and enables a prolonged sweat production time. Using
a commercial iontophoresis device, we compared the sweat rate stimulated by commercial pilogels loaded
with pilocarpine and custom made car
bagels (
Supplementary Fig. 30
), both with the same geometry and
the same dosing area (a circle with a 27 mm diameter). Using the same commercial sweat collectors and the
same sampling area (a concentric circle with a 28.4 mm diameter), the total sweat rate
s (of the dosed area
and the surrounding area) of three subjects induced by carbagels is much higher and lasts longer than those
induced by commercial pilogels (
Fig. 3g
). Moreover, with the same dosing area (a 11 mm diameter circle)
blocked (by a 13 mm dia
meter adhesive disk) and the same commercial sweat collector (a concentric circle
with a 28.4 mm diameter), the carbagels elicited significant SAR sweat rates in 3 subjects compared to none
by pilogels (
Fig. 3h
). To avoid the potential contamination from g
el, we harvest only the SAR sweat and the
high sweat rate obtained is sufficient for continuous chemical sensing (
Supplementary Fig. 33
).
Supplementary
Note 3
|
Refreshing time analysis and simulations
The refreshing time analyses were performed
using numerical simulations (COMSOL).
Three
-
dimensional
models of different microfluidic designs with same dimensions of the actual device were created in
Rhinoceros and imported into COMSOL Multiphysics. The mass transport process was simulated by
numeric
ally solving the Stokes equation for an incompressible flow coupled with convection
-
diffusion
equation.
∇
p
=
μ
∇
2
v
∂
C
∂
t
=
D
∇
2
C
-
v