In Vivo Glucose Measurement by
Surface-Enhanced Raman Spectroscopy
Douglas A. Stuart,† Jonathan M. Yuen,‡ Nilam Shah,† Olga Lyandres,‡ Chanda R. Yonzon,†
Matthew R. Glucksberg,‡ Joseph T. Walsh,‡ and Richard P. Van Duyne*,†
Departments of Chemistry and Biomedical Engineering, Northwestern University, 2145 Sheridan Road,
Evanston, Illinois 60208
This paper presents the first in vivo application of surface-
enhanced Raman scattering (SERS). SERS was used to
obtain quantitative in vivo glucose measurements from an
animal model. Silver film over nanosphere surfaces were
functionalized with a two-component self-assembled mono-
layer, and subcutaneously implanted in a Sprague-
Dawley rat such that the glucose concentration of the
interstitial fluid could be measured by spectroscopically
addressing the sensor through an optical window. The
sensor had relatively low error (RMSEC ) 7.46 mg/dL
(0.41 mM) and RMSEP ) 53.42 mg/dL (2.97 mM).
Since the discovery of surface-enhanced Raman scattering in
the 1970s, the number of publications on surface-enhanced Raman
spectroscopy (SERS) has grown almost exponentially. Whereas
the number of applications employing normal Raman scattering
have dramatically increased, those using SERS have languished.
SERS has been largely confined to fundamental or rarified
academic research or denigrated as a “solution in search of a
problem”. For example, while it would seem that SERS would be
ideally suited to the in situ study of catalytic reactions, where
detailed vibrational information about transitional species would
be highly advantageous, SERS has failed to materialize as a wide-
spread analytical technique. Similarly, it has been difficult to
effectively use SERS in bioanalysis. While indirect application of
SERS, i.e., as a detection modality for molecular probes or labels,
has been successful, attempts to use SERS to detect native
biomolecules in situ has been limited.1-3 Detection of nonresonant
molecules of biological relevance, such as glucose, is challenging.
There are pervasive hurdles to the more extensive use of SERS
detection: (1) the historic view that both SERS signals and SERS-
active surfaces exhibit poor reproducibility, (2) not all molecules
are highly SERS active, (3) analytes must be close to (�1-2 nm)
or adsorbed on a roughened metal surface, and (4) the complexity
and structural similarity of many important molecules (e.g.,
proteins) yield SERS spectra that are difficult to interpret.4-6
The challenge is even greater when attempting to obtain
relevant SERS data in living systems. A nontrivial problem is
placement of the SERS-active surface in vivo without damage to
either the host or the surface. For example, colloids are difficult
to introduce into cells and can aggregate in extracellular space.
Solid substrates are more robust but require surgical implantation.
Additional problems can manifest after a surface is surgically
implanted and immersed in a biological milieu. There is currently
no control over what species adsorb, perhaps irreversibly, to the
SERS-active surface. This condition creates undesired spectral
noise from nontarget molecules, while simultaneously blocking
the access of the desired species, thereby lowering the analyte’s
overall signal. The cellular and in vivo environments are awash
with a multitude of unknown interferants whose presence and
concentration are in a constant state of flux. Indeed, the concen-
tration of the target molecule itself is invariably changing. In an
in vivo animal model, the problems are compounded by the host’s
immune response, the clotting factors, and the low concentration
of target species in the extracellular matrix. These factors can
contribute to surface contamination and cause unwanted effects.
We have developed a technique that addresses the critical
problems previously limiting the use of SERS to glucose measure-
ments, particularly in vivo. The fundamental enabling advances
are the development of stable and strongly enhancing SERS-active
surfaces and the chemical functionalization of those surfaces with
self-assembled monolayers (SAMs).7-10 The SAMs perform mul-
tiple roles in the detection system: limit fouling,11,12 provide a
convenient internal standard,13 exclude entire classes of interfer-
ants,14 and improve the glucose signal.7-9,15-17
* To whom correspondence should be addressed. E-mail: vanduyne@
chem.northwestern.edu.
† Department of Chemistry.
‡ Department of Biomedical Engineering.
(1) Ni, J.; Lipert, R. J.; Dawson, G. B.; Porter, M. D. Anal. Chem. 1999, 71,
4903-4908.
(2) Cao, Y. C.; Jin, R. C.; Nam, J. M.; Thaxton, C. S.; Mirkin, C. A. J. Am. Chem.
Soc. 2003, 125, 14676-14677.
(3) Doering, W. E.; Nie, S. Anal. Chem. 2003, 75, 6171-6176.
(4) Kneipp, K.; Kneipp, H.; Itzkan, I.; Dasari, R. R.; Feld, M. S. J. Phys.: Condens.
Matter 2002, 14, R597-R624.
(5) Nabiev, I.; Chourpa, I.; Manfait, M. J. Raman Spectrosc. 1994, 25, 13-23.
(6) Cotton, T. M.; Kim, J. H.; Chumanov, G. D. J. Raman Spectrosc. 1991, 22,
729-742.
(7) Yonzon, C. R.; Haynes, C. L.; Zhang, X. Y.; Walsh, J. T.; Van Duyne, R. P.
Anal. Chem. 2004, 76, 78-85.
(8) Stuart, D. A.; Yonzon, C. R.; Zhang, X. Y.; Lyandres, O.; Shah, N. C.;
Glucksberg, M. R.; Walsh, J. T.; Van Duyne, R. P. Anal. Chem. 2005, 77,
4013-4019.
(9) Lyandres, O.; Shah, N. C.; Yonzon, C. R.; Walsh, J. T.; Glucksberg, M. R.;
Van Duyne, R. P. Anal. Chem. 2005, 77, 6134-6139.
(10) Stuart, D. A.; Yonzon, C. R.; Zhang, X.; Lyandres, O.; Shah, N.; Glucksberg,
M. R.; Walsh, J. T.; Van Duyne, R. P. Anal. Chem. In press.
(11) Ostuni, E.; Chapman, R. G.; Liang, M. N.; Meluleni, G.; Pier, G.; Ingber, D.
E.; Whitesides, G. M. Langmuir 2001, 17, 6336-6343.
(12) Love, J. C.; Estroff, L. A.; Kriebel, J. K.; Nuzzo, R. G.; Whitesides, G. M.
Chem. Rev. 2005, 105, 1103-1169.
(13) Loren, A.; Engelbrektsson, J.; Eliasson, C.; Josefson, M.; Abrahamsson, J.;
Abrahamsson, K. Nano Lett. 2004, 4, 309-312.
Anal. Chem. 2006, 78, 7211-7215
10.1021/ac061238u CCC: $33.50 © 2006 American Chemical Society Analytical Chemistry, Vol. 78, No. 20, October 15, 2006 7211
Published on Web 09/12/2006
Stability of the SERS signal and surface stability are primarily
determined by the material properties of the enhancing surface.17-20
We have theoretically predicted and experimentally verified the
parameters required to optimize the plasmonic properties of the
film over nanospheres (FON)-type surfaces. Although the FON
variant of nanosphere lithography (NSL) is intrinsically less
enhancing (EF ) 106) than other NSL varieties (108), FONs
provide higher overall SERS signals.20 This is because the total
signal is related to both the SERS EF and the number of analyte
molecules probed, which is quite high for FONs because of their
relatively large viable surface area. The high radius of curvature
imparted by the underlying nanospheres prevents annealing or
loss of the nanoscale roughness features that give rise to SERS.
The use of SAM-functionalized surfaces allows us to surmount
the other hurdles.9,21
The SAM improves the signal from the analyte by partitioning
glucose and localizing it within the first few nanometers of the
SERS-active Ag or Au surface. This is critical, since SERS has a
short-range distance dependence and can thus only detect analytes
within this narrow zone.22,23 Previously, we demonstrated the
efficacy of an ö-alkanethiol (i.e., decanethiol (DT)) and a glyco-
lated alkanethiol (1-mercaptoundeca-11-yl triethylene glycol (EG3)),
as a partitioning SAM in the SERS measurement of glucose.8,15,16
An important improvement is the development of a new mixed
SAM based on two commercially available components, DT and
mercaptohexanol (MH). The DT/MH SAM was designed to have
dual hydrophobic/hydrophilic functionality, analogous to the
ethylene glycol-terminated SAMs used in our previous work. In a
recent publication, we conclusively demonstrate that the DT/MH
mixed SAM performs better as a partitioning layer than the DT
and EG3 SAMs used previously.9 We conjecture, based on space-
filling models, that the shorter hydroxyl-terminated chains form
hydrophilic pockets, thus partitioning glucose closer to the SERS-
active surface than the previously used SAMs. The SAM also
excludes nontarget molecules, such as proteins, that could give
rise to spectral congestion. This facilitates detection by simplifying
the composition of the solution at the surface, while the SERS
spectra provide a vibrational “fingerprint” that is unique to each
molecule. It is this best-to-date partition layer that was used in
the in vivo experiments.
EXPERIMENTAL SECTION
Materials. All the chemicals were reagent grade or better and
used as purchased. Silver pellets (99.99%) were purchased from
Kurt J. Lesker Co. (Clairton, PA). Oxygen-free, high-conductivity
copper was obtained from McMaster-Carr (Chicago, IL) and cut
into 18-mm-diameter disks. An 18-mm-diameter Cu mesh was also
used effectively as a substrate. To clean substrates, NH4OH, H2O2,
and ethanol from Fisher Scientific (Fairlawn, VA) were used.
Surfactant-free, white carboxyl-substituted latex polystyrene nano-
sphere suspensions (390 ( 19.5-nm diameter, 4% solid) were
purchased from Duke Scientific Corp. (Palo Alto, CA). Ultrapure
water (18.2 M¿ cm-1) from a Millipore system (Marlborough,
MA) was used for surface, substrate, and solution preparation.
Glucose was purchased from Sigma (St. Louis, MO). Decanethiol
(CH3(CH2)9SH), and 6-mercapto-1-hexanol (HS(CH2)6OH) were
purchased from Aldrich (Milwaukee, WI).
AgFON Fabrication and Incubation Procedure. The SAM-
functionalized SERS-active surfaces were prepared in four steps,
as shown in Figure 1B. First, silver film over nanosphere (AgFON)
surfaces were fabricated by drop-coating 10 íL of 390-nm-diameter
nanosphere solution onto clean copper substrates and then
depositing a 200-nm-thick Ag film onto the nanosphere mask. This
thickness provided the optimal LSPR location (Figure 1D) for
SERS enhancement with 785-nm excitation.20 The AgFON surfaces
were then incubated in 1 mM DT for 45 min and subsequently
transferred to 1 mM MH solution for at least 12 h to form a mixed
DT/MH SAM. This produces a mixed monolayer in which DT
predominates. The FONs were kept in 1 mM MH in ethanol prior
to surgically implantation.
LSPR Reflectance Spectroscopy. Measurements were car-
ried out using a SD2000 spectrometer coupled to a reflection probe
(Ocean Optics, Dunedin, FL) and a halogen lamp (F-OLiteH,
World Precision Instruments, Sarasota, FL). The reflection probe
consists of a tight bundle of 13 optical fibers (12 illumination fibers
around a collection fiber) with a usable wavelength range of 400-
900 nm. All reflectance spectra were collected against a mirror-
like Ag film over glass surface as a reference.
Surgical Implantation. The surgical procedure followed the
protocol filed with Northwestern University and IACUC. Sprague-
Dawley rats (300-500 g, N ) 4) were anesthetized with pento-
barbital (Office of Research Safety, Northwestern University) with
an initial dose of 50 mg/kg. The animals were checked for pain
reactions by toe tug and blink tests. The rats were kept under
anesthetic by hourly administration of pentobarbital at 25 mg/
kg. After the anesthetic had taken effect, the surgical areas were
prepared by removal of hair (shaving and chemical depilatory)
and cleaning. Then, the femoral vein was cannulated using PE 50
tubing (Clay Adams) for glucose injections. The carotid artery
was cannulated with PE 90 tubing for blood glucose measure-
ments with FDA-qualified home medical equipment (One Touch
II Meter, Lifescan, Inc.). A tracheotomy was performed to enable
the attachment of a ventilator to aid respiration. The incisions were
shut with surgical clips. The rat was thermally stabilized by an
electric heating pad throughout the course of the surgery and
experiment. A metal frame containing a glass window was placed
along the midline of the rat’s back. A circular incision was made
to allow the positioning of a DT/MH-functionalized AgFON sensor
subcutaneously such that the substrate was in contact with the
interstitial fluid and optically addressable through the window.
The rodent was then gently positioned into a heated holder in
the conventional sample position on a lab-scale Raman spec-
(14) Mosier-Boss, P. A.; Lieberman, S. H. Appl. Spectrosc. 2003, 57, 1129-1137.
(15) Shafer-Peltier, K. E.; Haynes, C. L.; Glucksberg, M. R.; Van Duyne, R. P. J.
Am. Chem. Soc. 2003, 125, 588-593.
(16) Yonzon, C. R.; Stuart, D. A.; Zhang, X. Y.; McFarland, A. D.; Haynes, C. L.;
Van Duyne, R. P. Talanta 2005, 67, 438-448.
(17) Dieringer, J. A.; McFarland, A. D.; Shah, N. C.; Stuart, D. A.; Whitney, A.
V.; Yonzon, C. R.; Young, M. A.; Zhang, X. Y.; Van Duyne, R. P. Faraday
Discuss. 2006, 132, 9-26.
(18) Stuart, D. A.; Biggs, K. B.; Van Duyne, R. P. Analyst 2006, 131, 568-572.
(19) Zhang, X.; Young, M. A.; Lyandres, O.; Van Duyne, R. P. J. Am. Chem. Soc.
2005, 127, 4484-4489.
(20) McFarland, A. D.; Young, M. A.; Dieringer, J. A.; Van Duyne, R. P. J. Phys.
Chem. B 2005, 109, 11279-11285.
(21) Sulk, R.; Chan, C.; Guicheteau, J.; Gomez, C.; Heyns, J. B. B.; Corcoran, R.;
Carron, K. J. Raman Spectrosc. 1999, 30, 853-859.
(22) Jianxin, Q. Y.; Sun, L. J. Phys. Chem. B 1997, 101, 8221-8224.
(23) Kennedy, B. J.; Spaeth, S.; Dickey, M.; Carron, K. T. J. Phys. Chem. B 1999,
103, 3640-3646.
7212 Analytical Chemistry, Vol. 78, No. 20, October 15, 2006
troscopy system (Figure 1A). This system consists of a laser as a
monochromatic light source, a band-pass filter to clean up the
laser line, a series of steering, focusing optics to deliver the laser
light to the sample, and collection optics, which relay the scattered
light to the detector, a high-quality, long-pass filter to reject
Rayleigh scattering, a third meter spectrograph for wavelength
dispersion, and a CCD detector. Following the experiment, the
animals were sacrificed with an overdose of anesthetic and
bilateral thorachotomy, according to protocol.
Experimental Procedure and Spectroscopic Measure-
ment. Glucose was varied in the rat through intermittent
intravenous infusion for 3 h. An infusion of glucose was delivered
over 5-10 min, at a concentration of 1 g/mL in sterile phosphate-
buffered saline via the femoral cannula. A droplet of blood was
drawn from the rat, the glucose level was measured with the One
Touch II glucometer, and corresponding SERS measurements
were taken. The SERS spectra were acquired through the optical
window using a Ti:sapphire laser (ìex ) 785 nm, P ) 50 mW, t )
2 min). The data were collected and analyzed by the partial least-
squares method previously described.7-10,15
Time Constant Analysis. After being implanted in the rat for
5 h, a DT/MH-functionalized AgFON surface was removed and
placed in a flow cell containing bovine plasma to simulate the in
vivo environment. Step changes in glucose concentration were
made by introduction into the cell of high and low concentrations
of glucose dissolved in plasma. The data were processed using
PeakFit 4.12 software (Systat Software Inc, Richmond, CA). To
remove the varying background in SERS spectra, a fourth-order
polynomial was subtracted from the baseline using MATLAB
software. The spectra were further preprocessed in PeakFit with
linear best-fit baseline correction and Savitsky-Golay smoothing.
The amplitude of the Raman bands was obtained by fitting the
data to the superposition of the Lorentzian amplitude line shapes.
The data were then iteratively fit to an exponential curve to
minimize the residuals.
RESULTS AND DISCUSSION
Comparison of SERS to Electrochemical Measurement.
Figure 2 shows the glucose concentration variation in the rat
measured using SERS and the One Touch II blood glucose meter
with respect to time. The inset shows that autofluorescence is
Figure 2. Time course of the in vivo glucose measurement. Glucose
infusion was started at t ) 60 min. Triangles (2) are measurements
made using One Touch II blood glucose meter, and squares (9) are
measurements made using the SERS sensor. Glucose infusion was
started at �1 h, as demarcated by the arrow. The inset shows a typical
in vivo spectrum compared to a typical ex vivo spectrum of the same
surface. (ìex ) 785 nm, P ) 50 mW, t ) 2 min).
Figure 1. Schematic of (A) instrumental apparatus, (B) sensor preparation, (C) morphology, and (D) optical characterization. (A) A rat with a
surgically implanted sensor and optical window was integrated into a conventional laboratory Raman spectroscopy system consisting of a
Ti:sapphire laser (ì ) 785), band-pass filter, steering and collection optics, and a long-pass filter that rejected Raleigh scattered light. (B)
AgFONs were prepared by depositing metal through a mask of self-assembled nanospheres. The AgFON was then functionalized by successive
emersions in ethanolic solutions of decanethiol and mecaptohexanol. Glucose is able to partition in to and out of the DT/MH layer, as shown in
the left of the frame (C) The resultant structure is show in the atomic force micrograph. (D) After functionalization, a reflectance spectrum was
collected to determine the position of the LSPR.
Analytical Chemistry, Vol. 78, No. 20, October 15, 2006 7213
not a significant concern if 785-nm excitation is used, as the typical
SERS spectra of a DT/MH FON prior to implantation and in vivo
show only minor differences, attributable to the change in
environment. Both the standard glucometer and the SERS-based
measurements effectively tracked the change in glucose concen-
tration. A sharp rise in glucose concentration is detected by both
techniques after the start of the glucose infusion (t ) 60 min).
Figure 3 plots the time-independent data on the Clarke error grid
to more precisely gauge the performance of the SERS measure-
ment system.
Error Grid Analysis. The Clarke error grid was developed
as a convenient and modality-independent means to compare the
accuracy and performance of glucose sensors in the clinically
relevant range.24 The grid is divided into five zones, and predic-
tions within these zones lead to the following: (A) clinically correct
measurement and treatment, (B) benign errors or no treatment,
(C) incorrect measurements leading to overcorrection of accept-
able glucose levels, (D) dangerous failure to detect and treat, and
(E) treatments that further aggravate abnormal glucose levels.
The majority of measurements from all samples fell within the
acceptable range. Figure 3 shows a representative Clarke error
grid analysis of a single rodent. The 26 measurements were taken
from a single spot on the implanted DT/MH-functionalized
AgFON surface. The calibration set was constructed using 21 data
points, which were correlated with the commercial glucometer.
The validation set utilized the remaining five measurements as
independent data points. The sensor had relatively low error
(RMSEC ) 7.46 mg/dL (0.41 mM) and RMSEP ) 53.42 mg/dL
(2.97 mM)). These data compare favorably with our previous in
vitro results,8 as well as those of other optically based glucose
measurements,25-28 and with existing detection methods, which
have instrument-dependent coefficients of variation of 0.96-26.9%
(0.096-2.69 mM, 1.75-49 mg/dL at 10 mM).29,30
Time Constant Analysis. Important parameters governing the
overall efficacy of a given sensor are its response time, revers-
ibility, and long-term stability. Previous research has demonstrated
that FON-based sensors are quite stable with good reversibility
under a variety of conditions.9,16,31,32 The present work shows the
DT/M SERS sensor possesses sufficiently rapid response to
measure the glucose injection, keeping pace readily with the
conventional glucometer. However, the infusion rate used was
slow, and there is an inherent lag in the physiological glucose
levels, particularly in interstitial fluid. Because it was impossible
to rapidly and accurately vary glucos