View Article Online / Journal Homepage / Table of Contents for this issue

Analyst Dynamic Article Links < C

Cite this: Analyst, 2011, 136, 1192

www.rsc.org/analyst PAPER
Continuous glucose monitoring by means of mid-infrared transmission laser
spectroscopy in vitro†
Published on 27 January 2011. Downloaded by Iowa State University on 20/09/2013 11:13:14.

Christian Vrancic,a Anna Fomichova,a Norbert Gretz,b Carina Herrmann,a Sabine Neudecker,b
Annemarie Puccia and Wolfgang Petrich*a
Received 17th July 2010, Accepted 21st December 2010
DOI: 10.1039/c0an00537a

The continuous surveillance of glucose concentration reduces short-term risks and long-term
complications for people with diabetes mellitus, a disorder of glucose metabolism. As a first step
towards the continuous monitoring of glucose, reagent-free transmission spectroscopy in the mid-
infrared region has been carried out in vitro using a quantum cascade laser and an optical silver halide
fiber. A 30 mm gap in the fiber allowed for transmission spectroscopy of aqueous glucose solutions at
a wavelength of 9.69 mm, which is specific to a molecular vibration of glucose. A noise-equivalent
concentration as low as 4 mg/dL was achieved at an average power of 1.8 mW and an integration time
of 50 s. This is among the most precise of glucose measurements using mid-infrared spectroscopy. Even
with the very low average laser power of 0.07 mW the sensitivity of previous results (using a fiber optical
evanescent field analysis) has been improved upon by almost one order of magnitude. Finally, the
impact of potentially interfering substances such as other carbohydrates was analyzed.

1. Introduction precautions have to be taken to reduce or avoid adverse reactions

Diabetes mellitus is a disorder of glucose metabolism and one of
the most challenging diseases, both medically and economically.
The prevalence rate amounts to 280 million people worldwide
and an increase of more than 50% has been predicted by 2030,
particularly high in developing countries.1 The frequent testing of
the concentration of glucose in the blood has been shown to
reduce both the risk for long-term complications (e.g., diabetic
foot, retinopathy) and acute events (e.g., diabetic coma).2,3
While glucose testing is usually performed up to 7 times a day
by applying a drop of capillary blood from the finger tip to
a chemical test strip,4 recent technical advances target the
continuous surveillance of the blood glucose concentration. In
this context continuous refers to a monitoring interval that is on
the order of glucose reading per minute.
The current standard for continuous monitoring is based on
an enzymatic reaction of glucose and the electrochemical detec-
tion of the reaction products.5,6 However, a number of
a
Kirchhoff Institute for Physics, University of Heidelberg, Im Neuenheimer
Feld 227, 69120 Heidelberg, Germany. Tel: +49 6221 54-9893
b
Medical Research Center, University of Heidelberg, Theodor-Kutzer-Ufer
1-3, 68167 Mannheim, Germany
† This paper was submitted as part of an Analyst themed issue on Optical
Diagnosis. The issue includes work which was presented at SPEC 2010
Shedding Light on Disease: Optical Diagnosis for the New Millennium,
which was held in Manchester, UK June 26th–July 1st 2010. Other
papers on this topic can be found in issue 12 of vol. 135 (2010). This
issue can be found from the Analyst homepage
(http://www.rsc.org/analyst).
of the patient’s immune system to the implanted reagents.7
Furthermore, reagent-based, continuously measuring sensors
strongly depend on the stability of the required reagents such as
enzymes. Finally, from a principal standpoint, the reagent-based
approach consumes glucose, while reagent-free approaches only
observe the analyte under investigation.
A wide variety of optical methods has been proposed for the
reagent-free observation of glucose,8 such as polarimetry, or
refractometry. Spectroscopy of the molecular vibrations of
glucose by means of infrared or Raman spectroscopy9–11 has been
shown to be a sensitive and specific method in vitro12–16 and ex
vivo.17,18
The goal of the research described here is to evaluate the
potential of a reagent-free, fiber-based, and minimally invasive
glucose sensor in vitro. The fiber-based sensor may later be
implanted into the subcutaneous fatty tissue in order to contin-
uously monitor the body’s glucose levels within the interstitial
fluid (ISF) where the concentration of glucose is known to
correlate with the glucose concentration in blood with an average
time lag of about 10 min.19 The abdication of potentially biotoxic
reagents should lead to an improved biocompatibility and thus to
a prolonged operational life time in the human body. Further-
more, the non-consuming character of a reagent-free sensor
avoids the conceivable deterioration of reagent-based sensors.
Vibrational spectroscopy is known for its high sensitivity and
selectivity in the mid-infrared spectral range at wavelengths
around 10 mm. The presented approach is based in this so-called
fingerprint region in which strong vibrational bands of glucose

1192 | Analyst, 2011, 136, 1192–1198 This journal is ª The Royal Society of Chemistry 2011

overlap the broad absorption spectrum of water. A compact
setup seems feasible due to the commercial availability of
quantum cascade lasers (QCLs) in this region.20–25 Transmission
spectroscopy in an IR-element was also reported previously.26,27
Lambrecht et al. showed the potential of a QCL-based deter-
mination of glucose by means of fiber-based evanescent field
spectroscopy.28 In addition, first fiber-based transmission
measurements were reported in this reference. Starting from
these promising, preliminary results of fiber-based mid-infrared
transmission spectroscopy with QCLs, further research has been
Published on 27 January 2011. Downloaded by Iowa State University on 20/09/2013 11:13:14.
pursued in order to elucidate the detection limits of this method
in vitro.
2. Methods & materials
In the in vitro experiments, mid-infrared radiation was sent
through an optical fiber which had been positioned inside a flow-
through chamber. The transmission measurements were con-
ducted in a cavity within the fiber. The fluid transport to the fiber
was accomplished by pumping fluid through the flow-through
chamber, whereas the actual transport of glucose into the gap
had to rely on diffusion.
The mid-infrared radiation originated from a Fabry-Perot
quantum cascade laser (Alpes Lasers SA, Neuch^atel, Switzerland)
which operated at a wavelength of 9.69 mm with a pulse width of 50
ns and at a repetition period of 3.3 ms (duty cycle 1.5%) resulting in
a modulated, average output power of 1.8 mW. The laser was
Peltier-cooled to a temperature of 30  C and operated at 22 V.
The laser radiation allowed for an integral absorption measure-
ment over the interval from 1010 cm1 to 1065 cm1. The trans-
mitted radiation was detected by a pyroelectric detector (LME-
353, InfraTec GmbH, Dresden, Germany) using a lock-in tech-
nique with a modulation frequency of 110 Hz (SR830, Stanford
Research Systems, Inc., Sunnyvale, CA, USA).
Silver halide (AgCl/AgBr) was chosen as an optical fiber
material due to its small absorption in the mid-infrared region
and availability as fiber material. Fibers with an outer diameter
of 500 mm and a core diameter of 450 mm (A.R.T. Photonics
GmbH, Berlin, Germany) were modified by inserting a cavity
along the transmission pathway. The refractive indices of the
core and cladding were nCore ¼ 2.16 and nClad ¼ 2.14, respec-
tively. Previous calculations and optical simulations have shown
that the optimal path length through an aqueous glucose solu-
tion for a wavelength of 9.7 mm amounts to 16.5 mm.29 Such short
transmission path lengths were realized by milling an approxi-
mately 30 mm deep gap into the front facet of a silver halide fiber
using a 300 mm wide blade. The flat end of a second fiber was then
brought into contact with the modified facet of the first fiber. A
hull made out of polyether ether ketone (PEEK) with an inner
diameter of 500 mm was then used to adjust and fix the two facets
with a two-component adhesive (plus endfest 300, UHU GmbH
& Co. KG, B€ uhl, Germany). The hull had a circular opening in
order to allow for a diffusion-controlled transport of the solutes
into the fiber gap (see Fig. 1). This fiber sensor was then posi-
tioned in a flow-through chamber with a volume of 2.6 mL which
could be automatically filled with different test solutions. It
should clearly be pointed out that the filling process of the cavity,
which was oriented parallel to the flow direction, was diffusion-
controlled, since very high pressures would otherwise have been
This journal is ª The Royal Society of Chemistry 2011
View Article Online
Fig. 1 Schematic view of the setup of the fiber sensor.
required for a convection current through the cavity due to its
microscopic cross section.
The flow regulation of a peristaltic pump (LabDos; pump
drive: HiTec Zang GmbH, Herzogenrath, Germany; pump head:
Watson-Marlow GmbH, Rommerskirchen, Germany), valve
control (Smartline Valve Drive S12, Knauer GmbH, Berlin,
Germany) and lock-in amplifier were controlled and/or read out
by a PC with a control and measurement script written in Perl.
Further details about the setup can be found in the references.29
Ten different glucose solutions (Merck KGaA, Darmstadt,
Germany) were prepared with deionized water to result in
concentrations c [mg/dL] of 40, 70, 100, 150, 200, 250, 300, 400,
500 and 600. All solutions were phosphate buffered at pH ¼ 7.4
(10 mM KH2PO4/10 mM Na2HPO4$2H2O) and supplemented
with a bacteriostatic agent (0.095 wt% NaN3) to inhibit glucose
consumption by microorganisms.
To determine the impact of water absorption and the addition
of glucose to water, QCL emission and fiber transmission spectra
were recorded for different sample solutions of the fiber cavity.
Note that the pyroelectric detector was replaced by a FT-IR
spectrometer (MATRIX-M, Bruker Optik GmbH, Ettlingen,
Germany) for this measurement only. The resolution for 50 scans
each was set to 1 cm1 in the spectral range of 1010–1065 cm1.
After the transmission through air, the cavity was filled with
deionized water and finally with a glucose solution (cGlu ¼ 500
mg/dL).
For the determination of the glucose sensitivity, pure, deion-
ized water, a water–ethanol mixture (90/10 wt%) and a glucose
solution were alternately pumped through the measurement
chamber. Water and the water–ethanol mixture were used for
reference and cleaning purposes, respectively. Both were pumped
for 20 min at a rate of 15 mL/min. The aqueous solutions of
different glucose concentrations were pumped, in protocolled but
random order, with a rate of 10 mL/min for 30 min. The
measurement signals were baseline corrected, using one 4th grade
polynomial fit for all of the water plateaus in any series of
measurements. In the following analysis, the mean signals of the
glucose solutions were derived from the plateaus over a time
interval of 17 min in order to ensure that the flow-chamber and
transmission cavity were completely filled with the glucose
solution under investigation and for consistency with other
measurements. The differences between these mean glucose sig-
nals‡ and the water baseline were calculated in order to obtain
‡ Since glucose changes in absorptions are small, it is legitimate to use
a linearized approximation of the Lambert-Beer law where U(c) f c.
Analyst, 2011, 136, 1192–1198 | 1193
 View Article Online

the concentration-dependent signal differences U(c). The sensi-
tivity vU/vc of the fiber sensor was then derived by applying
a linear fit onto these signal differences. The analysis of the
measurements was automated with the help of MatLab (version
7.0, The MathWorks, Inc., Natick, MA, USA).
Following guidelines for interfering substances,30 the inter-
ference of further sugars was analyzed, namely fructose, lactose
(both Carl-Roth GmbH, Karlsruhe, Germany), galactose
(AppliChem GmbH, Darmstadt, Germany) and maltose (Merck
KGaA, Darmstadt, Germany). In addition, the impact of NaCl
Published on 27 January 2011. Downloaded by Iowa State University on 20/09/2013 11:13:14.
a prolonged measurement was performed over the course of
more than 48 h with a cavity filled with water only.
The dependence of the signal stability on the QCL duty cycle
was analyzed by increasing the repetition period by factors of 3
to values of 10 ms, 30 ms and 90 ms, leading to duty cycles of
0.50%, 0.17% and 0.06%. For each of the duty cycle settings,
both the glucose concentration measurement series and the long-
term measurement in water were repeated under identical
conditions.
The Allan variance s2Allan may be used to characterize the noise

(Mallinckrodt Baker B. V., Deventer, Netherlands), KCl, and drift as a function of a given time interval s.27 It can be
MgCl2, sodium lactate (all AppliChem GmbH), CaCl2, KH2PO4, derived according to
NaHCO3, urea (all Gr€ ussing GmbH, Filsum, Germany), ethanol
1 X
M1
2
(Sigma-Aldrich Chemie GmbH, Taufkirchen, Germany) and s2Allan ðsÞ ¼ *
Usþ1 ðsÞ  Us* ðsÞ (1)
ammonia (VWR International S.A.S., Fontenay-sous-Bois, 2ðM  1Þ s¼1
France) was tested. An overview of the substances, the recom-
summing up the squared deviations of the averaged signals U*(s)
mended and actual testing concentrations and the standard
over (M  1) adjacent time intervals of length s. The Allan
deviations within the group of healthy patients (as well as their
formalism was applied to the long-term measurement leading to
impact on the measurement signal, see Sec. 3) is given in Table 1.
a measure of signal stability and especially to the time interval for
Each measurement range was covered by 10 different concen-
optimum analyte sensitivity.
trations.
With the knowledge of sAllan(s) as well as the calibration
For technical reasons, the measurements for NaHCO3, sodium
sensitivity vU/vc from the glucose measurements, the concen-
lactate and ethanol were conducted with a second sensor with
tration can be derived at which the signal-to-noise ratio equals 2
a gap width of about 50 mm and the impact on the measurement
(noise-equivalent concentration, NEC):
signal was scaled according to the glucose sensitivity of this
sensor. The flush time of water–ethanol and the non-sugar ana- 2sAllan ðsÞ
NECðsÞ ¼ : (2)
lyte solutions was reduced to 3 min and 20 min, respectively, with vU=vc
a general pumping rate of 10 mL/min. The analysis time interval
was also set to 17 min. The drift compensation of the QCL signal
3. Results
was performed analogous to the analysis of the series of glucose
measurements. Fig. 2 shows a comparison between the emission spectra of the
Averaging over adjacent data points can help to reduce short- QCL after having passed through the fiber and its cavity. As the
term fluctuations but is susceptible to long-term drifts. To silver halide fiber does not have any pronounced absorption
quantify short-term noise as well as long-term stability, bands within the spectral region considered,31,32 it is evident from

Table 1 Molecular weight, guidance concentrations, standard deviations of healthy patients and measurement range for glucose and interfering
substances performing integral absorption measurements from 1010 to 1065 cm1. The slope reflects the signal change as a function of the concentration
of the analyte. The R2 value is a measure of the quality of the linear fit of the analysis described in Sec. 2. *The slope for NaHCO3, sodium lactate and
ethanol was scaled due to the use of a alternative fiber sensor based on their glucose sensitivity

molecular CLSI standard measurement

interferent weight [g/mol] guidance [mg/dL] deviation35 [mg/dL] range [mg/dL] slope [nV/mg/dL] R2

glucose 180.2 1000 — 4–600 376.5  3.7 0.9991

fructose 180.2 18 — 0.5–100 126.0  16.0 0.8714
galactose 180.2 15 — 0.5–100 291.9  18.7 0.9643
lactose 342.3 — — 0.5–100 254.5  28.4 0.8981
maltose 342.3 — — 1–450 376.6  3.3 0.9993
Na+ 23.0 414 5.8 8–3932
31.5  0.1 0.9998
Cl 35.5 420 9.8 12–6068
K+ 39.1 27 1.6 10–5241
25.9  0.2 0.9993
Cl 35.5 420 9.8 10–4759
Mg2+ 24.3 36 0.2 1–2550
7.3  0.2 0.9901
Cl 35.5 420 9.8 4–7450
Ca2+ 40.1 20 0.4 4–3609
13.8  0.3 0.9950
Cl 35.5 420 9.8 6–6391
Na+ 23.0 414 5.8 2–860
31.1  0.2* 0.9997
HCO3 84.0 294 8.4 8–3140
Na+ 23.0 414 5.8 2–1017
66.9  1.7* 0.9951
sodium lactate 90.1 59 4.1 8–3983
K+ 39.1 27 1.6 3–2873
61.4  0.3 0.9998
H2PO4 97.0 — — 7–7127
ethanol 46.1 400 — 10–10000 95.6  6.6* 0.9583
urea 60.1 257 9.0 10–10000 2.0  0.2 0.9013

1194 | Analyst, 2011, 136, 1192–1198 This journal is ª The Royal Society of Chemistry 2011

Published on 27 January 2011. Downloaded by Iowa State University on 20/09/2013 11:13:14.
Fig. 2 QCL emission spectra after transmission through the fiber and its
cavity with different sample solutions. The spectra were recorded by
coupling the fiber into a FT-IR spectrometer. The arrow indicates the
position of a known maximum of the absorbance spectrum of glucose.
the measurement with the air-filled cavity that the QCL emits
multi-line with a center-wavelength of 9.69 mm (wavenumber:
1032 cm1), a FWHM of the central mode of approximately
94 nm (10 cm1) and an emission range (at least 10% of the
maximum intensity) of 376 nm (40 cm1). If the cavity was filled
with water or aqueous glucose solution, the integral signal
declined to 27.0% and 26.0%, respectively, compared to the air-
filled cavity. More precisely, the detector signal was reduced by
3.7% upon the addition of 500 mg/dL glucose to water. Since
such a small signal change is expected to linearly depend on
glucose concentration, the relative change of signal amounts to
7.4  105 per mg/dL.
An example of the raw data of the in vitro experiments can be
found in Fig. 3. A signal reduction due to the additional
absorption of glucose is visible in the decline of the signal relative
to pure water. The larger decline of the signal of the water–
ethanol mixture is based on the strong absorption of ethanol in
this spectral region that exhibits a 0.74 mm (82 cm1) wide
absorption band around 9.51 mm (1051 cm1) which is caused by
an antisymmetric CCO stretch vibration of ethanol.33 A drift of
Fig. 3 Lock-in signal over the course of a measurement with 10 different
glucose solutions, water and water–ethanol. The glucose concentration
labels can be found on top of the corresponding glucose solution
plateaus. As a guide, the dashed line indicates the polynomial baseline fit
to the water plateaus.
This journal is ª The Royal Society of Chemistry 2011
View Article Online
the signal is observed among the baseline plateaus of both water
and water–ethanol.
While the rise time from one plateau to the next amounts to
approximately 60 s at a pumping rate of 15 mL/min, the total
duration between the switching of the valve and the observation
of a stable signal (variation < 0.2%) is around 90 s. This is most
likely due to the time it takes the glucose to penetrate the gap by
diffusion.
To establish a correlation between the change in the signal and
the concentration of the detected glucose solution, the difference
between the mean signal of the plateau and the water baseline is
analyzed as described above. This difference signal depends on
the glucose concentration linearly as shown in Fig. 4 (which in
turn confirms the expected validity of the linear approximation
to the Lambert-Beer law for small concentrations).
From the slope of this linear dependence it can be derived that
a change of the glucose concentration DcGlu of 1 mg/dL corre-
sponds to a decline of the detector signal U by (377  4) nV with
a coefficient of determination R2 of 0.9991. It is of note that the
extrapolation of the linear fit misses the origin by (30  1) mV
due to the additional absorption of the phosphate buffer in the
glucose solutions and, thus, a larger difference to the reference
water signal.
The impact of the 4 disturbing sugars under investigation is
summarized in Table 1. The measurements indicate that only
maltose exhibits a specific absorption comparable to glucose at
the designated QCL wavelength. The results of the other
interferents are also listed in Table 1. An increase in the signal
was observed for monatomic salt ions. This signal change is
caused by local density changes due to the displacement of
water by the added analyte and further effects as outlined in the
references.34
Finally, and as expected from the literature, very
high concentrations of ammonia dissolved the silver halide
fiber almost completely over the course of the measurement
(cammonia ¼ 0.01.250 g/L) and led to a degradation of the signal.
Under the crude assumption of the fiber degradation being linear
in time and linear to the concentration of ammonium, one might
speculate that a recommended maximum analyte concentration
Fig. 4 Correlation of the signal differences between glucose plateaus
and water baseline as a function of the concentration. The linear fit has
a slope of (377  4) nV/mg/dL with a coefficient of determination R2 of
0.9991.
Analyst, 2011, 136, 1192–1198 | 1195

Published on 27 January 2011. Downloaded by Iowa State University on 20/09/2013 11:13:14.
Fig. 5 Noise-equivalent concentration (NEC) as a function of the
measurement integration time for varying duty cycles and, thus, average
laser output powers. The results are compared to the work of Lambrecht
et al. which were obtained using a fiber optical evanescent field analysis
(FEFA) sensor.28
of 112 mg/dL30 would yield a signal reduction of less than 1% over
a duration of 20 days.
The analysis of the long-term measurements using water only
with regard to the NEC is shown in Fig. 5 for varying integration
time intervals s and repetition periods of the QCL. The results
using the evanescent field fiber sensor (fiber optical evanescent
field analysis, FEFA) are also displayed as a comparison. Note
that the QCL of the FEFA sensor operated at a wavenumber of
1010 cm1 which is expected to lead to a lower glucose sensitivity.
The optimal integration time for each duty cycle as well as the
minimum NECs can be found in Table 2.
4. Discussion
Based on the previous results using evanescent field spectroscopy
as well as the promising initial findings using transmission
spectroscopy,28 QCL-based mid-infrared transmission spectros-
copy was utilized for the continuous determination of glucose in
vitro.
The analysis of the system stability analogous to the Allan
variance shows that for short time intervals s, the decrease of the
pffiffiffi
NEC follows a 1= s behavior, which is indicative for thermal
noise: As the signal linearly increases with integration time, while
the thermal noise only increases with the square-root of the
integration time, it follows that the NEC should decrease as
observed. On the other hand, very long integration times tend to
increase the NEC due to a drift of the overall system. For
intermediate times, and especially for a time interval of s ¼ 50 s
Table 2 Optimal integration times and corresponding minimum NECs
for different average QCL output powers
average optimal integration minimum
power [mW] time [s] NEC [mg/dL]
1.80 50 4 deviations cause apparent signal changes that could lead to false
0.60 110 7 glucose predictions. Salt ions are known to have an indirect
0.20 340 10
0.07 1020 27
1196 | Analyst, 2011, 136, 1192–1198
View Article Online
and at an average output power of 1.8 mW (duty cycle 1.5%),
a minimum NEC of 4 mg/dL is achieved, which constitutes an
improvement by more than one order of magnitude as compared
to the NEC values obtained previously. It is also considerably
smaller than measurements of serum films using mid-infrared
attenuated total reflectance (ATR) or transmission-based
spectroscopy.36 However, the presented measurements were
performed with a lower grade of complexity and a very good
reproducibility compared to dry serum films. For smaller duty
cycles the minimum of the NEC shifts to larger values and longer
integration times. Note that at an average laser power as low
as 70 mW the NEC remains substantially lower than the best
QCL-based values published previously. This is valid even when
compared to measurements using the FEFA sensor with cryo-
genically cooled lead salt lasers (LSLs) as reported previously28
where a NEC of 40 mg/dL was achieved (the LSLs operated in
continuous wave mode at a wavenumber of 1035 cm1). The
reduction of the average power by almost one order of magni-
tude down to 0.2 mW still delivers a NEC of 10 mg/dL, which is
comparable to the accuracy of standard, hand-held glucose
meters and which, at an integration time of 6 min, would still
hold promise for the continuous surveillance of glucose. Thus,
fiber-based laser transmission spectroscopy appears to be
well-suited for the continuous monitoring of glucose, based on
these in vitro results.
A further estimation of the detection limit of the presented
method was performed by dividing the long-term measurement
into 48 parts with a length of 60 min each. Then the standard
deviation of the means of 50 s intervals was determined for each
part as this time showed to deliver the best compromise between
short-term noise and long-term drifts based on the Allan
variance analysis. The mean standard deviation s0 of all 48 parts
amounted to 1.1 mV corresponding to a glucose concentration
change of 2.9 mg/dL. Following recommendations for analyzing
detection capabilities,37 the detection limit (‘‘minimum detectable
value’’) LD can be derived according to LD ¼ 3.29s0 ¼ 3.6 mV or
11.9 mg/dL. This approach does not exclude the effect of
low-frequency drifts which can be compensated for by reference
measurements, e.g., at a second wavelength. On the other hand,
the Allan formalism delivers an intrinsic detection limit of the
method.
The impact of further carbohydrates maltose, fructose,
galactose and lactose was also determined. Unlike glucose, the
other tested monosaccharides, namely galactose and the diabetic
sugar replacement fructose, both have a slow physiological
absorption and are almost completely metabolized and partly
converted to glucose derivates in the liver.38 Although in special
situations, unusually high concentrations of maltose39 or fruc-
tose40,41 may impose some restrictions, the impact of other sugars
will generally be small due to their low concentrations in the
interstitial fluid.
Beyond the impact of the total mean concentration of any
interfering substance on the measurement signal, it is vital to
emphasize the importance of the variation of the concentration
for the glucose sensitivity. Instead of a constant signal offset,
influence on the spectrum of IR-active analytes.34 Together with
a high variability in concentration, the impact of these analytes
This journal is ª The Royal Society of Chemistry 2011

can be coarsely estimated by artificially assigning the total signal
change of 32 nV/mg/dL of NaCl to chloride only (see Table 1).
With a standard deviation of the concentration of chloride of 9.8
mg/dL, this would lead to a signal deviation of 314 nV.
Compared to a glucose-induced signal change of 377 nV/mg/dL
and a NEC of 4 mg/dL, it is obvious that the disturbance of the
possible interferents within their reference interval is low.
However, exceptionally high concentrations of interferents (such
as ethanol or lactate during power exercises) will require further
attention.
Published on 27 January 2011. Downloaded by Iowa State University on 20/09/2013 11:13:14.
In the experiments, the overall time to obtain stable signal
readings after changing the glucose concentration amounts to 90
s, which thus might serve as a crude upper limit for the diffusion
time. However, the active pumping process into the chamber
needs to be separated from the purely diffusive transport of the
biofluid into the cavity. As pointed out earlier, the pumps in this
study served only as a tool to change solutions in the surrounding
measurement chamber whereas the filling of the gap depended on
diffusion solely. Accordingly, the diffusion process presents
a restrictive factor to the size of the outer fiber diameter as the
diffusive path lengths for the fluid must be kept very small in an
implantable, continuously operating sensor. Given the diffusion
coefficient D of glucose in water (D ¼ 6.73  106 cm2/s)42 and
the radius of the fiber of 250 mm, it can be coarsely estimated that
the mere diffusion process amounts to approximately 45 s.
While further technical steps focus on improving the sensor
head as well as extending research concerning interfering
substances such as proteins and triglycerides, an increasingly
important part of future research will target the biocompatibility
and applicability in vivo. It is of note that a reliable and robust
long-term performance of the fiber sensor requires additional
effort in more complex measurements designed closer to an
in vivo environment. This is true for both interferents identified to
produce significant deceptive signal changes and those leading to
fiber degradation. We further suspect that, for example, a second
and/or tunable QCL should allow for the correction of baseline
drifts caused by variations of interfering substances in vivo as
exemplified previously (in vitro).16,43 In addition, a baseline
correction seems essential to correct for the impact of tempera-
ture changes, e.g., induced by variations of the local tissue
temperature.
In passing we would like to note that the research presented
here was accompanied by biotoxicity testing of the materials in
order to rank the priorities with respect to alternative sensor
materials. For reasons of clarity, brevity and focus, the results
of the biotoxicity testing and countermeasures will be reported
in detail elsewhere. In short, one concept along the progression
to in vivo applicability is the fact that polymer coatings may
strongly reduce the biotoxicity of silver halide fibers. At the
same time, a polymer coating can protect the fiber from dete-
rioration44,45 caused by interaction with dissolved biological
agents such as body fluid electrolytes.46 With the use of such
a biocompatible coating and/or membrane, it may be possible to
maintain the good performance of the system even under in vivo
conditions.
With the realization of a dependable, reagent-free continuous
glucose sensor, the connection to an insulin pump may one day
complete a minimally invasive closed-loop system for glucose
control, i.e., the artificial pancreas.
This journal is ª The Royal Society of Chemistry 2011
View Article Online
5. Disclosures
In addition to his affiliation with the Kirchhoff Institute for
Physics, W. Petrich is an employee of Roche Diagnostics GmbH,
Mannheim, Germany.
6. Acknowledgments
This work was financed by the Baden-W€ urttemberg Stiftung.
The authors thank the Nikon Imaging Center at the University
of Heidelberg for their assistance and support.
References
1 International Diabetes Federation, IDF Diabetes Atlas, 4th edn.
2 The Diabetes Control and Complications Trial Research Group,
N. Engl. J. Med., 1993, 329, 977–986.
3 The UKPDS Study Group, Lancet, 1998, 352, 837–853.
4 J. H€ones, P. M€uller and N. Surridge, Diabetes Technol. Ther., 2008,
10, S10–26.
5 T. Koschinsky and L. Heinemann, Diabetes/Metab. Res. Rev., 2001,
17, 113–123.
6 G. S. Wilson and R. Gifford, Biosens. Bioelectron., 2005, 20, 2388–2403.
7 A. Mang, J. Pill, N. Gretz, B. Kr€anzlin, H. Buck, M. Schoemaker and
W. Petrich, Diabetes Technol. Ther., 2005, 7, 163–173.
8 Handbook of optical sensing of glucose in biological fluids and tissues,
ed. V. V. Tuchin, Taylor & Francis, Inc., Boca Raton, FL, 2008.
9 W. Petrich, Appl. Spectrosc. Rev., 2001, 36, 181–237.
10 Vibrational Spectroscopy for Medical Diagnosis, ed. M. Diem, P. R.
Griffiths and J. M. Chalmers, John Wiley & Sons, Inc., Hoboken,
NJ, 2008.
11 Biomedical Vibrational Spectroscopy, ed. P. Lasch and J. Kneipp,
John Wiley & Sons, Inc., Hoboken, NJ, 2008.
12 D. Rohleder, W. Kiefer and W. Petrich, Analyst, 2004, 129, 906–911.
13 D. Rohleder, G. Kocherscheidt, K. Gerber, W. Kiefer, W. K€ ohler,
J. M€ocks and W. Petrich, J. Biomed. Opt., 2005, 10, 031108.
14 R. Vonach, J. Buschmann, R. Falkowski, R. Schindler, B. Lendl and
R. Kellner, Appl. Spectrosc., 1998, 52, 820–822.
15 M. K€ olhed, M. Haberkorn, V. Pustogov, B. Mizaikoff, J. Frank,
B. Karlberg and B. Lendl, Vib. Spectrosc., 2002, 29, 283–289.
16 W. B. Martin, S. Mirov and R. Venugopalan, Appl. Spectrosc., 2005,
59, 881–884.
17 H. M. Heise, A. Bittner, T. Koschinsky and F. A. Gries, Fresenius J.
Anal. Chem., 1997, 359, 83–87.
18 P. Magarian and B. Sterling, J. Diabetes Sci. Technol., 2010, 3, 1411–
1418.
19 E. Cengiz and W. V. Tamborlane, Diabetes Technol. Ther., 2009, 11,
S11–16.
20 U. Bindig, M. Meinke, I. Gersonde, O. Spector, I. Vasserman,
A. Katzir and G. M€ uller, Sens. Actuators, B, 2001, 74, 37–46.
21 H. M. Heise, L. K€ upper and L. N. Butvina, Spectrochim. Acta, Part
B, 2002, 57, 1649–1663.
22 L. Hvozdara, N. Pennington, M. Kraft, M. Karlowitz and
B. Mizaikoff, Vib. Spectrosc., 2002, 30, 53–58.
23 A. A. Kosterev, R. F. Curl, F. K. Tittel, C. Gmachl, F. Capasso,
D. L. Sivco, J. N. Baillargeon, A. L. Hutchinson and A. Y. Cho,
Laser Phys., 2001, 11, 39–49.
24 T. Beyer, M. Braun and A. Lambrecht, J. Appl. Phys., 2003, 93, 3158–
3160.
25 C. Charlton, A. Katzir and B. Mizaikoff, Anal. Chem., 2005, 77,
4398–4403.
26 C. C. Johnson and L. T. Taylor, Anal. Chem., 1984, 56, 2642–2647.
27 B. Lendl, J. Frank, R. Schindler, A. M€ uller, M. Beck and J. Faist,
Anal. Chem., 2000, 72, 1645–1648.
28 A. Lambrecht, T. Beyer, K. Hebestreit, R. Mischler and W. Petrich,
Appl. Spectrosc., 2006, 60, 729–736.
29 C. Herrmann, C. Vrancic, A. Fomichova, N. Gretz, S. Hoecker,
A. Pucci and W. Petrich, Proc. SPIE–Int. Soc. Opt. Eng., 2010,
75600E.
30 Clinical and Laboratory Standards Institute (CLSI), Interference
testing in clinical chemistry; Approved guideline, Clinical and
Laboratory Standards Institute, Wayne, PA, 2nd edn, 2005.
Analyst, 2011, 136, 1192–1198 | 1197

31 F. Moser, D. Bunimovich, A. DeRowe, O. Eyal, A. German,
Y. Gotshal, A. Levite, L. Nagli, A. Ravid, V. Scharf, S. Shalem,
D. Shemesh, R. Simchi, I. Vasserman and A. Katzir, IEEE J. Sel.
Top. Quantum Electron., 1996, 2, 872–879.
32 B. Dekel, Z. Barkay and A. Katzir, Opt. Commun., 2001, 199, 383–
388.
33 National Institute of Advanced Industrial Science and Technology
(AIST), Spectral Database for Organic Compounds, 2010, accessed
14-Jun-2010.
34 F. Vogt, M. Kraft and B. Mizaikoff, Appl. Spectrosc., 2002, 56, 1376–
1380.
35 K. D€orner, Klinische Chemie und H€
amatologie, Thieme, Stuttgart, 5th
edn, 2003.
Published on 27 January 2011. Downloaded by Iowa State University on 20/09/2013 11:13:14.
36 E. Diessel, P. Kamphaus, K. Grothe, R. Kurte, U. Damm and
H. M. Heise, Appl. Spectrosc., 2005, 59, 442–451.
37 L. A. Currie, Anal. Chim. Acta, 1999, 391, 105–126.
38 S. S. Gropper, J. L. Smith and J. L. Groff, Advanced Nutrition and
Human Metabolism, Wadsworth, Belmont, CA, 5th edn, 2009.
1198 | Analyst, 2011, 136, 1192–1198
View Article Online
39 A. R. Gaines, L. R. Pierce and P. A. Bernhardt, Fatal Iatrogenic
Hypoglycemia: Falsely Elevated Blood Glucose Readings with
a Point-of-Care Meter Due to a Maltose-Containing Intravenous
Immune Globulin Product, FDA technical report, 2009.
40 I. Macdonald, A. Keyser and D. Pacy, Am. J. Clin. Nutr., 1978, 31,
1305–1311.
41 A. S. Truswell, J. M. Seach and A. W. Thorburn, Am. J. Clin. Nutr.,
1988, 48, 1424–1430.
42 CRC Handbook of Chemistry and Physics, ed. D. P. Lide, CRC
PRESS, 81st edn, 2000.
43 S. Schaden, A. Dominguez-Vidal and B. Lendl, Appl. Phys. B: Lasers
Opt., 2006, 83, 135–139.
44 R. G€obel, R. Krska, R. Kellner and A. Katzir, Fresenius’ J. Anal.
Chem., 1994, 348, 780–781.
45 E. Bormashenko, R. Pogreb, S. Sutovski, I. Vaserman and A. Katzir,
Proc. SPIE–Int. Soc. Opt. Eng., 1999, 3570, 100–106.
46 R. Simhi, Y. Gotshal, D. Bunimovich, B.-A. Sela and A. Katzir, Appl.
Opt., 1996, 35, 3421–3425.
This journal is ª The Royal Society of Chemistry 2011