1.pdf
1.pdf
DIABETES TECHNOLOGY & THERAPEUTICS
Volume 10, Number 6, 2008
© Mary Ann Liebert, Inc.
DOI: 10.1089/dia.2008.0033
Preclinical In Vivo Study of a Fluorescence Affinity
Sensor for Short-Term Continuous Glucose Monitoring
in a Small and Large Animal Model
Ralph Dutt-Ballerstadt, Ph.D., Colton Evans, B.S., Ashok Gowda, Ph.D., and Roger McNichols, Ph.D.
Abstract
Background: The performance of a fiber-coupled fluorescence affinity sensor (FAS) was studied in vivo in small
and large animal models, in order to assess its feasibility and safety for short-term glucose monitoring in hu-
mans.
Methods: Determination of interstitial glucose concentrations in skin tissue of hairless rats and small pigs was
facilitated by measuring the fluorescence response of the implanted FAS over several hours and multiple days.
Blood sugar changes in animals were induced by injections of insulin and dextrose. The Medtronic Minimed
CGMS® (Medtronic Diabetes, Northridge, CA) was used for comparison.
Results: The acute in vivo performance study of the fiber-coupled FAS showed that more than 96% of the paired
FAS/venous blood glucose readings were in the clinically acceptable A and B regions of the Clarke Error Grid.
Mean absolute relative difference (MARD) and root mean squared error (RMSE) values for small and large an-
imal models were 18.5% and 19.8 mg/dL and 15.9% and 16.3 mg/dL, respectively. In comparison, MARD and
RMSE for the Medtronic Minimed CGMS in small and large animal models were similar (in rats, 25.4% and
19.8 mg/dL, respectively; in pigs, 18.4% and 16.2 mg/dL, respectively). No instance of irritation or infection
was observed at any implantation site. The in vivo performance of FAS over a 3-day period was successfully
demonstrated in both animal models.
Conclusions: Overall, the fiber-coupled FAS was safe, and its performance during 4-h and 3-day testing com-
pared favorably to the commercially available Medtronic Minimed CGMS, indicating its potential value for di-
abetes management.
453
Introduction
TO ENSURE A TRULY CONTINUOUS sensor readout for the suc-cessful development of a complication-free artificial pan-
creas, improvements in operational functionalities of the cur-
rent commercially implantable sensors for short-term
glucose monitoring in individuals with type 1 diabetes are
desirable.1 For example, the Medtronic Minimed CGMS®
Gold System™ from Medtronic Diabetes (Northridge, CA)
and the STS® sensor from DexCom (San Diego, CA), along
with the FreeStyle Navigator® Continuous Glucose Monitor
by Therasense/Abbott Diabetes Care (Alameda, CA) are the
most advanced sensor systems to date.2–6 All three systems
are based on a disposable electroenzymatic sensing platform,
relying on the amperometric detection of glucose by glucose
oxidase immobilized to an electrode. Both the Medtronic
Minimed CGMS and DexCom STS sensors require several
calibrations per day for reliable blood glucose detection over
a period of 3–7 days. More importantly, both sensors require
up to a 2-h warm-up period immediately after implantation
or, in the case of the Medtronic Minimed CGMS sensor,
whenever the sensor is reconnected to the readout unit (e.g.,
after bathing). Despite their promising value for diabetes
management, it remains to be seen how these electroenzy-
matic-based devices can be successfully integrated into an
artificial pancreas if they require extended warm-up periods
and frequent “finger stick” calibrations and/or if their per-
formance is unreliable because of shifts in the sensor sensi-
tivity.
Over the last several years, research into fluorescence
affinity sensors (FASs) for glucose detection has steadily
gained acceptance among scientists and clinicians through
the introduction of a number of improved glucose-sensitive
assays based on either the glucose-specific protein con-
BioTex, Inc., Houston, Texas.
canavalin A (ConA),7–22 artificial glucose-specific receptors
based on boronate derivatives,23,24 or an inactive form of the
enzyme glucose oxidase.25 There are several intrinsic ad-
vantages of an FAS over electrode-enzymatic sensors in
terms of practicality for in vivo sensing. For one, the light-
based signal detection system does not suffer from the need
for warm-up time when the sensor is disconnected from the
readout unit, since the light-based signal is immediately
available for glucose determination. Second, the absence of
an electrode-based system eliminates potential interferences
of electrode-active components that may enter interstitial
fluid, such as ascorbate, urate, or acetaminophen.26,27 Third,
the nature of ligand-binding interactions in receptor-based
sensors eliminates the occurrence of “autodestructive” side
products, e.g., hydrogen peroxide, as produced by electrode-
enzymatic sensors. And, last but not least, the binding reac-
tion in affinity-based sensors is equilibrium-driven, result-
ing in a signal sensitivity that is independent of the rate of
glucose diffusion into the sensor. This is an advantage when
compared to glucose-consumptive electroenzymatic sensors,
whose signal is rate dependent.
During the last 5 years, our group has made steady
progress toward improvements and optimization of a ConA-
based FAS for in vivo glucose monitoring.28,29 We have pre-
viously reported significant improvements in the chemical
stability of the FAS over earlier ConA-based sensors by im-
mobilizing ConA to a macroporous hydrogel, such as
Sepharose, which eliminates precipitation of ConA and in-
creases in vitro functionality of the FAS over a time period
of up to 6 months at 37°C.28 ConA-Sepharose at 37°C has
been maintained over 450 days with only 20% loss of activ-
ity.29 Besides our research on studying a transdermal FAS
for long-term glucose monitoring in patients with type 1 and
2 diabetes,29 we have been concentrating our interest on de-
veloping a fiber-coupled FAS for short-term (3–5-day) inter-
stitial glucose monitoring. The basic design of this prototype
sensor is facilitated by interrogating a hollow dialysis fiber
containing the fluorescent ConA-based assay with an opti-
cal fiber.30 This concept was first described by Schultz and
co-workers, who demonstrated measurement of blood sugar
in the jugular vein of a dog, as noted in the dissertation the-
sis of Mansouri.31 However, their sensor chemistry was ill-
suited for longer in vivo interrogation because of strong
photobleaching of short-wavelength fluorescent dyes (fluo-
rescein) and elaborate and inferior assay chemistry. In con-
trast, we have employed much brighter and more photo-
stable dyes (Alexa 647 and Alexa 750 [Invitrogen, Eugene,
OR]) at longer wavelength, enabling us to detect fluorescence
with off-the-shelf photodetectors instead of cumbersome and
power-consuming photomultipliers.
In this paper we have summarized results of acute and
chronic in vivo studies performed on a small and large ani-
mal model to evaluate efficacy and safety for potential pre-
liminary human trials over several days.
Materials and Methods
FAS description
The fiber-coupled FAS is a needle-type, fiber optic-based
glucose sensor whose detection principle is based on fluo-
rescence resonance energy transfer of the glucose-specific
chemistry—housed in a hollow fiber—in response to glucose
level changes. The detailed description of design, manufac-
ture, mechanism, and in vitro and in vivo performance of
the BioTex (Houston, TX) FAS was reported in two earlier
papers.28,30 In brief, one end of a 175-m-diameter multi-
mode polymer optical fiber was mechanically spliced to two
105-m-diameter silica optical fibers. The proximal ends of
the two smaller fibers were terminated with SMA-905 con-
nectors. One of the fibers was attached to a collimated laser
diode at 650 nm (Thorlabs, Newton, NJ), and the other was
attached to a miniature spectrometer (USB-2000, Ocean Op-
tics, Dunedin, FL). An individual hollow dialysis fiber (di-
ameter 210 m, length 5 mm) was carefully pushed onto the
end of the 175-m-diameter polymer optical fiber. The hol-
low fiber was then filled with sensing suspension by aspi-
ration and sealed with cyanoacrylate (Loctite®, Henkel
Corp., Düsseldorf, Germany) at both ends. An additional
bonding sleeve made of thin-walled polyimide tubing was
then attached over the junction between the optical fiber and
sensor fiber. The entire fiber sensor assembly fit inside of a
20-gauge hypodermic needle for insertion into skin tissue.
FAS implantation and testing in hairless rats
Male hairless rats without diabetes weighing approxi-
mately 300 g were anesthetized and maintained with isoflu-
rane by inhalation. The preclinical animal study included 25
hairless rats. Before implantation the sensors were bathed for
10 min in sterile saline. A 20-gauge hypodermic needle con-
taining the fiber-coupled FAS was inserted intradermally at
a shallow angle on the dorsum of the animal 1–4 cm from
the midline. After the sensor and needle were pushed ap-
proximately 2 cm into the skin, the hypodermic needle was
entirely withdrawn, leaving the sensor exposed to skin tis-
sue. A fresh Medtronic Minimed CGMS SOF-Sensor™
(Medtronic Diabetes) was co-implanted in each rat in the
subcutaneous tissue of the dorsal thoracic region for com-
parison purposes according to the manufacturer’s instruc-
tions. After 1 h of baseline acquisition by the implanted FAS,
DUTT-BALLERSTADT ET AL.454
TABLE 1. STRATIFIED CLARKE ERROR GRID OF FAS READINGS IN THE SMALL ANIMAL MODEL
Clarke Error Grid zone (%)
Glucose range (mg/dL) Number of paired readings A B A B C D E
80 10 70 70 0 0 30 0
81–120 50 100 50 50 0 0 0
121–240 166 98 62 36 2 1 0
241 33 91 58 33 6 3 0
Total 40–400 259 96 59 37 2 2 0
“regular” fast-acting insulin (0.5 unit/kg) was administered
subcutaneously. After another hour, a bolus glucose injec-
tion (50% dextrose, 3 mL/kg) was given intraperitoneally.
Serial blood samples from the tail vein were taken approxi-
mately every 15–20 min over the 2–3-h period and measured
using a FreeStyle glucometer (Abbott Diabetes Care). At the
end of the experiment, blood glucose was normalized by in-
traperitoneal bolus injection of 50% dextrose, and the animal
was returned to its cage. In a total of 25 animals, the proto-
col was repeated 2 days later. When a 3-day experiment was
performed, the sensor sites were protected with bandages
and tape to prevent sensor rupture or removal by the ani-
mal. On day 3, an identical glucose manipulation was per-
formed. At the end of the experiment, the FASs were care-
fully removed from the sites, and the animal was returned
to its cage. Overall, chronic experiments over more than 2
days were difficult to perform because of excessive move-
ment of the rats while awake, even when the sensors were
protected by bandages and tape around the animal’s body.
FAS implantation and glucose testing in pigs
A total of four juvenile farm pigs (weighing 25–30 kg)
without diabetes were included in this study. Pigs were
pre-anesthetized with ketamine (1 mL/50 kg), scrubbed
thoroughly on the dorsal site with povidone and warm wa-
ter, and then maintained with 1–4% isoflurane by mask. Af-
ter saline rinse for 10 min, up to three fiber-coupled FASs
were inserted in the upper back using a hypodermic nee-
dle as described in experiments with the small animal
model (see above). We implanted several sensors for re-
dundancy and to maximize our chances that sensors would
remain in the animals. Sensors were secured with tape and
covered with sterile Tegaderm™ (3M, St. Paul, MN) ban-
dages and masking tape when not in use. At the time of the
experiment, a Medtronic Minimed sensor was co-implanted
in a similar area in each pig for comparison purposes. To
modulate blood glucose, 25 mL of 50% dextrose was in-
fused intravenously through an ear vein. Blood samples
were collected from a vein in the contralateral ear. On day
3, all sensors remained implanted, and one was chosen for
monitoring. A new CGMS sensor was implanted, and the
glucose challenge test was repeated. After the experiments
on day 3, sensors were carefully removed, and the pig was
returned to its cage.
All animal studies were carried out at an Association for
the Assessment and Accreditation of Laboratory Animal
Care-accredited facility and in accordance with an Institu-
tional Animal Care and Use Committee-approved protocol.
Signal analysis
Determination of glucose concentrations with FAS was
performed retrospectively after simultaneously measuring
both the glucose-sensitive emission at 675 nm and the refer-
ence dye emission at 780 nm at rate of five to 10 measure-
ments per hour, which were then stored on a notebook com-
puter. After each experiment, the 675 nm and 780 nm signals
were analyzed for drift, most likely due to photobleaching,
and normalized accordingly. Normalization was performed
by determining the slope of the change in emission at 675
nm and 780 nm, respectively, before and after the experi-
ment (usually 2–3 h) at the same glucose concentration. Then
the corrected ratio signal was calculated. For calibration pur-
poses, the initial baseline period during which no changes
in blood glucose were measured (usually 50–60 min) was
used for one-point blood glucose calibration based on paired
readings of FAS and venous blood glucose levels. If the one-
point glucose calibration reading was not in agreement with
the in vitro calibration curve for FAS obtained 12–24 h be-
fore the in vivo experiment, the calibration curve was shifted
along the y-axis accordingly with the slope remaining con-
stant.
Calibration of the Medtronic Minimed CGMS was inter-
nally performed by the CGMS monitor unit (model MMT-
7310, version 3.0B). These data were extracted from the unit
using Minimed Solutions software and fed into Windows XP
Excel™ (Microsoft, Redmond, WA) by a Matlab® application
(version 4.2c.1, The Mathworks, Inc., Natick, MA). No fur-
ther calibration adjustments were performed.
PRECLINICAL IN VIVO STUDY OF FAS 455
TABLE 2. PERCENTAGE OF FAS READINGS WITHIN 20%, 30%, AND 40% OF
GLUCOMETER READINGS BY RANGE IN THE SMALL ANIMAL MODEL
Glucose range (mg/dL) Number of paired readings 20% 30% 40%
80 21 48 61 71
81–120 45 58 71 82
121–240 162 61 82 90
241 31 52 71 87
FIG. 1. Glucose values obtained from small animal model
experiments in hairless rats plotted in the Clarke Error Grid.
Data and safety analysis
Retrospectively analyzed and corrected data from the Bio-
tex FAS and the Medtronic Minimed CGMS were pooled for
each animal model and analyzed. For each animal model, a
correlation coefficient (R), a mean absolute relative differ-
ence (MARD), and a root mean standard error (RMSE) was
calculated by pooling all sensor data. A one-sided Student’s
t test was performed to test for differences in MARD between
the BioTex FAS and Medtronic Minimed CGMS. In addition,
for the BioTex FAS a Clarke Error Grid was constructed, and
the percentage of points falling into the A, B, C, D, and E re-
gions of the Clarke Error Grid were then calculated. For sen-
sor evaluation, we chose the “classical” error grid analysis
in combination with MARD rather than the continuous glu-
cose error grid analysis both developed by Clarke et al.,32,33
because it was less labor intensive and time-consuming and
the results were easier to interpret, as was also discussed and
recommended in detail by Wentholt et al.34 To evaluate the
agreement between glucose concentrations of the FAS sen-
sor reading and venous blood glucose measurement ob-
tained by fingerprick, the differences between the individ-
ual data pairs were plotted against their means as described
by Bland and Altman.35 The mean difference between both
types of glucose measurements and the limits of agreement
(including 95% of the differences) were calculated.
We assessed safety in two ways by (1) error grid analysis
and (2) visual examination of the skin site during sensor op-
eration and after sensor removal.
Results
In vivo response of fiber-coupled FAS in
small animal model
Acute in vivo performance studies with the fiber-coupled
FAS were performed in rats. The MARD of the FAS com-
pared to the Abbott FreeStyle glucometer was 18.5% (259
paired samples), and the average correlation coefficient (R)
was 0.62. A RMSE of 19.8 mg/dL was calculated. The Clarke
Error Grid indicated that of the total of 259 readings, 98.5%
of the readings were in the clinically acceptable A and B re-
gions, and 1.5% were in C and D regions, respectively (Fig.
1). Table 1 shows the percentage of points falling within each
zone, stratified according to the range of glucose concentra-
tions. Within the glucose range from 40 to 350 mg/dL, 55%
of readings were within 20% and 71% within 30%. Table 2
shows that almost at least 50% of paired readings were
within 20% of the glucometer readings. Figure 2 shows the
differences between capillary blood glucose and FAS values
plotted against their averages (Bland-Altman plot).35 The
mean of the differences was 1.1 mg/dL, and their standard
deviation was 25.8 mg/dL, which would indicate a 95% con-
fidence interval for agreement of 51 mg/dL. The difference
appears to be a function of concentration. The data in Fig-
ure 2 indicate a 1.1 mg/dL mean difference in FAS and
blood glucose readings, giving us confidence that we are not
introducing a bias into the FAS response.
For an animal implanted with the FAS for 3 days, the re-
sponse on day 1 was comparable to the response on day 3
(Fig. 3). The lower correlation coefficient (R) of 0.62 on day
3 versus 0.92 on day 1 was heavily influenced by the faster
rate of change in blood glucose compared to the response by
the sensor as illustrated by the data point (encircled) in Fig-
ure 3. The sensor was unable to track this artificially high
rate of blood glucose change (18 mg/dL/min). However,
several studies have shown that rates of blood glucose
changes larger than 3 mg/dL/min occur with a probability
of less than 1%.36 The corresponding correlation coefficient
(R) calculated without the influence of this single point was
0.95 for day 3, indicating excellent stability of sensor re-
sponse.
During the entire study, no major acute effects were found
as the result of sensor implantation. Minor adverse effects
that were observed included bleeding at the insertion site,
edema (swelling), and erythema (redness). All were mild and
required no treatment.
DUTT-BALLERSTADT ET AL.456
FIG. 2. Bland-Altman plots for FAS evaluated in the small
animal model. The x-axis shows the average of blood and
sensor glucose measurements, and the y-axis represents the
difference between sensor and concomitant blood glucose
(BG) measurements. The solid line is drawn at the mean dif-
ference (-1.1 mg/dL); dotted lines are drawn at the mean dif-
ference 1.96 times the SD of the differences (51 mg/dL).
FIG. 3. Comparison of stability of FAS response over a 3-
day period in a hairless rat. The sensor was implanted on
Day 1 in an anesthetized rat, and its acute response to vari-
ation in blood glucose due to injection of insulin and dex-
trose was measured (open squares). The animal was allowed
to freely move around on Day 2. On Day 3, the sensor re-
sponse was tested again (solid squares). R denotes correla-
tion coefficient between blood glucose values and FAS sig-
nal. The encircled data point was the result of a much faster
rise of glucose level (18 mg/dL) in blood than in interstitial
fluid measured by sensor (see text for further details).
In vivo performance of FAS in large animal model
Acute in vivo performance studies with the fiber-coupled
FAS were performed in pigs. The retrospective MARD of all
sensor data including acute experiments over a 4-h period
compared to the Abbott FreeStyle glucometer was 15% (37
paired readings), and the average R value was 0.61. The
Clarke Error Grid indicated that from the total of 37 read-
ings, 97% of the readings were in the clinically acceptable A
and B regions, and 3% were in the D region (Fig. 4). Table 3
shows the percentage of points falling within each zone,
stratified according to the range of glucose concentrations.
Within the glucose range from 40 to 350 mg/dL, 56% of read-
ings were within 20% and 83% within 30%.
Figure 5 shows the results of one sensor during a 3-day
implantation experiment. The correlation coefficient (R) val-
ues of 0.89 and 0.88 for Day 1 and Day 3, respectively, dem-
onstrate good stability of the sensor response over a period
of 3 days. Sensors were generally well tolerated by the ani-
mals. Visual assessment of one implantation site after ex-
plantation on Day 3 showed only slight redness at the in-
sertion site, possibly the result of minor bleeding after FAS
insertion on day 1, which subsided after a few days. No gross
evidence of inflammation, irritation, or infection was noticed
based on visual observation at any of the implantation sites.
In vivo performance comparison between FAS and CGMS
A summary of the performance of the BioTex FAS and the
Medtronic Minimed CGMS is shown in Table 4. Overall, both
sensors performed quite similar in both animal models. We
note, however, that the BioTex FAS showed a slightly better
performance in hairless rats compared to the CGMS, illus-
trated by a slightly higher correlation coefficient (0.62 vs.
0.55) and a lower MARD value (18.5% vs. 25.4%, P 0.005).
Conclusions
The purpose of this study was to assess feasibility and
safety of the fiber-coupled FAS in small and large animal
models. While the advantage of hairless rats rested in the
ease of handling, the benefits of using pigs for sensor test-
ing are similarities to skin anatomy and general physiology
with humans. Although pigs also tended to engage in be-
haviors that could result in dislodging or damaging the sen-
sors (e.g., rolling, scratching, rubbing), we experienced a re-
duced failure rate in the large animal model.
In both animal models more than 97% of the paired read-
ings were within the clinically relevant zone A and B regions.
Sensor accuracies expressed as RMSE and MARD in both an-
imal models were comparable (16.3 and 19.8 mg/dL and
18.5% and 15.9%, respectively). Moreover, one of the most
significant results of the acute in vivo study is the excellent
performance of our prototype FAS in comparison to the clin-
ically applied CGMS from Medtronic Minimed. In terms of
accuracy with regard to blood glucose level, no major dif-
ferences were found. Values for the correlation coefficient
and MARD were slightly better for the BioTex FAS than for
the CGMS. For the Guardian RT system studied in humans
over a 72-h period followed by retrospective data analysis,
Medtronic Minimed reported a MARD of 17.32%.37 Dex-
Com’s STS sensor performed similarly in humans over a 72-
h period with a reported MARD of 20.3% based on retro-
spective analysis.38 Although these MARD values, which are
very similar to the BioTex FAS, may not be directly compa-
rable because of differing experimental protocols, the BioTex
FAS performance appears very promising and suggests that
consideration of initial human trials is appropriate.
During 3-day sensor testing in both animal models, the
FAS sensor response was quite acceptable and only ham-
pered by the challenge of keeping the sensor tightly fastened
when the highly agile rats were awake. In both cases, the
sensor response after 3 days was still measurable. The de-
crease in response was probably caused by mechanically
based fatigue of sensor integrity. We are currently in the pro-
cess of significantly improving the mechanical robustness of
the FAS, which would be imperative when employed in hu-
man trials.
Time delay of the FAS in response to blood glucose mod-
ulators was modest and did not exceed more than 10 min,
even in pigs. We observed very short time lags in rats. We
noticed quite often in pigs longer lag times immediately
PRECLINICAL IN VIVO STUDY OF FAS 457
TABLE 3. PERCENTAGE OF FAS READINGS WITHIN 20%, 30%, AND 40% OF
GLUCOMETER READINGS BY RANGE IN THE LARGE ANIMAL MODEL
Glucose range (mg/dL) Number of paired readings 20% 30% 40%
80 6 67 83 100
81–120 13 92 92 100
121–240 16 68 75 87
241 1 0 0 100
FIG. 4. Glucose values obtained from large animal model
experiments in pigs plotted in the Clarke Error Grid.
...
