Anal. Chem.
1994,66, 3131-3138
Design, Characterization, and One-Point in Vivo Calibration
of a Subcutaneously Implanted Glucose Electrode
Eiisabeth Csoreg1,t Chris P. Quinn,* David W. Schmldtke,* Sten-Eric Lindquist,g Michael V. Pishko,*
Ling Ye,* Ioanis Katakis,* Jeffrey A. Hubbell,* and Adam Heiier’**
Department of Chemical Engineering, University of Texas at Austin, Austin, Texas, 787 12- 1062, Department
of Analytical Chemistry, University of Lund, P.O. Box, 124, S-221 00 Lund, Sweden, and Department of
Physical Chemistry, University of Uppsala, P.O. Box 532, S-751 21 Uppsala, Sweden
A 0.29-mm-diameter flexible electrode designed for subcutane-
ous in vivo amperometric monitoring of glucose is described.
The electrode was designed to allow “one-point” in vivo
calibration, Le., to have zero output current at zero glucose
concentration, even in the presence of other electroreactive
species of serum or blood. A valid zero point, along with a
measurement of the glucose concentration in a withdrawn
sample of blood at which the current is known, defined the
sensitivity in the linear response range. The electrode was
four-layered, with the layers serially deposited within a 0.125-
mm recess upon the tip of a polyimide-insulated 0.25-mm gold
wire. The recessed structure reduced the sensitivity to
movement and allowed, through control of the depth of the
recess, control of the transport of glucose and thus the range
of linearity. The recess contained the four polymeric layers,
with a total mass less than 5 pg and no leachable components.
The bottom glucose concentration-to-current transducing layer
consisted of the enzyme “wiring” redox polymer poly-
[(~inylimidazole)Os(bipyridine)2Cl]+~~+, complexed with re-
combinant glucose oxidase and cross-linked with poly(ethy1ene
glycol) diglycidyl ether, to form an electron-conducting
hydrogel. The layer was overcoated with an electrically
insulating layer of polyaziridine-cross-linked poly (allylamine),
on which an immobilized interference-eliminating horseradish
peroxidase based film was deposited. An outer biocompatible
layer was formed by photo-cross-linking derivatized poly-
(ethylene oxide). The current output of a typical electrode at
10 mM glucose and at 37 OC was 35 nA, the apparent K, was
20 mM, and the 10-90% response time was -1 min. The
sensitivity varied only by f5% over a 7 2 4 test period. The
electrode tracked the blood glucose concentration in a rat model
before, during, and following intraperitoneal glucose infusion.
Two failure modes were observed. The first, deactivation of
hydrogen peroxide-producing lactate oxidase in the interfer-
ence-eliminating layer, resulted in inadequate preoxidation of
interferants. The second was an abrupt drop in the sensitivity
of implanted electrodes, -7 h after implantation. The failed
electrodes promptly regained their sensitivity in buffer.
In response to the need for frequent or continuous in vivo
monitoring of glucose in diabetics, particularly in brittle
diabetics, a range of possible in vivo glucose electrodes have
been Thedesired characteristics of theseelectrodes
+ University of Lund.
t University of Texas at Austin.
I University of Uppsala.
0003-2700/94/0366-3131$04.50/0
0 1994 American Chemical Society
include (a) safety, (b) clinical accuracy and reliability, (c)
feasibility of in vivo recalibration, (d) stability for at least one
hospital shift of 8 h, (e) small size, ( f ) ease of insertion and
removal, and (g) a sufficiently fast response to allow timely
intervention.
Keys to safety are absence of leachable components,
biocompatibility, and limiting of the potentially hazardous
foreign matter introduced into the body to an amount that is
inconsequential in a worst case failure. The clinical accuracy
(1) Armour, J. C.; Lucisano, J. Y.; McKean, B. D.; Gough, D. A. Diabetes 1990,
39, 1519-1526.
(2)Shichiri, M.; Kawamori, R.; Yamasaki, Y. Merhods Enzymol. 1988, 137,
326-334.
(3)Shichiri, M.; Yamasaki, Y.; Nao, K.; Sekiya, M.; Ueda, N. In Zmplanrable
Glucose Sensors-The Srare of rhe Art; Pfeiffer, E. F., Reaven, G. M., Eds.;
Supplement Series/Hormone and Metabolic Research; (Reisensburg, 1987)
Thieme Medical Publishers, Inc.: New York, 1988;pp 17-20.
(4) Moatti-Sirat, D.;Capron, F.; Poitout, V.; Reach, G.; Bindra, D. S.; Zhang,
Y.; Wilson, G. S.; Thevenot, D. R. Diaberologia 1992,35,224230.
(5) Karube, 1.; Yokoyama, K.; Tamiya, E. Biosens. Bioelecrron. 1993,8, 219-
228.
(6) Brilckel, J.; Kerner, W.; Zier, H.;Steinbach, G.; Pfeiffer, R. F. Klin.
Wochenschr. 1989,67,491495.
(7)Gernet, S.; Koudelka, M.; Rwij de, N. F. S e w . Acruarors 1989,17,537-540.
(8)Pickup, J. C.; Claremont, D. J.; In Zmplanrable Glucose Sensors-The Srare
ojthe Art;Pfeiffer, E. F., Reaven, G. M., Eds.;Supplement Series/Hormone
and Metabolic Research; (Reisensburg, 1987)Thieme Medical Publishers,
Inc.: New York, 1988;pp 3636.
(9)Yokoyama, K.; Lee, S. M.; Tamiya, E.; Karube, I.; Nakajima, K.; Uchiyama,
S.; Suzuki, S.; Akiyama M.; Masuda, Y. Anal. Chim. Acra 1992,263,101-
110.
(10)Tamiya, E.; Karube, I. Sens. Acruarors 1988,15, 199-207.
(11) Bindra, D.S.; Zhang, Y.; Wilson, G. S.; Sternberg, R.; Thevenot, D. R.;
Moatti, D.; Reach, G. Anal. Chem. 1991,63,1692-1696.
(12)Cronenberg, C.; Grocn van, B.; Beer de, D.; Heuvel van den, H. Anal. Chim.
Acta 1991, 242,275-278.
(13) Johnson, K. W.; Mastrototaro, J. J.; Howey, D. C.; Brunella, R. L.; Burden
Brady, P. L.; Bryan, N. A.; Andrew, C. C.; Rowe, H. M.; Allen, D. J.; Noffke,
B. W.; McMahan, W. C.; Morff, R. J.; Lipson, D.; Nevin, R. S . Biosens.
Bioelectron. 1992,7, 709-714.
(14)Koudelka, M.; Gernet, S.;Rwij de, N. F. Sens. Acruarors 1989,18,157-165.
(15)Shaw, G. W.; Claremont, D. J.; Pickup, J. C. Biosens. Bioelectron. 1991, 6,
401406.
(16)Pickup, J. C.; Shaw, G. W.; Claremont, D. J. Diaberologia 1989,32,213-217.
(17)Gregg, B.; Heller, A. Anal. Chem. 1990,62,258-263.
(18) Gregg, 8.; Heller, A. J. Phys. Chem. 1991, 95,5976-5980.
(19)A h , T.; Lau, Y. Y.; Ewing, A. G. AMI. Chem. 1992,64,2160-2163.
(20)Bartlett, P. N.; Caruana, D. J. Analyst 1992,117, 1287-1292.
(21)Chen, C.-Y.; Gotoh, M.; Makino, H.;Su, Y.-C.; Tamiya, E.; Karube, I. Anal.
Chim. Acra 1992,265, 5-14.
(22)Kawagoe, J. L.; Niehaus, D. E.; Wightman, R. M. Anal. Chem. 1991, 63,
2961-2965.
(23)Murakami, T.; Nakamoto, S.; Kimura, J.; Kuriyama, T.; Karube, I. Anal.
Lett. 1986,19,1973-1986.
(24)Pishko, M. V.;Michael, A. C.; Heller, A. Anal. Chem. 1991,63,2268-2272.
(25)Tamiya, E.;Sugiura, Y.; Akiyama, A.; Karube, I. Ann. N.Y. Acad. Sci. 1990,
613. 396-400.
(26)Turner, R. F. B.; Harrison, D. J.; Rajotte, R. V.; Bakes, H. P. Sens. Acruarors
B 1990,21, 561-564.
(27)Urban, G.; Jobst, G.; Keplinger, F.; Aschauer, E.; Tilado, 0.; Fasching, R.;
Kohl, F. Biosens. Bioelectron. 1992,7,733-739.
(28)Velho, G.; Froguel, P.; ThCenot, D. R.; Reach, G. Diabetes Nurr. Merab.
1988,I , 227-234.
Analytical Chemistry, Vol. 66, No. 19, October 1, 1994 3131
 must be such that even when the readings are least accurate,
the clinical decisions based on these be still correct. Feasibility
of prompt confirmation of proper functioning of the sensors
and of periodic in vivo recalibration is of the essence if a
physician is to allow the life of a patient to depend on the
sensor. This one-point calibration, relying on the signal at
zero glucose concentration being zero and measuring the blood
glucose concentration at one point in time, along with the
signal, is essential, but has heretofore been elusive. The
sensitivity must be sufficiently stable for the frequency of the
required in vivo recalibration to not be excessive. The sensor
must be small enough to be introduced and removed with
minimal discomfort to the patient and for minimal tissue
damage. It is preferred that the sensor be subcutaneous and
that it be inserted and removed by the patient or by staff in
a physician’s office. Finally, its response time must be fast
enough so that the corrective measures, when needed, will be
timely.
In response to some of these needs, needle
typel.2A6.8.11-13,16,19,22,23,25-27,29-32and other subcutaneous am-
perometric sensors were considered. The majority of these
utilized platinum, platinum-iridium, or platinum black to
electrooxidize H202 generated by the glucose oxidase (GOx)
catalyzed reaction of glucose and oxygen.2,5q6J1-13~16~19~26,27*30
In these sensors, the GOx was usually in large excess and
immobilized, often by cross-linking with albumin and
g l ~ t a r a l d e h y d e . ~ ~ To
- ~ ~exclude
? ~ ~ , ~electrooxidizable
~ inter-
ferants, membranes of cellulose acetate and sulfonated
polymers including Nafion were ~sed.~~,~6,33,34 Particular
attention was paid to the exclusion of the most common
electrooxidizable interferants: ascorbate, urate, and acetami-
nophen. Also, to cope with the interferants, two-electrode
differential measurements were used, one electrode being
sensitive to glucose and electrooxidizable interferants and the
other only to i n t e r f e r a n t ~ .A~ ~strategy for overcoming the
problem of interferants, applied also in this work, involved
their p r e ~ x i d a t i o n . ~To
~ . ~make
~ the electrodes more bio-
compatible, hydrophilic polyurethanes,2.6*11.37,38 poly(viny1
a l ~ o h o l ) and
, ~ ~ o ~ ~ H Emembranes M A ~ ~have been used.
Several researchers tested GOx-based glucose sensors in
vivo and obtained acceptable results in rats,4*30,37*42 rabbit~,13?~3
d 0 g ~ , 2 , pigs,8-41142
~ + ~ ~ , ~sheep,6and h u m a n ~ . l ~These
, ~ ~ studies
>~
validated the subcutaneous tissue as an acceptable glucose-
(29) Wang, J.; Angnes, L. Anal. Chem. 1992,64, 456-459.
(30) Koudelka, M.; Rohner, J. F.; Terrettaz, J.; Bobbioni, H . E.; Rooij de, N . F.;
Jeanrenaud, B. Biosens. Bioelectron. 1991, 6, 31-36.
(31) Velho, G.; Froguel, P.; Sternberg, R.; Thevenot, D. R.; Reach, G . Diabetes
1989, 38, 164-171.
(32) Pishko, M. V.; Katakis, I.; Lindquist, S.-E.; Heller, A. Mol. Cryst. Liq. Crysr.
1990, 190, 221-249.
(33) Shiono, S.;Hanazato, Y.; Nakako, M. Anal. Sei. 1986, 2, 517-521.
(34) Gorton, L.; Karan, H. I.; Hale, P. D.; Inagaki, T.; Okamoto, Y.; Skotheim,
T.A. Anal. Chim. Acta 1990, 28, 23-30.
(35) Maidan, R.; Heller, A. Anal. Chem. 1992, 64, 2889-2896.
(36) Maidan, R.; Heller, A. J . Am. Chem. SOC.1991, 113, 9003-9004.
(37) Koudelka, M.; Rohner, J. F.; Terrettaz, J.; Bobbioni, H. E.; Rooij de, N . F.;
Jeanrenaud, B. Biomed. Biochim. Acta 1989.48.953-956.
(38) Velho, G.; Froguel, P.; Thevenot, D. R.; Reach, G. Biomed. Biochim. Acta
1989, 48, 951-964.
(39) Rebrin, K.; Fischer, U.; Woedtke, T.; Abel, P.; Brunstein, E. Diabetologia
1989, 32, 573-576.
(40) Fischer, U.; Ertle, R.; Abel, P.; Rebrin, K.; Brunstein, E.; Dorschevon, H. H.;
Freyse. E. J. Diabetologia 1987, 30, 940-945.
(41) A h , T.; Lau, Y. Y.; Ewing, A. G. J. Am. Chem. Soc. 1991,113,7421-7423.
(42) Claremont. D. J.; Pickup, J. C.; Sambrook, I. E. Life Support Sysr. 1986,
369-370.
(43) Ito, K.; Ikcda, S.;Asai, K.; Naruse, H.; Ohkura, K.; Ichihashi, H.; Kamei, H.;
Kondo, T. ACS Symp. Ser. 1986, 309, 373-382.
3132 AnalyticalChemistry, Vol. 66, No. 19, October 1, 1994
sensing site. Good correlation was observed between intra-
vascular and subcutaneous glucose concentration^.^^,^^ They
also demonstrated the need for in vivo sensor ~ a l i b r a t i o n . ~ ~
Another approach to in vivo glucose monitoring was based on
coupling subcutaneous microdialysis with electrochemical
detection.46 To control and adjust the linear response range,
electrodes have been made glucose-diffusion limited, usually
through glucose transport limiting membranes.
Diffusional mediators, through which the 02 partial
pressure dependence of the signals is reduced, are leached
from sensors. Such leaching introduces an unwanted chemical
into the body and also leads to loss in sensitivity, particularly
in small sensors. In microsensors, in which outward diffusion
of the mediator is radia1,47.48the decline in sensitivity is rapid.
This problem has been overcome in “wired” enzyme electrodes,
i.e., electrodes made by connecting enzymes to electrodes
through cross-linked electron-conducting redox hydrogels
Glucose oxidase has been wired with
polyelectrolytes having electron relaying [Os(bpy)2C1]+/2+
redox centers in their backbones. Hydrogels were formed
upon cross-linking the enzyme and its wire on electrodes. These
electrodes had high current densities and operated at a potential
of 0.3 V vs SCE.17*50The electrooxidizable interferants were
eliminated through peroxidase-catalyzed preoxidation in a
second, nonwired, hydrogen peroxide generating layer on the
wired enzyme ele~trode.~5.36
In this paper, we report on a 0.29-mm recessed gold wire
electrode for subcutaneous in vivo glucose monitoring, ap-
proaching in its performance all of the above listed require-
ments, including in vivo one-point calibration. The electrode
was constructed by depositing four layers into a recess. The
recess was formed by etching away gold from an insulated
gold wire. The polymer layers were protected within the recess
against mechanical damage. The recess and its polymer layers
also reduced the transport of glucose to the gold contacting
sensing layer. By limiting the glucose flux, the desired linear
response range, spanning the clinically relevant glucose
concentration range, was obtained. The sensor had no
-
leachable components, and its four cross-linked polymer layers
contained only 5 pg of immobilized material and only a few
nanograms of polymer-bound osmium. Preoxidation of the
interferants in one of the four layers made possible one-point
in vivo calibration of the sensor.
EXPERIMENTAL SECTION
Materials. The electrodes were made of a polyimide-
insulated 250-pm-diameter gold wire, 290-pm outer diameter
(0.d.) (California Fine Wire Co., Grover City, CA). Heat-
shrinkable tubing (RNF 100 3/64 in. BK and ‘ / I 6 in. BK,
Thermofit, Raychem, Menlo Park, CA) and a two-component
(44) Shichiri, M.; Asakawa, N.; Yamasaki, Y.; Kawamori, R.; Abe, H. Diabetes
Care 1986, 9, 298-301.
(45) Kerner, W.; Zier. H.; Steinbach, G.; Briickel, J.; Pfeiffer, E. F.; Wiess, T.;
Camman, K.; Plack, H. In Implantable Glucose Sensot-The State of the
Art; Pfeiffer, E. M., Raven, G. M., Eds.; Supplement Serics/Hormone and
Metabolic Rcaearch; (Reiscnsburg, 1987) Thieme Medical Publishers Inc.:
New York, 1988; pp 8-13.
(46) Moscone, D.; Pasini, M.; Mascini, M. Talanta 1992, 39, 1039-1044.
(47) Aoki, K. Electroanalysis 1993, 5, 627639.
(48) Fleischmann, M.; Pons, S.;Rolison, D. R.;Schmidt, P. P. Ultramicroelectrodes;
Publishers Press: Morganton, NC, 1987.
(49) Heller, A. J . Phys. Chem. 1992, 96, 3579-3587.
(50) Ohara, T.; Rajagopalan, R.; Heller, A. Anal. Chem. 1993, 65, 24-29.
silver epoxy (Epo-tek H20E; Epoxy Tech. Inc., Billerica, MA)
were used for electrode preparation.
The glucose-sensing layer was made by cross-linking a
genetically engineered glucose oxidase (rGOx) (35% purity,
Chiron Corp., Emeryville, CA) with a polymer derived of
poly(vinylimidazo1e) (PVI), made by complexing part of the
imidazoles to [O~(bpy)2Cl]+/~+. The resulting redox polymer,
termed PVI-Os, was synthesized according to a previously
published protocol.50 Poly(ethy1ene glycol) diglycidyl ether
400 (PEGDGE; Polysciences, Warrington, PA) was used as
the cross-linker.
The barrier layer between the sensing and interference-
eliminating layers was made of poly(ally1amine) (PAL;
Polysciences) cross-linked with a polyfunctional aziridine
(PAZ) (XAMA-7; Virginia Chemicals, Portsmouth, VA).
The interference-eliminating layer was prepared by co-
immobilizing horseradishperoxidase (HRP) type VI (Catalog
no. P-8375,3 10 units/mg, denoted herein as HRP-VI; Sigma,
St. Louis, MO) and HRP for immunological assay (No.
8 14407, minimum 1000 units/mg, denoted HRP-BM; Boeh-
ringer-Mannheim, Indianapolis, IN) with lactate oxidase from
Pediococcus sp. (Catalog No. 1361, 40 units/mg, denoted
LOX; Genzyme, Cambridge, MA) and a recombinant mi-
crobial source (Catalog No. 138l denoted rLOx, Genzyme).
Coimmobilization was performed using sodium periodate
(Catalog No. S-1147, Sigma).35
The biocompatible layer was made of 10% aqueous poly-
(ethylene oxide) tetraacrylate (PEO-TA). To form the photo-
cross-linkable polymer, PEO was acrylated by reaction with
acryloyl chloride. The 18 500 g/mol PEO (Polysciences) is
a tetrahydroxylated compound by virtue of two hydroxyl
groups on a bisphenol A bisepoxide that linked two a,
o-hydroxy-terminated 9000 g/mol PEO units. Acryloyl
chloride (Aldrich, Milwaukee, WI) in a 2-5 M excess was
used to acrylate the polymer (10% w/v PEO in benzene).
Triethylamine (Mallinkrodt, Paris, KY) was used as a proton
acceptor equimolar with the acryloyl chloride.51
Other chemicals used were ascorbic acid, uric acid,
4-acetaminophenol,L-(+)-lactic acid, and hydrogen peroxide
(30%), all from Sigma. All chemicals were used as received.
Solutions(if not otherwisespecified) were made with distilled,
deionized water. Glucose monitoring was performed in buffer,
in bovine serum (Sigma, Catalog No. S-6648) containing
antibiotic-antimycotic solution (Sigma, Catalog No. A-8909)
at 37 "C, and in rats.
Instrumentation. In making the recessed gold electrodes,
a potentiostat-galvanostat (PAR Model 173, Princeton
Applied Research, Princeton,NJ), operated in a galvanostatic
mode, and a sonicator (Fisher Scientific, Pittsburgh, PA) were
used. Cyclic voltammograms were recorded with a poten-
tiostat (PAR Model 273A) and a conventional electrochemical
cell having a Pt wire counter and a SCE reference electrode
and were evaluated with PAR 270 software. Glucose signals
were monitored with a bipotentiostat (Biometra, EP 30,
Gothingen, Germany) and a two channel strip-chart recorder
(Kipp & Zonen, Bohemia, NY). The recessed electrodes were
coated under a microscope (Bausch & Lomb) using a
micromanipulator (Narishige,Seacliff, NY). The micropipets
(51) Pathak, C. P.; Sawhney, A. S.;Hubbell, J. A. J. Am. Chem. Soc. 1993,115,
83 11-83 12.
I
Poly imide
1. Sensing Layer
2. Insulating Layer
3. Interference Eliminating Layer
4. Biocompatibie Layer
Flgure 1. (A) Photomicrograph of a recessed, uncoated electrode,
lying flat on a surface. Use of a light microscope afforded a view
through the polyimide insulation but not through the gold wire. (B)
Schematic drawing of the glucose sensor.
were pulled with a micropipet puller (Narishige). Tempera-
ture was controlled with an isothermal circulator (Fisher
Scientific).
Electrode Preparation. Lengths (5 cm) of polyimide-
insulated gold wire were cut with a sharp razor blade.
Electrical contact was made at one end with silver epoxy to
an insulated stainless steel wire, and the junction was covered
with insulating heat-shrinkable tubing. The recess-forming
electrochemical etching process was carried out in 10 mL of
3 M potassium cyanide, with the gold wire as the working
electrode and a platinum or gold wire as the counter electrode.
The wires were placed in contact with the bottom of the beaker,
all electrodes being equidistant from the counter electrode.
The beaker was sonicated during the etching procedure. The
ends of the Au wires were bent upward, so that agitation by
the sonicator caused the oxygen bubbles formed during the
etching process to rise and escape. The electrodes were then
thoroughly washed and immersed in water for 30 min. The
recession procedure is highly reproducible; a deviation off 10
pm was found (using an objective micrometer) for a batch of
30 recessed electrodes. Before coating, the electrodes were
examined under a microscope for flatness of the gold surface
and correct depth. Figure 1A shows a side view of a recessed,
uncoated electrode. The recessed gold surfaces were coated
by filling of the cavities with aqueous solutions containing the
cross-linkable components of the different layers, and their
cross-linkers. The solutions were introduced under a micro-
scope with a micropipet (connected to a microsyringe by
polyethylene tubing and shrink tubing), using a microman-
ipulator. After application of each of the individual layers,
the electrodes were cured overnight at room temperature in
air.
Analytical Chemistry, Vol. 66, No. 19, October 1, 1994 3133
 Electrode Structure. The electrodes were prepared by
sequentially depositing four layers: the sensing layer (l), the
insulating layer (2), the interference-eliminating layer (3),
and the biocompatible layer (4). Figure 1B is a schematic
drawing of the sensor. The thin polymer layers were well
protected by containment within the polyimide sleeve.
The sensing layer was made by “wiring” rGOx to the gold
electrode through a redox hydrogel to which the enzyme was
covalently bound. The electrodes were prepared as follows:
10 mg/mL solutions were made from (1) the PVI-Os redox
polymer in water, (2) the cross-linker, PEGDGE, in water,
and (3) the enzyme, rGOx, in a 10 mM HEPES solution
adjusted topH8.15. Theredox hydrogelwas formed bymixing
the three solutions so that the final composition (by weight)
was 52% redox polymer, 35% enzyme, and 13% cross-linker.
The insulating layer prevented electrical contact between
the redox hydrogel and the interference-eliminating enzymes
(HRP and LOX). PAL-PAZ was used as the insulating
material. The film was deposited from a solution obtained by
mixinginvolumeratioof 1:1, 1:2,or 1:3,aPALsolution (4.5
mg in 100 mM HEPES buffer at pH 7.0) and a freshly
prepared PAZ solution (30 mg/mL). The PAZ solution was
used within 15 min of preparation.
The interference-eliminating layer was prepared according
to a previously published protocol.3s A 50-pl aliquot of a 12
mg/mL freshly prepared sodium periodate solution was added
to 100 pl of a solution containing 20 mg/mL H R P (HRP-VI
or HRP-BM) and 100 mg/mL LOX (LOX or rLOx) in 0.1
M sodium bicarbonate, and the mixture was incubated in the
dark for 2 h. Alternatively, the oxidation of H R P was carried
out prior to adding LOX and cross-linking.
The biocompatible layer films were photo-cross-linked by
exposure to UV light (UVP, Inc., San Gabriel, CA; Blak-
Ray; spectral peak at 360 nm, UV irradiance at the sample
200 mW/cm2) for 1 min. The initiator used was 2,2-
dimethoxy-2-phenylacetophenone (Aldrich). A solution of
300 mg/mL of the initiator in 1-vinyl-2-pyrrolidinone (Al-
drich) was added to the prepolymer mixtures. Approximately
30 pL of the initiator solution was added per milliliter of 10%
(w/w) aqueous solution of the tetraacrylated PEO. The
prepolymers were cross-linked in situ inside the recess of the
electrode. The films were prepared by filling the recess with
the prepolymer solution twice and exposing the electrode to
the UV light source after each time the cavity was filled.
In vitro experiments were carried out in batch fashion at
25 and 37 OC, using a conventional three-electrode electro-
chemical cell with the enzyme-modified gold wire as the
working electrode, a platinum wire as the counter electrode
and a saturated calomel reference electrode (SCE). The
electrolyte was a 20 mM phosphate-buffered saline solution
containing 0.15 M NaCl at pH 7.15. Experiments in serum
were performed at 37 O C , adding 100 pL antibiotic-
antimycotic solution to 10 mL of serum. Phosphate-buffered
saline and serum were agitated during the experiments. The
working potential was +0.3 V vs SCE for experiments with
the PVI-Os polymers.
In vivo experiments (6-10 h) were carried out in 300-g
male SpragueDawley rats. The rats were fasted overnight
and prior to the experiment were anaesthetized with an
intraperitoneal (i.p.) injection of sodium pentobarbital (65
3134 AnalWalChemisby, Vol. 66, No. 19, October 1, 1994
mg/kg rat wt). An i.p. injection of atropine sulfate (166 mg/
kg rat wt) was then administered to suppress respiratory
depression. Once the rat was anaesthetized, a portion of the
rat’s abdomen was shaved and coated with a conductive gel,
and an Ag/AgCl surface skin reference electrode was attached.
Sensors were then implanted subcutaneously, using a 22-gauge
Per-Q-Cath Introducer (Gesco International, San Antonio,
TX), on the rat’s thorax or subcutaneously in the intrascapular
area through a small surgical incision. The sensors were taped
to the skin to avoid sensor movement. The sensors, along
with the reference electrode, were connected to an in-house-
built bipotentiostat. The operating potential of the sensors
was 0.3 V vs Ag/AgCl, with the Ag/AgCl electrode serving
as both the reference and counter electrodes. Sensor readings
were collected using a data logger (Rustrak Ranger, East
Greenwich, RI) and at the end of the experiment were
transferred to a computer. During the experiment, the rat’s
body temperature was maintained at 37 OC by a homeostatic
blanket. The sensors were allowed to reach a basal signal
level for at least 1 h before blood sampling was started. Blood
samples were obtained from the tail vein, and all blood samples
were analyzed by use of a glucose analyzer (YSI, Inc., Yellow
Springs, OH; Model 23A). Approximately 30 min after the
start of blood sampling, a i.p. glucose infusion was started
using a syringe pump (Harvard Apparatus, South Natick,
-
MA) at a rate of 120 mg of glucose min-I (kg of rat wt)-I.
The glucose infusion was maintained for 1 h. At the end
of the experiment, the rat was euthanized by sodium pento-
barbital injection i.p. or asphyxiation by C02, consistent with
the recommendations of the American Veterinary Association.
All animal experimentation was approved by the University
of Texas Institutional Animal Use and Care Committee.
RESULTS AND DISCUSSIONS
Structure of the Sensor. The Recessed Electrode. The
recess, i.e., channel, in the polyimide-insulated wire was formed
by electrochemical etching of the gold under galvanostatic
control. By controlling the charge, the total amount of gold
electrooxidized and dissolved as Au(CN)2- was defined. When
the conditions were set so that the CN- transport into the
channel and the Au(CN)2- transport out of it were not rate
limiting, a flat gold wire surface was produced at the bottom
of channels with aspect ratios of 0.5-2.0. Thus, when the
CN- concentration was high enough and the wires were
ultrasonically vibrated, the tips of gold wires were flat. Passage
of 1.5 C/electrode at 8 mA current/electrode produced 125-
pm-deep cavities. At theoretical efficiency for one-electron
oxidation, 3.08 mg of gold would have been etched. The
amount of gold actually etched was only 0.076 mg, showing
significant CN- or water oxidation. Nevertheless, the process
was reproducible, accurate, and fast, with 20 electrodes being
processed in each batch in less than 5 min.
Structure and Performance. The depth of the channel
and the thickness of the polymer layers in it controlled the
mass transport of glucose to the sensing layer. By controlling
these parameters, the apparent K,,,was adjusted to -20 mM
glucose. The polyimide wall of the channel also protected the
four polymer and polymer-znzyme layers against mechanical
damage and reduced the hazard of their loss in the body.
 ‘00 , , , , , , , , , , , , , , , , 160 , , , , , , , , , , I , , , , ,

i 150 -
-

__r
140

80 130 -
120 -
70 110 -
- 60 / h

s
100 -
c
v

F 50 - P
P
v

F
u
90
80
-
-
70 -
0
60 -
50 -
40 -

‘ 1 ‘ 1 1 1 1 1 1 1 1 1 1 1 1
0 5 1 0 15 2 0 2 5 3 0 3 5 40 45 5 0 5 5 6 0 6 5 70 75 80 8 5 0 5 1 0 15 2 0 2 5 3 0 3 5 40 45 50 5 5 60 6 5 70 75 80 8 5
Glucose Conc. (mM) Glucose Conc. (mM)
Figure 2. Dependence of the sensitivity on the Os loading of the wire: Figure 3. Dependenceof the sensitivity on the recess depth: (0)125-
(V)PVI,-Os and (V)PV15-Os. and (0)250-pmdeep electrodes modified with the same amount of
sensing layer; E = +0.3 V vs SCE; 20 mM phosphate buffer containing
0.15 M NaCi at pH 7.15 and 25 OC.
Because the glucose electrooxidation current was limited by 140 I I I I

glucose mass transport through the recess and its polymer

films, rather than by mass transport to the tissue-exposed tip,
120
the current was practically insensitive to motion. Evidently,
the electrooxidation rate of glucose in the recessed sensing
layer was slower than the rate of glucose diffusion to the 100

channel’s outer fluid contacting interface.

PVI5-Os was chosen as the wire of the sensing layer. The -
0 80
subscript 5 is used to indicate that, on the average, every fifth :
v
vinylimidazole mer carried an electron-relaying osmium 0
center.50 The results for PVI5-Os and for PVI3-Os (every > 60

third vinylimidazole mer carrying an osmium center) are

compared in Figure 2, showing higher current density of 40

glucose electrooxidation on electrodes made with PVIs-Os

than on those made with PVI3-Os. 20
Depth of the Recess and the Sensing Layer. Channels of
125- and 250-pm depth were investigated to assess the
1 I I 1 1 I
0
dependence of the current on the depth of the recess (Figure 2 4 6 6 10 12 14 16 18 20
3), with the total amount of PVIs-Os and rGOx being kept Sensing Layer Thickness (p”
constant. Much of the loss in current in the deeper cavities Figure 4. Dependence of the normalized current on the thickness of
resulted not from reduced glucose mass transport but from the sensing layer: 20 mM phosphate buffer containing 0.15 M NaCi
adsorptive retention of part of the enzyme and polymer on the at pH 7.15 and 37 OC; E = +0.3 V vs SCE. Q Is the number of
coulombs of Os redox centers in the layer.
polyimide wall when microdrops of the component solutions
were introduced into the recess in the process of making the used. The best results were obtained using 1:2 PAL-PAZ
electrodes. Through repeated rinsing with water, some of the (Figure 5 ) , with three coatings applied to form a -7-pm-
adsorbed polymer and enzyme on the walls were washed onto thick cross-linked film.
the electrode surface, increasing the current. The highest The Interference- Eliminating Layer. The interferants,
currents were seen after five washings. When the thickness particularly ascorbate, urate, and acetaminophenol, were
of the sensing layer was increased through increasing the oxidized in the third layer, containing LOXand HRP. In this
number of coatings (Figure 4), the current reached a maximum layer, lactate, the typical concentration of which in blood is

-
and then dropped. For the preferred 125-pm recess, 10
coatings, producing a 13-pm-thick wired-rGOx sensing
layer, yielded sensors that had the desired characteristics for
in vivo use.
The Insulating Layer. This layer electrically insulated
the redox enzymes of the interference-eliminating layer (HRP
and LOX) from the wired rGOx layer. PAL cross-linking
with PAZ, forming a polycationic network at pH 7.0, was
1 mM, reacted with 0 2 to form HzOz and pyruvate. H202,
in the presence of HRP, oxidizes ascorbate, urate, and
acetaminophenol, being reduced to ~ a t e r . The
~ ~ preferred
, ~ ~
coimmobilization process involved two separate steps: pe-
riodate oxidation of oligosaccharide functions of HRP to
aldehydes, followed by mixing with LOX and formation of
multiple Schiff bases between HRP aldehydes and LOXamines
(e.g., lysines) and between H R P aldehydes and amines. The

AnalyticalChemistty, Vol. 66, No. 19, October 1, 1994 3135

 l o9 o0 -I
60
v
Z 50
e value measured during the 72-h test conducted in 10 mM glucose in the
40
01 I 1
0 10 20 30 40 50 60 70
Time ( h )
Figure 5. Stability of the sensitivity of electrodes varying in sensing
layer thickness and Insulating layer composition and thickness. (0)
+
5-pm PV15-Os, 4-pm PAL PAZ (ratio 1:l): (0)7.5-pm PVIs-Os,
+
4-pm PAL PA2 (ratio 1:l): (V)7.5-pm PV15-Os, 4.5-pm PAL PA2
(ratio 1:2):and (V)12.5-pm PV15-Os, 7-pm PAL PA2 (ratio 1:2):
20 mM phosphate buffer containing 0.15 M NaCl at pH 7.15, 37 OC,
10 mM glucose; +0.3 V vs SCE.
Flgure 6. Interferenceelimination by the glucose sensor: A indicates
the injections of ascorbate, producing a 0.1 mM concentration: L
indicates inlections of lactic acki; 0, indicates glucose injections; 20
mM phosphate buffer containing 0.15 M NaCi at pH 7.15, 37 OC; E
= +0.3VvsSCE. Toaccommodateallthedata,thescale waschanged
between the first and second glucose Injection.
thickness of the interference eliminating layer was -85 pm.
The layer was made by applying six successivecoatings. Figure
6 shows that electrooxidizable interferants were eliminated in
the presence of lactate at physiological levels. LOX slowly
lost its activity in the cross-linked HRP-LOX layer. This led
to degradation of the ability of the layer to eliminate
interferants. After 36 h of operation at 37 O C , a measurable
current increment was noted when enough ascorbate was added
to produce a 0.1 mM concentration.
The Biocompatible Layer. The biocompatible layer
consisted of photo-cross-linked tetraacrylated 18 500-Da poly-
(ethylene oxide).51 The thickness of this layer, made by
sequential photo-cross-linking of two coatings, was 20 pm.
The 1-min UV exposure required for the photo-cross-linking
process reduced the sensitivity by 16 f 2%.
3136 AnalyticalChemistty, Vol. 66, No. 19, October 1, 1994
Table 1. Sensoor Characterlrtks'
i(nA) j(pA/cm*)
KmpPP(mM)
EH LB tr(s)
current
variance(%)
33.9 69.1 18.5 33.4 30-90 5.0
i is the current measured at 37 OC and at 10 mM glucoseconcentration;
j i s thecurrentdensitymeasuredat 37 'Cat 10mMglucoseconcentration;
Kmapp is the apparent Michaelis-Menten coefficient determined from an
electrochemical Eadie-Hoffstee (EH) or a Lineweaver-Burk (LB) plot;
1, is the 10-902 rise time, 90 s for 2 mM and 30 s for 20 mM glucose
concentration; current variance is the maximum deviation from the mean
presence of interferants. The current was monitored continuously at 37
OC.
Stability and Other Characteristics. In order to improve
i the stability, the more thermostable rGOx rather than GOx49
was used and glucose transport was reduced to make the sensor
, I current diffusion, not enzyme turnover, limited. The glucose
80 90 100
flux was attenuated by the three outer layers and the sensing
layer itself. Because the sensing layer had a large excess of
glucose oxidase, its activity greatly exceeding that needed for
electrooxidizing the attenuated glucose flux, the sensor's
+ stability was improved.
+ The stabilitywas tested in the presence of 0.1 mM ascorbate
in 10 mM glucose at 37 OC. The current output of a typical
optimized electrode was 35 nA and the apparent K m , derived
-
from an Eadie-Hofstee plot, was 20 mM (Table 1). The
10-90% response time was 1 min. As expected, and as can
be seen in Figure 5, with thinner films the glucose mass
transport increased, Le., the current was higher, while for
thicker films the stability improved. Because of the high
sensitivity of thin sensing film (- 1 pm) electrodes (>10-*A
cm-* M-l), a 1 order of magnitude decrease in sensitivity
could be traded for stability, while the currents remained high
enough to be easily measured. As seen in Figure 5, the
sensitivity of the stabilized sensors did not change by more
than f5% for 72 h of operation at 37 OC. After a small initial
decrease in sensitivity, it increased to a maximum after 40 h,
and the final 72-h sensitivity was almost identical with the
initial. There was no measurable difference in the apparent
stability of the sensitivity of continuously operated and resting
electrodes measured at 6-h intervals.
The characteristics of the electrodes are summarized in
Table 1. Each entry represents an average value for five
electrodes. The baseline currents were typically <0.5 nA and
the noise <10 PA. The currents observed throughout the
physiological glucose concentration range (2-20 mM) ex-
ceeded the noise equivalent current by at least a factor of 100.
The apparent Km was 20 mM, and the 10-90% response time
was, for aged electrodes, -90 s at the lowest physiologically
relevant glucose concentration (2 mM) and 30 s a t the highest
(20 mM). The baseline of nil at 0 mM glucose was stable for
36 h in the presence of 0.1 mM ascorbate. The stability
observed and the existence of a valid zero point in the presence
of interferants suggested that the sensor could be used in vivo
for 72 h and be tested/recalibrated in vivo through a single-
point calibration, Le., by withdrawing only a single sample of
- blood for independent analysis.
In Vivo Experimentation. The objective of this experiment
was to establish the validity of a one-point in vivo calibration.
Two sensors were simultaneously implanted subcutaneously
 1 ysI
Infusion
stop 1350

25 I I300

50

5 I I I I 4 I 1 Io
50 100 150 200 250 300 350 400 450 500 50
Time (min.)

Flgure 7. Current output of two subcutaneously Implanted sensors, 0 50 100 150 200 250 300 350 400
tracking the blood glucose levels before, during, and after an
intraperitonealInfusion of glucose In a rat. The glucose concentrations Reference Glucose(mg/dl)
in blood samples from the tail vein of the rat were measured with a Flgure 8. Clark-type clinical error grkl analysis for the two subcutane-
YSI analyzer. The circles (0)represent the blood glucose levels ously implantedelectrodes calibrated In VIVOusingan arbitrarilychosen
measured on the withdrawn samples, line A shows the output of the point (0)electrode implantedsubcutaneously in the lntrascapulararea
electrode implantedsubcutaneouslyin the chest, and line B represents and (0)electrode Implanted subcutaneously in the chest. Readings In
the output of the electrodeImplantedsubcutaneouslyinthe intrascapular zone A are clinically accurate, those in zone B are considered clinically
area of the rat. correct.

in a rat: oneon the thorax and the second between the scapulas.
To make the difference between the blood sampled and the 350
' A
subcutaneous fluid probed with the sensors as extreme as E c .
possible, i.e., to probe whether the one-point calibration holds
even if the organs sampled are different and the sampling
sites are remote, blood was withdrawn from the tailvein. Blood
glucose levels were periodically measured in withdrawn
samples, while the absolute uncorrected sensor current output
was continuously monitored. The results are shown in Figure
7. As seen in Figure 7, at 410 min the current of this electrode
dropped precipitously. Such a drop was observed in other
subcutaneously implanted electrodes between 400 and 600
min, but was never observed in electrodes operated in buffer
at 37 OC. When the failed electrodes were withdrawn and
retested in buffer, most of their original sensitivity was found L
:' A 0%
to be intact. The cause for this sudden and reversible 0%
6 I 1 I
deactivation is under study. By use of an arbitrarily chosen
point to calculate a calibration curve for each electrode, Le.,
one blood glucose level determination and one current
measurement to establish the scales, all the data from Figure
7 were plotted in a Clarke-types2 clinical grid (Figure 8),
without further correction. In this analysis, points falling in
region A of the grid are considered clinically accurate, while deactivation of the lactate oxidase, resulting in loss of
those in region B are considered clinically correct. Points interference elimination, one-point calibration should be
falling in region C are not correct, but would not lead to practical. After such calibration, the readings of the sub-
improper treatment. Points in regions D or E are incorrect, cutaneous sensors will provide, without any correction,
and if treatment would rely on these, it would be improper. clinically useful estimates of blood glucose levels.
All of the points, from both electrodes, are in regions A and Figure 9 shows the distribution of all possible correlations
B, with 43 of the 48 points being in region A. The three points obtained when each of the 24 glucose analyses was used for
in region B near 100 mg/dL glucose, for the electrode single-point calibration of either implanted electrode. There
implanted between the scapulas, were the last three points of are 2 X 24 X 24 = 1152 points in the distribution. Of these,
theexperiment, at -410 min. Notwithstanding theunknown 78% are in region A, 15% are in region B, 1% are in region
failure mode at 400-600 min and failure after 36 h by C, 6% are in region D, and no points are in region E. In
(52) Clarke, W. L.; Cox, D.; Gonder-Frederick, L. A,; Carter, W.; Pohl, S. L.
Figure 10, we test for the improvement of the single-point
Diabetes Care 1987, 5, 622427. calibration through use of redundant electrodes. First, the

Analytical Chemistry, Vol. 66, No. 19, October 1, 1994 3137


' (allylamine) with a polyaziridine; an interference-eliminating
0%
..i .! //i
r,(
0
D
0%
1%
50
C 30 s at 20 mM glucose. Their sensitivity, after 30-min
A 0%
, . I
0%
0 .'
0 50 100 150 200 250 300
Reference Glucoae(mg/dl)
Flguro 10. Distribution of 242 glucose concentrations, based on 11
single-point calibrations for each sensor, for redundant implanted
sensors. I f the difference in the readlngsof the two sensors (electrode
A readlng minus corrected electrode B reading) was greater than the
standard deviation for all of the differences, the data were rejected
both as possible one-point calibrations and as data points.
readings of electrode A were normalized with respect to those
of electrode B by multiplying each reading by the average
output of electrode B divided by the average output of electrode
A. Next, the standard deviation was calculated for the
differences between the 24 sets of readings of implanted
electrode B and corrected readings of implanted electrode A.
Then, all those sets of readings that differed by more than the
standard deviation were rejected. The number of sets was
reduced thereby from 24 to 11; 82% of the points were in
region A, 17% in region B, 1% in region D, and no points in
regions C and E. The distribution demonstrates that the
sensors can be calibrated through a single independent
measurement of the glucose concentration in a withdrawn
blood sample. They also demonstrate the improvement in
clinical accuracy resulting from the use of redundant sub-
cutaneous sensors. The selection of those data points that
differed by less than the standard deviation for the entire set
led to a 6-fold reduction in the probability of clinically erring
in a decision based on readings of the implanted sensors.
Finally, the observation that a particularly large number of
readings differed by more than the standard deviation at low
glucose concentrations shows that the sensors are less accurate
at the low glucose levels.
CONCLUSIONS
Insulated gold wire based 290-pm-0.d. subcutaneous
glucose sensors, allowing one-point calibration in vivo, can be
3138 Analytical Chemistiy, Vol. 66, No. 19, October 1, 1994
built. Their construction involved coating of a 250-pm-
diameter gold wire, in a 125-pm-deep polyimide-walled recess
formed by etching the gold, with four layers: an immobilized
wired genetically engineered glucose oxidase layer; an electri-
cally insulating layer formed by cross-linking of poly-
layer of cross-linked horseradish peroxidase and lactate
oxidase; a biocompatible film of poly(ethy1ene oxide), de-
i rivatized to allow its photo-cross-linking.
The electrodes had no leachable components. The total
amount of polymers and enzymes was -5 pg. The glucose
response through the physiologically relevant 2-20 mM
concentration range was close to linear. The electrodes did
not respond to ascorbate, urate, or acetaminophen01 for 36 h.
Their 10-90% response time was 90 s at 2 mM glucose and
equilibration, was stable for 72 h at 37 OC in 10 mM glucose,
350 400 the current deviating from the average by less than f5%. Two
electrodes implanted subcutaneously in a rat tracked the blood
glucose levels, and their absolute, uncorrected current output
was proportional to the blood glucose concentration. Analysis
of the correlation between the blood glucose levels in the tail
vein and the current output of the sensors in the subcutaneous
regions of the thorax and between the scapulas of the same
rat showed that even when the probed sites and organs differed
in the extreme, one-point in vivo calibration was valid. The
analysis also showed thevalue of implanting redundant sensors.
Had clinical decisions been made based on individual sensor
readings, calibrated at one point, 94% would have been
clinically correct. By using redundant sensors and accepting
only those pairs of readings that were within one standard
deviation, the percentage of the clinically correct decisions
was increased to 99%.
ACKNOWLEDGMENT
The National Institutes of Health (DK42015) is acknowl-
edged for supporting this work. The Royal Swedish Academy
of Sciences/Per-Eric Lindahl Foundation is acknowledged
by E.C. and the Swedish National Board for Industrial and
Technical Development is acknowledged by S.-E. L. for
financial help. The Chiron Corp. is acknowledged for its gift
of the rGOx. The authors thank Dr. Tim Ohara and Mr.
Ravi Rajagopalan for synthesizing the PVI-Os polymer, and
Mr. Mark Vreeke for valuable discussions. The authors thank
Mr. Andrew Young for the use of his bipotentiostats for the
in vivo experiments. We also acknowledge Dr. Wolfgang
Kerner for his assistance in constructing the recessed elec-
trodes.
Received for review January 26, 1994. Accepted June 1, 1994."
Abstract published in Aduonce ACS Abstracts. August 1, 1994.