Single-Walled Carbon Nanotube-Based Near-Infrared Optical Glucose Sensors toward In Vivo Continuous Glucose Monitoring

Abstract

This article reviews research efforts on developing single-walled carbon nanotube (SWNT)-based near-infrared (NIR) optical glucose sensors toward long-term in vivo continuous glucose monitoring (CGM). We first discuss the unique optical properties of SWNTs and compare SWNTs with traditional organic and nanoparticle fluorophores regarding in vivo glucose-sensing applications. We then present our development of SWNT-based glucose sensors that use glucose-binding proteins and boronic acids as a high-affinity molecular receptor for glucose and transduce binding events on the receptors to modulate SWNT fluorescence. Finally, we discuss opportunities and challenges in translating the emerging technology of SWNT-based NIR optical glucose sensors into in vivo CGM for practical clinical use.

Introduction

Diabetes affects 346 million people worldwide.^1–2 Frequent hyperglycemia can cause serious complications, including heart disease, stroke, blindness, nervous system disease, and kidney disease.^3–6 In order to avoid such complications, diabetes patients rely on glucose sensors to maintain target blood glucose levels.⁷ As a consequence, effective management of diabetes requires continuous monitoring of blood glucose levels.⁸ An easy-to-use and accurate continuous glucose monitoring (CGM) sensor would thus improve the quality of patients’ lives and help prevent and/or reduce complications that may arise from diabetes.^5–6,8–10 The next generation of glucose sensor technology therefore focuses on in vivo CGM.^8,10–18

The current standard for in vivo CGM is the electrochemical method that uses enzymatic reactions.^8,10,19,20 Several CGM electrochemical glucose sensors have been approved by the Food and Drug Administration and are commercially available.^2,8,21 However, these sensors still have several limitations:^8,10,16,22 they have a short lifetime (2–7 days in vivo), they require frequent recalibration (typically at least once a day), they use a needle-like transdermal electrode and thus leave an open wound, they often overestimate hypoglycemia [which can limit their use with a “closed-loop” insulin pump (i.e., artificial pancreas)], and they are intrinsically susceptible to biofouling because they measure the flux of glucose rather than the concentration.

Because of their advantages over electrochemical sensors for in vivo continuous monitoring, noninvasive or minimally invasive optical glucose sensors are of great interest.^16,23,24 A key advantage is that optical glucose sensors do not require an electrode that physically penetrates the skin as in commercial electrochemical CGM sensors; rather these implanted sensors can be optically interrogated into the skin. Optical glucose sensors include minimally invasive, implantable fluorescence- or surface-plasmon-resonance-based sensors and completely noninvasive spectroscopic sensors.^16,23–25 Fluorescence-based sensors typically use glucose-responsive fluorophores and fluorescence resonance transfer pairs of fluorophores that emit fluorescence at visible wavelengths. However, such fluorophores often photobleach, which is intrinsically unfavorable for long-term in vivo glucose sensing; furthermore, fluorescence at visible wavelengths does not penetrate well through the optically turbid biological tissue.^16,23,26 On the other hand, current noninvasive spectroscopic-based sensors lack specificity and dynamic range for glucose detection.^{11,12,21,24,25} For example, a noninvasive CGM sensor based on a multisensor concept, embedding dielectric spectroscopy and optical sensors, and a mathematical model is only valid for real-time monitoring of glycemic variation.²⁴

Single-walled carbon nanotubes (SWNTs) have unique physical and chemical properties for applications in life sciences and medicine, including in vivo sensing and imaging.^27,28 Particularly, nonphotobleaching, near-infrared (NIR) fluorescence of SWNTs, and the capability to modulate SWNT fluorescence selectively in response to specific analytes offer great opportunities to develop NIR optical biomedical sensors. For example, we and others have demonstrated the selective modulation of SWNT fluorescence in response to target analytes, including glucose,^29–32 and have developed various SWNT-based NIR optical sensors,^29–37 which even operate at the single-molecule level.^38–40 Here we review research efforts on developing completely passive, implantable SWNT-based NIR optical sensors toward long-term in vivo CGM. We first describe the unique optical properties of SWNTs and compare SWNTs with traditional organic and nanoparticle fluorophores regarding in vivo optical glucose sensing. We then present our development of SWNT-based glucose sensors that use glucose-binding proteins (GBPs) and boronic acids (BAs) as high-affinity molecular receptor for glucose. Finally, we discuss challenges and opportunities in translating the emerging technology of SWNT-based NIR optical glucose sensing into in vivo CGM for practical clinical use.

Optical Properties of Single-Walled Carbon Nanotubes for in vivo Sensing

Carbon nanotubes are a high-aspect ratio nanostructure, conceptually constructed by rolling graphene sheets into a seamless cylinder. Single-walled carbon nanotubes consist of a single layer of graphene whereas multiwalled carbon nanotubes (MWNTs) consist of multiple layers of graphene. The quasi-one-dimensionality of SWNTs imparts sharp electronic state densities at Van Hove singularities (Figure 1).⁴¹ This electronic structure renders unique optical properties of SWNTs. Single-walled carbon nanotubes absorb light at photon energy E₂₂ and emit fluorescence at E₁₁ in the NIR wavelength range (900 to 1600 nm; Figure 1A).^41–43 The values of E¹¹ and E₂₂ depend on the chiral structure of SWNTs, denoted by the chiral index (n, m) that indicates the direction in which a graphene sheet is rolled into a cylinder; accordingly, the fluorescence properties of SWNTs vary with nanotube chiral structures (Figure 1B).^41,42 All atoms of SWNTs are surface atoms, sensitive to surface adsorption events.

Figure 1.

Optical properties of SWNTs. (A) Schematic density of electronic states of SWNTs. Solid arrows indicate the optical excitation and emission transitions; dashed arrows depict nonradiative relaxation of the electron in the conduction band and hole in the valence band before emission. (B) Fluorescence excitation–emission profile map for SWNTs. Reprinted with permission from Science.41

Conventional fluorophores, including fluorescent quantum dots, typically emit fluorescence at visible wavelengths and often photobleach rapidly, whereas SWNTs emit fluorescence at NIR wavelengths (900 to 1600 nm), within the so-called biological “tissue transparency window,” in which biological tissues are optically transparent (Figure 2A). SWNT-based sensors could thus outperform conventional visible wavelength fluorophores for applications in biological tissues, because the tissues are strongly absorbing and autofluorescent at visible wavelengths. Additionally, SWNT-based sensors could persist indefinitely, because SWNTs are nonphotobleaching (Figure 2B).

Figure 2.

Fluorescence properties of SWNTs and absorption and scattering in biological tissues. (A) Fluorescence of SWNTs in the NIR II region. Blood (red) and water (water) absorption are significant in the visible and infrared region. The tissue transparency window, in which the tissue is optically transparent, lies between the water and blood absorption wavelength ranges. The figure includes data from Reference 26. (B) The fluorescence of SWNTs (red) is photostable even under high fluence (1.3 × 10⁷ Wm^-2). Indocyanine green (ICG) and most organic fluorophores show a rapid photobleaching upon continuous excitation (blue). (C) Reduced scattering coefficient µ_s’ for skin, adipose tissue, and mucous tissue as derived in Bashkatov and coauthors.45 Scattering in the tissues decreases with increasing wavelengths: µ_s’~λ^–w, where w depends on the size and concentration of scatters and thus varies with tissue types from 0.22 to 1.68. Reprinted with permission from ChemSusChem²⁸ and from Proceedings of the National Academy of Sciences.44 Abs, absorption.

Next we compare the quantum yield, optical tissue transparency, tissue penetration depth, and photostability of SWNTs with those of conventional organic and nanoparticle fluorescence probes, regarding in vivo sensing applications. The intensity of the signal emitted by optical sensors is an important design constraint. For in vivo optical sensing, the optical signal emitted by optical sensors depends both on the quantum yield and the absorption in tissues. The quantum yield is the efficiency of energy transfer from incident light to emitted fluorescence. The absorption in biological tissues strongly depends on the wavelength (Figure 2A).²⁶ A simple one-dimensional fluorescence and absorption model estimates the fluorescent signal intensity I_S from a sensor implanted in the tissue as I_S = I₀φe^-2μd-kτ, where I₀ is the incident excitation intensity, ϕ is the quantum yield, µ is the absorption coefficient, d is the distance from the surface of the tissue to the sensor implanted in the tissue, k is the pseudo-first-order photobleaching rate constant, and τ is the excitation exposure lifetime.²² Table 1 compares the calculated value of ϕe^–2md for common organic and nanoparticle fluorophores and SWNT, which accounts for the effect of the absorption (e^–2md) in I_S and determines the maximum signal emitted by an implanted sensor.²² Despite the high quantum yield of visible fluorophores, the high absorption by tissues at visible wavelengths significantly attenuates the fluorescence signal of visible fluorophores. Moreover, low autofluorescence of tissues at NIR wavelengths renders NIR fluorophores beneficial for in vivo sensing applications.

Table 1.

Quantum Yield, Absorbance in Human Whole Blood (µ), and ϕe–2μd of Common Visible and Near-Infrared Organic and Nanoparticle Fluorescent Probes^a

Fluorescent probes	Quantum yield (%)	Conditions for quantum yield measurement	Excitation (nm)	µ (oxygenated)	µ (deoxygenated)	Φe–2μdb
Cy5	27	PBS	620	2	60	3.20 × 10-28
Fluorescein	95	0.1 M NaOH	496	150	120	5.23 × 10-118
Rhodamine 6G	95	Water	488	200	105	3.30 × 10-133
Rhodamine B	31	Water	514	110	190	1.60 × 10-131
ICG	0.266	Water	820	0.96	0.77	4.72 × 10-4
ICG	1.14	Blood	830	1.01	0.78	1.91 × 10-3
Type II NIR quantum dots	13	PBS	840	1.05	0.78	2.09 × 10-2
SWNT	0.1	PBS	1042	0.889	0.12	3.65 × 10-4

PBS, phosphate-buffered saline

^aAdapted from Barone and coauthors.²²

^bCalculated based on d = 1 cm and an average µ.

In addition to absorption, scattering also attenuates the fluorescent signal from optical sensors (Figure 2C).⁴⁴ The optical penetration depth in the tissue depends both on absorption and scattering: δ = (3μ_aμ_a + μ_S^′))^-1/2, where μ_a is the absorption coefficient andis the reduced scattering coefficient. The scattering coefficient varies with the wavelength: μ_S^′˜λ^-w, where w is determined by the size and concentration of scatters and thus varies with tissue types.^44,45 Because of the inverse dependence of tissue scattering on the wavelength, long-wavelength NIR fluorophores could have a higher penetration depth in tissues than short-wavelength traditional visible fluorophores.

Experimental and simulation modeling studies of the effect of tissue scattering on the tissue penetration depth showed that the maximum penetration depth can be obtained at the wavelength range 1000–1400 nm, called the second NIR widow (NIR II).^44–47 Single-walled carbon nanotubes emit fluorescence within the NIR II, whereas most commercially available NIR fluorophores emit fluorescence in the NIR I (700 to 900 nm), including indocyanine green (ICG) and NIR fluorescent quantum dots.^44,48 The lower scattering and autofluorescence in the NIR II may compensate for higher absorption in NIR II (mostly due to water) than NIR I and make NIR II-emitting SWNTs outperform conventional NIR I-emitting fluorophores for in vivo applications (Figure 2).⁴⁴ Welsher and coauthors⁴⁴ studied the effect of tissue absorption and scattering on NIR I (ICG) and NIR II (SWNT) fluorophores in tissue phantoms (Figure 3).⁴⁹ This study shows a more pronounced loss of signal clarity as well as feature contrast and integrity in NIR I (ICG) than NIR II (SWNT; Figure 3A and 3B), confirming the inverse wavelength dependence of the scattering effect.⁴⁹ The intensity profile as a function of tissue depth shows a similar attenuation in the fluorescence intensity at NIR I (ICG) and NIR II (SWNT), attributed to the combined absorbance and scattering effects (Figure 3C). An additional benefit of SWNTs (NIR II fluorophores) over traditional NIR I fluorophores includes a lower autofluorescence in the NIR II region, resulting from a larger Stokes shift between excitation and emission, which can be larger than 300 nm, and the longer excitation and emission wavelengths (because autofluorescence is mostly in the visible wavelength region).⁴⁴ Figure 3D shows a dynamic contrast-enhanced image with SWNTs through principal component analysis (left)⁴⁴ and NIR fluorescence image (right) of mice.⁵⁰

Figure 3.

Optical penetration depth of NIR II-emitting SWNTs and NIR-I-emitting ICG in tissue phantoms. (A) Fluorescence images of capillaries filled with SWNTs (NIR II) and ICG (NIR I) at the depth of 0, 3, and 5 mm in Intralipid® excited at 785 nm. The fluorescence at NIR II (SWNT) shows a less loss of signal clarity and feature contrast and integrity (due to less scattering) than the fluorescence at NIR I (ICG) in the tissue phantom. Scale bars, 1.5 cm. (B) Feature width of SWNT and ICG capillary images as a function of depth in the tissue phantom, showing a more significant feature loss of NIR I-emitting ICG than NIR II-emitting SWNT. Control experiments done in water show no change in feature size for both ICG and SWNT. Error bars are derived from the uncertainty in the fitting of feature width. (C) Normalized intensity loss of ICG and SWNT as a function of depth in the tissue phantom (Intralipid) and water. Despite the higher absorption of water in NIR II, the decay of signal of ICG and SWNTs in the tissue phantom is similar, with the exponential decay depths of 1.04 ± 0.04 and 0.97 ± 0.03 mm, respectively. (D) Dynamic contrast-enhanced image with SWNTs through principal component analysis (left) and NIR fluorescence image (right) of mice. Reprinted with permission from Proceedings of the National Academy of Sciences⁴⁴ and Nature Nanotechnology.50 FWHM, full width at half maximum.

The photostability of fluorophores determines the lifetime of optical sensors. Assuming the maximum tolerable error of viable in vivo glucose sensors of 20% for clinical applications,⁵ a simple calculation from the attenuation factor of e^–kt in the fluorescent signal intensity I_S estimates the lifetime τ of optical sensors as τ = –ln(0.8/k). Table 2 shows the estimated sensor lifetime calculated with literature values of the photobleaching rate constant k. Organic fluorophores are expected to have short lifetimes mostly because of a rapid photobleaching. Single-walled carbon nanotubes are the only known NIR fluorescent probe with no photobleaching threshold, ideal for long-term sensing applications.^51,52

Table 2.

Photobleaching and Sensor Lifetime of Common Organic and Nanoparticle Fluorescent Probes^a

Fluorescent probes	Photobleaching rate constant (h-1)	Fluence (mW/cm2)	Irradiated sensor lifetimeb
Infrared dye 78-CA	250.9	600	3.2 s
Cy5	20.5	600	39.1 s
ICG	0.0412	28	5.4 h
Type II NIR quantum dots	0.0827	600	2.7 h
SWNT	0	1.0 × 106	∞

^aAdapted from Barone and coauthors.²²

^bEstimated sensor lifetime calculated from τ = –ln(0.8/k).

Selective Modulation of Single-Walled Carbon Nanotube Fluorescence

The fluorescence of SWNTs is highly responsive to the local physical and chemical environment (e.g., local dielectric environment), making SWNTs a unique nanoscale optical sensing platform.^53–55 Figure 4 shows five representative mechanisms for selective modulation of SWNT fluorescence.³⁵ Adsorption of a redox-active analyte to the nanotube surface can shift the nanotube Fermi level, inducing bleaching of the absorption transition and associated excitonic loss (mechanism 1, Figure 4A).^29,56 Examples of this mechanism include the binding of hydrogen peroxide and several other redox-active molecules and pH effects.^39,40,57 Secondly, exciton quenching through the nonradiative excited-electron transfer from nanotubes to the analyte adsorbed on the nanotubes can quench SWNT fluorescence (mechanism 2, Figure 4B).^32,33 Additionally, changes of the solvent environment lead to solvatochromic shift (mechanism 3, Figure 4C).⁵⁸ An example of this mechanism is the ion-induced conformational change of deoxyribonucleic-acid -encapsulated SWNTs.^59,60 Engineering polymer-wrapped nanotube complexes can also impart steric selectivity to the mechanism 3 (mechanism 4, Figure 4D). Another strategy for engineering selectivity is to conjugate an analyte-specific switchable polymer to the nanotubes (mechanism 5, Figure 4E).³⁵ An example is SWNT complexes wrapped by a polyvinyl alcohol (PVA) derivative functionalized with glucose-binding protein for glucose detection.³¹ In reality, the modulation of SWNT fluorescence is often resulted from the combination of these mechanisms.

Figure 4.

Representative mechanisms of SWNT fluorescence modulation. (A) Mechanism I, bleaching of the absorption transition induced by the shift of the Fermi level through the adsorption of redox-active analytes to the nanotube surface. (B) Mechanism II, fluorescence quenching resulted from nonradiative excited-electron transfer from nanotubes to the analyte adsorbed on the nanotubes induced by analyte binding. (C) Mechanism III, solvatochromic shifting resulting from the perturbation of SWNT-bound polymer. (D) Mechanism IV, selective analyte binding mediated by engineered polymer wrapping. (E) Mechanism V, analyte-activated polymer switching resulting in intensity and/or wavelength modulation. Reprinted with permission from Proceedings of the National Academy of Sciences.35

Single-Walled Carbon Nanotube-Based Near-Infrared Optical Glucose Sensors

We have demonstrated a proof-of-concept for modulating the NIR fluorescence of SWNTs in response to glucose and developed prototype SWNT-based glucose sensors.^29,30 A central challenge in designing SWNT-based sensors is to engineer the nanotube interface to be selective to glucose and effectively modulate SWNT fluorescence, while maintaining the colloidal stability of SWNT complexes. Because direct covalent functionalization of the surface of SWNTs disrupts SWNT fluorescent emission, noncovalent functionalization is required. Here we describe two SWNT-based glucose sensors: glucose-binding-protein-based and BA-based SWNT optical glucose sensors. Our earlier sensors have been reviewed previously.⁷

Glucose Binding Protein-Based Single-Walled Carbon Nanotube Glucose Sensors

Glucose-binding protein is a periplasmic-binding protein with high affinity for glucose that changes its conformation upon binding of glucose (from the open to the closed form; Figure 5).⁶¹ We developed a GBP-based SWNT glucose sensor by constructing PVA–GBP conjugates that wrap the nanotube (Figure 5A).³¹ The GBP mechanically actuates the SWNT complex, resulting in reversible exciton quenching in response to glucose. The calibration curve shows the increase in fluorescence quenching with increasing glucose concentrations (Figure 5B). Furthermore, no response to glucose is observed for the same SWNT construct without GBP, which confirms the quenching mechanism (Figure 5C). The observed fluorescence quenching is selective to glucose, when compared with two other saccharides, with fructose and mannose showing a response of 40% and 20%, respectively, of the response to glucose at 10 mM (Figure 5D). As an illustration of a potential application, the GBP-based SWNT sensor solution is loaded into a dialysis capillary (10,000 molecular weight cut-off) to allow glucose to diffuse across the membrane while retaining the SWNT (Figure 5E). Cycling of the glucose level also shows a rapid and reversible change in SWNT fluorescence (Figure 5F).

Figure 5.

Glucose binding protein-based SWNT optical glucose sensors. (A) Schematic of the glucose recognition of the GBP–PVA/SWNT sensor. The difference in the angle between the two structural domains is the maximum of 36°, which arises from torsion angle changes in a three-segment hinge. (B) Calibration curve for GBP–PVA/SWNTs: steady-state responses of the fluorescent intensity of three different types of nanotubes to varying concentrations of glucose. (C) Histogram showing the selectivity of GBP–cationic-modified PVA/SWNTs over two other saccharides at 10 mM. (D) Optical and NIR fluorescent images of a 200 µm microdialysis capillary that encapsulates GBP–PVA/SWNTs. (E) Rapid and reversible response of GBP–PVA/SWNTs in the microdialysis capillary upon cyclic exposure to glucose. The gray-colored region indicates the presence of glucose. Reprinted with permission from Angewandte Chemie International Edition.30

Boronic-Acid-Based Single-Walled Carbon Nanotube Glucose Sensors

Boronic acids are an excellent molecular receptor for glucose.^62,63 The reversible complexation of glucose with aromatic BAs produces a stable boronate anion, changing the electronic properties of the BAs.^64–67 This alternation in the electronic properties of aromatic BAs upon binding of glucose has been a basic scheme for various BA-based glucose-sensing approaches.^{63–66,68–70} It was hypothesized that the complexation of glucose with aromatic BAs conjugated on the surface of SWNTs could modulate the SWNT fluorescence signal in response to binding of glucose (Figure 6).³² To test this concept and identify BAs that can be exploited as a molecular receptor for glucose in SWNT-based sensing systems, we screened the fluorescence spectral response of sodium cholate (SC)-suspended SWNTs (SC/SWNTs) against ³⁰ BAs and the subsequent spectral change of the 30 BA–SWNT complexes in response to glucose.³² The screening identifies 4-chlorophenylboronic acid (Figure 6B) and 4-cyanophenylboronic acid (Figure 6C), which uniquely show a reversible wavelength shift and a turn-on fluorescence response when complexed with SWNTs upon glucose binding. Figure 6D shows the fluorescence spectra of the quenched BA–SWNT complex of 4-cyanophenylboronic acid upon stepwise introduction of glucose: the BA–SWNT complex recovers the quenched fluorescence and the red-shifted emission wavelength with glucose concentration. Figure 6E shows the fluorescence intensity (left) and the emission wavelength (right) of (6,5) nanotubes as a function of glucose concentration: the BA–SWNT complexes respond to glucose in a physiologically important range of 0 to 30 mM.

Figure 6.

Reversible fluorescence quenching of BA–SWNT complexes for glucose detection. (A) Schematic illustration of the reaction of BAs with SC-suspended SWNTs (SC/SWNTs) and the fluorescence spectral response of the BA–SWNT complex to glucose. (B),(C) Fluorescence spectra that compare the original spectrum of SC/SWNTs (black), the spectrum after adding 50 mM BAs to SC/SWNT solutions (blue), and the spectrum after adding 50 mM glucose to the BA-SWNT complex solutions (red). The BA–SWNT complexes were prepared with (B) 4-chlorophenylboronic acid and (C) 4-cyanophenylboronic acid. Reprinted with permission ACS Nano.32 FL, fluorescence.

This “turn-on” sensing scheme that uses the reversible fluorescence quenching and wavelength shift of the BA–SWNT complex has a potential to be applied to develop a highly stable and sensitive NIR optical glucose sensor. However, the current form of the BA-based SWNT sensors requires surfactant in the nanotube sensor solution to disperse the nanotubes in aqueous solutions.

Polymer complexes that can wrap SWNTs and disperse them in aqueous solutions and contain BA receptors to specifically detect not only glucose, but also other saccharides are thus under active research, including BA-conjugated polyethylene glycol-based polymer complexes.⁷¹ In this work, we demonstrated both functions using a homologous series of seven phenyl BAs conjugated to a polyethylene glycol, eight-member, branched polymer (PPEG8) that allows for aqueous dispersion of SWNT and quenching of the NIR fluorescence in response to saccharide binding. We compared the 2-carboxyphenylboronic acid (CPBA), 3CPBA, 4CPBA, N-(4-Phenylboronic)succinamic acid, 5-Bromo-3-carboxy, ³-Carboxy-5-fluoro, and 3-carboxy-5-nitro, demonstrating a clear link between SWNT photoluminescence quantum yield and BA structure (Figure 7). Saccharide recognition using the NIR photoluminescence of SWNT is demonstrated, including selectivity toward pentoses such as arabinose, ribose, and xylose to the exclusion of the expected fructose (Figure 8). This study is the first to conclusively link molecular structure of an adsorbed phase to SWNT optical properties and modulation in a systematic manner.

Figure 7.

Synthesis scheme of a series of phenylboronic acid conjugated amphiphilic polyethylene glycol polymers presents two steps, including hydrophobicity modification of PEG8 and BA conjugation. (B) Proposed wrapping geometry of different phenylboronic acids on a SWNT (from left to right: 2CPBA–PPEG8, 3CPBA–PPEG8, 4CPBA–PPEG8, N-(4-Phenylboronic)succinamic acid–PPEG8, 5-Bromo-3-carboxy–PPEG8, 3-Carboxy-5-fluoro–PPEG8, and 3-Carboxy-5-fluoro, and 3-carboxy-5-nitro–PPEG8). The highest photoluminescence quantum yield were observed for para-substituted phenylboronic acid (4CPBA), much higher than ortho (2CPBA)- and meta (3CPBA)-substituted phenylboronic acid. Reprinted with permission from Journal of the American Chemical Society.71

Figure 8.

Response of saccharides in different phenylboronic acid–PPEG8 polymer dispersed SWNT solutions. All polymer-wrapped SWNT solutions were diluted using phosphate-buffered saline buffer to a final SWNT concentration of 2 mg/liter. The concentration of SWNT was determined using absorbance at 632 nm with c = 0.036 (mg/liter)^-1 cm^-1. All analytes were dissolved in water at 1 M concentration, and 2 µl analyte solutions were added to 200 µl SWNT, such that the final analyte concentration was 10 mM. The mixture solution was incubated for 1 h before the NIR fluorescence measurement. The intensity change is calculated based on the fluorescence of SWNT (9,4). Error bar is given by repeating three times for each measurement. Dopamine completely quenched fluorescence of SWNTs as a positive control. Reprinted with permission from Journal of the American Chemical Society.71

Opportunities and Challenges for in vivo Continuous Glucose Monitoring

Testing in vivo responses of the SWNT glucose sensors is critical for their clinical applications. A challenge is to improve the sensitivity, selectivity, and lifetime of the sensors in physiologically relevant conditions, including varying oxygen levels, pH, temperature, and potential interferents. An approach to achieve this is to use an internal reference by using two or more different types of SWNTs that emit fluorescence at different wavelengths for sensors and references, respectively. Another task to consider for in vivo applications is to construct microscale structures that encapsulate SWNT-based sensors. For in vivo sensing applications, the direct injection of nanoscale water-soluble SWNT sensors would not be practical, because the nanoscale SWNT sensors can circulate and distribute through the body;⁴⁴ instead, microscale glucose-responsive structure that are small enough to be injected or implemented in the body yet large enough to remain at the implantation site for a long period of time would be required for stable in vivo sensing.^72,73 Because the SWNT sensor is in aqueous suspension, the final sensor architecture is flexible, including dialysis capillary (Figure 5E)^29,31 and hydrogel microparticles that encapsulate SWNT glucose sensors (~100 µm in diameter).^72,73

Another challenge is that sensors implanted in the body often fail because of biofouling of the sensor surface, inflammation, and fibrosis-induced vessel regression at the implantation site.^74–78 An approach to reduce the problems is to coat the surface of the glucose-sensing structure (e.g., hydrogel microparticles encapsulating SWNT glucose sensors) with the antifouling coating and incorporate anti-inflammatory agents, such as dexamethasone (DX), and angiogenic factors, such as vascular endothelial growth factor (VEGF), into the glucose-sensing structure.^74–78 For instance, delivering VEGF-producing cells to the site of glucose sensor implantation using a tissue-interactive fibrin biohydrogel induced significant neovascularization and improved the glucose sensor function in vivo.⁷⁶ The surface coating of hydrogels and microcapillaries with DX and VEGF and the release of DX and VEGF in the tissue reduced the biofouling and developed local environments favorable for long-term, reliable in vivo sensing.^74,79

Testing the toxicity of SWNT sensors is an important issue for in vivo applications. Many studies have investigated the potential toxicity of carbon nanotubes to in vitro cell cultures and in vivo animal models and have shown various results, depending on the type and geometry of nanotubes, nanotube materials preparation, and surface functionalization.²⁷ Properly surface-functionalized carbon nanotubes have not induced obvious toxicity in cell culture experiments and in vivo studies,^80–83 whereas raw carbon has induced pulmonary toxicity and mechanical blockage of the upper airways in animal models.^84–88 Long MWNTs introduced into the abdominal cavity of mice also developed asbestos-like, length-dependent pathogenicity, including inflammation and formation of granulomas.⁸⁸ For the proposed SWNT-based in vivo glucose sensing, the SWNT sensor-encapsulating microstructures, such as hydrogel microparticles and microcapillaries, provide another protective layer (i.e., biocompatible hydrogels) other than biocompatible functionalization layers on SWNTs, which reduce the potential toxicity resulting from the direct contact of SWNTs with biological entities and the needle-like shape of SWNTs.⁸⁸ Considering the growing interest and use of SWNTs for biomedical application, further investigations are required to fully address the SWNT toxicology.^27,88

Conclusion

SWNT-based NIR optical glucose sensors with their extraordinary NIR, nonphotobleaching fluorescence have a great potential for long-term in vivo CGM. A key challenge in designing SWNT-based glucose sensors is to identify and engineer molecular receptors for glucose that will work effectively in a SWNT sensing system and to understand how to effectively transfer glucose-binding events into the modulation of SWNT fluorescence to achieve the selectivity, sensitivity, and dynamic range of the sensors for in vivo glucose monitoring. Advancement of this technology also requires simultaneous research efforts toward translating the laboratory development of SWNT-based NIR optical glucose sensors into in vivo glucose sensing for practical clinical uses. The ability to sense glucose with nanoscale SWNT-based NIR optical sensors would also allow for new ways to answer important biological questions regarding glucose metabolism. The emerging technology of SWNT-based NIR optical glucose sensors can also be extended to various in vivo health monitoring systems.

Glossary

(BA)

boronic acid

(CGM)

continuous glucose monitoring

(CPBA)

carboxyphenylboronic acid

(DX)

dexamethasone

(ICG)

indocyanine green

(GBP)

glucose-binding protein

(MWNT)

multiwalled carbon nanotube

(NIR)

near-infrared

(PPEG8)

polyethylene glycol, eight-member, branched polymer

(PVA)

polyvinyl alcohol

(SC)

sodium cholate

(SWNT)

single-walled carbon nanotube

(VEGF)

vascular endothelial growth factor

Funding

This work is funded by a grant from Sanofi-Aventis.