Optical oxygen sensing with quantum dot conjugates

Sensor design criteria

In order to accurately measure oxygen levels in biological systems, several design elements must be incorporated. An ideal optical sensor should utilize red or near-infrared (NIR) light in the 600–1100 nm range, as tissue is transparent to these wavelengths [42]. This enables deep tissue imaging because photon absorption and scatter is minimized, resulting in maximal signal from the sensor. Ideally, both the excitation and emission wavelengths of the sensor would fall in the tissue transparency window. Two-photon microscopy satisfies this requirement, as it utilizes NIR excitation light. Moreover, this technique provides high spatial resolution (~1 μm) to accurately map the heterogeneity of biological microenvironments [43], [44], [45]. However, most common organic fluorophores are poor absorbers under two-photon excitation [46], [47]. Semiconductor quantum dots (QDs), such as CdSe or CdTe, have two-photon absorption cross-sections that are many orders of magnitude higher than those of traditional fluorophores [48]. As a result, QDs are attractive photon antennas for multiphoton microscopy. Moreover, they have additional unique optical properties, resulting in their widespread use for bioimaging [49], [50].

Perhaps the most important design element for a sensor is the ability to detect the analyte of interest over a desired concentration range. Since QDs are typically insensitive to changes in their environment [51], they must be paired with an analyte-responsive dye to render a functional sensor. This dye is selected on the basis of its optical properties (e.g. emission in the tissue transparency window) and analyte sensitivity over the desired concentration range. For the sensor to function, there must be communication between the QD and the attendant dye. Förster resonance energy transfer (FRET) [52] is a common method of signal transduction in QD/fluorophore dyads [27]; the energy absorbed by one component is non-radiatively transferred to the other without emission of a photon. With two fluorophores in a single construct, the emission intensity ratio of the two components may serve as a quantitative measure of analyte concentration, furnishing a ratiometric sensor. This is a robust sensing methodology that is independent of sensor concentration and is functional in scattering media, such as cells and tissue [28]. The following subsections will review these design principles, which are then applied to the QD-based oxygen sensors described herein.

Semiconductor quantum dots

Inorganic semiconductor nanocrystals, commonly referred to as quantum dots (QDs), are photostable fluorophores with high photoluminescence quantum yields. QDs have broad excitation profiles with high molar absorption coefficients (10⁵–10⁶ M⁻¹ cm⁻¹) and very narrow (~30 nm FWHM), Gaussian-shaped emission profiles (Fig. 1a) [26]. This is in stark contrast to traditional molecular fluorophores, which exhibit narrow absorption profiles with broad emission features that tail to the red and are prone to photobleaching. The emission profile of QDs is tunable with particle size as a result of quantum confinement (Fig. 1b). When a bulk semiconductor absorbs a photon with energy greater than the band gap, an electron is promoted from the valence band (VB) to the conduction band (CB), resulting in the formation of a free hole and free electron. The electron is then attracted to the localized positive charge by Columbic interactions, forming an exciton or bound electron–hole pair. As a result of this stabilizing interaction, the exciton has a lower energy than the free electron and free hole [53]. The spatial extension of the exciton is defined by the Bohr radius. For example, a photogenerated exciton in CdSe will delocalize over a distance of about 12 nm. When the size of the semiconductor particle is on the order of the Bohr radius, the wave function of the exciton is perturbed, modulating the optical properties of the material [54], [55]. The confinement energy of the exciton in the particle becomes larger than the Coulomb energy, giving rise to molecule-like states rather than bands for the bulk material. As a result, the exciton behaves like a particle in a box; the energy (E) depends on the size (r) of the box (E~1/r²) and the effective band gap (E_g) increases with decreasing particle size [56]. Additionally, discrete energy levels arise at the band edges, resembling a molecular HOMO–LUMO gap [26], [55]. In the case of CdSe, small QDs (~2 nm diameter) emit blue light, while large QDs (~6 nm) emit red light [55]. The range of accessible emission wavelengths depends on the band gap of the semiconductor. For example, UV emission is possible with high band gap materials like CdZnS [57], while IR emission can be achieved with low band gap materials like PbS or PbSe [58]. For a more detailed discussion of QD photophysics, the reader is directed to a recent review by Klimov and coworkers [59].

Fig. 1:

(a) Absorption (solid lines) and emission spectra (dotted lines) of a CdSe QD in toluene (—) and Nile blue, a molecular fluorophore, in ethanol (

). The QD exhibits a broad absorption spectrum and sharp emission profile. Conversely, the molecular fluorophore exhibits a broad emission spectrum that tails to the red region of the spectrum. (b) Schematic representation of quantum confinement. As the size of the particle decreases, the exciton becomes confined and behaves like a particle in a box (E~1/r²). The effective band gap E_g increases and discrete, molecule-like energy levels form at the band edges. Reprinted with permission from [28]. Copyright 2014 American Chemical Society.

One of the most common ways to synthesize QDs is by the rapid injection of organometallic precursors into a solvent such as tri-n-octylphosphine oxide (TOPO) in conjunction with a stabilizer (a phosphonic acid or hexadecylamine) at elevated temperatures (>300°C) [26], [60]. Alternatively, octadecene (ODE) may be used as the solvent with oleic acid or oleylamine as a stabilizer. This results in a short burst of homogenous nucleation of nanocrystallites. The sudden drop in temperature from adding room-temperature reagents to the hot reaction mixture prevents further nucleation, maintaining a narrow size distribution of the nanoparticles. Continued heating of the reaction mixture results in nanocrystal growth that may proceed by focused growth and/or Ostwald ripening. Under focused growth conditions, the temperature of the system is low enough to preclude the nucleation of additional particles and there are sufficient remaining precursors. As a result, smaller particles grow at a faster rate than larger particles. Conversely, the precursor concentration is low during Ostwald ripening. The high surface energy of the small particles promotes their dissolution, providing the necessary material to redeposit on larger particles [61]. Since the average size and distribution of particles depends on the growth conditions, the optical properties of the QD can be tuned with fidelity. QDs are often overcoated with another semiconductor in a core–shell motif to further enhance the optical properties by passivating surface defects. If a QD core is coated with a semiconductor that has a higher band gap, then a photogenerated exciton will be confined to the core, preventing it from interacting with the environment and thereby increasing the photoluminescence quantum yield [55], [60]. For example, CdSe QDs overcoated with CdS or CdZnS exhibit stable emission with near-unity (>95%) quantum yields [62], [63]. Conformal shells are usually synthesized using layer-by-layer methods, such as successive ion layer adsorption and reaction (SILAR), which enable the preparation of precise overcoats [64].

As synthesized, quantum dots are hydrophobic and are natively capped with the solvent used in their synthesis. Consequently, the QD surface must be modified to impart water solubility and biocompatibility. This may be achieved by adding a shell of silica in a core–shell motif [65] or using mercaptoacetic acid as a surface ligand [66]. The native hydrophobic capping ligand may be exchanged with a multidentate hydrophilic ligand [67], [68], [69], [70]. Such ligands may be polymeric and incorporate several functional groups for QD surface binding, water solubility, and further derivatization. One example is a dihydrolipoic acid–polyethylene glycol (DHLA–PEG) polymer [71], [72]. While DHLA is commonly used to solubilize QDs in aqueous media, such QDs are not readily derivatized, are unstable at pH<6, and exhibit nonspecific binding in cell cultures. The inclusion of a PEG unit stabilizes the QD in solutions of pH 5.0–9.5 and preserves high photoluminescence quantum yields. Moreover, the PEG polymer may be terminated with an additional functional group to enable conjugation of an analyte-responsive dye [72]. Alternatively, water solubility may also be achieved by encapsulating hydrophobic QDs with oligomeric phosphines [73], dendrimers [74], [75], [76], phospholipids [77], or amphiphilic polymers [78], [79].

Oxygen-responsive phosphorescent molecules

Several methodologies have been used to quantify oxygen in biological systems: electron paramagnetic resonance (EPR), positron emission tomography (PET), magnetic resonance imaging (MRI), hemoglobin saturation spectrometry, polarographic microelectrodes, and phosphorescence quenching [41]. Of these methods, phosphorescence quenching is non-invasive and offers the highest spatiotemporal resolution. When a photon is absorbed (A), a molecule is promoted to an excited singlet state (S₁) (Fig. 2a). It can then undergo vibrational relaxation (VR) to thermally release excess energy and subsequently fluoresce (F) from the lowest vibrational level of S₁. Alternatively, intersystem crossing (ISC) can occur to access the triplet manifold (T₁). Then, as with fluorescence, the molecule can undergo VR to relax to the lowest vibrational state and subsequently phosphoresce (P). In a competing process, the triplet can interact with molecular oxygen (a ground state triplet, ³O₂) via collisional quenching in a triplet–triplet annihilation process, generating singlet oxygen (¹O₂) and returning the molecule back to the ground state (S₀). Consequently, the triplet state will emit fewer photons at higher concentrations of oxygen. This phosphorescence quenching serves as a quantitative measure of pO₂ and this process follows Stern–Volmer kinetics [80]:

Fig. 2:

(a) Jablonski diagram illustrating photophysical processes that can occur after photon absorption. Photonic processes are drawn with straight arrows: absorption (A), fluorescence (F), and phosphorescence (P). Non-radiative processes are denoted with wavy arrows: vibrational relaxation (VR) and intersystem crossing (ISC). Chemical structures of common oxygen-responsive phosphorescent molecules: (b) homoleptic bipyridine (bpy) complexes, (c) heteroleptic phenanthroline (phen)/bipyridine (bpy) complexes, (d) tetraphenylporphyrin, (e) mesoporphyrin IX, (f) triphenylcorrole, and (g) pyrene.

where τ₀ is the natural radiative lifetime of the phosphorescent molecule in the absence of O₂, τ is the phosphorescence lifetime at a given oxygen concentration [O₂], and k_q is the bimolecular quenching rate constant.

In order to have an efficient optical oxygen sensor, the molecule should undergo rapid intersystem crossing upon excitation to maximize the population of the triplet state. The inclusion of heavy atoms will facilitate ISC by the so-called “heavy atom effect” through increased spin–orbit coupling, which scales with atomic number (Z⁴). As a result, the population of the triplet manifold increases at the expense of the singlet, resulting in fluorescence quenching. However, the molecule must phosphoresce rather than simply non-radiatively relax back to the ground state without emitting a photon. This can be achieved with closed-shell coordination complexes of second and third row transition metals. Figure 2 shows the chemical structures of common oxygen-responsive compounds. Perhaps the most widely utilized phosphorescent molecules are Ru(II) and Os(II) polypyridine complexes (Fig. 2b and c), which possess triplet metal-to-ligand charge transfer (³MLCT) states with lifetimes up to 6.4 μs and quantum yields as high as 30% in the case of [Ru(dpp)₃]²⁺, where dpp=4,7-diphenyl-1,10-phenanthroline [81]. In general, Os(II) derivatives emit at longer wavelengths than the analogous Ru(II) complexes, but they have significantly shorter lifetimes (~10² ns) and, consequently, lower quantum yields (φ~1%) [38]. With lifetimes of several microseconds, Ru(II) polypyridine complexes are sensitive over the 0–760 Torr O₂ range [82].

In addition to polypyridine complexes, metalloporphyrins (Fig. 2d and e) are common phosphorescent molecules for oxygen sensing applications. While a variety of metals have been utilized, including Ir(III) [83] and Gd(III) [84], the most prevalent examples are Pd(II) and Pt(II) because of their strong room-temperature phosphorescence in the 650–800 nm range and long (~10² μs) phosphorescence lifetime [85]. In general, longer phosphorescence lifetimes increase the sensitivity of the sensor at lower oxygen concentrations. These attributes render Pd(II) and Pt(II) porphyrins responsive over the biologically relevant 0–160 Torr O₂ range and emissive in the tissue transparency window. In general, Pd(II) porphyrins have longer lifetimes, redder emission profiles, and higher oxygen sensitivity than their Pt(II) analogs. Conversely, Pt(II) porphyrins have higher quantum yields than their Pd(II) congeners [40]. Commercial Pd(II) porphyrins (Oxyphor R2) and benzoporphyrins (Oxyphor G2), which are available as glutamate dendrimers, are water-soluble sensors that have been used for solution-based sensing applications [86]. Another approach to make these porphyrins water soluble is to incorporate them into a protein scaffold. Mesoporphyrin IX complexes (Fig. 2e) of Ru(II) [87], Pd(II) [88], and Pt(II) [88] have been incorporated into the heme binding pocket of myoglobin and horseradish peroxidase to furnish oxygen-responsive proteins. Recently, corroles (Fig. 2f) have also been explored as oxygen sensors. Corroles are 18 π-electron tetrapyrrole macrocycles, like porphyrins, but contain a contracted 23-atom core. As a result, corrole is a trianionic ligand when fully deprotonated, in contrast to the dianionic porphyrin ligand. Consequently, Au(III) corroles, which are isoelectronic with Pt(II) porphyrins, have been reported as optical oxygen sensors [89]. These complexes emit in the 780–795 nm region with triplet lifetimes of ~80 μs, rendering these effective sensors over the 0–160 Torr O₂ range. Similarly, Os(VI)≡N corroles are oxygen sensors that emit in the same spectral range with longer lifetimes and higher quantum yields [90].

One additional class of oxygen-responsive compounds is polycyclic aromatic hydrocarbons (PAH), such as perylene and pyrene (Fig. 2g) [38], [40]. These molecules typically emit in the near-UV to blue region of the spectrum and have high fluorescence quantum yields. For example, pyrene exhibits an emission maximum at 393 nm with ϕ=65% [91]. The emission profiles of PAHs may be red shifted by increasing the size of the aromatic π system (e.g. λ_max,em~445 nm for perylene [92]). While most typical fluorophores have excited state lifetimes between 1 and 10 ns that are independent of oxygen, PAHs have τ>100 ns that are readily quenched by oxygen. This is due to a triplet–triplet annihilation process known as P-type delayed fluorescence [93]. The “P” designation stands for pyrene [94], the first molecule that was discovered to exhibit this phenomenon [91]. In this process, two molecules in the triplet state interact to form an excited state complex that subsequently dissociates to give one molecule in the excited state and one molecule in the ground state. As a result, fluorescence is observed on the timescale of triplet states, enabling optical oxygen sensing. It should also be noted that molecular oxygen induces ISC in certain PAHs (e.g. 9,10-diphenylanthracene), resulting in fluorescence quenching [95], [96].

Förster resonance energy transfer

If an oxygen-responsive phosphorescent molecule is coupled to a QD, then energy must be transferred between the two components to observe oxygen-dependent emission from QD excitation. One of the most common methodologies for signal transduction is Förster resonance energy transfer (FRET) [52]. In this mechanism, energy is non-radiatively transferred via long-range dipole–dipole interactions from a donor (D) fluorophore to an acceptor (A) molecule (Fig. 3a), resulting in a decrease of the donor’s emission intensity. Upon absorption of a photon, one of the electrons in the HOMO of the donor is promoted to the LUMO (Fig. 3c). During FRET, the donor returns to the ground state without emitting a photon, while, simultaneously, the acceptor is promoted to an excited state [97]. In a FRET-based sensor, the donor is selected for its absorption properties, while the acceptor is chosen based on its emission properties and analyte sensitivity [27], [28]. The FRET pair is selected to maximize the spectral overlap between donor emission and acceptor absorption (Fig. 3b) for efficient energy transfer.

Fig. 3:

(a) Schematic of a QD donor (D)/fluorophore acceptor (A) construct. The QD serves as a photon antenna and transfers the absorbed energy to the acceptor, promoting it to an excited electronic state. (b) Illustration of spectral overlap between donor emission (

) and acceptor absorption (

). (c) Molecular orbital representation of FRET. Adapted with permission from [28]. Copyright 2014 American Chemical Society.

Mathematically, the FRET efficiency (E) may be quantified in terms of energy transfer rates or donor–acceptor distances:

where m is the number of acceptor molecules per donor molecule, k_D−A is the rate of energy transfer, τ_D is the lifetime of the donor in the absence of acceptor, R₀ is the Förster distance (or the distance at which FRET efficiency is 50%), and r is the distance between the donor and acceptor [97]. Based on this equation, E increases with increasing values of m, which can be readily achieved in QD-based sensors by incorporating many acceptor molecules per QD donor. When m>1, the parameter r reflects the ensemble average of D–A distances. The FRET efficiency may be determined experimentally by measuring the lifetime of the donor:

where τ_D−A is the lifetime of the donor in the presence of acceptor. The Förster distance R₀ can be calculated for a given donor–acceptor pair:

where κ² is the relative orientation factor of the transition dipoles (taken to be 2/3 for random D–A orientations), Φ_D is the fluorescence quantum yield of the donor, N is Avogadro’s number, n is the refractive index of the solvent, F_D(λ) is the normalized emission intensity of the donor, and ε_A(λ) is the molar absorption coefficient of the acceptor at wavelength λ. The latter half of this equation is known as the spectral overlap integral and is often denoted by the variable J. Critical FRET length scales for R₀ typically range from 2 to 9 nm [97]. Since the emission profile of QDs is tunable with size, QDs are attractive FRET donors because the overlap integral may be tuned with fidelity.

In an alternative to the dipolar Förster interaction, energy may be transferred in an exchange-driven process known as Dexter energy transfer [98]. This mechanism relies on wave function overlap and requires close contact between the donor and acceptor. This can be achieved by using high concentrations of both donor and acceptor, or through direct physical or chemical contact between the two components. In canonical Dexter energy transfer, an excited electron in the LUMO of the donor is transferred to the acceptor molecule. The acceptor then transfers an electron from its HOMO to the HOMO of the donor, resulting in an excited-state acceptor and ground-state donor. Typically, Dexter energy transfer is associated with fluorescence quenching and can be observed if J is small, as electron transfer rates then become significant. When J is high, FRET will occur before Dexter transfer [97]. This has been demonstrated experimentally with a series of QDs of varying size conjugated with a surface-bound squaraine dye [99]. QD photoluminescence experiments demonstrated that smaller QDs (low spectral overlap and high orbital overlap with the dye) exhibit higher energy transfer efficiencies than predicted by FRET and faster rates of energy transfer. These results suggest that Dexter transfer dominates in these constructs. Conversely, in conjugates with larger QDs (high spectral overlap and low orbital overlap with the dye), FRET is the predominate mechanism of energy transfer. This example demonstrates that both mechanisms may be operative in QD conjugates, but the dominant mode of energy transfer depends on spectral and orbital overlap.

Two-photon spectroscopy

The theory of two-photon absorption was first proposed by Maria Göppert-Mayer in 1931 [100]. It was not until 30 years later that this concept was experimentally verified when Kaiser and Garrett used red light from a ruby laser to excite a sample of Eu(II)-doped CaF₂ and detected blue fluorescence [101]. Until 1990, multiphoton spectroscopy was considered an exotic technique that was limited to the fields of optical spectroscopy and chemical physics. This was because the availability of photon sources that could provide sufficiently high peak powers to increase the probability of multiphoton absorption was limited [97]. With the advent of mode-locked solid-state femtosecond lasers (e.g. Ti:sapphire), the accessibly of multiphoton spectroscopy has dramatically increased [46]. In 1990, two-photon imaging of biological samples was first reported using an ultrafast laser source [102]. Since then, multiphoton laser scanning microscopy (MPLSM) has become a powerful, routine imaging technique that uses NIR light to exploit the tissue transparency window and enable deep tissue imaging (450–600 μm) with approximately 1 μm spatial resolution [42], [44], [45], [103]. Given the small focal volume of two-photon excitation (vide infra), photobleaching and photodamage are minimized while generating high-resolution images to accurately map the heterogeneity of the tumor microenvironment [104], [105], [106].

Under linear or one-photon excitation, a molecule is directly promoted to the first excited state through the absorption of a photon with an energy comparable to the HOMO–LUMO gap (Fig. 4a). Multiphoton processes require two or more photons to simultaneously interact with a molecule. In a two-photon process, the first photon excites the molecule to an intermediate state while the second photon promotes the molecule to its final state (Fig. 4b). The intermediate state can be an eigenstate of the molecule (i.e. a real state for sequential two-photon absorption) or a superposition of molecular states (i.e. a virtual state for simultaneous two-photon absorption) [46]. The intensity of a two-photon transition, or the two-photon absorption cross-section σ₂, is analogous to the molar absorption coefficient ε for linear excitation and can be estimated using a single intermediate state approximation [107]:

Fig. 4:

Comparison of one- versus two-photon excitation processes. (a) A fluorophore is promoted to an excited state by a single blue photon and subsequently emits a green photon. (b) The same transition in (a) is achieved with two red photons whose total energy is equivalent to that of the blue photon. (c) One- and (d) two-photon excitation of fluorescein. Under linear excitation, fluorescence is observed throughout the sample. Conversely, two-photon excitation exhibits sharp, pinpoint emission at the focal point of the excitation source, highlighted with a yellow circle. Reprinted with permission from [28]. Copyright 2014 American Chemical Society.

where σ_ij is the one-photon absorption cross-section for the transition from initial state i to intermediate state j, σ_jf is the one-photon absorption cross-section for the transition from intermediate state j to final state f, and τ_j is the lifetime of the intermediate state and determines the time scale for photon coincidence. For a virtual state, τ_j≤10⁻¹⁵ s and simultaneous photon absorption occurs. Alternatively, sequential photon absorption can occur if the intermediate state is a real state with a lifetime of 10⁻⁹ to 10⁻¹² s. A one-photon absorption cross-section σ₁ may be estimated using the length of the transition dipole, which is ~10⁻⁸ cm for a typical organic fluorophore, to give σ₁~10⁻¹⁶ or 10⁻¹⁷ cm². Using these values and Eq. 5, the corresponding two-photon absorption cross-section may be estimated as 10⁻⁴⁹ cm⁴ s/photon or 10 Göppert-Mayer (GM), where 1 GM=10⁻⁵⁰ cm⁴ s/photon.

Under linear excitation, the excitation density in the focal region is proportional to the intensity of the incident light. Since the excitation density depends on the square of the intensity for a two-photon process, the excitation density rapidly decreases as the distance from the focal point increases. As a result, the two-photon excitation volume is significantly smaller than that of one-photon excitation, thereby increasing the spatial resolution of the signal. This phenomenon is illustrated in Fig. 4c and d. Under one photon excitation, a streak of emission is visible along the focal plane, while emission is visible only at the focal point of the excitation source under two-photon excitation. Consequently, two-photon excitation minimizes photobleaching and photodamage.

While multiphoton imaging is a powerful technique, there are few dyes that have sufficiently high two-photon absorption cross-sections. Most typical fluorophores have σ₂~10–100 GM [45], [47], [108], [109]. For fluorescein, a common dye used in biological labeling, σ₂ varies from 8 GM to 37 GM over the 690–960 nm range [108]. Some conventional fluorophores exhibit unusually high cross-sections, such as Cy3 and Rhodamine 6G, which have maximal σ₂ values of 140 GM and 150 GM, respectively, under 700 nm excitation [108]. Although organic molecules have been developed with σ₂~10³–10⁴ GM [110], [111], common analyte-responsive dyes have prohibitively low two-photon absorption cross-sections for MPLSM applications. Conversely, QDs possess large two-photon absorption cross-sections [112], [113], [114], with σ₂ as high as 47000 GM for CdSe/ZnS core–shell QDs [48]. This property renders QDs suitable two-photon antennas for an appended analye-sensitive dye to enable biological sensing with MPLSM.

Solid-state sensors

The most common types of optical oxygen sensors are solid-state devices where the phosphorescent molecule is immobilized in a polymer matrix, on a solid surface, or in mesoporous silica [85]. Therefore, it is not surprising that the majority of QD-based oxygen sensors are solid-state constructs where a QD and O₂-responsive phosphorescent molecule are blended in a polymer or sol–gel matrix. Since there are two emissive components, the construct is ratiometric; the QD serves as an internal emission intensity standard and the ratio of QD to phosphorescent molecule emission is a quantitative measure of pO₂. This is a significant improvement over sensors with only a phosphorescent molecule that rely on changes in absolute emission intensity. The first sensor of this type consisted of a CdSe/ZnS QD (λ_em=520 nm) and [Ru(bpy)₃]Cl₂ blended in a sol–gel matrix and coated on a glass slide [115], [116]. A later iteration of this sensor design utilized [Ru(dpp)₃]Cl₂ as the phosphorescent molecule with CdTe QDs (λ_em=680 nm) [116].

Solid-state sensors have primarily utilized Pd(II) and Pt(II) porphyrins because they offer greater oxygen sensitivity than Ru(II) polypyridine complexes. Both metalloporphyrins have been paired with CdSe QDs (λ_em=515 nm) in an organically modified silica (ormosil) xerogel and coated on a fiber optic probe [117]. A ratiometric sensor strip was prepared by coating a glass slide with CdTe/CdS QDs (λ_em=552 nm) dispersed in a poly-silane. This optical reference layer was then coated with an ormosil matrix containing a Pt(II) porphyrin (λ_em=648 nm). The sensor exhibited a gradual red to green optical transition with increasing pO₂ (Fig. 5) [118]. Variations on these methods have also led to the development of dual sensors for measuring oxygen in conjunction with another analyte of interest. For O₂ and temperature sensing, CdSe QDs (λ_em=532 nm) overcoated with a silica shell and a Pt(II) porphyrin were embedded in a sol–gel matrix. The QD exhibited temperature-dependent emission over the 30–100°C range. At constant temperature, QD emission serves as an internal intensity standard for ratiometric oxygen sensing [119]. A similar approach was utilized for measuring oxygen and Cu(II) ions, where the photoluminescence of CdSe QDs (λ_em=580 nm) was quenched by copper ions [120]. A later iteration of this sensor included a coumarin dye as an internal reference that was invariant to both analytes [121]. Dual O₂ and H₂O₂ detection was accomplished by blending CdSe/ZnS QDs (λ_em=517 nm) and [Ru(dpp)₃]Cl₂ in an ethyl cellulose matrix and coating on an optical fiber. QD photoluminescence was quenched by H₂O₂ and served as the means of detection of this analyte [122].

Fig. 5:

Optical output of a solid-state sensor strip comprised of a Pt(II) porphyrin and a CdTe/CdS QD. The top row shows the emission of the Pt(II) porphyrin alone, while the bottom row shows the ratiometric output of the combined porphyrin/QD sensor. Adapted with permission from [118]. Copyright 2008 Wiley-VCH Verlag GmbH & Co. KGaA.

In addition to these fiber optic and other monolithic devices, ratiometric oxygen sensors have been prepared in polymeric micro- or nanoparticles. The first example of this type of sensor was reported by McShane and co-workers in 2011 [123]. Amine-functionalized porous silica microspheres (10.3 μm diameter) were treated with QDs (λ_em=786 nm) and a Pt(II) porphyrin, both bearing carboxylic acid moieties for conjugation via amide bond formation. One advantage of this sensor is that both components emit in the NIR. The sensor exhibited a linear optical response at [O₂]<400 μM (240 Torr). Recently, a ratiometric nano-sensor was reported where a cyclometalated Ir(III) complex was used as the O₂-responsive phosphorescent molecule [124]. QDs (λ_em=690 nm) and Ir(piq)₃ (piq=1-phenylisoquinoline with λ_em=620 nm) were dispersed in a liquid crystalline phase of monooleoylglycerol; sonication was employed to exfoliate particles with a 35.6 nm average diameter. The small size of the particles makes the sensor amenable for biological imaging applications (vide infra). The 750 nm to 620 nm emission intensity ratio was used to quantify pO₂ and this parameter exhibited a linear response from 1 to 21% (8–160 Torr) O₂.

In all of the sensors discussed so far, the QD serves as an internal intensity standard for ratiometric sensing, as the phosphorescent molecule and QD are spectrally separated and do not interact. There has been one example of a solid-state sensor where the QD was utilized as a FRET donor. A CdSe/CdS QD (λ_em=590 nm) was selected such that its emission profile overlapped with the absorption of Pt(II) octaethylporphyrin ketone and the two components were blended in a 1:3 ratio in a polyvinyl chloride (PVC) matrix [125], [126]. This sensor has been utilized for simultaneous O₂ sensing and voltage monitoring in biological tissue (vide infra). Since this sensor operates via FRET, it is functional under both one- (λ_ex=480 nm) and two-photon (λ_ex=850 nm) excitation. The 590 nm to 757 nm emission ratio served as a quantitative measure of pO₂ [126].

Covalent conjugates

While solid-state sensors may be the most prevalent, these devices are generally not amenable for biological sensing, particularly for in vivo applications. Instead, a soluble, solution-based sensor is preferred so that it may be efficiently distributed throughout a biological sample. One of the most common ways in which to prepare such a sensor is to covalently attach an analyte-responsive fluorophore to an appropriately functionalized QD. This approach has been widely utilized to prepare QD-based sensors for a variety of analytes including pH, metal cations, proteins, and other molecules of interest [28], [32], [33], [36]. The first example of a solution phase QD-based oxygen sensor was reported in 2009 by Nocera and co-workers [127]. A ZnSe/CdSe/CdZnS core–shell–shell QD (λ_em=548 nm) was paired with [Os(bpy)₃][PF₆]₂ or [Os(bpy)(phen)₂][PF₆]₂ as the oxygen-responsive phosphorescent molecule. The QD was solubilized in water using n-octylamine-functionalized poly(acrylic acid) with terminal carboxylic acid groups to couple to amine-functionalized Os(II) complexes (Fig. 6a). Since this QD ligand features many surface carboxylates, high coupling efficiencies are achieved to give 57 [Os(bpy)₃]²⁺ or 135 [Os(bpy)(phen)₂]²⁺ complexes per QD. The emission intensity of the QD is attenuated while that of the Os(II) complex is enhanced in the conjugate relative to the free components, indicative of a FRET interaction. This is further substantiated by an increase in the ³MLCT quantum yield of the phosphorescent molecule with a concomitant decrease in the QD photoluminescence lifetime. The FRET efficiency for these sensors is quite high: 67% (R₀=3.4 nm) for [Os(bpy)₃]²⁺ and 50% (R₀=4.0 nm) for [Os(bpy)(phen)₂]²⁺. This property makes the sensor functional under two-photon excitation (λ_ex=920 nm); emission from the Os(II) complexes alone is nearly undetectable. The ³MLCT emission is quenched by ~20% under 760 Torr O₂, making the sensor sensitive at higher oxygen tensions.

Fig. 6:

(a) Schematic representation of the ZnSe/CdSe/CdZnS QD (NC)–Os(II) polypyridine sensor. Reprinted with permission from [127]. Copyright 2009 American Chemical Society. (b) Schematic representation of the AgInS₂/ZnS QD–perylene sensor. Reprinted with permission from [128]. Copyright 2016 American Chemical Society.

Snee and co-workers have reported constructs with perylene [128] and pyrene [129] as the oxygen-responsive dye. For these sensors, the QD and PAH are spectrally separated and do not interact by FRET. A non-toxic AgInS₂/ZnS core–shell QD (λ_em=775 nm) was solubilized in water with n-octylamine-functionalized poly(acrylic acid). The surface carboxylic acid moieties were conjugated with amine-functionalized perylene (λ_em=472 nm) to afford a ratiometric sensor (Fig. 6b). Since this construct is responsive over physiologically relevant pO₂ (0.3–213 μM or 0.2–129 Torr), it was used to measure oxygen levels in HeLa cells (vide infra) [128]. Similarly, a CdSe/CdZnS QD (λ_em=595 nm) was treated with amine-functionalized pyrene, giving ~100 chromophores per QD. This sensor exhibited oxygen sensitivity from 0–100% (0–760 Torr) O₂. Next, this sensor design was applied to Mn(II)-doped magnetic QDs: ZnSe/ZnMnS/ZnS and CdSe/CdMnZnS. Interestingly, pyrene did not exhibit oxygen-dependent emission in either of these constructs. The authors speculate that the microenvironment of the QD restricts the oxygen–pyrene interaction into orientations that are perturbed by the magnetic field induced by the Mn(II) ions. However, the underlying physical basis for this phenomenon remains an open question [129].

Self-assembled sensors

Self-assembly is a facile and scalable alternative to covalent strategies for the preparation of QD conjugates. This methodology circumvents the laborious preparation of amphiphilic polymers and allows the donor–acceptor ratio to be tuned with fidelity. The QD can be viewed as a large metal ion that can bind a variety of functional groups including imidazole, amines, and carboxylic acids on its surface. An acceptor molecule with one of these functional groups can exchange with the hydrophobic ligands on the QD surface. The first example of this type of sensor was reported by Credi and co-workers in 2011 [130]. A hydrophobic, TOPO-capped CdSe/ZnS QD (λ_em=636 nm) was treated with imidazole-functionalized pyrene (λ_em=398 nm), which binds directly to the QD surface. As with the PAH sensors described the previous section, the two components are spectrally separated and do not interact via FRET. The sensor is functional over the 0–760 Torr O₂ range and the 632 nm to 398 nm emission ratio serves as a suitable parameter for ratiometric sensing.

In order to red shift the emission profile of the sensor and increase the sensitivity at physiological O₂ pressures, Nocera and co-workers paired Pd(II) porphyrins (λ_em=690 nm) with CdSe core–shell QDs (λ_em=519 nm) [131]. Pyridyl groups at the meso positions of the porphyrin enable direct binding to the QD surface (Fig. 7). The sensors were prepared by adding 10 equivalents of porphyrin per QD. It was found that porphyrins with two pyridyl substituents in a cis arrangement have high binding affinity (K_A~10⁷ M⁻¹), whereas porphyrins with a single pyridyl group exhibit weaker binding. Time-resolved porphyrin emission demonstrates that there are two populations in the self-assembled constructs, corresponding to surface-bound (75%) and free porphyrin (25%). This is also reflected in titration data, which exhibits binding saturation with ~8 equivalents of added porphyrin. The QD was selected such that its emission profile overlaps with Pd(II) porphyrin absorption (λ_abs=523 nm), resulting in high FRET efficiencies (94% with R₀=4.1 nm). As a result, the sensor functions under two-photon excitation (λ_ex=800 nm), while emission from the porphyrin alone is undetectable. Under ambient air, porphyrin phosphorescence is significantly quenched and the emission intensity increases 100-fold in the absence of oxygen, rendering the sensor sensitive over the 0–160 Torr O₂ range. Since QD photoluminescence is unaffected by O₂, ratiometric sensing is possible [131].

Fig. 7:

Schematic representation of the QD–Pd(II) porphyrin self-assembled sensor. Upon porphyrin binding, QD photoluminescence is quenched and the absorbed energy is transferred to the surface-bound porphyrin by FRET, promoting it to an excited state. Porphyrin phosphorescence is quenched by oxygen while the QD is unaffected, resulting in a ratiometric optical response. Reprinted with permission from [131]. Copyright 2013 American Chemical Society.

This approach was then extended to Au(III) corroles to further shift the optical readout of the sensor to the NIR (λ_em=780 nm) [132]. Since Au(III) corroles have redder absorption profiles than those of Pd(II) porphyrins [89], a yellow-emitting CdSe/ZnS QD was selected (λ_em=572 nm) to maximize spectral overlap and FRET efficiency. Corroles with a meso-carboxylic acid group enable high-affinity binding to the QD surface with K_A~10⁶ M⁻¹. Conversely, the corresponding methyl ester exhibits drastically diminished affinity (K_A~10⁴ M⁻¹), as this molecule can only associate with the QD through non-specific hydrophobic interactions. This difference in affinity is also reflected in the FRET efficiencies and average donor–acceptor distance (r): 84% (R₀=5.12 nm, r=5.70 nm) for the carboxylic acid and 22% (R₀=5.03 nm, r=9.12 nm) for the corresponding methyl ester. The high FRET efficiency renders the sensor functional under two-photon excitation (λ_ex=965 nm) [132]. Since these QDs are larger than those used for the Pd(II) porphyrin sensors [131], binding saturation is not observed, even up to 20 equivalents of corrole. Additionally, monoexponential decay kinetics are observed for Au(III) corrole emission in these constructs, indicating that all the corrole molecules are bound to the QD. The oxygen sensitivity of these conjugates is comparable to that of the free Au(III) corrole and is responsive over the 0–160 Torr O₂ range [132].

Micelle-encapsulated sensors

The sensors described in the previous section rely on the hydrophobicity of the QD for self-assembly and are functional in organic solvents rather than in aqueous media. However, another self-assembly process may be leveraged to solubilize those sensors in water: micelle formation. Micelles are comprised of amphiphilic molecules that assemble such that the hydrophobic ends of the molecule reside in the micelle interior, while the hydrophilic groups on the exterior interact with solvent. Through hydrophobic interactions with surface ligands, natively capped QDs can readily be solubilized in water by micelle encapsulation. Phospholipids have been used to accomplish this and these micelles are amenable for biological imaging [77]. This method has been used to encapsulate the self-assembled sensors with Pd(II) porphyrins [133] and Au(III) corroles [132] described in the previous section. The organic-soluble sensors were treated with a PEG-2000-functionalized phosphoethanolamine lipid using sonication processing to afford sensors with a hydrodynamic diameter of 18.2 nm in the case of Pd(II) porphyrins (Fig. 8). Electron microscopy demonstrates that a single QD is contained within each micelle. Based on the optical spectra of the sensors, there are eight porphyrins per QD [133], which is consistent with the photophysical characterization of the organic-soluble sensors [131]. The sensor exhibits 99% FRET efficiency (R₀=4.06 nm) to give an average donor–acceptor distance of 2.67 nm. This is identical to the QD to porphyrin center-to-center distance based on structural metrics, where the QD radius is 1.90 nm and the distance between the pyridyl nitrogen and the Pd(II) center is 0.77 nm. Under linear excitation, two porphyrin populations are observed: one that is surface-bound to the QD and one that is dispersed in the phospholipids of the micelle. Conversely, a single population of porphyrins is observed under two-photon excitation (λ_ex=850 nm), as only the QD-bound porphyrins are sufficiently close to the QD to be excited via FRET. As a result, two-photon excitation is ideal for in vivo applications of this sensor, as the experiment naturally selects for a single population of porphyrins that is directly associated with the QD, simplifying the photophysics of the system. An oxygen-dependent ratiometric response is achieved with the 682 nm to 528 nm ratio. The sensor was calibrated at both 25°C and 37°C to give k_q values of 9.02×10⁸ M⁻¹ s⁻¹ and 1.36×10⁹ M⁻¹ s⁻¹, respectively. This latter value was used to quantitatively measure oxygen concentrations in vivo (vide infra).

Fig. 8:

Schematic representation of the micelle-encapsulated QD–Pd(II) porphyrin sensor. Oxygen can freely diffuse into the micelle and quench the Pd(II) porphyrin. QD photoluminescence is unaffected, resulting in a ratiometric response. Reprinted with permission from [133]. Copyright 2015 American Chemical Society.

Similarly, the Au(III) corrole self-assembled sensors were encapsulated in the same phospholipid. It was found that both the carboxylic acid and methyl ester derivatives were incorporated into the micelle constructs, demonstrating that direct surface binding is not a prerequisite for micelle encapsulation. However, the carboxylic acid, which directly binds to the QD surface, resulted in higher FRET efficiency and closer average donor–acceptor distances (r) than the methyl ester: 93% (R₀=4.13 nm, r=3.94 nm) and 66% (R₀=4.06 nm, r=5.24 nm), respectively [132]. The D–A distance of 3.94 nm determined from the FRET analysis is consistent with the structural metrics of the construct, where the QD radius (2.88 nm) and the distance between the carboxylic acid group and the Au(III) center (1.11 nm) results in a center-to-center distance of 3.99 nm [134]. As in the case of the Pd(II) porphyrin micelles [133], two corrole populations are observed under linear excitation and correspond to QD-bound corrole and free corrole dispersed in the lipids of the micelle. The surface-bound corrole population serves as a reliable metric to measure pO₂ and this population exhibits a k_q value of 1.6×10⁹ M⁻¹ s⁻¹ [132].

While micelle-encapsulation of self-assembled sensors is a facile way to produce water-soluble QD conjugates, there are several limitations to this methodology. First, the acceptor must be appropriately functionalized to enable sufficient QD surface binding and to maximize incorporation in the micelle. Moreover, the size of the QD limits the number of acceptors that can be accommodated on the surface. For example, only ~8 Pd(II) porphyrins could bind to the surface of a 3.5 nm diameter QD [133], while at least 20 Au(III) corroles can bind to the surface of a 5.8 nm diameter QD [132]. If the quantum yield of the acceptor is low, the emission may not be comparable with QD photoluminescence, resulting in QD emission that is off-scale relative to the phosphorescent molecule. In the case of the Au(III) corrole sensors, QD emission (ϕ=0.70) is substantially greater than corrole phosphorescence (ϕ=0.0024) and 20 equivalents are insufficient to quench QD photoluminescence and increase corrole emission such that the intensity of the two components are comparable [132]. Additional acceptor molecules are needed than can be physically accommodated on the QD surface. Finally, micelle constructs can only be used to sense analytes that readily diffuse though the lipid layer to reach the analyte-responsive dye buried in the micelle core. To overcome these limitations, the micelle approach can be extended beyond the encapsulation of self-assembled constructs by utilizing phospholipids modified with functional groups at the end of the PEG chain. In this way, the QD can readily be solubilized in water and covalent conjugation strategies may be used to functionalize the exterior of the micelle with acceptor molecules. By using a mixture of functionalized and unfunctionalized lipids, the number of surface groups may be adjusted to easily tune the donor–acceptor ratio. This method has been applied to the preparation of a sensor construct where a water-soluble Au(III) corrole with carboxylic acid moieties was covalently attached to a micelle-encapsulated QD with surface amines. As a result, more corroles were coupled to the QD than are possible via self-assembly methods, resulting in comparable emission intensity from both components [135].

As an alternative to phospholipid micelles, QD constructs may also be encapsulated in amphipilic polymers. Polyethyleneimine was used to encapsulate hydrophobic CdSe/CdS/ZnS QDs (λ_em=610 nm) to give particles with a hydrodynamic diameter of ~100 nm that contained dozens of QDs. In order to furnish a functional oxygen sensor, a chloroform solution of QDs (λ_em=470 nm) and [Ru(dpp)₃]²⁺ in a 1:3 ratio was treated with the polyethyleneimine polymer to give <100 nm particles. The two components are spectrally separated and do not interact via FRET. The construct exhibited a ratiometric response from 0–152 Torr O₂ (0–20% O₂). These nanoprobes were used to measure oxygen concentrations in cells and tumor spheroids (vide infra) [136].

Biological applications

Despite the variety of construct designs and diversity of O₂-responsive phosphorescent molecules, only a few of these sensors have been applied to study biological systems: ex vivo tissue samples, live cells, and animals. The CdSe/CdS QD and Pt(II) octaethylporphyrin ketone sensor dispersed in a PVC matrix was used for oxygen sensing in ex vivo brain tissue [125], [126]. First, a glass slide coated with the sensor was used to monitor oxygen levels in a murine hippocampal slice. The tissue exhibited a radial oxygen gradient from 18 mg/L (560 μM or 340 Torr) to 5 mg/L (160 μM or 100 Torr) at a distance of 250 μm from the edge of the sample, despite perfusing the tissue with gas containing 31 mg/L (970 μM or 590 Torr) O₂ during data acquisition. Next, the potassium channel blocker 4-aminopyridine was added to induce hyperexcitability. This resulted in spontaneous seizure-like events (SLE) that caused a decrease of oxygen that was attributed to oxygen consumption in the tissue. The sensor was also coated onto a microelectrode, enabling simultaneous voltage and O₂ sensing in a small 25 μm×25 μm region of the hippocampus. These measurements also demonstrated a decrease in pO₂ during the spontaneous SLEs that resulted in a prolonged period of hypoxia that persisted for >30 s after completion of the SLE. This approach was further miniaturized to measure pO₂ using a fine glass electrode with the sensor matrix in conjunction with an intracellular electrode to measure these two parameters in a single cell, reducing the area to 3 μm×3 μm [126].

In contrast to solid-state devices, soluble oxygen sensors are more amenable for measuring pO₂ in live cells. Human colon cancer cells (HCT116) were treated with the polyethyleneimine-encapsulated QD/[Ru(dpp)₃]²⁺ sensor and cultured under normoxic or hypoxic conditions [136]. It was found that the pO₂ in these samples was 20.5% (156 Torr) and 1.1% (8 Torr), respectively. Similarly, this sensor was used to study a tumor spheroid, which mimics metastatic tumors, using the HCT116 cell line. The spheroid exhibited a radial oxygen gradient with the core being most hypoxic (1.3% or 10 Torr); pO₂ increased to 8.4% (64 Torr) halfway from the center, and finally 19.9% (151 Torr) at the periphery of the spheroid. Similarly, the QD/perylene sensor was used to measure oxygen concentrations in HeLa cells [128]. A 30% increase in the perylene:QD emission ratio was observed for cells cultured overnight under 6.9% (52 Torr) vs. 19.8% (150 Torr) oxygen (Fig. 9a). The Ir(piq)₃/QD nanoparticle sensor embedded in a monooleoylglycerol matrix was also used to examine pO₂ of HeLa cells cultured under different oxygen tensions [124]. The Ir(piq)₃:QD emission ratio increased from 1.42 to 3.96 for cells cultured under 21% (160 Torr) and 1% (8 Torr) O₂, respectively.

Fig. 9:

(a) Ratiometric response of the QD/perylene oxygen sensor in HeLa cells cultured under 19.8% (150 Torr) or 6.9% (52 Torr) O₂. Reprinted with permission from [128]. Copyright 2016 American Chemical Society. Three-dimensional depth projections of murine brain vasculature under two-photon excitation (λ_ex=850 nm), showing the (b) QD and (c) Pd(II) porphyrin emission channels. Reprinted with permission from [133]. Copyright 2015 American Chemical Society.

To date, there have been few reports of in vivo oxygen sensing with QD conjugates. In one experiment, HeLa cell tumor xenografts were implanted on the leg of nude mice. Oxygen levels, as measured by the Ir(piq)₃/QD sensor, in the tumor and muscle tissue were compared and it was found that the Ir(piq)₃:QD emission ratios were 1.79 and 0.93, respectively [124]. As expected, the tumor exhibited higher Ir(piq)₃ phosphorescence and was more hypoxic than muscle tissue and the sensor was able to qualitatively differentiate these two tissues. However, differences between the emission ratios observed for in vitro calibration and in vivo measurements prevented the translation of these values into precise oxygen concentrations. Indeed, this remains an unsolved problem in the field. The identity of the tissue, as well as the depth at which the image is acquired, has a profound effect on the observed intensity ratio [137]. Optical properties, such as absorption and scattering coefficients, are tissue-dependent [138]. Consequently, the collection efficiency of photons from the QD and phosphorescent molecule may not be identical, resulting in differences between in vivo ratio measurements and in vitro calibrations [135]. One way to circumvent this limitation is to perform lifetime-based measurements, as this parameter is independent of the optical properties of the tissue. The micelle-encapsulated QD/Pd(II) porphyrin sensor was systemically injected into mice with surgically implanted cranial windows or dorsal skinfold chambers. Mice were anesthetized and subsequently imaged using two-photon microscopy (λ_ex=850 nm). Emission from the QD was homogenous throughout the vessels (Fig. 9b), while the porphyrin emission was more variable (Fig. 9c), reflecting differences in oxygenation of different vessel types. In order to obtain quantitative information, lifetime measurements were acquired at various points and converted to oxygen concentrations using Eq. 1 with k_q from in vitro calibrations and τ₀ measured under two-photon excitation. As a result, quantitative oxygen concentrations were measured for arteries (51–72 Torr), veins (20–30 Torr), and capillaries (22–29 Torr) that are consistent with previous observations measured by phosphorescence quenching [139], [140]. This study demonstrates that the micelle construct is a viable sensor for in vivo imaging and sensing using lifetime-based measurements [133].