Rare earth-doped glass microbarcodes

Abstract

The development of ultraminiaturized identification tags has applications in fields ranging from advanced biotechnology to security. This paper describes micrometer-sized glass barcodes containing a pattern of different fluorescent materials that are easily identified by using a UV lamp and an optical microscope. A model DNA hybridization assay using these “microbarcodes” is described. Rare earth-doped glasses were chosen because of their narrow emission bands, high quantum efficiencies, noninterference with common fluorescent labels, and inertness to most organic and aqueous solvents. These properties and the large number (>1 million) of possible combinations of these microbarcodes make them attractive for use in multiplexed bioassays and general encoding.

Encoded bead bioassays are emerging as an attractive alternative to traditional slide-based microarrays because bead-based bioassays offer multiplexing of both probes and samples (the “analyte”), and they have significantly fewer drawbacks related to mass transport-limited binding of analytes to the immobilized probes. Several approaches have been described for the fabrication of encoded beads: those in which the coding material is randomly distributed in the bead (1, 2) and those in which the coding material is present in a defined pattern on the bead (3). Because different patterns of the same coding materials (e.g., position and thickness of metal stripes on cylindrical particles) result in distinguishable beads (3), a larger number of uniquely encoded beads can be obtained relative to beads with randomly distributed coding materials (e.g., polymer beads infused with mixtures of quantum dots) (2).

Current methods for fabricating encoded beads are limited in terms of either the number of possible codes or the compatibility of the beads with bioassays and fluorescence detection. The most widely used method for making encoded beads, infusing polymer microspheres with mixtures of fluorescent dyes in predefined ratios, is not well suited for the fabrication of large (>10⁵) numbers of uniquely distinguishable beads. Trau and coworkers have used silica microspheres containing fluorescent dyes for encoding polymer beads by using split-pool methods, and have also described the formation of dye-doped concentric silica layers around core silica particles (4). There are only a limited number of spectrally well-resolved dyes that do not also interfere with commonly used biological labels. Moreover, measurements of intensities and their ratios are inherently difficult, which limits the number of levels at which a dye can be incorporated to give distinguishable beads. Mixtures of quantum dots embedded in polymer microspheres offer significant advantages over conventional fluorescent dyes because they are relatively more photostable and have narrow emission linewidths (2). However, quantum dots are made of toxic materials (e.g., CdS, CdSe, CdTe) (5), and difficulties distinguishing between codes based on different amounts of the same quantum dots are similar to those for organic dyes. For both organic dyes and quantum dots, the use of polymer beads as a host matrix raises issues related to background fluorescence and poor compatibility with organic solvents (1). The use of cylindrical beads with metal stripes as codes described by Natan and coworkers (3) overcomes many of the limitations described above. Nevertheless, there are significant limitations. Decoding of beads based on differences in reflectivity can be challenging (6). The use of particles consisting of several different surfaces (e.g., Au, Ag, Pt) and lateral widths (for each stripe) requires careful consideration of the surface chemistry and results in different surface areas for immobilization of the probe and binding of the analyte. Although the use of chemistries based on monolayers of alkanethiolates on gold is well established, gold-based surfaces significantly compromise the sensitivity of detection by fluorescence because of quenching by the gold (7).

Rare earth (RE) ions in a silicate glass matrix are an ideal material for the fabrication of encoded beads. REs are a spectroscopically rich (8) species (Fig. 1), a property that facilitates their use as optical codes in a spectral window distinct from fluorescent dyes used for labeling biological samples. The lanthanide 4f orbitals are buried beneath the 6s, 5p, and 5d orbitals; hence, spectra arising from f–f transitions are narrow and insensitive to their environment, unlike transition metal (3d) spectra (9, 10). Triply ionized RE ions in glassy hosts typically have emission line widths of ≈10–20 nm (FWHM, full width at half maximum) in the visible portion of the spectrum, which is about half that observed for quantum dots (25–40 nm) and much narrower than that observed for organic dyes (30–50 nm) or transition metal ions (≈100 nm) (2, 11, 12). This feature allows more resolvable bands to be packed into the same spectral bandwidth, which enables a larger number of distinct combinations. Because RE emissions involve only atomic transitions, they are extremely resistant to photobleaching. The energy level structure in RE ions also creates the possibility for large shifts between the excitation and emission bands. This shift can be several hundred nanometers containing discrete gaps with zero absorption. By comparison, the HOMO–LUMO (highest occupied molecular orbital–lowest unoccupied molecular orbital) transition in organic dyes typically results in overlapping excitation and emission bands and a Stokes shift of only 10–30 nm between the absorption and emission maxima. Thus, multiple RE ions can be simultaneously excited in the UV and conveniently decoded by observing their emission in the visible, without interfering with organic tags that have excitations in the visible. The silica-based glass matrix itself offers significant advantages relative to polymer-based materials: lower phonon energies that alleviate issues with nonradiative decay, readily available silane chemistries that facilitate the immobilization of biomolecules, compatibility with organic solvents that enable procedures requiring nonaqueous solvents, and low background fluorescence that provides lower limits of detection. In addition, glass preforms can be efficiently drawn down into miles of very thin fiber or ribbons whose structure is an exact miniature of the parent preform, allowing large complex structures to be replicated down to the desired size.

Figure 1.

Energy level diagram of trivalent RE ions with dominant visible emission transitions observed in silicate glasses designated by arrows representing the approximate color of the fluorescence. Horizontal lines extending across the diagram designate the location of common organic label excitation sources labeled by wavelength in nm. The 365-nm line was used to excite the RE ions.

Materials and Methods

Glass Fabrication.

RE-doped alkaline earth aluminosilicate glass compositions for a particular “color” were mixed, melted in a platinum crucible at 1,650°C, cast into 10-mm-thick patties, and annealed for 1 h at 750°C. The optimal doping levels were obtained by maximizing the relative fluorescence intensity as a function of RE (RE₂O₃) concentration. Ce, Dy, and Tm exhibited optimal brightness between 0.25 and 0.5 mol % RE₂O₃, which corresponds to ≈10²⁰ RE ions per cm³ and an average RE ion separation of 2 nm. Because emission intensity from the ⁵D₄ level in Tb³⁺ is free of concentration quenching and continues to increase above 1 mol % RE₂O₃, a level of ≈0.2 mol % was chosen as optimal to match the intensities of the other dopants and to avoid overwhelming them with the strong fluorescence from Tb³⁺.

Fabrication of Barcodes.

Conventional optical fiber draw methods were used to fabricate the encoded fiber ribbons. First, the optimized glasses were melted and cast into 25 × 25 mm square bars and annealed for 1 h at 750°C. These bars were drawn into lengths of square (3.5-mm sides) canes and stacked in a predetermined order to define a barcode pattern. The assembly was then fused in a graphite press in a furnace at 900°C under N₂. The fused preform was drawn at 1,200°C into a ribbon fiber (20 μm thick, 100 μm wide). The ribbon fiber was scribed every 20 μm at a rate of 5 mm/s with 800-nm femtosecond laser pulses (100 mW average power) by using a computer-controlled stage. The scribed ribbon fiber was then sonicated for 60 s in water to break the ribbon along the scribes into individual barcodes.

Hybridization Assay.

The HSPCB gene (human heat shock 90-kDa protein, β1) and the ybac gene (from Bacillus subtilis) were obtained by PCR amplification. The cDNAs were immobilized on γ-aminopropylsilane (GAPS)-coated beads by incubating the beads with the DNA suspended in 50% DMSO/≈5 mM sodium citrate buffer (pH 5.5) for 30 min, at a concentration of 0.25 mg/ml. Excess DNA was removed by washing the beads with water. Hybridization with complex human lung probe (Cy3 labeled) was carried out in buffer (pH 7.2) containing 50% formamide, 2× SSC (1× SSC = 0.15 M sodium chloride/0.015 M sodium citrate), and 0.1% SDS at 42°C for 40 min, with shaking. After hybridization, the beads were sequentially washed with 2× SSC, 0.1% SDS (10 min), 0.1× SSC/0.1% SDS (5 min), and 0.1× SSC (twice, 2 min each), and 0.1× SSC.

Barcode Decoding and Fluorescence Reading.

The barcodes were decoded and imaged by using a spectral imager (SpectraCube, Applied Spectral Imaging, Migdal Haemek, Israel) mounted on a fluorescence microscope (Olympus AX 70) equipped with a mercury lamp (100 W). A dichroic filter was used to select the excitation wavelength, a 420-nm long-pass filter was used for observing the RE fluorescence, and a 570-nm bandpass filter was used to collect the Cy3 fluorescence.

Results and Discussion

Considerations for Choosing REs.

The ideal RE ions for microbarcodes should have nonoverlapping, bright visible luminescence for ease of detection, a common excitation source for simultaneous interrogation of all barcode elements, and no overlap of excitation (and/or emission) with commonly used fluorescent labels (e.g., FITC, Cy3, Cy5, and Texas red) to enable their use in quantitative bioassays. Our choice of candidate REs and excitation and emission wavelengths was based on a consideration of these desired properties.

Energy Levels.

For a RE ion to emit in the visible, there must be an energy gap between an upper excited state and a lower excited state (or the ground state) that is between 14,000 and 25,000 cm⁻¹ (corresponding to visible emission wavelengths ≈700–400 nm). The parasitic nonradiative relaxation rate is exponentially dependent on the integer number of phonons required to bridge the gap between the upper excited state and the next lower level. In practical terms, this gap must be at least 4 phonons wide to prevent quenching by nonradiative multiphonon decay (13). Silicate glasses have a maximum vibrational or phonon energy of ≈1,100 cm⁻¹ (14); therefore, any candidate emitting levels in a silica-based glass must be >4,400 cm⁻¹ above the next lower lying level. For example, upon excitation at 365 nm, Eu³⁺ ions in a silicate glass nonradiatively relax from the ⁵L₈ absorbing level to the metastable ⁵D₀ level, because all of the intermediate levels are separated by fewer than 3 phonons, which is why emission from these levels is absent from silicate-based materials (15). The gap between the ⁵D₀ and next lower lying level, the ⁷F₆, is >10 phonons, so relaxation from the ⁵D₀ is nearly 100% radiative to the ⁷F_J levels, resulting in fluorescent emission centered at 610 nm.

Examination of Fig. 1 suggests that the following seven transitions should emit in the visible and be reasonably efficient: Pr³⁺, ¹D₂ → ³H₅, ³H₄; Sm³⁺, ⁴G_5/2 → ⁶H_J; Eu³⁺, ⁵D₀ → ⁷F_J; Tb³⁺, ⁵D₄ → ⁷F_J; Dy³⁺, ⁴F_9/2 → ⁶F_15/2, ⁶F_13/2; Tm³⁺, ¹D₂ → ³F₄; and Tm³⁺, ¹G₄ → ³F₄, ³H₆. While Nd³⁺, Ho³⁺, and Er³⁺ have visible transitions, they are all quenched in silicate glasses because of multiphonon decay. Even though Ce³⁺ does not have any intra-f transitions in the visible, the lowest lying 5d orbitals are ≈24,000 cm⁻¹ above the ground state; therefore, the Ce³⁺ 5d → ²F_5/2, ²F_7/2 transitions provide an eighth possibility for encoding, although these transitions are much broader than intra-f transitions.

To enable multiplexed excitation of the barcodes, we used a UV source. Mercury lamps are cheap, efficient, and powerful UV sources that emit at 254 and 365 nm. We chose the 365-nm line because Ce³⁺, Sm³⁺, Eu³⁺, Tb³⁺, Dy³⁺, and Tm³⁺ all have absorption bands that overlap with the 365-nm Hg emission line and because damage to biological samples would be minimized relative to illumination at 254 nm.

Determination of Spectral Interference.

As mentioned earlier, a key requirement for a good RE marker is low spectral interference with fluorescent dyes commonly used in bioassays. Relative to fluorescent dyes, lanthanide ions are typically more efficient emitters and also have significantly longer fluorescent lifetimes. If there were overlap, these properties would make deconvolution of the fluorescence emission from the organic label and the RE difficult and would compromise the sensitivity of fluorescence assays. The background contribution of the RE ions was evaluated by fabricating microscope slides from RE-doped glasses and comparing their fluorescence to that of a regular glass slide (Fig. 2). Laser-based microarray scanners optimized to be maximally sensitive to FITC, Cy3, Texas red, and Cy5 were used for detection. The experiment showed that Eu³⁺, Sm³⁺, and Pr³⁺-doped glasses exhibit higher background counts than the bare glass slide, whereas the Dy³⁺, Tm³⁺, Ce³⁺, and Tb³⁺-doped slides have backgrounds that are equal to or lower than that of a conventional glass slide. The high Pr³⁺ background is due to excitation overlap of the 488-nm (FITC) and 594-nm (Texas red) pumps with the ³P₀ and ¹D₂ levels that radiatively relax to the ³H₄ and ³H₅ levels, respectively. The unexpected excitations of Eu³⁺ on the Cy3 (532 nm) and Cy5 (632 nm) excitation channels are due to absorption tails arising from the feebly thermally populated ⁷F₁ and ⁷F₂ states to the ⁵D₁ and ⁵D₀ excited states, respectively. Sm³⁺ is similar to Eu³⁺ in that it is absorption from the thermally excited ⁶H_7/2 (not the ground state) to the ⁴G_5/2 that creates absorption overlap with the 594-nm pump and a high background in the Texas red band. While the ⁴I_9/2 and ⁴F_3/2 energy levels in Sm³⁺ exhibit excitation overlap with the 488- and 532-nm pumps, these absorptions are weak, and its emission bands do not strongly overlap with the FITC and Cy3 emission filters, although they are extremely close and could change dramatically with different filter sets. Pr³⁺, Eu³⁺, and Sm³⁺ were therefore eliminated as potential RE markers for fluorescence-based bioassays, but they are potential candidates for other applications.

Figure 2.

Background fluorescence of RE-doped glasses relative to a bare microscope slide.

Spectral Properties of Candidate REs.

Because of their low background fluorescence, we investigated the use of Dy³⁺, Tm³⁺, Ce³⁺, and Tb³⁺-doped glasses as candidate microbarcodes. Fig. 3 shows fluorescence spectra of glasses doped with Ce³⁺, Tm³⁺, Tb³⁺, and Dy³⁺; the spectra are well resolved and easily distinguished with the naked eye. UV excitation makes the Ce³⁺, Tm³⁺, Tb³⁺, and Dy³⁺-doped glasses glow cyan, blue, green, and pale orange/yellow, respectively. The electronic transitions corresponding to these emissions can be deduced from Fig. 1. Excitation at 365 nm pumps the Ce³⁺ ion into an upper 5d level, which nonradiatively relaxes to the lowest 5d level. Thereafter, the ions radiatively relax to the ²F_5/2 and ²F_7/2 levels, creating broad 5d → 4f peaks at about 390 and 430 nm, respectively. The Tb³⁺ transitions are the most complex. Excitation at 365 nm populates the ⁵G₅ and ⁵L₁₀ levels resonant with the Hg emission. The excited ions quickly cascade down to the metastable ⁵D₃ state. Some light is emitted from the ⁵D₃ to the ⁷F_J manifold as evident in the weak peaks at 380, 415, 435, and 450–470 nm. However, most of the energy is transferred to the ⁵D₄ state by pairwise cross-relaxation, which leads to emissions at 490, 540, 585, and 620 nm, caused by transitions to the ⁷F₆, ⁷F₅, ⁷F₄, and ⁷F₃ levels_, respectively. Tm³⁺ emits directly from the pumped ¹D₂ excited state at 455 nm (¹D₂ → ³F₄) with a small shoulder at 475 nm (¹G₄ → ³H₆). Dy³⁺ is excited to the ⁶P_3/2 level, which nonradiatively cascades down to the ⁴F_9/2 level, resulting in ⁴F_9/2 → ⁶H_15/2 and ⁴F_9/2 → ⁶H_13/2 transitions at 480 and 575 nm, respectively. All of the ions have emissions near 480 nm; therefore, this region of the spectrum is impractical for readout as barcodes. Diagnostic peaks at 400, 455, 540, and 575 nm for Ce³⁺, Tm³⁺, Tb³⁺, and Dy³⁺, respectively, offer the best spectral separation and quantum yield, and are therefore optimal wavelengths for readout.

Figure 3.

False-color image of two 100 × 20 μm barcodes (Inset) and corresponding fluorescence spectrum barcode elements. The same color scheme is used for the spectra and the image [e.g., the yellow band in the barcode corresponds to the yellow (combination Tm+Dy) line spectrum].

Increasing Microbarcode Combinations.

The maximum number of possible bar code combinations in a ribbon-shaped fiber with N members and C colors is given by C^N. The actual number of nondegenerate barcodes is ≈C^N/2 (when N is even, number of codes = [C^N + C^N/2]/2; when N is odd, number of codes = [C^N + C^(N+1)/2]/2). For example, a six-member ribbon with a palette of four colors will yield a maximum of 4⁶ or 4,096 combinations and 2,080 nondegenerate combinations. Although it is possible to increase the number of barcodes by varying the scribe-length of the ribbon, the number of bands in a ribbon, and the concentration of the candidate lanthanide, the most convenient option for decoding involves a simple yes/no determination of color and sequence within a ribbon. With this objective in mind, we created five “combination colors”: binary combinations of Ce³⁺–Tb³⁺, Ce³⁺–Dy³⁺, Tm³⁺–Tb³⁺, Tm³⁺–Dy³⁺, and Tb³⁺–Dy³⁺. These doped glasses were observed to have clearly resolvable fluorescence and negligible quenching. The emission spectrum of the Tm³⁺–Dy³⁺ combination is illustrated in Fig. 3; the spectrum is a sum of the spectra of the individual ions. The expanded nine-color palette brings the number of barcode combinations to 9⁶ or ≈530,000 (or ≈265,000 nondegenerate). The Ce³⁺–Tm³⁺ combination was not used because the Ce³⁺ emission tail overlaps with the Tm³⁺ emission at 454 nm, which could complicate the readout. Triplet combinations of Ce³⁺–Tb³⁺–Dy³⁺, and Tm³⁺–Tb³⁺–Dy³⁺ are also possible. Given these encoding options, fabricating >10⁶ uniquely distinguishable barcodes by using RE-doped glass fibers is theoretically achievable.

Microbarcode Assay.

The use of the barcodes in a fluorescence assay was demonstrated by using a DNA hybridization experiment. After silanization of the barcodes with γ-aminopropylsilane (GAPS), one set of barcodes (Tb-DyTm-Tb-Dy-Tb-DyTm; set A) was treated with DNA corresponding to the human HSPCB gene; the second set of barcodes (TbTm-Tm-TbTm-Tb-Ce-TbTm; set B) was treated with DNA corresponding to the Bacillus subtilis gene ybac. The immobilization of DNA on GAPS (and other amine-presenting surfaces such as polylysine) is primarily attributed to polyvalent electrostatic interactions between the negatively charged DNA backbone and positively charged amino groups on the surface (16). We chose to fabricate barcode B because it represents a particularly difficult scenario for decoding. TbTm (shown as purple in Fig. 4C) is present in three positions of the six-member bead, sandwiched between Tb (green) and Tm (blue), and next to Ce (cyan), which has a similar light-blue emission. The barcodes were pooled and hybridized for 40 min with a sample consisting of Cy3-labeled cDNA obtained by reverse transcription from human lung poly(A)⁺ RNA (1 pmol of Cy3 per μl). This sample, which represents the expression of all of the genes found in the human lung, was used because it simulates a real hybridization experiment and provides a higher stringency test for nonspecific binding relative to hybridization with only the PCR-amplified DNA corresponding to the HSPCB gene. After the assay, the barcodes were dispensed onto a microscope slide for observation. Fig. 4a shows images of the barcodes illuminated with white light; the barcodes are indistinguishable. When illuminated at 530 nm and monitored at 570 nm (conditions corresponding to the detection of Cy3), only some of the beads fluoresced (Fig. 4b). Fig. 4c shows the same beads illuminated with 365-nm light and viewed through a 420-nm long-pass filter. Two sets of beads are clearly revealed. Matching Fig. 4 b and c shows that there is an exact correspondence between beads comprising set A and beads that exhibit positive Cy3 signals (from Fig. 4b). These data demonstrate the expected specific hybridization of the labeled probe to beads presenting the HSPCB gene and the lack of nonspecific binding under the conditions of the assay.

Figure 4.

Fluorescence false-color images of barcode particles A and B used in a DNA hybridization assay using Cy3-labeled DNA. (a) “White light” image. (b) Cy3 channel image. (c) RE images obtained by using a 420-nm long-pass filter.

We have also carried out experiments in which biotin (derivatized with a terminal NH₂ group) was covalently attached to epoxide-functionalized barcodes and demonstrated specific binding of Cy3-labeled anti-biotin to these beads (data not shown). As mentioned earlier, the entire repertoire of silane chemistries for immobilization of biological molecules should be feasible on these barcodes.

Conclusion

Although encoded bead assays offer the possibility of significantly improved assay performance relative to conventional microarrays, the development of this technology has been hampered by the limited attributes of materials used to fabricate the beads. RE-doped glass beads have properties that overcome most of these identified limitations and are therefore ideally suited for use in encoded-bead bioassays. While the number of uniquely distinguishable beads (>10⁶) is more than sufficient for highly multiplexed bioassays, removing the constraints imposed by spectral compatibility with fluorescent biological tags expands the palette of combinations even further. The fabrication of RE-doped barcodes is not limited to fiber draw techniques, and alternative approaches are possible.

Abbreviation

RE

rare earth