Illumina

Randomly Ordered Fiber Optic Arrays

The randomly ordered BeadArrayTM technology was invented by Professor David Walt and colleagues at Tufts University and further developed at Illumina, Inc., where it has been used as a platform for a wide range of assays (Ferguson et al., 2000; Michael et al., 1998; Walt, 2000). An array of wells is patterned into an optical imaging fiber bundle. This process takes advantage of the intrinsic structure of the optical fibers in the bundle (Figure 1).

Figure 1. Structure and useful properties of an individual optical fiber. Each fiber has a light-conducting inner core that is surrounded by a cladding of different refractive index. The core can be chemically etched at a different rate from its surrounding cladding. By treating the polished end of an optical fiber with acid, an array of microwells is generated. The geometry and dimensions of the array are determined by the physical specifications of the optical fiber and are chosen so that one bead can fit in each well in the array. Once a labeled target nucleic acid is hybridized to beads in the array, a fluorescent signal can be generated by making use of the optical properties of the fiber. An excitation beam is guided to the bead through the fiber bundle, and emitted fluorescence is guided back up the fiber, allow the array to be imaged at the opposite end of the optical fiber bundle.

The optical imaging fiber bundles used by Illumina consist of ~ 50,000 individual fibers fused together into a hexagonally packed matrix, and therefore can hold up to ~ 50,000 beads, each ~ 3 microns in diameter and spaced ~ 5 microns apart. The entire array is ~ 1.4 mm in diameter. The preparation of a bead library and assembly into the array are illustrated in Figure 2. The beads are stably associated with the wells under standard hybridization conditions. A wide variety of bead types can be used, including silica and polystyrene.

Figure 2. Assembly of a randomly ordered fiber optic array. (A) A collection of bead types, each with a distinct oligonucleotide capture probe, is pooled. An etched fiber optic bundle is dipped into the bead pool, allowing individual beads to assemble into the microwells at the bundle’s end. (B) Scanning electron micrograph of an assembled array containing 3 micron diameter silica beads.

A

B

Decoding Randomly Assembled Arrays

Since the assembly of beads into wells is a random process, the location and identity of beads in the array must be determined post-assembly. This process is called decoding. Randomly assembled arrays can be decoded by simple DNA hybridization techniques. DNA hybridization is sufficiently specific to resolve easily at least several thousand simultaneous oligonucleotide hybridizations. The approach is illustrated by the following example. Imagine an array that contains 16 types of bead, each differing by the oligonucleotide sequence attached to them.

In order to decode the array, a set of 16 “decoder” probes complementary to each of the 16 probes in the array is synthesized. Each decoder probe is then labeled with each of 4 different fluorophors to generate 4 versions of each decoder probe. Thus, for decoder 1, there are four versions 1A, 1B, 1C, and 1D, where A, B, C and D represent different labels. Individual decoders are then pooled so that the pool contains one version of each of the 16 decoders. For example:

Pool 1 = (1A, 2A, 3A, 4A, 5B, 6B, 7B, 8B, 9C, 10C, 11C, 12C, 13D, 14D, 15D, 16D)

Hybridization of Pool 1 to the array partially decodes the array, because positions that contain probes hybridizing to sequences 1, 2, 3, and 4 are illuminated by the “A” fluorophor. Positions hybridizing to sequences 5, 6, 7 and 8 are illuminated by “B”, and so on. At this stage, the identities within each group of four sequences are still ambiguous. This is resolved by a second hybridization:

Pool 2 = (1A, 2B, 3C, 4D, 5A, 6B, 7C, 8D, 9A, 10B, 11C, 12D, 13A, 14B, 15C, 16D)

A combination of the data yields a unique identifier for each location in the array (Figure 3).

Figure 3. An array of 16 different bead types is decoded in two sequential rounds of hybridization. The top panel shows the label associated with each decoder oligonucleotide in the two decode pools, Decode hyb 1 and Decode hyb 2. A portion of the image from the first hybridization, Decode hyb 1, is shown in the lower left panel. The lower right panel shows the same region of the array after stripping the signal from the first hybridization and carrying out the second hybridization, Decode hyb 2. To illustrate the decoding principle, two beads are identified. One of the beads is identified as carrying Sequence 4 (blue followed by red) and the other as Sequence 9 (yellow followed by blue).

The decoding process scales well and is very efficient. If two fluors are used, then 2n sequential hybridizations can distinguish 2n sequences. With 4 fluors, 4n sequences can be distinguished, and so on. The important point is that an exponential number of codes is obtained using only a small number of fluors and a small number of decoding steps. The accuracy of decoding is estimated to be 99.99%, which is more than sufficient for our needs (because each bead type is represented by at least five individual beads, the impact on assay results of an error rate of 1 in 10,000 beads decoded is negligible).