Key Considerations When Purchasing A Microarray Scanner

With a variety of formats, and numerous instruments to support them, microarrays have become the tool of choice for discerning the expression patterns of normal and diseased tissue, as well as answering a host of other questions.

In most instances, data is generated by hybridization of fluorescent samples of interest with ligands organized on the surface of the array (called “features”). Given the diminutive size of the data points—and unlike their predecessors, such as agarose and acrylamide gels and Southern, Northern, and Western blots—reading and interpreting microarrays requires the enlistment of specialized equipment. Depending on the application and the array itself, these microarray scanners may be part of an integrated system, or purchased ad hoc.

“Cost is going to be boundary condition, always,” notes Jeff McMillan, product manager for Genomics Automation at Agilent Technologies. But with the market evolving so quickly, even the purchaser of a low-end scanner should avoid being short sighted. Ask your vendor: “What’s your upgrade path to get me scanning these [higher density] arrays next year when they come out?”

Second: What’s the application? In some instances, researchers may be limited by the fact that the arrays they plan to use are proprietary and supported by a proprietary platform. In that case, the choice has already been made for them.

Yet for researchers who spot their own arrays, or who wish to take advantage of the myriad of commercially available standard format (1” x 3” glass slide) arrays already on the market (or soon to appear), an “open platform” may be the way to go. And for these, paying attention to certain parameters may help guide a purchasing decision.

Some arrays are designed to allow quantification of sample material, and thus might require an instrument capable of scanning over a very large dynamic range. McMillan cites microRNA detection as an example. In contrast, gene expression analysis—especially for discovery of novel patterns—may require high sensitivity to find faint signals. “All the easy calls have been done,” he notes. “You need to dig down further into the weeds.”

How comfortable the user is with analyzing raw data may also help determine which scanner to purchase. Most scanners come equipped with software meant to streamline the analysis of data it has acquired (often optimized for the scanner manufacturer’s own arrays). Since the most popular applications still involve the hybridization of nucleic acids onto a DNA-featured array, most scanner software is designed with this in mind.

The majority of open-platform scanners can physically read most arrays printed on standard 1” x 3” glass slides, yet since most scanners were developed for DNA arrays, they may not be specialized to interpret the results of a protein or cellular array. In these cases, the user would either have to adapt and optimize to the scanner, or export the data to other software.

There are many proprietary technologies on the market as well. Some represent a nearly turnkey operation to very specialized niches (such as on-the-go specific pathogen detection), some, like Geniome, specialize in user-designed “biochips,” while others offer near-comprehensive sets of more broadly applicable assays (like Affymetrix’s human genome-wide SNP GeneChips)—at a cost, of course.

“We provide a complete system that takes you all the way through the workflow,” explains Shantanu Kaushikkar, Affymetrix’s product manager of systems and software. The platform is intuitive, and customers would rather spend their time designing experiments and collecting samples than tinkering with the technology, he adds.

“It provides a complete system that works without optimization,”¹ echoes Martin Dufva, of the Technical University of Denmark, in a 2005 Biomedical Engineering review. Dufva likens the attitudes of Affymetrix users (in comparison with users of an open platform) to those of Macintosh computer users (compared to users of Windows-based machines), “where the Mac stands for user friendliness but traditionally at the cost of flexibility and economics.” He continues, “there are many situations where the open platform is the only choice for both economical and [sic] practical reasons.”

Users of a closed platform are locked into products offered by the manufacturer for that system—either off-the-shelf or custom-fabricated to an investigator’s specifications.

Researchers looking to analyze the binding of biomolecules to both DNA and protein arrays, for example, must either invest in separate systems, or find one that is compatible with both applications.

Arrays laid out in the proper grid pattern on 1” x 3” glass slides will be scannable by standard readers. If the array was made by Agilent (or was prepared according to specifications published by Agilent) and was read by an Agilent reader, then users can “take advantage of most of the conveniences that our workflow offers,” McMillan notes, including quality control and statistical measures, as well as a summary data report. “They simply can look … and judge whether the array results were good or not.”

“If it’s a homebrew,” he continued, “we won’t know exactly how to treat that array, and a customer may need to fine tune some of the parameters on their own to get the best statistical calculations.” An example of a homebrew might be a protein array. The adoption of protein arrays had originally lagged far behind those of DNA arrays, and scanner manufacturers’ support for them has consequently lagged as well.

Yet they have “gotten to the point where it was worth an investment to write software around doing protein array quantitation,” explains Mary Duseau, Molecular Medicine’s business unit leader for detection and analysis systems at PerkinElmer Life and Analytical Sciences. That software package—sold as part of PerkinElmer’s ProScan Array line of scanners—“gives more of a quantitative number” for the generation of standard curves. In many cases, protein arrays utilize a single fluorophore, and it is the intensity of that fluor that indicates the amount of ligand bound to the array. In contrast, most DNA arrays rely on the ratio of two fluors in competition with each other for binding to the feature, to determine the relative hybridization of the sample to the array.

Cellular and tissue arrays—in which cultures are grown or deposited right on the slide—can be imaged with standard scanners, too, but it is unlikely that any current scanner comes equipped with software specifically written for them. What instrument-makers like PerkinElmer do is output the image file as a standard TIFF, says Duseau. “That then gives customers the ability to take the image and put it into a third party analysis package.”

The application will determine the size of distinct spots the scanner needs to be able to resolve. Winston Patrick Kuo, who is creating a center for emerging technologies at Harvard University Medical School, likes the flexibility of having a wide range of resolutions available. Fifty micrometer scans are sufficient to give a good, quick overview of what’s on a slide, and five micrometers is ideal for even the finest of most current applications.

Determining resolution may not be as easy as looking at a machine’s specs, however. Duseau claims that PerkinElmer delivers a true five-micrometer resolution—with a five-micrometer beam size hitting the feature—while other vendors use a larger beam size and stitch four pixels together. “It’s not true resolution.”

There is currently a drive toward denser microarrays—witness Affymetrix’s new 1.8 million feature Genome-Wide Human SNP Array 6.0, and NimbleGen’s 2.1 million feature custom arrays—pushing the limits of current technology. At some point, writes Dufva, scanners with higher resolution will need to be developed that are capable of differentiating smaller and smaller spots.

However, he explains, “the fluorescent signal will be weaker for each pixel since fluorescence is collected from a smaller area.” Thus when scanning at higher resolutions, scanners tend to dwell longer on any given spot to compensate. Yet at some point noise from the instrument will need to be taken into account.

PMTs deliver a fairly wide dynamic range—they’re pretty linear over four or five orders of magnitude, Duseau says. CCD (charge-couple device) camera technology—found in some instruments in place of PMTs—was once a poor stepchild of the PMT, buy it has come a long way in the past few years, she says, and can now deliver roughly equivalent sensitivity and dynamic range.

Yet it’s often necessary to choose between detecting low signals and quantifying bright ones. Like photographing dark objects against a strong backlight, if the instrument is set for high sensitivity, bright spots will become saturated, while if the gain is set to record information from the bright spots, low levels will be lost.

“In the past it has been a tradeoff—if you wanted a large dynamic range you backed off on sensitivity,” notes McMillan. Agilent employs a computer algorithm to merge the images from a high sensitivity scan with those from a broader-range scan. “Anything that’s saturated on the first scan is going to be picked up on the second scan.”

For a new purchaser, then, it may be important to look not only at the instrument’s dynamic range capabilities, but at any compensatory software that comes with it as well.

A standard gene expression assay puts two different sample aliquots—each labeled with its own fluorophore—onto the array. Thus most commercial scanners come equipped with two lasers—a red and a yellow—to (simultaneously or sequentially) excite these fluors. By employing the appropriate emission filters, the PMT (or CCD) can distinguish the emission from each as well.

Yet many applications, says McMillan, are headed in the direction of single color assays. Agilent’s micro RNA platform, for example, relies on only the fluorophore Cy3.

Some assays may utilize more colors. This may be accomplished by filters on the emission end, as is the case with Affymetrix’s targeted genotyping system.

In other cases—especially when looking at cellular or tissue arrays, notes Duseau—users may want to add a third (or even fourth) laser, such as a green laser capable of exciting FITC or green fluorescent protein (GFP). Labs with such current or future needs should make sure their instrument has (or can be upgraded to) the capacity for multiple lasers.

With arrays becoming denser, and in some cases multiple assays being placed on a single array, however, “we don’t see many people filling carrousels,” notes McMillan.

High-throughput users will probably want to opt for a barcode reader as well, to help keep track of the samples being automatically loaded.

For individual users interested in single applications, a basic box may do. But, points out Kuo, the field is always evolving, and so flexibility is key. It’s probably best—even when purchasing a basic box—to anticipate future needs and (at least) make sure that box can adapt with those needs.