¹TeleChem International, Inc., Sunnyvale, CA, USA

Correspondence: Dr M Schena, Telechem International, Inc., 524 East Weddell Drive, Sunnyvale, CA 94089, USA. Tel: +1 408 744 1331; Fax: +1 408 744 1711; E-mail: mark@arrayit.com

Microarrays can be used to identify toxin effects on gene expression, delineate detoxification pathways in vivo, and potentially as sensors to monitor human exposure.

'Why is a raven like a writing desk?' asked the memorable Mad Hatter in Lewis Carroll's Alice in Wonderland,¹ a riddle that has puzzled and amused readers for more than a century. This was not the only example of the Mad Hatter's strange behavior at Alice's infamous tea party. The Mad Hatter criticized Alice's long hair, an unusually forward and perplexing criticism of a Victorian girl, and sported a wristwatch that reported the day of the month but not the time of the day. So bizarre was the episode that an exasperated Alice regarded it as the 'stupidest tea party I ever was at',¹ perhaps not surprising in view of the well-documented cases of mercury poisoning that afflicted felt hatmakers in the 19^th century, producing a myriad of symptoms including severe neurological dysfunction.² Although Carroll's Mad Hatter was a fictional character, the 'mad hatter's' syndrome was anything but that—so prevalent in fact that it resulted in a complete banning of mercury nitrate from the hat-making industry.

Exposure to environmental toxins continues to be a challenge in industrialized countries, with more than 70 000 chemical compounds registered for use in the United States alone, and relatively little detailed information available concerning the toxicity of these compounds in humans.³ Organic solvents, pesticides, herbicides, and heavy metals are known or suspected contributors to cancer, birth defects, immunological disorders, and a host of neurological diseases when exposed to humans at elevated levels⁴. Carbon tetrachloride (CCl₄) is a clear, colorless, toxic liquid that enters the atmosphere in a highly efficient manner because of its high vapor pressure. The atmospheric lifetime of this stable compound is approximately 50 years.⁵ CCl₄ is used commercially in dry cleaning, degreasing, and in the production of refrigerants and other chlorinated hydro-carbons, and acute human exposure via inhalation produces serious symptoms including nausea, vomiting, dizziness and headache. The liver, kidney and brain are three of the main target organs for this toxic substance, and chronic exposure in laboratory rats and mice produces embryonic lethality, hepatomas, and hepatocellular carcinoma.⁵ The stability and toxicity of CCl₄ suggest the need for new tools to study and monitor human exposure to CCl₄ and other environmental toxins.

It is in this context that our discussion turns to a recent paper by Young et al⁶ 'Analysis of gene expression in carbon tetrachloride-treated rat livers using a novel bioarray technology' published in The Pharmacogenomics Journal. In a paper destined to be a classic, Young et al⁶ combine high-quality microarrays, powerful computational tools, and a well-established rat model of liver toxicity to identify genes that are induced and repressed by CCl₄. The expression signature identified in this study provides fresh biochemical clues to the mechanism of CCl₄ toxicity in rat, and such data are likely to be informative in humans. The work also suggests a general approach for studying environmental toxins, providing a straight path to developing analytical microarrays to assess patient exposure.

The paper builds on the earliest microarray publications, which demonstrated the usefulness of chips for expression profiling⁷ and toxin-induced gene discovery,⁸ and in these respects the Young et al⁶ paper does not break new ground. However, in terms of all the technical details of the approach and in identifying a specific subset of CCl₄-induced genes, Young's publication is new and impressive.

The paper begins with a description of microarray platforms, describing some of the pros and cons to the two main types of nucleic acid microarrays (cDNA and oligonucleotide). Young et al⁶ correctly point out that for organisms in which there is abundant sequence information, oligonucleotide microarrays⁹ offer some advantages over cDNA microarrays,⁷ including the fact that single-stranded array elements do not self-hybridize and therefore can produce stronger signals than double-stranded cDNA elements. Oligos can be designed with pinpoint accuracy against unique gene regions, essentially eliminating crosshybridization between homologous genes and providing greater assay accuracy. There are also some important differences between oligonucleotide arrays synthesized in situ¹⁰ and microarrays made by deposition using pins or ink jets. A major distinction between traditional arrays¹⁰ and microarrays⁷ is the use of fluorescence detection instead of radioisotopes, which allows a high degree of assay miniaturization in the latter case. A major difference between in situ synthesis and deposition is the ability in the latter to assess oligonucleotide quality and purity prior to microarray manufacture.

The microarrays used in this study contained 1137 unique targets representing 1040 rat genes and 97 controls deposited by ink jetting. Although the gene subset represented only about 3% of genes in rat, their functions were broad, spanning cellular metabolism, apoptosis, DNA repair, cell cycle regu-lation, detoxification, and other functions related to CCl₄ toxicity. The computer-designed 30-mers were made 'off line' by phosphoramidite synthesis, and examined for concentration, purity, and sequence identity prior to ink jetting to ensure high quality. The adherence to quality control and quality assurance contribute to the outstanding results, and refute the mis-guided notion that statistical tools can reconcile poor-quality chips (never!).

Sample preparation is always a challenge in gene expression studies because stimuli unrelated to the intended treatment can alter expression patterns and produce artefacts. Young et al⁶ take appropriate care in setting up the animal model, and in harvesting and preparing the samples from treated and control rats. The quality of the data substantiates the validity of the results, and 'quality in–quality out' is a lesson well taken and one that we need to drum loudly as microarray data continue to be produced in dizzying quantities.

Microarray innovation is happening at a feverish pace, and it is important to continue to spawn innovation by distinguishing methodological principles from methods, the former defining 'what' and 'why', and the latter generally outlining 'how'. Although Young et al⁶ used the same methodological principles reported in the original microarray publication,⁷ nearly every technical aspect of the work was different from the 95' Science paper. Schena et al⁷ used double-stranded cDNA targets, direct labeling, two-color fluorescence, tweezer-based contact printing, polylysine surface chemistry, confocal scanning, and spreadsheets for data analysis. By contrast, Young et al⁶ employed single-stranded oligonucleotide targets, indirect probe labeling, single-color fluorescence, noncontact printing, acrylamide-based surface chemistry, nonconfocal scanning, and sophisticated scatter plots, clustering, and self-organizing maps (SOMs) for data analysis. Both papers identified differentially expressed genes by microarray (what) to better understand a biological process (why), but used very different experimental tools (how) to achieve the desired outcome.

Organisms express genes when and where they are needed, and consequently changes in gene expression correlate tightly with function.¹¹ Of the 1040 rat genes examined, 19 were induced and 21 were repressed by 4-day CCl₄ treatment, and several of the 40 genes have been identified in previous nonmicroarray studies. The transcription factor ATF3 has been shown in previous work to be activated following liver damage,¹² consistent with the fact that CCl₄ is detoxified in this organ. A number of other genes including those of the cytochrome P450 family exhibited expression changes, not surprising in the light of their role in xenobiotic metabolism. The study also identified a number of novel genes including a protein tyrosine phosphatase, which is thought to play a role in cell growth and differentiation,¹³ and is presumably induced in response to CCl₄- induced liver injury. The data correlate nicely with independent approaches, and extend those observations by affording new information. In this respect, the work is classic microarray science in that it validates the use of chips, and showcases the value of nonhypothesis-driven (discovery-based) research.

Data mining and modeling are usually more difficult than generating raw results, and the authors chose their computational tools wisely and applied scatter plots, clustering, and SOMs to their microarray data. Scatter plot information on same–same and control vs experimental samples demonstrates the precision of the data and provides a global view of altered expression as a function of the duration of CCl₄ treatment. What is striking about the scatter plot data is that the 4-day treatment produced a more profound effect than the 7- or 14-day treatments, emphasizing the importance of 'looking early' in signaling pathways.¹⁴ A key aspect of the work is that the dendrograms (clusters) and the SOMs corroborated the scatter plot data, which underscores the value of using multiple tools for data analysis. It might be informative to look more globally at CCl₄-induced changes using a more extensive rat microarray¹⁵ or a whole-genome microarray containing 30 000 targets.

Young et al⁶ provide a clear path to the development of diagnostic microarrays containing CCl₄-responsive genes for human exposure monitoring. Microarray analysis of blood cell gene expression might be useful in identifying blood-based markers for human studies. 'Toxin chips' might find broad medical use by allowing physicians to monitor symptoms of acute exposure with quantitative microarray assays that would presumably reflect duration and exposure levels.