gerona      J Gerontol A Biol Sci Med Scigerona      The Journals of Gerontology Series A: Biological Sciences and Medical Sciences      J Gerontol A Biol Sci Med Sci      1079-5006      1758-535X              Oxford University Press                    020035BS10.1093/gerona/57.11.B400                        Journal of Gerontology: Biological Sciences                            Designer Microarrays        From Soup To Nuts                                          Wang            Eugenia                                a                                                Lacelle            Chantale                                a                                b                                                Xu            Suying                                a                                                Zhao            Xuechun                                a                                                Hou            Michael                                a                          aDepartment of Biochemistry and Molecular Biology, University of Louisville School of Medicine, Kentucky        bDepartment of Anatomy and Cell Biology, McGill University, Montréal, Canada                    Eugenia Wang, Department of Biochemistry and Molecular Biology, University of Louisville School of Medicine, 570 South Preston Street, Baxter Building Room 304, Louisville, KY 40292 E-mail: eugenia.wang@louisville.edu.                    1        11        2002            57      11      B400      B405                        7          8          2002                          28          5          2002                            The Gerontological Society of America        2002                    The recognition that multigene mechanisms control the pathways determining the aging process renders gene screening a necessary skill for biogerontologists. In the past few years, this task has become much more accessible, with the advent of DNA chip technology. Most commercially available microarrays are designed with prefixed templates of genes of general interest, allowing investigators little freedom of choice in attempting to focus gene screening on a particular thematic pathway of interest. This report describes our “designer microarray” approach as a next generation of DNA chips, allowing individual investigators to engage in gene screening with a user friendly, do-it-yourself approach, from designing the probe templates to data mining. The end result is the ability to use microarrays as a platform for versatile gene discovery.                              hwp-legacy-fpage          B400                          hwp-legacy-dochead          RESEARCH ARTICLE                            Decision Editor: James R. Smith, PhD            OVER the past century, as a result of increasing medical knowledge, better sanitation, and improved nutrition, developed countries are experiencing an unprecedented increase in average human life span, along with a higher incidence of multifactorial diseases such as cardiovascular diseases, neurodegenerative disorders, type 2 diabetes, and cancers (1). These age-dependent diseases, plaguing people as young as their mid-50s, are products of the combined influences of genetics and environment (2). Nature and nurture together provide predispositions to cancer, cardiovascular disease, diabetes, and neurodegenerative disorders, presenting a complex picture for the development of these perils in the fast-growing middle- and old-age subpopulations of our society.    Although recent advances in medical research have enabled us to diagnose several age-associated diseases, alleviate pain associated with them, and retard the onset of their acute stages, we remain largely incapable of identifying at an early stage those individuals bearing genetic predispositions to these diseases, and thus of administering preventive medicine or treatment. Because the human genome contains some 30,000 genes (3), and modern industrialized society yields increasing environmental complexity, it is an ever-greater challenge to perceive how the integration of our genes and surrounding environment creates disease-predisposed states. For example, why do certain individuals suffer lung cancer at an early age, after a few years of cigarette smoking, whereas some centenarians tolerate lifelong smoking without dying of the same disease? Such questions led to the idea of the need to identify “genetic signatures.” Once genetic signatures are secured, one may develop means to prevent and/or treat diseases in an individual manner, creating individualized medicine for prognostic, diagnostic, and therapeutic purposes.    In general, large-sample gene chips, bearing perhaps 10,000 genes, are applicable only to early-stage screening and may yield voluminous lists of potential positive results; here we describe a next-generation, medium-density microarray approach embodying considerable quality control in both chip design and analysis, which results in fewer hits of enhanced accuracy and pertinence.          Genetic Signatures and Microarray Technology      During the 1980s and early 1990s, biologists were busy dissecting the functions of single genes, by using a reductionist approach, which, although thorough in its exact methodological analysis of genetic impact, was restricted in explaining how each particular single gene functions in the context of many homologous genes or partners to accomplish a biological task. In an attempt to shed light on these biological tasks, “biochips” were introduced in the mid-1990s (4) as a new tool for molecular medicine; they now constitute a rapidly developing field of biomedical research, which combines molecular medicine, nanotechnology, computer science, and engineering. Biochip technology was developed for high-throughput gene screening, capable of simultaneously identifying changes in the expression of hundreds or thousands of genes, and thus genetic signatures defining particular physiological states. Consequently, as a result of its power, the young field of DNA microarray technology has rapidly gathered speed and popularity within the biomedical research community. However, as universities across North America have started to establish microarray core facilities, they are realizing that the next generation of microarrays must be more versatile, user friendly, and inexpensive to ensure that these facilities will meet the divergent needs of their researchers, and ultimately provide practical answers to fundamental biological problems. Thus, 4 years ago, parallel to the development of commercial microarrays, we began devoting significant time and effort to developing pathway-specific “designer biochips”; here we provide an account of our quest for designer arrays, and we discuss some of the challenges that lie ahead for those seeking the perfect array.              The Theory Behind Microarrays      Unlike Northern blots, in microarrays the probes, not the targets, are immobilized to a physical platform. Microarrays consist of a platform to which are independently attached numerous nucleic acid sequences known as probes, to screen targets of prelabeled nucleic acids obtained from donors (human or animal). The quantification of probe–target complexes formed during hybridization permits measuring gene distribution and intensity, as complementary probe-labeled target sequences bind together.      Although the principle behind microarrays is simple, creating and implementing microarray technology is difficult, as several parameters (discussed in the paragraphs that follow) can drastically affect the validity of the results obtained from microarrays. Furthermore, because the number of probes included on each microarray platform is great, the magnitude of results obtained from microarrays is huge, and thus requires powerful computerized image processing and statistical software to classify and analyze the data; without these, little significant gain can be obtained from using microarrays. Thus, microarray core facilities must integrate expertise in biology, computer science, engineering, and statistics. It is with this in mind that we started our quest for designer biochips.              Platform and Printing Robots      The first consideration when developing microarray technology is the type of platform, commonly either membrane or glass slide (5)(6)(7)(8)(9). Membrane-based microarrays use nylon or nitrocellulose membranes, and they are generally used when radioactive or colorimetric tags are used to label the targets, whereas glass slides are more suitable for microarrays using fluorescent tagging; we chose to use a membrane-based platform for our microarrays. Unlike glass slides, which must be chemically treated to permit the attachment of probes and decrease background fluorescence before use as a platform for microarrays (5)(8)(9)(10)(11)(12)(13), most membranes (whether bearing positive, negative, or neutral charge) require no such pretreatment. Because nucleic acids are negatively charged, the use of positive or neutral membranes greatly improves the signal; however, neutral membranes are generally more suitable, as positive-charged membranes yield higher background readings as well. Negatively charged membranes, although providing very little background, are usually not suitable because they yield poor signal. Thus we chose to use neutral-charged membranes for our arrays.      Once a microarray platform has been chosen, probes must be attached to the platform. Two obvious methods exist: synthesis of probes directly on the platform (14)(15)(16)(17)(18), and probe-spotting by use of a contact or noncontact printing robot (4)(19)(20)(21)(22). Although leading biochip companies often synthesize oligos directly on their microarrays by using techniques such as photolithography, this method is not easily mastered, nor accessible to most laboratories. In contrast, probe-spotting can be accomplished using any of several commercially available printing robots (22). Because we use membranes attached to glass slides as our platform instead of glass slides directly, we encountered several problems, including skipped dots and uneven printing, when we first attempted to print arrays. We had to substantially modify the printing heads of the first robot we purchased, and we had to build an enclosure over it that permitted maintenance of constant humidity, to ensure even printing of the probes.              Probes      Almost any type of nucleic acid can be printed with a printing robot; nucleic acids obtained from complementary DNA (cDNA) libraries, as well as polymerase chain reaction (PCR) products generated by reverse transcription PCR (RT-PCR), are commonly used as probes for microarrays. The use of cDNA libraries is of considerable value for screening previously unknown genes or a great many genes with no predefined preferences for certain gene families. However, mistakes arising from mislabeling of clones or contamination can cause problems (23). We spot PCR products on our chips to generate thematic arrays of particular known genes of interest. Commercially available biochips are often restricted to specific sets of genes contained on each biochip, posing user-unfriendly conditions. In the gene discovery task, users are generally conditioned to screen according to the preset configuration of genes, without the possibility of generating thematic or pathway-specific microarrays covering genes known to belong to a specific family, as demanded by a particular research program. The use of designer biochips can circumvent this problem. The following sections describe the strategy we used to fabricate thematic arrays.              Thematic Microarray Design              Gene Selection and Primer Design        To generate thematic microarrays bearing genes from a particular family or the same pathway, we use several public databases, including Unigene and GenBank (http://www.ncbi.nlm.nih.gov/GenBank, http://www.ncbi.nlm.nih.gov/Unigene), and conduct an extensive literature search (http://www.ncbi.nlm.nih.gov/PubMed) to obtain a repertoire of genes. Once a list of potential genes is constructed, we use GenBank and Unigene to obtain the sequences of all candidate genes. Primers for each gene sequence are designed using Primer3 software (http://www.genome.wi.mit.edu/genome_software/other/primer3.html) to generate a PCR product, or “amplicon,” with a length between 300 and 700 bp and a melting constant that ranges from 75°C to 89°C. For each gene we choose a pair of sense and antisense primers with an annealing temperature of approximately 60°C and an average length of 23 bp, for amplicon production using 96-well PCR plates. These clustered amplicon sizes and melting constants support standardized hybridization conditions for all probes across the microarray platform, including stringent washing, thus decreasing nonspecific signals while maximizing a specific signal. The length of the PCR product, as documented by Stillman and Tonkinson (6), is particularly important in maximizing a specific signal.        Primer design is perhaps the most time-consuming step in our microarray production, because once a primer pair is selected, an analysis must be performed with Blast (proprietary software available on the National Center for Biotechnology Information website) to ensure that each primer pair amplifies only the gene of interest. This is crucial, because results obtained from the microarray are dependent on the specificity of the amplicons. However, in some instances, the specificity of the primer may not guarantee the specificity of the generated amplicon, when a conserved or shared domain lurks somewhere within the sequence. It is therefore highly recommended that the entire amplicon sequences themselves be Blasted to identify homologous regions, which can cause nonspecific binding. With probes obtained from cDNA libraries this may become a pitfall, especially when the spotted nucleic acid sequence is unknown; often highly homogenous sequences may result in nonspecific binding between genes of high homology. Stringent hybridization conditions and washing can generally eliminate this nonspecific binding, if the homologous region is not too long.                    Controls        As in any biological experiment, and most importantly for microarrays, controls must be carefully selected. It is important to spot on all microarrays negative and positive controls as well as “housekeeping genes,” used in more traditional experiments such as quantitative RT-PCR, which show little or no physiological change in expression among the subjects or conditions being studied. The inclusion of housekeeping genes is useful for data normalization; for our designer microarrays targeted to mouse models, we selected six mouse genes (glyceraldehyde phosphate dehydrogenase, ribosomal S6, beta-actin, hypoxanthine-guanine phosphoribosyltransferase, phospholipase A2, and ubiquitin) commonly used in the literature as controls. In general, the validity of these controls must be determined a priori by using independent tests, such as Northern blotting assays or quantitative RT-PCR (24). For instance, EF-1α would be a poor choice for a housekeeping gene if the target nucleic acids were obtained from skeletal muscle, as it is not expressed in adult muscle cells; it would, however, be a good control when cDNA from liver is used as a target (25). The use of housekeeping genes permits the measuring of changes in gene expression against a gene whose expression does not vary significantly; in some cases this can be of great value. Negative controls should include buffer, bacterial, and viral DNA, as well as amplicons from genes known not to be expressed in target tissues. Negative controls are used to assess the level of background noise arising from nonspecific nucleic acid binding during probe–target hybridization. Positive controls such as total cDNA or genomic DNA permit the detection of suboptimal conditions of hybridization and staining, which may obscure appropriate signal intensity.                    Quality Control for Amplicon Production        In order to avoid producing the wrong amplicon for printing as a result of contaminated PCR reactions, the use of dedicated equipment and reagents in the PCR setup and reaction areas is recommended. For each PCR reaction with a unique amplification primer pair, a negative control should be used to ensure the absence of reagent contamination, often caused by the presence of exogenous nucleic acids. This control reaction is identical to the regular reaction, except that no template is present. Agarose gel electrophoresis is used to verify the amplicons and ensure that they are of expected size. In instances where multiple bands result from the PCR, products can be resolved on an agarose gel, and the fragment of expected size excised; these amplicons can then be sequenced to confirm their identity. It is our experience that, when a primer pair is well chosen, multiple bands seldom result from the PCR reaction.                    Printing the Arrays        Once amplicons have been produced for all genes of interest as well as housekeeping genes, arrays can be printed. To avoid positional bias, arrays should be printed in a scattered fashion, with several repeats of the same amplicon located in different regions of the chip. It is important to avoid positional bias, as uneven distribution of charges on the membrane can result in regions of increased background. A typical microarray manufactured in this fashion carries arrayed triplicates or quadruplets of amplicons from selected genes, positioned on the array among many control spots. The rationale for triplicate printing is to provide three data points for statistical analysis of significance; ideally, the three could be expanded to four or five repeats, to yield more data points for statistical analysis. This approach of scattered array printing requires considerable careful analytical software design, to enable tracking of amplicon repeats across the platform; however, it approaches an ultimate solution to resolving positional bias.        Although the spots of microarrays printed onto membranes affixed to glass slides are usually colorless, it is possible to monitor quality to detect gross errors in printing, such as missing, smeared, or non-uniform spots; immediately before a batch of microarrays are printed, a colored dye can be used to print a test array of dots onto a membrane. Microscopic visual inspection of the spots enables any necessary adjustments to be made to the robot before sample printing begins. While large batches of chips are being printed, quality can be monitored by inserting poly-L-lysine-coated glass slides among the membrane platforms. Unlike membranes, the clear surface of glass slides permits the researcher to see printed spots by breathing on the slide and viewing it through a transmitted light source.        Once the probes are printed on the membranes, they are cross-linked to the microarray to permit better attachment of the nucleic acids to the substratum; probes are denatured by boiling the membranes before hybridization.                    Target Labeling      Donor nucleic acids can be labeled by adding a labeled base to the RT reaction used to generate target cDNA. Whereas most commercial arrays use fluorescence-conjugated or radiolabeled bases, we developed our microarrays based on a nonradioactive colorimetric method. The bulkiness of some of the fluorescent tags, fluorescent quenching over time, and the need of specialized scanners to read fluorescent signals dissuaded us from using the fluorescent approach (26). Similarly, we wanted to avoid using radioactive labels for safety reasons, because they often give saturated signals, and to save the time required for exposure of the labeled arrays to x-ray film.      Using commercially available digoxigenin (DIG)-labeled dUTP (Roche, Palo Alto, CA) and alkaline phosphatase (AP)-conjugated anti-DIG antibody (27), we have developed a new application for DIG in microarrays (28). In our method, the cDNA to each donor RNA is synthesized with a DIG-labeled base. Following hybridization of the DIG-labeled target with the probes, positive reactions are revealed by incubating with anti-DIG antibody conjugated to AP, and subsequent staining with Nitro-blue-tetrazolium/5-Bromo-4-chloro-3-indolyl phosphate (NBT/BCIP, Roche) to detect AP (29). Taking advantage of the fact that two complementary nucleotide strands can hybridize with each other, we generate microarray results by quantifying the signal obtained from the labeled targets bound to the immobilized probes. Thus the positive loci are visible as bluish spots, easily identified as round deposits for each positive locus. The final detection is revealed as a matrix of many round dots of varying intensity of staining.              Microarray Inventory      Extensive records should be maintained on the microarray printing process, including logistical parameters of the physical status of the microarrayer (e.g., relative humidity of the chamber during printing, or how often the printing pins are cleaned during the run), the specifics of printing, including the printing sequence, the identity of each amplicon-containing spot, and all other applicable data or comments. To ensure that identical records are made for each printing, we record all data for our microarrays onto a standardized form that we have developed over the past several years. We are also developing a two-dimensional bar-coding system to keep detailed track of our inventory of microarrays.              Image Acquisition and Data Processing      The work just described supports the process of data generation; the following sections describe the mining of the resulting data.              Array Normalization and Background Subtraction        As our arrays are based on a colorimetric detection method, a high-resolution scanner is used to scan them into digital images. Before a normal office scanner is used, it is important to ensure that it digitizes accurately without transforming the image (26). If the image is transformed by the scanner, mathematical correction transformation should be applied to the result. Following acquisition, the digitized images can be normalized and subtracted as desired. We have developed a software program, GeneAnalyzer, which accomplishes background subtraction, array normalization, and quantification. When colorimetric microarrays are analyzed, several types of background must be considered; for instance, regional background subtraction is useful when the array shows differential intra-array background expression, whereas global background subtraction is suitable when the background value is constant within arrays but variable between arrays, as a result of experimental conditions. For interarray comparison to be supported, arrays may be normalized by several methods, including reference to housekeeping gene levels and median chip values. However, investigators should think carefully about the effects of performing such background standardization or normalization before they start analyzing their results; they should especially consider the effect of background subtraction on diminishing the signal of low-expression genes.                    Software for Microarray Data Acquisition        In general, image acquisition and data analysis include the following processes: (a) image grabbing and digitizing; (b) image processing; and (c) data mining, including a qualitative and quantitative analysis of all digitized images, and a statistical analysis of data. We developed our software with a user-friendly interface and a limited number of preset functions, to enable researchers to analyze their own data. The main features of our program are as follows.        First, we provide users a personal identification number, which allows optimal security of their data and access to the interactive functions of our web server facility. Second, users can upload their electronic images from remote sites over the Web. Third, our system processes the users' initial data to enhance the image profile, through standard computer software such as MatLab. Fourth, our system supports the users' data archiving and database organization for the next stage of data analysis.                    Statistical Analysis and Data Mining      Although it is necessary to use statistical methodologies to analyze voluminous microarray data, it is equally important to generate adequate microarray-derived data to support statistical testing. Thus it is recommended for most studies to use at least three mice (or individuals) chosen at random from each test group (e.g., young vs old mice). RNA extracted from each donor tissue source (e.g., lung) is then subjected to three separate RT reactions, and each of these nine independently obtained cDNAs is used as targets for three different microarrays. Because on our microarray each amplicon is spotted in triplicate, we have a total of 81 (34) spots, or data points, for each gene being analyzed (RNA from three mice multiplied by three RT reactions multiplied by three microarrays multiplied by three spots on each slide). The use of at least triplicates for each step of microarray fabrication enables us to obtain statistically meaningful data. Without this replication, simple variations in the efficiency of the RT reaction, interanimal variation, or misprinting of a spot could all result in falsely perceived changes in gene expression.      Because the statistical analysis of microarrays presents a challenge to many biologists, it is recommended that a statistician be consulted as necessary. Statistical consultants can be extremely helpful, not only at the final stage of data analysis, but also at the initial experimental design step; for example, they may provide answers to fundamental questions, such as how many animals are needed to establish a statistically significant data analysis, or whether or not one may pool RNA samples.      Once microarray data are processed through statistical analysis, data entry points deemed of true “significant” value, that is, gene expression changes as effects of an experimental physiological change, should be subjected to the next level of data analysis, now popularly termed the data mining process (30). Many established methods have been popularized among microarray users, including principal component analysis (31)(32)(33), hierarchical clustering (34)(35)(36)(37), multidimensional scaling (38)(39)(40), and self-organizing maps (41). In general, the selection of any of these methods is dependent on the individual investigator's preference and expertise. For example, GeneSpring software, sold by Silicon Genetics, Inc. (San Carlos, CA), and Significance Analysis for Microarrays from Stanford University (42), are preferred for many gene screening data mining tasks because they can analyze data generated by several different microarray platforms. These data mining software packages enable researchers to display their data in forms suitable for publication, easily conveying the essence of the results.      Following data mining, microarray data that seem to be significant should be validated by using one of two popular methods: Northern blotting or quantitative RT-PCR. In general, it is advisable that microarray data be validated by the selection of four or five randomly designated genes from each of three categories: those showing high, intermediate, and low levels of significant difference. Because we use amplicons to generate our probes, we can easily validate our results, using the same primers used to generate our amplicons by quantitative or semi-quantitative RT-PCR.      During data analysis, special consideration must be given to low-expression genes, which generally exhibit the greatest variance in expression levels. On any given microarray, these genes show very weak intensities, and in some cases they are barely visible above the background value. Here, standard global normalization and thresholding are not practical, because the signals are so weak. Often we find that global thresholding is too crude, allowing in one case the gene expression to be quantified as a gain, and in another case allowing the same gene expression to be quantified as a loss. One possible solution for this problem is to use “segmental thresholding,” localized thresholding for each individual weak spot. Then the local background level is calculated against the global background level to obtain confidence level indices. The actual gene expression level for these low-abundance genes is then the “minithresholding level” divided by the confidence level. We realize that this is not a perfect solution; often we have to disregard these data points altogether.              Conclusions      The notion that the bioinformatics of microarray studies is still in its infancy pertains not only to studies in the aging area, but also to many other biological systems as well. The entire field of microarrays is experiencing volatile changes in methodological approaches, technological design, and bioinformatics development for data interpretation. Part of the growing pains in the use of microarray technology is the constant need for new cutting-edge technology and the reevaluation of methodology. Therefore, for any designer microarray projects, it is necessary to be vigilant for any new methodology and technology developments, to improve the design and fabrication process as well as the bioinformatic aspects of gene screening tasks.      As with all technologic advances, the microarray approach is not an end in itself; it is just a beginning. Obviously, one wants to know whether the genes identified as significantly changed at the RNA level are truly manifested at the protein level. For this purpose, the recent explosion of proteomic technology is certainly a testimony to the need for follow-up to microarray data. Ultimately, gene expression microarray studies have to be followed with experiments to examine protein changes, thus permitting a comprehensive examination of gene expression changes from RNA to protein levels.                  This work was supported by Grant R01 AG07444 from the National Institute on Aging of the National Institutes of Health, and from the Defense Advanced Research Project Agency of the Department of Defense, to E. Wang.      We express our sincere gratitude to Ms. Sherry Chen, Ms. Angel Wang, Mr. Keith Liang, Dr. Yih-Jing Tang, Dr. Nagathihalli Nagaraj, Dr. Bo Yu, and Ms. Jane Williams for their excellent technical assistance, and to Mr. Alan N. Bloch for proofreading this manuscript.                      1        Hayflick L, 2000. The future of ageing. Nature408:267-269.                     2        Perls T, 2002. Genetic and environmental influences on exceptional longevity and the AGE nomogram. Ann NY Acad Sci959:1-13.                     3        International Human Genome Sequencing Consortium2001. Initial sequencing and analysis of the human genome. Nature409:860-921.                     4        Schena M, Shalon D, Davis RW, Brown PO, 1995. Quantitative monitoring of gene expression patterns with a complementary DNA microarray. Science270:467-470.                     5        Beier M, Hoheisel JD, 1999. Versatile derivatisation of solid support media for covalent bonding on DNA-microchip. Nucleic Acids Res27:1970-1977.                     6        Stillman BA, Tonkinson JL, 2001. Expression microarray hybridization kinetics depend on length of the immobilized DNA but are independent of immobilization substrate. Anal Biochem295:149-157.                     7        Bertucci F, Bernard K, Loriod B, et al. 1999. Sensitivity issues in DNA array-based expression measurements and performance of nylon microarrays for small samples. Hum Mol Gen8:1715-1722.                     8        Stillman BA, Tonkinson JL, 2000. FAST slides: a novel surface for microarrays. Biotechniques29:630-635.                     9        Afanassiev V, Hanemann V, Wolf S, 2000. Preparation of DNA and protein microarrays on glass slides coated with an agarose film. Nucleic Acids Res28:e66                    10        Linftood K, Lijedahl U, Raitio M, Syvänen A-C, 2001. Minisequencing on oligonucletide microarrays: comparison of immobilization chemistries. Nucleic Acids Res29:e69                    11        Bordoni R, Consolandi C, Castiglioni B, et al. 2002. Investigation of the multiple anchors approach in oligonucleotide microarray preparation using linear and stem-loop structured probes. Nucleic Acids Res30:e34                    12        Podyminogin MA, Lukhtanow EA, Reed MW, 2001. Attachment of benzaldehyde-modified oligodeoxynucleotide probes to semicarbazide-coated glass. Nucleic Acids Res29:5090-5098.                     13        Maskos U, Southern EM, 1992. Oligonucleotide hybridizations on glass supports: a novel linker for oligonucleotide synthesis and hybridization properties of oligonucleotides synthesized in situ. Nucleic Acids Res.20:1679-1684.                     14        Beier M, Hoheisel JD, 2000. Production by quantitative photolithographic synthesis of individually quality checked DNA microarrays. Nucleic Acids Res28:e11                    15        Beier M, Hoheisel JD, 2002. Analysis of DNA-microarrays produced by inverse in situ oligonucleotide synthesis. J Biotechnol94:15-22.                     16        Fodor SPA, Read JL, Pirrung MC, Stryer L, Lu AT, Solas D, 1991. Light-directed spatially addressable parallel chemical synthesis. Science251:767-773.                     17        Singh-Gasson S, Green RD, Yue Y, et al. 1999. Maskless fabrication of light-directed oligonucleotide microarrays using a digital micromirror array. Nat Biotechnol17:974-978.                     18        LeProust E, Zhang H, Yu P, Zhou X, Gao X, 2001. Characterization of oligonucleotide synthesis on glass plates. Nucleic Acids Res29:2171-2180.                     19        O'Donnell-Maloney MJ, Smith CL, Cantor CR, 1996. The development of microfabricated arrays for DNA sequencing and analysis. Trends Biotechnol14:401-407.                     20        Hughes TR, Mao M, Jones AR, et al. 2001. Expression profiling using microarrays fabricated by an ink-jet oligonucleotide synthesizer. Nat Biotechnol19:342-347.                     21        Ramakrishnan R, Dorris D, Lublinsky A, et al. 2002. An assessment of Motorola CodeLink microarray performance for gene expression profiling applications. Nucleic Acids Res30:e30                    22        Bowtel DDL, 1999. Options available—from start to finish—for obtaining expression data by microarray. Nat Genet21:25-32.                     23        Knight J, 2001. When the chips are down. Nature410:860-861.                     24        Selvey S, Thompson EW, Matthaei K, Lea RA, Irving MG, Griffiths LR, 2001. Beta-actin—an unsuitable internal control for RT-PCR. Mol Cell Probes15:307-311.                     25        Khalyfa A, Bourbeau D, Chen E, et al. 2001. Characterization of elongation factor-1A (eEF1A-1) and eEF1A-2/S1 protein expression in normal and wasted mice. J Biol Chem276:22,915-22,922.                     26        Ramdas L, Coombes DR, Baggerly K, et al. 2001. Sources of nonlinearity in cDNA microarray expression measurements. Genome Biol2: (11) 0047                    27        Hotke JH, Ankenbauer W, Muhlegger K, et al. 1995. The digoxigenin (DIG) system for non-radioactive labeling and detection of nucleic acids—an overview. Cell Mol Biol.41:883-905.                     28        Semov A, Marcotte R, Semova N, Ye X, Wang E, 2002. Microarray analysis of E-box binding-related gene expression in young and replicatively senescent human fibroblasts. Anal Biochem.302:38-51.                     29        Thompson D, Larson G, 1992. Chromogenic phosphatase substrate producing a blue-colored precipitate at the site of enzymatic activity. Biotechniques12:656                    30        Miller RA, Galecki A, Shmookler-Reis RJ, 2001. Interpretation, design, and analysis of gene array expression experiments. J Gerontol Biol Sci56A:B52-B57.                     31        Hilsenbeck S, Friedrichs W, Schiff R, et al. 1999. Statistical analysis of array expression data as applied to the problem of tamoxifen resistance. J Natl Cancer Inst.91:453-459.                     32        Crescenzi M, Giuliani A, 2001. The main biological determinants of tumor line taxonomy eludicated by a principal component analysis of microarray data. FEBS Lett.19:114-118.                     33        Fellenberg K, Hauser NC, Broors B, Neutzner A, Hoheisel JD, Vingron M, 2001. Correspondence analysis applied to microarray data. Proc Natl Acad Sci USA.98:10,781-10,786.                     34        Yan PS, Chen CM, Shi H, Rahmaatpanah F, Wei SH, Caldwell CW, 2001. Dissecting complex epigenetic alterations in breast cancer among CpG island microarrays. Cancer Res23:8375-8380.                     35        Hoffman KF, McCarty TC, Segal OH, et al. 2001. Disease fingerprinting with cDNA microarrays reveals distinct gene expression profiles in lethal type 1 and type 2 cytokine-mediated inflammatory reactions. FASEB J.15:2545-2547.                     36        Huang J, Qi R, Quackenbush J, Dauway E, Lazaridis E, Yeatman T, 2001. Effects of ischemia on gene expression. J Surg Res99:222-227.                     37        Dysvik B, Jonassen I, 2001. J-Express: exploring gene expression data using Java. Bioinformatics17:369-370.                     38        Khan J, Simon R, Bittner M, et al. 1998. Gene expression profiling of alveolar rhabdomyosarcoma with cDNA microarrays. Cancer Res58:5009-5013.                     39        Hess KR, Fuller GN, Rhee CH, Zhang W, 2001. Statistical pattern analysis of gene expression profiles for glioblastoma tissues and cell lines. Int J Mol8:183-188.                     40        Helgason A, Hickey E, Goodacre S, et al. 2001. mtDNA and the islands of the North Atlantic: estimating the proportions of Norse and Gaelic ancestry. Am J Hum Genet68:723-737.                     41        Tamayo P, Slonim D, Mesirov J, et al. 1999. Interpreting patterns of gene expression with self-organizing maps: methods and application to hematopoietic differentiation. Proc Natl Acad Sci USA.96:2907-2912.                     42        Tusher VG, Tibshirani R, Chu G, 2001. Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci USA98:5116-5121.