High-density microarrays for gene expression analysis

Changes in gene expression are associated with many important biological phenomena including morphogenesis and aging, cancer and disease states, and adaptive responses to the environment. To detect these changes, molecular biologists have, over the years, developed a variety of techniques including RNA blots (1), differential display (2), ribonuclease protection assays (3), and reverse transcription–polymerase chain reaction (RT-PCR) (4). Although these methods are effective, being sensitive and accurate, their dependence on gel electrophoresis places an empirical limit on the number of samples that can be simultaneously analyzed. High-density microarrays, first described by Schena et al. (5), provide the important advantage of allowing parallel quantification of the expression levels of an almost unlimited number of genes from a given genome. This greatly accelerates the characterization of individual genes and provides the means to decipher complex, multigenic patterns of gene expression (6). The importance of the high-throughput capability of microarrays is particularly relevant in the context of the recent explosion in genomic sequence acquisition. Each day, the human genome sequencing project alone deposits an estimated 2 million bases of DNA sequence into the GenBank database (7). For other eukaryotes, including Saccharomyces cerevisiae (8), Caenorhabditis elegans (9), and Drosophila melanogaster (10), the complete genomic sequence has already been reported. By the end of the year 2000, the list of organisms having fully sequenced genomes should comprise at least 36 different species (11), including the higher plant Arabidopsis thaliana. The rapid growth in accumulation of genomic sequences has been paralleled by the growth in collections of expressed sequence tags (ESTs), each of which contains the partial sequence of an expressed gene (12). For genomic or EST sequences to be truly informative, the genes that they identify must be assigned biological functions. Expression profiling using microarrays provides a powerful tool for correlating gene functions with DNA sequences. DNA microarrays also are useful for identifying differentially expressed genes in organisms for which the available DNA sequence data are limited. This advantage is not shared by other high-throughput methods of expression analysis such as serial analysis of gene expression (SAGE) (13). The use of microarrays for expression profiling is based on two fundamental principles. First, for many genes, a predominant factor underlying changes in expression is an alteration in the abundance of the cognate mRNA. It is clear that posttranscriptional factors also can affect gene expression, but these factors are not considered in this review because they are not generally amenable to microarray analysis. Second, only DNA strands possessing complementary sequences can hybridize to each other to form a stable, double-stranded molecule. Microarrays exploit this property through the immobilization of millions of single-strand copies of a gene as individual array elements on a solid support surface. The array surface is then incubated with a mixture of labeled DNA molecules; the surface contains a proportional representation of all of the genes that are being expressed in a given tissue sample. Out of this mixture, only the labeled molecules that represent the same gene as the immobilized DNA elements can form heteroduplexes. By measuring the amount of label that is bound to each array element at the end of the hybridization reaction, a researcher can determine the relative transcript abundance level of each gene. Because each microarray comprises many elements, RNA abundance levels for thousands of genes can be measured in a single experiment. By comparing abundance levels from several experiments, the investigator can correlate patterns of gene expression with particular tissues or experimental conditions. Despite being a relatively new field of research, some aspects of microarray nomenclature have become ambiguous. In this review, we follow the convention established by the editors of Nature Genetics who argued that the immobilized DNA should be termed the probe and the labeled DNA the target because of the similarities between microarraying and the older technique of reverse dot blotting (14). Two types of microarray are commonly used for expression profiling. These arrays differ primarily in the length of the probes that make up the array elements. For DNA microarrays, which are printed on membranes or glass, the array elements are large fragments of genes (∼400–2,000 bp) typically produced through PCR amplification. In contrast, oligonucleotide-based microarrays [otherwise termed DNA chips (15)] have array elements comprising short (∼20 bp), synthetic DNA molecules. The production of oligonucleotide microarrays is usually done commercially because it requires specialized equipment and procedures that are not typically available to research laboratories. Therefore, we provide only a brief description of oligonucleotide microarrays, followed by a detailed discussion of DNA microarrays. We then discuss analyses and applications of both types. Most oligonucleotide arrays in current use are produced by Affymetrix Inc. (Santa Clara, CA; Fig. 1a–e). As the first step in the production of Affymetrix GeneChips, unigene EST databases are scanned for sequences representing genes of interest (16). Based on these sequences, a set of 16 different complementary 25-mer oligonucleotides is designed for each selected gene. In addition, for each oligonucleotide, a mismatch probe is also designed, which differs from the original sequence at only one base position and therefore can be used during hybridization as a control for stringency. Production of Affymetrix GeneChip oligonucleotide microarrays. a: Each 1.28- × 1.28-cm microarray consists of up to 260,000 square features. Each feature contains more than 106 copies of a particular oligonucleotide probe. b: GeneChip arrays are packaged in cartridges for handling during hybridization and analysis. c: Oligonucleotides are synthesized on linker molecules attached to the array substrate. The linker molecules are capped by a photolabile protecting group, represented here as diamonds. d: Protecting groups are removed in specific features by masking parts of the array and exposing it to light. e: Nucleotides, which also have photolabile protecting groups, flood on to the array, where they bind to deprotected sites. The cycle of deprotection and coupling is repeated with different nucleotides and different masks to synthesize a different set of probes in each feature of the array. All probes on Affymetrix chips are synthesized in parallel, on the array surface, using a photolithographic process (15, 17). This process begins with an array substrate that is coated with photolabile linker molecules (Fig. 1c). Examples of photolabile protecting groups used in this context include 5′-O-dimethoxybenzoincarbonate (DMBOC) and {[(α-methyl-2-nitropiperonyl)-oxy]carbonyl} (MeNPOC) compounds (18, 19). Exposure to light removes the protecting group (Fig. 1d), leaving a hydroxyl that is then available for coupling to reactive DNA monomers (i.e., MeNPOC- or DMBOC-protected phosphoramidites of adenosine, cytosine, guanosine, or thymidine; Fig. 1e). Photolithographic masks are used to illuminate only those array locations at which a given monomer is to be attached. Unincorporated monomers are washed away, and deprotection is repeated with a different mask. This is followed by the coupling of the second of the four possible monomers at the newly deprotected sites. Subsequent cycles of washing, spatially directed light exposure, and coupling extend the oligonucleotides, thereby providing sequences that are predefined for each feature on the array. The result is the synthesis of approximately 106 copies of up to 260,000 different probes on a 1.28- × 1.28-cm array (16). Because each gene is represented by 40 different probes (including mismatch probes), the expression patterns of approximately 7,000 genes can be examined on the largest of the oligonucleotide arrays produced by Affymetrix. Affymetrix arrays are available for expression analyses of genes from six species: human, rat, mouse, yeast, Arabidopsis, and Escherichia coli (www.affymetrix.com). These arrays are expensive relative to the DNA microarrays that are produced by robotic spotting technology (described below), must be purchased from Affymetrix, and require an expensive proprietary fluorescence reader for analysis. Another disadvantage of Affymetrix chips is the lack of flexibility inherent in the synthesis process. Thus, should modifications to the chips be required, based on ongoing discoveries in the laboratory, a considerable lead time is required for the expensive process of probe and lithographic mask design before production of the modified chips. An alternative means of synthesizing oligonucleotide arrays, using ink-jet deposition, has recently been developed and promises to reduce costs and increase the flexibility of oligonucleotide array production (Agilent Technologies, Palo Alto, CA). With this technology, the time from completion of the design to fabrication of the array can take less than 1 day. Current state of the art permits the creation of arrays with more than 25,000 features (individual oligos up to 60 mer in length) on a 1- × 3-inch microscope slide (M. N. Kronick, personal communication). The production of DNA microarrays can be divided into three stages: (a) preparation of probe DNA, (b) preparation of a suitable array substrate, and (c) robotic deposition of array elements. Preparation of probe DNA is one of the most resource-intensive aspects of microarray production. Although linear DNA of any source in principle can be immobilized on microarrays (Fig. 2a), probes are generally produced by PCR amplification under high-throughput conditions (in 96- or 384-well plates; Fig. 2b). Inserts are typically amplified from cDNA and EST libraries. Library clones can be chosen randomly for arraying, particularly when the microarray is to be used as a screening tool for gene discovery. Alternatively, bioinformatics techniques can be used during clone selection (20), e.g., to construct arrays containing all members of a specific gene superfamily. Vector-specific primers are commonly employed for large-scale amplifications because the same pair of primers can be used to amplify every insert within the library and because they do not require sequence information of the cDNA inserts. However, because parts of vector sequences and poly(A) tail can be included in each probe amplified in this manner, the specificity of hybridization on the microarray can be diminished. An alternative is the use of unique pairs of gene-specific primers for the amplification of each probe. These primers can be designed to exclude the poly(A) tail and regions of sequence similarity between members of a gene family. Gene-specific primers also can be employed for amplification of probes from genomic DNA, in some cases obviating the need for EST/cDNA libraries. The use of gene-specific primer pairs is limited by expense and the degree to which genome sequencing has been completed. Production of DNA microarrays by microspotting. Starting with genomic DNA and/or cDNA and EST libraries (a), individual genes are amplified in multiwell plates (b). The polymerase chain reaction products are purified, concentrated, and then robotically deposited on coated glass slides with devices such as slotted steel pins (c). Tens of thousands of spots can be deposited on each slide, and each spot contains roughly 108 copies of a specific gene fragment. Before microarray printing, the amplified probes are purified and analyzed by gel electrophoresis. DNA amounts are then sometimes quantified by fluorometry (21). Electrophoretic analysis confirms the presence of a single amplicon and provides an estimate of probe concentration. To ensure that microarray hybridization signal intensity is proportional to target abundance, the final concentration of probe DNA must be in excess relative to the expected concentrations of labeled target for each gene. In most published reports, probes for microarrays on glass are printed at concentrations of 100–500 ng/μl (22, 23). The minimum concentration of probe DNA required likely depends on factors such as probe length, base composition, and the binding capacity of the array substrate. Heller et al. (24) examined the effects of printing various dilutions of different probes with lengths of 0.2–1.2 kb. Dilution of probe solutions up to eightfold (i.e., a minimal DNA concentration of 125 ng/μl) affected hybridization signal intensities for only the shortest probes. Other reports have indicated that short probes (<300 bp) produce unreliable data in expression analyses (25). Microarrays are formed by robotically depositing probes in a regular pattern on a solid support (Fig. 2c). Nylon filters can be used as substrates for arrays containing many elements (e.g., 36,000; Incyte Microarray Systems, Fremont, CA) (26, 27). However, because of the deformability of these filters, array elements cannot be printed on nylon filters at high density, meaning that a 36,000-element array necessarily occupies an area of at least 22 × 22 cm. High-density microarrays, which can be operationally defined as having element spacings (center-to-center) of 0.5 mm or less, are printed on solid substrates, typically glass microscope slides that have been coated to increase hydrophobicity and enhance probe adhesion. The rigidity of glass, its desirable optical properties, and the fact that it can be manufactured to exceptional surface flatness allow arrays to be printed at element spacings as close as 60 μm. In current practice, elements are generally arrayed at a minimal spacing of 150 μm (28); thus, more than 10,000 elements can be arrayed within the area of a standard (22 × 22 mm) coverslip. Minimizing the array area also minimizes the total volume required for the hybridization reaction. This is particularly important for experiments involving targets from low-abundance sources, e.g., tumor biopsies or plant meristems. Two other properties of glass that make it an excellent substrate for high-density microarrays are its low porosity, which helps minimize hybridization volumes, and its low inherent fluorescence, which minimizes background signals during analysis (29). Glass slides are coated before printing to increase the binding efficiency of probe DNA to the surface. A common coating material is poly-L-lysine, which binds to glass through noncovalent linkages involving the primary amines of the lysine residues. These amines also adsorb electrostatically to the phosphate groups of DNA to immobilize the probe on the slide (30). Other noncovalent interactions, including hydrophobic and hydrogen bonds, also likely contribute to the coupling of DNA with poly-L-lysine. Amine terminal silanes, which are a second class of coating, likewise provide positively charged amine groups for binding probe DNA but differ from poly-L-lysine in their covalent attachment to glass. A third type of surface coating, aldehyde terminal silanes, bind covalently to the glass surface and the probe DNA with Schiff's base formation that is subsequently reduced (TeleChem International, Sunnyvale, CA). Schiff's base formation occurs between aldehydes of the coating material and amine groups within the DNA molecules, either those occurring naturally in the bases of DNA or those provided in the form of C6-amino modified PCR primers used during probe production (22). Although covalent attachment of DNA to the slide might be expected to diminish the loss of probes during denaturation and washing, the numerous published reports that used slides coated with poly-L-lysine or amine terminal silane have indicated that noncovalent attachment of DNA to coated slides is sufficient for microarray production (23, 31). Specialized robotic instruments are used to deposit probe DNA onto the coated glass slide. Some instruments use capillary tubes, piezoelectric ink-jets, solid pins, or pin-and-loop systems for probe deposition (32). However, the most common design uses slotted steel pins to transfer probes from their source plate to the slide. Much like a quill pen, these slotted pins collect the probe by capillary action and then deposit small aliquots of it each time they contact the slide surface. Each pin typically takes up 0.1–0.6 μl of DNA solution from the source plate and deposits approximately 0.3–1 nl as each array element (33, 34). This means that at least 400 spots can be printed from a single loading of the pin, although in practice this number can be limited by factors such as evaporation or by the total numbers of slides that can be positioned on the printing platen. The printing times are typically one to spots per second per pin, depending on the instrument and microarray parameters (26). This estimate does not include the time required for the rinsing, drying, and pin sonication cycles, which are usually programmed to precede each loading of the pin and change of source plates. Most robotic printers can print simultaneously with up to 32 pins, which greatly enhances the total printing rate. For example, GeneMachines (San Carlos, CA) has reported that their OmniGrid equipped with 32 pins (Majer Precision Engineering, Tempe, AZ) can print a 34,000-element microarray on 100 slides in 17 h. For the preparation of labeled target, RNA is extracted by standard methods from any tissue of interest. Total cellular RNA (22, 35-37), poly(A)+ mRNA (25, 31), or polysomal fractions of poly(A)+ mRNA (38) can be used. Hybridization to DNA microarrays typically requires 100 μg of total RNA, which is the approximate yield from 100 mg of mammalian tissue or 300 mg of Arabidopsis leaves. Poly(A)+ RNA normally comprises ∼1–2% (w/w) of the total RNA (36). For 18- × 18-mm DNA microarrays printed on glass, hybridization can be successful with as little as 1 μg of poly(A)+ RNA (31, 39). RNA from microscopic amounts of tissue also can be successfully employed for microarray hybridization, if it is first amplified. Procedures for the linear amplification of RNA populations are based on in vitro transcription from an introduced T7 promoter and can increase RNA abundance up to 1 millionfold (40, 41). Before hybridization, target RNA is reverse transcribed into cDNA. For filter-based arrays, targets are generally labeled by incorporating radioactive nucleotides (27). In the case of oligonucleotide arrays and DNA microarrays printed on glass, targets are ultimately labeled with fluorescent dyes (42) (Fig. 3a,b). Nucleotides that have been conjugated directly or indirectly to fluorochromes, such as and have been used with oligonucleotide and DNA arrays The dyes and are the most for DNA microarrays because of their and This is important because it two different populations of labeled targets to be simultaneously to a single DNA Affymetrix oligonucleotide arrays use methods that do not Hybridization and of microarrays. or DNA microarrays of DNA attached to a solid substrate (a) are with the targets from experimental and RNA which labeled with 1 and (b). c: target molecules hybridize to complementary with a proportional to their relative abundance in the original mixture of hybridization and washing, the amounts of the two labeled targets that have at each spot on the array are by with or at specific for those targets for DNA microarrays printed on glass, fluorescent nucleotides can be directly into the cDNA However, some a in which labeled nucleotides are during a synthesis by the of DNA polymerase Another the of nucleotides during cDNA followed by reaction of the cDNA with Although is the the is to possible of some reverse the of as to However, not all this to be targets for Affymetrix oligonucleotide arrays are as which is transcribed from approximately 1 μg of cDNA Because many are transcribed from each cDNA this process the original mRNA nucleotides are during and these targets are hybridization by with target molecules are to all types of microarrays under conditions to those used for blots (Fig. such as standard are used to reduce between probe and target molecules. such as DNA, DNA, and DNA are to reduce of double-stranded probes are in to the probes. arrays are in or Hybridization on glass slides is under a in a Affymetrix oligonucleotide arrays are in cartridges that also as hybridization all types of arrays, as for standard and can be by the concentration or the during hybridization and In is with a in hybridization and in which has not been can lead to by of genes that small regions of sequence with DNA microarrays indicated that short regions of with up to sequence produced only low levels of However, when several of these short regions of a DNA the of Other experiments, which used oligonucleotides as that sequences with complementary sequence between probe and target not contribute to the hybridization the of to a of or more bases We levels of specificity when using PCR products as probes during a analysis of the of Arabidopsis and types of for hybridization can be included as probes on DNA microarrays, including genes from genomic sequences, or probes containing only or hybridization on Affymetrix oligonucleotide arrays is gene by gene by comparing signals between probes and their probes. hybridization and washing, the amount of target bound at each spot is quantified (Fig. labeled targets are with a labeled targets are by or light and are with specialized instruments with or instruments for arrays on glass slides at as high as with a of and a of four of e.g., instruments from and CA). Although some are of and in up to four different is most are and then reported as a typically with a the of each spot in the must be usually with This the signal intensity at each most as the or of all intensities within the spot for the analysis of DNA microarrays also the background intensity the spot filters can be during e.g., to exclude with low or high because these are by background fluorescence or A detailed discussion of methods for array quantification has been The of most microarray experiments is to correlate changes in gene-specific RNA abundance with particular or conditions. As RNA samples are usually analyzed in with one being from a control state and the other from the experimental The signal intensities from each must then be or relative to each other to in which are not to changes in relative RNA abundance for individual of between microarray experiments include in the amount of in each target preparation and in the efficiency of hybridization, and The of some not of these is by simultaneously targets from two different to a single array, as is common with most DNA microarray experiments but not with oligonucleotide array experiments (Fig. hybridization to an DNA sequence tags representing more than Arabidopsis genes in arrays at spacing on an microscope slide. This slide with targets from with and control with Arabidopsis that most with the or targets or and represent genes that are differentially expressed in to an EST to both the spots or of of the can be by during or data can be used to determine the degree of Each depends on the selection of a set of signals levels are not expected to change as a result of the experimental For example, some a of genes that are not expected to be differentially expressed under most conditions with this is its dependence on the of the expression patterns of the As an a set of probes from genes with sequence similarities to other array elements can be included as of the of in vitro RNA to these are then to the control and experimental RNA samples before (36). The factor is then by comparing the signal intensities of these in the two different This is limited by the with which proportional of the genes can be into each RNA and is less than other because of the and experimental it A and more of uses the set of signals from each or as the This on the that only a small of genes represented on a given array be differentially when the two complete data are the is to the measured between the two of data The data can therefore be by each signal in the by a factor that is proportional to the This factor is normally by linear of the two populations or by the of the of all signals from the two data We the of as a for all DNA microarray experiments, for cases in which the selection of probes is small or one set of experimental as when are