| Nucleic Acids Research | Pages |
Rapid analysis of gene expression (RAGE) facilitates universal expression profiling
Introduction
Materials And Methods
Preparation of cDNA
Preparation of templates
Preparation of ICMLN62
PCR
Informatics
Product analysis
Results
Description of the RAGE technique
Specificity, reproducibility and sensitivity of RAGE
Analysis of expression changes in K5E2F1 transgenic mice
Gene discovery by RAGE
Discussion
Supplementary Material
Acknowledgements
References
Rapid analysis of gene expression (RAGE) facilitates universal expression profiling
Received August 6, 1999; Revised October 1, 1999; Accepted October 22, 1999
ABSTRACT Current techniques for analysis of gene expression either monitor one gene at a time, for example northern hybridization or RT-PCR methods, or are designed for the simultaneous analysis of thousands of genes, for example microarray hybridization or serial analysis of gene expression. To provide a flexible, intermediate scale alternative, a PCR-based method for the rapid analysis of gene expression has been developed which allows expression changes to be determined in either a directed search of known genes, or an undirected survey of unknown genes. A single set of reagents and reaction conditions allows analyses of most genes in any eukaryote. The method is useful for assaying on the order of tens to hundreds of genes in multiple samples. Control experiments indicate reliable detection of changes in gene expression 2-fold and greater, and sensitivity of detection better than 1 in 10 000. Analyses of over 400 genes in a mouse system transgenic for the E2F1 gene have identified several new downstream targets of E2F1, including Brca1 and Cdk7, in addition to several unidentified genes that are upregulated in the transgenic mice. Changes in expression of several genes related to apoptosis suggest a possible potentiation of apoptotic pathways in the transgenic keratinocytes.
INTRODUCTION
Our knowledge of the nucleotide sequences of eukaryotic genes has been expanding dramatically for the past decade. This increase has been driven by advances in DNA sequencing technologies derived from the human genome project, and should culminate in a complete human genome sequence in the near future. This has raised the possibility of determining the patterns of expression of all genes, and asking how this pattern changes during development, in disease states, or after exposure to external stimuli. Global gene expression patterns have been reported in Arabidopsis (1), yeast subjected to altered growth conditions (2-6), in several types of human tumors (7-10) and in quiescent human fibroblasts stimulated to grow by serum addition (11).
Two major strategies for global gene expression analysis are currently available. Serial analysis of gene expression (SAGE) provides a comprehensive picture of gene expression, but cannot be directed specifically at a small subset of the transcriptome (10,12). Data on the most abundant transcripts are obtained first, and ~1 Mb of sequencing data is needed for analysis of low abundance transcripts. Hybridization to microarrays of cDNAs or oligonucleotides can be used to direct a search towards specific subsets of genes, but these techniques cannot identify novel genes and the arrays are costly to produce (1,7). Both approaches are best suited for analysis of large numbers of genes, and lack the flexibility to easily incorporate different subsets of the transcriptome into the analysis or to focus on smaller subsets of genes for more detailed analyses. These factors make it impractical for many investigators to use either SAGE or microarray methods, especially in projects involving large numbers of samples.
To address these issues we have developed a method for the rapid analysis of gene expression (RAGE) which allows expression changes for most genes to be determined in either a directed search of known genes, or a more comprehensive but undirected survey. In both modes, unknown genes are also detected. Unique, gene-specific target sequences averaging ~128 bp in length are prepared from each cDNA by poly(A) selection and digestion with two restriction nucleases, ligated to common primer binding sites, and amplified by PCR. The PCR primers contain a common region, derived from the linker, plus a 3[prime]-specificity region of 3 or 4 nt, allowing them to selectively amplify cDNAs from a small number of genes. For known genes, the identity of the cDNA amplified can be deduced from the sequence of the primers and the length of the amplimer produced. The intensity of the amplimer band on the gel is a relative measure of the frequency of the corresponding mRNA in the total population of mRNAs.
We recently described the generation of transgenic mice expressing the cell cycle-regulated transcription factor E2F1 under the control of a keratin 5 (K5) promoter (13,14). Deregulated expression of E2F1 in basal keratinocytes results in epidermal hyperplasia, hyperproliferation and aberrant p53-dependent apoptosis. We have used keratinocytes from these mice as a model system for exploring the utility of the RAGE method for determining changes in gene expression.
MATERIALS AND METHODS
Sequence information for EIG-2, EIG-3 and EIG-4 has been deposited in GenBank (AF176514, AF176515, AF176516).
Preparation of cDNA
Mice transgenic for the human E2F-1 gene under the control of the bovine keratin 5 promoter were maintained as heterozygotes. All animal manipulations were carried out in accordance with Institutional regulations. Epidermal keratinocyte cultures were derived from newborn mice carrying the transgene or from their wild-type siblings and maintained as described (14). Total RNA was prepared by extraction into a chaotropic salt solution and organic solvent extraction using either a Qiagen (Valencia, CA) or a Gibco BRL (Gaithersburg, MD) kit, and used directly for northern analyses. Alternatively, mRNA was prepared using a Qiagen kit, and double-stranded cDNA was synthesized using a Gibco BRL kit but substituting biotinylated p(dT)18 as the primer for first strand synthesis. To prepare mouse epidermal RNA, adult mice were sacrificed and dorsal skin was dissected, heated to 55°C in DEPC-treated H2O for 30 s, and cooled to 4°C for 30 s. Skin samples were placed epidermal side down in 5 ml Trizol (Gibco BRL) for 30 s, and the epidermal layer was then scraped into the Trizol with the edge of a glass microscope slide. RNA extraction proceeded as described above.
SKBR3 human breast tumor cells were grown in DMEM containing 10% fetal bovine serum. A normal human mammary epithelial cell line, HME87 (15), was grown in serum-free medium (MEGM, Clonetics, Walkersville, MD). Cell monolayers were rinsed with saline, and RNA isolated as described above for keratinocytes.
Preparation of templates
Double-stranded linkers with overhangs complementary to the ends created by restriction with NlaIII (A-linker) and DpnII (B-linker) were prepared separately by mixing equal amounts of the following oligonucleotides, warming to 90°C for 2 min and slowly cooling to room temperature: A-linker, 5[prime]-CGTC-TAGACAGC (previously phosphorylated with T4 polynucleotide kinase) and 5[prime]-GCTGTCTAGACGCATG; B-linker, 5[prime]-CG-GTGATGCATC and 5[prime]-GATCGATGCATCACCG (previously phosphorylated with T4 polynucleotide kinase).
cDNA (1.5 µg) was digested with DpnII, the 3[prime]-most DpnII fragment of each cDNA was adsorbed to streptavidin/magnetic beads (Dynal, Lake Success, NY) and non-biotinylated fragments were removed. B-linker (217 ng) was added to the cDNA fragments bound to the beads, warmed to 50°C for 2 min, cooled to room temperature for 15 min, then cooled on ice. Ligation was accomplished by adding 10 U T4 DNA ligase (Gibco BRL) and incubating in a final volume of 50 µl for 2 h at 16°C. The preparation was then digested with NlaIII, and fragments released from the beads were recovered and ligated to the A-linker (217 ng) under similar conditions. These fragments of cDNA, containing the gene-specific targets ligated to the B and A linkers, are referred to as B/A bitags. A second preparation, A/B bitags, was obtained by performing NlaIII restriction and A-linker ligation prior to DpnII restriction and B-linker ligation.
Preparation of ICMLN62
To prepare an internal control template that would mimic the human MLN62 bitag, we first created a 10 bp insertion in the MLN62 bitag by PCR. Two primers were used: one a normal RAGE primer specific for the B-end of MLN62, 5[prime]-CGGTGATGCATCGATCTCA; the second containing the normal A-end RAGE primer for MLN62, a (GT)5 insertion, and the next 8 nt of the MLN62 bitag, 5[prime]-GCTGTCTAGACGCATGCCTT(GT)5GGTGCCGG. An amplimer ~10 bp longer than the normal MLN62 product was produced using these primers with bitags prepared from the SKBR3 cell line, and gel purified. This amplimer was cut with DpnII, a biotinylated linker based on the B-end primer was ligated on, and the resulting molecule (see Fig. 3A) was gel purified and quantitated.
PCR
Two sets of primers for RAGE PCR reactions were synthesized, corresponding to the A- and B-linkers above, but containing 3 or 4 nt specificity regions at the 3[prime] end. The sequences of these primer sets were: A-end (256 primers)-5[prime]-GCTGTCTAGACGCATGNNNN and B-end (64 primers)-5[prime]-CGGTGATGCATCGATCNNN.
PCR reactions contained (total volume 25 µl) bitags equivalent to 0.1-4 ng cDNA, 40 pg of each RAGE primer, 0.25 µl AmpliTaq (Perkin-Elmer) and 1× `D' buffer (Epicentre Technologies, Madison, WI). Reactions were run in a Stratagene RoboCycler with an initial denaturation of 5 min at 95°C, 2 min at 60°C and 1 min at 72°C followed by 26 cycles of 0.5 min at 95°C, 1 min at 60°C and 1 min at 72°C. The final extension at 72°C was increased to 6 min.
Informatics
mRNA sequences for the genes to be assayed were downloaded from GenBank and scanned for the presence of NlaIII and DpnII sites beginning at the 3[prime] end of the sequence. Where several accessions for the same gene were available, an accession containing an explicit poly(A) site and/or a poly(A) signal (AATAAA or ATTAAA) close to the 3[prime] end was used whenever possible. For each gene, the pattern of these restriction sites was used to predict the pair of RAGE primers and the bitag orientation to be used for amplification and the expected length and sequence of the amplification product. The locus names, common names and accession numbers of the genes described herein are as follows.
Murine genes. Actb, [beta]-actin, X03672; Actg, [gamma]-actin, M21495; Anx3, annexin III, AJ001633; Bax, Bax-[alpha], L22472; Brca1, BRCA1, U36475; Ccne, cyclin E, X75888; Ccnb2, cyclin B2, X66032; Ccng, cyclin G, Z37110; Cdc2d, cdc2, U58633; Cdk7, cyclin activating kinase, X74145; Cdkn1a, p21/WAF1, U24173; Cdkn2a, p19ARF, L76092; Gapd, glyceraldehyde 3-phosphate dehydrogenase, M32599; Hfh2, homolog of forkhead 2, AF067421; Lmna, lamin A, D13181; Mdm2, MDM2, X58876; Odc, ornithine decarboxylase, S64539; Rpl5, ribosomal protein L5, X83590; Rps17, ribosomal protein S17, D25213; Yy1, YY-1, M73963.
Human genes. ARF3, ADP-ribosylation factor 3, M74491; CAPN4, Calpain, X04106; CASP7, caspase 7, Y13088; CASP8, caspase 8, AF067834; CCNA2, cyclin A, Z26580; CREB2, CRE binding protein 1, U16028; ETR101, transcription factor ETR01, M62831; HMG14, HMG14, X53476; HSPB1, heat shock protein HSP27, X54079; IGFBP2, IGF binding protein 2, X16302; MYBL2, B-myb, X70472; MYC, c-myc, X01023; POLA2, DNA polymerase-[alpha], D13546; PSMD12, 26S proteasome subunit p55, AB003103; RPL31, ribosomal protein L31, X69181; RPS5, ribosomal protein S5, U14970; RPS16, ribosomal protein S16, M60854; TCEB3, elongin A, L47345; TK1, thymidine kinase, M11945; TOP3, DNA topoisomerase III, U43431; TRAF4, oncogene MLN62, X80200; VEGF, vascular endothelial growth factor, M27281; XPC, XPC, X65024.
Product analysis
After addition of 1/10 vol 10× sample buffer (7.0 M urea, 0.4% bromphenol blue, 50 mM Tris, 20 mM EDTA, pH 7.5), portions of the reactions were analyzed by electrophoresis on 8% polyacrylamide gels. DNA fragments were stained with Vistra Green (Molecular Probes, Eugene, OR) and digitized fluorescent images were obtained with a FluorImager (Molecular Dynamics, Sunnyvale, CA). Data was collected under conditions where the fluorescent signal increased linearly with template concentration. For some low abundance amplimers, the number of PCR cycles was increased to 30 to obtain a more reliable signal.
RESULTS
Description of the RAGE technique
The details of the method are given in Figure 1 using the cDNA for annexin III as an illustrative example. cDNA was synthesized with biotinylated oligo (dT) as first strand primer, digested with a frequent-cutting restriction endonuclease (DpnII), and the 3[prime]-most fragment recovered by binding to a streptavidin-coated bead. To provide a common priming sequence, a 16-bp adapter with a DpnII compatible sticky end (B-linker) was ligated onto the cDNA fragments. The fragments were then digested with a second frequent cutting restriction endonuclease (NlaIII) and a second common priming site (A-linker) was added by ligation. This procedure resulted in a template preparation (B/A bitags) that contained a single gene-specific target sequence from each cDNA, with common A and B linkers at the two ends. Because each of the two enzymes used has a recognition site on average every 256 bp, the average size of the gene-specific target sequences was expected to be ~128 bp. Thus, the sequence complexity of the bitag preparation was reduced by ~15-fold relative to the cDNA population, assuming an average mRNA size of ~2 kb.
Figure 1. Preparation of bitag templates. The method for preparation of B/A bitags described in the text is illustrated for the Anx3 gene.
The scheme illustrated in Figure 1 will not produce an amplifiable bitag from cDNAs in which there is no NlaIII restriction site between the last DpnII restriction site and the poly(A) tail. To prepare a template suitable for analysis of these cDNAs, a second preparation, A/B bitags, was made by reversing the order in which the DpnII and NlaIII cuts were made. cDNAs in which DpnII and NlaIII restriction sites are absent or separated by <6 bp are refractory to this analysis; empirically this corresponds to ~5-10% of the transcriptome.
PCR reactions using primers containing only the A- and B-linker sequences would be expected to amplify all of the gene-specific targets in these bitag preparations. To provide specificity to the PCR reactions, RAGE primers were constructed containing the A-linker sequence (16 nt) followed by 4 variable nt (256 different primers) or the B-linker sequence (16 nt) followed by 3 variable nt (64 different primers). These primers can be combined pairwise with the two orientations of bitags to produce (256 × 64 × 2) = 32 768 unique reactions. The presence of relatively long common regions in the RAGE primers allows near-optimal amplification with all primers under a single set of PCR conditions. For cDNAs of known sequence, a single pair of primers that will amplify the gene-specific target in one bitag orientation can be predicted from the sequence, as well as the size of the resulting fragment (amplimer). Current estimates of the number of genes in the human transcriptome range from ~60 000 to 100 000. Since the RAGE method effectively divides the transcriptome randomly among 32 768 unique reactions, each pair of RAGE primers tested may give several amplimers (2-3 on average), only one of which corresponds to the gene being assayed. In general, these amplimers can be distinguished by size.
Specificity, reproducibility and sensitivity of RAGE
As an example of the specificity of the method, reactions were performed with primers predicted to produce a 291 bp amplimer from the murine Brca1 gene. As template for these reactions, mRNA was prepared from cultures of mouse keratinocytes and bitags were prepared as described above. To test the selectivity of the method, two pairs of primers that differed by a single nucleotide from the Brca1 primers were also chosen that were expected to produce amplimers of 117 and 197 bp from the genes for annexin III and an anonymous cDNA (clone 2C11B), respectively. An amplimer of the expected size was produced in PCR reactions with each of these three pairs of primers (Fig. 2A), and in each case the expected amplimer was the major product. Significantly, production of the Brca1-specific amplimer was not detected in reactions designed to amplify annexin III or 2C11B, and vice versa. Densitometry of the stained gel indicated that the integrated intensity of the amplimer bands increased linearly with the amount of bitag template up to ~0.6 ng cDNA for annexin III and ~4 ng cDNA for Brca1 and 2C11B (Fig. 2B). In addition, when the number of PCR cycles was varied, the integrated intensities increased exponentially with increasing cycle number from 22 to 28 (Fig. 2C). A similar experiment was carried out using the RAGE primers predicted for three human genes, HSP27, RPS5 and MLN62, with bitags prepared from the human breast cancer cell line SKBR3. Again the primers selectively gave rise to amplimers of the expected sizes (Fig. 2D), the intensity of the bands increased linearly with template concentration over a defined range, and exponentially with PCR cycle number from 24 to 28 cycles (data not shown).
Figure 2. Selective detection of expressed genes. (A) RAGE primers specific for Brca1, Anx3 (annexin III), or an anonymous cDNA clone, 2C11B (U01139) were combined with B/A bitags prepared from murine keratinocytes and PCR amplification carried out for 27 cycles. PCR products were analyzed by PAGE. Wedges above the lanes indicate increasing concentrations of template (0.15-1.2 ng for Anx3-specific reactions, 2-8 ng for Brca1- and 2C11B-specific reactions). -, No template controls; M, molecular size markers. (B) The integrated intensities of the Anx3-specific amplimers in (A) were determined by densitometry, and are plotted as a function of template concentration. (C) Anx3-specific RAGE reactions were carried out as described, but the number of PCR cycles was varied from 22 to 30. The natural logarithm of the integrated intensity is plotted as a function of the number of cycles. (D) RAGE primers specific for HSPB1 (HSP27), RPS5 (ribosomal protein S5) or TRAF4 (MLN62 oncogene) were combined with A/B bitags prepared from normal human mammary epithelial cells, and PCR products were analyzed as in (A). Amplimer intensities increased linearly with template concentration up to ~0.4 ng/reaction for HSP27 and S5, and up to at least 3.8 ng/reaction for MLN62.
Figure 3. Sensitivity of detection of MLN62. (A) The structure of the ICMLN62 DNA is shown. (B) cDNA from normal human mammary epithelial cells (500 ng) was mixed with the ICMLN62 DNA (25 pg), and bitags were prepared. The bitags were used as template for duplicate MLN62-specific RAGE reactions as follows: 1, no template; 2, 0.3 ng; 3, 0.5 ng; 4, 0.7 ng; 5, 0.9 ng. The reactions analyzed in lane 6 contained 20 fg of ICMLN62 as template. The intensity of the ICMLN62-specific amplimer band (open arrowhead) increased linearly with template at least up to 80 fg/reaction (data not shown). (C) The intensities of the endogenous MLN62-specific amplimer (151 bp, closed symbols) and the ICMLN62-specific amplimer (161 bp, open symbols) were determined and are plotted as a function of bitag concentration. The ratio of the slopes of the two linear least-square fit lines was 1.28.
To verify that the products were indeed those expected, the 291 bp amplimer obtained with primers specific for the Brca1 gene (see Fig. 2A) was gel-purified and sequenced in both directions with the same primers used in the initial reaction; the experimentally determined sequence exactly matched the predicted sequence. To further validate the specificity of RAGE reactions, primer pairs were selected for nine other murine genes: several high abundance, housekeeping genes, Odc, Rpl5 and Actg; lower abundance genes, Mdm2, Ccng, Cdkn2a and Cdk7; and transcription factors Yy1 and Hfh2. Similarly, primer pairs were selected for 14 human genes: HSPB1, TRAF4, RPS5, RPS16, RPL31, CAPN4, XPC, IGFBP2, ARF3, PSMD12, CREB2, TOP3, TCEB3 and ETR101. Reactions were run with the appropriate primer pairs and bitag templates, reaction products of the expected size were again gel-purified and sequenced, and in each case the sequence obtained was >95% identical to the expected product.
To assess the reproducibility of the technique, replicate bitag preparations were assayed for the expression of three representative genes that displayed a wide range of expression. Four samples of A/B bitags were prepared independently from a single cDNA preparation, and quadruplicate reactions were carried out using three non-overlapping pairs of RAGE primers specific for the small subunit of calpain, MLN62, or the gene for the p55 component of the proteasome. The amplimers expected in these reactions, as well as two unidentified products of approximate lengths 76 and 130 bp obtained in the calpain-specific reactions, were quantitated. The means and standard deviations of the band intensities (in arbitrary units) were: p55, 3.0 ± 0.8; MLN62, 26.7 ± 7.1; calpain, 75.4 ± 19.9; unknown130, 46.1 ± 11.4; and unknown76, 41.6 ± 9.8. Although the relative expression levels for this group of amplimers varied over a 25-fold range, the overall coefficients of variation were all between 0.23 and 0.27.
To assess the sensitivity of the technique, an internal control template for the MLN62 gene (ICMLN62) was constructed as described in Materials and Methods. This template contained (Fig. 3A) the gene-specific target for the MLN62 gene with a 10 bp insertion, the A-end and B-end linkers, and a biotin label at one end. As expected, amplification of this template with MLN62-specific RAGE primers gave rise to a product 10 bp longer than the MNL62-specific amplimer (Fig. 3B, lane 6, arrowheads). The ICMLN62 DNA was mixed with cDNA prepared from HME87 normal human mammary epithelial cells at a 1:20 000 weight ratio, and A/B bitags were prepared. PCR reactions were performed using the MLN-specific RAGE primers, and amplimers corresponding in size to the MLN62 cDNA (closed arrowhead) and to the ICMLN62 (open arrowhead) were obtained (Fig. 3B, lanes 2-5). The intensity of both bands increased linearly with template concentration (Fig. 3C). From the ratio of intensities of these two bands in Figure 3B and the known level of addition of ICMLN62 to the cDNA, the molar concentration of the endogenous MLN62-specific bitag could be calculated as 0.6 amol/ng cDNA. Assuming an average mRNA length of 2000 nt this corresponded to an mRNA abundance of ~0.07%. Expression levels at least 8-fold lower than that of MLN62 can easily be detected in this system (compare the mean intensities of p55 and MLN62 described above), suggesting that the lower limit of detectability is <0.01% or ~30 molecules of mRNA per cell.
Analysis of expression changes in K5E2F1 transgenic mice
The RAGE technique is well suited to performing comparisons of relative expression levels of hundreds of selected genes in paired bitag preparations that differ in some biologically meaningful way. As an interesting test system, mouse keratinocytes overexpressing the cell-cycle regulated transcription factor E2F1 were compared to keratinocytes lacking this transgene.
Paired reactions were performed with bitag preparations derived from wild-type or K5 E2F1 transgenic keratinocytes (13,14). The relative concentrations of wild-type and transgenic bitags were adjusted to give approximately equal expression ratios for a set of control genes (ribosomal proteins Rpl5 and Rps17, Gapd, Actb and Actg) whose expression was not expected to change significantly with E2F1 overexpression. In all, reactions were performed for >400 known murine genes, including genes previously shown to be regulated by E2F1 and other genes related to cell proliferation, apoptosis, transcriptional regulation and signal transduction. The expected amplimers were detected in reactions with 223 pairs of primers. The remaining genes may not be expressed in keratinocytes, or their expression levels may be below the detection limit of the RAGE technique. A complete list of the genes assayed is available as Supplementary Material for this article in NAR Online. Information on the primers needed to assay each of these genes is available at our website (http://sciencepark.mdanderson.org/ggeg ). The amplimers produced by RAGE primers for several representative genes are discussed below. For genes that were of special interest, either intrinsically or due to expression changes detected in the first analyses, at least three independent determinations were carried out.
Eight genes previously shown to be transcriptionally regulated by E2F (14,16-18) were analyzed first. Replicate, paired analyses for two of these E2F1-inducible genes, cyclin E and cdc2, as well as analyses of a control gene, [beta]-actin, are shown in Figure 4. While production of the [beta]-actin-specific amplimer was identical with the wild-type and transgenic templates, the cdc2-specific amplimer was ~2-fold more abundant in transgenic keratinocytes and the cyclin E-specific amplimer was increased by ~5-fold. The 5-fold increase in expression of cyclin E seen here agrees well with a previous determination by northern hybridization (~6-fold) (14). Recent studies have shown an upregulation of the ARF mRNA product of the Cdkn2a locus by E2F1 overexpression, and implicated this induction in E2F1-induced, p53-mediated apoptosis (18-20). RAGE analyses indicated at least a 3-fold induction of Cdkn2a/p19ARF in the transgenic keratinocytes (Fig. 5, lane 7; Table 1). In all, six of the eight known E2F1-target genes exhibited 2-5-fold increases in steady-state expression in E2F1 transgenic keratinocytes (Table 1), while two more targets changed <2-fold. In addition, several other cell-cycle related genes, including Ccnb2, the cyclin activating kinase Cdk7 and Odc were upregulated 3-6-fold in the transgenic keratinocytes. Increased expression of the genes for several transcription factors (Hfh2 and Yy1) and the Brca1 tumor suppressor gene, genes not known to be E2F1-regulated, was also seen; several of these are illustrated in Figure 5. The Brca1 gene has been shown to be cell-cycle regulated (21,22), and potential E2F sites are present in the Brca1 promoter. Evidence for cell-cycle regulation of Cdk7 expression in fibroblasts has recently been presented (11).
Figure 4. Changes in expression of E2F1-target genes. cDNA and bitags were prepared from keratinocyte cultures derived from newborn wild-type mice (-) or their K5 E2F1 transgenic siblings (+). Replicate, paired PCR reactions were analyzed using primers specific for Cdc2 and Ccne (A) and Actb (B).
Figure 5. E2F1-dependent changes in expression of selected genes. RAGE PCR analysis was performed, using bitags prepared from wild-type (-) or K5-E2F1 transgenic (+) keratinocytes. Reactions contained RAGE primers chosen to amplify specific genes and the expected amplimers are indicated by black dots between the lanes. The selected genes and the size of the expected amplimers were: 1, Actg, 98 bp; 2, Rpl5, 130 bp; 3, Lmna, 291 bp; 4, Cdk7, 392 bp; 5, Yy1, 248 bp; 6, Hfh2, 254 bp; 7, Cdkn2a/p19ARF, 508 bp; and 8, Brca1, 291 bp.
Table 1. Determination of gene expression changes by RAGE
| Gene | Expression ratio |
| E2F target genes | |
| Ccne | 5.1 ± 1.0 |
| Cdkn2a | 2.7 ± 0.6 |
| Tk1 | 2.6 ± 1.2 |
| Pola1 | 2.4 ± 0.4 |
| Cdc2 | 2.4 ± 0.9 |
| Mybl2 | 2.1 ± 0.2 |
| Myc | 1.4 ± 0.2 |
| Ccna2 | 1.1 ± 0.1 |
| Other genes | |
| Ccnb2 | 6.4 ± 4.1 |
| Hfh2 | 4.9 ± 3.3 |
| Vegf | 3.7 ± 1.7 |
| Odc | 3.6 ± 1.0 |
| Yy1 | 3.1 ± 1.8 |
| Cdk7 | 3.1 ± 0.3 |
| Brca1 | 2.6 ± 1.0 |
| Casp7 | 2.4 ± 1.5 |
| Casp 8 | 2.1 ± 0.3 |
| p53 target genes | |
| Ccng | 2.3 ± 0.5 |
| Bax | 2.0 ± 0.7 |
| Mdm2 | 2.0 ± 0.2 |
| Cdkn1a | 1.4 ± 0.2 |
| Control genes | |
| Actb | 1.1 ± 0.2 |
| Rps17 | 1.1 ± 0.1 |
| Gapd | 1.0 ± 0.1 |
| Hmg14 | 0.9 ± 0.1 |
| Lmna | 1.1 ± 0.2 |
| Actg | 1.1 ± 0.2 |
| Rpl5 | 1.0 ± 0.2 |
To confirm the magnitude of changes in expression seen with the RAGE technique, we compared the expression of several selected genes between wild-type and transgenic keratinocytes by northern analyses. While the expression of Actg, the gene for [gamma]-actin, was approximately equal in wild-type and transgenic keratinocytes, the genes for Cdc2 and Cdkn2a/p19ARF were upregulated by E2F1 overexpression (Fig. 6A), in good agreement with the RAGE results (Figs 4 and 5, Table 2). Expression ratios were determined for nine genes by both RAGE and northern analyses, and the ratios were plotted as a scattergram (Fig. 6B). The data were well-fit by a straight line with slope close to 1.0, indicating a high degree of concordance between the two techniques.
Figure 6. Validation of expression changes by northern analyses. (A) Northern analyses were performed using 20 µg of total RNA from wild-type (-) or K5E2F1 transgenic (+) keratinocytes and probes specific for Cdkn2a/p19ARF, Cdc2 or Actg. (B) Similar analyses were carried out for six additional genes, and the expression ratio was determined after quantitation of the hybridized bands using a phosphorimager. The expression ratios for each of the nine genes determined from northern analysis (abcissa) is compared with the expression ratios obtained by RAGE analysis (ordinate). The line is that expected for perfect agreement between the two techniques.
Table 2. Expression changes in non-target genes
| Gene | RAGE primers | Amplimer length | Database match |
| EIG-1 | CATGCGGG/GATCCAG | 147 | EST AA245406 |
| EIG-2 | CATGCGCA/GATCTGA | 210 | AF119955 |
| EIG-3 | CATGCTTT/GATCCTG | 95 | No match |
| EIG-4 | CATGGCCA/GATCTTC | 157 | No match |
| EIG-5 | CATGATTT/GATCAGC | 132 | EST AV076207 |
To extend these findings to the in vivo situation, RNA was prepared from the epidermis of wild-type and E2F1 transgenic mice, and either used in northern analyses or to prepare bitags for RAGE analysis. As shown in Figure 7A and B, RAGE analysis indicated a dramatic upregulation of expression of several genes, including Odc, Ccng and Cdkn2a/p19ARF, due to overexpression of the E2F1 transgene, in both newborn keratinocytes and adult epidermis. The increase in Cdkn2a/p19ARF was particularly striking, and this induction was confirmed by northern analyses (Fig. 7C).
Figure 7. Expression changes in mouse skin and keratinocytes. RNA, cDNA and bitags were prepared from newborn keratinocytes (A) or epidermal extracts derived from adult wild-type (-) or K5 E2F1 transgenic (+) mice (B). In (A) and (B), PCR reactions contained RAGE primers chosen to amplify: 1, Rps5, 108 bp; 2, Actb, 176 bp; 3, Odc, 202 bp; 4, Ccng, 109 bp; and 5, Cdkn2a/p19ARF, 508 bp. Other symbols as in Figure 5. (C) Northern analyses were performed using 20 µg of total RNA from either newborn keratinocytes or adult skin of wild-type or transgenic mice and a probe specific for Cdkn2a/p19ARF.
Brca1, Cdkn2a/p19ARFand CDK7 were among the 23 genes that were upregulated by E2F1 overexpression (Table 1; Fig. 6, reactions 4, 7 and 8); in each of these cases the sequence of the amplimer was experimentally verified. Interestingly, the products of these three genes have been reported to increase the transactivation activity of the tumor suppressor protein p53 by post-transcriptional mechanisms (23-28). These results suggested the possibility that one or more aspects of p53 activity might be enhanced in the K5 E2F1 transgenic cells. Consistent with this possibility, RAGE analysis indicated that expression of three known downstream targets for p53 transactivation, Ccng, Bax and Mdm2 (29), was increased ~2-fold in E2F1 transgenic keratinocytes (Table 1). A fourth target of p53, Cdkn1a (p21/WAF1), was increased slightly in the transgenic cells.
Gene discovery by RAGE
As noted above, each RAGE reaction has the potential to amplify several genes, and many non-targeted amplimer bands were noted in the course of these studies (see for example reaction 6 in Fig. 5). Five non-targeted amplimer bands that exhibited changes in expression >3-fold in transgenic keratinocytes were identified (Table 2), and these amplimers were sequenced. Two amplimer sequences matched several ESTs, exemplified by AA245406 for EIG-1 and AV076207 for EIG-5. Subsequent to sequencing of EIG-2, an apoptosis-related gene, AIP1 (30), was described (AF119955) that contains a sequence 99% identical to EIG-2. EIG-3 and -4 had no matches in GenBank.
DISCUSSION
The RAGE technique was conceived as an accessible and flexible alternative or adjunct to currently available global gene expression technologies. In this manuscript we have established the extent to which the technique can give quantitative information on the expression of individual genes. The data in Figure 2 indicate a high degree of selectivity in the priming reaction in that single nucleotide differences in primers routinely select the appropriate template, even in the face of a more abundant closely related template (compare the low abundance MLN62 gene with that of more abundant ribosomal protein S5, Fig. 2D). In every case that we have tested, including fairly low abundance genes such as Brca1 and Cdk7, the actual sequence of the amplimers after gel-purification has agreed with that predicted by our database, confirming the selectivity.
The most important requirements for achieving quantitative measurements in PCR techniques are establishing that the signal increases linearly with template concentration and exponentially with the number of cycles (31). In both human and murine systems we have found that these requirements are satisfied under our conditions (Figs 2 and 3). The data on reproducibility indicate that the dispersion of measurement in the system is ~25% of the mean. This fairly high coefficient of variation is probably to be expected in a PCR-based system, and limits the extent to which quantitative variation can be assayed. Analysis of these data by multiple ANOVA (not shown) indicated that the overall dispersion is due to variance in repeated PCR measurements of a single bitag preparation, and to differences between the bitag preparations. While the use of a set of control genes helps to control for preparative variations, the inherent variability of PCR measurements remains. However, by performing multiple analyses on selected genes of interest, statistical significance can be attained for changes in gene expression that extend below the 2-fold level (Table 1). Power calculations (assuming a log-normal distribution of the ratio data) indicate that with a minimum of four observations, a 2-fold difference in the expression ratio can be detected with a power of 95%.
We have not attempted to show that the efficiency of detection of all genes is the same for the RAGE technique, and in fact control experiments suggest some differences between genes in the efficiency of preparation of the respective bitags (data not shown). For example, differences in G + C content of the gene-specific target sequences may result in different efficiencies of amplification for different genes. Therefore, absolute expression levels cannot routinely be measured except by performing internal control experiments (Fig. 3 for example). Thus, the technique is most suitable for measuring relative differences in expression between two cell populations. By computing relative expression ratios independently for each gene, the postulated differences in detection efficiency effectively cancel out. This is analogous to the approach taken in SAGE and microarray measurements (8,12). Nevertheless, the data obtained with an internal control template indicate that the technique can be extremely sensitive, giving reliable detection below the level of 0.01% of the total mRNA population.
Several genes are known to be upregulated by overexpression of the E2F1 transcription factor, and transgenic mice expressing human E2F1 in basal keratinocytes provided a useful initial test system for validating the RAGE technique. As shown in Table 1, several of the known targets of E2F1 transcriptional activation were found to be upregulated in the transgenic keratinocytes by RAGE (Fig. 4), and this difference was also found in extracts prepared from adult epidermis (Fig. 7). The RAGE results in both systems were confirmed by the more conventional northern hybridization technique (Figs 6 and 7).
The K5 E2F1 mice exhibit several interesting phenotypes, and the RAGE expression results have provided intriguing suggestions as to possible underlying mechanisms. The transgene is expressed in basal keratinocytes, and histochemical analyses of skin sections revealed an aberrant, increased level of apoptosis in the skin of the transgenic mice, as well as areas of hyperproliferation. Overexpression of E2F1 in mouse fibroblasts leads to Cdkn2a/p19ARF mediated apoptosis (18-20), and Brca1 overexpression has also been linked to p53-dependent apoptosis (23,32). An E2F1-mediated increase in the expression levels of Cdkn2a/p19ARF and Brca1 is also seen in keratinocytes and the epidermis of K5 E2F1 transgenic mice, and this correlates with increased apoptosis in the skin which is reduced in a p53-null background (14). A third gene that can increase p53 activity post-transcriptionally, Cdk7, is also upregulated in the transgenic keratinocytes, and several genes that are downstream targets of p53 are also activated in the K5 E2F1 transgenics. Although neither Brca1 nor Cdk7 has been previously shown to be E2F-regulated, we have found that the murine Brca1 promoter is E2F1 and Rb1 responsive, and that this responsiveness depends on consensus E2F1 binding sites (A. Wang et al., manuscript in preparation). The human BRCA1 gene is also responsive to E2F1. This suggests that the ability of E2F1 to induce p53-mediated apoptosis may involve several different intermediate effector molecules. Interestingly, while K5 E2F1/p53 null mice have an increased incidence of spontaneous skin tumors, the single K5 E2F1 transgenics are resistant to chemical carcinogenesis in the skin (33). It will be of interest to determine how Brca1 and/or Cdk7 participate in E2F1-mediated apoptosis.
The RAGE technique has proven to be of great utility in these studies. It provides expression information on selected genes that is reliable and at least semi-quantitative, at a relatively low cost and with a high degree of flexibility. Simultaneously, information on non-targeted genes is obtained giving the possibility of identifying novel genes whose expression changes in the experiment being analyzed. By sequencing the novel amplimer enough information is generally obtained (the average gene specific target length is ~128 bp) to design primers that allow the entire cDNA to be obtained. The sets of genes to be analyzed can easily be altered based on the results of initial experiments, and studies with multiple time points or dose regimens are easily accommodated. Indeed, RAGE may have utility as a complement to SAGE or microarray methods. For example, an initial SAGE screen for differentially expressed genes can provide a focussed set of genes that can then be studied in detail by RAGE in a cost-effective manner. The set of reagents that must be maintained for RAGE analyses (256 A-end primers and 64 B-end primers) is fairly small, can be generated easily by synthetic methods, and is the same for all organisms. This contrasts with microarray technologies where much larger sets of clones must be maintained and where each set of reagents is species specific. We are currently working on ways to speed up the analyses of the amplimers, possibly allowing RAGE to be performed in a high-throughput mode in which the entire transcriptome could be analyzed in a short period of time.
SUPPLEMENTARY MATERIAL
See Supplementary Material available at NAR Online (1.435 MB PDF File).
ACKNOWLEDGEMENTS
We gratefully acknowledge the help of Rebecca Deen, Michelle Gardiner, Judy Ing and Chris Yone in the preparation of this manuscript. This work was supported by grants from the American Cancer Society (RPG-96-001-03-CNE to M.C.M.) and the National Institutes of Health (CA79648 to D.G.J., CA35581 to M.C.M.), and by NIEHS Center Grant ES07784 and NCI Cancer Center Grant CA16672.
REFERENCES
*To whom correspondence should be addressed. Tel: +1 512 237 9541; Fax: +1 512 237 2475; Email: mmacleod{at}odin.mdacc.tmc.edu
This page is run by Oxford University Press, Great Clarendon Street, Oxford OX2 6DP, as part of the OUP Journals
Comments and feedback: jnl.info{at}oup.co.uk
Last modification:
Copyright© Oxford University Press, 1999.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  M. Gowda, H. Li, J. Alessi, F. Chen, R. Pratt, and G.-L. Wang Robust analysis of 5'-transcript ends (5'-RATE): a novel technique for transcriptome analysis and genome annotation Nucleic Acids Res., November 14, 2006; 34(19): e126 - e126. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Afshar, N. Jelluma, X. Yang, D. Basila, N. D. Arvold, A. Karlsson, G. L. Yount, T. B. Dansen, E. Koller, and D. A. Haas-Kogan Radiation-Induced Caspase-8 Mediates p53-Independent Apoptosis in Glioma Cells. Cancer Res., April 15, 2006; 66(8): 4223 - 4232. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 Y. Ma, J. Yuan, M. Huang, R. Jove, and W. D. Cress Regulation of the Cyclin D3 Promoter by E2F1 J. Biol. Chem., May 2, 2003; 278(19): 16770 - 16776. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
L. Smith, P. Underhill, C. Pritchard, Z. Tymowska-Lalanne, S. Abdul-Hussein, H. Hilton, L. Winchester, D. Williams, T. Freeman, S. Webb, et al. Single primer amplification (SPA) of cDNA for microarray expression analysis Nucleic Acids Res., February 1, 2003; 31(3): e9 - e9. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. Wang, J. Gu, K. Judson-Kremer, K.L. Powell, H. Mistry, P. Simhambhatla, C. M. Aldaz, S. Gaddis, and M. C. MacLeod Response of human mammary epithelial cells to DNA damage induced by BPDE: involvement of novel regulatory pathways Carcinogenesis, February 1, 2003; 24(2): 225 - 234. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. A. Boorman, S. P. Anderson, W. M. Casey, R. H. Brown, L. M. Crosby, K. Gottschalk, M. Easton, Hong Ni, and K. T. Morgan Toxicogenomics, Drug Discovery, and the Pathologist Toxicol Pathol, January 1, 2002; 30(1): 15 - 27. [Abstract] [PDF]
|
|
|
|
|
|
Z. J. Chen, H. Shen, and K. D. Tew Gene expression profiling using a novel method: amplified differential gene expression (ADGE) Nucleic Acids Res., May 15, 2001; 29(10): e46 - e46. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. N. Hines, Z. Luo, T. Cresteil, X. Ding, R. A. Prough, J. L. Fitzpatrick, S. L. Ripp, K. C. Falkner, N.-L. Ge, A. Levine, et al. Molecular Regulation of Genes Encoding Xenobiotic-Metabolizing Enzymes: Mechanisms Involving Endogenous Factors Drug Metab. Dispos., April 13, 2001; 29(5): 623 - 633. [Abstract] [Full Text]
|
|
|
|
 |
|
 S E Wildsmith and F J Elcock Microarrays under the microscope Mol. Pathol., February 1, 2001; 54(1): 8 - 16. [Abstract] [Full Text]
|
|
|
|
 |
|
 G. B. Carey and D. W. Scott Role of Phosphatidylinositol 3-Kinase in Anti-IgM- and Anti-IgD-Induced Apoptosis in B Cell Lymphomas J. Immunol., February 1, 2001; 166(3): 1618 - 1626. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. Wang, R. Schneider-Broussard, A. P. Kumar, M. C. MacLeod, and D. G. Johnson Regulation of BRCA1 Expression by the Rb-E2F Pathway J. Biol. Chem., February 11, 2000; 275(6): 4532 - 4536. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||