The adenosine A2A receptor (A2AR) is highly expressed in the striatum, where it modulates motor and emotional behaviors. We used both microarray and bioinformatics analyses to compare gene expression profiles by genetic and pharmacological inactivation of A2AR and inferred an A2AR-controlled transcription network in the mouse striatum. A comparison between vehicle (VEH)-treated A2AR knockout (KO) mice (A2AR KO-VEH) and wild-type (WT) mice (WT-VEH) revealed 36 upregulated genes that were partially mimicked by treatment with SCH-58261 (SCH; an A2AR antagonist) and 54 downregulated genes that were not mimicked by SCH treatment. We validated the A2AR as a specific drug target for SCH by comparing A2AR KO-SCH and A2AR KO-VEH groups. The unique downregulation effect of A2AR KO was confirmed by comparing A2AR KO-SCH with WT-SCH gene groups. The distinct striatal gene expression profiles induced by A2AR KO and SCH should provide clues to the molecular mechanisms underlying the different phenotypes observed after genetic and pharmacological inactivation of A2AR. Furthermore, bioinformatics analysis discovered that Egr-2 binding sites were statistically overrepresented in the proximal promoters of A2AR KO-affected genes relative to the unaffected genes. This finding was further substantiated by the demonstration that the Egr-2 mRNA level increased in the striatum of both A2AR KO and SCH-treated mice and that striatal Egr-2 binding activity in the promoters of two A2AR KO-affected genes was enhanced in A2AR KO mice as assayed by chromatin immunoprecipitation. Taken together, these results strongly support the existence of an Egr-2-directed transcriptional regulatory network controlled by striatal A2ARs.
- A2A receptor antagonist
- A2A receptor knockout
- promotor analysis
adenosine is a naturally occurring nucleoside that is distributed ubiquitously throughout the body as a metabolic intermediary. Extracellular adenosine acts through multiple G protein-coupled receptors, namely, adenosine receptors (subtypes A1, A2A, A2B, and A3), to exert a variety of physiological effects (27). The adenosine A2A receptor (A2AR) is expressed at a high level in the striatum and at moderate levels in the thymus, leukocytes, and other tissue types (60, 82, 93). Activation of A2ARs affects a broad range of physiological and pathophysiological processes, including inflammation and immune responses, myocardial activity and altered blood pressure, cell death and tissue protection, wound healing, sleep-wake cycle, and locomotor and anxiety behavior (14, 26, 27, 47, 58, 82).
The regulation of motor and emotional behaviors by the A2AR in the striatum has been largely attributed to the modulation of dopaminergic activities (30, 55). The A2AR is coexpressed with the dopamine D2 receptor (D2R) in striatopallidal neurons, providing an anatomic basis for the cellular interaction between these two receptors (25, 70). Indeed, antagonistic interaction between A2ARs and D2Rs has been demonstrated at neurochemical (e.g., release of acetylcholine and GABA) and behavioral (e.g., locomotor activity) levels (21, 59). On the basis of the regulatory mechanisms of A2AR, A2AR antagonists have been proposed to target neuropsychological disorders with hypodopaminergic activity, such as Parkinson's disease, and A2AR agonists for disorders associated with hyperdopaminergic activity, such as schizophrenia and drug addiction (8, 20, 69, 72). Further therapeutic prospects for A2A adenosinergic drugs come from recent demonstration of the novel neuroprotective effect of A2AR inactivation in various animal models of brain injury (26, 47, 58, 72). Several laboratories, including ours, have shown that genetic inactivation or pharmacological blockade of A2ARs protects against brain cell injury induced by excitotoxicity (such as quinolinic acid and kainate) (42, 43, 65), ischemic attack (10, 17, 54, 61), neurotoxins [such as N-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), 6-hydroxydopamine, and 3-nitropropionic acid] (11, 24, 41), and β-amyloid aggregation (15). As a result, the A2AR has recently emerged as an attractive therapeutic target for Parkinson's disease, stroke, and Huntington's disease (5, 26, 72).
The short- and long-term neurochemical and behavioral effects of adenosinergic agents are likely to involve A2AR-mediated gene expression in the striatum. They cause altered striatal gene expression, which may underlie the neuromodulative and neuroprotective effects of A2AR inactivation. Pharmacological studies show that the activation of A2ARs increases c-Fos expression, whereas inactivation of A2ARs decreases the expression of immediate-early genes, including c-Fos, Zif268, and c-Jun/JunD (16, 62, 64, 82). The former is presumably associated with the positive coupling of A2AR to Golf protein in the striatum, leading to the stimulation of adenynyl cyclase and an increase in cAMP levels and DARPP-32 (dopamine- and cAMP-regulated phosphoprotein of 32 kDa) phosphorylation (81, 83, 84). Similarly, genetic inactivation of A2ARs reduces basal expression levels of several immediate-early genes, including Zif268 and members of the AP-1 complex c-Fos, c-Jun, and junD (16). These are transcription factors (TFs) presumably involved in regulating other striatal genes (such as neuropeptides, substance P, and enkephalin) that respond to A2AR modulation (13, 40, 73, 82). Thus understanding the transcriptional regulatory network modulated by A2AR is an important step toward deciphering the molecular actions of A2A adenosinergic agents.
DNA microarrays have been recently introduced to monitor gene expression patterns in various brain regions under normal and pathophysiological conditions such as aging, Parkinson's disease, schizophrenia, and tissue repair (1, 32, 51, 53). Inspired by the availability of genomic sequences and the ability to identify large cohorts of coregulated genes by DNA microarray, bioinformatics researchers have developed a battery of computational tools to facilitate genomic sequence analysis. Various algorithms have been developed to predict promoter regions (22, 80, 88) and have been successfully applied to the search for cis elements in the entire genomes of yeast, worm, and fly (2, 39, 46, 67, 68, 87) and, in a few cases, mouse (49). We have recently developed a suite of statistical algorithms to detect cis element overrepresentation in a set of coregulated genes (29, 37).
In this study, we compared genome-wide mRNA expression changes in the striatum resulting from genetic and pharmacological inactivation of A2ARs to understand the possibly distinct molecular actions of these two treatments. We also adapted a drug target validation approach to confirm the specificity of SCH-58261 (SCH) as an A2AR antagonist by treating A2AR knockout (KO) mice with SCH. Furthermore, we computationally analyzed cis element overrepresentation in core promoters of A2AR KO-affected genes and validated DNA binding activities for two predicted cis elements by chromatin immunoprecipitation (ChIP). By integrating the results from microarray and promoter sequence analyses, we identified a transcriptional regulatory network that responded to genetic inactivation of A2ARs.
Mice were handled in accordance with the guides and protocols approved by the Institutional Animal Care and Use Committee at Boston University School of Medicine.
Generation of A2AR KO mice and A2AR antagonist treatment.
A2AR KO mice in a near congenic (N6) C57BL/6 genetic background were generated as previously described (10, 11). Heterozygous cross-breeding was used to generate A2AR KO and wild-type (WT) littermate mice. For each experiment, A2AR KO mice and WT littermates were matched for gender and age. A2AR KO and WT mice (male, ∼6–8 wk of age) were administered SCH (intraperitoneally; 5 mg/kg a generous gift of the Schering-Plough Research Institute) or vehicle [VEH; intraperitoneally; 15% DMSO, 15% ethoxylated castor oil (Alkemuls), and 70% water] daily for 8 days. Mice were killed 120 min after the last injection. The mouse striatum was dissected out and immediately frozen with dry ice for RNA isolation.
Affymetrix GeneChip processing.
Total RNA was isolated using TRIzol reagent according to the manufacturer's protocol (Invitrogen; Carlsbad, CA). RNA concentration and integrity were assessed by spectrophotometry and gel electrophoresis. The labeling of RNA samples, mouse GeneChip (Mu11ksubB) hybridization, and array scanning were performed according to the Affymetrix GeneChip Expression Analysis Manual (Affymetrix) at the Harvard Genome Research Center. An average yield of 40 μg of biotin-labeled target cRNA was obtained from 5–7 μg of total RNA from each sample, and 20 μg of cRNA were applied to each chip. The chips were hybridized overnight in a rotating oven at 45°C, washed and stained on a fluidics station, and then scanned at a resolution of 3 μm in a confocal scanner (Agilent Affymetrix GeneArray Scanner). RNA from each mouse was processed individually to allow independent DNA chip analysis. One RNA sample from the WT-SCH group did not produce high quality RNA and consequently was not processed for microarray hybridization. A total of 15 DNA chips was used in this study (see Fig. 1, inset, for details).
Quantitative PCR analysis.
In a separate experiment, A2AR KO and WT mice were treated with either VEH or 5 mg/kg SCH (n = 5 for each group) for 8 days. At the end of the treatment, striata were isolated, and total RNA was extracted, as described above. We then reverse-transcribed cDNA from total RNA using an Omniscript RT Kit (Qiagen; Valencia, CA) and an oligo(dT) primer (Invitrogen). We carried out quantitative PCR (qPCR) for all genes deemed significantly affected by two different comparisons (all colored dots/genes in Fig. 2 and Table 1), using a SYBR green kit (Applied Biosystems; Warrington, UK). PCRs were performed in an ABI PRISM 7900HT Sequence Detection System (PE Applied Biosystems). Reaction conditions were 50°C for 2 min and 95°C for 10 min, followed by 45 cycles of the amplification step (95°C for 15 s, 60°C for 30 s, and 72°C for 45 s). An endogenous control mouse cDNA, pgk1, was included in each amplification using a TaqMan PCR Core Reagent kit (Perkin-Elmer; Roch, NJ). The reaction consisted of 0.3 μl cDNA, 2.5 μl of 2× Taqman, 2× Master Mix, and 100 nM TaqMan Probe in a final volume of 5 μl. The specificity of qPCR products was verified by melting curve analysis and by visualizing PCR products in 1% agarose gel at the expected molecular sizes. The relative abundances of target genes were obtained by application of a standard curve (generated by serial dilution of a reference cDNA of the normal mouse) and normalizing to pgk1. The qPCR primer sequences for 23 genes tested are provided as Supplemental Table S1 (the Supplemental Material for this article is available at the Physiological Genomics web site).1
Processing and statistical analysis of microarray data.
The Affymetrix Microarray Suite software package (version 4.0) was used to calculate the overall noise of each image (Qraw) to ensure that background noise was similar across arrays in all groups. Normalized log2 abundance values were obtained using the robust multiarray analysis (RMA) method as implemented in the affy (31) package (version 1.3) for R (version 1.8.1, http://www.r-project.org). P values for the significance of changes in mRNA abundance between control and experimental conditions were calculated using a permutation test. This test computes the significance of a simplified t-test statistic proposed by Golub and colleagues using a null distribution of the statistic consisting of all possible permutations of the experimental condition labels (19, 34). This test requires fewer statistical assumptions than a standard t-test but provides a narrow range of P values with the small samples sizes available for our study. Multiple testing could not be performed given this characteristic of the test. Genes showing an absolute fold change of ≥2.0 and a permutation P value ≤0.05 were considered as significantly altered in expression and selected for promoter analysis. This set of genes with significant expression changes is referred to as the positive set compared with the negative set of genes described below. The Affymetrix annotation database (http://www.affymetrix.com) and BLAST searches were performed to identify the gene associated with each probe set. The RMA-processed data for all 15 samples (including a sample description, a data list, and four Excel files for the four comparisons described in Fig. 1 and methods) are provided as Supplemental File S1.
Sequence retrieval by PromoSer and transcription factor binding site analysis by ROVER.
Promoters for the genes represented on the microarray chip were downloaded from PromoSer (36) (http://biowulf.bu.edu/zlab/PromoSer/) on 9/22/03 using the following parameters: quality 1, support 1, most aggressive 5′ extension, and return promoters even when they overlap genes and assembly gaps. All promoters were specified as the sequence region from −1,000 to +100 (transcription start site = +1). The sequences were obtained from PromoSer with known repeat elements masked. PromoSer was able to extract unique promoter sequences for 3,647 of 6,586 probe sets on the chip, including 31 of the 45 genes affected by A2AR KO in the microarray analysis for network inference. An equal sized set of promoter sequences was selected for the genes showing the least-significant changes in expression for comparison. These control genes were limited to those genes among the top 60% in log2 abundance values as a rough requirement for their actually being expressed. They are referred to as the negative set. The promoter sequences for the positive and negative gene sets are provided as Supplemental File S2.
TF binding site searches were conducted using 466 vertebrate position-specific scoring matrixes (PSSMs) from TRANSFAC (version 7.2 professional) (91). Multiple PSSMs are annotated in TRANSFAC as belonging to a particular factor, and we assigned these PSSMs as a redundant group. When PSSMs belonged to multiple factors, we merged PSSMs from these factors into the same group. We filtered the PSSM significance results from ROVER to include only the highest scoring member of each redundant group, resulting in a set of 231 PSSMs.
The ROVER promoter analysis tool (37) was used to screen the two sets of promoters with the TRANSFAC library and compute the statistical significance of any overabundance of high-scoring PSSM matches. ROVER was run using default P value cutoffs of 0.01 for the significance of individual sequences and 0.001 for single PSSM matches. Sequence regions corresponding to masked known repeats were ignored in the ROVER calculation.
Network inference analysis using CARRIE.
Inference of a transcriptional network was accomplished using the software package CARRIE as described previously (37). CARRIE is a computational method that analyzes microarray and promoter sequence data to infer a transcriptional regulatory network from the response to a specific stimulus. CARRIE combines two complementary approaches for detecting transcriptional regulation. First, microarray data were used to reveal genes that respond to a given stimulus through changes in mRNA abundance. These genes are believed to form a coregulated group. If there are TFs in the group, we propose that they regulate the observed expression changes of the other genes. Second, we identified TFs with binding sites that are statistically overrepresented in promoter regions of the coregulated group of genes using ROVER. Even if their expression levels do not change upon stimulation, these TFs are also predicted to regulate the group of genes. A network diagram was generated using the specific TF/promoter interactions predicted by ROVER. The stimulatory or inhibitory impact of each TF on its regulated genes was inferred from the directions of change in their expression levels.
ChIP of genomic DNA associated with Egr-2 protein was carried out according to the manufacturer's protocol with minor modifications (Upstate Biotechnology). Briefly, mouse striatum was cross-linked in 1% formaldehyde (10 μl/mg) for 15 min, washed twice in PBS, and lysed with a pellet pestle. The resulting extracts were sonicated to fragment chromatin and centrifuged for 15 min at 14,000 g. Chromatin was immunoprecipitated with an antibody against Egr-2 (1:40, Santa Cruz Biotechnology). Immunoprecipitated (enriched) and whole cell genomic DNAs (unenriched, “input”) were incubated in 200 mM NaCl for 4 h at 65°C to reverse cross-links. DNA was purified by proteinase K treatment, phenol-chloroform extraction, and ethanol precipitation.
PCR amplification (50 μl final volume) of the chromatin-immunoprecipitated genomic DNA (∼100 ng) was performed with specific primer pairs targeting for the promoters of Egr-2 (forward 5′-CAGCTAGCCTCTATTAGCAC-3′ and reverse 5′-GAT TGCCGCTACCAACCTTC-3′); protein phosphatase 2A (PP2A; forward 5′-GGCCAGAGATATGTGATCT GCATCA-3′ and reverse 5′-GCTATCAAGGTTGTTAAAGCGTGGC-3′), Sui 1 (forward 5′-CTGACAC AGTAGTTATCAGC-3′ and reverse 5′-GAAGCAGAGACAGGCGGATT-3′), and Rap1 (forward 5′-CTGAAATCTGGCCAAGTCCTGCAAG-3′ and reverse 5′-AGCAGACCTTCACTATCATC-3′). Egr-2 binding sites for these four promoters were selected based on their high P value for overrepresentation in promoter analysis. A second set of primers was designed against the coding sequence and served as negative control sets. A PCR thermal profile was performed for 40 cycles as follows: holding at 95°C for 3 min, denaturing at 95°C for 30 s, extending at 60°C for 1 min, and annealing at 72°C for 1 min. PCR products were detected by electrophoresis in 1% agarose gel.
Microarray and bioinformatics analyses identified similarities and differences in striatal gene expression profiles for A2AR KO mice and SCH-treated WT mice.
Figure 1 is a flow chart of the microarray and bioinformatics analyses of striatal gene expression by A2AR inactivation. The process includes scaling, normalization, and statistical (permutation test) analysis of the microarray data, the retrieval and analysis of the promoter regions of A2AR KO-affected genes, and transcription network inference. The inset table in Fig. 1 lists the number of animals/microarrays for each treatment group.
A statistical cutoff (P value ≤0.05 and fold change ≥1.8) was used to select genes with significant expression changes. These cutoffs were selected due to the relatively small number and magnitude of expression changes observed. This statistical analysis yielded 90 genes for the A2AR KO-VEH vs. WT-VEH comparison, 31 genes for WT-SCH vs. WT-VEH, 30 genes for A2AR KO-SCH vs. WT-SCH, and 4 genes for A2AR KO-SCH vs. A2AR KO-VEH (see Table 2 for the complete list of these genes).
qPCR confirmed altered gene expression detected by microarray.
To validate the microarray results, we used qPCR to measure expression levels of all genes deemed significantly affected by two different comparisons (all colored dots/genes in Fig. 2 and Table 1). These include 11 genes that were upregulated by both A2AR KO and SCH (i.e., significantly affected genes revealed by the comparison between KO-VEH vs. WT-VEH and WT-SCH vs. WT-VEH), 9 genes that were downregulated by A2AR KO either in the KO-VEH vs. WT-VEH or the KO-SCH vs. WT-SCH comparisons, and 3 upregulated genes that were significantly affected by the KO-SCH vs. WT-SCH and KO-SCH vs. KO-VEH comparisons (see Table 1). Data from microarray and qPCR experiments are presented as fold change in expression of genes in the experimental group relative to the control group in four types of comparisons (A–D) as illustrated in Table 1 and Fig. 2. In total, qPCR analysis confirmed microarray results in 36 of 46 (or 78%) assays (Table 1). Specifically, qPCR validated the microarray results in 15 of 19 genes in comparison A (A2AR KO-VEH vs. WT-VEH), 9 of 11 genes in comparison B (WT-SCH vs. WT-VEH), 8 of 12 genes in comparison C (A2AR KO-SCH vs. WT-SCH), and 3 of 3 genes in comparison D (A2AR KO-SCH vs. A2AR KO-VEH) (Table 1).
A2AR KO produced both downregulation and upregulation of striatal gene expression, with downregulation predominating.
By comparing the microarray results of A2AR KO-VEH and WT-VEH mice, we identified 90 genes with significant expression changes, among which 54 genes were downregulated and 36 genes were upregulated (Table 2, comparison A, and Fig. 2A). Similarly, comparison between microarray results for A2AR KO-SCH and WT-SCH revealed 25 downregulated genes and 5 upregulated genes (Table 2, comparison B). Notably, 9 of these 25 downregulated genes were also detected in the A2AR KO-VEH vs. WT-VEHe comparison (green dots in Fig. 2, A and B). Thus both comparisons demonstrated that genetic depletion of A2AR produced both upregulation and downregulation of gene expression in the striatum, with downregulation dominating.
Many of the genes impacted by A2AR KO are associated with energy metabolism and protein synthesis. For example, six various types of ATPase were downregulated by A2AR KO. This finding, together with downregulation of eukaryotic translation elongation factor 2, might be related to predominantly downregulation of gene expression in A2AR KO mice, possibly as result of inactivation of A2AR during early development (as discussed below). In addition, A2AR KO also affected some genes that have been implicated in the pathogenesis of neurological disorders. For example, β-amyloid precursor protein (APP) is the protein product responsible for β-amyloid accumulation in Alzheimer's disease (86), and PP2A activity has been postulated to be associated with microtubule-associated protein-τ, which is a major feature of neurofibrillary tangles in Alzheimer's disease (90). Both genes (APP and PP2A) were found downregulated by A2AR KO. Interestingly, PARK7 (DJ-1), an autosomal recessive early-onset Parkinson's disease gene (6), has been found to be upregulated by SCH treatment.
The A2AR antagonist SCH produced almost exclusively upregulation of striatal gene expression partially overlapping with the A2AR KO effect.
With the same statistical and fold change threshold, microarray analysis yielded 31 genes that were significantly affected by the daily treatment of WT mice with SCH compared with VEH treatment (Table 2, comparison C, and Fig. 2C). Almost all of these 31 genes (29 of 31) were upregulated with SCH treatment. Most interestingly, 11 of the 31 genes were also upregulated by A2AR KO (A2AR KO-VEH vs. WT-VEH; red dots in Fig. 2, A and C). This significant overlap between the genes affected by A2AR KO and SCH (∼30% of SCH-affected genes) is supportive of the same molecular target being affected by genetic and pharmacological approaches. Thus the SCH treatment partially duplicated one aspect of the A2AR KO effect on striatal gene expression, i.e., the upregulation of a set of genes, but did not replicate the downregulation of a larger set of genes. Of note, some differences observed between A2AR KO and SCH may be due to the effect of the drug VEH (including 15% DMSO and 15% castor oil).
Finally, we validated that the A2AR was the drug target for SCH by administering SCH to A2AR KO mice. As expected, the comparison between A2AR KO-SCH and A2AR KO-VEH produced only four genes with significant expression changes (Fig. 2D). These genes were not among the genes showing a significant expression change in the WT-SCH vs. WT-VEH comparison. This result confirms that SCH largely acts on the A2AR to elicit the gene expression response.
Two TFs have significantly overabundant binding sites in the promoters of A2AR KO-affected genes.
To uncover a potential transcriptional regulatory network responsive to A2AR stimulation in the striatum, we employed several bioinformatics algorithms developed in our laboratory to detect TF binding sites in core promoter regions of A2AR KO-affected genes. Two groups of genes were used for promoter analysis and transcriptional network inference. The first group, the positive set, consisted of genes that showed significant changes in mRNA abundance in A2AR KO mice vs. WT mice. We used a permutation test with a P value ≤0.05 and fold change ≥2.0 (a stricter cutoff than the cutoff of 1.8 in the above analysis; Table 2) to ensure the inclusion of only the most responsive genes. Forty-five microarray probe sets passed the threshold for significant expression change. We used the PromoSer tool developed in our laboratory (36) to determine transcription start sites and corresponding core promoter sequences for these genes. PromoSer was able to determine the promoters for 31 of these genes.
The second set of genes, the negative set, consisted of an equal number of promoters for genes that were expressed at similar levels in both A2AR KO and WT mice. The negative set was used as a set of control sequences that should not contain significant numbers of the TF binding sites believed to be enriched in the positive set. As heterochromatin organization may render the promoters of some genes inaccessible, we required all genes in the negative set to be expressed in the striatum.
Two methods were used to identify TFs regulating the response to genetic depletion of A2AR. The first method identified TF-encoding genes among the A2AR KO-affected genes from the microarray data analysis. We thus identified two TFs, Egr-2 and c-Fos. The second method made use of our computer algorithm ROVER to identify TFs whose binding sites were significantly overabundant in promoters of the positive set of genes compared with those of the negative set (37). The second method supplemented the first by including TFs that may be activated posttranscriptionally. TFs were represented with their PSSMs as annotated in TRANSFAC (91). Figure 3 shows a plot of overabundance P values calculated for each PSSM. The PSSM list has been filtered for redundancy, as described in methods, and includes a total of 231 nonredundant PSSMs. ROVER identified two PSSMs with dramatically more significant overabundance P values than the remaining PSSMs (Fig. 3). The most significant PSSM represented the Egr family of TFs, reinforcing the selection of Egr-2 based on its significant change in mRNA abundance. The PSSM of AP-2 was the second most significant. These two PSSMs had P values of 6.2 × 10−22 and 9.9 × 10−21, respectively, whereas the next most significant PSSM, electron-transfer flavoprotein (ETF), had a P value of 5.9 × 10−11. The TF identified only by the expression change of its gene, c-Fos (AP-1), is represented by a group of 19 PSSMs (see methods). The highest scoring of these had a P value of 1.3 × 10−3. In light of the false positive rate mentioned above, it is important to note that these results were robust to different levels of stringency in selecting genes for analysis and to the presence or absence of redundant promoters in the analysis due to the presence of multiple microarray probe sets representing the same gene. Gene lists generated using microarray P value cutoffs from 0.1 to 0.01 consistently produced ROVER results showing Egr-2 and AP-2 as the most significant TFs. However, those predicted TF binding sites should be treated with extreme caution because <50% of the predicted binding sites for Egr-2 (see Fig. 5) and AP-2 binding (data not shown) could be confirmed by our ChIP analysis (also see the discussion below).
Other high-scoring PSSMs in Fig. 3 may represent biologically important TFs as well. Zinc finger protein of the cerebellum 2 (Zic2) is known to regulate D1R (92). Zic1 has been implicated in the regulation of a gene selected in the microarray analysis–or downregulated in the Zic1-deficient cerebellum (Dorz1). Pax-3 is involved in neurogenesis (7). Nerve growth factor-induced protein C is a member of the EGR family (Egr-4) according to the National Center for Biotechnology Information (NCBI) database, although it is not annotated as such by TRANSFAC. The involvements of these TFs merit further investigation.
CARRIE analysis revealed an Egr-2/AP-2/c-Fos-mediated transcriptional regulatory network responsive to A2AR inactivation in the striatum.
The computational method CARRIE was used to infer the transcriptional regulatory network regulated by A2ARs. The details of CARRIE have been described previously (37). Three TFs were selected based on their significant changes in expression (Egr-2 and c-Fos) and/or the overabundance of their binding sites in promoters of the positive set of genes as described above (Egr-2 and AP-2). The significant binding sites for these TFs, as detected by ROVER using the relevant PSSMs, were used to draw direct regulatory connections between them and genes showing significant expression changes in microarray data (Fig. 4). The inhibitory/stimulatory relationship between each TF and the genes it regulates were inferred by comparing the directions of gene expression changes observed for the regulated gene and that of the TF, according to microarray data. For example, in A2AR KO mice, mRNA abundance for Egr-2 increases, whereas mRNA abundance for PP2A decreases, and Egr-2 is shown to bind the promoter of PP2A. From this, we inferred that Egr-2 inhibits the transcription of PP2A. Indirect regulatory connections were drawn between A2AR and the three TFs shown in Fig. 4.
The network diagram resulting from this analysis (Fig. 4) shows examples of potential competitive regulation, combinatorial regulation, and feedback control. AP-2 is shown to regulate Egr-2, and these two TFs are inferred to regulate some of the same genes. The directions of changes in expression seen for Egr-2 and coregulated genes implies that Egr-2 and AP-2 have opposite effects on expression of these genes and that AP-2 further accentuates this difference by downregulating the expression level of Egr-2 itself. Additionally, some genes in the network appear to be regulated by c-Fos and Egr-2 in a coordinated fashion. Finally, the Egr-2 promoter has significant matches to the Egr PSSM, suggesting that Egr-2 controls its own expression through a feedback mechanism. This self-regulation of Egr-2 was confirmed by ChIP analysis.
Egr-2 binding in promoters of Egr-2 and PP2A was enhanced in A2AR KO mice as assayed by ChIP.
To provide experimental evidence for the inferred transcription network control by striatal A2ARs, we evaluated Egr-2 DNA binding activities for promoters of four A2AR-affected genes with top P values for their overrepresentation in promoter analysis (from Fig. 4). After ChIP with Egr-2 antibody, PCR analysis detected Egr-2 DNA binding activity for promoters of Egr-2 (214 bp; Fig. 5, top) and PP2A (265 bp; Fig. 5, bottom) but failed to detect any PCR products for Sui 1 and Rap 1 in the striatum of WT mice (data not shown). Importantly, Egr-2 binding activities in promoters for both Egr-2 and PP2A genes were significantly enhanced in A2AR KO mice compared with those of WT littermates (Fig. 5, top and bottom). This finding is consistent with the inferred transcriptional network controlled by A2ARs. As a negative control for the ChIP assay, we also PCR amplified the coding sequence of these two genes. As expected, we did not detect any PCR products for these coding sequences in chromatin-immunoprecipitated genomic DNA, although we detected PCR products at the expected molecular sizes in “input” genomic DNA (Fig. 5).
Multiple cross-comparisons between A2AR KO and SCH-58261 treatment and independent qPCR analysis are critical for validating the gene expression profiles identified by microarray analysis.
The quality of this study rests heavily on the validation of the consistency of the gene expression pattern using independent methods. We performed 46 qPCR analysis for all 23 genes that were jointed affected by at least 2 comparisons (see Table 1 and Fig. 2, all colored dots). For the A2AR KO-VEH vs. WT-VEH comparison, 7 of 11 genes showed similar changes in expression by qPCR and microarray analysis. For the comparison between WT-SCH and WT-VEH, 9 of 11 genes were consistent. Notably, this qPCR confirmation was done with an independent set of animal/tissue samples. This consistency rate (78%) is similar to previously published microarray studies with qPCR validation (4, 12, 18) and substantiates our confidence in the microarray analysis.
In addition, we designed our experiments with several comparisons built in to cross-confirm the main biological effect of A2AR inactivation on striatal gene expression. First, two sets of comparisons (i.e., A2AR KO-VEH vs. WT-VEH and A2AR KO-SCH vs. WT-SCH) were designed to cross-confirm the effect of A2AR KO. The former comparison produced all of the genes that were affected by A2AR KO, including genes mimicked by SCH and genes not mimicked by SCH. The latter comparison further dissected out the effects unique to A2AR KO. Second, we performed two comparisons to address the specific effect of SCH on striatal gene expression. The comparison between WT-SCH and WT-VEH yielded 30 genes that were specifically affected by SCH. To validate that A2AR was the molecular target of SCH treatment, we adapted a drug target validation strategy (52) to compare A2AR KO-SCH with A2AR KO-VEH. If the molecular target for SCH is indeed the A2AR, which is completely depleted in A2AR KO mice, then SCH is not expected to produce any additional biological effect on gene expression. Indeed, in contrast to the 30 genes identified by the WT-SCH vs. WT-VEH comparison, this comparison yielded only 4 genes, indicating that only a small portion (∼13%) of genes with altered expression were due to the impact of SCH on non-A2AR targets. Third, we compared the effects of genetic depletion (A2AR KO) and pharmacological inactivation (SCH treatment) of A2AR in parallel so that some common as well as distinct features of the A2AR-mediated gene expression pattern could be deduced and cross-validated. Interestingly, comparing the gene list affected by A2AR KO (A2AR KO-VEH vs. WT-VEH) and by SCH (WT-SCH vs. WT-VEH) showed that about 1/3 of the genes in the SCH group overlap with the upregulated genes by A2AR KO, indicating that SCH partially mimics upregulation of gene expression by A2AR KO. This cross-comparison between genetic and pharmacological approaches supports the validity of our microarray analysis of gene expression in the striatum. Despite these strengths of cross-validation and qPCR confirmation, the genes listed in Table 1 should be treated with caution until additional validation, because at least 22% of these genes are false positive, as estimated by qPCR analysis.
Distinct gene expression profiles in A2AR KO and SCH-58261-treated striatum may indicate different phenotypes from KO and pharmacological approaches.
Microarray analysis of striatal gene expression demonstrated that genetic (A2AR KO) and pharmacological (SCH treatment) approaches produce distinct gene expression profiles in the striatum. Genetic depletion of A2ARs produced two groups of striatal genes with altered expression: 54 downregulated genes and 36 upregulated genes. In contrast, SCH treatment produced almost all upregulated genes (29 of 31 affected genes). Most importantly, 11 genes were upregulated by both A2AR KO and SCH treatment, supporting the hypothesis that this aspect of the two gene expression profiles are mediated by the same molecule target, the A2AR. Thus SCH treatment partially reproduced one aspect of the A2AR KO effect, i.e., upregulation of one group of striatal genes, but did not mimic the other aspect of the A2AR KO effect, i.e., downregulation of another group of striatal genes. The large number of upregulated genes that are regulated by both pharmacological and genetic inactivation of A2ARs (11 common genes of 29 SCH-upregulated genes and 36 A2AR KO-upregulated genes) strengthened our confidence in these findings.
The major effect of downregulation of gene expression by A2AR KO was not mimicked by SCH, which indicates a unique and possibly developmental effect of A2AR KO. This unique downregulation effect of A2AR KO was further supported by the comparison between A2AR KO-SCH and WT-SCH, which removed the effect of SCH and produce only unique A2AR KO effects. Indeed, this comparison yielded largely downregulation of gene expression with nine of these downregulated genes overlapping with the downregulated genes deduced from the A2AR KO-VEH vs. WT-VEH comparison.
The predominant downregulation of gene expression by A2AR KO is consistent with an early study showing that A2AR KO resulted mainly in downregulation of a large number of immediately-early genes and some neuropeptide genes (16, 44). A2AR is coupled to Golf proteins in the striatum to activate adenylyl cyclase, leading to increased PKA activity and DARPP-32 phosphorylation (82, 83). The alteration in the PKA signaling pathway may contribute in part to the gene expression pattern by A2AR KO or SCH. Thus genetic inactivation of A2AR may result in the downregulation of gene expression in the striatum. However, in contrast to the almost-complete upregulation of gene expression by SCH in our microarray analysis, previous pharmacological activation or inactivation of A2AR have shown both increased and decreased striatal gene expression (21, 56, 63). The different treatment paradigms between our study (repeated administration for 8 days) and previous pharmacological studies (largely a single injection) may account for the difference in these results.
Several possible explanations may account for the different expression profiles generated by genetic and pharmacological approaches. First, the difference in gene expression profiles may be in part due to the partial specificity of SCH in contrast to the complete and specific inhibition by A2AR KO. SCH has ∼100-fold selectivity for A2AR over A1R (59). Nonetheless, it is possible that SCH acts on other molecular targets (such as A1Rs) in addition to A2ARs to produce a gene expression profile that differs from that in A2AR KO mice. However, by adapting a novel approach of drug target validation with a combined drug and genetic KO model (52), our microarray analysis showed that, in contrast to the 30 SCH-affected genes in WT mice (WT-SCH vs. WT-VEH), administering SCH to A2AR KO mice yielded only 4 affected genes (A2AR KO-SCH vs. A2AR KO-VEH). This strongly suggests that the A2AR is the main molecular target for SCH.
A second possible explanation is that, unlike the SCH treatment in WT mice (which produces partial blockade of A2AR for 8 days in an adult animal), A2AR KO mice are completely deficient in A2AR function throughout development and adult life. We therefore speculate that the unique downregulated genes in A2AR KO mice (i.e., the effect that cannot be mimicked by SCH) may be related to the developmental effects of global A2AR deletion. Although our initial characterization of A2AR KO mice did not reveal gross abnormalities in several neurochemical markers in the cortex and striatum or in behavior (9, 10), some subtle changes at the level of gene expression may exist throughout life, leading to altered neuroadaptation. Indeed, neurochemical studies suggest that endogenous adenosine acting at adenosine receptors plays an important role in brain development and Schwann cell differentiation (78, 79). The facilitative role of adenosine receptors during early development is also consistent with the finding of a predominant downregulation of striatal gene expression in A2AR KO mice (16). This notion of a unique, early development effect by the genetic KO approach has been suggested to account for the discrepancy between genetic KO and pharmacological blockade approaches (33). Thus some effects may not be due to the absence of the receptor in the adult mouse but to the lack of the receptor at some earlier point in development. This has been shown for serotonin (5-HT) 1A receptors: a developmentally controlled rescue strategy showed that postnatal developmental expression of 5-HT1A receptors is important to establish anxiety-like behavior in adult mice (35).
As with other genetic KO models, important discrepancies were noted between pharmacological and genetic inactivation of A2ARs (8, 9). Regardless of the underlying causes, the distinct gene expression profiles revealed by our microarray analysis provides a plausible molecular correlate of the different and sometimes opposing phenotypes seen in genetic versus pharmacological blockade of neurotransmitter receptors. For example, A2AR antagonists produced motor stimulant effects in normal and dopamine-depleted rodent and nonhuman primates, whereas genetic deletion of A2ARs resulted in a motor depressant effect (8, 9). On the basis of our analysis, we hypothesize that the A2AR exerts a facilitative influence on gene expression and brain maturation during development. Further studies are needed to test this hypothesis.
Promoter and microarray analyses reveal an Egr-2-mediated transcriptional network controlled by A2AR in the striatum.
How to deduce regulatory relationships among a large, diverse set of coregulated genes in a microarray represents one of the major challenges in understanding gene regulatory programs responsible for tissue specificity and coordinated responses to signaling in the brain. Early microarray analyses were limited to clustering coregulated genes (77). Additional methods have been developed to make use of DNA-TF binding data to reveal a potential transcription network (68, 76). However, this TF-based approach has been largely limited to simple eukaryotic systems such as the yeast Saccharomyces cerevisiae (2, 39, 46, 68), C. elegans (50), and Drosophila (67). Application of this approach in higher eukaryotic species (such as rodents and humans) so far has met with very limited success (38, 49, 89). We have developed and successfully applied the computational algorithms ROVER and CARRIE to decipher a transcriptional regulatory network under six experimental conditions in S. cerevisiae (37), and we now extended our approach to these mouse microarray experiments to gain new insights into A2AR neurobiology.
One feature of the ROVER-CARRIE algorithm is the combination of the gene expression data with the promoter sequence information for A2AR KO-affected genes. The validity of identifying TFs by their own changes in gene expression is supported by several previous studies in network inference (3, 74, 94). Two TFs (Egr-2 and c-Fos) identified by microarray analysis (i.e., their expression was deemed to be significantly affected by A2AR KO) were proposed to participate in transcriptional regulation. The key TF (Egr-2) was found to be upregulated, a finding that was confirmed by qPCR. However, previous pharmacological studies showed downregulation of Egr-1, a close Egr family member, in the striatum of A2AR KO mice (16). Because the structure of the Egr-2 protein is very similar to that of Egr-1 and they bind to similar sites (48), we reasoned that Egr-2 expression may represent a compensatory response to decreased expression of Egr-1. Consistent with this notion, our qPCR showed that Egr-1 expression was indeed decreased in A2AR KO mice (data not shown), in parallel with increased Egr-2 expression by both A2AR KO and SCH treatment.
The TFs identified through their own expression changes are further supplemented by the TFs identified through promoter analysis with ROVER, which detected TF binding sites statistically overrepresented in promoters of the positive set of genes (compared with the negative set of genes with unchanged expression). The promoter approach complements microarray data by including TFs that are activated posttranscriptionally but do not show altered levels of expression. To detect overrepresented TF binding sites, we used a novel approach to define the cutoff score, determined by the negative set of equal size, rather than random genomic sequences or other promoter sequences. The negative set is further limited to the top 60% of intensity values, thus ensuring that it is comprised of striatally expressed genes and is largely independent of chromatin structure issues. Thus the comparison between the positive and negative sets was designed specifically to detect a transcriptional regulatory mechanism influenced by genetic inactivation of A2ARs rather than by tissue-specificity issues or other stimuli. ROVER analysis identified Egr-2 as the TF with the most significantly overrepresented binding sites in the promoters of A2AR KO-affected genes. The convergence of both microarray and promoter analyses on the same TF, Egr-2, enhances confidence that these analyses are valid and that Egr-2 has a role in the A2AR-mediated transcription network. The TF with the second most overrepresented binding sites, AP-2, has a PSSM that is similar to that of Egr-2. This may account for its overrepresentation, and thus the biological function of AP-2 on the A2AR-mediated transcription network requires further investigation.
The inferred Egr-2-mediated transcription regulatory network received experimental support from the assessment of Egr-2 binding activity in promoters of Egr-2 and PP2A genes by ChIP. The ChIP assay demonstrated that the binding of Egr-2 to promoters of Egr-2 and PP2A was enhanced in A2AR KO compared with WT littermates, a conclusion consistent with the predicted relationship between Egr-2 and its target genes and with the increased levels of Egr-2 in the striatum by microarray analysis. The lack of Egr-2 binding to Sui 1 and Rap 1 promoters is probably an indication that these sites are not active in vivo, as is true for many sites identified by in vitro experiments (89).
This network analysis reveals several important regulatory features of A2AR-mediated gene expression: positive and negative regulation of other genes and autoregulation. The network diagram resulting from this analysis (Fig. 4) shows examples of potential competitive regulation, combinatorial regulation, and feedback control. These inferred regulatory features have been shown to be common in transcriptional networks (46). For example, AP-2 and Egr-2 are inferred to regulate some of the same genes. The directions of the changes in expression seen for Egr-2 and coregulated genes imply that Egr-2 and AP-2 may have opposite effects on expression that clearly warrant further experimental validation. Additionally, some genes in the network may be coregulated by c-Fos and Egr-2 in a coordinated fashion. The Egr-2 promoter has significant matches to the EGR PSSM, suggesting that Egr-2 acts to control its own expression through a feedback mechanism. It is yet to be determined whether this mechanism would have a positive or negative effect on Egr-2 expression.
Identification of a large set of A2AR KO-affected genes and the inferred transcription regulatory network provide a road map for the global understanding of striatal A2AR function.
This study identified a group of genes that were commonly affected by both genetic inactivation and pharmacological blockade of A2ARs. This subgroup represents potentially important downstream target genes of the A2AR. The validity of this finding is enhanced by similar results using qPCR to analyze expression of these genes. Notably, three genes involved in protein folding were included in this group, indicating a possible effect of A2ARs on protein folding in vivo. In addition, the discovery of several neurological disease-related genes affected by A2AR KO, including early-onset Parkinson's disease genes DJ-1 and APP, raises the possibility that A2AR may influence the pathogenesis of several neurological disorders. For example, APP was reduced in A2AR KO mice by 3.6-fold. Concurrently, the expression of cyclooxygenase (COX)-2 in A2AR KO mice and in SCH-treated WT mice was reduced by 1.5- and 2.8-fold, respectively. The finding of downregulation of COX-2 and APP is in agreement with previous studies showing that A2AR antagonism inhibits COX-2 (23) and that COX-2 inhibitors (such as nonsteroidal anti-inflammatory agents) can partially downregulate the expression of APP (45, 66). These results identify a potential link for A2AR involvement in the pathological progress in Alzheimer's disease through the COX-2/APP inflammatory regulatory pathway, which clearly merits further investigation.
Identification of Egr-2 as a proposed key TF in an A2AR KO-induced transcription network suggests that A2AR may be implicated in functions related to the Egr-2 KO phenotype. The Egr TF family was originally identified as an immediately-early gene in response to nerve growth factor stimulation (75) and has been shown to play a critical role in neuronal plasticity in response to a wide range of neuronal stimuli (57). It is known that Egr-2 KO mice displayed defects in hindbrain development as well as peripheral nerve myelination (71, 85). This finding prompted a screen of patients with congenital abnormalities of peripheral nerve myelination and led to the discovery of a mutation in Egr-2 (71, 85). The proposed center role of Egr-2 in the A2AR-mediated transcriptional regulatory network proposed in this study would suggest that A2AR may also affect Schwann cell myelination through Egr-2-mediated transcription. Indeed, recent studies in cultured cells have clearly demonstrated that adenosine acting at the A2AR is a signal to stop the cell proliferation and promote Schwann cell myelination (78, 79).
In summary, our combined microarray and bioinformatic analyses revealed distinct striatal gene expression patterns modulated by genetic and pharmacological inactivation of A2ARs. Particularly, it uncovered a predominant downregulatory effect of A2AR KO on striatal gene expression, which cannot be reproduced by treatment with SCH, suggesting that A2AR KO may exert early developmental effects on striatal gene expression. Inference analysis of the transcriptional regulatory network triggered by A2ARs in the striatum indicates that Egr-2 may be the key TF in A2AR-mediated gene expression. The inferred transcriptional regulatory network should facilitate the generation of testable hypotheses about the mechanisms underlying integrated function of A2ARs in vivo. Finally, this analysis indicates that an integrated computational and experimental approach to complex transcriptional regulatory networks triggered by neurotransmission can be very successful and is applicable to other systems.
This study was supported by National Institutes of Health Grants DA-019362 and NS-41083 (to J.-F. Chen), DA-13508 and ES-10804 (to M. A. Schwarzschild), and HG-03110 and GM-066401 (to P. Haverty and Z. Weng); the Bumpus Foundation (to J.-F. Chen); and National Science Foundation Grants DBI-0078194, MRI DBI-0116574, and IGERT-9870710 (to P. Haverty and Z. Weng).
↵* L. Yu and P. M. Haverty contributed equally to this work.
↵1 The Supplemental Material for this article (Supplemental Tables S1 and S2 and Supplemental Files S1 and S2) is available online at http://physiolgenomics.physiology.org/cgi/content/full/00068.2005/DC1.
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).
Address for reprint requests and other correspondence: J.-F. Chen, Dept. of Neurology, Boston Univ. School of Medicine, 715 Albany St., E301, Boston, MA 02118 (e-mail:), or Z. Weng, Dept. of Biomedical Engineering, Boston Univ., Boston, MA (e-mail: ).
- Copyright © 2005 the American Physiological Society