Identification of differentially regulated transcripts in mouse striatum following methamphetamine treatment – an oligonucleotide microarray approach

Authors


Address correspondence and reprint requests to Donald M. Kuhn, Department of Psychiatry and Behavioral Neurosciences, Wayne State University School of Medicine, 540 E. Canfield, Room 2125, Detroit, MI 48201, USA. E-mail: donald.kuhn@wayne.edu

Abstract

Methamphetamine is an addictive drug of abuse that can produce neurotoxic effects in dopamine nerve endings of the striatum. The purpose of this study was to identify new genes that may play a role in the highly complex cascade of events associated with methamphetamine intoxication. Using Affymetrix oligonucleotide arrays, 12 488 genes were simultaneously interrogated and there were 152 whose expression levels were changed following methamphetamine treatment. The genes are grouped into broad functional categories with inflammatory/immune response elements, receptor/signal transduction components and ion channel/transport proteins among the most populated. Many genes within these categories can be linked to ion regulation and apoptosis, both of which have been implicated in methamphetamine toxicity, and numerous factors associated with microglial activation emerged with significant changes in expression. For example, brain-derived neurotrophic factor (BDNF), chemokine (C-C) receptor 6 (CCr6) and numerous chemokine transcripts were increased or decreased in expression more than 2.8-fold. These results point to activated microglia as a potential source of the reactive oxygen/nitrogen species and cytokines that have been previously associated with methamphetamine toxicity and other neurotoxic conditions.

Abbreviations used
BDNF

brain-derived neurotrophic factor

CCr6

chemokine (C-C) receptor 6

DA

dopamine

DAT

dopamine transporter

DMT

Data Mining Tools

GAPDH

glyceraldehyde-3-phosphate dehydrogenase

GFAP

glial fibrillary acidic protein

HRP

horseradish peroxidase

HSP

heat-shock protein

i.p.

intraperitoneal

MAS

Microarray Suite 5.0

METH

methamphetamine

Osmr

oncostatin receptor

SLR

signal log2ratio

SOM

self-organizing map

Methamphetamine (METH) is an addictive stimulant drug of abuse that causes damage to dopaminergic nerve endings of the striatum. The molecular mechanisms underlying METH toxicity are not fully understood, but several have emerged as important mediators. METH targets the dopamine (DA) neuronal system by gaining access to the nerve ending as a substrate for the DA transporter (DAT). Once inside the dopaminergic nerve ending, METH collapses the pH gradient across the synaptic vesicle membrane (Cubells et al. 1994) and inhibits the vesicle monoamine transporter (Brown et al. 2000), allowing DA to leak into the cytoplasm and eventually into the synaptic space by reverse transport through the DAT. Drugs that deplete neuronal DA (Schmidt et al. 1985) or that block the DAT protect against METH toxicity (Schmidt and Gibb 1985; Hanson et al. 1987). For these reasons, DA itself, reactive species derived from DA (Gibb et al. 1989; Yamamoto et al. 1998; LaVoie and Hastings 1999; Larsen et al. 2002) and various reactive oxygen/nitrogen species have been implicated as mediators of METH damage to the dopaminergic neuronal system (Cadet and Brannock 1998; Yamamoto and Zhu 1998; Fleckenstein et al. 2000; Davidson et al. 2001; Imam et al. 2001).

The neurotoxicity of METH is facilitated by its ability to cause hyperthermia. Lowering of core temperature abates METH neurotoxicity (Bowyer et al. 1993, 1994; Miller and O'Callaghan 1994; Albers and Sonsalla 1995) whereas increasing the ambient temperature exacerbates the condition (Miller and O'Callaghan 2003). Drug-induced changes in body temperature can also be associated with disruption of energy production (Stephans et al. 1998; Burrows et al. 2000; Nixdorf et al. 2001) and deficits in ion regulation (Haughey et al. 2000; Callahan et al. 2001; Xie et al. 2002), both of which have also been implicated in METH toxicity.

It seems clear that METH toxicity is closely and specifically associated with the dopaminergic nerve ending, but it can be modulated by more general physiological processes such as ion flux and mitochondrial energy production. Therefore, it is possible that factors arising outside dopaminergic neurons might interact with them to influence METH-mediated neurotoxicity. In an attempt to gain some insight into factors that influence METH toxicity, but which are extrinsic to the dopaminergic nerve ending, we turned to global analysis of gene expression using oligonucleotide arrays. Gene chip analysis has been applied to the study of METH, yielding interesting insights into how this drug may damage the dopaminergic nerve ending. For example, Ricaurte and colleagues identified a set of genes related to ion function whose expression was changed significantly by METH (Xie et al. 2002). Cadet and colleagues have also used microarray analysis to implicate apoptosis as the mechanism by which METH damages the dopaminergic system (Cadet et al. 2001, 2002; Jayanthi et al. 2001). In the present study, we have expanded the search for genes whose expression is differentially regulated by METH through the use of the Affymetrix MGU74Av.2 chip, which contains 12 488 independent transcripts. Several functionally important classes of genes emerge as targets for METH-induced toxicity, including those associated with activated microglia, inflammatory/immune elements, receptor/signal transduction components and ion channel/transport proteins.

Materials and methods

Materials and reagents

METH (methamphetamine hydrochloride) was obtained from Sigma (St Louis, MO, USA). Trizol was acquired from Invitrogen Life Technologies (Grand Island, NY, USA), and RNAlater, QIAshredder Kits, RNeasy Kits and Dnase I were purchased from Qiagen (Valencia, CA, USA). Superscript Choice cDNA Synthesis Kits were obtained from Invitrogen, and Enzo BioArray High Yield RNA Transcript Labeling Kits from Enzo Life Sciences, Inc. (Farmingdale, NY, USA). Antibodies were purchased from the following sources: anti-heat-shock protein (HSP)70, anti-glyceraldehyde-3-phosphate dehydrogenase (GAPDH), anti-brain-derived neurotrophic factor (BDNF), anti-oncostatin receptor (Osmr), anti-chemokine (C-C) receptor 6 (CCr6) and anti-β-actin (Santa Cruz Biotechnology, Santa Cruz, CA, USA), anti-glial fibrillary acidic protein (GFAP) (BD Biosciences, Palo Alto, CA, USA), anti-SOCS-3 (Zymed, South San Francisco, CA, USA), horseradish peroxidase (HRP)-conjugated sheep anti-mouse and HRP-conjugated donkey anti-rabbit (Amersham Biosciences, Amersham, UK) and HRP-conjugated rabbit anti-goat (Cappel, West Chester, PA, USA). Western Lightning Chemiluminescence Reagent Plus was purchased from PerkinElmer Lifesciences (Boston, MA, USA).

Animals

Male C57BL/J6 mice, 6–8 weeks of age, were purchased from Charles River Laboratories (Wilmington, MA, USA) and were acclimatized to their new surroundings for 5 days before experimentation. Animals received food and water ad libitum, and were housed in clear acrylic cages in a temperature-controlled room (22 ± 1°C) on a 12-h light–dark cycle. All animal care and experimental manipulations were approved by the Animal Investigation Committee and Division of Laboratory Animal Resources at Wayne State University. For gene expression studies, mice (n = 2 per time point) received a single intraperitoneal (i.p.) injection of METH (40 mg/kg), and were killed 3, 6, 12 or 24 h after treatment. This dosage was chosen based on its proven ability to induce striatal neurotoxicity in mice (Fukumura et al. 1998; Cadet et al. 2001; Deng et al. 2001; Jayanthi et al. 2001). Control mice (n = 2) received a single i.p. injection of physiological saline and were killed immediately. Striatal tissue was harvested from each animal, and samples from like time points were pooled in RNAlater and stored at − 80°C. Results reported here are from four complete replicates of such experiments. For protein expression studies, mice (n = 5 per time point) received a single i.p. injection of METH (40 mg/kg), and were killed 3, 6, 12, 24, 48 or 72 h after treatment.

RNA isolation and quantification

Each striatal tissue pool was used to prepare hybridization targets for a single microarray. The striatum was chosen based on the vast literature that characterizes the neurotoxic events known to occur in this region following METH treatment. Samples were thawed and RNA isolated by manual homogenization in Trizol, followed by passage through QIAshredder columns. The RNA was purified using RNeasy Kits according to the manufacturer's instructions, treated with Dnase I to remove any contaminating DNA, and quantified by spectrophotometry. The quality of the RNA was assessed using an Agilent 2100 Bioanalyzer RNA6000 Nano Assay, and only RNA with a 28S rRNA : 18S rRNA ratio exceeding 1.3 was used in subsequent target preparation.

Microarray target preparation and hybridizations

Synthesis of cDNA, synthesis and biotinylation of cRNA, fragmentation and hybridization were all performed according to the manufacturer's instructions (Affymetrix, Santa Clara, CA, USA). Briefly, double-stranded cDNA was synthesized from total RNA using a Superscript Choice Kit with a T7-dT24 primer incorporating a T7 RNA polymerase promoter. Biotinylated cRNA was prepared by in vitro transcription using the Enzo BioArray High Yield RNA Transcript Labeling Kit and then fragmented. Some 15 µg of fragmented, biotinylated cRNA was hybridized for 16 h at 45°C to an Affymetrix Mouse genechip, MG-U74A.v2. Hybridizations and scans were done using an Affymetrix Fluidics Station 400 and Agilent Gene Array Scanner respectively (Applied Genomics Technology Center, Center for Molecular Medicine and Genetics, Wayne State University, MI, USA).

Microarray analyses

A minimum of three replicate experiments provides substantial protection against the potential variability of cDNA microarray technology (Lee et al. 2000). Here, we have used the statistical power of four replicate experiments to minimize that likelihood. Data were processed using Affymetrix Microarray Suite 5.0 (MAS) software. A signal value was calculated as a measure of the relative abundance of a transcript. To remove differences between arrays and facilitate comparisons, the data were scaled to a user-defined target intensity. For further analysis, data obtained from MAS were formatted as a tab-delimited text file and imported into Vector Xpression software (Informax Inc., Frederick, MD, USA). Replicates were combined and averaged based on the mean signal value for each gene across the four experiments. The average signal values were adjusted to a minimum threshold value of 20 to exclude outlier data that fell in the noise range. M versus A plots of METH-treated mice to saline-treated control mice from t = 0 were generated for each time point. Lowess normalization was performed and these values were then filtered to include only those transcripts with signal log2ratio (SLR) values of ≤ −1.5 or ≥ 1.5 at least once across the four time points. Most published cDNA microarray analyses to date require an SLR value of ≤ −1 or ≥ +1 (a twofold change in expression) to characterize a gene as differentially regulated. We have increased the stringency of this parameter by requiring a SLR of ≤ −1.5 or ≥ + 1.5 (a 2.8-fold change). Hierarchical cluster analysis, using complete linkage with correlation coefficient of the SLR, was performed on these genes. From the resulting dendogram, a cut-off point of nine clusters was chosen. Within these nine clusters, the average expression profiles of the most differentially regulated genes (2.8-fold) based on SLR are presented. To validate these clustering results, additional cluster analyses were performed on the same genes using Vector Xpression: non-hierarchical QT clustering with correlation coefficient (QT_clust), non-hierarchical k-means clustering with correlation coefficient, and hierarchical clustering using single linkage with correlation coefficient.

To examine the data further using an entirely different approach, the MAS data were imported into Affymetrix Data Mining Tools 3.0 (DMT). Unlike Vector Xpression, DMT filters and analyzes absolute signal values rather than SLR values. Replicates of the signal values were combined and averaged based on the mean for each time point. The merged data were clustered using the self-organizing map (SOM) algorithm. In doing so, the data were first filtered using the DMT default settings and the remaining genes organized into a 3 × 3 grid. Each of the nine clusters represents an average expression profile of the most differentially regulated genes (≥ 2-fold) based on individual signal values. The transcripts obtained following the initial hierarchical cluster analysis were characterized using Affymetrix NetAFFX gene ontology descriptions (http://www.affymetrix.com/analysis/index.affx), and supplemental literature reviews.

Western blot analyses

Striatal tissue was harvested from control mice (t = 0 h; n = 5) and METH-treated mice (t = 3, 6, 12, 24, 48 and 72 h; n = 5 at each time point), then flash frozen before storing at − 80°C. Protein expression was examined over a more expansive time frame than gene expression with the anticipation that changes in translation would lag behind changes in transcription. Striatal lysates were prepared by sonication in boiling lysis buffer (1% sodium dodecyl sulfate, 10 mm Tris pH 7.4, 1.0 mm sodium orthovanadate). Lysates were cleared by centrifugation at 14 000 g for 5 min, and protein concentrations determined using the bicinchoninic acid protein assay. Equivalent amounts of protein from different time points in the experiment were fractionated on sodium dodecyl sulfate–polyacrylamide gels, and then transferred to nitrocellulose. Before analyses, primary antibody dilutions were optimized so that densitometric values of immunoreactive bands fell within the linear range of the scanned images. Blots were probed with diluted primary antibodies (anti-CCr6 1 : 100; anti-GFAP 1 : 250; anti-HSP70 1 : 500; anti-SOCS-3 1 : 250, anti-GAPDH 1 : 400; anti-BDNF 1 : 300; anti-Osmr 1 : 200, anti-β-actin 1 : 800) in blocking buffer (1 × Tris-buffered saline containing 5% non-fat dry milk and 1% Tween-20) overnight at 4°C. Following three washes, blots were incubated for 1 h at room temperature (23°C) with appropriate HRP-conjugated secondary antibodies diluted 1 : 3000 in blocking buffer. Blots were washed three times, and proteins detected using enhanced chemiluminescence. Autoradiographs were scanned for relative densitometric measures using Scion Image 1.59 software (Scion Corporation, Frederick, MD, USA). Data are presented as the mean ± SEM of five replicate experiments.

Statistical analysis

All data were evaluated statistically with Prism version 2.01 (GraphPad Software Inc., San Diego, CA, USA). Gene expression values and protein levels at all time points were subjected to one-way anova followed by post-hoc analysis using Dunnett's multiple comparison test. The significance level was set at p < 0.05.

Results

A total of 152 genes exhibiting 2.8-fold differential regulation remained after filtering all 12 488 genes across four independent experiments. Following an approach similar to one using the Cluster Program (Eisen et al. 1998), hierarchical clustering using complete linkage analysis with correlation coefficient (referred to hereafter as ‘original clustering method’) was performed on these genes (Fig. 1a). Clustering in this manner generated a dendogram that arranged the genes in tight clusters according to similar patterns of expression across all time points. Depending on where one chooses to ‘cut’ the tree, clusters of varying sizes can result (Heyer et al. 1999); in our dataset, a cut-off point of nine clusters provided the most average-sized clusters. Both the complete linkage and average linkage approaches attempt to represent the expression profile data so that meaningful patterns can be readily identified. However, when using Affymetrix data it has been demonstrated that the complete linkage method is superior to that of average linkage in forming less random clusters (Gibbons and Roth 2002). Furthermore, of the five distance metrics available for use with complete linkage in Xpression software, correlation coefficient was chosen because of its superior performance when applied to single-channel array data (Gibbons and Roth 2002), like that presented here. The magnitude of expression for each gene is shown on the color intensity map adjacent to the dendogram. Green indicates lower than median expression (black) and red indicates higher than median expression. The color intensity depicts the degree of gene expression relative to control across all time points. An average expression profile for each cluster was generated (Fig. 1b).

Figure 1.

Expression profiles and clustering of the most differentially regulated genes in mouse striatum following METH treatment. Mice were killed 3, 6, 12 or 24 h following METH treatment (40 mg/kg, i.p.) and striatal tissue was harvested. Following RNA and Affymetrix processing, microarray analyses using Vector Xpression software generated a list of 152 genes with SLR values ≤ − 1.5 or ≥ 1.5 (2.8-fold change in expression). Hierarchical cluster analysis using complete linkage with correlation coefficient generated a dendogram that grouped the genes into nine clusters according to similar patterns of expression across time points (a). Green indicates lower than median expression (black) and red indicates higher than median expression. The color intensity indicates the degree of gene expression relative to control. An average expression profile for each cluster was generated (b ). The light gray lines represent the individual gene expression profiles. The bold, colored line represents the average expression profile for all the genes within each cluster.

One-way anova was performed on all transcripts within each cluster to determine the statistical significance of the representative gene expression profile. All nine clusters were shown to be statistically significant at p < 0.05 (data not shown). The representative gene expression profile for each cluster is a composite of both up- and down-regulated transcripts. To refine the statistical analysis, the genes within each cluster were grouped according to the direction of regulation and then subjected to one-way anova followed by Dunnett's multiple comparison test. The results of this analysis are presented in Table 1. The majority of differentially regulated genes from this experiment (68%) fall in the up-regulated category. It is apparent from these results that they exert more of an influence on the statistical significance of a cluster than do the down-regulated genes. Three of the clusters do not significantly change relative to control with respect to down-regulated genes. This was expected, as the variance in microarray data is largest at lower signal values (Draghici 2002).

Table 1.  Statistically significant clusters as determined by one-way anova with Dunnett's multiple comparison test
ClusterUp-regulatedDown-regulated
3 h6 h12 h24 h3 h6 h12 h24 h
  1. The average signal log2ratio values (mean ± SD) for the genes within a cluster are presented at each time point. *p < 0.05, **p < 0.01 (one-way anova with Dunnett's multiple comparison test), n/a, only one gene was up-regulated in cluster F and therefore anova could not be applied.

A0.2 ± 0.40.3 ± 0.41.6 ± 0.62.0 ± 0.5**− 1.5 ± 0.5*− 1.7 ± 0.0*– 0.3 ± 0.6− 0.8 ± 0.7
B1.1 ± 0.70.3 ± 0.61.2 ± 0.8*1.6 ± 0.4**− 0.6 ± 0.5**− 1.7 ± 0.2**− 0.5 ± 0.4**− 0.5 ± 0.4**
C0.6 ± 0.41.3 ± 0.5**1.7 ± 0.4**0.8 ± 0.4− 0.9 ± 0.9− 0.4 ± 1.6− 0.4 ± 1.1− 1.7 ± 0.1
D2.1 ± 0.9**0.9 ± 0.80.2 ± 0.50.0 ± 0.70.5 ± 0.9− 0.2 ± 0.6− 0.3 ± 0.7− 1.8 ± 0.4
E1.8 ± 0.2**0.5 ± 0.50.9 ± 0.3*0.9 ± 0.4*− 0.2 ± 0.6– 1.2 ± 0.6*− 1.5 ± 0.3**0.7 ± 0.7
Fn/an/an/an/a− 0.4 ± 0.3*− 1.3 ± 0.3**− 0.5 ± 0.2**− 1.5 ± 0.2**
G0.6 ± 0.51.5 ± 0.3**0.8 ± 0.61.5 ± 0.4**− 1.4 ± 0.3**– 0.5 ± 0.4– 1.6 ± 0.2**– 0.3 ± 0.3
H0.1 ± 0.51.7 ± 0.1**1.4 ± 0.3**1.3 ± 0.2**− 1.7 ± 0.2**− 0.6 ± 0.3**– 0.3 ± 0.2– 0.5 ± 0.3**
I0.6 ± 0.61.8 ± 0.2**0.3 ± 0.40.4 ± 0.5− 0.9 ± 0.80.2 ± 0.7− 1.4 ± 0.2− 1.1 ± 0.4

To verify the clustering results, we attempted to validate the original clustering method by applying three additional clustering algorithms to these same 152 genes. Among them were two non-hierarchical clustering algorithms: QT_clust (Heyer et al. 1999) and k-means (Hartigan 1975) with correlation coefficient. QT_Clust analysis resulted in a number of clusters ranging in size from one to 12 members, the largest of which overlapped with significant portions of original clusters A, B, D and I (Fig. 2). K-means did support the hierarchical clustering results; however, there is an inherent bias in that the user must define the final number of resulting clusters before analysis. To diminish this bias, before applying the k-means algorithm to the data set, the number of clusters was predefined at five based on the Silhouette statistic (Kaufman and Rousseeuw 1990) and Gap statistic (Hastie et al. 2001) cluster validation algorithms. As seen with the QT_clust confirmation of our original clustering method, large groups of genes once again partitioned together (data not shown). The third clustering algorithm applied to the data was hierarchical clustering using single linkage with correlation coefficient, and, again, a cut-off of nine clusters was chosen. As expected, the cluster sizes were significantly skewed; clusters were either very large or very small. The original clusters were considerably rearranged, a result fully expected owing to the random nature of cluster formation inherent in this approach (Gibbons and Roth 2002). Even with this significant rearrangement, distinct portions of the routinely migrating groups were evident (data not shown). Clusters A, B, D and I appear to have groups of genes that routinely migrate together, suggesting their constituents might be biologically related.

Figure 2.

Validation of hierarchical clustering using complete linkage by comparison with non-hierarchical QT_clust. The dendograms from selected clusters generated by hierarchical clustering using complete linkage (clusters A, B, D and I) are presented alongside shaded gray boxes (numbered 1–3) that represent the regions encompassed by the three largest gene clusters generated by QT_clust. The gene accession numbers for genes within these specific regions of the dendogram are presented in the adjacent boxes. Bold numbers within each box represent components of the gene clusters generated by QT_clust.

In addition to rigorous analysis of these results using various algorithms within Xpression, we sought to add additional layers of power to our approach by further examining these results using a second software package. Affymetrix DMT 3.0 offered the option of analyzing the data using absolute signal values as opposed to SLR values. It also allowed us to take advantage of another clustering algorithm, SOM, not available in the first package. Repeated validations of this sort should overcome any limitations of a one-sided approach, and thereby impart confidence to the results. Following the application of DMT's filtering criteria, 89 of 12 488 genes remained. These genes were processed using the non-hierarchical SOM algorithm (Tamayo et al. 1999) set for three columns and three rows, generating nine clusters. Upon examination of these 89 genes, it was noted that 61 were common to the 152 genes obtained using the original clustering method (bold entries, Tables 2 and 3). The 28 genes that did not overlap failed to do so as they did not meet our stringent criterion of at least 2.8-fold differential regulation. Again, as with previous validations, substantial numbers of genes partitioned together (data not shown). More importantly, most members of original clusters A, D and I can be found in the SOM clusters 9, 3 and 4 respectively. The fact that this list was smaller than the original list was fully expected, as the filtering criteria used in this algorithm are more stringent than those used in Xpression. For a given gene, DMT requires that the range in signal values across all time points exceeds a certain threshold. No such filtering is present in the Xpression analysis and therefore will allow those genes with smaller ranges in expression over time to be defined as differentially regulated.

Table 2.  Transcripts up-regulated at least 2.8-fold in mouse striatum 3, 6, 12 or 24 h following METH treatment
Gene
accession #
ClusterTranscript IDEncoded protein, EST, RIKEN cDNASLR value
3 h 6 h12 h24 h
  1. All transcripts result from hierarchical clustering with complete linkage using Xpression software. Bold entries highlight the transcripts resulting from SOM clustering using Affymetrix DMT software that also meet the additional criterion of 2.8-fold up-regulation of expression.

Cell cycle
 M38724BCdc2acell division cycle 2 homolog A (S. pombe)0.450.100.811.64
 AW048937DCdkn1acyclin-dependent kinase inhibitor 1 A (P21)1.511.070.02− 0.13
 AV377480IMcmd7mini chromosome maintenance deficient 7 (S. cerevisiae)1.131.50− 0.080.64
Cell growth and maintenance
 X55573ABdnfbrain-derived neurotrophic factor0.970.632.471.76
 L01991ICntn3contactin 3− 0.151.57− 0.090.36
 X98471AEmp1epithelial membrane protein 10.160.190.591.77
 X14897DFosbFBJ osteosarcoma oncogene B1.910.25− 0.53− 0.53
 X69619DInhbainhibin beta-A1.820.56− 0.17− 0.21
Cell structure, adhesion and motility
 X54511ACapgcapping protein (actin filament), gelsolin-like; mbh1 gene for
Myc basic motif homologue-1 (mbh1)
0.09− 0.401.081.87
 U52925DF5coagulation factor V1.730.420.36− 1.94
 X02801AGfapglial fibrillary acidic protein0.030.361.522.15
 X02801AGfapglial fibrillary acidic protein− 0.070.161.672.59
 D86420IKrtap6–1keratin-associated protein 6–11.291.660.761.13
 U43298ILamb3laminin, beta 3− 0.501.51− 0.540.00
 X16834ALgals3lectin, galactose binding, soluble 3; Mouse mRNA for Mac-2 antigen− 0.18− 0.161.801.87
 AW124470ATm4sf7transmembrane 4 superfamily member 7− 0.240.201.652.56
Inflammatory and immune response
 AI643420EBag3Bcl2-associated athanogene 31.530.480.510.18
 AV373612DBag3Bcl2-associated athanogene 32.140.321.160.49
 M19681ACcl2chemokine (C-C motif) ligand 20.24− 0.562.161.59
 J04491CCcl3chemokine (C-C motif) ligand 30.691.882.031.19
 X62502CCcl4chemokine (C-C motif) ligand 4; ccl-4; MIP-1b gene− 0.550.921.530.20
 U50712ACcl12chemokine (C-C motif) ligand 12− 0.280.112.583.35
 U79525CCmklr1chemokine-like receptor 10.310.321.550.07
 X15591CCtla2acytotoxic T lymphocyte-associated protein 2 alpha1.101.741.640.59
 U12560IDefcr5defensin-related cryptdin 50.211.710.31− 0.17
 L23636IFlt3lFMS-like tyrosine kinase 3 ligand0.001.840.000.00
 AA983101AGlipr2GLI pathogenesis-related 20.240.401.901.97
 M65027AGp49aglycoprotein 49 A0.000.002.281.90
 M35525CHchemolytic complement0.951.411.590.74
 AF109906DHsp70–1heat-shock protein, 70 kDa 14.812.05− 0.24− 0.08
 M12571DHsp70–3heat-shock protein, 70 kDa 32.191.290.210.46
 M14639CIl1ainterleukin 1 alpha0.781.362.091.49
 L32838BIl1rninterleukin 1 receptor antagonist0.81− 0.181.661.09
 X83601APtx3pentaxin-related gene0.350.602.161.57
 M83218IS100a8S100 calcium-binding protein A8 (calgranulin A); MRP80.942.290.410.39
 M83219IS100a9S100 calcium-binding protein A9 (calgranulin B); MRP141.122.250.400.55
 U41341AS100a11S100 calcium-binding protein A11 (calizzarin); endothelial
monocyte- activating polypeptide I mRNA
0.290.581.251.87
 AV374868CSOCS3suppressor of cytokine signaling 30.631.171.951.31
 U88328HSOCS3suppressor of cytokine signaling 30.211.631.701.19
 M20985G mouse MHC class I H2-Qa-Mb1 gene0.771.341.221.74
Ion channels and transport proteins
 L06234ICacna1scalcium channel, voltage-dependent, L type, alpha 1S subunit0.531.640.560.83
 X93038GFxyd3FXYD domain-containing ion transport regulator 3, mRNA for MAT8 protein0.381.520.350.89
 AI591702FKif23kinesin family member 231.511.021.620.92
 X81627DLcn2lipocalin 2 (24p3 gene)3.273.331.021.57
 AJ006036ISlc22a2solute carrier family 22 (organic cation transporter), member 21.361.870.770.67
 U30840IVdac1voltage-dependent anion channel 10.611.55− 0.160.89
 AF075267IXtrp2X transporter protein 2; orphan transporter isoform A80.871.550.761.00
Kinases and phosphatases
 AB021861DMap3k6mitogen-activated protein kinase kinase kinase 61.630.980.33− 0.48
 AV317524DPtpns1protein tyrosine phosphatase, non-receptor type substrate 11.610.08− 0.020.07
 X60980ETk1thymidine kinase 11.531.261.261.48
Metabolism
 AI326963CAngptl4angiopoietin-like 40.781.581.170.78
 U04204AAkr1b8aldo-keto reductase family 1, member B80.240.440.801.90
 M32599GGapdglyceraldehyde-3-phosphate dehydrogenase− 0.382.19− 0.181.20
 AA120675BPrps2phosphoribosyl pyrophosphate synthetase 21.590.381.481.38
Protease inhibitors
 AI850558AA2malpha-2-macroglobulin0.620.141.281.52
 AB015224ACst7cystatin F (leukocystatin); mRNA for murine CMAP0.340.211.742.70
 X15592BCtla2bcytotoxic T lymphocyte-associated protein 2 beta1.901.431.981.90
 M64086BSerpina3nserine (or cysteine) proteinase inhibitor, clade A, member 3 N0.460.350.581.56
 X16490ESerpinb2serine (or cysteine) proteinase inhibitor, clade B, member 22.060.060.440.33
Protein turnover
 AW408912BUbce8ubiquitin-conjugating enzyme 80.11− 0.240.191.57
Receptors and signal transduction
 AF004326DAgpt2angiopoietin 21.871.630.420.33
 L02844DCd22CD22 antigen1.700.380.080.22
 J03857ECd79bCD79B antigen1.60− 0.190.520.85
 X03818EChrngcholinergic receptor, nicotinic, gamma polypeptide1.670.581.120.92
 L22181DFpr1formyl peptide receptor 11.610.790.710.32
 U10551DGemGTP binding protein (gene overexpressed in skeletal muscle)1.930.981.190.18
 L10666CGnat2guanine nucleotide-binding protein, alpha transducing 20.742.011.680.99
 AI850277GNmuneuromedin0.341.710.351.25
 AB015978BOsmroncostatin receptor1.420.402.382.34
 AV361022EPde6bphosphodiesterase 6B, cGMP, rod receptor, beta polypeptide1.750.801.051.27
 X62700HPlaururokinase plasminogen activator receptor (muPAR1)0.551.531.011.30
 AA656014HTm7sf1transmembrane 7 superfamily member 10.551.571.381.16
 M83649ETnfrsf6tumor necrosis factor receptor superfamily, member 61.660.591.000.91
 L24495GTnfrsf7tumor necrosis factor receptor superfamily, member 70.311.281.161.52
RNA related
 AV126830GSars2seryl-aminoacyl-tRNA synthetase 20.751.040.921.61
Transcription factors and DNA-binding proteins
 X61800BCebpdCCAAT/enhancer binding protein (C/EBP), delta1.831.131.791.77
 M60285DCremcAMP responsive element modulator1.730.46− 0.02− 0.14
 M60285DCremcAMP responsive element modulator1.580.230.04− 0.70
 X86368IFoxd4forkhead box D40.351.560.190.17
 U36340CKlf3Kruppel-like factor 3 (basic) (CACCC-box binding protein BKLF mRNA)0.371.550.820.56
 J04620HPrim1DNA primase, p49 subunit (priA)0.471.931.301.20
 AF064088ITiegTGFB -inducible early growth response (transcription factor GIF)− 0.052.090.34− 0.30
 AW124134C Mus musculus 10, 11 days embryo whole body cDNA, RIKEN full-lengthsequence enriched library, clone:2810490L08: scleraxis, full insert0.891.322.150.96
Transferases
 AF004108IAanatarylalkylamine N-acetyltransferase0.161.610.550.04
 AV174251HCmascytidine monophospho-N-acetylneuraminic acid synthetase− 0.791.601.471.69
 AF015768GLfnglunatic fringe gene homolog (Drosophila)1.411.590.741.98
Others
 AI510131CHurp-pendinghepatoma up-regulated protein0.611.321.720.78
 AW230891DLrg-pendingleucine-rich alpha-2-glycoprotein2.391.040.220.95
 X76652GRai2retinoic acid induced 21.101.661.661.96
 AF004941AS100a3S100 calcium-binding protein A3− 0.370.162.101.94
 M16465AS100a10S100 calcium-binding protein A10 (calpactin)− 0.20− 0.140.361.63
 AF064748IS3-12-pendingplasma membrane-associated protein, S3-120.521.790.160.00
 X87671ISh3bp1SH3-domain-binding protein 1; 3 BP-11.282.021.241.45
ESTs, RIKEN genes and hypothetical proteins
 AW212532H ESTs− 0.291.721.661.02
 AI415109A ESTs, moderately similar to solute carrier family 4 (anion exchanger),
member 8; sodium bicarbonate cotransporter isoform 3 kNBC
1.141.001.491.67
 AI553024D ESTs, highly similar to 2118318A promyelocyte leukemia
Zn finger protein (M. musculus)
1.950.70− 0.91− 0.09
 AW125432ALOC227699hypothetical protein LOC2276991.161.151.721.38
 AA986100BMGC37245hypothetical protein MGC372451.74− 0.151.421.02
 AA080253CB930025N23RikRIKEN cDNA B930025N23 gene0.890.781.720.55
 AI461631G1110025G12RikRIKEN cDNA 1110025G12 gene0.801.560.731.31
 AW208938D1200008D14RikRIKEN cDNA 1200008D14 gene1.620.250.120.27
 AW215736A2310057H16RikRIKEN cDNA 2310057H16 gene0.420.441.412.00
 AI841689B9430096L06RikRIKEN cDNA 9430096L06 gene0.69− 0.130.081.51
Table 3.  Transcripts down-regulated at least 2.8-fold in mouse striatum 3, 6, 12 or 24 h following METH treatment
Gene
accession #
ClusterTranscript IDEncoded protein, EST, RIKEN cDNASLR value
3 h6 h12 h24 h
  1. All transcripts result from hierarchical clustering with complete linkage using Xpression software. Bold entries highlight the transcripts resulting from SOM clustering using Affymetrix DMT software that also meet the additional criterion of 2.8-fold down-regulation of expression.

Cell growth and maintenance
 AA756568BGmfbglia maturation factor, beta− 0.19− 1.57− 0.55− 0.22
 AA647799EOgnosteoglycin− 0.23− 1.51− 1.37− 1.42
Cell structure, adhesion and motility
 U52925DF5coagulation factor V1.730.420.36− 1.94
 AV074196EKrt1–13keratin complex 1, acidic, gene 130.21− 1.59− 1.09− 0.09
Inflammatory and immune response
 AJ001101HC1qbpcomplement component 1, q subcomponent-binding protein− 1.94− 0.88− 0.47− 0.83
 AJ222714ICCr6chemokine (C-C) receptor 6− 0.121.16− 1.32− 1.72
 X04418BPrlprolactin− 1.04− 1.73− 0.51− 0.80
 X03278BTcrb-V13T-cell receptor beta, variable 13− 0.37− 1.68− 0.89− 0.70
 X72904G MVA5T mRNA for T cell receptor alpha chain− 1.71− 0.88− 1.52− 0.35
Ion channels and transport proteins
 AA986344FKcnj1potassium inwardly rectifying channel, subfamily J, member 1− 0.53− 0.94− 0.40− 1.59
 U60091FTap2transporter 2, ATP-binding cassette, subfamily B (MDR/TAP)0.02− 1.70− 0.37− 1.19
 D00073DTtrtransthyretin1.180.34− 0.38− 2.69
 AV365676HVps45vacuolar protein sorting 45 (yeast)− 1.65− 0.31− 0.27− 0.49
Kinases and phosphatases
 U26589BAdkadenosine kinase− 0.24− 1.90− 0.040.01
 U93848GEef2keukaryotic elongation factor-2 kinase− 1.51− 0.28− 1.49− 0.49
 U36488FPtprvprotein tyrosine phosphatase, receptor type, V− 0.46− 1.20− 0.29− 1.64
Protease inhibitors
 AF010254BSerping1serine (or cysteine) proteinase inhibitor, clade G, member 1− 0.52− 1.50− 0.21− 0.39
Protein synthesis
 AW108045HEIF4A2eukaryotic translation initiation factor 4A2− 1.54− 0.44− 0.07− 0.18
Protein turnover
 M57401BMcptlmast cell protease-like− 0.08− 1.85− 0.71− 0.13
Receptors and signal transduction
 X72862DAdrb3adrenergic receptor, beta 3− 0.53− 0.70− 1.04− 1.71
 AV035020DFolr1folate receptor 1 (adult)1.09− 0.310.20− 1.55
 AF031127DItpr5inositol 1,4,5-triphosphate receptor 5− 0.52− 1.07− 1.58− 1.24
 C80388EPrkczprotein kinase C, zeta0.08− 0.27− 1.590.02
 D10214DPrlrprolactin receptor0.910.580.30− 1.80
 X58289HPtprbprotein tyrosine phosphatase, receptor type, B− 1.66− 0.86− 0.57− 0.62
 U67326ARasgrf2RAS protein-specific guanine nucleotide-releasing factor 2− 1.64− 1.73− 0.68− 1.26
 AF100778BWisp2WNT1 inducible signaling pathway protein 2; Mus musculus connective tissue growth factor-related protein WISP-2 (Wisp2) mRNA− 0.48− 1.67− 0.42− 0.68
Transcription factors and DNA-binding proteins
 AW213865BAsf1bASF1 antisilencing function 1 homolog B (S. Cerevisiae)− 1.28− 1.95− 1.210.07
 M24377CEgr2early growth response 2− 1.53− 1.59− 1.17− 1.79
 AF010405BFoxq1forkhead box Q1− 0.08− 2.00− 0.62− 0.50
 D13803BRad51RAD51 homolog (S. cerevisiae)− 0.38− 1.61− 0.22− 0.32
Transferases
 AV238359DCratcarnitine acetyltransferase− 0.39− 0.61− 0.86− 1.97
 AJ002141IDsppdentin sialophosphoprotein− 1.62− 0.78− 1.42− 1.28
 M88694GTemtthioether S-methyltransferase− 1.18− 0.83− 1.56− 0.57
Tumor suppression
 U89652ABrca2breast cancer 2− 1.00− 1.750.41− 0.05
Others
 AI047912FD1Pas1DNA segment, Chr1, Pasteur Institute 1− 0.67− 1.58− 0.20− 1.27
 AW120971ANulp-1-pendingnuclear localized protein 1− 1.87− 1.76− 0.64− 1.20
 X12809IPhxr1per-hexamer repeat gene 1− 0.370.20− 1.64− 0.62
 AF062476IStra6stimulated by retinoic acid gene 6− 1.850.50− 1.12− 1.00
ESTs, RIKEN genes and hypothetical proteins
 AA408251E ESTs− 1.09− 1.52− 1.85− 1.16
 AU020743B ESTs− 0.63− 1.69− 0.50− 0.86
 AW047736I ESTs− 0.64− 0.33− 1.57− 0.86
 C76472G ESTs, weakly similar to S70642 ubiquitin ligase Nedd4 – rat
(fragment) (R. Norvegicus)
− 1.160.05− 1.860.04
 AV301574BG430025P05RikRIKEN cDNA G430025P05 gene− 0.90− 1.67− 0.26− 0.32
 AW121336D1600023A02RikRIKEN cDNA 1600023A02 gene0.940.000.22− 1.57
 AW047445B3110041O18RikRIKEN cDNA 3110041O18 gene− 1.50− 1.97− 1.04− 1.28
 AV362664F5033402L14RikRIKEN cDNA 5033402L14 gene− 0.65− 1.07− 0.77− 1.55
 AA873956F5830404H04RikRIKEN cDNA 5830404H04 gene− 0.36− 1.30− 0.74− 1.68
 AW121848C9130005 N23RikRIKEN cDNA 9130005N23 gene− 0.200.740.37− 1.60
 AI849831B9130022B02RikRIKEN cDNA 9130022B02 gene− 0.07− 1.52− 0.11− 0.19

The data obtained using Xpression software are presented in Tables 2 and 3. Within the tables we have broadly classified each gene based on gene ontology information presented on the Affymetrix NetAFFX website (http://www.affymetrix.com), and supplemental literature reviews. HSP70-1 is the most up-regulated transcript, reaching peak expression at 3 h. Most of the up-regulated transcripts fall within the inflammatory and immune response group (Table 2). In fact, among the 152 differentially regulated genes identified in this study, 19% fall within this category. Hierarchical clustering will produce clusters of genes that are often functionally related; however, when applied to large data sets, the clusters are often populated with uncharacterized genes that might share common functions (Eisen et al. 1998). Of those that could be categorized, 14% are receptors and/or are involved in signal transduction. The largest group of down-regulated genes contains those that have yet to be characterized (expressed sequence tags [ESTs], RIKEN genes and hypothetical proteins; Table 3). Transthyretin is the most down-regulated transcript in our experiment (6.5-fold decrease at 24 h). Only one gene, coagulation factor V, appears in both lists as it was up-regulated at 3 h and then down-regulated at 24 h.

Many of the 152 transcripts presented in this study are relevant to METH-induced neurotoxicity. These include HSP70, GFAP, BDNF, Osmr, SOCS-3 and CCr6. A number of other genes exhibiting differential regulation may not be intuitively correlated with METH-induced neurotoxicity. Among these is GAPDH, classically considered a ‘housekeeping gene’. These genes of interest were chosen to examine their protein expression after METH treatment. The resulting protein expression profile for each is presented in Fig. 3 with its respective transcriptional expression profile. It can be seen that the pattern of change in protein expression was in general agreement with changes in transcript levels. For example, BDNF and GFAP were increased in expression by about 12–24 h after METH treatment and continued to increase through the endpoint of 72 h (2–2.5-fold increase for each). Similarly, HSP70, Osmr, and SOCS-3 were gradually increased in expression between 12 and 72 h after METH. CCr6, the only down-regulated gene examined by western analysis, shows a statistically significant decrease in protein levels from 6 to 24 h after METH, and increases slightly thereafter. GAPDH protein expression was steady for about 48 h after METH treatment and then increased significantly by 72 h, or about 2 days after increases in its transcript. In general, it appeared that changes in protein levels lagged behind changes in their respective genes, as expected. Neither the expression levels of the β-actin gene nor its protein were changed by METH.

Figure 3.

METH effects on protein and gene expression profiles. Protein expression levels for these transcripts are presented as densitometric values relative to control from five replicate experiments (mean ± SEM). Examination of protein levels was extended to 48 and 72 h. Gene expression signal values for 3, 6, 12 and 24 h are presented as a percentage of control from four replicate experiments (mean ± SEM). Statistically significant changes in protein expression (#p < 0.05; ##p < 0.01) and gene expression (*p < 0.05; **p < 0.01) are indicated (one-way anova followed by Dunnett's multiple comparison test). A representative Western blot for each protein is presented.

Discussion

The present results show that METH causes significant changes in expression of numerous genes in the striatum. Interrogation of 12 488 genes on Affymetrix MG-U74A.v2 chips revealed that 152 genes were changed in expression (up- and down-regulated) more than 2.8-fold. Many of these genes have been shown previously by microarray analysis to be responsive to METH, including apoptosis-related (Cadet et al. 2001) and ion-regulating transcripts (Xie et al. 2002). GFAP (O'Callaghan and Miller 1994; Pu and Vorhees 1995), HSP70 (Barrett et al. 2001; Cadet et al. 2001; Xie et al. 2002) and neuromedin (Adams et al. 2001), all known to be induced by METH, were also identified by the present microarray approach as being significantly increased in expression. Confirmation of METH-induced changes in expression of these genes, along with the present use of multiple analytical approaches (hierarchical and non-hierarchical clustering, cluster validation, and one-way anova analysis) adds substantial validation to our findings of new METH-responsive genes.

Gene expression changes caused by METH were clustered into nine distinct patterns with an average of 17 genes in each cluster, evidence of the complexity of the neurobiological effects of this drug of abuse. These genes could also be sorted into more biologically ‘functional’ groups, possibly revealing new pathways involved in METH neurotoxicity. For instance, numerous genes commonly associated with inflammatory and immune responses, signal transduction pathways, cell structure and motility, and ion channels and transport proteins were changed significantly in expression by a toxic regimen of METH.

Several classifications of transcripts presented here can be readily linked to the processes thought to be involved in METH-induced neurotoxicity. Callahan et al. (2001) provide compelling evidence implicating a role for ion dysregulation. Although the relationship between the two has not been fully characterized, our data indirectly support this possibility, as we have identified a significant number of differentially regulated genes encoding ion channel components and transport proteins. For example, SLC22a2 (homologous to OCT2) is an organic cation transporter that is up-regulated almost fourfold 6 h after treatment. OCT2 is implicated in the translocation of agmatine, which may function as a neurotransmitter and inhibit cellular proliferation (Grundemann et al. 2003). The uptake of agmatine is dramatically increased with decreasing pH, a condition likely to be present in a neurons exposed to METH, in which H+ transport is drastically altered (Callahan et al. 2001). The most up-regulated transport gene in this study is lipocalin 2, a protein believed to bind small lipophilic substances like lipopolysaccharide, and may also function as a modulator of inflammation. Lipocalin has also been shown to possess pro-apoptotic properties (Devireddy et al. 2001; Tong et al. 2003).

In addition to ion dysregulation, apoptosis has been implicated in METH-induced neurotoxicity. Using TαT (terminal deoxynucleotidyl transferase)-mediated dUTP nick end labeling (TUNEL) immunohistochemistry, Deng et al. (2001) discovered significant numbers of apoptotic cells in mouse brain following METH treatment. Further investigation suggested a role for c-Jun in this process (Deng et al. 2002). We have identified transcripts that encode both pro- and anti-apoptotic elements. Among the pro-apoptotic proteins is calprotectin, a complex of two calcium-binding proteins belonging to the S100 family (S100a8 and S100a9; both transcripts are up-regulated at 6 h in this study), which possesses apoptosis-inducing activity (Yui et al. 2003). Although not necessarily involved in the process itself, Ptx3, up-regulated at 12 and 24 h in this study, has been shown to specifically bind to apoptotic cells (Rovere et al. 2000). GAPDH, once considered the classical housekeeping gene, is now purported to be involved in the initiation of apoptosis (Berry and Boulton 2000).

The absolute number of transcripts present within the ‘inflammatory and immune response’ category strongly suggests such a reaction is occurring following METH treatment. In addition to the aforementioned up-regulation of HSP70 transcripts, four chemokine ligands, interleukin-1α (Patel et al. 2003) and CapG (Witke et al. 2001) also demonstrate substantial increases. Although categorized in the ‘receptors and signal transduction’ category, OSMR transcripts have been shown to be up-regulated during inflammatory responses in human cerebral endothelial cells (Ruprecht et al. 2001). S100a8/a9 and Ptx3, in addition to having apoptotic properties mentioned above, are also associated with inflammatory/immune responses. S100a8/a9 has been shown to be a potent activator of neutrophils and may be involved in neutrophil migration to inflammatory sites (Ryckman et al. 2003). Ptx3 has the potential to activate or inhibit the classical complement pathway, suggesting a regulatory role in the innate immune response (Nauta et al. 2003). SOCS-3, up-regulated at 6 and 12 h in this study, has been shown to counter-regulate the hypothalamo-pituitary-adrenal axis response to inflammation (Chesnokova et al. 2002). CCr6 is expressed at low levels on both microglia and astrocytes (Flynn et al. 2003). CCr6 and its ligand CCl20 have both been shown to be up-regulated in the CNS of mice during experimental autoimmune encephalomyelitis (Serafini et al. 2000), and it has been suggested that through the secretion of CCl20 astrocytes may be involved in recruiting specific leukocyte subsets to the inflamed CNS thus regulating CNS-targeted immune responses (Ambrosini et al. 2003). Previous cDNA microarray analysis has shown CCr6 to be down-regulated in patients with relapsing–remitting multiple sclerosis (Ramanathan et al. 2001). In this study CCr6 gene and protein expression were shown to be down-regulated suggesting that METH treatment may lead to an impaired immune response in the brain.

Further examination of the genes within this category suggests a strong link between METH and microglia, the CNS equivalent of systemic macrophages. Cytokines and their receptors play a major role in the defense of the CNS. They are expressed at constitutively low levels in microglia, and are induced by inflammatory mediators (Mennicken et al. 1999). BDNF (up-regulated approximately fivefold at 12 h) is produced by both activated microglia and neurons. It inhibits intracellular oxyradical stress triggered by DA and partially blocks basal and DA-induced apoptosis (Petersen et al. 2001). BDNF is also implicated in METH-induced release of DA and the induction of DA-related behaviors (Narita et al. 2003), and can also induce neuronal necrosis accompanied by reactive oxygen species production (Kim et al. 2002). The significant up-regulation of a multitude of inflammatory and immune response factors further supports the hypothesis that microglia are involved in METH-induced neurotoxicity. The METH-induced ‘activation’ of striatal microglia in rats has been suggested previously (Bowyer et al. 1994). In the light of recent studies showing that microglia mediate the neuronal damage associated with 6-hydroxydopamine (He et al. 2001), 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (Du et al. 2001; Wu et al. 2002), cerebral ischemia (Yrjanheikki et al. 1998; Yrjanheikki et al. 1999) and excitotoxins (Tikka et al. 2001), and are not simply responding to local damage to neurons, the possibility that METH-induced damage to the dopaminergic neuronal system is likewise mediated by microglia seems likely and is currently under investigation in our laboratory.

Acknowledgements

Supported by National Institute on Drug Abuse (NIDA) grants DA13753 and DA14692, and by a Department of Veterans Affairs (VA) Merit Award. We also thank Dr Susan J. Land, Tara Twomey and Daniel Lott of the Wayne State University (WSU) Applied Genomics Technology Center for their technical assistance and helpful discussions.

Ancillary