Gene expression following acute morphine administration
A. V. LOGUINOV1,
L. M. ANDERSON2,4,
G. J. CROSBY3,4 and
R. Y. YUKHANANOV3,4
1 Department of Nutritional Sciences and Toxicology, University of California, Berkeley, California
2 Gene Array Technology Center, Department of Medicine
3 Neurogenomic Laboratory, Pain Research Center, Department of Anesthesiology, Perioperative and Pain Medicine, Brigham and Womens Hospital
4 Harvard Medical School, Boston, Massachusetts 02115
 |
ABSTRACT
|
---|
The long-term response to neurotropic drugs depends on drug-induced neuroplasticity and underlying changes in gene expression. However, alterations in neuronal gene expression can be observed even following single injection. To investigate the extent of these changes, gene expression in the medial striatum and lumbar part of the spinal cord was monitored by cDNA microarray following single injection of morphine. Using robust and resistant linear regression (MM-estimator) with simultaneous prediction confidence intervals, we detected differentially expressed genes. By combining the results with cluster analysis, we have found that a single morphine injection alters expression of two major groups of genes, for proteins involved in mitochondrial respiration and for cytoskeleton-related proteins. RNAs for these proteins were mostly downregulated both in the medial striatum and in lumbar part of the spinal cord. These transitory changes were prevented by coadministration of the opioid antagonist naloxone. Data indicate that microarray analysis by itself is useful in describing the effect of well-known substances on the nervous system and provides sufficient information to propose a potentially novel pathway mediating its activity.
opioids; cDNA microarray; regression analysis; cluster analysis; spinal cord; striatum; skeletal protein
 |
INTRODUCTION
|
---|
THE ACTIVITY OF NERVOUS SYSTEM depends on fast and transitory changes in neuronal excitability. It is believed that only repeated stimuli alter gene expression in neurons. However, the changes in gene expression do not always require long exposure to the initial stimulus. Morphine, which is known to induce changes in the neuronal physiology that persist for a long time following cessation of drug exposure (23), alters gene expression even following single injection (8, 25, 30). These changes may be markers/mediators of neuronal responses to opioids and indicators of regulatory pathways involved. With the advent of gene array technology, it becomes practical to evaluate the expression of many genes simultaneously. However, it is unknown to what extent this technology is useful to study such a heterogeneous tissue as neural tissue, and it is unknown whether it is possible to see any changes and make any conclusion regarding intracellular pathways involved. Therefore, in this study a relatively simple question was addressed, whether it is possible to detect the effect of a single morphine injection on gene expression in neural tissue using microarray technology. Two areas responsible for mediation of the reward properties of opioids (medial striatum) and their analgesic activity and withdrawal reaction (lumbar enlargement of the spinal cord) were chosen for analysis. To compare multiple samples and to find coregulated groups of genes, we combine the regression and cluster analysis. As a result we showed that a single morphine injection alters the gene expression both in the spinal cord and medial striatum, which can be detected by cDNA microarray. The majority of altered genes represents mitochondrial and cytoskeletal proteins that may belong to a novel pathway mediating some effects of morphine.
 |
MATERIAL AND METHODS
|
---|
Subjects
Male, B6CBAF1/J, 8-wk-old mice (Jackson Laboratories, Mt. Desert Island, ME) were housed in groups of 4 with a 12:12-h light-dark cycle and food and water available ad libitum. Animals were injected with morphine (M group, 10 mg/kg ip), saline (S group, saline 10 ml/kg), or morphine+naloxone (N group, 10 mg/kg ip of each). This dose of morphine produces strong analgesia and the phenomenon of acute dependence that can be observed as soon as 2 h after injection (29). To confirm the time course of morphine-induced analgesia, tail-withdrawal latency was measured for 2 h following morphine injection in a separate group of mice (we did not measure the pain sensitivity in mice used for RNA preparation, since extra stress can modify gene expression).
All animal protocols were approved by the Animal Care Committee at Harvard Medical School, Boston, MA.
Analgesia
Analgesia was assessed by the tail immersion assay at a water temperature of 52°C as described elsewhere (35). The mouse tail was immersed in the heated water. The latency to a rapid flick of the tail was recorded as the withdrawal latency. After drug administration, we used a maximum cutoff of 15 s to minimize tissue damage.
Naloxone-Precipitated Jumping
Naloxone-precipitated jumping was determined by observation of the animal in a 3-liter glass cylinder containing 23 cm of bedding for 15 min following naloxone injection as described elsewhere (34). To confirm that even a single morphine injection produces an acute withdrawal reaction (28), the separate group of mice was injected with naloxone (15 mg/kg ip) 2 h following morphine administration. The response to naloxone was considered positive if the mouse jumped at least four times.
Tissue Dissection and RNA Isolation
To prepare the total RNA for array hybridization, morphine-injected (M group) or saline-injected (S group) mice were killed 30 or 120 min after injection, and morphine+ naloxone-injected mice (N group) were killed 30 min after injection. Thus there were five groups of at least six mice each: M30, M120, S30, S120, and N30. After decapitation, the brain was quickly removed and placed in mouse brain matrix (TedPella) with division 1 mm, cooled on ice. The prefrontal cortex was removed at the level of about 2.0 mm; the distances are given from the Bregma line according to the stereotaxic atlas of mouse brain (7). The next section cut at 0.5 mm (just in front of the optical chiasma) was used to dissect the medial striatum with nucleus accumbens by a diagonal cut just above the anterior commissure. The spinal cord was removed under hydraulic pressure using a 10-ml syringe with saline (21), and the lumbar enlargement was saved. Tissues were frozen and kept at -80°C until RNA isolation. (We have found that it is better to keep the brain tissue rather than extracted RNA; even at -80°C, extracted RNA slowly deteriorates.) The tissue was homogenized at high speed in Trizol (Life Sciences), and total RNA was isolated according to the manufacturers instructions. The concentration of total RNA was measured by ultraviolet (UV) spectrophotometry at 260/280 nm, and RNA quality was assessed by electrophoresis on a 1% agarose gel. Only those samples with a 260/280 ratio more than 1.9 and no signs of degradation based on agarose electrophoresis were used for analysis.
RNase Protection Assay
RNase protection assay was done as described elsewhere (1) with some modifications. In brief, 32P-labeled antisense riboprobes were synthesized with T7 or SP6 RNA polymerase from linearized plasmids and purified by electrophoresis on polyacrylamide gels. Specific activity of the labeled probes was
13 x 108 cpm/µg. The aliquots of the sample (2 µg of total RNA) were incubated with a molar excess of 32P-labeled antisense RNA (105 cpm) for 1620 h at 55°C in 30 µl of the hybridization buffer (80% formamide, 40 mM PIPES, 1 mM EDTA, and 400 mM NaCl, pH 6.7). The nonhybridized antisense RNA was digested with RNase A and T1. The samples were treated with SDS + proteinase K (0.6% and 290 µg/ml, respectively) for 15 min at 37°C followed by phenol:chloroform:isoamyl alcohol extraction and ethanol precipitation; these were resuspended in the 80% formamide, 10 mM Tris, and 10 mM EDTA, pH 7.0, heat denaturated, and electrophoresed on a polyacrylamide gel (5% polyacrylamide and 7% urea). Gels were exposed to film (Biomax, Kodak) and quantified by image analysis software (Imaging Research)
RNA Labeling
Total RNA (1060 µg, depending on type of experiments) was labeled with a corresponding fluorescent nucleotide (Cy3- or Cy5-labeled dUTP) as described elsewhere (6), with some modifications. In brief, total RNA was mixed with oligo dT18 primers (20 µM) and incubated at 72°C for 5 min and then quickly cooled on ice. In some experiments, before heating, control RNA from Escherichia coli with poly-A linker (pGIBS-Phe from ATCC) was added at molar ratios between 1:100 and 1:50,000 (assuming a 1.0-kb average length for mRNAs) to evaluate the efficiency of reverse transcription. Unlabeled nucleotides (dCTP, dATP, and dGTP) in final concentrations of 0.5 mM, 200 U of reverse transcriptase (RNase H-) and its corresponding buffer, and labeled dUTP in final concentration 0.2 mM were added. The reaction mixture was incubated for 5 min at 25°C and then for 1 h at 42°C in the dark; 200 U more of reverse transcriptase (RNase H-) was added, and the reaction continued for another 50 min. After incubation, the sample was heated at 95°C for 3 min to denature cDNA/mRNA hybrids and chilled on ice immediately. RNA was hydrolyzed by the addition of 5 µl of 0.5 M EDTA and 5 µl of 0.2 M NaOH, incubated at 60°C for 10 min, and cooled to room temperature; 5 µl of 1 M Tris·HCl (pH 7.5) and 5 µl of 0.2 N HCl were added to neutralize the reaction. The sample was purified from unlabeled nucleotide on a Sephadex G50 minicolumn, and the eluate was kept frozen at -20°C until hybridization.
Hybridization to Array
DNA microarray analysis of gene expression was done essentially as described (6) with some modifications. In brief, for each experiment, fluorescent cDNA probes were prepared from two samples and labeled with Cy5- or Cy3-dUTP. In this study we have used Cy5-dUTP to label samples from morphine-injected animals and Cy3-dUTP for samples from saline- or morphine+naloxone-injected mice, to enable comparison of the samples on the same slide. Cy3- and Cy5-labeled cDNAs were mixed and dried on Speedvac (Savant, Farmingdale, NY) and dissolved in hybridization buffer (Tris·HCl, 20 mM, 0.04% Denhardts solution, 0.02% salmon sperm DNA, 0.01% poly-A DNA, 20% dextran sulfate, 0.01% SDS, and 50% formamide). Prior to application to the array, the solution was heated at 85°C for 4 min, chilled on ice, and centrifuged at 14,000 g for 1 min. Hybridization was performed in a humidified chamber. Twenty microliters of hybridization solution was added to each DNA array and covered with a 22-mm coverslip. Slides were placed in a humidified chamber and incubated overnight at 48°C. Following hybridization, coverslips were removed in a glass staining dish containing 2x SSC with 0.1% SDS. Slides were washed twice for 4 min with the same solution, washed in 1x SSC for 10 min at 65°C, rinsed twice with 0.1x SSC for 2 min, rinsed with water twice for 1 min, and quickly dried by filtered air. Arrays were usually immediately scanned with a GMS 418 scanner (Affymetrix); however, we found that an array may be kept for 23 days in the dark without loss of fluorescence intensity.
Preparation of cDNA Array
cDNA fragment preparation.
The cDNA array contains 1,667 cDNA fragments and 61 spots used as a negative control (3 bacterial cDNA in duplicate and the spots with primers used for fragment amplification, buffer, and other solutions used in fragment preparation). Some cDNA fragments were different sequences of the same gene. About 1,100 of the cDNA fragments were prepared from a sequenced mouse brain cDNA library arrayed in 96-well plates (kindly provided by Dr. D. Beier, Genomic Center, Brigham and Womens Hospital). About 600 fragments were prepared by subtractive cloning (32). Using a representational difference analysis (17), we prepared libraries from cortex, striatum, medulla oblongata, spinal cord, and thalamus enriched in genes altered during tolerance and abstinence. The original methodology was modified to increase the probability of cloning genes that are slightly different in target and driver populations. Amplification of PCR fragments from the plasmid vector employed a T7 and SP6 primer pair or a T3 and T7 primer pair. The PCR reaction contained dNTP mix (1 mM), primer (0.5 µm),
12 ng of plasmid, and 1 U of Taq polymerase for 100 µl of reaction. Plasmids were amplified in a 96-well plate. Cycling protocol was as follows: 4 h 95°C-1 cycle, 45 min at 56°C; 2.5 h at 73°C, 50 min at 95°C; 38 cycles. The PCR reaction product was purified by ethanol precipitation, and the presence of fragments was verified by agarose gel electrophoresis. The concentration of fragments after PCR was 0.10.2 µg/µl.
cDNA arraying.
Amplified cDNA fragments were arrayed on a poly-L-lysine-coated slide with GMS 417 (Affymetrix). After arraying, slides were rehydrated, dried, UV cross-linked, and blocked as described (6). Arrays were kept at room temperature and used within 2 mo after preparation.
Data Analysis
Scanning parameters (light intensity and detection levels) were set such that the brightest spots are just below saturation. Images were analyzed with ArrayVision (Imaging Research), and fluorescence intensity values (along with quality control parameters) were stored in a database. Spots with obvious defects were tagged and excluded from subsequent analyses. Nontagged array elements were considered positive when the fluorescence intensity in each channel was greater than 1.93 times the local background (the threshold ratio for each slide was chosen such that negative control spots were at the background level). After background subtraction, the two channels were normalized to channel-specific median signal intensity. This approach compensates for possible biases in intensity resulting from inefficiencies in sample extraction and labeling and is less sensitive to outliers than normalization by total signal intensity. Fluorescence intensity values were log-transformed (base 2) and stored in a single table. The replicates of identical samples were analyzed for homogeneity (two-way nonparametric ANOVA by Friedman) and then averaged for regression analysis. The analysis of data was based on the combination of two procedures: regression and self-organizing map (SOM) clustering. This regression/clustering approach combines the statistical power of regression analysis and the exploratory strength of SOM-based clustering. Up- and downregulated genes were detected by regression analysis (18) with robust and resistant linear regression (MM-estimator) and simultaneous prediction confidence intervals (SPCIs) (20). The basic assumption of this model is that genes whose expression levels significantly deviate from the regression line (outliers) are the most likely to be up- or downregulated. The model is based on the fact that intensity values in two channels for a given set of spots are highly correlated (18, 26), if there are no print-tip and/or spatial effects (5). Since the standard error of replicates is smaller than intensity value range, we can use the regression analysis to estimate the confidence interval (4, 18). The MM-estimator was used to build linear regression, since an ordinary least-squared linear regression approach is sensitive to outliers itself. The SPCIs were used to provide a desirable confidence level across the whole range of intensity values and to estimate P values for up- and downregulated genes.
The results of regression analysis were combined with SOM clustering. SOM was applied by the weighted pair-group method with centered average and Pearson correlation implemented in the program xCluster (kindly provided by Gavin Sherlock, Stanford University). The clusters were modified with the subset of genes selected by regression analysis so that P values for the differentially expressed genes are less than or equal to 0.05, at least in one pair used to build the corresponding scatter plot for regression analysis.
To organize selected genes according to the intracellular pathways involved, we used a context-based information search and graphical tools (GeneSphere, Boston, MA).
 |
RESULTS
|
---|
Animal Experiment
Morphine analgesia is maximal at 30 min after injection and not significant at 2 h (Fig. 1). Coinjection of morphine with its antagonist naloxone did not produce any changes in withdrawal latency compared with control (Fig. 1). All morphine-injected mice (n = 6) were positive in the naloxone precipitated jumping test, and the number of jumps was 7.5 ± 1.2 jumps/15 min (mean ± SE), whereas none of the saline-injected (n = 4) or morphine+naloxone-injected (n = 4) mice were positive, indicating that morphine-treated mice developed acute dependence 2 h after morphine injection.

View larger version (18K):
[in this window]
[in a new window]
|
Fig. 1. Tail withdrawal latency. Mice were injected with morphine (10 mg/kg, n = 6), saline (n = 4), or morphine+naloxone (10 mg/kg, n = 4). Analgesia was measured by tail immersion test, with water temperature set at 52°C. Cutoff time was 15 s. *P < 0.05; data are means ± SE.
|
|
Array Characterization
To evaluate the properties of the array (for detailed description, see Ref. 32), we used total RNA pooled from three whole mouse brains. Using a positive control (pGIBS-Phe), we found that the sensitivity of the array was more than 1:50,000 (data not shown), which means our array can measure RNA expressed at the level of only 23 copies per cell. This experiment was done using 15 µg of total RNA; therefore, to measure the rare transcript present at 12 copies per cell requires about 2530 µg of total RNA per array. Therefore we pooled two medial striata (each yields about 1520 µg of total RNA) and two lumbar cords (each yields 2025 µg of total RNA) for each sample.
The use of amplified RNA (by reverse transcribing RNA using T7 polymerase promoter as a primer) may decrease the required amount of total RNA to 0.5 µg. However, it distorted the RNA expression profile: after amplification, the level of expression of at least 450 genes out of 1,667 was altered more than twofold. Amplification also affected linearity and decreased the dynamic range of detection (data not shown). Therefore, in this experiment we have not used amplification.
In a separate experiment, we tested the detection range of the array using different concentrations of pGIBS-Phe from E. coli with an attached poly-A tail. In a range of 1:100 to 1:50,000, the intensity of signal was linearly dependent on the amount of RNA used (r = 0.94, P < 0.05).
There is not yet a theoretically and experimentally verified best design for two-channel array experiments, particularly where multiple comparisons are involved. Currently there are two variants, one based on the cross-reference sample, which is labeled with one dye, while experimental and control samples are labeled with another dye, and a second design, that also includes two dyes, one for experimental and another for control samples. It seems that the first type of design poses significant difficulties for the analysis of nervous tissue in which different brain areas are involved. The great variability of gene expression between brain areas makes a reference sample practically useless (32). Therefore, in this study there were Cy5-labeled cDNA (in triplicate) from morphine-injected mice and Cy3-labeled cDNA from saline- and morphine+ naloxone-injected animals (in duplicate), for a total of 24 samples hybridized to 12 slides. Because this design does not automatically compensate for the difference between Cy3 and Cy5 properties, we independently normalized the data from the Cy5 and Cy3 channels prior to further analysis on the basis of median intensity. The replicates were homogenous according to two-way nonparametric ANOVA by Friedman and thus, for further analysis, were averaged. After removing all genes that lacked at least one value (to perform logarithmic transformation), the final set contained 522 rows and 10 columns corresponding to five groups (S30, N30, S120, M30, and M120) and two brain areas [5 from medial striatum (s) and 5 from lumbar part of the spinal cord (d)].
Regression Analysis
We used scatter plots and regression analysis to detect up- and downregulated genes in different groups within each brain area. The common practice of selecting up- and downregulated genes based on the predetermined fold differences in the expression assumes that variance of ratios is a constant for the microarray of interest. However, because of between-slide variability, a ratio value, e.g., 2, may be statistically significant for one slide and nonsignificant for another; therefore, the use of fold difference without appropriate statistical analysis may be misleading. One way to solve this problem is to collect more replicates and to apply ANOVA and a multiple comparison procedure. The successful use of replication and estimates for variation of intensity to reduce misclassification rates has been described in literature (5, 13, 16). MM-estimator, a robust and outlier-resistant method of regression analysis (see MATERIALS AND METHODS, Figs. 2 and 3) used in this study, is less sensitive to data distribution and more economical (compare with use of multiple replicates) but less accurate from a statistical point of view. Nevertheless, it enables us to determine the statistical significance of the observed differences in gene expression. Using the SPCIs (see MATERIALS AND METHODS), we were able to estimate the probability that a given gene is differentially expressed, i.e., that the gene is up- or downregulated. A total of 12 pairs were compared on the basis of three factors: time (30 and 120 min after injection), drug (S, M, and N groups), and brain area (s, medial striatum; and d, spinal cord; Fig. 2). Significance of regression coefficients varied for different scatter plot pairs but in all cases P < 0.0001 at least, except for scatter plot dM120 vs. dS120 where P = 0.0127.

View larger version (56K):
[in this window]
[in a new window]
|
Fig. 2. Regression analysis of gene expression. The regression analysis was performed using MM-estimator. Simultaneous prediction confidence intervals (SPCIs) were constructed as described in MATERIALS AND METHODS. Data are log2 values of fluorescent intensities. Blue dots indicate downregulated genes, and red dots indicate upregulated genes. For color coding of SPCI lines see Fig. 3. Significance of regression coefficients (a measure of signal-to-noise ratio) varied for different scatter plot pairs but in all cases P < 0.0001, except for dM120 vs. dS120 P = 0.0127. M, morphine; S, saline; N, morphine+naloxone; s, striatum; d, spinal cord; 30 and 120 indicate 30 and 120 min after injection, respectively.
|
|

View larger version (35K):
[in this window]
[in a new window]
|
Fig. 3. Regression analysis of gene expression with SPCIs. The regression graph from top left of Fig. 2 is zoomed to show different SPCIs. See Fig. 2 for further description. Blue dots indicate downregulated genes, and red dots indicate upregulated genes. The dot numbers correspond to "OrderN" in Fig. 4. Some numbers are omitted for clarity.
|
|
To select differentially expressed genes, we set a threshold of P
0.05. The 100 gene fragments that had such a P value in at least one pair used to build a scatter plot for regression analysis were selected (Fig. 4). The number of unique genes is fewer, since some fragments were different sequences of the same gene (see below). As expected, there is an inversely proportional correlation between the ratio magnitude and its statistical significance; i.e., the higher the ratio the less its P value. However, as mentioned above, there is no single threshold to apply to the entire data set. Thus, for gene g1194, the ratio = 2.68 from the scatter plot sM30 vs. sS30 and the ratio = 1.3 from the scatter plot sM30 vs. sN30 have P values in the same range (between 0.01 and 0.05), whereas for gene g1191 the ratio = 1.94 from the scatter plot dM30 vs. dN30 is more statistically significant (0.001 < P
0.01) than the ratio = 3.43 from the scatter plot sM30 vs. sS30 (0.01 < P
0.05). Moreover, the ratio = 0.72 (g1547, 0.01 < P
0.05; Fig. 4) is usually considered too small to be an indicator of alteration in gene expression, and the important difference may be lost. The approach based on MM-estimator and SPCIs is able to detect these genes as differentially expressed, since it takes into account variability characterized by residual error of linear regression and specific to a given comparison. The fact that 80% of all selected genes were significantly altered in more than one pair further confirms the exploratory power of the approach. However, the interpretation of the results even for this relatively simple experimental design is constrained by the fact that we did not take explicitly into account the variability between replicates in expression value specific for a given gene. We use a relatively high P value (P = 0.05) for gene selection; therefore, our analysis may have high misclassification rates (i.e., the relative numbers of false-positive cases may be high). However, it enables us to complement the regression analysis used to detect differentially expressed genes with cluster analysis (SOM), which by itself has no statistical value and often may lead to spurious conclusions.

View larger version (40K):
[in this window]
[in a new window]
|
Fig. 4. Table of the genes selected using regression clustering. Genes were included if at least one P value was <0.05. Data are presented as log2 of expression level ratio. Color indicates P value at which expression level is different between two samples on the basis of SPCI (see MATERIALS AND METHODS). "OrderN" corresponds to numbers on Fig. 3. See legend to Fig. 2 for description of abbreviations.
|
|
The same data set was analyzed using SOM (2 x 3) (because the higher dimension of SOM did not reveal any additional clusters, Fig. 5), which reveals the following three patterns of gene expression in both the medial striatum and the lumbar spinal cord: genes that downregulated following morphine administration (clusters C1A, C1B, and C1C), upregulated (cluster C2), and were essentially unchanged (clusters C0A and C0B). The patterns of alteration of gene expression are similar in the medial striatum and the lumbar part of the spinal cord; however, in the medial striatum the decrease of mRNA levels (cluster C1) is maximal at 30 min, whereas in the spinal cord the maximal effect of morphine is observed 2 h following injection.

View larger version (24K):
[in this window]
[in a new window]
|
Fig. 5. Cluster analysis. Data are means ± SE of all genes in the cluster. Clusters in the spinal cord and medial striatum were similar.
|
|
Such cluster analysis is an exploratory tool, which does not provide any P values associated with selected clusters but helps to detect more general regularities. To statistically validate the cluster analysis results, we superimposed the cluster and regression analysis results (see MATERIALS AND METHODS). The number of genes in each cluster was dramatically decreased by this approach, but the pattern of alteration remained the same in clusters C1 and C2 (Fig. 6). The few genes selected from cluster C0 most likely responded to the injection-associated stress, since expression was greater at 30 min than at 120 min after morphine and saline injection, or to action of morphine and naloxone not mediated by opioid receptor (Fig. 4). Thus further analysis was focused on clusters C1R and C2R.

View larger version (20K):
[in this window]
[in a new window]
|
Fig. 6. Combination of regression and clustering. The modified clusters marked by "R" are similar to the original but emphasize gene expression alterations. Data are means ± SE of all genes in the cluster. *P < 0.05 for the majority of genes in cluster compared either with S30 or S120.
|
|
The genes selected after regression clustering can be divided in two groups (Table 1): downregulated (C1R) and upregulated (C2R). Both of these effects of morphine were likely mediated by activation of opioid receptors, since no differences between saline and morphine+naloxone injection were observed (Figs. 5 and 6). Some of the genes are represented by several fragments (from 2 up to 6) that 1) are all selected and 2) belong to the same clusters. The gene for NADH dehydrogenase subunit 4 (ND4) is represented by six fragments that cover the entire sequence of the gene. The expression patterns of these fragments are similar (Fig. 7). The ND4 expression level decrease 30 min after morphine injection (P < 0.0001, by two-way nonparametric ANOVA with Friedman statistics = 22.51) is consistent with the result of regression analysis. The ND4 data set shows not only low variability between different spots on the array, but also the absence of bias toward the 3'-end of the transcript.

View larger version (20K):
[in this window]
[in a new window]
|
Fig. 7. Expression of NADH dehydrogenase (ND4) in the striatum. Each line represents the fluorescence intensity of each fragment in different samples. The numbers correspond to the sequences of the mitochondrial genome; avr, average intensity of all ND4 fragments. *P < 0.001 based on two-way nonparametric ANOVA by Friedman.
|
|
RNase Protection Assay
To provide independent verification of the array findings, the levels of four genes were analyzed by RNase protection assay in the medial striatum and the lumbar part of the spinal cord 30 min after morphine administration. The amplitude of the changes was not identical to the array data, but the direction of change remained the same (Fig. 8).

View larger version (31K):
[in this window]
[in a new window]
|
Fig. 8. Gene expression levels measured by RNase protection assay (RPA). The mice were injected with morphine (10 mg/kg, n = 4) or saline (n = 4) and killed 30 min later. The medial striatum and the spinal cord were collected as described in MATERIALS AND METHODS. RPAs were performed using 32P-labeled riboprobes incubated with 2 µg of total RNA. *P < 0.05 and **P < 0.01, based on regression analysis (SPCIs, see Fig. 4). #P < 0.05, based on Students t-test. First letter signifies brain area: s, medial striatum; d, spinal cord. Second letter signifies treatment: M, morphine (10 mg/kg), S, saline. B, background (riboprobes only). The riboprobes [ribosomal protein L23a (similar to X65228, bp 132342), apolipoprotein E (M12414, bp 89361), 1-tubulin (NM_011653, bp 9181225), H-2D(q) antigen (X52915, bp 18871991)] were transcribed from corresponding plasmids. The gels were exposed to film, and the images were quantified with image analysis software. Data are presented as an average ratio (log2) from four gels ± SE. Array data are presented as intensity ratio morphine/saline (see Fig. 4).
|
|
Functional Analysis
As a result of the combined application of regression and cluster analysis, we have identified a set of genes [38 known genes and 16 novel or expressed sequence tags (EST), Table 1] that are significantly altered following a single morphine injection. The genes with known function can be divided into three major groups: proteins involved in mitochondrial oxidation, cytoskeleton proteins, and regulatory proteins (Table 1). The first group consists of genes concerned with mitochondrial oxidation; these genes control expression of protein involved in electron transport and phosphorylation, glycolysis, and protein (BAP37) and fatty acid (FABP) transport into mitochondria (13 genes). A closely related group of genes controls proteins involved in sphingomyelin metabolism (ceramidase, saposin, Ggm2) and protein synthesis (ribosomal proteins L23a, L40, and 60S). The second group consists of gene for cytoskeletal proteins. These can be divided into three subgroups: actin-filament interacting proteins (
-actin, cofilin, SPARC, and csk, which negatively regulates actin filament formation), microtubule interacting proteins (tubulin alpha, stathmin), and a number of genes whose products are involved in secretory vesicle formation and cell adhesion (Table 1). We include in this group the sodium channel ß1-subunit, since it may function as an adhesion molecule (19). Based on this classification and analysis of available publications, we tried to reconstruct the pathway modulated by opioid receptor activation. There is no exact match with known intracellular pathways, and the closest match is a new "isotype" of integrin-activated skeletal protein regulatory pathway (Fig. 9).

View larger version (28K):
[in this window]
[in a new window]
|
Fig. 9. Intracellular pathway mediating the effect of morphine. For gene abbreviations, see Table 1. Red indicates upregulated genes, and blue indicates downregulated genes. Solid line, known effect; dotted lines, hypothetical interactions. OR, opioid receptor; Gi, inhibitory G protein; SNARE, soluble N-ethylmaleimide-sensitive factor attachment protein receptor; for all other definitions of protein abbreviations, see Table 1.
|
|
 |
DISCUSSION
|
---|
Gene expression profiling using DNA array is a powerful methodology with the potential for new discoveries as well as for great pitfalls. Experimental variability makes interpretation of the results and design of new experiments challenging. In this study we asked the relatively simple question of whether a single morphine injection produces significant changes in gene expression. The answer was definitively positive as confirmed by RNase protection assays of at least four genes. Moreover, these changes can be detected even in relatively big anatomical subdivisions of brain and spinal cord, which may contain thousand neuronal cell types.
However, the reliability of microarray data is still uncertain, and the interpretation of the results is confounded by potential methodological problems. Therefore, we used a new procedure based on regression analysis and clustering to analyze the gene expression data. The regression analysis provides a statistical approach to selecting up- and downregulated genes, whereas SOM-based cluster analysis adds the power of pattern recognition to select groups of coregulated genes. This hybrid procedure (clustering followed by regression analysis) seems to combine the advantages of both methods. However, this simple and empirically useful approach requires further theoretical analysis to understand its validity and limitations. Data preprocessing plays an important role in our approach: to safeguard against spurious results, we included in the analysis only spots that were positive in all samples. In this way useful information can be lost, since most variable genes in the low-intensity area are removed. A possible alternative of including in the analysis even highly variable genes (a compendium approach as described in Ref. 11) requires many more replicates and may not be practical for certain experiments. By using the hybrid approach (regression clustering), we detected two predictable effects of morphine on gene expression: inhibition and activation. Both of these effects are probably mediated by specific opioid receptors, since naloxone coinjection prevents them. The unexpected finding was that only 9 genes were upregulated and 45 genes were downregulated. Currently we do not have a simple explanation for this phenomenon. It might be a consequence of the general inhibitory effect of opioids on metabolic function, since it is known that morphine decreases temperature, respiration, heart rate, and heat production. Whether the alteration of these physiological parameters is directly connected to the metabolic rate of neurons is not known but seems unlikely, since neurons are generally protected from fluctuation of metabolic rates. The changes in mitochondrial protein expression may still be secondary to the general depressant effect of morphine on metabolism. The observed changes, however, were prevented by the opioid receptor antagonist naloxone and limited to the specific subset of genes that belong to distinct intracellular pathways. It was also reported that a single morphine injection decreased mRNA turnover in the brain but not in the liver (9). The relatively large number of altered genes is not an argument in favor of general metabolic inhibition, since part of the gene fragments on array were preselected based on their potential changes following morphine administration, and 75% of all altered genes belong to this group.
The activation of the opioid receptor inhibits cAMP levels and ion currents via pertussis toxin-sensitive and -insensitive Gi proteins. The decrease of cAMP and/or Ca2+ concentrations may inhibit gene transcription, but it is unclear how to explain the particular effect of morphine observed in this study. We do not know whether any agent that decreases the level of cAMP or Ca2+ will produce a similar effect, nor are we sure how information to induce a specific gene expression pattern in response to extracellular receptor activation is transmitted and coded. The ratio of cAMP, diacylglycerol, and ion concentrations may define the specific alterations in gene expression in response to opioid receptor activation. Yet, we cannot exclude other more sophisticated mechanisms. At this point there is not sufficient sequence information to decide whether the altered genes have similar promoter regions, but certainly the data might be reanalyzed at a later time.
Morphine decreases the expression of multiple genes following repeated administration (22), but there are limited data regarding its acute effect. Nevertheless, these data confirm morphines ability to alter mRNA levels even following a single injection (2, 3, 8, 14, 15, 33).
Functionally, the altered genes can be divided into three groups (Table 1). One includes proteins involved in mitochondrial functions: oxidative phosphorylation, glycolysis, and transport into mitochondria. Another group are the proteins involved in broadly defined cytoskeletal functions: synaptic vesicle formation from the Golgi complex and its fusion to membrane, axon growth, and adhesion. The third more heterogeneous group consists of regulatory proteins. The observed changes in gene expression are consistent with the fact that activation of the opioid receptor reduces neuromediator secretion, axon formation, and protein synthesis (9), consistent with proapoptotic and antiproliferative activity of opioids (10). It was also reported that even a single morphine injection inhibits axon regeneration (27) and diminishes the structural integrity of cytoplasm in the hypothalamus (31). Moreover, the incubation of opioids with kidney cells induces transitory reorganization of actin filaments (24) as early as 15 min after addition. There were no changes in the level of ß-actin mRNA, which is consistent with our data, but the levels of
-actin or cofilin were not measured (24).
It is less clear how changes in the changes in gene expression related to the analgesic, locomotor, and other pharmacological effects of opioids. However, this is beyond the scope of the current project and would require not only the measurement of the level of corresponding proteins but also intensive pharmacological analysis to confirm whether the changes in mRNA level are consequence or mediator of the opioid effects on neuronal activity.
It is accepted that these opioid effects are mediated by inhibition of neuromediator release via a decrease of cAMP concentration or Ca2+ current. However, downstream events are less clear. From our data the following downstream pathway is feasible: opioids attenuate the formation of synaptic vesicles from Golgi apparatus, prevent turnover of microtubules and actin filaments, and induce redistribution of adhesion proteins on the cell surface. In this way opioids may prevent the fusion of intracellular vesicles with the extracellular membrane and decrease the release of neurotransmitters. There are few data to support this hypothesis but at least one of the adhesion proteins, SPARC, potentiates morphines effects on locomotor activity following direct administration into the amygdala (12). In any way, the data indicate that expression of cytoskeletal and mitochondrial genes may be sensitive indicators of drug effects and may be used to find subtle distinctions among different opioid analogs and to detect the specific "isotypes" of regulatory pathways that mediate drug activity. The present findings also help to focus the future experiment on the new protein subset that would be difficult to select by other methods. It is noteworthy that observed alterations of gene expression were detected with use of a limited number of gene fragments that, however, were preselected according to their response to morphine administration.
The difference between the two areas studied was not dramatic. In part it may be explained by the fact that we measured alterations that occur simultaneously in many neurons. Further studies will show whether this is true for other brain areas and other drugs. In the striatum, however, the maximal effect was observed at 30 min after morphine administration vs. 2 h in the spinal cord. It is not clear why the effect of morphine occurs first in the striatum but lasts longer in the spinal cord (Figs. 5 and 6). There are no data indicating different pharmacokinetics in the brain and the spinal cord following systemic morphine injection. It remains to be proved that these lasting changes in the lumbar enlargement of the spinal cord are relevant to the development of acute dependence on or involved in the chronic effect of morphine.
In conclusion, the expression profile analysis proved to be able to detect changes in gene expression following single morphine injection. Moreover, these changes are mediated by a specific opioid receptor and might be useful in clarifying the effect of a known substance by providing insight into the regulatory pathways that otherwise evade detection. It seems that this approach may not only complement the pharmacological analysis of the drugs but may actually be a primary method to describe the properties of new substances, in particular those with neurotropic activity, and provide targets for traditional pharmacological and neurochemical experiments. However, to unleash its full potential will require further development of data analysis procedures and more comprehensive coverage of the genome.
 |
ACKNOWLEDGMENTS
|
---|
We are grateful to Dr. R. Pratt for helpful discussion during time course of this experiment and to Dr. G. Davar for critical comments.
This work was supported by grants from the Anesthesia Foundation of the University of Wisconsin to R. Y. Yukhananov and by National Institutes of Health Grant GM-42466 to G. J. Crosby.
 |
FOOTNOTES
|
---|
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).
Address for reprint requests and other correspondence: R. Y. Yukhananov, Anesthesia Research, MRB 611, Dept. of Anesthesia, Brigham and Womens Hospital, Harvard Medical School, 75 Francis St., Boston, MA 02115 (e-mail: ryyukhan{at}zeus.bwh.harvard.edu).
 |
REFERENCES
|
---|
-
Burgess LH and Handa RJ. Chronic estrogen-induced alterations in adrenocorticotropin and corticosterone secretion, and glucocorticoid receptor-mediated functions in female rats. Endocrinology 131: 12611269, 1992.[Abstract]
-
Chang SL, Squinto SP, and Harlan RE. Morphine activation of c-fos expression in rat brain. Biochem Biophys Res Commun 157: 698704, 1988.[ISI][Medline]
-
Ding XZ and Bayer BM. Increases of CCK mRNA and peptide in different brain areas following acute and chronic administration of morphine. Brain Res 625: 139144, 1993.[ISI][Medline]
-
Draper NR and Smith H. Applied Regression Analysis. New York: Wiley, 1998.
-
Dudoit S, Yang YH, Callow MJ, and Speed TP. Statistical methods for identifying differentially expressed genes in replicated cDNA microarray experiments [Online]. Technical report .Dept. of Statistics, UC Berkeley. http://www.stat.berkeley.edu/users/terry/zarray/html/matt.html [August 2000].
-
Eisen MB and Brown PO. DNA arrays for analysis of gene expression. Methods Enzymol 303: 179205, 1999.[ISI][Medline]
-
Franklin LBJ and Paxinos G. The Mouse Brain in Stereotaxic Coordinates. New York: Academic, 1996.
-
Gutstein HB, Thome JL, Fine JL, Watson SJ, and Akil H. Pattern of c-fos mRNA induction in rat brain by acute morphine. Can J Physiol Pharmacol 76: 294303, 1998.[ISI][Medline]
-
Harris RA, Harris LS, and Dunn A. Effect of narcotic drugs on ribonucleic acid and nucleotide metabolism in mouse brain. J Pharmacol Exp Ther 192: 280287, 1975.[Abstract]
-
Hauser KF and Mangoura D. Diversity of the endogenous opioid system in development. Novel signal transduction translates multiple extracellular signals into neural cell growth and differentiation. Perspect Dev Neurobiol 5: 437449, 1998.[ISI][Medline]
-
Hughes TR, Marton MJ, Jones AR, Roberts CJ, Stoughton R, Armour CD, Bennett HA, Coffey E, Dai HY, He YD, Kidd MJ, King AM, Meyer MR, Slade D, Lum PY, Stepaniants SB, Shoemaker DD, Gachotte D, Chakraburtty K, Simon J, Bard M, and Friend SH. Functional discovery via a compendium of expression profiles. Cell 102: 109126, 2000.[ISI][Medline]
-
Ikemoto M, Takita M, Imamura T, and Inoue K. Increased sensitivity to the stimulant effects of morphine conferred by anti-adhesive glycoprotein SPARC in amygdala. Nat Med 6: 910915, 2000.[ISI][Medline]
-
Kerr MK, Martin M, and Churchill GA. Analysis of variances for gene expression microarray. J Computational Biol 7: 819837, 2000. [Technical report available online at http://www.jax. org/research/churchill/pubs/index.html].[ISI]
-
Kluttz BW, Vrana KE, Dworkin SI, and Childers SR. Effects of morphine on forskolin-stimulated pro-enkephalin mRNA levels in rat striatum: a model for acute and chronic opioid actions in brain. Mol Brain Res 32: 313320, 1995.[ISI][Medline]
-
Le Greves P, Huang W, Zhou Q, Thornwall M, and Nyberg F. Acute effects of morphine on the expression of mRNAs for NMDA receptor subunits in the rat hippocampus, hypothalamus and spinal cord. Eur J Pharmacol 341: 161164, 1998.[ISI][Medline]
-
Lee MT, Kuo FC, Whitmore GA, and Sklar J. Importance of replication in microarray gene expression studies: statistical methods and evidence from repetitive cDNA hybridizations. Proc Natl Acad Sci USA 97: 98349839, 2000.[Abstract/Free Full Text]
-
Lisitsyn NA. Representational difference analysis: finding the differences between genomes. Trends Genet 11: 303307, 1995.[ISI][Medline]
-
Loguinov AV, Yukhananov RY, Mian SI, and Vulpe C. Using robust and resistant regression analysis (MM-estimator) to find differentially expressed genes in microarray data. In: Proc MSRI Workshop On Nonlinear Estimation and Classification [Online]. Berkeley, CA: Mathematical Research Sciences Institute. http://www.msri.org/publications/ln/msri/2001/nle/loguinov/1/index.html [3 May 2001].
-
Malhotra JD, Kazen-Gillespie K, Hortsch M, and Isom LL. Sodium channel beta subunits mediate homophilic cell adhesion and recruit ankyrin to points of cell-cell contact. J Biol Chem 275: 1138311388, 2000.[Abstract/Free Full Text]
-
MathSoft. S-Plus 2000: Guide to Statistics. Seattle, WA: MathSoft, 1999, p. 255298.
-
Meikle AD and Martin AH. A rapid method for removal of the spinal cord. Stain Technol 56: 235237, 1981.[ISI][Medline]
-
Nestler EJ and Aghajanian GK. Molecular and cellular basis of addiction. Science 278: 5863, 1997.[Abstract/Free Full Text]
-
Nestler EJ, Hope BT, and Widnell KL. Drug addiction: a model for the molecular basis of neural plasticity. Neuron 11: 9951006, 1993.[ISI][Medline]
-
Papakonstanti EA, Bakogeorgou E, Castanas E, Emmanouel DS, Hartig R, and Stournaras C. Early alterations of actin cytoskeleton in OK cells by opioids. J Cell Biochem 70: 6069, 1998.[ISI][Medline]
-
Przewlocka B, Lason W, and Przewlocki R. The effect of chronic morphine and cocaine administration on the Gs and Go protein messenger RNA levels in the rat hippocampus. Neuroscience 63: 11111116, 1994.[ISI][Medline]
-
Sapir M and Churchill GA. Estimating the posterior probability of differential expression from microarray data [Online]. Technical report. Bar Harbor, ME: Jackson Laboratory. http://www.jax.org/research/churchill/pubs/index.html [2000].
-
Sinatra RS, Ford DH, and Rhines RK. The effects of acute morphine treatment on the incorporation of [3H]L-lysine by normal and regenerating facial nucleus neurons. Brain Res 171: 307317, 1979.[ISI][Medline]
-
Sofuoglu M, Sato J, and Takemori AE. Maintenance of morphine dependence by naloxone in acutely dependent mice. J Pharmacol Exp Ther 254: 841846, 1990.[Abstract]
-
Sofuoglu M and Takemori AE. Naloxone produces protracted tolerance in mice treated with a single dose of morphine. J Pharmacol Exp Ther 254: 836840, 1990.[Abstract]
-
Tirumalai PS and Howells RD. Regulation of calbindin-D28K gene expression in response to acute and chronic morphine administration. Brain Res Mol Brain Res 23: 144150, 1994.[ISI][Medline]
-
Vaskina LV and Pushkin AS. Dynamics of changes in the ultrastructure of hypothalamic nerve cells in acute morphine poisoning. Zh Nevropatol Psikhiatr Im S S Korsakova 78: 980984, 1978.[ISI][Medline]
-
Yukhananov RY. Variability of gene expression in the brain. DNA 2000 Abstract . Boston, MA: 2000, p. 186.
-
Yukhananov RY and Handa RJ. Effect of morphine on proenkephalin gene expression in the rat brain. Brain Res Bull 43: 349356, 1997.[ISI][Medline]
-
Yukhananov RY, Klodt PM, IlIna AD, Zaitsev SV, and Maisky AI. Opiate withdrawal intensity correlates with the presence of DSLET high-affinity binding. Pharmacol Biochem Behav 49: 11091112, 1994.[ISI][Medline]
-
Yukhananov RY, Zhai QZ, Persson S, Post C, and Nyberg F. Chronic administration of morphine decreases level of dynorphin A in the rat nucleus accumbens. Neuropharmacology 32: 703709, 1993.[ISI][Medline]