Materials and Methods

Study Population.

With informed consent, peripheral blood anticoagulated with heparin was obtained by venipuncture from 10 patients (8 females, 2 males, mean age 42 years, standard deviation = 10.8 years) with relapsing-remitting multiple sclerosis (MS) and 10 apparently healthy age- and sex-matched controls (mean age 41.8 years, standard deviation = 11 years). Additional samples were obtained from the MS patients 24 h after i.m. administration of 30 μg of interferon(IFN)-β1a, an approved drug. Thus, a total of 30 separate samples were analyzed using the method.

Total RNA Preparation.

Peripheral blood mononuclear cells were isolated from the anticoagulated blood within 4 h of collection using the Hypaque-Ficoll method (Histopaque reagent; Sigma Chemical, St. Louis, MO). Monocytes were depleted from the peripheral blood mononuclear cells by the plastic adhesion method. Total RNA was prepared from the monocyte-depleted peripheral blood mononuclear cells using the TRI reagent method (Molecular Research Center Inc., Cincinnati, OH) (Chomczynski and Mackey, 1995). The TRI reagent is the improved ready-to-use version of the popular single-step method of total RNA isolation (Chomczynski and Sacchi, 1987; Chomczynski, 1993) that allows processing of a large number of samples for the isolation of total RNA or the simultaneous isolation of RNA, DNA, and proteins from diverse biological samples. The TRI reagent was used according to the manufacturer's recommended protocol.

DNA Array Protocol.

The DNA arrays used were GeneFilters GF211 (Research Genetics Inc., Huntsville, AL), which contain immobilized PCR fragments (typically, around 1000 bases long) from the 3′-untranslated region of sequence verified IMAGE/LLNL cDNA clones of named human genes. Briefly, 5 μg of total RNA were radioactively labeled with [³³P]CTP using reverse transcriptase and oligo-dT primers. The labeled cDNA was used to probe the GF211 membrane. The membranes were washed, and the bound radioactivity was visualized using a Cyclone phosphorimager (Packard Instrument Co., Meriden, CT). The manufacturer's protocols were used essentially as recommended (for detailed protocol see http://www.resgen.com/).

Data Analysis.

The TIFF images from the phosphorimager were imported directly into Pathways software program (Research Genetics Inc., Huntsville, AL). The images were aligned, gridded, and quantitated according to recommended procedures. Filters were normalized using the intensity from all spots, and the software also normalizes for intensity ranges in two-filter comparisons. Scripts called Path files were specifically set up to facilitate extraction of the normalized data corresponding to the CYPs, other drug-metabolizing enzymes, and the transporters. The data were exported to an Excel spreadsheet (Microsoft Corp., Bellevue, WA) for further analysis.

Statistical analysis was carried out using SPSS 6.0 (SPSS Inc., Chicago, IL) and Excel. The t test was used for comparisons involving the control and MS patient groups, whereas the pairedt test was used for comparisons involving MS patients before and after treatment with IFN-β1a. Linear regression and the Pearson product moment regression coefficient (r) were calculated using Kaleidagraph (Synergy Software, Reading, PA). The Spearman rank correlation coefficient was computed using SPSS 6.0. Pvalues are directly reported, and a value of 0.05 was used to determine significance.

PCR.

A total of six mRNA samples, three from MS patients and three from healthy controls were analyzed using PCR.

The reverse transcription conditions were similar to those used for labeling mRNA for DNA arrays except that [³³P]CTP was not included. The PCR conditions were derived from Baron et al., 1998. The following splice junction spanning primers were used to minimize the amplification of any contaminating genomic DNA (Baron et al., 1998). The sense (S) and antisense primers (AS) were: CYP 1B1-S, 5′-GTA TAT TGT TGA AGA GAC AG-3′ and CYP 1B1-AS, 5′-AAA GAG GTA CAA CAT CAC CT-3′, 316-base pair product; 2E1-S, 5′-AGC ACA ACT CTG AGA TAT GG-3′ and CYP 2E1-AS, 5′-ATA GTC ACT GTA CTT GAA CT-3′, 366-base pair product; 2B6/7-S, 5′-CCA TAC ACA GAG GCA GTC AT-3′ and CYP 2B6/7-AS, 5′-GGT GTC AGA TCG ATG TCT TC-3′, 377-base pair product; β-Actin-S, 5′-ACC CAC ACT GTG CCC ATC TA-3′ and β-Actin-AS, 5′-CGG AAC CGC TCA TTG CC-3′, 290-base pair product. The PCR conditions were 35 cycles with 1 min of annealing at 56°C, 2 min of extension at 72°C, and 1 min of denaturation at 93°C, with 5 min extension. The PCR products were separated on 1% agarose gels. The amplification of CYP 1A1 was according to Vanden Heuval et al. (1993). The primers were CYP 1A1-S, 5′-TAG ACA CTG ATC TGG CTG CAG-3′ and CYP 1A1-AS, 5′-GGG AAG GCT CCA TCA GCA TC-3′, 148-base pair product. The PCR conditions were 30 cycles with 30 s of annealing at 54°C, 1 min of extension at 72°C, and 15 s of denaturation at 94°C. The PCR products were separated on 3% agarose gels.

Results

Overview of mRNA Expression on DNA Arrays.

The distribution of normalized intensities was examined to determine whether the healthy controls and the MS patients differed in their global patterns of mRNA patterns. Because DNA arrays provide measurements of mRNA levels in arbitrary optical units, an important additional goal of the analysis was to provide an internal reference scale against which the expression of the drug-metabolizing enzyme mRNAs could be compared. The mean background was 19 arbitrary units (range 10–53).

The distribution of normalized intensities of the spots obtained using DNA arrays is shown on probability paper in Fig.1 for two representative controls (solid lines) and two patients (dashed lines). The data in Fig. 1 show that the overall distribution of the mRNA intensity distribution was similar in patients and healthy controls. The similarity of the medians of two distributions demonstrates that the normalization technique corrects adequately for differences in labeling intensities and loading. The steep initial slope shows that distributions are strongly skewed toward lower normalized intensities: 60 to 70% of the spots had normalized intensities less than 1,000 arbitrary units. Approximately 10% of the spots had intensities greater than 5,000 units and 5% of the spots had intensities greater than 10,000 arbitrary units.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

The distribution of normalized intensities of the spots obtained using DNA arrays is shown on probability paper for two representative controls (solid lines) and two patients (dashed lines).

CYP Expression.

The normalized intensity of drug-metabolizing CYP mRNA expression in peripheral blood in the 10 samples from MS patients and 10 controls is summarized in Table 1.

The data in Table 1 are sorted so that the magnitude of the mean decreases down the table. The signals corresponding to CYP 4A11, CYP 2J2 and CYP2E1 were strong, whereas the other CYPs were weaker by comparison. In the controls, the coefficient of variation ranged from 18% for CYP 2J2 to 114% for CYP 2A6. There were no statistically significant differences (P ≤ .05) in CYP expression between MS patients and controls and in MS patients before and after IFN treatment.

Because IFN treatment did not have a statistically significant effect on CYP expression, we assessed the intraindividual variability involved with CYP detection using DNA arrays by graphing the normalized intensities of CYP 4A11 and CYP 2J2 in each MS patient before and after IFN treatment (Fig. 2). For CYP 4A11, the average change was 17% and only 2 of 10 patients showed percent changes greater than 30%. For CYP 2J2, the average change was 12% and 3 of 10 patients showed percent changes greater than 30%.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

The normalized intensities for CYP 4A11 (A) and CYP 2J2 (B) in individual multiple sclerosis patients before and after treatment with interferon.

The results demonstrate the sensitivity of the method and the feasibility of detecting the expression of multiple CYPs from single peripheral blood samples. Additionally, they suggest the possibility that the technique could be used to assess interindividual variability.

Expression of Other Drug-Metabolizing Enzymes.

The signals corresponding to several drug-metabolizing enzymes are summarized in Table 2. The uridine diphosphate glucuronosyl transferases (UGTs) are an important class of Phase II conjugating enzyme and probes corresponding to three isozymes, UGT 2B4, UGT 2B10, and UGT 2B15 were available on the array. The strongest signal corresponded to UGT 2B10.

We also examined a range of other transferase enzymes such as catecholamine-O-methyltransferase (COMT), thiopurineS-methyltransferase, and the steroid sulfotransferases, DHEA-preferring sulfotransferase and hydroxysteroid sulfotransferase. Robust signals corresponding to COMT and the two steroid sulfotransferases were detected (Table 2). No statistically significant differences in the expression of the Phase II enzyme mRNAs were found between MS and control groups. However, expression of hydroxysteroid sulfotransferase was reduced upon IFN treatment (P = .05) in MS patients.

Several glutathione S-transferase (GST) probes were also immobilized on the DNA array, and we were able to examine the expression of several isozymes (Table 2). The signals corresponding to GST M4, GST A3, and a GST homolog were particularly notable. The GST signal strengths ranged over a greater than 200-fold range of intensity depending on the isozyme. These findings support the premise that considerable interisozyme selectivity is potentially achievable using the method.

Expression of Transporters.

Because transporter activity contributes significantly to the clearance of many drugs as well to the emergence of drug resistance, we examined the expression of transporter mRNAs using the method in Table3. Probes corresponding to five proteins, MDR1, MDR3, MRP1, MRP3, and MRP5, which have been linked to multidrug resistance, were available. A strong signal for MRP1 was consistently detected.

Probes corresponding to a wide range of transporters were available on the DNA array used and only a subset of these are presented in Table 3. The signals corresponding to several transporters of potential pharmaceutical interest, e.g., transporters involved in creatinine, betaine, monocarboxylic acid, and nucleoside transport were readily and consistently detected. There were no statistically significant differences in the expression of transporter mRNAs between MS patients and controls. However, the IFN-treated MS patient group showed statistically significant changes in the expression of mRNA for the creatine transporter.

Comparison of DNA Array Data to Hepatic Protein Levels.

We compared the peripheral blood CYP expression levels obtained using DNA arrays to the values for CYP expression in the liver previously reported by Shimada et al. (1994). These authors examined the levels of CYPs 1A2, 2A6, 2B6, 2C, 2D6, 2E1, and 3A4 using immunochemical methods. The DNA arrays used provided corresponding mRNA levels for 2A6, 2B6, and 2E1. Figure 3 plots the mean values for the normalized intensity of the CYP spots from the DNA array (n = 20 because controls and patients were included) against the immunochemical measure of CYP protein level from Shimada et al. (1994). The r value of the linear regression line was 0.89, and the correlation achieved a P value of .15. There was exact correspondence between the rank orders for the three CYPs among the two methods (Spearman r = 1.00). This data supports the premise that DNA arrays measurement in peripheral blood may prove useful for assessing expression of certain CYPs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

The mean concentration of the CYP in human hepatic microsomes obtained from Shimada et al. (1994) versus the mean normalized intensity of the same CYPs in peripheral blood mononuclear cells obtained using DNA arrays.

The solid line represents the best fit line through the mean values. The regression coefficient and the CYP isozymes are indicated.

In Fig. 4, we compare Phase II enzyme levels from DNA arrays to the activities of five Phase II enzymes reported by Iyer and Sinz (1999) in human liver fractions. Iyer and Sinz examined six Phase II enzyme activities: GST, UDP glycosyltransferase, sulfotransferase, N-acetyl transferase, thiopurine methyl transferase, and COMT. We summed the normalized signals of the spots on the DNA array corresponding to these activities (the extracellular matrix sulfotransferases were excluded from the analysis for sulfotransferase; protein N-acetyl transferases were excluded for N-acetyl transferase) and plotted the data against the activities reported by Iyer and Sinz. The UDP glucuronsyl transferase activities did not correlate between the two methods, but the r value for the regression line for the five remaining enzymes was excellent (r = 0.99), and the correlation achieved a P value of .002. The rank order correspondence for these five Phase II enzymes was exact (Spearmanr = 1.00).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4

The mean activity of the Phase II drug-metabolizing enzymes in human liver fractions were obtained fromIyer and Sinz (1999) versus the mean normalized intensity of the spots corresponding to the same activities in peripheral blood mononuclear cells obtained using DNA arrays.

The solid line represents the best fit line through the mean values. The regression coefficient and the activities are indicated.

Confirmation Using Reverse Transcriptase (RT)-PCR.

We used RT-PCR to qualitatively confirm the presence of several CYPs in a representative subset of three MS patient and three control mRNA preparations. Figure 5 shows the PCR products using primer pairs for CYPs 2E1, 1B1, 1A1, 2B6/7, and actin. For each of these CYPs, products of the expected length were observed demonstrating that the mRNA is present in each sample examined.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5

The results using RT-PCR.

Three patients samples (lanes P1, P2, P3) and three control samples (lanes C1, C2, C3) were analyzed using primers specific for β-actin (A), CYP 2E1 (B), CYP 1B1 (C), CYP 1A1 (D), and CYP 2B6 (E). A band corresponding to P3 was clearly noted in the gel for panel E but is not evident in this image. The arrowheads mark the fragment corresponding to the CYP. The lane marked M is a molecular weight marker containing 100-base pair ladder and the lines to the side of each gel align with 100-, 500-, and 1000-base pair markers.

Discussion

In this report, we have presented evidence that a subset of CYPs is expressed in peripheral blood cells and can be identified using DNA arrays. A wide range of drug-metabolizing enzymes and transporter signals were also detected using the technique. In Fig. 3, we demonstrated that the rank order of mean values of normalized intensity from the DNA array matched the rank order of mean protein expression in human liver microsomes for three CYPs, and in Fig. 4, we extended the results to demonstrate that the mean activity of five (of a possible six) Phase II drug-metabolizing enzymes was strongly correlated with the mean normalized intensity from DNA arrays. The strength of the correlation between the two techniques in Fig. 4 was unexpected and the mechanistic basis for the correlation is currently unclear. There were no significant differences between controls and MS patients, and IFN treatment altered mRNA levels of only a few enzymes/transporters.

The findings on CYP expression complement other reports that have examined CYP expression in peripheral blood using the RT-PCR. Raunio et al. (1998) found CYPs 1B1, 2B6/7, 2C, 2E1, 2F1, 3A5, and 4B1 in bronchoalveolar macrophages; CYPs 1B1 and 2E1 mRNA were consistently observed in peripheral blood samples, but CYPs 2B6/7, 2C8–19, 3A5, 4B1 were expressed in some samples. Dassi et al. (1998) reported CYP1B1 in mononuclear cells; Baron et al. (1998) reported CYPs 1B1, 2E1, and 2B6/7 in human monocytes; and Vanden Heuvel et al. (1993) were able to quantitate CYP 1A1 mRNA levels in human blood lymphocytes. Janardan et al. (1996) demonstrated the presence of CYP 3A5 mRNA and a protein that was recognized by an anti-CYP3A polyclonal antibody but failed to detect CYP3A5 activity in peripheral blood cells. Using DNA arrays, we were able to examine a wider range of CYPs than these previous studies.

The possibility that the pharmaceutically important CYP 3A family of isozymes can also be estimated from peripheral blood with some experimental manipulation is suggested by Baron et al. (1998) who were able induce 3A3/4 in monocytes in vitro using inducers such as cyclosporin A, phenobarbital, and benzanthracene. Additionally, Vanden Heuvel et al. (1993) were able to measure ethoxyresorufin-O-dethylase activity in human peripheral blood cells and were also able to induce the expression of the CYP mRNA with 2,3,7,8-tetrachlorodibenzo-p-dioxin. Using immunochemical methods on human peripheral blood and lymph nodes,Sempoux et al. (1999) demonstrated the presence of CYP 3A proteins on B cells but not T cells; the isozymes present were not defined.

There are many legitimate arguments to justify both the existence and the absence of correlations between mRNA levels and protein levels. The primary determinants of mRNA levels are the transcription rate and the half-life of mRNA, whereas the protein level depends on the mRNA levels, the translational rate constant, and the half-life of protein. According to the Hargrove-Schmidt model, mRNA level will be proportional to the level of the corresponding protein at steady state (Hargrove and Schmidt, 1989; Hargrove et al., 1990; Ramanathan et al., 1993). Thus, for a family of closely related, constitutively expressed proteins, the combined effects of mRNA and protein half-lives and translational constant may be sufficiently close to provide strong correlations for different mRNA-protein pairs. However, because the transcription rates and the half-lives of individual mRNAs and proteins vary considerably, this does not necessarily imply a strong correlation in expression of a random selection of mRNA-protein pairs at steady state. Differences in mRNA and protein half-lives are also likely to cause poor correlations if mRNA-protein levels are monitored under transient conditions that deviate significantly from steady state. If the rate of protein production from translation is relatively small compared with the total size of protein pool, the correlation between protein and mRNA is likely to be poor because large changes in mRNA levels may cause only small changes in protein levels. Likewise, if only a small fraction of the mRNA pool effectively contributes to protein levels, the correlation is likely to be poor as well.

The strength of the hybridization signal from a DNA array depends on a variety of factors in addition to the amount of specific mRNA in a sample. The efficiency of labeling and hybridization are two key variables. The labeling efficiency is determined by the guanosine content of the mRNA sequence because cytidine was used for labeling the cDNA and the hybridization efficiency of the primer. We used oligo-dT primers containing a mixture of 10- to 20-mers instead of random hexamers, and this may reduce variations due to primer hybridization efficiency. Usually, the length of a poly(A) tail is around 200 residues (Lewin, 1995), and the affinity of primer binding can reasonably be expected to be similar across mRNAs if the length of the tail exceeds the length of the primer and it is not extensively involved in secondary structures; protein binding is not likely to be an issue because of the mRNA isolation method. Unlike oligonucleotide arrays, the immobilized cDNA probes on the arrays used are PCR fragments that, typically, are around 1000 bases long, and the quantitation is less likely to be sensitive to differences in hybridization unless the probes are severely compromised during immobilization. Long probes, however, are more prone to specificity and selectivity problems than short probes or oligonucleotides.

On the methodological front, it is important to note that the absence of a signal on a DNA array does not always imply that the mRNA is not expressed. It merely suggests that optimization of the sequence of the immobilized probe is necessary. For clinical and diagnostic applications involving polymorphic isozymes, the optimization process must necessarily include both signal strength and selectivity considerations. Analogously, because long probes were used in our DNA arrays, additional research to definitively exclude the possibility that mRNAs with high sequence similarity are responsible for some signals is needed and is underway.

The DNA arrays that are currently used for measuring mRNA expression cannot discriminate between active and null variants of various drug-metabolizing enzymes. However, simultaneous analysis of the DNA and mRNA for certain well known CYPs using separate array-based systems is certainly feasible. Thus, in theory, it is possible to obtain both drug-metabolizing enzyme expression level and genotype using arrays.

There are well known interindividual differences in CYP and drug-metabolizing enzyme levels, and we emphasize that the correlations in Figs. 3 and 4, which use average values, should not accidentally be misconstrued as demonstrating that individual-specific predictions are feasible at this time. Clearly, for individual-specific predictions, a higher standard of proof, namely, demonstration of strong correlation on a sample-by-sample basis is required. The experiments in this report were not intended to test this hypothesis.

Using a more specific polyclonal antibody for CYP 2B6, Stresser and Kupfer (1999) suggested that initial estimates of CYP 2B6 levels such as those used in Fig. 3 may have been lower than the true hepatic values. There is no doubt that further validation of the significance of the correlation in Fig. 3 is needed and that any such validation would also benefit from simultaneous measurements of multiple CYPs in individual samples using the most specific, highest affinity antibodies. Subjectively, from the position of the Fig. 3 regression line, it appears the DNA arrays appear to provide an estimate that is higher than that reported by Shimada et al. (1994). Both the comparison and the correlation in Fig. 3 are weakened because of the small number of CYPs for which the protein levels are available and because the literature values are variable. Concomitant measurements of mRNA (using DNA arrays) and protein levels (using immunochemical methods) on the same blood samples are planned to directly test the signficance of the correlation.

In conclusion, DNA arrays offer the potential for revolutionizing drug development because they are powerful tools for pharmacogenomic, drug mechanism, pharmacodynamic, and drug toxicity studies. However, further validation of the selectivity and the quantitative capabilities of the technique are required before it can be used in a clinical setting for patient care.