The Roche PA-1 oligonucleotide array contained 66,176 features that correspond to 517 inflammation-related genes and was fabricated by Affymetrix. Among these, 250 genes were of human origin whereas the others represented mouse and rat genes. Each gene is monitored by 64 pairs of features that consist of 25-mer oligonucleotides, one perfectly complementary to the gene sequence and the other identical except for a single mismatch located in the center of the oligonucleotide.

The Specificity and sensitivity of the current type of oligonucleotide array used are based on several criteria. First, the design of the antisense oligonucleotides was targeted toward a well annotated full-length cDNA sequence (without introns) for each gene. Second, the number of both perfectly and mismatching oligonucleotides (features) for each sequence was 64. When directly converted, the 32 specific oligonucleotides (25-mers) overlap 800 bp of sequence data. Third, for this preliminary screen the amount of mRNA used for in vitro transcription reaction was rather high, 1 μg. In principle this amount should represent even the low abundant messages needed for difference detection [16,17,18]. Lastly, the hybridization mixture was used only once and discarded. However, from our earlier studies [18] it was known that for abundant messages (such as IFN-γ) or very rare transcripts (IL-4) the linearity of detection in this type of hybridization-based method is largely compromised (10² to 10³). We therefore have chosen a real-time RT-PCR method that facilitates quantitation of the transcripts over at least five orders of magnitude [19]. To confirm the linearity of the detection for each gene dilution curves were created (see Materials and methods). In addition, on the basis of our earlier work [20], we knew that the differences in the expression for several of the key genes in effector type 1 and type 2 cells would be very small. To minimize individual variation the oligonucleotide array was first hybridized twice with a type 1 or type 2 sample that had been previously validated with known marker genes. A second aliquot of the same mRNA was then independently prepared and hybridized to a new oligonucleotide array manufactured within the same batch. After this, the analyzed data was confirmed in samples from other individuals with a real-time RT-PCR method.

The expression of a total of 250 human inflammation-related genes was screened with an oligonucleotide array by performing two independent hybridizations with a Th1 or Th2 cell sample. The phenotype of these cells had been determined using ELISA as type 1 or type 2 (secretion of high IFN-γ, low IL-4, or high IL-4, low IFN-γ, respectively, data not shown). The differentially expressed human RNA transcripts of Th1 and Th2 cells are shown in Figures 1 and 2. The level of expression is shown as the average fluorescence intensity value for each gene in each sample. Collectively, 34 transcripts were recorded as differentially expressed in these cells (14 in type 1 and 20 in type 2 cells). The transcript of a proinflammatory cytokine INF-γ is abundant in type 1 T cells and very low in type 2 cells (Figure 1a). IFN-γ mRNA had routinely been measured with real-time RT-PCR from our cell cultures and the actual difference between the type 1 and type 2 cells was in the order of 10³- to 10⁵-fold. An oligonucleotide array, however, indicated less than 10²-fold difference in type 1 and type 2 cells (78-fold, calculated from average fluorescence intensity values for IFN-γ: Th1 1567.8, SD 129.1 and Th2 < 20, SD 3.3). In clear contrast to IFN-γ are the transcripts of the IL-4/IL-13 cluster known to promote humoral type 2 responses. Despite their all (IL-4, IL-5 and IL-13) being detected as differentially expressed in T-cell samples by an oligonucleotide array (Figure 2b), reliable detection of both IL-4 and IL-5 failed in a customized RT-PCR setup, probably because of their low level of expression.

The other transcripts expressed at a higher level by Th1 than Th2 cells are presented in Figure 1a,b. In addition to IFN-γ, these include IL-8, tumor necrosis factor-α (TNF-α) and granzyme B (fluorescence intensity > 400) as well as granulocyte-macrophage colony-stimulating factor (GM-CSF), macrophage inhibitory protein-1α (MIP-1α), MIP-1β, RANTES, IL-7 receptor (IL-7R), IL-12Rβ2, SLAM, CCR1, CCR2, CCR5, IκB, TIMP-1, c-Jun, IRF-1, caspase-1 and clusterin (fluorescence intensity < 400). Transcripts of MIF, IL-4R and STAT4 gave the highest relative fluorescence intensity values (> 300) in Th2 samples and their relative expression is shown as a separate graph (Figure 2a). The fluorescence values for all other genes that were expressed at a higher level in Th2 cells compared to Th1 cells were lower (< 300; Figure 2b). In Th2 cells, preferentially expressed genes include in addition to IL-4, IL-5 and IL-13, the transcripts for IFN-αβR, IFN-γRβ, fibroblast growth factor receptor (FGFR), CCR4, Smad2, BAX, CAS and caspase-6.

Hybridizations of oligonucleotide arrays were used to screen a panel of genes in T cells that are differentially expressed by polarized Th1 or Th2 cells (Figures 1,2). To validate the observations, the same Th1 and Th2 samples were quantitated together with samples of cultured Th1/Th2 cells originating from other individuals by using a real-time RT-PCR (TaqMan) method. We designed probe and primer sets for the quantitative detection of mRNAs for 14 genes (Table 1; caspase-1, CCR1, CCR2, CCR4, CCR5, clusterin, FGFR, GM-CSF, IκB, IL-4R, RANTES, STAT4, TIMP-1 and TNF-α). In addition, we had earlier confirmed a differential expression of SLAM, IFN-γRβ and IL-12Rβ2 transcripts in these cells [20]. All TaqMan RT-PCR measurements were carried out for each gene in duplicate. After three individual measurements the fold change in mRNA expression in Th1 cells compared to Th2 cells was calculated (Table 2). The overall transcriptional preferences recorded with a real-time RT-PCR method correspond well with the differences recorded by the oligonucleotide microarrays. The measurements from Th1 or Th2 cultures of other individuals further confirm this. When the calculated fold differences by the different methods in Th1 and Th2 samples are compared they are in most cases reminiscent but not equal. Because of a broader dynamic range of the real-time RT-PCR detection, we consider TaqMan results to be more reliable in this respect. In the case of the chemokine receptors CCR1, CCR2 and CCR3, where the expression level in Th2 cells is low, the sensitivity of the method becomes limiting and leads to higher values of standard deviation. The reason for the significantly higher fold difference values recorded by the oligonucleotide arrays for TIMP-1 and IL-4 is currently unknown. Most notably, even when the fold difference for some of the target genes (CCR5, IκB, STAT4) is low (< twofold) according to an oligonucleotide array analysis, the transcriptional preference in all Th1/Th2 samples is similar and the statistical significance high (p < 0.01).

Because cell activation is known to affect the transcription of genes in T cells [21], we next asked whether the cell activation also changes the level of transcription of genes in polarized T helper cells. Real-time RT-PCR measurements of gene expression were carried out with a set of 14 genes (Table 1) on three separate T-cell cultures. Polarized (day 14 cells) CD4⁺ Th1 or Th2 cells were triggered with antibodies for CD3 and CD28 for 2, 6 or 24 hours and gene expression was quantitated (Figure 3). During the time course of activation a general trend of downregulation of transcription was observed in the genes studied. The expression of a housekeeping gene, however, was unaffected (data not shown). The downregulation occurred within 24 hours in both type 1 and type 2 cells. Even the transcripts for GM-CSF, TNF-α and IκB, which were rapidly upregulated in the early activation period were downregulated soon after. Downregulation of expression was also seen for CCR1, CCR2, CCR5, caspase-1 and clusterin. Both T-cell subsets seem to exhibit very similar patterns of changes in the mRNA levels during activation of these genes, with the following exceptions. First, the early induction of GM-CSF and TNF-α expression is more pronounced in Th1 cells than in Th2 cells. Second, the downregulation of transcripts at 24 hours of activation for CCR1, CCR2 and CCR5 in Th2 cells is more effective compared to Th1 cells. The mRNA levels for RANTES, CCR4, FGFR and STAT4 in Th1 or Th2 cells did not respond to an anti-CD3/CD28 mediated activation signal or the observed changes were low (< two- to threefold). In conclusion, genes that responded promptly to the activation signal were GM-CSF, TNF-α, IκB, CCR1, CCR2, CCR5, caspase-1 and clusterin. Most importantly, activation-induced changes in expression of these genes did not affect the preferential expression pattern in Th1 and Th2 cells.

Human CD4⁺ T helper cells were isolated from neonatal cord blood and cultured in polarizing conditions for 14 days as previously described. Briefly, the stimulation of the cells was carried out by plating 1.0 × 10⁶ T cells per ml and 0.5 × 10⁶ per ml of irradiated (6400 rad) CD32-B7 transfected mouse L fibroblasts [54] and 100 ng/ml PHA (Murex Diagnostics, France) on 24-well flat-bottomed plates (Lindbro, ICN Biomedicals). Cells were grown in Yssel's medium supplemented with 1% of human AB serum (Gemini Bioproducts) and 100 U/ml human recombinant IL-2 (Cellular Products). Cell differentiation was primed with either 2.5 ng/ml human recombinant IL-12 (R & D Systems) or with 10 ng/ml human recombinant IL-4 (R & D Systems,) for Th1 or Th2 cells, respectively. Culture media for Th2 cells contained 10 μg/ml human recombinant α-IL-12 (R & D Systems). The cells were fed every other day and split at day 3-4 after stimulation. At day 7, cells were restimulated and cultured as described above for another 7 days. When cells were harvested for RNA isolation, an additional immunomagnetic purification step of CD4⁺ cells was performed (DYNABEADS M-450 CD4, Dynal A.S). The protocol for activating the cells through CD3/CD28 for 2, 6 and 24 h has been previously described [55]. Monoclonal antibodies for CD3 and CD28 were from Immunotech and goat F(ab')₂ antibody from BioSource International. The phenotype of the cells was defined by measuring secreted IFN-γ and IL-4, corresponding to type 1 and type 2 cells, respectively, by ELISA (Genzyme).

Poly(A)⁺ mRNA (1 μg) was isolated by two rounds of Oligotex purification (Qiagen). cDNA synthesis of mRNA, in vitro transcription and the production of a biotin-labeled RNA probe for oligonucleotide arrays were performed as previously described [8]. The hybridization mixture contained 12.5 μg of the fragmented sample cRNA together with defined amounts of control Escherichia coli and bacteriophage P1 gene cRNAs, produced for spiking purposes in an identical manner as the sample, and a biotin labeled control oligo (5'-biotin-GTCAAGATCGTACCGTTCAG-3') from Genset. Subsequent hybridization and washing steps were done as described by Wodicka et al. [17]. Chips were washed with GeneChip WashB program on a fluidics station and scanned with the GeneChip scan program (Affymetrix).

The principles of quantitative analysis of hybridization intensities have previously been described [16,17,56]. Data analysis was carried out under low or high stringency using the GeneChip software program. Low-stringency parameter values were: difference threshold (DT) = 20; ratio threshold (RT) = 1.5; change threshold (CT) = 15, and percent change threshold (PCT) = 30. High-stringency parameter values were: DT = 50; RT = 1.5; CT = 50 and PCT = 30. Comparative data were obtained by analyzing results from IL-4-treated RNA samples (Th2 cells) compared to those from IL-12-treated RNA samples (Th1 cells) as baseline. All intensity values below 20 were assigned a threshold value 20. The data reported here is from the more rigorous high stringency analysis (cutoff range > 1.5-fold). The GenBank accession numbers of the sequences used for the initial microarray design and reported here are as presented in Table 1 and as follows: BAX (U19599); c-Jun (J04111); granzyme B (M28879); IFN-αβR (X77722); IFN-γ (X13274); IFN-γRβ (U05875); IL-4 (M13982); IL-5 (X04688); IL-8 (M17017); IL-13 (L06801); IL-7R (M29696); IL-12Rβ2 (U64198); IRF-1 (X14454); MIF (M25639); MIP-1α (M23452); MIP-1β (J04130); SLAM (U33017) and Smad2 (U59911).

The principles of the real-time RT-PCR detection with hydrolysis probes (TaqMan) have been previously described [19,57]. TaqMan probes and primers for the quantitative detection of target mRNAs were designed (Table 1) by using Primer Express computer software (Applied Biosystems). The real-time measurements were carried out with the ABI PRISM 7700 SDS instrument (Applied Biosystems) as described in Hamalainen et al. [20]. Samples were analyzed in duplicate in three independent runs. Statistical significance was determined by ANOVA model. The C_T value is defined as the cycle number in which the detected fluorescence exceeds the threshold value [19,57]. In all experiments the threshold value used to determine C_T during analysis was kept constant. For each probe and primer pair the linearity of detection was confirmed to have a correlation coefficient of at least 0.98 over the detection area by measuring a fourfold dilution curve with cDNA isolated from T helper cells. Fold difference in expression in Th1 and Th2 cells was therefore calculated assuming 100% efficient PCR where each C_T was normalized to GAPDH:

(1) Fold difference =

2 ^{||C_T1(target) - C_T1(GAPDH)| - | C_T2(target) - C_T2(GAPDH)| |}

(2) Fold difference =

2^{||deltaC_T1| - |deltaC_T2||}

Where C_T1(target) and C_T2(target) represent the C_T values for the target gene of Th1 and Th2 samples, respectively. C_T1(GAPDH) and C_T2(GAPDH) represent the C_T values for the GAPDH gene of the Th1 and Th2 samples, respectively.