Cell-type and Donor-specific Transcriptional Responses to Interferon-α

A sensitive, specific, reproducible, robust, and cost-effective customized cDNA array system based on established nylon membrane technology has been developed for convenient multisample expression profiling for several hundred genes of choice. The genes represented are easily adjusted (depending on the availability of corresponding cDNAs) and the method is accordingly readily applicable to a wide variety of systems. Here we have focused on the expression profiles for interferon-α2a, the most widely used interferon for the treatment of viral hepatitis and malignancies, in primary cells (peripheral blood mononuclear cells, T cells, and dendritic cells) and cell lines (Kit255, HT1080, HepG2, and HuH7). Of 150 genes studied, only six were consistently induced in all cell types and donors, whereas 74 genes were induced in at least one cell type. IRF-7 was identified as the only gene exclusively induced in the hematopoietic cells. No gene was exclusively induced in the nonhematopoietic cell lines. In T cells 12, and in dendritic cells, 25 genes were induced in all donors whereas 45 and 42 genes, respectively, were induced in at least one donor. The data suggest that signaling through IFN-α2 can be substantially modulated to yield significant cell-type and donor-specific qualitative and quantitative differences in gene expression in response to this cytokine under highly standardized conditions.

Type I (predominantly ␣ and ␤) and type II (␥) IFNs 1 play a central role in mediating antiviral, antiproliferative, and immunomodulatory responses. The pathways that are involved in IFN-induced gene expression include specific type I and II receptors, JAKs and STATs (1). Upon ligand binding, STATs form homo-or heterodimers through phosphotyrosine-SH2 interactions following activation by JAKs. Whereas STAT dimers bind to ␥-activated sequence elements, both STAT1-2 heterodimers and STAT1 homodimers bind to p48 (ISGF-3␥/ IRF-9) resulting in a trimer that binds to interferon-stimulated regulatory elements in promoters of responsive genes (2).
To date the gene expression profile induced by IFN-␣2 has been studied predominantly in fibrosarcoma and melanoma cell lines (3,4). Little is known about the transcriptional profiles for other cell lines and nontransformed cells or of donorspecific differences. The definition of cell-type and donor-specific quantitative and qualitative differences is, however, central to a full understanding of the biology of the IFNs and their mechanisms of action.
Approaches through expression profiling are also of potential clinical importance. IFN-␣2 is widely used in the treatment of diseases including chronic viral hepatitis B and C and several malignancies (5,6). Only a minority of patients, however, respond to this therapy (7). The definition of gene expression profiles that correspond to "response" or "nonresponse" should ultimately result in further optimization of IFN treatment. Genes that are abnormally expressed in "nonresponders" to IFN-␣2 may define novel pharmacological targets and provide further insight into the pathophysiology of the underlying disease.
To address these questions it is critical to use read-out systems that cover expression from a large number of genes. Technological advances have made possible the simultaneous detection of thousands of gene transcripts using small tissue or cell samples. These technologies include DNA chips (high density oligoarrays (8,9) or microarrays (10)), differential display (11), differential cDNA arrays (12)(13)(14), serial analysis of gene expression (15), and expressed sequence tag data base comparison (16). These methods have been used to analyze gene expression in colon, breast, ovarian, and renal cell carcinomas, multiple sclerosis lesions, leukemic cells, and to monitor gene expression in resting, activated, and anergic lymphocytes (17)(18)(19)(20)(21)(22)(23)(24)(25). Although large scale array techniques are particularly useful to give a broad view of gene expression changes between samples and to discover "novel" genes that are induced by a particular cytokine or drug, they are, in general, costly, labor intensive, and unsuitable for the assay of multiple samples necessary for the detailed analysis of cytokine responses. Appropriate, customized, nylon membrane-based filter arrays, however, are attractive for precisely such analyses.
Here, we describe a customized cDNA array system that is specific, sensitive, robust, reproducible, convenient to use, and cost-effective. Using this technology we have defined significant quantitative and qualitative differences in the response of cells of hematopoietic and nonhematopoietic origin to IFN-␣2a * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. under highly standardized conditions. Substantial quantitative and qualitative donor-specific differences for T cells and DC were observed in response to this cytokine.

EXPERIMENTAL PROCEDURES
Cell Culture-Human HT1080 (fibrosarcoma), HepG2 (hepatoma), HuH7 (hepatoma), and Kit255 (T cell lymphoma) were cultured in Dulbecco's modified Eagle's medium supplemented with 5% fetal calf serum, 2 mM L-glutamine, penicillin, and streptomycin at 37°C in a humidified atmosphere containing 10% CO 2 . Recombinant human IFN-␣2a was provided by Roche Molecular Biochemicals. The human IL-2dependent T cell line Kit255 (26) was maintained in RPMI 1640 supplemented with 10% heat-inactivated fetal calf serum and 20 ng/ml rIL-2 (Proleukin, Chiron, Emeryville, CA). Prior to treatment of cells with IFN-␣2, Kit255 cells were washed and then cultured for 48 h in the absence of rIL-2. PBMC were isolated from buffy coat by density centrifugation on Lymphoprep (Nycomed, Norway). To obtain TC, PBMC were activated with phytohemagglutinin (Murex, UK) and maintained in RPMI 1640 supplemented with 10% inactivated fetal calf serum and human rIL-2 (20 ng/ml) for 1 week. Prior to treatment of cells with IFN-␣2, TC were washed and then cultured for 48 h in the absence of rIL-2. To generate DC, monocytes were isolated from PBMC by magnetic cell sorting using anti-CD14-conjugated magnetic microbeads (Miltenyi Biotec, Cologne, Germany) and cultured for 6 days in RPMI 1640 supplemented with 10% inactivated fetal calf serum, 50 ng/ml granulocyte-macrophage colony-stimulating factor, and 50 ng/ml IL-4 (both from R&D Systems). All experiments were performed under stringent endotoxin-free conditions. RNA Extraction-Total RNA was isolated from cells using Trizol (Invitrogen) according to the instructions of the manufacturer. RNA quantity and quality was analyzed by spectrophotometry and additional visualization by agarose gel electrophoresis.
Selection and Propagation of IMAGE Clones-Genes of interest were selected from the UniGene data base (www.ncbi.nlm.nih.gov/UniGene/ Hs.Home.html). 5Ј IMAGE clones with 0.5-0.8 kb length were chosen and ordered from the Human Genome Mapping Project, Hinxton, UK (www.hgmp.mrc.ac.uk). Bacteria were streaked out onto 1.5% LB-agar plates containing 75 g/ml ampicillin and cultured overnight at 37°C. Single clones were picked, transferred to 96-well plates with 200 l of LB medium containing 75 g/ml ampicillin and 10% glycerol, and grown overnight at 37°C in an incubator without shaking. A 1/10 dilution of individual clones was set up in 96-well plates by adding 10 l of bacterial culture to 90 l of sterile ddH 2 O. Throughout the duration of the experiments, the number of genes present on the filters was constantly extended from 150 to 231 genes (Fig. 1A), reflecting the flexibility of the method. The data presented here are, however, restricted to the initial 150 genes (Table I).
Expansion of DNA from IMAGE Clones by PCR-30-l aliquots from the 10-fold diluted bacterial cultures were transferred into PCR strips on ice. The cDNA inserts were amplified in the presence of 50 mM KCl, 10 mM Tris, pH 9.0, 0.1% Triton X-100, 1.5 mM MgCl 2 , 0.2 mM of each of the deoxynucleotide triphosphates, and 50 units of Taq polymerase using the following conditions: initial denaturation 94°C, 3 min; denaturation 94°C, 40 s; annealing 55°C, 30 s; elongation 72°C, 1 min for 30 cycles followed by a final elongation at 72°C, 7 min using the M13 forward primer (5Ј-GTAAAACGACGGCCAGT-3Ј) and the M13 reverse primer (5Ј-CAGGAAACAGCTATGAC-3Ј). The PCR-amplified DNA was diluted 1/2 with Tris/EDTA, pH 8.0, in round-bottom 96-well plates, and stored at Ϫ20°C. This DNA was used for spotting onto nylon membranes. To confirm the identity of the IMAGE clones, DNA was PCR amplified in 50-l reactions, purified (QIAquick PCR purification kit, Qiagen, Crawley, UK), and sequenced (ABI Prism, Applied Biosystems, Foster City, CA).
Preparation of Membranes-Membranes (Hybond N ϩ , Amersham Biosciences) were cut (12.5 ϫ 8 cm), placed on top of one 3MM sheet (12.5 ϫ 8 cm) and assembled on an aluminum board using a registration device (V&P Scientific, San Diego, CA, catalog number VP382). 96-pin replicators (V&P Scientific, catalog number VP409) were treated with surfactant (V&P Scientific, catalog number VP110) prior to use according to the manufacturer's instructions. Library copiers (V&P Scientific, catalog number VP381) were used for exact positioning of the replicator on the 96-well plates containing the amplified IMAGE clones. DNA was then transferred from 96-well plates to the membranes using 96-pin replicators (1 stroke). Each clone was spotted in triplicate. Membranes were dried overnight at room temperature. Batches of four membranes were transferred to plastic boxes (20 ϫ 20 ϫ 5 cm) and the DNA was denatured in 1 liter of 0.66 M NaCl, 0.5 M NaOH for 10 min at room temperature. The membranes were washed in 1 liter of deionized water, neutralized (40 mM Na 2 HPO 4 /NaH 2 PO 4 , pH 7.2), and rewashed with deionized water all for 10 min at room temperature in an orbital shaker. Prior to use, the DNA on the membranes was UV cross-linked (120,000 J/cm 2 ).
Generation of Labeled cDNA, Hybridization, Washing of Membranes, and Analysis-Radiolabeled cDNA was generated from 5 to 40 g of total RNA by reverse transcriptase at 42°C for 2 h using 360 units of reverse transcriptase (Superscript II, Invitrogen), dATP, dTTP, dGTP (0.5 mM each, Amersham Biosciences), and dCTP (2 M, Amersham Biosciences) in the presence of 30 Ci of [␣-33 P]dCTP (PerkinElmer Life Sciences, catalog number NEG613H), T 23 ACG anchored primers (1 g), and RNase inhibitor (40 units Stratagene, Amsterdam, Netherlands, catalog number 300 -151) in a total volume of 30 l. After reverse transcription, residual RNA was hydrolyzed by alkaline treatment (15 l of 0.1 M NaOH) at 70°C for 20 min followed by neutralization with 15 l of 0.1 M HCl. To remove unincorporated nucleotides the 33 P-labeled cDNA was purified using Sephadex columns (ProbeQuant G-50, Amersham Biosciences, catalog number 27-5330-01). Before hybridization to the arrays, the labeled cDNA was mixed with 50 g of COT1-DNA (Invitrogen, catalog number 15279-011) and 10 g of poly(dA) DNA (Amersham Biosciences, catalog number 27-7836-02) in 4ϫ SSC, 0.1% SDS, denatured at 95°C for 5 min, and hybridized for 1 h to minimize nonspecific binding to repetitive sequences and the poly(A) tail. After denaturation, the cDNA was added directly to medium sized hybridization bottles (260 ϫ 40 mm, Amersham Biosciences, catalog number RPN2516) containing the membrane arrays prehybridized in 20 ml of CHURCH buffer for 30 min in a rotary hybridization oven. Hybridization with the 33 P-labeled cDNA was for 16 h at 65°C. After hybridization the hybridization buffer was discarded and replaced by 150 ml of washing buffer: the membranes were washed once in 2ϫ SSC, 0.1% SDS, twice in 0.2ϫ SSC, 0.1% SDS, and once in 0.1ϫ SSC, 0.1% SDS for 20 min each at 65°C. The membranes were transferred to a sheet of 3MM paper, immediately wrapped in Saran wrap, exposed to intensifying screens for 48 h, and scanned with a PhosphorImager at 200micron resolution (Storm 820, Amersham Biosciences). Images were subsequently analyzed with ImageQuant (Amersham Biosciences) and converted into a table of signal intensities. Further data analysis was performed using Excel (Microsoft). For normalization between samples data were corrected for glyceraldehyde-3-phosphate dehydrogenase present in 18 copies on each filter. A detailed laboratory protocol for the cDNA array method described here is available on request from the Schlaak (joerg.schlaak@uni-essen.de) or Kerr (ian.kerr@cancer.org.uk) labs.
RNase Protection Assay-RPAs were carried out as described previously (27). Briefly, probes were synthesized from SP6/T7 transcription vectors and labeled with [ 32 P]UTP to a specific activity of 2-5 ϫ 10 8 cpm/g of input DNA. Aliquots equivalent to 1-3 ϫ 10 5 cpm of each probe and 13 g of total RNA were used per assay. The intensities of radioactive bands were quantified using a PhosphorImager (Storm, Amersham Biosciences). Bands of interest were quantitated and corrected for background. Data are expressed as -fold induction compared with unstimulated samples.
Statistical Methods-Statistical analysis was performed using the two-sample Wilcoxon test.

RESULTS
Here we have developed a customized cDNA array methodology for ISGs based on nylon filter technology. At present this system permits the analysis of between 288 (triplicate spot) and 864 genes (single spot) for up to 12 samples per day (Fig.  1A). Throughout the duration of the experiments it was constantly extended from 150 to 231 genes. The analysis presented here, however, is restricted to the initial 150 genes (Table I). A substantial spectrum of known ISGs can be assayed with this macroarray (''macro''array: spot size Ͼ300 m; ''micro''array: spot size Ͻ300 m) and it allows the convenient investigation of complex experimental settings including, for example, extensive kinetic and dose-response curves. Moreover the processing, analysis, storage, and recovery of the data is significantly easier and quicker compared with that for microarrays, because only genes of interest are investigated. Accordingly, the analysis of the data for 12 arrays takes ϳ60 -90 min using standard Microsoft Excel software. Currently, each cDNA is spotted in triplicate (mean coefficient of variation for triplicates: 6 -8%) to yield maximal reproducibility and sensitivity (Fig. 1B). The high flexibility of the spotting procedure also permits the generation of filters with only one spot per gene that are particularly useful for scanning higher numbers of target genes with lower sensitivity.
To enhance the performance of the system it is critical to block nonspecific hybridization through repetitive sequences or the poly(A) tail using COT1-DNA and poly(dA) (Fig. 2). This is particularly useful for genes only marginally (1.5-2.5-fold) induced: differentials for highly induced genes are still detectable in the absence of prehybridization with COT1-DNA and poly(dA). The system offers a high degree of reproducibility as indicated by its low inter-and intra-assay variation (Fig. 3, A and B, Supplementary Materials for Table II, Table IV, experiment TC V/a-d). To achieve this it is essential to use strict endotoxin-free culture conditions because lipopolysaccharide can induce IFN-␤ and the expression of ISGs (28). 2 For RNA extraction, to avoid artifacts induced by prolonged trypsinization and centrifugation, adherent cells were lysed directly on the tissue culture plates and suspension cells directly after spinning down without washing.
All of the IMAGE clones used were sequence verified. Comparability of the macroarray data for known genes with data obtained by alternative accepted RNase protection methodology was established by data from experiments carried out as an integral part of our ongoing program. An example of the data from one such experiment (involving an analysis of the responses obtained through the endogenous type I and II IFN receptors and a receptor chimera 2fEg⌬B (29)), reveals, for the ISG-56k, IRF-1, and 9 -27 ISGs, a good correlation (r ϭ 0.89 - Genes of interest were selected from the UniGene data base. These genes comprise known ISGs and genes of intrinsic interest that might or might not be induced by IFNs in different cell systems. They include genes involved in cell proliferation, immune responses, and the responses to a variety of cytokines. 5Ј IMAGE clones with 0.5-0.8 kb length were chosen and obtained from the Human Genome Mapping Project.  0.99) between the data from the two approaches (Fig. 4). Similarly good correlations have been obtained in a number of further experiments comparing the results by the two methods for the above and additional ISGs including IP-10, GBP-1, 6-16, MxA and 2-5OAS (for example, Fig. 7, Table V). 3 The sensitivity of the method has also been assessed. As a rule of thumb, in most micro-and macroarray systems a 2-fold change in the expression level is regarded as being significant. Statistical analysis showed that the macroarrays are capable of detecting smaller differences, after stimulation with very low concentrations of IFN-␣ (e.g. 10 IU/ml, Fig. 5), changes in gene expression of 30% or less are detectable with a high degree of significance (p value Ͻ 0.05) by this method. This permits the analysis of dose-response curves for poorly induced (Ͻ2-fold, Fig. 6) genes considered marginally significant by other methods. Using more replicates of the spotted DNA this high sensitivity may be enhanced even further. The physiological relevance of these relatively small changes, however, still have to be determined for the individual genes. Accordingly, we have retained 2-fold as the threshold level for significant inducibility for comparative expression profiling (Tables II-IV), which is, in addition, associated with a very high degree of statistical significance (p ϭ 0.001 and less, Fig. 5).
Cell Type-specific Responses to IFN-␣2a-To date little is known about the cell-type specificity for the expression profiles induced by the ␣-IFNs. The most substantial current data were obtained for human fibrosarcoma cell lines (3). Only limited information is available for nontransformed or hematopoietic cells. Accordingly, here expression profiles were compared for HT1080, HepG2, HuH7, and Kit255 cell lines and the nontransformed PBMC, T cells, and DC. Stimulation was with 1000 units/ml rhIFN-␣2a for 6 (T cells) or 8 h (remainder): conditions known from further extensive experiments to yield maximal responses for each of the cell types represented. 150 genes (Table I) were represented on the arrays. Examples of the data from all seven cell types (Table II) and for different donors for the DC and T cells (Tables III and 4) are presented. 74 genes were induced more than 2-fold in at least one cell type (Table II). Of these only ISG-56k, ISG-15, GBP-1, 9-27, 6-16, and LMP-2 were consistently induced in all cell types and all experiments (Tables II-IV). Even within this group, however, there was considerable quantitative variation, for example, ISG-56k and ISG-15 were predominantly highly induced (Ͼ10fold, with two exceptions, TC III and V, Table IV), whereas LMP-2 induction was low (2-4-fold) throughout and for the 6-16 variable (2-21-fold with no obvious pattern, Tables II-IV). IRF-7 was consistently induced in the primary cells but not in the cell lines (Tables II-IV). CCR1 and HIF-1 appeared to be predominantly induced in PBMC with low or no induction in T cells, DC, or the cell lines (Tables II-IV). VEGF-C, IL-6, E2F-1,

FIG. 3. Assay variation.
A, intra-assay variation. Two independent samples of total RNA were extracted from 10-cm dishes of unstimulated HT1080 cells, reverse transcribed into labeled cDNA, and hybridized to target cDNA spotted onto nylon membranes. In this scatter plot analysis data are shown as absolute intensity values for each gene. B, inter-assay variation. HT1080 cells were stimulated in two independent experiments with 1000 units/ml IFN-␣2a for 8 h. Total RNA was extracted, reverse transcribed into labeled cDNA, and hybridized to target cDNA spotted onto nylon membranes. In this scatter plot analysis data are shown as -fold induction for each gene in the two experiments.
Pim-1, and SOCS2 were up-regulated in PBMC and T cells but not in DC, which might indicate that under these conditions these genes are predominantly induced in the T cell population. These genes were not, however, induced in all 5 T cell donors (Table IV). Of the genes represented none was exclusively induced in nonhematopoietic cells although RING4 and phospholipid scramblase 1 showed only low induction in 4 of 5 of the T cell and 2 of the 3 DC samples (Tables III and IV). In contrast to the donor variability observed for DC and T cells (Tables III  and IV and below), for the cell lines similar results to those presented in Table II were obtained in additional experiments (see Supplementary Materials).
To establish further the validity of the comparative macroarrays and the cell type-specific differences between the T and DC cells for IP-10 in particular, a number of RPAs were also carried out on the T and DC cell RNAs for a spectrum of ISGs (e.g. Fig. 7, Table V). The data confirm the major differences in the induction of IP10 (strongly induced in DC but not T cells) and the similar patterns of induction observed for the ISGs by the two types of assay. Lower -fold inductions are frequently observed using macro-or microarray versus alternative assays. This is particularly obvious here for IP10 in DCs (Table V, a). In this instance the higher -fold induction by RPA (60 -349-fold) versus macroarray (15.6 -20.6-fold) reflects, in part at least, the very low background values obtained for IP10 in the RPAs (Fig.  7, lanes 1-4). For the 2-5OAS and 6-16 genes the generally lower -fold induction values by macroarray versus RPA almost certainly reflect, in part, similar differences in background values, but likely also donor variability (see below) and possibly the relative hybridization efficiencies of the particular image clones ''chosen'' for the macroarrays. Despite these differences in the -fold inductions observed by the RPAs and macroarrays for some of the ISGs, in general the agreement between the two types of assay is good with those ISGs registering as substantially or minimally induced by one assay registering similarly by the other. Overall, for the induced genes, the data appear consistent with limited cell-type specificity superimposed upon fairly wide inducibility for most genes, but with considerable quantitative variation between cell types and, see below, between the same cell type from individual donors.  1-4) and JAK2-negative ␥2AEg⌬BY440 cells (lane [5][6][7][8] were stimulated with 1000 units/ml IFN-␣ (lanes 2 and 6), 100 units/ml Epo (lanes 3 and 7), or 1000 units/ml IFN-␥ (lanes 4 and 8) for 6 h, respectively. Aliquots of cytoplasmic RNA (13 g) were analyzed by RNase protection assay using probes for ISG-56k, IRF-1, 9-27, and ␥-actin as loading control. B, comparison of the data obtained by RNase protection and macroarray analyses. Total RNA (40 g) from the experiment shown in A was analyzed by macroarray as previously described. Quantitative data from both analyses are compared. Data are shown as -fold induction compared with the matched control (samples 1 and 5, respectively).

Donor-type Specific Responses in DC and T Cells-
Little is known about donor-specific differences in the transcriptional response to IFN-␣. To address this question DC and T cells from a number of donors were isolated, cultured, and treated with IFN-␣2a under identical conditions.
Dendritic Cells-In three independent experiments with DC derived from 3 different healthy individuals (DC I-III, Table  III), a total of 42 genes was found to be induced in at least one donor, whereas 30 (71%) genes were up-regulated in at least 2 donors and 25 (60%) in all 3 donors (Table III). 11 (26%) ISGs were found with apparent qualitative rather than quantitative differences as defined by inductions of Ͼ2-fold and Յ1.3-fold each for at least one donor: TTF-2, collagen type XVI, PRAME, phospholipid scramblase 1, Mig, IP-30, RAP46, HLA-E, p48/ ISGF-3␥/IRF-9, iNOS2A, and granzyme B. 5 genes were exclusively induced in donor DC I, 7 genes were found to be up-regulated only in donor DC III, whereas no gene was exclusively induced in donor DC II. The larger among these differences, e.g. those for TTF2, collagen type XVI, and PRAME, at least, are unlikely to reflect simple quantitative differences because, overall, a surprising degree of quantitative consistency was observed between donors for the majority of the induced genes (for example, IP-10, MxB, IRF-7, IFP-35, and 2-5OAS).
T Cells-In five independent experiments with T cells from 5 different healthy donors (TC I-V, Table IV), a total of 45 genes was found to be induced (Ͼ2-fold) by IFN-␣2a in at least one donor, whereas 35 (78%) of these were up-regulated in at least 2 donors, 22 (49%) in at least 3 donors, 14 (31%) in at least 4 donors, and 10 (22%) in all 5 donors (Table IV). For donor TC V, data were obtained in quadruplicate. For this donor, the median of the 4 replicates was used to define the inducibility of the individual genes. 28 (62%) ISGs were found with apparent qualitative rather than quantitative differences, i.e. substan-  tial induction for one or more and Յ1.3-fold in at least one donor, extreme examples being MxA (52-versus 0.7-fold) and IFI-16 (15-versus 0.7-fold). As indicated by the lower percentage of "conserved"' responses among the donors (22 versus 60%) and the higher percentage of qualitative differences (60 versus 28%), the donor-specific differences appeared to be more prominent in T cells than DC. In addition to the conserved responses found in all cell types (ISG-56k, ISG-15, GBP-1, 9-27, 6-16, and LMP-2), HCV-associated protein 44 kDa, BST2, IRF-7, Hou, and STAT1 were induced in all 5 donors.
Because T cells and DC were obtained from different donors and no matched preparations, i.e. T cells and DC from one donor, were available, it is not clear whether the observed donor-specific responses are specific for the cell type investigated. Some ISGs (TTF2, PRAME, granzyme B, RAP46, and p48/ISGF-3␥/IRF-9) were, however, identified that showed apparent qualitative differences (induction Յ1.3 in at least one donor) in both T cells and DC suggesting that, for these genes, the factor(s) responsible for differential donor-dependent induction by IFN-␣2a may be shared between different cell types. DISCUSSION A highly sensitive, specific and reproducible customized cDNA array system has been established. It is suitable for the routine assay of multiple samples including, for example, those necessary for the dose response and kinetic analyses required for detailed comparisons of the effect of a given cytokine in different cell types (as here) or under different conditions or in response to the inhibition of a particular signal transduction pathway. These ''custom'' arrays are clearly designed to study a set of known genes, not to discover ''novel'' genes involved in particular signaling pathways or biological conditions. Cell lines offer the advantage of virtually unlimited availability of RNA. Access to larger amounts of RNA from possibly more biologically relevant nontransformed cells is, however, often limited. An advantage of the macroarrays is the relatively small amount of total RNA required (5-10 g) per sample. Accordingly, ''macros'' are suitable for the assay of sufficient samples for significant analyses of readily (PBMC) or less readily available (DC) nontransformed cells. Systematic information on the transcriptional response of such cells to IFN-␣2a has been generated here. The modular design of this technology allows easy adjustment to address alternative questions across species borders provided the genes of interest are known and cDNAs, ideally in the form of IMAGE clones, are available. Based on the methodology described here, macroarrays for expression profiling in response to human IFN-␥, for type I and II murine IFNs, and for woodchuck ISGs have also been established and the technology is being used, for example, to study in vivo responses in PBMC from patients treated with IFN-␣2.
In addition to signaling through ISGF-3 (STATs 1, 2, and p48/ISGF-3␥/IRF-9) type I IFNs variably activate additional STATs. STAT3 is activated in a variety of cell types (30) and IFN-␣2a directly activates (in addition to STATs 1, -2, and -3) STAT5 and -6 in Daudi cells (31), STAT4 in T cells and natural killer cells (32), and STAT5 in undifferentiated promonocytic U937 cells (33). Moreover, JAK/STAT signaling is essential but not sufficient for full IFN responses. Additional pathways activated include the ERK1/2 and p38 MAP kinases, the phosphatidylinositol 3-kinase/Akt pathway and those pathways responsible for the phosphorylation of STAT1 on Ser-727 (34 -38). Differential activation of these and other additional pathways  1, 2, 5, and 6) and T cells (lanes 3, 4, 7, and 8) were stimulated with 1000 units/ml IFN-␣ for 8 and 6 h, respectively. Aliquots of cytoplasmic RNA (13 g) were analyzed by RNase protection assay using probes for IP-10 (lanes 1-4) or STAT1␣, STAT1␤, p48, ISG-56k, 2-5OAS, 6-16, 9-27, and GBP-1 (lanes 5-8). Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (lanes 1-4) or ␥-actin (lanes 5-8) were used as loading control. yet to be identified, together with differential STAT activation might a priori lead to the differential induction of subsets of genes in different cell types. Indeed an initial objective was to determine whether such subsets are in fact differentially induced. If so, this could, in turn, through shared promoter elements, shed light on the involvement of additional known or novel signaling pathways in the type I IFN response. Although there is little evidence here for discrete subsets of genes highly induced uniquely in one or only a few related cell types, some degree of cell-type specificity was observed. Relatively few genes appeared to be induced in all cell types (Table II). More particularly the induction of IRF-7 appeared to be specific to primary PMBC, DC, and T cells of the immune system, whereas numerous genes were substantially induced in one or more of the primary PBMC, DC, or T cells but not the cell lines. In addition, IFI-16, IP-10, and PRAME provide striking examples of substantial cell type and donor differentials in DC and T cells. Because the ISGs analyzed were mainly selected based on data obtained from experiments with fibrosarcoma cells, further differentials in the induction of, yet unknown, ISGs can be anticipated. Accordingly, to build up a comprehensive custom array for human IFNs, the multisample macroarray analyses will be complemented by further limitedsample microarrays to identify additional novel ISGs in a variety of cell types. The appropriate combined application of the two types of array should permit the detection of additional cell type-specific ISGs and convenient detailed analysis of their expression. IRF-7 has previously been reported to be predominantly expressed in cells of hematopoietic origin (39). Of the genes analyzed IRF-7 was the only one consistently induced in PBMC, DC, and T cells but not in nonhematopoietic cells. IRF-7 is induced by a variety of agents including viruses, lipopolysaccharide, and type I IFNs. It regulates the production of the ␣-IFNs, IFN-␤, and the chemokines RANTES and IP-10 in virus-infected cells (40 -42). It has also been identified as a key regulator for monocyte differentiation to macrophages (43). The lack of expression of IRF-7 in the fibrosarcoma cell line 2fTGH correlates with hypermethylation of the CpG island in the human IRF-7 promoter (44). Such hypermethylation may represent one mechanism whereby differential cell type-specific responses to IFNs are generated.

FIG. 7. Assessment of ISG induction in dendritic cells and T cells by RNase protection assay. Dendritic cells (lanes
Little is known about donor-specific responses to IFNs and their molecular background. It is, however, well established that both the clinical side effects of IFN-␣ therapy and its efficacy vary largely within groups of patients treated for chronic viral hepatitis or malignancies (45)(46)(47)(48)(49)(50). There is, therefore, large donor-dependent variation in the clinical response to this cytokine. The data here support this in that substantial donor-specific IFN responses were observed. It is important to emphasize that differences similar to those observed between the expression profiles for different donors (Tables III and IV) were not observed for the remarkably reproducible profiles obtained on different occasions for the cell lines (e.g. Fig. 2). Overall the data are consistent with complex modulation ranging from possible specificity for particular lineages or cell-type (IRF-7) to universal high induction (ISG-56k and ISG-15), with the majority of genes showing substantial quantitative variation between cell types and donors indicative of the likely modulatory effects of differential activation of STATs and additional pathways. Superimposed on this one can reasonably expect some degree of quantitative cell type and donor variation with IFN dose and kinetics. SNP in the promoter and coding regions of ISGs may also affect their transcription and function thus contributing to donor-specific differentials. SNPs in IRF-1 in several human liver cancer cell lines result in different antiproliferative effects of type I IFNs (51) and have been associated with juvenile idiopathic arthritis (52) and childhood atopic asthma (53), whereas SNPs in IRF-2 were found to be associated with atopic dermatitis (54). Although donor-specific responses to IFNs were not analyzed in these studies, accepting the secondary requirement for IRFs in the induction of many ISGs, one can assume that such SNPs can lead to altered function and result in differential responses to the IFNs. In principle, SNPs in promoter or coding regions of JAKs or STATs could similarly affect donor-specific responses. SNPs at phosphorylation sites could profoundly affect the quality and quantity of downstream signaling events. Relevant data are not yet available for STAT1 and -2 but SNPs in the first exon of the STAT6 gene are associated with the development of allergic diseases (55).
Here, nylon membrane-based cDNA array technology has been further developed and optimized to provide a sensitive, robust, and convenient method for the analysis of expression profiles for a substantial ''customized'' set of genes of interest. The method has been used to study, in a number of both primary cells and cell lines, cell-type and donor-specific responses to IFN-␣2, the most widely used IFN for the treatment of viral hepatitis and malignancies. The data have revealed that signaling in response to IFN-␣2 is substantially modulated leading to significant cell-type and donor-specific qualitative and quantitative differences in the response to this cytokine. Further analysis of the in vivo responses to IFN-␣2 and definition of the modulatory factors responsible for the heterogenous transcriptional responses to this agent will lead to further insight into the biology of IFN-␣ signaling and ultimately to further improvements in IFN therapy. T cells using RNase protection assays and macroarrays Cells from 6 (DC) or 7 (T cells) different donors, respectively, were stimulated with 1000 units/ml IFN-␣2a for 6 h. Data are shown as -fold change compared to a matched control (ND; not determined). Shaded areas denote inductions Ͼ2-fold.