Identification of Novel Cytokine-Induced Genes in Pancreatic β-Cells by High-Density Oligonucleotide Arrays

Pancreatic islets were isolated from male Wistar rats 10 weeks of age by collagenase digestion, and islet β-cells were purified by autofluorescence-activated cell sorting (29) (FACStar, Becton-Dickinson, Sunnyvale, CA). β-cell preparations were cultured at 37°C as aggregates in suspension in Ham’s F10 medium (Gibco Brl-Life Technologies, Paisley, U.K.), as previously described (30). For the microarray analysis, purified rat β-cells were precultured in Ham’s medium for 16 h and then exposed to the following conditions: control for 6 h (no cytokines added), IL-1β + INF-γ for 6 h, control for 24 h, IL-1β + INF-γ for 24 h, and IL-1β for 24 h. The number of experimental conditions tested (5) was adapted to the number of microarrays present in the Affymetrix package (5). For the RT-PCR confirmation experiments, four groups were studied (control, IL-1β, INF-γ, and IL-1β + INF-γ conditions), and all groups were studied at both 6 and 24 h. IL-1β (tested at 50 U/ml, 38 U/ng) was a kind gift from Dr. C.W. Reinolds from the National Cancer Institute, Bethesda, MD, and INF-γ (tested at 1,000 U/ml, 10 U/ng) was purchased from Holland Biotechnology, Leiden, the Netherlands. The choice of cytokine concentration and the time of exposure was based on our previous data (10,13,14,16) and aimed to identify genes that are either directly induced by the cytokine(s) and/or result from β-cell responses to cellular stress (mostly after 24 h). After a 24-h exposure to INF-γ + IL-1β, most β-cells are still viable, but ∼10% of the β-cell population is already committed to undergo apoptosis. However, the morphological changes of cell death (concomitant to nonspecific changes in gene expression) are only apparent after a subsequent 48- to 72-h culture (A.K.C. and D.L.E., data not shown). Culture media from the cells used for the microarray analysis were collected after 24 h for nitrite determination (nitrite is a stable product of NO oxidation), as previously described (31).

For the microarray analysis, samples of control and cytokine-treated cells were harvested, and total RNA was isolated using RNeasy kit (Qiagen). Because it was difficult to obtain a sufficient number of rat β-cells in a single occasion, and in order to decrease eventual biases due to biological variation, the cells were pooled from six separated experiments, using 2.5 × 10⁵ cells per group in each experiment. Ten micrograms total RNA were obtained from each pooled experimental group, and the RNA was converted into double-stranded cDNA using a cDNA synthesis kit (Superscript Choice; Gibco, Gaithersburg, MD) with a special oligo(dT)24 primer containing a T7 RNA promoter site added 3′ of the poly-T tract. Biotinylated cRNAs were generated from purified cDNAs using the Bioarray high-yield RNA transcript labeling kit (Enzo Diagnostics, Farmingdale, NY). The cRNA sample was purified with the RNeasy kit (Qiagen), and 20 μg of each cRNA sample was prepared and hybridized as previously described (32). Analysis of differential expression was performed by GeneChip software (version 3.3). Normalization was performed by global scaling, with the arrays scaled to an average intensity of 150. Duplicate hybridizations using separate sets of chips were performed for all conditions. Cytokine-induced differences in gene expression were considered present when the fold-change was ≥3.0 in both experiments. In case genes were previously described as modified by cytokines by other techniques or present in more than one experimental condition (with values >3 in at least one condition), we considered changes ≥2.0 sufficient for inclusion in Table 1. Note that the GeneChip software allocates an arbitrary value of 20 for genes below detection limit. This is done to allow calculation of the gene expression in the experimental group as the fold variation compared with the reference group. Thus, in cases in which the reference genes were undetectable and the allocated fold-variation was between 3 and 10, we indicated the values as >3.0 or <–3.0. If the allocated variation was >10, we indicated it as >10 or <–10.0 (Table 1). Genes were classified on different functional clusters (Table 1) based on the putative biological function of the encoded protein, as determined by database searches on PubMed, gene cards from the Weizmann Institute of Science (http://bioinfo.weizmann.ac.il/bioinfo.html), and a previously published classification scheme for cellular functions (33).

Note that our β-cell preparations have a purity of ∼95%, but also contain 1–4% of α-cells, δ-cells, and PP-cells (29). We observed the presence of both glucagon, somatostastin, peptide YY, and prepropancreatic polypeptide mRNAs among the genes detected by the present analysis (data not shown). Thus, it cannot be excluded that a minor proportion of the presently observed changes in gene expression occurred in these other cell types.

RT-PCR using specific primers was performed to confirm the differential expression of 17 mRNAs detected with the microarray analysis. The selection of RT-PCR instead of Northern blot analysis was motivated by the limited availability of primary β-cells. mRNA isolation and RT-PCR were performed as previously described (16). The number of cycles was selected to allow linear amplification of the cDNA under study. For semiquantitative PCR, the housekeeping gene glyceraldehyde 3-phosphate dehydrogenase (GAPDH) was used as control. We have previously shown (34) and confirmed in these experiments (data not shown) that IL-1β and INF-γ do not affect GAPDH mRNA expression in insulin-producing cells. The primers sequences and their respective PCR fragment lengths were as follows: GAPDH F 5′-TCCCTCAAGATTGTCAGCAA-3′, R 5′-AGATCCACAAACGGATACATT-3′ (308 bp); iNOS F 5′-GACTGCACAGAATGTTCCAG-3′, R 5′-TGGCCAGATGTTCCTCTATT-3′ (308 bp); thyrotropin-releasing hormone F 5′-CCTAACTGGTATCCCTGAAT-3′, R 5′-TGAGAACCAGGAATCCAGAA-3′ (392 bp); Isl-1 F 5′-TTTGGACGAAAGCTGTACCTG-3′, R 5′-TGGATGCAAGGGACTGAGAG-5′ (304 bp); mob-1 F 5′-TGAGTCTGAGTGGGACTCAA-3′, R 5′-CCTTGCTGCTGGAGTTACTT-3′ (452 bp); growth arrest and DNA damage (GADD)-153: F 5′-CTTTGAGACAGTGTCCAGCT-3′, R 5′-TTCAGCAAGCTGTGCCACTT-3′ (377 bp); lactate dehydrogenase B F 5′-ACTCCGTGACAGCCAATTCT-3′, R 5′-AGCATGGATTCGATGAGGTC-3′ (543 bp); gastric inhibitory peptide (GIP) receptor F 5′-ATGTGCCTGGAGATGAGGTC-3′, R 5′-ACCTGCATTCCTCTCACTGG-3′ (541 bp); GLUT1 F 5′-AACCTGTTGGCCTTTGTGTC-3′, R 5′-ATCTGCCGACCCTCTTCTTT-3′ (454 bp); IL-15-F5′-AAAGAGGAGGCGTTCTGGAT-3′, R 5′-GCTGTTTGCAAGGTAGAGCA-3′(452 bp); prolactin receptor (PRL) F 5′-CAAATGGGAAGCAGTTCCTC-3′, R 5′-CCACTGCCCAGACCATAATC-3′ (448 bp); growth hormone (GH) receptor F 5′-ATGAGCCCGATATGGTCAAC-3′, R 5′-TGGTACGTCCAGAATCGTCA-3′ (547 bp); Janus protein tyrosine kinase-2 (JAK-2) F 5′-CTGCAGGACAACACTGGAGA-3′, R 5′-GGGACTTTCACCTGGTTCCT-3′ (441 bp); nuclear factor (NF)-κB-p105 F 5′-TCATCCACCTTCATGCTCAG-3′, R 5′-GCCAACGAGATGTTGTCGTA-3′ (447 bp); C/EBPδ F 5′-GAGACTCCGAACGACCGATA-3′, R 5′-GTCAGAGCTGGTGCCTCTTT-3′ (310 bp); osteoprotegerin F 5′-CGCATCTTGATGGAGAGCTT-3′, R CATTTCAAGAGCCGGAGAAG-3′ (402 bp); macrophage inflammatory protein (MIP)-3α F 5′-TGAGAATGGCCTGCAAGCAT-3′, R 5′-TCCATTGGACAAGACCACTG-3′ (425 bp); and fractalkine F 5′-CAATCTGGATCTAGCCTCTG-3′, R 5′-AGTCGGGACAGGAGTGATA63′ (379 bp). The abundance of the PCR products of interest was expressed in pixel intensities (optical density [OD]), normalized using the maximum signal in each amplification as 10, and divided for the abundance of the GAPDH signal amplified in parallel from the same cDNA sample. The data are presented as a percentage of the respective controls. When control values were below detection limits, the data were presented as their normalized pixel intensities.

Data are presented as means ± SE, and comparisons between groups were performed by Student’s paired t test.

To identify early and late cytokine-induced genes in pancreatic β-cells, these cells were exposed to the following treatments: control condition for 6 h, IL-1β + INF-γ for 6 h, control condition for 24 h, IL-1β + INF-γ for 24 h, and IL-1β for 24 h. After a 24-h culture, there was no detectable nitrite production by the control (non–cytokine-exposed) β-cells. On the other hand, IL-1β–treated cells released (mean ± SE) 7.0 ± 1.1 pmol nitrite · 10^–3 cells · 24 h^–1, whereas cells exposed to IL-1β + INF-γ released 13.6 ± 1.5 pmol nitrite · 10^–3 cells · 24 h^–1 (n = 6; P < 0.001 vs. cells exposed to IL-1β alone). These results are similar to our previous observations (5,35) and confirm that both IL-1β and INF-γ were biologically active.

Cells from the six separate experiments described above were pooled for RNA extraction, and the resulting biotinylated cRNAs were hybridized in duplicate to the Affymetrix rat U34-A oligonucleotide array containing ∼8,000 probes (77% known genes and 23% expression sequence tags [ESTs]). Approximately 3,000 genes or ESTs were scored as present in each of the five conditions (2,700– 3,300). The cytokine-modified known genes are shown in Table 1.

Scatter plot analysis comparing expression levels in control and cytokine-exposed β-cells at both 6 and 24 h showed a large number of cytokine-responsive genes (data not shown). After a 6-h exposure of β-cells to IL-1β + INF-γ, 96 known genes were differentially expressed (Table 1). Exposure of β-cells to the same combination of cytokines for 24 h modified the expression of 147 known genes (Table 1). IL-1β alone induced the differential expression of 105 known genes (Table 1).

To validate the microarray results, we initially compared the data with the available information on cytokine-induced gene expression (15,17). Approximately 27 genes and/or proteins have been previously described as modified by IL-1β and/or INF-γ in whole islets or β-cells using a time schedule similar to the one presently used (e.g., 6- and 24-h exposure to cytokines). Of these 27 genes, 22 were detected in the present analysis (80%). These genes are insulin (decreased), iNOS (increased), arginase (decreased), argininosuccinate synthase (AS) (increased), ornithine decarboxilase (increased), hsp-70 (increased), MnSOD (increased), HO-1 (increased), intracellular adhesion molecule (ICAM)-1 (increased), glucokinase (decreased), GLUT2 (decreased), prohormone convertase-1 (decreased), GAD65 (decreased), MCP-1 (increased), cytokine-induced neutrophyl chemoattractant (CINC)-1 (increased), CINC-3 (increased), major histocompatibility complex (MHC) class I (increased), interferon regulatory factor (IRF)-1 (increased), c-jun (increased), adenine nucleotide translocator-1 (increased), COX-2 (increased), and phospholipase-D1 (decreased). Of the five genes not detected in the present analysis, as modified by cytokines, two genes (pancreatic duodenal homeobox-1 [Pdx-1] and serine protease inhibitor-3) were not present in the array, whereas three genes were present in the array but were either considered as below detection limit (Fas and caspase-1), or were detected but considered “no change” after cytokine exposure (prohormone convertase-2).

To further validate the results of the microarray analysis, we selected 17 genes for confirmation by RT-PCR (Fig. 1). Induction of iNOS, a well-known effect of cytokines in β-cells (15), was used as a positive control. As previously described (15), IL-1β and IL-1β + INF-γ induced a high iNOS expression after 6 h, with a subsequent decline after 24 h of continuous exposure to the cytokines (Fig. 1). All 17 genes considered “changed” by the microarray analysis were confirmed by the RT-PCR as modified by cytokines. The RT-PCR analysis and the comparisons with genes and proteins previously described as modified by cytokines (see above) confirm that microarray analysis, performed in duplicate using pooled cells from several experiments, is a reliable method to detect massive variations in β-cell mRNA expression. Similar conclusions were found in a recent study in which microarray analysis was performed to determine glucose regulation of secretory and metabolic pathway genes in MIN6 insulin-producing cells (36). Note that microarrays are a relatively new technique; the present study, for instance, is the first in which this method is applied for characterization of gene expression in primary β-cells. Ideally, all genes described as modified by the microarray analysis should be confirmed by additional techniques, such as RT-PCR or Northern blot. Thus, the presently described modifications in mRNA expression, which were not confirmed by RT-PCR, should be viewed with caution.

The cytokine-responsive genes were clustered according to the putative biological function of their encoded proteins, as indicated in Table 1. The most frequent changes were observed in β-cell metabolism, with 19, 20, and 20% of all differentially expressed genes induced, respectively, by IL-1β + INF-γ for 6 h, IL-1β + INF-γ for 24 h, and IL-1β alone for 24 h. IL-1β + INF-γ induced an early suppression of β-cell metabolism, with a decrease in ∼80% of the 18 modified genes after 6 h exposure (Table 1, item 1.0). This suppression was maintained at 24 h, with nearly 70% of the metabolism-related genes inhibited by IL-1β + INF-γ or IL-1β alone. These alterations occurred in genes related to the metabolism of carbohydrates, amino acids (others than arginine), lipids, and ATP production.

Among the carbohydrate-related genes, it is noteworthy that cytokines decreased the expression of mRNAs for GLUT2 and glucokinase, whereas they increased expression of GLUT1 (Table 1, item 1.1). IL-1β–induced decreases in GLUT2 have been previously shown at the protein level (37), whereas the present microarray results on GLUT1 were confirmed by RT-PCR (Fig. 1). GLUT2-null knockout mice are hypoinsulinemic and hyperglycemic, but they regain normal insulin secretion and glycemia after transgenic expression of GLUT1 in β-cells (38). Thus, the presently observed upregulation of GLUT1 in cytokine-treated β-cells may represent an adaptive/compensatory mechanism for the decrease in GLUT2 expression. This compensatory response seems to be effective, i.e., rat islets exposed to IL-1β or IL-1β + TNF-α preserve normal glucose utilization but have defective glucose oxidation (39), probably caused by NO-induced aconitase blocking (40).

After a 24-h exposure to IL-1β or IL-1β + INF-γ, there were modifications in the expression of three key genes related to arginine metabolism and nitric oxide formation (Table 1, item 1.2). Thus, there was a parallel induction of iNOS and AS expression at 6 and 24 h, whereas arginase was inhibited. iNOS uses arginine as substrate for NO production, generating citrulline as a byproduct. Citrulline can be recycled into arginine by AS activity (41,42). This and the concomitant inhibition of arginase, an enzyme responsible for arginine degradation (43), will allow a continued arginine supply for NO production by the β-cells.

One of the well-known effects of cytokines in β-cells is inhibition of insulin mRNA expression (confirmed in the present array analysis), total protein and pro-insulin biosynthesis, and decreased insulin release (4,5,44). We presently observed downregulation after both 6- and 24-h exposures to cytokines of several genes related to protein synthesis, modification, and secretion (Table 1, item 2.0), which may at least partly explain the above described decrease in protein synthesis. Moreover, we observed modifications in the expression of two transcription factors, C/EBPβ (increased) and Isl-1 (decreased) (Table 1, item 9.0), which may contribute to decreased insulin mRNA expression. Thus, upregulation of C/EBPβ after β-cell exposure to high glucose was shown to inhibit insulin promoter activity (45), whereas downregulation of Isl-1 (Table 1, Fig. 1), a transcription factor involved in β-cell development and possibly in insulin gene transcription (46), may have a negative effect on insulin mRNA expression.

Cytokine-induced inhibition of insulin release in rat islets is related to decreased glucose oxidation and ATP production (44), but decreased insulin secretion in mouse and human islets is dissociated from inhibition of glucose metabolism (22,47). This suggests that part of the inhibitory effects of cytokines is affected at two distal steps of the insulin release process: ionic fluxes (48) and granule exocytosis (49). In line with this possibility, there were modifications in the expression of several genes encoding for ionic channels (Table 1, item 3.0). Moreover, we detected inhibitory effects of cytokines on the expression of mRNA for soluble NFS attachment protein (SNAP)-25 (mean decrease of –2.9 by IL-1β + INF-γ for 6 h, data not shown), vesicle-associated membrane protein (VAMP)-2, and rab3A (Table 1, item 2.0). These proteins are potential regulators of trafficking, docking, and fusion of secretory vesicles (50). Morphological and biochemical studies demonstrated the presence of VAMP-2 in insulin and in γ-aminobutiric acid (GABA) secretory vesicles of β-cells (51), and cleavage of both VAMP-2 and SNAP-25 by tetanus or botulinum neurotoxins block insulin exocytosis in β-cells (52,53). The rab small G protein family consists of nearly 30 members implicated in intracellular vesicle trafficking (54). Rab 3A is associated with the membrane of secretory granules of rat pancreatic β-cells, and overexpression of Rab 3A mutants decreases nutrient-stimulated insulin secretion (55). The expression of the rab-3 GDP/GTP exchange protein, which stimulates the conversion of the inactive form of Rab-3A into the active form (56), is also severely downregulated in cells treated with IL-1β or IL-1β + INF-γ for 24 h (Table 1, item 2.0). This may further aggravate an eventual decrease in insulin secretion mediated by cytokine-induced Rab-3A expression.

Cytokines downregulated expression of mRNAs encoding receptors for the incretins cholecystokinin (CKK)-A, GIP receptor (confirmed by RT-PCR) (Fig. 1), and GLP-1 receptor (Table 1, item 4.0). The roles of GLP-1, GIP, and CKK-A as in vitro and in vivo potentiators of insulin release, via cAMP generation (GIP and GLP-1) and protein kinase C activation (CKK-A), have been confirmed in experimental models, and inhibition of their respective receptors decrease insulin secretion after food intake (57). On the other hand, the fact that cytokines increase expression of the mRNAs for glucagon receptor and downregulate expression of mRNAs for somatostatin, receptor type 2 (Table 1, item 4.0), may to some extent prevent the intracellular decrease in cAMP putatively caused by the downregulation of incretin receptors. The expression of mRNAs encoding receptors for GH and PRL (both confirmed by RT-PCR) (Fig. 1) were also decreased after cytokine treatment. In vivo and in vitro studies have shown that GH and PRL increase mitotic activity in islet cells and stimulate insulin release (58). During pregnancy, upregulation of both PRL and GH receptors contributes to the compensatory increase in β-cell mass and insulin secretion (59). Thus, the observed cytokine-induced downregulation of receptors for incretins and growth factors may hamper both β-cell function in vivo and decrease the ability of these cells to compensate for the progressive immune-mediated β-cell loss.

Several chemokines, cytokines, and cell adhesion molecules were induced by either IL-1β alone or IL-1β + INF-γ (Table 1, items 5.0 and 8.0). Chemokines and cytokines were already highly induced after a 6-h exposure to IL-1β + INF-γ and, with two exceptions, their expression was maintained after a 24-h exposure to IL-1β + INF-γ. IL-1β alone induced expression of four cytokines and chemokines (mob-1 [human IP-10 confirmed by RT-PCR] [Fig. 1], CINC-1, MCP-1, and MIP-3α [confirmed by RT-PCR] [Fig. 1]), while the addition of INF-γ both potentiated the effects of IL-1β and induced expression of five additional chemokines or cytokines, namely fractalkine, osteoprotegerin (both confirmed by RT-PCR) (Fig. 1), macrophage inhibiting cytokine-1, IL-15, and IL-6. The IL-1β–induced expression of CINC-1, CINC-3, and MCP-1 by rat β-cells was previously observed by differential display with RT-PCR (16). MCP-1 attracts mononuclear cells, and recent data indicate that IL-1β also induces MCP-1 mRNA and protein expression in human islets, and that the chemokine is present in pancreatic islets of prediabetic NOD mice (M.-C. Chen, P. Proost, C. Gysemans, C. Mathieu, and D.L.E., manuscript submitted for publication). Besides these previously described chemokines, new cytokine-induced chemokines were observed, including mob-1, fractalkine, and MIP-3α. Mob-1 is a specific chemoattractant for T-helper 1 cells (60) and contributes to autoimmune diseases such as systemic lupus erythematosus (60), autoimmune encephalomyelitis (61), and autoimmune neuritis (62). Fractalkine has adhesive and chemoattractant properties for IL-2–activated NK-cells and CD8⁺ T-cells (63), whereas MIP-3α has a role in the migration of dendritic cells (64) and was shown to induce adhesion of memory T-cells to ICAM-1 (65) (note that ICAM-1 is also highly induced by cytokines) (Table 1, item 8.0). Among the cytokines expressed in β-cells, IL-15 and IL-6 are of special interest. IL-15 is a potent growth factor for T-, B- and NK-cells, a T-cell chemoattractant, an enhancer of the cytolitic function of effector T- and NK-cells, and a potent inducer of INF-γ production by NK-cells (66,67). The expression of this cytokine was confirmed by RT-PCR (Fig. 1). The expression of IL-6 by β-cells in response to INF-γ and TNF-α has been previously demonstrated (68), and transgenic mice overexpressing IL-6 in the β-cells develop insulitis (69). These results suggest that β-cells exposed to IL-1β and/or INF-γ express several chemokines, cytokines, and adhesion molecules that may potentially contribute to the homing, adhesion, and activation of mononuclear cells in the course of insulitis.

Cytokines also induced several genes related to antigen presentation in β-cells. This was mostly an effect of INF-γ, because IL-1β alone induced only two of these genes, whereas a combination of IL-1β + INF-γ induced 14 genes (Table 1, item 7.0). Previous studies also indicated that INF-γ is the main inducer of MHC class I mRNA and protein in rat and human islet cells (70). We presently observed that IL-1β + INF-γ upregulated several components of the “machinery” for MHC class I antigen presentation (71), including mRNAs for several MHC class I–related components, proteasome subunits, and both MTP-1 and MTP-2 (Table 1, item 7.0). MTP-1 and -2 genes encode proteins that transport peptides (released from the proteasome) from the cytosol to the endoplasmatic reticulum, where they are loaded for MHC class I presentation (71).

It was previously demonstrated that IL-1β induces activation of extracellular signal–related kinase (ERK)-1/2 and p38 mitogen-activated protein kinase (MAPK) in whole islets (72) and in purified β-cells (13). We presently observed that cytokines upregulate both p38 (mean increase 2.9 by IL-1β + INF-γ for 24 h; data not shown) and ERK-3 (Table 1, item 6.0). There were also modifications in the expression of several of mRNAs encoding enzymes, which may affect signal transduction by MAPKs. Thus, there was a decrease in the expression of MAPK-2 and MAPK phosphatase and an increase in CL100 protein tyrosine phosphatase (Table 1, item 6.0).

Cytokines lead to up- and downregulation of several transcription factors and associated proteins. Among them, the expression of the NF-κB inhibitor I-κBα was highly upregulated after a 6-h exposure by IL-1β + INF-γ (Table 1, item 9.0). After 24 h, the level of expression of this mRNA was still clearly above control levels. IL-1β alone also upregulated I-κBα but to a lesser extent. In parallel, the NF-κB–p105 gene, a precursor of both NF-κB–p50 subunit and the repressor I-κBγ (73), was upregulated at both 6 and 24 h (confirmed by RT-PCR) (Figs. 1 and Table 1, item 9.0). The transcription factor NF-κB is formed by homodimers or heterodimers of Rel/NF-κB proteins, most commonly p50/p65 (74). In nonstimulated cells, NF-κB is sequestered in the cytoplasm associated with the inhibitory molecule I-κB (74). β-cell exposure to IL-1β leads to degradation of I-κBα and nuclear translocation of NF-κB (15). Interestingly, both NF-κB–p105 and IκBα genes are responsive to NF-κB stimulation, which may explain the presently observed increase in the expression of these genes after cytokine exposure (75). The increased expression of I-κBα mRNA and consequent translation of the protein probably functions as a negative feedback for NF-κB activation. Indeed, newly synthesized I-κBα can enter the nucleus, remove NF-κB from the DNA, export the complex back to the cytoplasm, and thus restore the original latent state (74). This negative feedback may explain the decrease in iNOS expression after 24 h in spite of the continuous presence of stimulating cytokines (Fig. 1). Besides I-κBα and NF-κB–p105, we observed expression of 19 additional cytokine-modified genes that are putative target genes for NF-κB. These genes are: AS (increased), iNOS (increased), c-myc (increased), mtp-1 (increased), ICAM-1 (increased), MHC class I (increased), IL-15 (increased), IL-6 (increased), mob-1 (increased), MCP-1 (increased), CINC-1/3 (increased), IRF-1 (increased), COX-2 (increased), Osteoprotegerin (increased), MnSOD (increased), HO-1 (increased), and cyclins D1 and D3 (decreased) (41,75). This places NF-κB as a central transcription factor in the process of cytokine-induced β-cell gene expression. In line with this, recent data from our group suggest that blocking NF-κB with an I-κB super-repressor prevents cytokine-induced β-cell apoptosis (H. Heimberg, Y. Heremans, C. Jobin, R.L., M. Darville, D.L.E., manuscript submitted for publication). The transcription factors C/EBPβ and C/EBPδ were also upregulated by cytokines (Table 1, item 9.0) The microarray results for C/EBδ were confirmed by RT-PCR (Fig. 1). C/EBPβ and δ can interact with NF-κB, and both are involved in cytokine-induced MnSOD and Fas expression in pancreatic β-cells (76) (M. Darville and D.L.E., unpublished observations).

The induction of putative β-cell defense/repair genes was more evident after 24-h exposure to IL-1β + INF-γ than after exposure to IL-1β alone (Table 1, item 12.0). This is in line with previous observations suggesting that a combination of IL-1β + INF-γ induces more severe β-cell damage than IL-1β alone (4,5). Besides previously described cytokine-induced genes, such as hsp-70 and MnSOD (17), we observed induction of hsp-27, hsp-40, GADD-153 (confirmed by RT-PCR) (Fig. 1), O-6 methylguanine-DNA methyltransferase (MGMT), gas-5 growth arrest homolog, and methallothionein and MX1 gene (Table 1, item 12.0). Several of these genes are involved in DNA repair and are probably a β-cell response to nitric oxide–induced DNA damage (5). Upregulation of these genes may explain the protective effects of a short-term (24 h) β-cell exposure to low concentrations of IL-1β against a subsequent assault by alloxan, streptozotocin, or NO (37). However, this increased expression of defense/repair genes is not sufficient to prevent β-cell apoptosis after a prolonged (6–9 days) exposure to IL-1β + INF-γ. This may, at least in part, be caused by the fact that cytokines also downregulate important defense/repair genes, such as gas-6 growth arrest (a ligand of receptor tyrosine kinases AXL, Sky, and Mer that have protective effects against apoptosis caused by serum deprivation, myc overexpression, and TNF-α in NIH3T3 cells) (77) and glutathione peroxidase (an antioxidant enzyme) (Table 1, item 12.0).

Only one gene directly related to apoptosis regulation was found modified more than threefold by cytokines in the microarray analysis. Thus, the death protein-5 was increased at 24 h by both exposure to IL-1β alone and IL-1β + INF-γ. Death protein-5 is a gene induced in neuronal apoptosis, and it possesses a BH3 domain that allows interaction with Bcl-2 and Bcl-xl (78). It will be of interest to investigate whether this gene has a similar role in β-cells. The lack of observed cytokine-induced changes in several of the known pro- and antiapoptotic transcripts present in the Affymetrix array (such as Bcl-2, Bcl-x, Bad, and caspases) raises concern whether the time points selected for the present analysis (6 and 24 h) were the most adequate to detect modifications in these genes. Cytokine-induced β-cell death increases mostly after 3–6 days (10,11,12,13,14), and we cannot exclude that an array analysis performed after 36 or 48 h of cytokine exposure would detect modifications in the classic pro- and antiapoptotic genes. An alternative possibility is that β-cell death is not determined by changes in the known pro- and antiapoptotic genes but instead depends on modifications in the expression of several genes related to maintenance of β-cell phenotype and function. Additional experiments are required to clarify this issue.

A model of the effects of cytokines in pancreatic β-cells, based on the present array analysis, is provided in Fig. 2. Cytokines decrease the expression of several genes related to differentiated β-cell function and preservation of β-cell mass, including insulin, GLUT2, glucokinase, and diverse receptors for incretins and growth hormones. This loss of β-cell–specialized functions is probably associated to cytokine-induced decrease in the expression of Isl-1 (present data) and Pdx-1 (37) and may explain the impaired glucose-induced insulin release observed in prediabetic NOD mice (79). On the other hand, there is upregulation of stress-response genes and, surprisingly for such a differentiated cell, expression of genes encoding several chemokines, cytokines, and adhesion molecules, several having the potential to increase mononuclear cell homing and activity during insulitis. Most of these genes are probably upregulated as a consequence of NF-κB activation. Note that NF-κB–responsive genes may also directly contribute to β-cell apoptosis (H. Heimberg, Y. Heremans, S. Jobin, R.L., A.K.C., M. Darville, D.L.E., manuscript submitted for publication).

The present findings open new avenues for research on the mechanisms of immune-mediated β-cell death, and we intend to characterize in detail the function of several of the novel β-cell genes, both at the mRNA and protein levels. The picture that emerges from the present data is that β-cells are not passive bystanders of their own destruction. They respond to immune-mediated damage by triggering complex patterns of gene expression, with some of these genes aggravating β-cell damage, whereas others probably contribute to cell defense/repair (Fig. 2). At some point in this response, the balance is tilted toward β-cell death. A more detailed characterization of the gene patterns described in the present study and of the transcription factors regulating them may allow us to understand what tilts the balance in this direction. Hopefully, this knowledge will point to novel targeted approaches to improve β-cell survival in early type 1 diabetes.

This work was supported by grants from the Juvenile Diabetes Foundation International, the Fond for Scientific Research Flanders, and the Karen Elisa Jensen Fond.

The assistance of the Diabetes Research Center personnel involved in β-cell purification and that of Meng-Chi Chen and Hanne Steen is gratefully acknowledged.