Discovery of Gene Networks Regulating Cytokine-Induced Dysfunction and Apoptosis in Insulin-Producing INS-1 Cells

Insulin-producing INS-1E cells, a kind gift from Prof. C. Wollheim (Center Medical Universitaire, Geneva, Switzerland), were cultured in RPMI 1640 (with Glutamax-1) supplemented with 10 mmol/l HEPES, 10% heat-inactivated fetal bovine serum, 100 units/ml penicillin, 100 μg/ml streptomycin, 1 mmol/l sodium pyruvate, and 50 μmol/l 2-mercaptoethanol (14,16). For the time-course microarray analysis, INS-1E cells (1.3 × 10⁶ cells/condition) were exposed for 1, 2, 4, 8, 12, and 24 h to recombinant human IL-1β (10 units/ml; kindly provided by Dr. C.W. Reynolds, National Cancer Institute, Bethesda, MD) plus recombinant murine IFN-γ (100 units/ml; Invitrogen, Baesley, Scotland), with or without 1.0 mmol/l of the iNOS inhibitor, NMA. The time points and the concentrations of cytokines and NMA used in the time-course analysis were selected based on our previous dose-response and time course studies on INS-1E cells, using cell survival, nitrite production and expression of iNOS, MCP-1, and insulin mRNAs as end points (Fig. 1) (data not shown). The time-course microarray experiments were performed on two separate occasions (i.e., two independent experiments) using a total of 6.2 × 10⁷ INS-1E cells from passages 64 and 65. Culture media were collected at different time points for nitrite determination (nitrite is a stable product of NO oxidation), which was performed spectrophotometrically at a 546-nm wavelength after colored reaction with the Griess reagent (17).

The percentage of viable, apoptotic, and necrotic INS-1E cells was determined after 2, 6, 12, and 24 h exposure to IL-1β + IFN-γ with or without NMA. For this purpose, INS-1E cells were incubated for 15 min with propidium iodide (PI) (10 μg/ml) and Hoechst HO 342 (20 μg/ml) (18). This method has been successfully used to evaluate apoptosis/necrosis in human, rat, and mouse β-cells (18–21). A minimum of 500 cells was counted in each experimental condition.

Total RNA isolated from INS-1E cells (at least 10 μg/sample) was used to prepare biotinylated cRNA. The marked cRNA was hybridized to the rat U34-A oligonucleotide array (Affymetrix, Santa Clara, CA) as previously described (8,10). A total number of 48 U34-A arrays were utilized in the present series of experiments. Analysis of differential expression was performed by the GeneChip Suite software (version 4.0.1). Arrays were normalized by global scaling, with the arrays scaled to an average intensity of 150. Samples from duplicate experiments (INS-1E cells from different passage numbers) were hybridized to increase the robustness of the results. Gene expression was considered as modified according to previously described criteria (8,10). Briefly, genes were considered as modified by cytokines when they fulfilled the following criteria for at least one of the six time points studied: 1) the mRNA was present in either control or cytokine-treated cells in both experiments, 2) the mean average fold change (experimental group versus control) was ≥2.5, and 3) the fold change in each individual duplicate was ≥2.0. We have used our own curated “β-Cell Gene Bank” to assign the filtered genes into their respective functional clusters. The ESTs that had homology to a known sequence were annotated using the Resourcerer 6.0 database (22). A cytokine-modified gene, according to the criteria described above (mean average fold change ≥2.5 and the fold change in each individual ≥2.0) was considered NO-dependent if its expression in the presence of cytokines was increased/decreased >50% by NMA (mean of two experiments) in at least one time point and increased/decreased >30% at a second time point (mean of two experiments). These criteria of considering genes as NO dependent is arbitrary and may result in the inclusion of genes that are only partially regulated by the radical. Clustering analysis was performed to classify the genes according to their temporal variation in response to cytokines. For this purpose, the log-transformed expression values of the cytokine-induced mRNAs (genes + ESTs) were mean and variance normalized. The preprocessed data were used as input data in the J-Express clustering software (23), and a distance matrix was created with the Pearson correlation distance measure. K-means clustering method analysis was performed to create 15 clusters after the initial observation of 15 patterns with the hierarchical clustering method (13). The k-means algorithm is explained in detail elsewhere (15,24).

RT-PCR was performed on poly(A)+ RNA as described (10). Primers used for PCR amplification were for notch-1 F-CTCACGCTGATGTCAATGCT, R-GTGTGGGAGACAGAGTGGGT (366 bp); delta-1 F-AAGGCCCGAGTCTGTCTACT, R-TGCTAACTCCGAGATGAACC (256 bp); jagged-1 F-AGCCTGTGAGCCTTCCTTAT, R-AAGCCACTGTTAAGACAGAGC (241 bp). The primers for glyceraldehyde 3-phosphate dehydrogenase (GAPDH) were described previously (8). The ethidium bromide-stained agarose gels were photographed under ultraviolet transillumination using a Kodak Digital Science EDAS 290 camera (Eastman Kodak, Brussels, Belgium). The abundance of the PCR products were assessed by Biomax one-dimensional image analysis software (Kodak), and mRNA contents were expressed as optical densities corrected for GAPDH.

Plasmid constructs containing the iNOS and MCP-1 gene promoters were prepared and studied as described previously (25,26). Transfected INS-1E cells were exposed to cytokines with or without NMA for 12–24 h (same concentrations as above). Luciferase activities were assayed with the Dual-Luciferase Reporter Assay System (Promega). Test values were corrected for the luciferase activity value of the internal control plasmid, pRL-CMV. The results for cytokine-exposed cells were expressed as a fold induction of the luciferase activity in control condition, taking control (no cytokine added) value as 1. NMA alone did not affect the promoter activity in any of the conditions studied (data not shown).

Results are given as means ± SE. Comparisons versus the respective control groups were performed using the Student’s paired t test or Wilcoxon’s signed-rank test, as indicated.

In the initial part of the study, we examined the effects of cytokines on INS-1E cell viability and NO production. Exposure of INS-1E cells to IL-1β + IFN-γ for 2 h and 6 h did not affect viability, but after 12 and 24 h, there was an increase in cell death by apoptosis, without significant changes in the percentage of necrotic cells (Fig. 1). The increase in INS-1E cell death was paralleled by progressive medium nitrite accumulation after 12 and 24 h (but not after 2 or 6 h, data not shown), indicating cytokine-induced NO production (Fig. 2). Addition of the iNOS blocker NMA prevented cytokine-induced nitrite production (Fig. 2) but did not protect INS-1E cells against apoptosis (Fig. 1). These findings indicate that cytokine-induced apoptosis in INS-1E cells is mostly NO independent. Similar observations were made in human and rodent purified primary β-cells (1,19,20), suggesting that the well-differentiated INS-1 cells are an adequate model for the present experiments.

For the microarray experiments, INS-1E cells were exposed to NMA, IL-1β + IFN-γ, NMA + IL-1β + IFN-γ, or left untreated (control) for 1, 2, 4, 8, 12, and 24 h. The microarray analysis was performed as previously described (8,10), utilizing the Affymetrix rat U34A oligonucleotide array containing around 8,700 probes (77% known genes and 23% ESTs). Around 4,000 (3,415–4,202) genes and ESTs were scored as present in each of the 48 conditions. To evaluate whether NMA by itself modifies INS-1E cell gene expression, we compared INS-1E cells exposed to NMA for different time points to their respective controls, i.e., INS-1E cells not exposed to NMA but cultured for the same time period. NMA induced changes in 48 probes at the different time points tested (data not shown), i.e., <0.6% of the genes detected as expressed. These genes were excluded from further analysis. These findings indicate that NMA alone has a minor effect on INS-1E cell mRNA expression and suggest that the most relevant effect of NMA on cytokine-induced gene expression is associated to inhibition of NO production (Fig. 2).

After filtering the expression values following the stringent criteria outlined in research design and methods, 936 mRNA probes encoding for 698 known genes and ESTs were detected as cytokine regulated (in the Affymetrix array, there are sometimes 2–4 distinct probes for the same gene or EST). We confirmed by RT-PCR eight selected genes detected as cytokine-modified by the microarray analysis, namely MCP-1, HO, insulin, iNOS, and isl-1 (data not shown), jagged-1, delta-1, and notch-1 (see below). Moreover, we confirmed that, as previously described for primary β-cells (8,10), cytokines induce a nearly 50% decrease in pancreatic duodenal homeobox factor (Pdx)-1 mRNA expression in INS-1E cells after 24 h, an effect prevented by NMA (data not shown).

The known genes were further assigned into functional clusters, as shown in Tables 1 and 2 (the total list of genes is provided in an online appendix available at http://diabetes.diabetesjournals.org). Of these genes, nearly half were NO dependent. The distribution of NO-dependent/independent genes was ∼50% in most functional groups of genes, with the following exceptions. 1) In the metabolism group, 60% of the genes are NO dependent. The most NO-dependent metabolism subgroups are amino acids (67%) and ATP (87%). 2) In the cell cycle group, 71% of the genes are NO dependent. 3) In the cytokine/chemokine and major histocompatibility complex (MHC)-related gene groups, 87 and 100%, respectively, of the genes are NO independent.

NO seems to induce a late (after 12 h) negative feedback for some genes, such as iNOS, argininosuccinate synthetase (AS), MCP-1, and IκB, which is prevented by NMA (Table 2; sections 1.2, 5.0, 9.0). We have previously observed an NO-mediated negative feedback during cytokine-induced iNOS expression in RINm5F cells (27). This negative feedback has been explained in other cell types by putative effects of the radical on NF-κB activation and promoter regulation (28,29). To test this possibility, we evaluated cytokine-induced iNOS and MCP-1 promoter activity using luciferase reporter constructs (25,26) in the presence or absence of NMA (same experimental conditions as in Fig. 1; exposure time to cytokines and/or NMA of 24 h). Cytokine-induced MCP-1 promoter activity was respectively 6.4 ± 2.0- and 4.7 ± 0.7-fold above control level in the presence or absence of NMA (results as means ± SE of three experiments), and for iNOS it was 25.5 ± 1.7- and 18.7 ± 2.9-fold, respectively, above control level in the presence or absence of NMA (results as means ± SE of three experiments). Thus, NMA did not modify cytokine-induced MCP-1 and iNOS promoter activity. NMA also did not interfere with promoter activity in the absence of cytokines (data not shown). These results suggest that the late NO-mediated negative feedback on iNOS and MCP-1 expression in INS-1 cells is not exerted at the level of promoter activity and may be related to mRNA stability and/or another level of regulation still to be determined.

A general analysis of cytokine-modified genes (Table 1) revealed that 19.6% of the filtered genes are metabolism related, making it the most prevalent group (Table 1; section 1.0). Surprisingly, the largest group of metabolism genes were those involved in lipid metabolism (5.4% of the total number of genes; Table 1; section 1.4), with nearly 60% of these genes classified as NO dependent. The transcription factor peroxisome proliferator–activated receptor-δ, which may enhance lipid uptake (30), was upregulated by IL-1β + IFN-γ after 4 h of exposure, and the increased expression was maintained up to 24 h (Table 2; section 1.4). In line with this observation, CD-36 (scavenger receptor class B) and adipophilin, both downstream targets of peroxisome proliferator–activated receptor-δ (30), were upregulated by IL-1β + IFN-γ (Table 2; section 1.4). Cytokines caused a parallel and late decrease in the expression of genes involved in free fatty acid β-oxidation (acetylcoenzyme A dehydrogenase, 1-3-oxacyl-CoA thiolase, peroxisomal enzymes, and carnithine palmitoyltransferase II) and cholesterol biosynthesis (cytosolic 3-hydroxy 3-methylglutaryl-CoA synthase and reductase and steroid 5α-reductase) (31) (Table 2; section 1.4). Moreover, cytokines induced genes that may lead to increasing exogenous free fatty acid apport (lipoprotein lipase and LDL receptor) and genes involved in lipid storage (adipophilin) (Table 2; section 1.4).

Cytokines decreased the expression of several genes related to differentiated β-cell functions and preservation of β-cell mass, including Pdx-1, Isl-1, insulin, GLUT2, glucokinase, and diverse receptors for incretins and growth hormones (10) (Table 2). An intriguing finding was the upregulation of jagged-1 and delta-1, with peak of expression after 8–12 h in an NO-dependent manner, followed by a progressive decrease at later time points (Table 2; 6.0). These findings were confirmed by RT-PCR in INS-1E cells (Fig. 3A and B) and primary β-cells (M.I.D. and D.L.E., unpublished data). Jagged-1 and delta-1 are two ligands of the notch receptors (32,33). Notch-1 is expressed in INS-1E cells but remains unchanged following exposure to cytokines as indicated by RT-PCR analysis (Fig. 3C). In good agreement with the RT-PCR data, notch-1 mRNA was detected as “present” but not changed by cytokines in the microarray analysis (data not shown).

It has been previously suggested (8,34) that β-cell exposure to elevated NO concentrations, either via NO donors or cytokines, leads to endoplasmic reticulum stress and consequent β-cell apoptosis. The endoplasmic reticulum stress response is based on the coordinated expression of several different genes. The present time-course microarray experiments provided interesting insights into the parallel up- and downregulation of this gene network (Table 2; section 13.0) (Fig. 4H) and their temporal correlation with β-cell apoptosis (Fig. 4A). Thus, we observed an important (>15-fold) increase in growth arrest and DNA damage (GADD)153/CHOP expression after 12 h, which was prevented by the iNOS blocker NMA (Fig. 4H). Notably, there was also an early (1–8 h) and more modest (two- to threefold) increase of GADD153/CHOP expression, which seems to be independent of NO formation (Table 2). The late increase in GADD153/CHOP may be secondary to sarco(endo)plasmic reticulum Ca⁺² ATPase type 2 b (SERCA2b) inhibition, which is first detected after 8 h exposure to cytokines, and is also NO dependent (Fig. 4H; Table 2). Cytokine-induced β-cell apoptosis increased in parallel with GADD153/CHOP expression (Fig. 4A) but, contrary to the observations with SERCA2b and GADD153/CHOP, apoptosis was not prevented by the iNOS blocker NMA.

Complex formation between pro- and antiapoptotic proteins from the Bcl-2 family may prevent apoptosis (35,36). Against this background, we examined the expression of known pro- and antiapoptotic genes from the Bcl-2 family (Table 2; section 13.0) (Fig. 4). Exposure of INS-1E cells to cytokines modified expression of several proapoptotic genes, including Bid, Bad, Bak, and Bcl-xs. The apoptotic inducer Bax was present in the array but was not changed by cytokines (data not shown). The two genes most consistently induced by cytokines were Bid and Bak. Both mRNAs increased after 8 h and maintained increased expression for up to 24 h. Although Bid induction was prevented by NMA (Fig. 4B), the iNOS blocker did not affect Bak expression (Fig. 4C). Interestingly, the proapoptotic protein Bcl-xs (Fig. 4F), a splice variant of Bcl-xL (Fig. 4E), showed a similar profile as Bcl-xL, including decreased expression at 4 h and subsequent NO-independent increase in expression at later time points.

Two genes that may participate in β-cell defense against cytokines are manganese superoxide dismutase (MnSOD) and IκB, both with the property of decreasing NF-κB activation (37,38). MnSOD and IκB mRNAs were already induced by cytokines after 1 h (Fig. 4I) in an NO-independent manner and maintained an increased expression during the 24-h follow-up. Despite the upregulation of IκB and MnSOD, and of other putative defense/repair genes such as HO, heat shock protein (hsp)70, Gas-6, and glutathione-S-transferase (Table 2; section 12.0), prolonged exposure of β-cells to IL-1β and IFN-γ culminates in apoptosis, as shown in Fig. 1.

To further examine the general pattern of cytokine-modified genes, all mRNAs described in online supplement S1 (available at http://diabetes.diabetesjournals.org) were reclassified based on their temporal pattern of gene expression, using the k-means clustering method (Fig. 5). (The complete list of genes in the different clusters is outlined in online supplement S2, available at http://diabetes.diabetesjournals.org.) To decide on the number of clusters to be used, we initially performed hierarchical clustering and observed 15 major profiles of gene expression. To further validate this selection, we performed k-means clustering using 14 or 16 clusters and obtained similarly shaped profiles (data not shown). Genes related to signal transduction and transcription factors were present in all clusters at similar proportions (mostly around 10–20% of the total number of genes). Examination of the clusters reinforces the notion that NO, which starts to be synthesized in large amounts after 6–8 h, is crucial for late-expressed genes. Thus, the two most “NO-related clusters” were clusters 6 and 12 with 75 and 60%, respectively, of the genes listed as NO dependent (Fig. 5) (online supplement S2). In both cases, the major variation in gene expression, either stimulation (cluster 6) or inhibition (cluster 12), was observed between 8–24 h. Cluster 6 contains a large proportion of MHC-related genes, whose expression peaks at around 8–12 h. These genes are mostly IFN-γ–induced and NO independent (10). The other cluster containing a high proportion of MHC-related genes is cluster 11, with a transitory peak of gene expression at 8–12 h and with only 32% of the genes as NO dependent. If we remove the HLA-related genes from the calculation in cluster 6, the number of NO-dependent genes climbs to 86%. Clusters 6 and 12 also contain a high proportion of “defense/repair” (around 9% of the total) genes but not of apoptosis-related genes. The highest number of apoptosis/endoplasmic reticulum stress-related genes (8.2%) is located in cluster 8, a cluster characterized by early (2 h) and transitory increase in gene expression. This cluster also contains the highest number of genes related to cell cycle (six genes). As an example of apoptosis-related genes, both caspase 3 and voltage-dependent anion channel (VDAC) are present in cluster 8 (online supplement S2; section 13.0). They have an NO independent peak of expression after 2 h (3.7- to 8.1-fold increase) and return to basal levels after 4–8 h. Most cytokines, chemokines, and adhesion molecules are classified in the same clusters (clusters 5 and 10).

We have currently detected nearly 700 cytokine-affected mRNAs in INS-1E cells. To a major extent, these observations broaden the pool of known cytokine-modified genes in insulin-producing cells, in addition to providing for the first time a comprehensive view of cytokine-induced and NO-dependent genes in these cells. Of these genes, nearly 50% were NO dependent, emphasizing the role for this radical in the late (i.e., after 6–8 h, the time point when increased nitrite accumulation is first detected) (39) (data not shown) effects of cytokines on insulin-producing cells. Synthesis of NO in primary β-cells is regulated by an “NO production module” (8,10,40), which includes, among others, iNOS (upregulated after 4 h, with additional increase after 8–12 h, followed by an NO-dependent decrease after 24 h) and AS, an enzyme that recycles citrulline into arginine, allowing continuous NO production (40,41). AS was also upregulated by cytokines with a similar temporal pattern as iNOS (Table 2; section 1.2). We identified another potential component of the “iNOS production module” in insulin-producing cells, namely the enzyme guanosine triphosphate (GTP) cyclohydrolase I (GTPCH). GTPCH is the initial and rate-limiting enzyme in the production of tetrahydrobiopterin, which serves as a cofactor for iNOS in pancreatic islets (42) and other tissues (43,44). Notably, increased expression of GTPCH has been observed in the target tissues of an animal model of autoimmune arthritis (45), and it will be of interest to evaluate whether this is also the case for islet cells from nonobese diabetic mice and BB rats.

Microarray analysis is a precise and reproducible technique to measure the dynamics of a genetic network at the mRNA level (46,47). The rate of confirmation by semiquantitative RT-PCR for 43 cytokine-modified genes detected in our previous (8,10) (J. Rasschaert, D. Liu, A.K.C., B.K., M.K., T.Ø., and D.L.E., unpublished observations) and present microarray studies is >90%. This rate of confirmation increases to >95% with the use of real-time PCR (B.K., A.K.C., M.I.D., M.K., N.M., T.Ø., and D.L.E., unpublished observations). An important issue is whether the observed modifications in mRNA expression correspond to changes in protein expression. This is often the case, as suggested by the following observations. 1) Studies on the expression of mRNA and protein/enzyme activity for specific cytokine-modified genes in β-cells, such as insulin, MnSOD, AS, ornitine decarboxylase, Pdx-1, hsp70, iNOS, A20, MCP-1, IL-15, MIP-3α, and IP-10 showed a good correlation between these parameters (8–10,40,48–55). 2) Total protein biosynthesis is not inhibited in rat β-cells or mouse islets exposed to cytokines for 24 h (50,56). If mRNA and protein expression often correlates well, as mentioned above, how can we explain that proteomic studies on IL-1–exposed neonatal rat pancreatic islets (57,58) failed to identify most of the genes described as modified in our present and previous microarray analysis (8,10)? Notably, these proteomic analyses (57,58) also failed to detect many cytokine-induced proteins previously identified as modified by Western blot and enzyme-linked immunosorbent assay, including abundantly expressed proteins, such as insulin (50), iNOS (51), MnSOD (48), GADD153/CHOP (A.K.C., F. Ortis, and D.L.E., unpublished observations), and the chemokines MCP-1 (52), IL-15 (9), MIP-3α (9), and IP-10 (9). This suggests that the proteomic approach utilized in these previous studies (57,58), although clearly of interest, is not yet sensitive enough to allow adequate comparison with the sensitive and reproducible microarray approach.

We have previously observed by microarray analysis and RT-PCR that cytokines decrease the expression of several genes related to differentiated β-cell functions and preservation of β-cell mass, including Pdx-1, Isl-1, insulin, GLUT2, glucokinase, and diverse receptors for incretins and growth hormones (8,10). Our present data confirm and extend these findings, providing a broader picture of the genes potentially involved in β-cell dedifferentiation during an immune-mediated assault. An intriguing and novel finding was the upregulation of jagged-1 and delta-1. Jagged-1 and delta-1 are two ligands of the notch receptors (32,33), and we observed that notch-1 is expressed in INS-1E cells but remains mostly unchanged following exposure to cytokines. Notch signaling plays an important role in the control of embryonic endodermal endocrine development (59). This pathway activates the expression of Hes genes, encoding transcription repressors, which in turn inhibit expression of genes promoting endocrine differentiation, such as neurogenins (33). Differentiation is repressed in the endocrine precursor cells expressing the activated notch receptors, whereas the signaling cells are free to express neurogenins and differentiate into endocrine cells (33). In line with this model, expression of NeuroD1/BETA2 and NeuroD2, transcription factors that induce and maintain the differentiated state of insulin-producing β-cells, was downregulated in response to cytokines (Table 2; section 9.0). The targets of NeuroD1/BETA2, namely glucokinase (60), insulin (61), secretin (62), and glucagon (63), were also decreased in the present experiments (Table 2; sections 1.1 and 4.0). The cytokine-induced reexpression of jagged-1 and delta-1, and of other genes potentially involved in the notch signaling pathway, raises the question of the putative effects of this pathway in differentiated β-cells exposed to an autoimmune attack. Two possibilities are a contributory role for the loss of the differentiated β-cell phenotype and prevention, in a paracrine fashion, of the differentiation of newly generated β-cells formed as a compensatory response to progressive β-cell destruction (64). Of note, in another model of autoimmune disease, multiple sclerosis in mice, cytokine-induced reexpression of the notch pathway prevents maturation of oligodendrocytes and hence efficient remyelination of axons in the multiple sclerosis lesions (65).

The relative concentration of pro- and antiapoptotic proteins from the Bcl-2 family may decide cellular outcome following some proapoptotic stimuli (35,36). The two Bcl-2-related genes most consistently induced by cytokines in the present experiments were Bid and Bak. Bid, a proapoptotic protein, is cleaved by caspase 8 and increases mitochondrial permeability by releasing Bax-like factors from Bcl-2 and stimulating the oligomerization and membrane insertion of Bax and/or Bak (36). Bax forms tetrameric channels in the outer membrane of the mitochondria, allowing the release of apoptogenic factors such as cytochrome c from the mitochondrial intermembrane space (35,36). Many forms of cell death require either Bax or Bak (66). The proapoptotic activity of Bak is mostly neutralized by binding to Bcl-xL, whereas Bax is inhibited to a major extent by Bcl-2 (35,36). Taking this into account, we examined the ratios of Bak/Bcl-xL and Bax/Bcl-2 expression in cytokine-treated cells, utilizing the absolute values of expression obtained in the arrays (data not shown). The ratios of Bak/Bcl-xL showed a sharp increase after 4 h (13.2 and 15.1, respectively, as compared with 2.7 and 4.1, respectively, in control; n = 2 similar experiments), returning to basal levels at later time points. On the other hand, cytokines did not change the ratios Bax/Bcl-2 at any of the different time points studied (data not shown) and also failed to decrease Bcl-2 expression. Two other genes found upregulated in the initial hours of cytokine exposure and potentially related to apoptosis are caspase 3 and VDAC. Caspase activation is partially regulated at the transcriptional level, and upregulation of caspases contributes to cell death in chronic neurological diseases (67). VDAC is located in the outer mitochondrial membrane, and participates in the formation of the mitochondrial permeability transition pore, allowing the release of proapoptotic molecules and the sequential activation of caspases 9 and 3 (35,36). The observed early (2–4 h) upregulation of caspase 3, VDAC, and the ratio of the expression of Bak/Bcl-xL may predispose the β-cells to enter the apoptosis program, pending additional stimuli.

To further examine the general pattern of cytokine-modified genes, all mRNAs described in online supplement S1 were reclassified based on their temporal pattern of gene expression, using the k-means clustering method. K-means is an efficient clustering method to process large datasets (24), and it has been successfully used to determine time-course gene expression data in synchronized yeast (15) and in mouse adipocytes (68). Genes related to signal transduction and transcription factors were present in all clusters at similar proportions (mostly around 10–20% of the total number of genes). This homogeneous dispersion is puzzling; we expected transcription factors to be mostly present in “early induced” clusters (1–2 h). If we bear in mind, however, that rodent β-cells continuously exposed to cytokines go through different functional phases, including stimulation in the first 1–3 h, functional inhibition after 6–8 h (69) and progressive cell damage after 8–12 h (1), it becomes evident that β-cells will need to continuously activate diverse transcription factors and signal transduction mechanisms to adapt to the continuous changes in internal homeostasis induced by cytokines. We also observed that two of the clusters, namely clusters 5 and 10, contain many genes previously described as NF-κB dependent (10). For instance, we have shown by detailed promoter studies that NF-κB regulates transcription of iNOS (present in cluster 5) (25), MnSOD (70) (present in cluster 10), and MCP-1 (26) (present in cluster 10). Moreover, the chemokine cytokine-induced neutrophil chemoattractant-1 and the cytokine IL-15, whose expression is at least partially prevented by an NF-κB blocker in β-cells (10), are also present in these clusters, as is also the case for IL-1α, tumor necrosis factor-β, and cyclooxygenase-2, three genes shown to be NF-κB dependent in other tissues (71). Temporal coexpression studies may provide useful information on the nature of the transcription factors regulating groups of genes relevant for β-cell fate, and we intend to perform “in silico” analysis (72) to search for binding sites for key transcription factors present in genes from the different clusters.

We have presently conducted the first systematic time course microarray and cluster analysis of cytokine-exposed insulin-producing cells in the presence or absence of the iNOS blocker NMA. The results obtained increased by more than twofold the number of known cytokine-modified genes, and also provide the first comprehensive analysis of cytokine-induced and NO-dependent mRNAs in insulin producing cells. This collection of genes and gene clusters is an exciting resource for researchers interested in understanding the functional inhibitory and proapoptotic effects of cytokines in β-cells. Moreover, these data provide novel and often surprising insights into the molecular patterns triggered in β-cells faced with a protracted immune assault.