Gene Expression Profiles Define a Key Checkpoint for Type 1 Diabetes in NOD Mice

NOD/LtJ, NOD.B10Sn-H2^b, and C57BL/6J (B6) mice were purchased from The Jackson Laboratory and then housed and bred under specific-pathogen-free conditions following the Committee on Animal Use for Research and Education (IACUC) guidelines at the University of Florida. The NOD.Idd3/Idd10 breeders were a gift from Dr. Edward Leiter. The NOD.B10Sn-H2^b and NOD.Idd3/Idd10 mice have been back-crossed for 15 and 11 generations, respectively. Only female mice were used in this study.

Total RNA (RNeasy Midi Kit, Qiagen) and then mRNA [Poly(A)Pure kit, Ambion] were isolated from multiple spleens of NOD and B6 female mice at 10 weeks of age. Three micrograms from each pool was used to create the library with the PCR-Select cDNA Subtraction Kit (BD Biosciences Clontech) (12) and then cloned into pCR2.1 vectors (TA cloning, Invitrogen). Each clone was amplified by PCR. Briefly, each reaction contained 22 μl 100 mmol/l dNTP, 11 μl 10× PCR buffer, 2.2 μl 20 pmol/μl TA-F primer (5′CCGCCAGTGTGATGGATATCTG), 2.2 μl 20 pmol/μl TA-R primer (5′TCCACTAGTAACGGCCGCCAG), 0.8 μl Taq polymerase, and 61.7 μl deionized water. This mixture was subjected to 40 cycles of 94°C for 30 s, 64°C for 30 s, and 72°C for 1 min. The PCR product was isopropanol precipitated then resuspended in 20 μl of 150 mmol/l sodium phosphate buffer (combine 1.1 ml 150 mmol/l NaH₂PO₄, 48.9 ml 150 mmol/l Na₂HPO₄, and 50 μl 10% SDS, then pH to 8.5).

Clear microscope slides were cleaned in a sodium hydroxide/ethanol solution (70g NaOH dissolved in 280 ml deionized water, to which 420 ml of 95% ethanol was slowly added) for 2 h. After a thorough rinsing, the slides were placed for 1 h in poly-l-lysine solution (560 ml deionized water with 70 ml poly-l-lysine and 70 ml PBS). The slides were briefly rinsed, centrifuged dry, and aged at least 3 days before printing. The MicroGrid TAS II (BioRobotics) was used to print the mouse array 1 (MAR1) microarrays that contained the 11,520 clones created above. Following printing, the slides were postprocessed by rehydrating, ultraviolet cross-linking (60 mJ), and then incubating in a blocking solution (335 ml 1-methyl-2-pyrrolodinone into which 6 g succinic anhydride is dissolved followed by 15 ml of 1 mol/l boric acid) for 15 min. Boiling water was used to denature the cDNA on each slide, followed by a brief rinse in 95% ethanol before drying by centrifugation.

Ten micrograms of total splenic RNA (RNeasy Midi Kit, Qiagen) was reverse transcribed to incorporate aminoallyl dUTP (Sigma) into the cDNA for each of the 117 individual mice (Table 1). The reaction was purified (QIAquick PCR Purification Kit, Qiagen), reduced to 5 μl volume, and then coupled to the monofunctional NHS ester Cy3 (Amersham Biosciences) for 1 h. The reference RNA, a pool of total RNA from 10 NOD and 10 B6 mice at 4 weeks, was processed in the manner described above and labeled with Cy5. The Cy3-labeled sample and Cy5-labeled reference were combined and purified (QIAquick PCR Purification Kit). The combined labeled product plus 15 μl 20× sodium chloride-sodium citrate, 1.8 μl Cot-1 DNA (0.1 μg/μl), and 2.25 μl 10% sodium dodecyl sulfate was applied to an MAR1 microarray and incubated at 65°C for 16 h. The arrays were washed briefly and then scanned using the Affymetrix 418 Scanner (MWG Biotech). Each of the 117 samples was compared with our reference using the MAR1 microarrays.

MolecularWare (Cambridge, MA) and ScanAylze programs were used to extract intensity values from the individual spots. The extracted data were flagged (13) and then uploaded into Another Microarray Database (AMAD), where the normalized intensity values were calculated for each array during submission. The data were extracted and divided into biologically relevant groups for analysis using a nonparametric Mann-Whitney U test (Statistica) to identify those genes that can best distinguish the groups. Hierarchical clustering was used to cluster genes with the similarity metric correlation (uncentered) and average linkage clustering in the Cluster program. The results were then viewed using TreeView.

Reverse transcription (RT) was performed from 3 μg total RNA from an independent set of 8 3-week, 10 6-week, and 11 10 week NOD mice using 2 μl poly T primer and sufficient diethyl pyrocarbonate (DEPC) water to total 10 μl. The RT reactions were placed at 70°C for 10 min and then on ice for 5 min. The following were added to each reaction and put into a thermocycler (MJ Research) at 42°C for 2 h before being diluted 1:1 with purified water: 1 μl 10 mmol/l dNTP, 2 μl 10× RT buffer, 1.5 μl RT enzyme (Stratagene), and 5.5 μl DEPC water. A control RNA pool was created (30 NOD samples above in equal amounts) and run as an internal control at the following dilutions: 1:1, 1:4, and 1:8. Each of the sample and control RT dilutions were set up in triplicate for real-time PCR analysis with 2 μl diluted RT, 2 μl 10× PCR buffer, 2 μl 10 mmol/l dNTP, 0.3 μl forward and reverse primer (20 pmol/μl), 0.5 μl SYBR green, 0.2 μl FDC, 0.13 μl Taq enzyme, and 12.37 μl water. Using the iCycler (BioRad), 35 cycles of 94°C for 30 s, 58°C for 30 s, and 72°C for 30 s were performed. PCR yields at the cycle threshold were calculated based on the standard curve. Twelve genes were investigated: Igj, Igh-6, B2m, Ddx5, Fech, Car2, Ms4a4b, Irf4, Rab1, Cd24a, Rik2610305D13, and Rik2310075C2. To normalize each gene, β-actin was amplified for each RNA sample.

To identify differentially expressed genes in immune tissues, a spleen cDNA library was created using a subtractive approach. Spleen RNA samples from the B6 and NOD mice were used to reduce the representation of abundant and common genes between NOD and B6 mice in such a way that genes with higher expression in the NOD than in the B6 mouse are enriched. However, this approach is unlikely to identify genes with weaker expression in NOD than in B6 mice. We elected to focus on the genes with higher expression in the NOD mouse because the autoimmune process is expected to induce hyperproliferation and activation of lymphocytes, leading to higher expression of a number of genes. From this library, 11,520 clones were selected, and the inserts were amplified to create the MAR1, which was used throughout this study.

We profiled the expression of 11,520 clones from a splenic cDNA library in 61 NOD mice ranging in age from 1 to 10 weeks. Table 1 presents the number of mice analyzed for each time point and strain. Hierarchical clustering analysis of the complete clone set identified two major groups of mice: immature (1–4 weeks) and mature (6–10 weeks). Over 1,000 clones exhibited differential expression patterns between the two groups of NOD mice. These clones were sequenced and BLAST (basic local alignment search tool) analysis revealed 362 unique genes. The nonparametric Mann-Whitney U test was then applied to this unique gene dataset, and 80 genes with known function were significantly different between the immature and mature NOD mice, with a P value of <10⁻⁴ (Fig. 1). Analysis of the 5-week-old NOD mice indicated that they have the immature expression profiles (Fig. 1).

The mean expression levels for the 80 genes in immature NOD mice (group 1) and mature NOD mice (group 2) are presented in Table 2. Please note that the expression levels have been normalized to the immature NOD group. Of the 80 differentially expressed genes, 62 have higher and 18 have lower expression levels in mature NOD compared with immature NOD mice. A number of the genes are lymphoid-specific, and many are involved in the normal cellular processes such as transcription, translation, DNA replication, signal transduction, and apoptosis.

Real-time PCR was used to confirm the expression differences revealed by the microarray technique. We included nine genes that have a P value of <5 × 10⁻⁵: immunoglobulin joining chain (Igj), immunoglobulin heavy chain 6 (Igh-6), β-2-microglobulin (B2m), DEAD box polypeptide 5 (Ddx5), ferrochelatase (Fech), carbonic anhydrase 2 (Car2), membrane-spanning 4-domains, subfamily A, member 4B (Ms4a4b), interferon regulatory factor 4 (Irf4), and CD24a. We also included two genes that showed differences but did not quite reach our stringent statistical criterion: RAB1 member of the RAS oncogene family (Rab1) (P = 0.0008) and Riken cDNA 2310075C12 (Rik231) (P = 0.0001). Finally, the Riken cDNA 2610305D13 clone (Rik261), which had a 3.8-fold difference by microarray, was analyzed by real-time RT-PCR even though it was not highly significant (P = 0.02). Splenic total RNA samples from 8 NOD mice at 3 weeks, 10 NOD mice at 6 weeks, and 11 NOD mice at 10 weeks were isolated and analyzed for these 12 genes (Table 3). Real-time PCR was able to confirm the microarray expression differences for all nine genes that have highly significant differences between immature and mature NOD mice. Real-time PCR was also able to confirm that Rik261 was not different between immature and mature NOD mice. However, the real-time PCR and microarray data disagreed for the two genes, Rab1 and Rik231, that did not reach the stringent statistical threshold. Indeed, the real-time PCR data were significantly different from the microarray data. The discrepancies between the real-time PCR and microarray data are most likely due to the low intensities on the microarrays for these two genes.

The switch in gene expression profiles around 5 weeks of age in NOD mice may be the result of two confounding biological processes: the progression to autoimmunity and the maturation of splenic cells. To dissect these two processes, we analyzed two NOD congenic strains that have little or no incidence of insulitis or diabetes: NOD.Idd3/Idd10 and NOD.B10Sn-H2^b (<1 and 0% incidence of diabetes, respectively). These mice were chosen as controls instead of B6 because they have the NOD background. Any differences seen are more likely to be due to the disease process than to strain differences. Figure 2 displays the expression profiles of the 56 congenic NOD mice for the 80 genes differentially expressed between immature and mature NOD. Interestingly, many of the genes also exhibit a similar temporal change in the congenic mice (Fig. 2A). Fifteen of the 80 genes did not show the same trend of temporal changes in NOD congenic mice (Fig. 2B). These 15 genes exhibiting temporal changes in NOD but not in NOD congenic mice are in boldface in Table 2 (compare the P values for immature and mature NOD [group 1 vs. 2] with the immature and mature NOD congenics [group 3 vs. 4]). These genes do not fall within one functional category, but are distributed across multiple functional groups. One of the genes, hypoxia inducible factor-1α (Hif1a), is upregulated in the NOD and downregulated in the NOD congenic mice.

To identify those genes that are different in the preinsulitis stage, the mean expression levels in the immature NOD (group 1) and the immature congenic (group 3) mice for the 80 genes were compared (1 vs. 3 in Table 2. The vast majority of the genes are not different between these two groups of mice. Twelve genes displayed a trend of lower expression in NOD when compared to the congenic mice (P < 0.05). Six of the 12 genes (Bsg, B2m, Ii, Igh-6, Igk-V28, and Igj) are lymphoid-specific and expressed in antigen-presenting cells. Among the 12 genes, only the Hif1a gene has a 1.8-fold higher expression in the congenics than the NOD with a reasonable P value (0.0008). Other confirmatory tests will need to be performed to determine whether any of these genes are truly different between the NOD and congenic NOD mice before the onset of insulitis.

Of the 80 genes that showed temporal changes in the NOD mice, 56 have a higher expression level in the mature NOD than the mature NOD congenic mice (Table 2, group 2 vs. 4). This increase is significant for 19 of the genes (0.001 < P < 0.05) and highly significant for 10 other genes (P < 0.001). These latter genes are involved in MHC class II processing (Hspa8), B-cell development, apoptosis control (Hif1a), transcription (Hnrpa2b1 and Taf10), translation (Eif4g2), and protein transport (Dnaja2). These data are consistent with the hyperactivated state of lymphocytes in the NOD mouse at the time of ongoing autoimmunity.

Because of its complex nature, it is a challenging task to understand the immunopathogenesis of type 1 diabetes. Since the identification of a spontaneous mouse model (NOD) two decades ago (14), our knowledge about the key players involved in type 1 diabetes has expanded exponentially. However, the underlying molecular mechanism of each of these key players and how they function together to create the disease phenotype remains elusive. The development of microarray technology has provided a new approach to understanding diabetes pathogenesis through the analysis of global gene expression profiles. In this study, we used microarrays to monitor the global splenic gene expression of NOD and NOD congenic mice, ages 1–10 weeks. The purpose of this experiment was to gain a better understanding of the molecular events that occur in the immune system of the NOD mouse during the development of insulitis [the infiltration of lymphocytes into the pancreatic islets that begins around 3 weeks (15)], but before the onset of overt diabetes. Because our microarray experiments analyzed the expression of over 10,000 clones simultaneously, a large number of genes may show differences due to random chance. Therefore, a stringent statistical threshold and appropriate gene selection method must be used to reduce the false-positive rate. In this study, we have generated real-time PCR data for a small number of genes selected based on the microarray data in order to select the most appropriate statistical methods for gene selection and establish the appropriate statistical threshold. The Mann-Whitney U test coupled with the statistical threshold of P < 5 × 10⁻⁵ seems to be the best compromise between higher sensitivity and lower false-positive rate. When this is applied to our microarray dataset, 100% of the genes tested by real-time PCR are concordant with the microarray data. Genes with a P > 5 × 10⁻⁵ may be considered, but the false-positive rate will increase at higher values. This should be kept in mind when interpreting any microarray data.

Our microarray dataset can be used to address several questions for autoimmune diabetes. The first question relates to the temporal changes of gene expression in NOD mice. Analysis of 61 NOD (1–10 weeks) splenic profiles revealed many genes rapidly changing expression levels between 5 and 6 weeks of age. This switch of gene expression defined two distinct groups of mice: immature (1–4 weeks) and mature (6–10 weeks). Among the top 80 differentially expressed genes, 62 had increased levels of expression in the mature NOD mice. These genes fall into a broad range of functional groups. The largest expression increases occurred in lymphoid-specific genes, suggesting that the splenic immune cells have increased cellular activities beginning around the 6-week time point. Specifically, significant increases were seen in B-cell genes such as Igj, Igk-V28, and Igh-6, suggesting higher antibody production. Increased antigen presentation is suggested by increased expression of MHC class I and class II antigen processing genes (Tra1, Hspa8, Hspa9a, Ii, and H2-Ab1) and β2-microglobulin (B2m). Interestingly, only a few T-cell-related genes (Ms4a4b and Bsg) are increased in the mature NOD mice. Several upregulated genes are involved in cell proliferation (Mki67 and Ddx5) and signal transduction in lymphocytes (Gnb2-rs1, Grb2, and Gnai2), suggesting increased lymphocyte proliferation and activation. Several genes (Hif1a, Ubl1, Bnip3l, Prdx2) are implicated in apoptosis control. Also, a large number of genes are related to basic cellular activities, which is not surprising given the higher proliferation and activation of splenocytes. These findings are consistent with the ongoing autoimmune response when NOD mice progress to invasive insulitis by 8–10 weeks of age (15). Our data suggest that an active autoimmune response is well underway by 6 weeks of age in the NOD.

The discovery of the rapid transition from immature to mature phenotype in the NOD spleens has significant implications for future research and may help explain the results from diabetes prevention studies. For example, administration of linomide (16), recombinant human interleukin-13 (17), or insulin (18–21) to 4- to 5-week-old NOD mice has been shown to significantly delay or prevent diabetes, along with a long list of other treatments (22). However, administration of one of these three substances to “mature” NOD mice is less effective. In light of our results, the mechanisms of action of linomide, human interleukin-13, and insulin could modulate the splenic immune cells, delaying or preventing the change that occurs at 6 weeks and thus delaying or preventing the onset of diabetes. This would explain why treatment after the 6-week transition results in reduced efficacy of the treatments.

Analysis of the two NOD congenic strains with near zero incidences of insulitis and diabetes, NOD.Idd3/Idd10 and NOD.B10Sn-H2^b, is critical to understanding the microarray data. Interestingly, the majority of the temporal changes observed in NOD mice also occur in the NOD congenic mice (Table 2. These results are not surprising given that many of the changes may be related to the normal development of the immune system. However, 15 genes were significantly different in the immature versus mature NOD mice but not significantly different between immature and mature NOD congenic mice (Figs. 2B and boldfaced genes in Table 2). Four of these genes (Hspa8, Sdcbp, Gnb2-rs1, and Hif1a) are lymphoid-specific or function in apoptosis or signal transduction. These genes may play important roles in the pathogenesis of type 1 diabetes.

The comparison between the mature NOD and diabetes-protective NOD congenic mice revealed very interesting findings. First, the vast majority of the genes in Table 2 have higher expression in the mature congenic mice (group 4 in Table 2) than in either immature NOD (group 1) or immature congenic (group 3) mice but lower expression than in the mature NOD mice (group 2). However, the differences between the mature NOD and congenic mice are not statistically significant for most genes. It is tempting to suggest that the propensity of developing type 1 diabetes is due more to quantitative differences than to qualitative differences. Ten genes are significantly different between the two groups at P < 0.001. These 10 genes include Hspa8, Hif1a, and genes involved in basic cellular functions such as transcription. The data, taken together, seem to indicate hyperactivation and proliferation of lymphocytes in the NOD mice with active autoimmunity.

A critical issue for the immunopathogenesis of type 1 diabetes is to identify the key molecules implicated in the initiation of autoimmunity. We attempted to address this question by comparing age-matched NOD and protective congenic mice before or right at the beginning of insulitis. We only found one gene (Hif1a) that is significantly higher in the congenics than in the NOD mice (1.8-fold increase, P = 0.0008). Further studies are required to assess the role of Hif1a.

Of the 80 genes identified, 7 are located within confirmed diabetes susceptibility regions. Two of the seven genes, β2-microglobulin (B2m) in the Idd13 region (23,24) and H2-Ab1 within the I-A locus in the Idd1 interval (25,26), have already been implicated in type 1 diabetes. The five other genes that map to type 1 diabetes susceptibility regions are heat shock protein 8 (Idd2), ubiquitin-like 1 (Idd5), retinoblastoma binding protein 4 (Idd9.1), proliferating cell nuclear antigen (Idd13), and interferon regulatory factor 4 (Idd14).

The retinoblastoma binding protein 4 (Rbbp4) functions as a mediator of chromatin assembly in DNA replication and repair (27). It is universally expressed, and there is no significant difference in expression between the NOD and congenic NOD mice for Rbbp4 (Table 2); therefore, Rbbp4 in not a likely candidate gene for type 1 diabetes even though it maps to the Idd9.1 region.

Heat shock protein 8 (Hspa8) has recently been shown to aid in the targeting of the invariant chain/MHC class II complex to endocytic compartments (28). Hspa8 is upregulated in the NOD mice from 1–4 to 6–10 weeks, but not in the NOD congenic mice (Table 2). The map position, pattern of differential expression, and function of this gene suggest that it is an ideal candidate gene within the Idd2 interval.

Ubiquitin-like 1 (Ubl1), also known as sentrin, GMP1, SUMO-1, or PIC1, was identified in 1996 by several different groups (29–33). Ubiquitin-like protein 1, NEDD8, and Apg12 are a newly discovered group of ubiquitin-like proteins that are involved in protein modification (34). Okura et al. (30) demonstrated that when Ubl1 is over- expressed, cells are protected from both anti-Fas/APO-1 and tumor necrosis factor-induced cell death. Ubiquitin-like protein 1 has also been shown to modify IκBα (35), making it resistant to degradation. Because IκBα inhibits NF-κB, an increase in ubiquitin-like protein 1 could decrease the expression of genes controlled by NF-κB, including genes involved in immune and inflammatory responses. Thus, overexpression of Ubl1 in a specific subset of splenic cells could indicate resistance to apoptosis or a decrease in immune response, both of which could significantly affect the pathogenesis of type 1 diabetes. Although Ubl1 is increasing in both the NOD and congenic NOD mice, it appears to have a higher expression level in the mature NOD mice (Table 2, group 2 vs. 4, P = 0.04) making it a good candidate gene for Idd5.

Interferon regulatory factor 4 (Irf4) is a lymphocyte-specific transcription factor involved in B- and T-cell function (36,37). The immature NOD mouse has the highest expression when compared with both the mature NOD and immature and mature congenic NOD mouse. The real-time data (Table 3) confirm that the immature NOD mouse has a higher expression than the mature NOD mouse. The location of Irf4 within a diabetes susceptibility region, its immune function, and its differential expression in NOD versus congenic NOD mice make Irf4 another excellent candidate gene for type 1 diabetes.

In summary, our studies have revealed a number of genes that are differentially expressed between NOD and NOD congenic mice at various time points. Some of these genes may play critical roles in the initiation and progression of type 1 diabetes. Two of the 80 genes exhibiting differential expression have previously been implicated in type 1 diabetes. Our study revealed another three excellent candidate genes, Hspa8 (Idd2), Ubl1 (Idd5), and Irf4 (Idd14). The data also allowed us to define a critical checkpoint (around 5 weeks) during the development of autoimmune diabetes in NOD mice. This checkpoint is marked by the maturation/activation of lymphocytes and a massive switch of gene expression. Detailed analysis of the molecular events in individual cell types occurring at the checkpoint should further shed light on the pathogenesis of the disease.

We gratefully acknowledge Dr. Mark Yang and Dr. James Yang from the University of Florida for their help with the statistical programs used in our data analysis and Dr. Ed Leiter from The Jackson Laboratory for generously providing the NOD.Idd3/Idd10 breeders. Funding was provided by a National Institute of Diabetes and Digestive and Kidney Diseases biotechnology center grant (NIH 5 U24 DK58778-02) and project 1 of program project 2P01 AI-42288-05 from the National Institutes of Health.