and Immunity (2010) 11, 269–278 & 2010 Macmillan Publishers Limited All rights reserved 1466-4879/10 $32.00 www.nature.com/gene

ORIGINAL ARTICLE Specific expression signature associated with development of autoimmune type-I diabetes using whole-blood microarray analysis

F Reynier1, A Pachot1, M Paye1,QXu2, F Turrel-Davin1, F Petit1, A Hot1, C Auffray3, N Bendelac4, M Nicolino4,6, B Mougin1 and C Thivolet5,6 1Joint Unit Hospices Civils de Lyon-bioMe´rieux, Hoˆpital Edouard Herriot, Lyon, France; 2Fudan University Cancer Hospital-bioMe´rieux Laboratory, Shanghai, China; 3Functional Genomics and Systems Biology for Health, CNRS Institute of Biological Sciences, Villejuif, France; 4Division of Pediatric Endocrinology and Diabetology, Hoˆpital Femme-Me`re-Enfant, Lyon, France; 5Department of Endocrinology and Diabetes, Hoˆpital Edouard Herriot, HCL Lyon, France and 6INSERM, U870; INRA, U1235; INSA-Lyon, RMND; University Lyon 1, France

Understanding the pathogenesis of type-I diabetes (T1D) is hindered in humans by the long autoimmune process occurring before clinical onset and by the difficulty to study the pancreas directly. Alternatively, exploring body fluids and particularly peripheral blood can provide some insights. Indeed, circulating cells can function as ‘sentinels’, with subtle changes in occurring in association with disease. Therefore, we investigated the gene expression profiles of circulating blood cells using Affymetrix microarrays. Whole-blood samples from 20 first-degree relatives of T1D children with autoimmune diabetes-related antibodies, 19 children immediately after the onset of clinical T1D and 20 age- and sex-matched healthy controls were collected in PAXgene tubes. A global gene expression analysis with MDS approach allowed the discrimination of pre-diabetic subjects, diabetic patients and healthy controls. Univariate statistical analysis highlighted 107 distinct genes differently expressed between these three groups. Two major gene expression profiles were characterized, including type-I IFN-regulated genes and genes associated with biosynthesis and oxidative phosphorylation. Our results showed the presence of early functional modifications associated with T1D, which could help to understand the disease and suggest possible avenues for therapeutic interventions. Genes and Immunity (2010) 11, 269–278; doi:10.1038/gene.2009.112; published online 21 January 2010

Keywords: type-I diabetes; gene expression; microarray; type-I IFN signature

Introduction loss in predisposed subjects. In addition, recent obser- vations of persistent b-cells in the pancreas of patients Type-I diabetes (T1D) is an autoimmune disease that with longstanding diabetes4 are puzzling and suggest results from the destruction of insulin-producing b-cells that regeneration and destruction of b-cells is a lifelong in the pancreas. Antibodies against b-cell compo- process. Although most clinical cases of T1D appear nents can be detected in the sera of pre-diabetic subjects de novo, first-degree relatives of T1D patients have a (PT1D), indicating an ongoing autoimmune process relative risk of 7–89% to develop the disease depending during a long asymptomatic preclinical period.1 These on the degree of genetic concordance and antibody antibodies serve as markers of a b-cell-targeted immune status.5 Prediction of T1D is commonly based on the reaction, although b-cell damage remains generally number and titer of diabetes-associated autoantibodies, regarded as a process that is mainly T-cell-mediated.2 including anti-insulin (IAA), anti-tyrosine phosphatase Most of our understanding of the mechanisms leading to (IA-2) or anti-glutamic acid decarboxylase-65 (GAD65) selective b-cell loss is derived from animal models such antibodies.6 However, some relatives with multiple as the NOD mouse.3 However, this does not necessarily autoantibodies do not develop diabetes and the time to reflect the human situation and several questions remain develop clinical diabetes in antibody-positive relatives to be answered. Little is known about the genetic and/or cannot be calculated accurately, indicating that addi- environmental factors that modulate the rate of b-cell tional markers that directly reflect the balance between pathogenic and protective mechanisms are required.5 Genome-wide gene expression profiling using micro- Correspondence: Dr F Reynier, Joint Unit Hospices Civils de Lyon- arrays is a powerful technology. It was recently applied bioMe´rieux, Hoˆpital Edouard Herriot, 5 place d’Arsonval, Lyon to the prediction of the outcome of T1D7 and the 69003, France. 8 E-mail: [email protected] identification of genes and pathways dysregulated from Received 27 July 2009; revised 6 November 2009; accepted 11 the analysis of peripheral blood mononuclear cells of November 2009; published online 21 January 2010 T1D patients. However, this approach has never been Functional genomics in type-I diabetes F Reynier et al 270 Set 1 Set 2

682 49 363

144 314 190

431

Figure 1 A MDS of the three groups from the whole microarray data. On the basis of the normalized expression data, samples were represented in a three-dimensional Euclidean space. The Set 1 and 2 color indicates the different groups of samples; blue represents Figure 2 A Venn diagram of differentially expressed genes the healthy controls, green the pre-diabetic subjects and red the obtained from statistical analyses of the three groups (SAM test, diabetic patients. MDS, multidimensional scaling. false discovery rate at 5%). The figure represents the number of probe sets differently expressed between the three groups (healthy controls, pre-diabetic subjects, diabetic patients) and shared among employed directly on peripheral whole-blood cells. In the three analyses (Set-1, Set-2 and Set-1 þ 2). From the intersection this study, we have explored the hypothesis that gene (Set-1-Set-2-Set-1 þ 2), 107 genes were selected (see Supplemen- tary Table S1). SAM, Significant Analysis of Microarrays. expression profiling of peripheral whole-blood cells may provide new insights into the pathophysiology of T1D.

Results To obtain further details on the description of these 107 Transcriptomic analysis of whole blood genes, pairwise comparisons (controls versus T1D Multidimensional scaling. Taking each gene expression patients, controls versus PT1D, T1D patients versus profile as a whole (n ¼ 54 675 probe sets), we first used a PT1D), associated with a false discovery rate (FDR) of multidimensional scaling (MDS) approach to display the 0.1, were performed on the Set-1 þ 2 combination. Over- position of each T1D patient (n ¼ 19), PT1D (n ¼ 20) and all, the majority of the genes identified among the 107 control (n ¼ 20) sample in a three-dimensional Euclidean genes came from comparisons of the PT1D versus T1D space. In the MDS plot in Figure 1, the distance between patients and control versus T1D patients (See supple- the samples reflect their approximate degree of transcrip- mentary Table S1). tional similarity. PT1D, T1D patients and controls appear distributed into three separate groups, with few excep- Cluster and ontology analysis of microarray data. Follow- tions, suggesting different global gene expression patterns. ing the description of the 107 genes differentially expressed between the three groups, we analyzed the Identification of differentially expressed genes. To underline distribution of the expression patterns generated by the the most discriminant genes, the samples were randomly SAM test. K-means clustering allowed us to group divided into two experimental sets (Set-1: n ¼ 10 PT1D, the expression profiles into a limited number of clusters. 10 T1D patients, 10 control and Set-2: n ¼ 10 PT1D, 9 T1D The clearest representation of the expression profiles patients, 10 controls). Analysis was performed by within the data set was obtained with a number of studying independently each set as well as the combina- clusters k ¼ 2 (Figure 3a), with a stability of 0.90 for tion of both. For each analysis, invariant genes were cluster-1 and 0.94 for cluster-2. Cluster-1 corresponds to removed as described under section ‘Materials and 39 genes mostly upregulated in PT1D and expressed at methods’. Using bioinformatic and statistical analyses, lower levels in controls and particular T1D patients. In 4288, 4694 and 10 713 probe sets were selected from Set-1, contrast, cluster-2 contains 68 genes with expression Set-2 and Set-1 þ 2, respectively. After Significant Analy- decreased in PT1D and increased in T1D patients when sis of Microarrays (SAM) statistical analyses between the compared with that in controls. three groups (PT1D, T1D and controls), we focused our To draw a clear picture of the gene products in terms analyses on the 144 probe sets (107 distinct UniGenes) of their associated biological processes, we used the commonly found in Set-1, Set-2 and Set-1 þ 2 compar- DAVID Functional Annotation Tool Suite and the EASE isons (Figure 2; see Supplementary Table S1). Statistical score P-value to further classify the selected genes. power was 0.65 for Set-1, 0.55 for Set-2 and 0.88 for Set- Ontological analysis of the 107 genes revealed 16 1 þ 2, confirming the relevance of the combinations used. significantly enriched (GO) biological

Genes and Immunity Functional genomics in type-I diabetes F Reynier et al 271 1

0.5 16 response to biotic stimulus 20 7.10E-05, 4.30E-09 14 immune response 18 9.70E-05, 5.40E-08 14 0 defense response 18 3.90E-04, 2.10E-07 01020 30 40 50 60 70 80 90 -0.5 31 biosynthesis 32 5.0E-18, 9.40E-24 31 macromolecule biosynthesis 32 1.70E-16, 3.40E-22 -1 32 cellular biosynthesis 35 1.8E-13, 9.90E-18 32 biosynthesis 35 4.60E-12, 2.50E-16 C PT1D T1D 38 protein metabolism 49 1.08E-8, 8.10E-10 cellular protein metabolism 35 1.30E-07, 8.50E-9 1 45 cellular macromolecule metabolism 35 45 2.20E-07, 1.30E-08 macromolecule metabolism 39 51 4.20E-05, 1.40E-06 cellular metabolism 50 0.5 66 3.80E-03, 1.40E-05 oxidative phosphorylation 5 5 5.20E-03, 1.00E-03 primary metabolism 47 0 64 5.70E-03, 1.90E-04 metabolism 51 68 9.20E-03, 4.80E-05 physiological process 0 3.80E-02 88 -0.5 01020 30 40 50 60 70 80 90

-1

C PT1D T1D Figure 3 Classification of differentially expressed genes. (a) K-means clustering of the expression data (k ¼ 2) based on 107 genes. The y-axis represents the Z-score-normalized median expression of each probe set and the x-axis the group of subjects: healthy controls (C), pre-diabetic subjects (PT1D) and diabetic patients (T1D). The solid yellow lines represent the learned cluster centers. (b) Ontological analysis (http:// david.abcc.ncifcrf.gov) of the genes identified 16 significant GO biological processes (Po0.05 after EASE correction). Each biological term is presented by at least four UniGene entries, with EASE score for the 107 genes (gray), the genes of cluster-1 (red) and the genes of cluster-2 (blue). Numbers of genes per category are indicated. GO, gene ontology; C, healthy controls, PT1D, pre-diabetic subjects; T1D, type-I diabetes patients. processes (EASE score P-value o0.05) associated with 13 at least four UniGene entries (Figure 3b). The enriched GO categories included in cluster-1 are related to 12 immunological processes, with genes mainly involved in the interferon (IFN) response (IFI27, OASL, SERP- 11 ING1, ISG15, IFIT3, GBP1, IFIT2, IFIT1, OAS3, STAT1, RSAD2 and IFI44), while in cluster-2 cellular metabolism 10 (for example, biosynthesis: RPL7/9/23/31/35, RPS3A/ 7/10/17/21/24/27/27L; oxidative phosphorylation: 9 COX7B, COX7C, NDUFA4, NDUFB3, UQCRB, ATP5J)is more represented. 8 expression (log2 Affymetrix data) IFN signature analysis. We studied the distribution of Average IFN-induced genes mRNA 7 samples based on the 12 IFN-response genes selected C SLE T1D from GO analysis of cluster-1. After determination of a PT1D high threshold at 10 by distinguishing individuals IFN Figure 4 low Patient profiling based on IFN signature. A subgroup of from IFN , by calculating the 90 percentile limits of the pre-diabetic individuals showed increased expression of type-I IFN- controls, we identified six PT1D patients (30%) with response genes (IFNhigh), similar to patients with SLE. Each point an average expression level above the normal value, represents a single individual, with the average of microarray defining the IFNhigh group (Figure 4). Interestingly, expression data from all 12 type-I IFN-response genes, which are we observed also that PT1D (IFNhigh) have a mean present in cluster-1 (Figure 3a, IFI27, OASL, SERPING1, ISG15, IFIT3, GBP1, IFIT2, IFIT1, OAS3, STAT1, RSAD2, IFI44). The gray expression equal to or higher than systemic lupus box indicates the normal range within the 90th percentile erythematosus (SLE) patients known to be the paradigm confidence limits. Individuals outside the gray box are defined as of IFN-induced disease. the IFNhigh group. IFN, interferon; C, healthy controls; PT1D, To go further in the description of this biological pre-diabetic subjects; SLE, systemic lupus erythematosus patients; signature, Ingenuity Pathway Analysis (IPA) was con- T1D, type-I diabetes patients. ducted on the 12 genes, which composed the IFN signature. IPA is capable not only of constructing additional molecules not associated in the main list of associations of genes identified in our analysis (‘focus’ genes. All genes were found in the IPA knowledge genes), but also of predicting the involvement of database, and are labeled ‘focus genes’. On the basis of

Genes and Immunity Functional genomics in type-I diabetes F Reynier et al 272

Figure 5 The two gene networks (a, b) derived from the 12 genes, which composed the IFN signature, using the IPA software. Edges (gene relationship) are displayed with labels that describe the nature of the relationship between nodes (genes). The nodes are displayed using various shapes that represent the functional class of the gene product. Genes in red belong to the list of the 12 IFN-inducible genes. Genes in blue were integrated into the computationally generated networks on the basis of the evidence stored in the IPA knowledge memory indicating a relevance to this network. The pink arrows represent the direct and indirect interactions for genes of type-I IFN (IFN-a, IFN-b). Network-a (score 28, focus genes 10) and Network-b (score 4, focus genes 2) show central connection represented by IFN-a, IFN-b. IFN, interferon; IPA, Ingenuity Pathway Analysis.

these focus genes IPA generated two biological networks obtainable and provides a large biosensor pool in the identified around the IFNA, IFNB. The IFN signature was form of gene transcripts that respond to changes in the segregated into two separate networks. Network-a human body and its environment. Gene expression (score 28, focus genes 10; Figure 5a) and Network-b, patterns might be influenced by quantitative variations the less informative network (score 4, focus genes 2; of specific blood-cell subsets within subjects. However, Figure 5b).These findings provide evidence that type-I we did not observe any significant change in blood-cell IFNs rather than type-II IFNs are responsible for the counts in T1D patients versus controls (data not shown). increased expression of IFN-induced genes in PT1D. Consistently, a study describing gene expression in peripheral blood mononuclear cells showed that altera- Validation of microarray gene expression data. For a subset tions in the composition or activation status of cell of patients (n ¼ 28), 10 selected genes were tested subsets did not account for the observed differences in by quantitative reverse transcription-PCR (RT-PCR) gene expression between healthy and autoimmune (Table 1). A significant correlation was observed between populations such as T1D.10 We used the PAXgene Blood the expression levels of these genes, as determined by RNA system that stabilizes the transcriptome and DNA microarray hybridization and quantitative reverse provides a standardized and reproducible method using transcription-PCR (RT-PCR) analysis (Table 2). small volumes of whole blood adapted to the study of young children. The PAXgene sampling system has been promulgated in a number of recent clinical studies Discussion looking at blood gene expression patterns based on microarray analyses or RT-PCR.11,12 Moreover, we illu- To investigate the mechanisms underlying the patholo- strated that mRNA quantification in PAXgene blood gical process of T1D, we examined global gene expres- samples could help to describe the immune function and sion profiles in pre-diabetic siblings and newly status of patients with septic syndromes.13 Several diagnosed T1D children. Thus, we highlighted new additional elements can justify the use of whole blood pathways affected at the early stages of the disease. with PAXgene as study material. This procedure reduces We hypothesized that blood transcriptome reflects RNA degradation and prevents any post-sampling physiological and pathological events occurring in stimulation.14,15 In contrast, previous published studies inflamed islets, with blood cells acting as sentinels. showed that blood collection using EDTA tubes or Indeed, peripheral blood cells express a large proportion leukocyte purification was accompanied by profound of genes encoded in the ,9 which makes it changes in mRNA expression.15,16 possible to study clinically relevant changes in gene In this study, the conducted analyses have shown the expression in a reproducible manner. Moreover, periph- interest provided by the combined approach (Set-1 þ 2) eral blood is an ideal surrogate tissue as it is readily in comparison with that provided by individual sets (Set-

Genes and Immunity Functional genomics in type-I diabetes F Reynier et al 273 Table 1 qRT-PCR performance, parameters and primers

Gene symbol UniGene Efficiencya Errorb Primer sequencesc

IFIT1 Hs.20315 1.94 0.0034 50-CAACCATGAGTACAAATGGTGA-30 50-TGGTTGTCATGTTCTTCCTGC-30 IFI44 Hs.82316 1.96 0.0043 50-TCATTGAGCTCAGGAAGAGCT-30 50-TCCTATACTTCTCAGATATCCC-30 IFI27 Hs.532634 1.89 0.0121 50-ATGGTGCTCAGTGCCATGGG-30 50-TCCAGTTGCTCCCAGTGACT-30 OAS3 Hs.528634 1.66 0.0087 50-TTGACGCCCTAGGCCAGC-30 50-ATCTTGGTACACTGCTGGTAC-30 COX7B Hs.522699 1.98 0.0030 50-CGTCTCCAAGTTCGAAGCATT-30 50-ACTTGTGTTGCTACATATGTCC-30 COX7C Hs.430075 1.81 0.0125 50-GCCCTGGGAAGAATTTGCCA-30 50-AAATGCTCTTCATATCTGTTAAAT-30 NDUFA4 Hs.50098 1.86 0.0096 50-AAGCATCCGAGCTTGATCCC-30 50-ACTGAGTAGAACTTGTATTGATC-30 RPL31 Hs.469473 1.84 0.0025 50-TCCATGGAGTGGGCTTCAAG-30 50-TGGCACATTCCTTATTCCTTTG-30 RPS17 Hs.433427 1.91 0.0054 5-0AGATAGCAGGTTATGTCACGC-30 50-TGCCGAAGTCCAAAAGCTTCA-30 RPS24 Hs.356794 1.964 0.0022 50-CGCCATCATGAACGACACCG-30 50-ACAGGCCATGTCTTGCAAGTC-30

Abbreviation: qRT-PCR, quantitative reverse transcription-PCR. aAmplification efficiency of standard curve E ¼ 10À1/slope. bThe standard curve error value. cThe top sequence is forward primer and the bottom sequence is reverse primer.

Table 2 Correlation of gene expression analyzed by Affymetrix genes. IFNs are important immune system mediators and qRT-PCR that initiate or modulate autoimmunity and tissue damage through diverse actions on cells with immune Spearman correlation test function. Type-I and type-II IFNs can have both positive and negative effects in animal models of T1D.17 Gene symbol ra P-value In our analysis, we explored the relationships among IFN-induced genes by IPA and showed that type-I IFIT1 0.94 o0.0001 IFN signature was preferentially expressed in PT1D. IFI44 0.53 0.0036 Recently, upregulation of these genes was observed in IFI27 0.91 o0.0001 OAS3 peripheral blood of patients with others diseases like 0.87 o0.0001 18,19 20 21 COX7B 0.85 o0.0001 SLE, dermatomyositis, systemic sclerosis, multiple COX7C 0.91 o0.0001 sclerosis22 and rheumatoid arthritis.23 Our analysis NDUFA4 0.90 o0.0001 revealed a striking heterogeneity in the profiles of IFN- RPL31 0.88 o0.0001 regulated genes in PT1D. Approximately, 30% of PT1D RPS17 0.93 o0.0001 carry the type-I IFN signature (Figure 4) as compared RPS24 0.93 0.0001 o with half in SLE or in rheumatoid arthritis patients,18,23 suggesting that other pathways may contribute more Abbreviation: qRT-PCR, quantitative reverse transcription-PCR. a directly to the development of the organ-specific auto- Non-parametric Spearman correlation coefficient. immune diseases. Recently, it was shown that incubation of peripheral blood mononuclear cells of healthy blood donors with the sera of recent-onset T1D or PT1D results in the expression of IFN-induced genes, leading to the 1 and Set-2). In fact, false-positive and false-negative hypothesis that the IFN signature emerges years before rates are minimized when statistical power and sample the onset of T1D.7 In our study, type-I IFN gene size are increased. Moreover, interaction of the three expression was not detected in blood. This suggests that analyses (Set-1-Set-2-Set-1 þ 2) also significantly re- the type-I IFN gene is slightly expressed in blood. duced the risk of false positive by selecting genes that Indeed, it was demonstrated that the levels of IFNa, appeared independently of sampling. On the basis of this IFNa-producing plasmacytoid dendritic cells and the method, exploratory and confirmatory data analysis expression of IFNa-inducible genes in CD4 þ T-cells identified an autoimmune signature detectable years were increased in the pancreatic-draining lymph nodes before the onset of disease using a transcriptome of NOD mice.24 Those results observed in mice were approach using whole blood. confirmed in humans, with the detection of higher levels Many of the immune-response genes selected were of IFNa mRNA and protein in the pancreas of T1D grouped in cluster-1 (Figure 3a) composed mainly of patients as compared with those in controls.25,26 More- genes upregulated in PT1D. The most statistically over, the delayed onset and the decreased incidence of relevant gene family was represented by IFN-regulated T1D in NOD mice after blockade of IFNa receptor-1 by a

Genes and Immunity Functional genomics in type-I diabetes F Reynier et al 274 neutralizing antibody indicate that IFNa produced by plasmacytoid dendritic cells in the pancreatic-draining lymph nodes may be a primary initiator of the T1D 10 (%) ¼ 24 process. In SLE patients, neutralization of IFNa was n shown to prevent the maturation of dendritic cells and subsequently prevented the expansion of autoreactive T- 27

cells. Recently, clinical trials using monoclonal antibody 20 (%)

against IFNa have shown promising results in SLE ¼ patients,28 which could be extended to T1D; one possible n area where IFNa-neutralizing or IFNa-blocking antibody may be useful for treating autoantibody-positive or high- 59) ¼ emoglobin; NA, not available;

risk-genotype pre-diabetic individuals when they first n

develop anti-islet autoantibodies during the long pro- 19 (%) ¼

dromal phase of pre-diabetes. However, at this stage of n our study, we were not able to demonstrate that type-I IFN is an important player as in SLE or in rheumatoid arthritis.19,23 No statistical association was observed between the molecular stratification of PT1D (IFNhigh/ 20 (%) low IFN ) and the demographic and clinical characteristics ¼ of the PT1D presented in Table 3. n Hyperglycemia and/or exogenous insulin influence cytokine production from circulating immune cells;

however, the mechanisms are still unclear. To date, the 10 (%) ¼

influence of hyperglycemia on type-I IFN has not yet n been studied. However, Lang et al.29 have shown a profound inhibition of IFN-g expression in osmotically shrunken cells. We can assume that type-I IFN-inducible genes could be temporarily switched-off during the acute 29) Set-1+2 ( 9 (%) ¼ n glycemic stress period characterizing the onset of T1D. ¼ Thus, hyperglycemia by itself might explain the dis- n continuity of our results during the transition stage from pre-diabetes to diabetes. Another possibility might be that late b-cell failure results more from glucotoxicity and/or ER stress factors than from an immune-mediated

insult. 10 (%) ¼

In contrast, cluster-2 (Figure 3a) included genes mainly n associated with metabolic changes consistent with insulin deficiency and hyperglycemia, reinforcing the relevance of the results obtained using whole blood. In diabetes, cell growth is increased by high glucose 30 10 (%)

concentration. In our study, we identified ribosomal ¼ protein genes transcriptionally activated in the diabetic n group, reflecting increase in ribosome biogenesis. This observation is corroborated by a study that has 30) Set-2 (

showed the effects of high glucose concentrations on ¼ n

Saccharomyces cerevisiae. In fact, ribosomal protein gene 10 (%)

transcription is activated and ribosomal protein mRNA ¼ n degradation is transiently inhibited.31 These two effects Set-1 ( increase ribosomal protein mRNA abundance, thereby accelerating ribosome biogenesis. Additional evidence for metabolic changes in our study of T1D comprises 10 (%) NA 10.8 (9.5–11.7) NA NA 9.6 (8.1–13.2) NA NA 10.7 (9.4–12) NA NA

upregulation of genes of the oxidative phosphorylation PT1D T1D C PT1D T1D C PT1D T1D C SLE ¼ 12 (10–13) 9.5 (7.3–11) 8 (7.3–10.5) 11.5 (4.5–15.75) 12 (9–14) 8 (5–12) 12 (4–18) 10 (5–16) 7.5 (4–18) 37 (34–44) pathway. This observation was confirmed in different n studies showing that insulin deprivation in T1D results ) NA 17.4 (14.9–20.7) NA NA 18.1 (10.5–22.3) NA NA 18 (12.4–21.7) NA NA 1

in increased consumption, suggesting increased À rate of oxidative phosphorylation.32,33 The recent immunohistochemical analysis of a pan- (mmol l

creatic gland from a patient with longstanding diabetes a provided interesting clues to disease pathogenesis.4 Interestingly, the remaining b-cells were detected in

islets and within the exocrine pancreas in close vicinity a of T-lymphocytes or macrophages This is consistent Demographic and clinical characteristics of the subjects with the existence of a persistent autoimmune process a and the inability of b-cells to regenerate from ductal cell Median (Q1–Q3). Table 3 a KetosisKeto-acidosisAbbreviations: anti-GAD, glutamic acid decarboxylase antibodies; anti-IA2, protein tyrosine phosphatase IA2 antibodies; HbA1C, NA glycosylated h NA 4 (40) 6 (60) NA NA NA NA 2 (22) 6 (67) NA NA NA NA 6 (32) 12 (63) NA NA NA NA C, healthy controls; PT1D, pre-diabetic individuals; SLE, systemic lupus erythematosus patients; T1D, type-I diabetes patients. Anti-GAD65Anti-IA2HbA1C 9 (90) 6 (60) 8 (80) 8 (80) 0 (0) 0 (0) 10 (100) 7 (70) 5 (56) 8 (89) 0 (0) 0 (0) 19 (95) 13 13 (65) (68.4) 16 (84.2) 0 (0) 0 (0) NA NA Blood glucose levels precursors. Uncertainty about the ability of peripheral MaleFemaleAge (years) 4 6 (40) (60) 4 6 (40) (60) 5 5 (50) (50) 8 2 (80) (20) 5 4 (56) (44) 7 3 (70) (30) 12 (60) 8 (40) 9 10 (47.4) (52.6) 12 (60) 8 (40) 0 10 (0) (100)

Genes and Immunity Functional genomics in type-I diabetes F Reynier et al 275 blood to reflect the pathological process ongoing in the (IFNhigh). Additional blood samples of SLE, not age- and pancreas has hampered many studies of humans. sex-matched, were collected during medical follow-up of Currently, antibodies to b-cell antigens are the best their lupus. Selection of these patients was based on their validated markers to determine the risk of developing SLE Disease Activity Index (SLEDAI), median (Q1–Q3), diabetes.6 Peripheral T-cells that proliferate in the 13 (12–17.5). presence of GAD6534 or insulin have been detected at disease onset. Moreover, T-cell clones have been pro- Sample processing and microarray hybridization duced from peripheral blood mononuclear cells of newly Peripheral blood samples were collected in PAXgene diagnosed patients that are home to the pancreas of Blood RNA tubes (PreAnalytix, Hilden, Germany). This NOD–SCID mice after intravenous transfer, suggesting technology uses a reagent that immediately stabilizes that autoreactive T-cells circulate at the periphery.35 intracellular RNA and inhibits RNA degradation and Evidence shows that transcriptome analysis will offer gene induction as blood is drawn directly into the tube.15 new opportunities to identify genes that can be used for After sample collection, the PAXgene tubes were early diagnosis of diabetes. Recently, through the incubated at room temperature for 2 h and then stored combined approach of microarray profiling and genetic at À80 1C until RNA extraction. Extraction was performed linkage analysis, it has been shown that gene expression according to the manufacturer’s instructions. Briefly, RNA patterns can be inherited under a variety of conditions was isolated using the PAXgene Blood RNA kit (Pre- across a range of species.36,37 Consequently, we expect Analytix). Following cell lysis, nucleic acids were pelleted that the association of genetic, serological and transcrip- and treated with a buffer containing proteinase-K. After tomic data could improve the monitoring of pre- digestion with RNase-free DNase (Qiagen, Valencia, CA, diabetes. The present study provides new perspectives USA), the RNA was subsequently purified on PAXgene on the nature and extent of immune dysregulation, and spin columns and eluted in 80 ml of elution buffer. RNA lends further support for a central role of type-I IFN in integrity was assessed using RNA 6000 nanochips and T1D pathogenesis. Our data presented a strong rationale an Agilent 2100 Bioanalyzer (Agilent Technologies, for the development of new therapies to block IFN Waldbronn, Germany), according to the manufacturer’s pathways in human T1D. Finally, patterns of gene instructions. The RNA integrity number (RIN) software expression in blood cells may be useful in identifying was used to estimate the integrity of total RNA samples and monitoring those individuals most likely to benefit from the entire electrophoretic trace of the RNA from these therapies. sample.38 cRNA was synthesized using 3 mg of total RNA following the Affymetrix (Santa Clara, CA, USA) protocol. cRNA was transcribed in vitro by incorporating Materials and methods a biotinylated pseudouridine molecule using GeneChip Expression 30-Amplification Reagents for IVT Labeling, Ethics statement overv16 h at 37 1C (Affymetrix). Hybridization was There are no submissions to ethical committee. This performed using GeneChip Human Genome U133 Plus study is out of the scope of the French law related to 2.0 arrays (Affymetrix) containing 54 675 probe sets patients participating in clinical research. Good clinical corresponding to 38 500 identified genes. After washing, practices under ICH E6 (R1)/135/95 were implemented the chips were stained with streptavidin–phycoerythrin by the sponsor and the investigators involved in this according to the Affymetrix EukGE-WS2v5 protocol study. Blood samples were taken during usual biological using a Fluidic FS450 station. The microarrays were read monitoring. All patients and controls provided written with the GeneChip Scanner 3000 (Affymetrix). The informed consent for the collection of samples and Affymetrix GeneChip Operating Software version 1.2 subsequent analysis. (GCOS) was used to manage the Affymetrix GeneChip array data and to automate the control of the GeneChip Patient and control selection fluidics stations and scanners. Age and sex-matched recently diagnosed T1D (n ¼ 19), PT1D (n ¼ 20) and control (n ¼ 20) Caucasian children Data analysis were enrolled in the study (Table 3). Additional blood Expression data were generated by the Robust Multi- samples from T1D and PT1D children were taken during array Average (RMA) method39 implemented in the medical follow-up of their disease. T1D children were qAffy package of the Bioconductor microarray analysis antibody-positive for GAD65 and/or IA-2, and were environment (http://www.bioconductor.org). The RMA selected within 48-h after initiation of insulin therapy. method consists of three steps: background adjustment, PT1D children were siblings of T1D children. PT1D were quantile normalization40 and probe set summary of the obtained from the Rhoˆne-Alpes prospective diabetes log-normalized data by applying a median polishing registry and were selected by the presence of antibodies procedure. to GAD65 and/or IA-2. Additional blood samples from Initially, the normalized data were analyzed using an control children were collected from preoperative test approach based on MDS techniques and visualized using before a benign surgery (exostosis, testicular ectopia, the Spotfire Decision Site 8.2.1 for Functional Genomics hernia). The Controls were antibody-negative, without software (Spotfire, Gothenburg, Sweden). To gain insight any familial history of T1D, autoimmune disease or into the molecular pattern among the three classes, MDS infectious diseases, and had no concomitant medication. made it possible to obtain a projection of high-dimen- For the microarray experiments, each group was sional data into a lower dimensional space in order to randomly divided into two subgroups (Set-1 and Set-2; examine relationships between sample profiles. Controls, Table 3). In a second step, we incorporated SLE patients PT1D and T1D were placed in a three-dimensional (SLE, n ¼ 10) used as an IFN-positive control group Euclidean space, according to their expression patterns,

Genes and Immunity Functional genomics in type-I diabetes F Reynier et al 276 with the distance between samples reflecting their average expression of the IFN genes identified in the approximate degree of correlation. ontological analysis (Figure 3a; cluster-1; immuno- Before subsequent statistical comparisons between logical processes; IFI27, OASL, SERPING1, ISG15, IFIT3, groups, we applied filter based on variability. The GBP1, IFIT2, IFIT1, OAS3, STAT1, RSAD2, IFI44). We interquartile range was used as the filtering criterion. defined individuals showing altered IFN-response by Thus, we considered genes with an inter-quartile range calculating the 90 percentile limits of the controls below 0.5 as invariant genes. The genes differentially (normal values, defined by a non-parametric ranking expressed were identified by comparing the three groups method based on the log2 mean expression of the 12 (PT1D, T1D and controls) using the SAM method IFN genes). Thus, individuals below the threshold were (FDR ¼ 0.05; number of permutations: n ¼ 1000).41 This classified as IFNhigh and vice versa, individuals present- approach was applied to Set-1, Set-2 and Set-1 þ 2 ing expression measures below the threshold were separately, and only genes common between the three assigned as IFNlow. analyses were selected. In parallel, to assess the From the 12 genes, which comprised the IFN signa- reliability of the identified genes, post hoc analyses were ture, networks of the IFN genes were constructed using conducted to establish the statistical power of the study. IPA (www.ingenuity.com). Out of the 12 IFN genes, there The Sizepower package (version 1.8) from Bioconductor are those that were found in the IPA knowledge Environment was used to address power analysis for database, and are labeled ‘focus’ genes. On the basis each set.42 To go further into the description of genes of the focus genes, IPA generated a set of molecular identified as differentially expressed, three pairwise networks, with a cutoff of 35 genes for each network, comparisons (controls versus T1D, controls versus based on interactions between uploaded genes and all PT1D, T1D versus PT1D) were also performed using other genes/ stored in the knowledge base. Each the same approach (FDR ¼ 0.1; number of permutations: network is assigned a score according to the number of n ¼ 1000). focus genes in our data set. These scores are derived from To visualize the expression profiles of genes differen- negative logarithm of the P-value, indicative of the tially expressed among the three groups, and to likelihood that that focus genes are found together in a investigate their relationships, a k-means clustering network due to random chance. Scores of 4 or higher was performed with Spotfire Decision Site 8.2.1. This have 99.9% confidence level of significance as defined in method used the median expression levels adjusted to detail elsewhere.47 the Z-score-normalized data set. Cluster initialization was performed using the data centroid-based search parameter and the level of similarity was calculated mRNA expression analysis by quantitative real-time RT-PCR based on Euclidean distance. The k-means clustering was For a subset of patients (n ¼ 28), 0.5 mg of total RNA was applied using a k-value (that is, number of clusters) of 2, reverse transcribed into cDNA using the ThermoScript 3, 4, 5 and 6. To assess the reliability of each cluster, the RT–PCR system (Invitrogen, Carlsbad, CA, USA) accord- Clusterv package based on R 2.4 was applied to the ing to the manufacturer’s instructions. Quantitative PCR Z-score-normalized data set to redo the k-means cluster- analysis was performed using the LighCycler 2.0 (Roche ing.43 Estimation of stability was achieved through Diagnostics, Mannheim, Germany) using the Light- random projections of the original high-dimensional CyclerÀ FastStart DNA MasterPLUS SYBR Green I kit data to lower dimensional subspaces. These random according to the manufacturer’s instructions (Roche projections were repeated many times and each time a Diagnostics). Thermocycling was performed in a final

new clustering was performed. The multiple clusterings volume of 20 ml containing 3 mM MgCl2 and 0.5 mM of obtained were then compared with the initial clus- each of the required primers. PCR was performed with tering we wanted to evaluate. Intuitively, a cluster was an initial denaturation step of 10 min at 95 1C, followed considered reliable if it could be maintained across by 40 cycles of touch-down PCR protocol (10 s at 95 1C, multiple clusterings performed in the lower dimensional 10 s annealing at 68–58 1C and 16 s extension at 72 1C). subspaces. mRNA expression of the housekeeping gene peptidyl- Each gene set identified with the k-means clustering propyl isomerase-B (PPIB) encoding for cyclophilin-B approach was classified using an ontological method to was investigated using cDNA standards and ready-to- structure biological knowledge. For this purpose, the use primer mixes obtained from Seach-LC (Heidelberg, DAVID Bioinformatics Resources (http://david.abcc. Germany) as previously described.48 The cDNA stan- ncifcrf.gov) was used to enhance biological interpreta- dard for selected genes was prepared from purified tion.44,45 This tool incorporates information from differ- PCR amplicons. Primer designs and reaction perfor- ent public resources and provides an easy way to make mance parameters are provided in Table 1. LightCycler biological sense out of large sets of genes.46 In order to 2.0 was used to determine the crossing point for indivi- identify GO categories, the EASE score (modified Fisher dual samples. Serial dilutions of the cDNA standard exact P-value) was used for gene-enrichment analysis were prepared in quadruplicate to generate standard (significance level at 0.05). curves. Relative standard curves describing the PCR efficiency of selected genes and the housekeeping IFN molecular pathway analysis gene were created and used to perform efficiency- For IFN signature analysis, we used the IFN-response corrected quantification with the LightCycler Software genes (selected in the previous analysis: Set-1-Set-2- version 4.05. Gene expression values were expressed Set-1 þ 2). Before selection of IFN-response genes, ex- as a concentration ratio of target gene mRNA/PPIB. pression data were generated again by RMA method, Correlation between the concentration ratio and Affyme- including the fourth group composed of the SLE trix-normalized data was studied using Spearman patients. For each individual we calculated the correlation test.

Genes and Immunity Functional genomics in type-I diabetes F Reynier et al 277 Conflict of interest gene expression profiles with high density microarrays. J Immunol Methods 2003; 283: 269–279. The authors declare no conflict of interest. 15 Rainen L, Oelmueller U, Jurgensen S, Wyrich R, Ballas C, Schram J et al. Stabilization of mRNA expression in whole blood samples. Clin Chem 2002; 48: 1883–1890. 16 Hartel C, Bein G, Muller-Steinhardt M, Kluter H. Ex vivo Acknowledgements induction of cytokine mRNA expression in human blood samples. J Immunol Methods 2001; 249: 63–71. We thank Professors H Dodat, R Kohler and P 17 Baccala R, Kono DH, Theofilopoulos AN. Interferons as Mouriquand, and Dr CL Gay (Hospices Civils de Lyon), pathogenic effectors in autoimmunity. Immunol Rev 2005; Dr P Pelissier (CHU de St Etienne) and Dr C Bony 204: 9–26. (CH d’Annonay) for their contribution to the study. This 18 Baechler EC, Batliwalla FM, Karypis G, Gaffney PM, Ortmann work was supported by bioMe´rieux and grants from the WA, Espe KJ et al. Interferon-inducible gene expression ‘Association Nationale de la Recherche Technique’ signature in peripheral blood cells of patients with severe (ANRT) and the ‘Association Franc¸aise des diabe´tiques’ lupus. Proc Natl Acad Sci USA 2003; 100: 2610–2615. (AFD). 19 Kirou KA, Lee C, George S, Louca K, Peterson MG, Crow MK. Activation of the interferon-alpha pathway identifies a subgroup of systemic lupus erythematosus patients with distinct serologic features and active disease. Arthritis Rheum 2005; 52: 1491–1503. References 20 Baechler EC, Bauer JW, Slattery CA, Ortmann WA, Espe KJ, Novitzke J et al. An interferon signature in the peripheral 1 Bach JF. Insulin-dependent diabetes mellitus as an autoim- blood of dermatomyositis patients is associated with disease mune disease. Endocr Rev 1994; 15: 516–542. 13 2 Bach JF. Insulin-dependent diabetes mellitus as a beta-cell activity. Mol Med 2007; : 59–68. targeted disease of immunoregulation. J Autoimmun 1995; 8: 21 Tan FK, Zhou X, Mayes MD, Gourh P, Guo X, Marcum C et al. 439–463. Signatures of differentially regulated interferon gene expres- 3 Atkinson MA, Leiter EH. The NOD mouse model of type 1 sion and vasculotrophism in the peripheral blood cells of diabetes: as good as it gets? Nat Med 1999; 5: 601–604. systemic sclerosis patients. Rheumatology (Oxford) 2006; 45: 4 Meier JJ, Bhushan A, Butler AE, Rizza RA, Butler PC. 694–702. Sustained beta cell apoptosis in patients with long-standing 22 van Baarsen LG, van der Pouw Kraan TC, Kragt JJ, Baggen JM, type 1 diabetes: indirect evidence for islet regeneration? Rustenburg F, Hooper T et al. A subtype of multiple sclerosis Diabetologia 2005; 48: 2221–2228. defined by an activated immune defense program. Genes 5 Achenbach P, Warncke K, Reiter J, Naserke HE, Williams AJ, Immun 2006; 7: 522–531. Bingley PJ et al. Stratification of type 1 diabetes risk on the 23 van der Pouw Kraan TC, Wijbrandts CA, van Baarsen LG, basis of islet autoantibody characteristics. Diabetes 2004; 53: Voskuyl AE, Rustenburg F, Baggen JM et al. Rheumatoid 384–392. arthritis subtypes identified by genomic profiling of periph- 6 Bingley PJ, Gale EA. Progression to type 1 diabetes in islet cell eral blood cells: assignment of a type I interferon signature antibody-positive relatives in the European nicotinamide in a subpopulation of patients. Ann Rheum Dis 2007; 66: diabetes intervention trial: the role of additional immune, 1008–1014. genetic and metabolic markers of risk. Diabetologia 2006; 49: 24 Li Q, Xu B, Michie SA, Rubins KH, Schreriber RD, McDevitt 881–890. HO. Interferon-alpha initiates type 1 diabetes in nonobese 7 Wang X, Jia S, Geoffrey R, Alemzadeh R, Ghosh S, Hessner MJ. diabetic mice. Proc Natl Acad Sci USA 2008; 105: 12439–12444. Identification of a molecular signature in human type 1 25 Huang X, Yuang J, Goddard A, Foulis A, James RF, Lernmark diabetes mellitus using serum and functional genomics. A et al. Interferon expression in the pancreases of patients with J Immunol 2008; 180: 1929–1937. type I diabetes. Diabetes 1995; 44: 658–664. 8 Kaizer EC, Glaser CL, Chaussabel D, Banchereau J, Pascual V, 26 Foulis AK, Farquharson MA, Meager A. Immunoreactive White PC. Gene expression in peripheral blood mononuclear alpha-interferon in insulin-secreting beta cells in type 1 cells from children with diabetes. J Clin Endocrinol Metab 2007; diabetes mellitus. Lancet 1987; 2: 1423–1427. 92: 3705–3711. 27 Blanco P, Palucka AK, Gill M, Pascual V, Banchereau J. 9 Liew CC, Ma J, Tang HC, Zheng R, Dempsey AA. The Induction of dendritic cell differentiation by IFN-alpha in peripheral blood transcriptome dynamically reflects system systemic lupus erythematosus. Science 2001; 294: 1540–1543. wide biology: a potential diagnostic tool. J Lab Clin Med 2006; 28 Wallace J, Petri M, Olsen N, Kirou K, Dennis G, Yao Y et al. 147: 126–132. Abstract 1315: MEDI-545, an anti-interferon alpha monoclonal 10 Maas K, Chan S, Parker J, Slater A, Moore J, Olsen N et al. antibody, shows evidence of clinical activity in systemic lupus Cutting edge: molecular portrait of human autoimmune erythematosus (ACR 71st annual scientific meeting). Arthritis disease. J Immunol 2002; 169: 5–9. Rheum 2007; 56(9 suppl): S526. 11 Batliwalla FM, Li W, Ritchlin CT, Xiao X, Brenner M, 29 Lang KS, Weigert C, Braedel S, Fillon S, Palmada M, Schleicher Laragione T et al. Microarray analyses of peripheral blood E et al. Inhibition of interferon-gamma expression by osmotic cells identifies unique gene expression signature in psoriatic shrinkage of peripheral blood lymphocytes. Am J Physiol Cell arthritis. Mol Med 2005; 11: 21–29. Physiol 2003; 284: 200–208. 12 Li D, Butt A, Clarke S, Swaminathana R. Real-time quantita- 30 Yeshao W, Gu J, Peng X, Nairn AC, Nadler JL. Elevated tive PCR measurement of thyroglobulin mRNA in peripheral glucose activates protein synthesis in cultured cardiac blood of thyroid cancer patients and healthy subjects. Ann N Y myocytes. Metabolism 2005; 54: 1453–1460. Acad Sci 2004; 1022: 147–151. 31 Yin Z, Wilson S, Hauser NC, Tournu H, Hoheisel JD, Brown 13 Pachot A, Lepape A, Vey S, Bienvenu J, Mougin B, Monneret AJ. Glucose triggers different global responses in yeast, G. Systemic transcriptional analysis in survivor and non- depending on the strength of the signal, and transiently survivor septic shock patients: a preliminary study. Immunol stabilizes ribosomal protein mRNAs. Mol Microbiol 2003; 48: Lett 2006; 106: 63–71. 713–724. 14 Thach DC, Lin B, Walter E, Kruzelock R, Rowley RK, Tibbetts 32 Nair KS, Halliday D, Garrow JS. Increased energy expenditure C et al. Assessment of two methods for handling blood in in poorly controlled type 1 (insulin-dependent) diabetic collection tubes with RNA stabilizing agent for surveillance of patients. Diabetologia 1984; 27: 13–16.

Genes and Immunity Functional genomics in type-I diabetes F Reynier et al 278 33 Charlton MR, Nair KS. Role of hyperglucagonemia in 40 Bolstad BM, Irizarry RA, Astrand M, Speed TP. A comparison catabolism associated with type 1 diabetes: effects on leucine of normalization methods for high density oligonucleotide metabolism and the resting metabolic rate. Diabetes 1998; 47: array data based on variance and bias. Bioinformatics 2003; 19: 1748–1756. 185–193. 34 Peakman M, Tree TI, Endl J, van Endert P, Atkinson MA, Roep 41 Tusher VG, Tibshirani R, Chu G. Significance analysis of BO. Characterization of preparations of GAD65, proinsulin, microarrays applied to the ionizing radiation response. Proc and the islet tyrosine phosphatase IA-2 for use in detection of Natl Acad Sci USA 2001; 98: 5116–5121. autoreactive T-cells in type 1 diabetes: report of phase II of the 42 Lee ML, Whitmore GA. Power and sample size for DNA second international immunology of diabetes society work- microarray studies. Stat Med 2002; 21: 3543–3570. shop for standardization of T-cell assays in type 1 diabetes. 43 Valentini G. Clusterv: a tool for assessing the reliability of Diabetes 2001; 50: 1749–1754. clusters discovered in DNA microarray data. Bioinformatics 35 van Halteren AG, Kardol MJ, Mulder A, Roep BO. Homing of 2006; 22: 369–370. human autoreactive T cells into pancreatic tissue of NOD–scid 44 Dennis Jr G, Sherman BT, Hosack DA, Yang J, Gao W, Lane mice. Diabetologia 2005; 48: 75–82. HC et al. DAVID: database for annotation, visualization, and 36 Monks SA, Leonardson A, Zhu H, Cundiff P, Pietrusiak P, integrated discovery. Genome Biol 2003; 4: P3. Edwards S et al. Genetic inheritance of gene expression in 45 Hosack DA, Dennis Jr G, Sherman BT, Lane HC, Lempicki RA. human cell lines. Am J Hum Genet 2004; 75: 1094–1105. Identifying biological themes within lists of genes with EASE. 37 Morley M, Molony CM, Weber TM, Devlin JL, Ewens KG, Genome Biol 2003; 4: R70. Spielman RS et al. Genetic analysis of genome-wide 46 Huang da W, Sherman BT, Tan Q, Kir J, Liu D, Bryant D et al. variation in human gene expression. Nature 2004; 430: DAVID bioinformatics resources: expanded annotation database 743–747. and novel algorithms to better extract biology from large gene 38 Schroeder A, Mueller O, Stocker S, Salowsky R, Leiber M, lists. Nucleic Acids Res 2007; 35 (Web Server issue): W169–W175. Gassmann M et al. The RIN: an RNA integrity number for 47 Calvano SE, Xiao W, Richards DR, Felciano RM, Baker HV, assigning integrity values to RNA measurements. BMC Mol Cho RJ et al. A network-based analysis of systemic inflamma- Biol 2006; 7:3. tion in humans. Nature 2005; 437: 1032–1037. 39 Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis 48 Pachot A, Blond JL, Mougin B, Miossec P. Peptidylpropyl KJ, Scherf U et al. Exploration, normalization, and summaries isomerase B (PPIB): a suitable reference gene for mRNA of high density oligonucleotide array probe level data. quantification in peripheral whole blood. J Biotechnol 2004; Biostatistics 2003; 4: 249–264. 114: 121–124.

Supplementary Information accompanies the paper on Genes and Immunity website (http://www.nature.com/gene)

Genes and Immunity