Emergence of Autoimmunity in C57BL/6.NOD-Aec1Aec2 mice

p

r

e a

- u

t t o o

i

s

m

t u

i m m b

u c e l

n i

l n i i

t n i

y c

e a l

f r

o

m s

u

b 4

c

l s

t i o n t a

i c b

1 a l 6 l e

p

w h e a

s

e e k

s

S s

u

o j

ö b

f

g c a r l e in

g n i e c ’ s a l s t y o n d o r v o e m r t e

transcriptional landscape of salivary glands targeted by autoimmunity innate immunity (darkened = dependence on susceptibility regions) adaptive immunity (darkened = dependence on susceptibility regions) Transcriptional landscapes of emerging autoimmunity: transient aberrations in the targeted tissue’s extracellular milieu precede immune responses in Sjögren’s syndrome Delaleu et al.

Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 http://arthritis-research.com/content/15/5/R174 Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 http://arthritis-research.com/content/15/5/R174

RESEARCH ARTICLE Open Access Transcriptional landscapes of emerging autoimmunity: transient aberrations in the targeted tissue’s extracellular milieu precede immune responses in Sjögren’s syndrome Nicolas Delaleu1*, Cuong Q Nguyen2, Kidane M Tekle3, Roland Jonsson1,4 and Ammon B Peck2

Abstract Introduction: Our understanding of autoimmunity is skewed considerably towards the late stages of overt disease and chronic inflammation. Defining the targeted organ’s role during emergence of autoimmune diseases is, however, critical in order to define their etiology, early and covert disease phases and delineate their molecular basis. Methods: Using Sjögren’s syndrome (SS) as an exemplary rheumatic autoimmune disease and temporal global -expression profiling, we systematically mapped the transcriptional landscapes and chronological interrelationships between biological themes involving the salivary glands’ extracellular milieu. The time period studied spans from pre- to subclinical and ultimately to onset of overt disease in a well-defined model of spontaneous SS, the C57BL/6. NOD-Aec1Aec2 strain. In order to answer this aim of great generality, we developed a novel bioinformatics-based approach, which integrates comprehensive data analysis and visualization within interactive networks. The latter are computed by projecting the datasets as a whole on apriori-defined consensus-based knowledge. Results: Applying these methodologies revealed extensive susceptibility loci-dependent aberrations in salivary gland homeostasis and integrity preceding onset of overt disease by a considerable amount of time. These alterations coincided with innate immune responses depending predominantly on located outside of the SS-predisposing loci Aec1 and Aec2. Following a period of transcriptional stability, networks mapping the onset of overt SS displayed, in addition to natural killer, T- and B-cell-specific gene patterns, significant reversals of , cell-cell junctions and neurotransmitter receptor-associated alterations that had prior characterized progression from pre- to subclinical disease. Conclusions: This data-driven methodology advances unbiased assessment of global datasets an allowed comprehensive interpretation of complex alterations in biological states. Its application delineated a major involvement of the targeted organ during the emergence of experimental SS.

Introduction of the molecular basis of autoimmunity is skewed toward Common to autoimmune diseases is a long and clinically late and overt disease phases. To conclusively assign etio- silent phase. As a consequence, affected individuals are di- logical relevance to any biological process altered in such agnosed only after immune system–mediated functional specimens is difficult, considering the causality di- deficiencies of the affected tissues result in overt disease lemma. Nevertheless, stratifying the chronology of these [1,2]. Hence, owing to the unavailability of human speci- events is crucial in estimating whether genetic predis- mens reflecting subclinical disease stages, understanding position to develop a specific autoimmune disease might also involve genes associated with tissue develop- * Correspondence: [email protected] ment and homeostasis or if the genes exclusively cluster 1Broegelmann Research Laboratory, Department of Clinical Science, University of Bergen, Laboratory Building, Jonas Liesvei 65, 5021 Bergen, Norway in processes associated with specific phases of innate Full list of author information is available at the end of the article adaptive immune maturation [3-5].

© 2013 Delaleu et al.; licensee BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 2 of 19 http://arthritis-research.com/content/15/5/R174

One approach to delineate, and to a certain extent strat- General Medical Sciences, Bethesda, MD, USA) [17] and ify, the molecular events associated with subclinical phases design of an advanced visualization methodology. By of autoimmune diseases is the use of adequate experimen- exploiting this approach, we sought to significantly im- tal models [6,7]. For this purpose, a suitable experimental prove our ability to analyze such “-omics” data sets com- strain must, in correspondence with humans, develop its prehensively and systematically and, in turn, to minimize relevant autoimmune phenotype over an extended period the introduction of personal bias. of time and in the context of its genetic background. C57BL/6.NOD-Aec1Aec2 mice fulfill these criteria as a Methods model of primary Sjögren’s syndrome (SS) because they Animals develop, in the absence of other inflammatory conditions, C57BL/6.NOD-Aec1Aec2 and C57BL/6 male mice were all major features relevant to the diagnosis of SS in bred and maintained under specific pathogen-free condi- humans spontaneously and over a period of several tions at the Department of Pathology mouse facility at the months [7,8]. University of Florida, Gainesville, FL, USA. To dissect the With a prevalence of 0.1% to 0.3% in the total popula- SGs, mice were killed by cervical dislocation after deep tion, SS is considered a relatively common autoimmune anesthetization. All procedures were approved by the disease. It mainly involves the exocrine glands. Nearly all University of Florida’s Institutional Animal Care and Use patients complain about persistent symptoms of dry Committee (protocols B317-2007 and 2008011756). mouth, and many present with hyposalivation. Severe disease outcomes also include disabling fatigue and de- Isolation of RNA from salivary glands velopment of non-Hodgkin’s lymphoma. To date, all Total RNA was isolated according to the protocol de- therapies tested have been ineffective in reversing the scribed in detail elsewhere [18]. When the mice were 4, 8, course of SS [9,10]. Similar to patients with systemic 12 and 16 weeks of age, the SGs free of lymph nodes were lupus erythematosus, a subpopulation of individuals with excised in parallel from five C57BL/6.NOD-Aec1Aec2 and SS exhibit a type 1 IFN signature, suggesting that a viral five C57BL/6 mice, then snap-frozen in liquid nitrogen. agent may be involved in triggering the disease [11]. As Total RNA from each mouse was isolated concurrently a consequence, studies designed to discover genetic as- using the RNeasy Mini Kit (QIAGEN, Valencia, CA, sociations have focused either on innate immunity [12] USA), then RNA concentrations and purities were evalu- or on genes that might explain the dominant role of B ated using UV spectroscopy. The ratio of absorbance cells in the pathogenesis of SS [10]. Unfortunately, these (260 nm and 280 nm) of the RNA samples averaged 1.976. studies have yet to yield results that allow estimation of Subsequently, each sample was hybridized separately on a an individual’s risk of developing SS. GeneChip Mouse Genome 430 2.0 Array and 3′ IVT Ex- Histological evaluations of minor salivary glands (SGs) press Kit (Affymetrix, Santa Clara, CA, USA) according to obtained from patients with SS commonly show focal in- the manufacturer’s instructions (annotation: build 32; 6 flammation that may coincide with epithelial cell atro- September 2011). Microarrays were assessed using Affy- phy and the presence of adipose tissue and fibrosis. metrix Expression Console Software 1.1 without changing Morphologically, these glands may also display structural the default settings (Affymetrix), and the data quality was disorganization, including loss of cell–cell and cell– deemed adequate for further analyses. (ECM) adhesion [13,14]. However, organizing these findings chronologically and conclu- Submission of data to Omnibus sively as etiological, pathogenic or bystander processes All the data sets reported herein have been deposited has not yet been possible [9]. and are publicly available in the Gene Expression Omni- Thus, the aim of this study was to delineate the tran- bus [GSE15640, GSE36378]. scriptional landscape associated with the extracellular milieu (EM) of the SGs during spontaneous emergence of Verification of microarray data experimental SS. The global scope of our aim favors inte- In addition to the experiments performed to validate the gration over reduction and is ideally based on a data- quality of the microarray data presented previously driven approach that ensures impartial interpretation of [19,20], verification experiments were expanded to include data sets as a whole. For this purpose, we developed a groups of genes in accordance with the specific aims of novel data analysis pipeline that combines gene set enrich- this study. Real-time polymerase chain reactions (PCRs) ment analyses (GSEAs) [15], leading edge (LE) analyses were carried out using the Extracellular Matrix & Adhe- [15] and Markov cluster algorithm (MCL) clustering [16] sion Molecules PCR Array (PAMM-013Z; SABiosciences, for analysis of biological states. This set of data analyses Valencia, CA, USA) and the PI3K-AKT Signaling PCR formed the basis for computation of interactive networks Array (PARN-058Z; SABiosciences) according to the within the Cytoscape software suite (National Institute of instructions provided by the manufacturer. These arrays Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 3 of 19 http://arthritis-research.com/content/15/5/R174

were analyzed using RT2 Profiler PCR Array Data Ranking differential gene expression between two Analysis software (SABiosciences) to calculate the fold biological states changes in gene expression occurring within the respective To identify coordinated and significant changes between time periods. These data were subsequently plotted the two chronologically closest time points from each against the values yielded by the GeneChip Mouse strain, the collapsed gene lists (21,673 genes) were Genome 430 2.0 Array and 3′ IVT Express Kit array ranked based on the observed relative difference upon (Additional file 1: Figures S1A to S1C) and subjected to performing significance analysis of microarrays (SAM) correlation analyses (Additional file 1: Figure S1D). in J-Express 2009. SAM makes no assumption about the distribution of the data and effectively introduces a non- arbitrary fold increase criterion, thus superseding the Data analysis pipeline introduction of a subjective fold-change threshold. These A flow diagram of the data analysis pipeline is depicted ranked lists were loaded into the GSEA v2.07 database in Figure 1. (Broad Institute, Cambridge, MA, USA).

Normalization of microarray probe cell intensity files Compilation of gene sets for gene set enrichment analysis Probe cell intensity files (.CEL) were quantile-normalized Ageneset(GS)isana priori-defined groups of genes and underwent general background correction. Control compiled, curated and annotated to reflect one specific metrics were generated and passed for each array (Robust trait that its members share, such as they are all Multichip Analysis performed with Affymetrix Expression [15]. The first GSs compiled for this study were extracted Console Software 1.1; Affymetrix). Genes covered by mul- from the following bioinformatics resources as described tiple probes on the microarray chip were collapsed to previously [22]: (1) (GO) (n = 12,467), (2) genes by selecting the probe yielding the highest signal National Cancer Institute (NC) (n = 219), (3) PFAM (PF) (J-Express 2009 software; MolMine AS, Hafrsfjord, (n = 4,147) and (4) Kyoto Encyclopedia of Genes and Norway). Genomes (KE) (n = 225). Using the WhichGenes 1.0 GS

SG Global Gene Expression GS Enrichment Analysis Requirements for Nodes C57BL/6.NOD-Aec1Aec2 (disease 7871 GS retained (>10 & <1000 For EM-related GS (ancestor prone) & C57BL/6 (asymptomatic genes); 1000 GS permutations either GO_0071944, controls) GO_0031012, GO_0005911): n=5 per strain per time point (04-, 08-, (FDRq <0.05) 12-, 16-weeks) Normalization for Age- For EM-associated GS: related SG Development 1st degree neighbor of a EM- Exclusion of GS showing the related GS ( 8% overlap in the Ranking Differential Expression same trend in both strains during LEs of the EM-related GS and Collapsing probes to genes the same time period (FDRq <0.05; the EM-associated GS) SAM based ranking of the nom. p-value<0.005; TAGS 50%) FDRq <0.05; nom. p-value chronologically closest time-points <0.005; TAGS 50%

Leading Edge (LE) Analysis GS Resources n=21855 Gene Ontology (GO) 12467 to each GS Requirements for Edges CONNECTIVITY score 0.08 NCI (NC) 219 computation of LE metrics (TAGS, (corresponding to 8% overlap PFAM (PF) 4147 LIST, SIGNAL) in the LEs the two GSs) KEGG pathways (KE) 225 Flagging LE genes located in Biocarta pathways (BI) 249 Aec1 and Aec2 Reactome pathways (RE) 943 Computation of a CONNECTIVITY Transcription factor targets (TF) 615 Network Clustering MicroRNA targets (MI) 793 overlap between the LE genes MCL clustering based on -log of Genebands (GB) 393 of all GSs paired one on one CONNECTIVITY

Data acquisition & resources Data analysis Network building Figure 1 Schematic representation of the data analysis and data visualization pipeline. Designed for unbiased mapping of alterations in the transcriptional landscape of the extracellular milieu (EM). CONNECTIVITY, degree of overlap in LE-members between the GSs the edge is connecting; FDR, false discovery rate; GS, gene set; KEGG, Kyoto Encyclopedia of Genes and Genomes; LE, leading edge; LIST, percentage of genes in the ranked gene list before (for positive enrichment score (ES)) or after (for negative ES) the peak in the running ES (indication of where in the list the enrichment score is attained); MCL, Markov cluster algorithm; NCI, National Cancer Institute database; PFAM, Pfam families database [21]; SAM, significance analysis of microarrays; SG, salivary gland; SIGNAL, ES strength that combines LIST and TAGS; TAGS, percentage of gene hits before (for positive ES) or after (for negative ES) the peak in the running ES, (indicates percentage of genes contributing to ES). Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 4 of 19 http://arthritis-research.com/content/15/5/R174

building tool [23], we compiled the remaining GSs from with GSs uniquely altered in C57BL/6.NOD-Aec1Aec2 (5) BioCarta pathways (BC) (n = 249), (6) Reactome path- mice (Figure 2A; exclusive), selected for network building. ways (RE) (n = 943), (7) transcription factor (TF) binding motifs (mouse orthologs were inferred from human genes) Network building (n = 615), (8) microRNA binding motifs as defined in the Network analysis is the study of a system that is depicted miRDB database (http://mirdb.org/miRDB/) (MI) (n = as connections (that is, edges) between discrete objects 793) and (9) close genomic localization (Ensembl genes in (that is, nodes). To define the EM, GSs that yielded a FDR bands resource) (GB) (n = 393). All GSs (N = 21,855) were less than 0.05 and had either GO_0071944 (cell periphery), downloaded between 14 and 20 August 2011. GO_0031012 (ECM) or GO_0005911 (cell–) as an ancestor in the GO tree were selected and qualified Running gene set enrichment analysis and identification as EM-related GSs. of leading edge genes GSs connected by an edge (≥8% of shared LE members) GSEA aids in overcoming the analytical challenges posed to an EM-related GS were qualified as an EM-associated by pleiotropy, as genes are assigned to GSs that repre- process when they passed the significance criteria (FDR sent each of their traits, and by the fact that biological <0.05, nominal P-value <0.005, TAGS ≥50%). Defining the processes commonly depend on a coordinated change in forces of attraction, the degree of overlap in LE members the expression of several genes [15]. Statistical analyses between the GSs also determined their position in the net- are performed for each GS by assessing the expression work computed using the edge-weighted, spring force– pattern formed by its members within the entire data set directed layout in Cytoscape 2.8.2 [17]. Cytoscape is an (21,673 genes). Thus, an asymmetrical distribution open source software platform utilized for visualizing skewed significantly to the overexpressed end of the complex networks and integrating these networks with ranked list signifies significant enrichment. In contrast, any type of attribute data [17]. The connectivity parameter, such an asymmetrical distribution indicates significant defined by the degree of overlap in LE members between depletion in cases where the expression pattern of the the GSs, could thereby also be applied as an edge weight GS is skewed significantly to the underexpressed end of for the subsequent MCL clustering [16] computed within the ranked list. This step of computational interpretation Cytoscape. The clusters identified by the MCL are defined based on a priori-defined and consensus-based bio- by simulating the stochastic flow within the networks [16]. logical knowledge without setting arbitrary cutoffs, such as fold change or significance level, prevents the intro- Results duction of bias and increases the robustness and com- Extent of alterations across the three time periods parability of results. GSEA was performed for GSs larger Application of the data analysis pipeline outlined in than 10 and smaller than 1,000 (7,871 of 21,855 GSs Figure 1 revealed that the most thematically diverse al- retained). Permutation number was deemed adequate at terations specific for C57BL/6.NOD-Aec1Aec2 mice, in- 1,000 iterations, and default values were used for all volving the most EM-related GSs and EM-associated other parameters. GSs (Figure 2B), occurred between 4 and 8 weeks of age. LE analysis identifies the genes of each GS that appear The same was true for the number of genes accounting in the ranked list at or before the point at which the for the GSs’ significant enrichment or depletion, that is, running sum reaches its maximum deviation from zero. LE genes (Figures 2C and 2D). Hence, genes assigned to a GS’s LE (LE genes) are the Significant enrichment at 8 weeks of age involved 79 genes accounting for the individual GS significant en- GSs that depended on coordinated upregulation of 481 richment or depletion signal [15]. LE analyses were com- LE genes. Interestingly, 43% of these GSs were simultan- puted after GSEA using GSEA v2.07. eously becoming depleted in age-matched C57BL/6 mice (Figure 2A; reciprocal). Over the same period of time, Subtraction of alterations associated with age-related downregulation of 359 LE genes led to significant deple- salivary gland development tion of 29 GSs (Figures 2B and 2C). Between 8 and To normalize for changes in gene expression associated 12 weeks of age, a single GS was becoming depleted with normal SG activities, we discarded GSs that yielded (Figures 2A and 2B) in conjunction with downregulation significant enrichment or depletion in both strains in of 12 LE genes (Figures 2C and 2D). The transition from parallel and over the same period of time (false discovery 12 to 16 weeks of age, which chronologically coincided rate (FDR) <0.05, nominal P-value <0.005, TAGS ≥ 50% in with the onset of overt SS-like disease in C57BL/6. C57BL/6 mice) from all subsequent analyses (Figure 2A; NOD-Aec1Aec2 mice, was marked by enrichment of 12 parallel). Reciprocal changes over the same period of time GSs comprising a total of 182 LE genes, as well as deple- (for example, enriched in C57BL/6.NOD-Aec1Aec2 while tion of 15 GSs as a consequence of downregulation of depleted in C57BL/6 mice) were retained and, together 227 LE genes (Figures 2B and 2C). Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 5 of 19 http://arthritis-research.com/content/15/5/R174

A B 80 80 exclusive 70 70 EM-related 60 reciprocal EM-associated 60 50 parallel 50 40 30 40 20 30 10 20 5 10 no. significant of gene sets 0 sets gene significant of no. 5

-5 0 d

08wk enriched08wk depleted12wk enriched12wk depleted16wk enriched16wk depleted 08wk enriched08wk depleted12wk enriche12wk depleted16wk enriched16wk depleted

C D 500 700 LE genes not in Theme 1 Aec1 or Aec2 600 Theme 2 400 LE genes in Theme 3 Aec1 or Aec2 500 Theme 4

300 enes

g 400

200 300

no. of LE genes LE of no. LE of no. 200 100 100

0 0

08wk enriched08wk depleted12wk enriched12wk depleted16wk enriched16wk depleted 08wk enriched08wk depleted12wk enriched12wk depleted16wk enriched16wk depleted Figure 2 Extent of significant alterations during emergence and onset of overt Sjögren’s syndrome. (A) Number of exclusive GSs (GSs yielding significance in C57BL/6.NOD-Aec1Aec2 mice only), reciprocal GSs (GSs yielding significance in C57BL/6.NOD-Aec1Aec2 and C57BL/6 mice, but with opposite trends) and parallel GSs (GSs yielding significance in C57BL/6.NOD-Aec1Aec2 mice and C57BL/6 mice with the same trend (excluded)). (B) Number of extracellular milieu(EM)-related and EM-associated gene sets (GSs) significantly altered for each time period. (C) Number of leading edge (LE) genes underlying the changes displayed in (A), categorized with respect to localization inside or outside the susceptibility regions Aec1 and Aec2. (D) Number of LE genes underlying each of the major biological themes.

Major biological themes involving the extracellular milieu allowed bidirectional integrin growth factor signaling during emergence of Sjögren’s syndrome pathway cross-talk as well as all effector processes re- Progression from pre- to subclinical disease occurs between lated to cell motility; and (4) the three major classes of 4 and 8 weeks of age intercellular junction complexes engaging in cell-cell sig- The network displaying all GSs enriched by 8 weeks of naling via E- (CDH1) and involving transform- age (Figure 3), together with interpretation of the re- ing growth factor β (TGFβ). spective LE genes (Figure 4), allowed us to identify four GSs that were depleted in 8-week-old C57BL/6.NOD- major biological themes: (1) activation of pathways char- Aec1Aec2 mice clustered in three independent networks acteristic of innate immune responses to long double- (Figure 5A), each depending on distinct sets of LE genes stranded RNA viruses; (2) insulin receptor (Insr) and (Figure 5B): (1) deceleration of ECM turnover, (2) down- insulin-like growth factor 1 (Igfr1)-mediated signaling regulation of genes encoding and via phosphoinositide 3-kinase (PI3K) and protein kinase (3) loss of positive regulation of nerve impulses in con- B (AKT) further guiding cell fate, proliferation and dif- junction with downregulation of genes encoding members ferentiation; (3) remodeling of epithelial cell–ECM an- of all classes of cysteine (Cys) loop neurotransmitter chorage via focal adhesions (FAs) whose specificities receptors and fewer metabotropic receptors. Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 6 of 19 http://arthritis-research.com/content/15/5/R174

Figure 3 (See legend on next page.) Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 7 of 19 http://arthritis-research.com/content/15/5/R174

(See figure on previous page.) Figure 3 Enrichments in the transcriptional landscape of the extracellular milieu during transition from pre- to subclinical Sjögren’s syndrome. Gene sets (GSs) enriched at 8 weeks of age delineate activation of the innate immune system coinciding with significant alterations in the targeted tissue’s homeostasis and integrity. Proportions of leading edge (LE) genes shared between GSs defined distance, organization and clustering of the GSs. Dashed lines, separators between major biological themes; annotations in italics, interpretation of transcriptional activity inferred from the LE gene clouds displayed in Figure 4; node color, Markov cluster algorithm (MCL) cluster number. Node shapes: triangles, extracellular milieu (EM)–related; circles, EM-associated; node size, relative to number of detected genes that are members of this GS (reference node = 50 genes). Node label type size, relative to percentage of genes belonging to this GS’s LE (TAGS) (reference node = 75%). Node border: none, alteration of this GS exclusive to C57BL/6.NOD-Aec1Aec2 mice; present, reciprocal trend in C57BL/6 mice. Edge color: degree of overlap in LE genes between the two GSs connected by this edge. ECM, extracellular matrix; Igfr1, insulin-like growth factor receptor 1; Insr, insulin receptor.

Stabilization of subclinical disease state between 8 and genes located in Aec1 or Aec2 (Figure 8 and Additional file 12 weeks of age 1: Figures S3 and S6). Regarding the subclinical phase of GO_0005581 () was the only GS yielding SS, the greater reliance on genes located in the congenic significance during this time period (Additional file 1: regions of the themes associated with the SGs’ homeosta- Figure S2A) and representing 12 LE genes (Additional sis and integrity compared to innate immunity may indi- file 1: Figure S2B). cate that the latter occur subsequently and in response to these tissue-specific alterations (Figure 8 and Additional Transition from subclinical to overt disease between 12 and file 1: Figures S3 and S4). 16 weeks of age The GSs enriched at 16 weeks of age (Figure 6A), in Detailed annotation of networks based on interpretation conjunction with their respective LE genes (Figure 6B), of gene set parameters and leading edge gene patterns delineated two distinct themes: (1) emergence of an ef- As described in the Methods section, the number of LE fector immune response characterized by reinforcement genes shared between GSs determined their position, in- of the IFNα signature and a natural killer (NK) cell terconnectivity and cluster membership in correspond- population, together with formation of the primary im- ence with all other GSs of a network. Thus, GSs in close munological synapse and late costimulatory signals de- proximity to each other share distinct similarities in livering survival, proliferation and maturation signals to their LE gene patterns. To resolve redundancies, com- T cells and B cells; and (2) resumption of gene transcrip- monly caused by GSs representing complex pathways, tion for Cys loop receptors with (ACh), γ- highly interconnected network areas require additional aminobutyric acid (GABA) and glycine (Gly) binding interpretation. The same accounts for large GSs anno- specificities and initial upregulation of specific subsets of tated with terms too general to reflect the true theme metabotropic receptors. shared by their LE members. At 16 weeks of age, 80% of the depleted GSs (Figure 7A) The basis for this curated annotation, written in italic and a large number of LE genes (Figure 7B) showed partial type in Figures 3, 5A, 6A and 7A, is formulated upon ana- reversal of the alterations pertaining to FAs and cell–cell lysis of (1) the LE members of each MCL cluster displayed junctions observed earlier between 4 and 8 weeks of age as LE gene clouds generated by using a vector graphics– (Figures 3 and 4). The remaining 20% of the GSs, such as capable adaptation of the WordCloud Cytoscape plugin those GSs not subject to earlier alterations, reinforced the [24] (Figures 4, 5B, 6B and 7B), (2) each GS’sLEgenes cellular component (CC) terms GO_0005923 (TIGHT (Additional file 2) and (3) current literature–based interac- JUNCTION) and GO_0005925 (FOCAL ADHESION). tome maps (Additional file 1: Figure S8). Additional file 3 comprises the networks displayed in Figures 3-7 as Major biological themes dependent to significantly infinitely scalable and electronically searchable vector different degrees on genes located in Aec1 and Aec2 graphics, thereby allowing the visualization of network The quantitative contributions of the SS-predisposing detail. genomic regions Aec1 ( 3; 0 to 46 cM) and Aec2 (; 29.7 to 106.1 cM) to each GS are, Transcriptional changes underlying themes being enriched together with the LE genes located in these susceptibility during progression from pre- to subclinical Sjögren’s regions per biological theme, presented in Additional file syndrome–like disease 1: Figures S3 to S7. Comparing average proportions of LE In Figure 4, the LE gene cloud for Cluster_01-01, in com- genes located in Aec1 and Aec2 per GS and per biological bination with the percentage of each pathway covered by theme showed that the innate immunity theme was least its LE members (Additional file 2; TAGS), points toward dependent (mean = 1.75%), and that the adaptive immun- two pattern recognition receptors, namely, Toll-like ity theme was most dependent (mean = 18.71%), on LE receptor 3 (TLR3) and IFN-induced helicase C domain– Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 8 of 19 http://arthritis-research.com/content/15/5/R174

REFERENCE ikbke ikbkg madd NODE atf2 cd14 map2k3 1) Innate immunity map2k4 chuk arhgdia dusp3 map2k7 CLUSTER bcl10 fos gab1 pdpk1 pik3r2 pik3r3 adrb2 01-01 grb2 itgb1 abca1 cbl dlg1 map3k7 atg12 atg13 prkca exoc1 exoc3 exoc5 pik3ca atg16l1 atg3 ly96 atg4c atg5 exoc6 foxo3 cblb jun pik3cb atg9a becn1 prkcd dusp3 mapk14 chmp1a elf1 malt1 pawr cln3 grb14 CLUSTER dusp1 ikbkap pik3r1 ctsd eif2ak4 prkcz 01-02 mapk3 gas6 hdac6 ptpre grb2 ptprj pip5k1a plcg1 hras1 mapk8 ikbkb hspa5 jun src map1lc3a rela rnf31 traf3 vasp zyx mll5 mtor nbr1 npc1 mapkapk3 ogt pik3c3 psen1 rptor traf6 slc2a1 mef2a ube2n ube2v1 srebf1 mef2c tnrc6a vdr ptpn11 wrn nfkb2 pik3c3 rps6ka3 ptpra rapgef1 adipor1 rasa1 sgk1 shc1 pik3r4 rps6ka5 akt1 hk1 igf1r ppp2cb sik2 tab3 tbk1 CLUSTER tirap tlr3 akt2 tlr4 tollip sorbs1 insr irs1 02 rela ctnna1 ctnnd1 sos1 trip10 jup jak2 phip met pak1 bcar1 ctsb pik3ca eif4ebp2 dlc1 pik3cb tsc2 flot1 pik3r1 map4k2 rffl zfp106 enpp1 inppl1 flot2 pten map4k3 ptk2 nfkb1 ripk1 irak1 ptk2b map2k1 nfkbia tank ptpn11 ngf mavs pxn tradd irf3 prkci raf1 rasa1 prkcz sos1 src 2) Insr / Igfr1 signaling stat3 smpd2 bag4 acta1 sqstm1 birc2 crk pik3r2 nfkbib cyld casp2 casp1 stat1 birc3 dock1 pik3r3 cflar dap3 ifih1 ubc

cldn4 itgb4 krt14 krt5 met pik3ca pik3r1 col17a1 prkca sfn lama3 CLUSTER shc1 ywhab 03-01 egf erbb3 grb2 erbb2 ywhag lamb2 ywhah lamb3 ywhaz hras1 itgb1 lamc2 capzb cdk5 bsnd pip5k1c tesc myo1c akt1 dlg1 cadm1 tmod3 erbb2ip ctnna1 cd9 dst cftr rac1 tsc1 plec ctnnd1 cttn col17a1 dnm1l dnm2 ddr1 spp1 vasp slco2b1 gab1 src slco3a1 dock1 CLUSTER dlc1 thbs1 trf vcl ilk ilkap 03-02 dlg1 dsp iqgap1 jup dst zfp384 kcnq1 tln1 ebag9 zyx keap1 adam9 afap1 ywhaz lama3 ank3 ap2a1 arrb1 enah 3) Focal adhesion & motility arrb2 atp7a epb4.1l5 abcc1 abi1 lims2 abi2 epcam lpp actr3 bcar1 ephx2 actn1 akt1 luc7l3 erbb2 arhgap26 cdkn1b grb2 grb7 arhgef6 erbb2ip arhgef11 nexn numb grlf1 itgav arhgef7 add3 ahnak erbb3 amotl2 keap1 krit1 apc strn sympk ldb1 pak1 flnb rictor lama3 pdk2 pdpk1 pdlim7 itgb1 amot arhgap17 magi1 git1 tns1 arhgef2 lin7c pik3ca anxa2 synpo pkd1 plec jam2 prkar1a igsf8 tns4 prkar1b ptk2b pip5k1a pld1 ash1l cadm1 marveld2 trf dlc1 enah tjap1 ptpra fermt2 cgn cldn23 micall2 ptk2 cbl cldn3 cldn4 ap1m1 itgb5 fat1 tjp1 ptprm cblb flnb hdac4 mpdz ccnd1 jun fgd4 cldn7 crb3 cblc gsn raf1 ilk kat2b tjp2 tjp3 cd2ap cdc37 map2k1 rock1 mpp5 cdc42 diap1 inppl1 map2k4 myh11 pxn cltb git2 myo10 shc1 ctnna1 mapk8 myo9b mpp7 CLUSTER cltc hsp90aa1 mapk8ip3 nckap1 myh14 rfwd2 scrib hsp90ab1 ctnnd1 mtdh 04 crk nme1 palld slc22a5 slc2a1 erbb2 farp2 fgfr1 slc9a1 slco2b1 arpc1a igf1r insr iqgap1 ocln pak1 slco3a1 bcr rapgef1 pdlim7 sept7 acta1 pik3cb arpc1b lmo7 map3k7 capn1 rasa1 mapk3 ptk2 sorbs1 rock2 arpc3 ptpn11 mapk3 met pik3r1 st14 stx4a capns1 rras mylk arpc4 pten mllt4 nlk stxbp3a egf sos1 arpc5l sorbs1 actn1 syne2 src ssx2ip tcf7l1 rdx tcf7l2 tgfbr1 baiap2 akt2 parp1 actn4 tenc1 scyl3 wasl cd14 acta1 als2 ppp1r12a tgfbr1 tjp1 slc39a6 yes1 cd81 coro1c amot ppp1r14a erbb3 vav2 stxbp2 tmsb15l dlg1 cyfip1 pik3r2 tpm1 dlg5 ptpn6 ptprj ptprk epcam dbnl pik3r3 met capn2 2900073g15rik arhgap5 prkar2a xpo1 jup scrib tpm3 dsc2 ptprm fzd6 kifc3 mpz f11r nf2 ppp1r12b synm plcg1 nf2 numb dsg2 igsf5 vcl pard3 shroom2 inadl dsp slc2a1 plekha7 shroom3 jam2 zyx pmp22 gnpat prkci itgb1 gipc1 gm2a itch itgb4 itpr2 ndfip1 nedd4 ppp1r9a clasp1 clasp2 lamc2 lancl2 lasp1 llgl1 macf1 cnksr1 ctgf ccm2 dstn pkp1 pkd1 map2k1 mapre1 mark2 pnn cd9 lancl2 prkcz cttn cytip lasp1 prkcd spna2 cdc42 9030425e11rik abi2 llgl1 daxx dbnl spnb2 spnb3 cdh1 ppp1r9a rac3 prkcz myo9b dstn utrn crebbp scnn1a pvrl2 pvrl4 nf2 palld pdlim7 shroom2 epb4.1l5 rac1 spna2 spnb2 spnb3 tchp eps8l2 rac3 ehd2 smad7 pip5k1a smad2 epb4.1 tgfb1 ezr frmd4b smad4 epb4.1l1 tgfb2 wasl plekha1 rab34 rab5a vcl vps11 tpm3 inpp5j snai2 epb4.1l2 tgfb3 yes1 akap7 tln1 stx3 gja1 anxa5 rasa1 rdx s100a6 sh2b1 vps16 synj2 tesc tln1 CLUSTER 03-03 tmod3 fermt2 fkbp15 gng12 tmsb15l flna sept2sept7 actn1 actn4 myo1c add3 capza1 capza2 tpm1 cd2ap trpm7 myo6 dlc1

acta1 afap1 amot arhgap1 psen1 4) Cell-cell junctions & psen2 tsc1 epb4.1 arhgef2 ptk2b usp2 exoc6 organization bcar1 scnn1a utrn exoc8 acap2 tnfrsf12a

Figure 4 (See legend on next page.) Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 9 of 19 http://arthritis-research.com/content/15/5/R174

(See figure on previous page.) Figure 4 Annotation of the Markov cluster algorithm clusters with their respective leading edge gene clouds. The individual gene sets (GSs) of each Markov cluster algorithm cluster shown in Figure 3 were collapsed into a metanode. Network: metanode color represents the original color of the ancestor GSs and node size and node label font size are proportional to the number of GSs collapsed into this metanode (reference node = 15 GSs). Clustered leading edge (LE) gene clouds: Font color represents clustering of the LE genes based on the connections between the original GSs. Font size is proportional to the frequency of the gene in the LEs of the GSs collapsed into this metanode. containing protein 1 (IFIH1), also known as MDA5. KINASE; TAGS = 68% [31]. Thus, 31 of 77 genes anno- Both these receptors are key molecules upstream of IFN tated in the CC GO_0005925 (FOCAL ADHESION) GS regulatory factor 3 (Irf3) and signal transducer and activa- were located in its LE (TAGS = 40%). tor of transcription 1 (STAT1) (Figure 4, Cluster_01-01). Matching the integrin genes (that is, Itgav, Itgb1, Itgb4 Also delineated by this cluster are upregulation of Tlr4 and Itgb5) with the dominant growth factor receptor genes and its coreceptors Cd14 and lymphocyte antigen 96 (that is, Met, Insr, Igfr1, fibroblast growth factor receptor 1 (Ly96). These may, via their upregulated signaling cascade, (Fgfr1)andTgfbr1) in the LE profiles displayed in Figure 4 deliver the strongest trigger for the observed canonical ac- suggests that integrin αvβ5, upstream of the enriched tivation of nuclear factor κ-light-chain-enhancer of acti- integrin-linked kinase signaling-associated GSs, provides a vated B cells (NF-κB) and mitogen-activated protein basis for IGFR1-integrin cross-talk [32]. Similarly, αvβ5 kinase 8 (MAPK8) [4,25]. In addition, the gene nerve and αvβ1 may allow for TGFBR1 signaling by collaborating growth factor (Ngf), which encodes another important in- with integrin pathways (Figure 3) [33]. Tgfb1, Tgfb2 and ducer of NF-κB, was upregulated, even though NGF’scru- Tgfb3, together with the TF Smad family member 2 cial receptor p75 neurotrophin receptor (p75NTR) [26] (Smad2)andSmad4 downstream of Tgfbr1 and the nega- was absent from the list of LE genes for RE_P75NTR SIG- tive feedback–associated Smad7, are all present in the LE NALS VIA NF-KB; TAGS = 73% (Additional file 2). of Cluster_04 (Figure 4). The presence of osteopontin Determining the effect of the interconnecting (Spp1), another ligand of integrins αvβ1 and αvβ5 in the LE NC_TRAIL SIGNALING PATHWAY; TAGS = 63% and of Cluster_03-02 (Figure 4), indicates that FA maturation BI_MET PATHWAY; TAGS = 74% GSs is more difficult may also occur in relation to innate immune cells. because of incomplete coverage of the different arms of Supporting a critical role of FA remodeling during this the TNF-related apoptosis-inducing ligand (TRAIL) path- transition from pre- to subclinical SS-like disease, cal- way [27] and the hepatocyte growth factor receptor pain 1 (Capn1) and Capn2 (Figure 4), which regulate the (MET) pathway [28] by their respective LE genes (Add- dynamics of FA assembly and disassembly, are at the itional file 2). However, their involvement in determining center of the two calpain-specific GSs (Figure 3). In cell fate and proliferation is reflected by their central pos- addition, all other effector phases of non-muscle-cell itioninFigure3. movement are represented by GSs and LE genes of Clus- The GSs and LE genes associated with Cluster_02 ter_03-02, Cluster_03-03 and the intercalated section of (Figures 3 and 4) suggest that INSR and IGFR1, via their Cluster_04, respectively (Figures 3 and 4) [30]. shared downstream signaling cascade involving PI3K The fourth biological theme shares 14.3% of its LE and AKT, upregulate mammalian target of rapamycin genes with Cluster_03-02 described above. This is due to complex 1 (mTorC1) and mTorC2. The mTor system in molecular similarities between CC GO_0030055 (CELL- turn is pivotal in determining cell fate [29]. Increased SUBSTRATE JUNCTION); TAGS = 41% and CC autophagy is inferred by the presence of PI3K pathway GO_0005913 (CELL-CELL ); members and several autophagy-related protein (Atg) TAGS = 46%. Multiple LE genes belonging to the encoding genes (Figure 4, Cluster_02). and the gene families further indicate increased In close proximity, Cluster_03-01 and Cluster_03-02 formation of tight junctions at CC GO_0016327 (API- delineate cell-matrix adhesion complexes that transmit COLATERAL PLASMA MEMBRANE); TAGS = 51% regulatory signals and mechanical forces (Figures 3 and 4) [30]. These two types of cell–cell junction complexes de- [30,31]. Cluster_03-01, including GO_0031581 (HEMI- pend critically on CDH1 expressed by epithelial cells ASSEMBLY); TAGS = 73%, defines the [30]. Correspondingly, NC_E-CADHERIN SIGNALING -mediated, -5-dependent anchor- IN THE NASCENT ADHERENS JUNCTION; TAGS = age of epithelial cells’ intermediate filaments to the basal 60% and CDH1 anchorage-related GS GO_0017166 lamina of the ECM. Cluster_03-02 represents, in large (VINCULIN BINDING); TAGS = 60% were significantly part, signaling pathways that are activated by alterations in enriched and are mapped at the center of Figure 3. Fur- acell’s immediate surroundings and are transmitted via thermore, enrichment of GS GO_0030057 (DESMO- actin cytoskeleton–anchored FAs, such as NC_SIGNAL- SOME); TAGS 40% [30] delineates a third class of ING EVENTS MEDIATED BY FOCAL ADHESION intercellular junction complexes associated with genes that Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 10 of 19 http://arthritis-research.com/content/15/5/R174

1) ECM turnover 3) Neurotransmission

GO_0051971 REFERENCE (POSITIVE REGULATION NODE OF TRANSMISSION OF NERVE IMPULSE)

GO_0030199 RE_NCAM1 (COLLAGEN INTERACTIONS Positive regulation FIBRIL ORGANIZATION) GO_0005234 (EXTRACELLULAR-GLUTAMATE-GATED of nerve impulses ION GO_0008066 CHANNEL PF_01410 ACTIVITY) (FIBRILLAR (GLUTAMATE COLLAGEN RECEPTOR GO_0022834 GO_0004970 C-TERMINAL ACTIVITY) DOMAIN) GO_0005583 PF_00093 (LIGAND-GATED (IONOTROPIC (FIBRILLAR (VON CHANNEL GLUTAMATE COLLAGEN) WILLEBRAND ACTIVITY) GO_0008328 FACTOR RECEPTOR Collagens TYPE C (IONOTROPIC ACTIVITY) DOMAIN) PF_00060 GLUTAMATE (LIGAND-GATED RECEPTOR (all structural classes) ION COMPLEX) CHANNEL)

GO_0005581 GO_0015276 GO_0005230 (COLLAGEN) Cys-loop receptors (LIGAND-GATED (EXTRACELLULAR ION LIGAND-GATED PF_01391 CHANNEL (COLLAGEN ION (all categories) ACTIVITY) TRIPLE CHANNEL HELIX GO_0005231 REPEAT (20 (EXCITATORY ACTIVITY) COPIES)) Contact activation PF_00090 EXTRACELLULAR PF_01094 (THROMBOSPONDIN LIGAND-GATED (RECEPTOR TYPE 1 DOMAIN) ION FAMILY pathway (coagulation) CHANNEL LIGAND ACTIVITY) BINDING Angioinhibitory REGION) BI_INTRINSIC PATHWAY signals

GO_0005578 (PROTEINACEOUS EXTRACELLULAR MATRIX) GO_0031012 2) Gap junctions (EXTRACELLULAR MATRIX)

GO_0005921 PF_10582 (GAP (GAP JUNCTION) PF_00045 JUNCTION CHANNEL (HEMOPEXIN) PROTEIN CYSTEINE-RICH DOMAIN) Formation and structural

Core components PF_00029 Mmp-mediated () organization of the ECM of gap junctions collagenolysis

GO_0005922 ( GO_0043205 PF_00413 A COMPLEX) (FIBRIL) (MATRIXIN) 1) ECM turnover 3) Neurotransmission

accn1 accn2 accn3 REFERENCE accn5 glra1 NODE kcnh2 kcnh7 kcnj1 kcnj10 kcnj12 kcnj14 kcnj2 kcnj3 kcnj4 kcnj5 kcnj6 kcnj8 kcnj9 c1ql2 c1ql3 c1qtnf1 c1qtnf2 c1qtnf3 c1qtnf6 c1qtnf7 col4a1 pex5l ryr2 trpc1 trpc2 col10a1 col11a1 col5a1 trpc6 col4a2 col11a2 col12a1 col13a1 col4a3 col14a1 col15a1 col19a1 col5a2 col4a5 shisa9 col4a6 f10 grm1 casr col1a1 col6a1 f12f2f8f9fga col5a3 grm2 htr6 gabbr1 fgb fgg klkb1 ifng gprc2a-rs5 col6a2 gdnf proc serpinc1 grm3 il6 clca5 cnga4 cngb3 grm5 itga2 lama2 camk2b col1a2 grm7 col6a3 gabra1 lgi1 cartpt col23a1 col25a1 col6a4 grm8 ncs1 cckbr col8a2 gucy2c nmu ccl2 CLUSTER nr2e1 gucy2e oxt col9a1 CLUSTER gucy2g drd1a oxtr col2a1 col9a2 ecm2 npr1 egfr ptgs2 01-01 colec10 gfap npr3 slc1a3 colec11 nell1 03 gip tas1r1 snca colq eda nell2 tas1r2 tacr1 nov vmn2r10 tnf cnih2 emilin1 pxdn col3a1 vmn2r84 tnr cnih3 fcna sspo vmn2r88 uts2 gm7455 thbs2 foxc2 mbl2 vwc2l lmx1b bmper scara3 vwce plod3 chrdl1 sftpa1 wisp1 serpinh1 chrdl2 acan cntn2 sftpd wisp2

a930038c07rik acan ahsg alb ambn hmcn1 igf1 amelx angptl4 mmp9 ache ihh kera lama1 aspn bmp7 clec3b lama2 lama4 muc2 ndp nepn adamts12 col10a1 nid1 nid2 nov lamb1 loxl1 lpl adamts16 2) Gap junctions col11a1 lum mamdc2 npnt ntn3 oc90 col11a2 matn3 mepe olfml2a olfml2b optc adamts19 col12a1 mfap2 mfap4 otoa papln calb2 col14a1 adamts20 prss36 ptn col15a1 col1a1 adamts3 gjb6 mmp10 pxdn rbp3 rptn gja10 col1a2 col24a1 adamts5 serac1 serpine1 sftpa1 col2a1 col3a1 adamts6 mmp11 sftpd shh slc1a3 gja5 gjc3 col4a1 col4a2 adamts7 col4a5 col4a6 smoc2 snca sod3 gjd2 mmp13 sparc sparcl1 adamts8 gja8 CLUSTER spock1 mmp14 adamts9 gjb2 gjd4 01-02 spon2 col5a1 col5a2 mmp16 tff3 tgfbi adamtsl1 gjb3 gje1 col5a3 col6a1 thbs2 thbs4 adamtsl2 mip panx3 col6a2 col6a3 mmp19 bai2 bai3 c6 c8b col8a2 col9a1 thsd4 timp1 gjb4 thsd7a CLUSTER unc5c col9a2 cpa6 cpz thsd1 rspo1 sspo mmp1a tnr ush2a vcan unc5d 02 crtap dmbt1 wisp1 fbn1 wisp2 hpx gjb5 dspp ecm2 vit vtn wnt1 wisp3 fbn2 efemp2 eln mmp1b wnt10a wnt10b mfap2 emid2 emilin1 wnt2 wnt2b slc1a3 emilin3 enam mmp2 wnt3a wnt8a epyc fbln5 fbn1 wnt9a wnt9b zg16 snca GO_0043205 (FIBRIL) fbn2 fras1 mmp20 zp1 zp3 frem1 thsd4 mmp23 mmp7 gabra6 gpc2 gpc3 gpc5 gria3 grin1 grin2b mmp8 B hapln1 hapln3

Figure 5 (See legend on next page.) Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 11 of 19 http://arthritis-research.com/content/15/5/R174

(See figure on previous page.) Figure 5 Depletions in the transcriptional landscape of the extracellular milieu during transition from pre- to subclinical Sjögren’s syndrome. (A) Gene sets (GSs) depleted when the mice were 8 weeks of age define marked deceleration of extracellular matrix (ECM) turnover and significantly decreased transcription of genes associated with gap junction formation and neurotransmission. The layout parameters of Figure 4A correspond precisely to the layout parameters of Figure 3. The reference node allows estimation of scaling and direct comparison of Figures 3, 5A, 6A and 7A and Additional file 1: Figure S2A. Mmp, matrix metalloproteinase. (B) Annotation of the Markov cluster algorithm clusters displayed in (A) with their respective leading edge gene clouds. The layout parameters correspond precisely to the layout parameters of Figure 4. The reference node allows estimation of scaling and direct comparison of Figures 4, 5B, 6B and 7B and Additional file 1: Figure S2B. are upregulated approximately 8 weeks prior to the onset (Figures 5A and 5B) and thus represents the only class of of SS-like disease in C57BL/6.NOD-Aec1Aec2 mice. cell-cell junctions not enriched at 8 weeks of age. The third theme is dominated by genes coding for ligand-gated ion channels essential for neurotransmis- Transcriptional changes underlying themes being depleted sion (Figure 5B, Cluster_03) [36]. The largest part of during progression from pre- to subclinical Sjögren’s these genes encodes subunits of anionic Cys loop recep- – syndrome like disease tors (GABAA 12/12, GABAA-ρ 2/3 and GlyR 5/5), cat- In Figure 5A, CC term GO_0031012 (EXTRACELLULAR ionic Cys loop receptor subunits (serotonin-gated 5- MATRIX) is located at the center of the first major bio- HT3A and 5-HT3B and nicotinic ACh receptor 14/16 logical theme, becoming depleted during this time period subunits), 18 of 20 ionotropic sub- (Figure 5A). It was the largest GS that yielded significance units and ATP-gated channels P2X purinoceptors P2X1, in this study, with 159 of its 320 members contributing to P2X3, P2X5 and P2X6, as well as subsets of voltage- its significance (TAGS = 50%). The LE genes grouped in gated and acid-sensing potassium channels (for example, Cluster_01-01 delineate broad downregulation of genes amiloride-sensitive cation channels 1 to 3 (ACCN1 to encoding collagens of the ECM (Figure 5B) [34]. Contrib- ACCN3) and ACCN5). The remaining clusters of this uted by GS BI_INTRINSIC PATHWAY; TAGS = 73% and gene cloud represent mainly metabotropic receptors in- suggesting endothelial cell activation, this cluster also in- volved in sensory perception, whereas the LE genes asso- cludes coagulation factor–encoding genes. ciated with GO_0051971 (POSITIVE REGULATION OF The LE gene cloud of Cluster_01-02 in Figure 5B lists TRANSMISSION OF NERVE IMPULSE; TAGS = 62%) genes associated with all categories of specialized ECM also include inflammatory mediators such as IFNγ, proteins [30]. These include laminin (LAM) encoding sub- tumor necrosis factor (TNF) and interleukin 6 (IL-6), all units, such as Lama4 and Lamb1; proteoglycans, such as of which are known to decrease the threshold for nerve versican (Vcan); and glycoproteins, such as fibrillin 1 impulse generation (Additional file 2) [37]. (Fbn1)andFbn2. Genes coding for all matrix metallopro- teinases (MMPs) capable of degrading collagens, as well Transcriptional changes underlying stabilization of as distinct members of the disintegrins and metallo- subclinical disease between 8 and 12 weeks of age proteinases with thrombospondin motif (ADAMTS) LE genes associated with the continued depletion of family, also contributed to the significance of the GSs GO_0005581 (COLLAGEN); TAGS = 46% encode all pep- grouped in Cluster_01-02 (Figure 5A). ADAMTS pepti- tide chains for collagen type I, the most abundant collagen dases catalyze procollagens (for example, Adamts3)and of the ECM, and collagen type III (Additional file 1: Figure inhibit angiogenesis (for example, Adamts5, Adamts8, S2B and Additional file 2). Collagen type IV, which has Adamts9 and Adamts20) [35]. Genes annotated as coverage of 67%, is associated with basal membranes [34]. inducers of wingless-type mouse mammary tumor virus (MMTV) integration site family members (Wnt)(for example, Norrie disease (Ndp)), several Wnt genes (for Transcriptional changes underlying themes being enriched example, Wnt1) and all Wnt1-inducible signaling pathway during transition from subclinical to overt Sjögren’s proteins (Wisp1, Wisp2 and Wisp3) [30] completed the LE syndrome–like disease of Cluster_01-02. These changes complement the marked In Cluster_01 of Figure 6A, GO_0009897 (EXTERNAL and broad deceleration of ECM turnover as a potential SIDE OF PLASMA MEMBRANE); TAGS = 33% intercon- consequence of the ongoing innate immune response nects the EM-associated GSs that delineate the adaptive and/or delayed conclusion of developmental processes in effector immune response. The LE pattern of integrins the SGs of C57BL/6.NOD-Aec1Aec2 mice. (Cluster_01; Figure 6B) suggests an increase of CDH1 The second theme delineates downregulation of genes adhesive integrin αEβ7-expressing intraepithelial T cells, associated with GSs annotating gap junction core proteins whereas CD11c, encoded by Itgax and Itgb2, points to- (for example, PF_00029 (CONNEXIN); TAGS = 67%) ward antigen-presenting cells (APCs) of myeloid origin Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 12 of 19 http://arthritis-research.com/content/15/5/R174

1) Adaptive immunity 2) Neurotransmission

KE_04672 (partial reversal) (INTESTINAL GO_0030594 REFERENCE IMMUNE (NEUROTRANSMITTER NODE NETWORK RECEPTOR FOR IGA Interferons (Inf -1, PRODUCTION) ACTIVITY) KE_05320 -5,-9, Ifitm1) (AUTOIMMUNE THYROID DISEASE) B-cell activation and differentiation Neurotransmitter (BAFF-system/Cd40(l)/Icos(l)/Il4/Aicda) receptor expression (specific for several NK-cell-mediated pos. & neg. regulation classes of ligands) of cytotoxic activity GO_0005892 KE_05340 GO_0009897 (NICOTINIC (PRIMARY (EXTERNAL ACETYLCHOLINE-GATED IMMUNODEFICIENCY) SIDE OF PLASMA RECEPTOR-CHANNEL MEMBRANE) COMPLEX) Ag-presentation via MHCII BI_THELPER Immunological synapse Nicotinic acetylcholine receptors PATHWAY & tertiary signals (neuronal & muscle type)

RE_IMMUNOREGULATORY INTERACTIONS BETWEEN A GO_0042165 LYMPHOID (NEUROTRANSMITTER AND A BINDING) T-cell activation NON-LYMPHOID CELL PF_02931 PF_02932 (NEUROTRANSMITTER-GATED (NEUROTRANSMITTER-GATED ION-CHANNEL BI_TCYTOTOXIC ION-CHANNEL PATHWAY LIGAND TRANSMEMBRANE ACh-, GABA- & Gly- BINDING REGION) A gated ionotropic receptors DOMAIN)

1) Adaptive immunity 2) Neurotransmission (partial reversal) REFERENCE NODE

5830411n06rik itgae ache anxa9 brs3 cckbr drd1a drd2 gabra3 ace aqp4 btla h2-ab1 cd8a ccl19 ccr4 ccr7 drd3 h2-dma itgal gabra4 cd19 chrm4 cd8b1 h2-dmb2 itgb2 chrna1 gabra6 cd2 gabra2 gabrb1 cr2 ctla4 cxcl10 h2-eb1 ptprc chrna3 galr2 galr3 gabrb2 cxcl12 cxcl9 cd200r4 h2-m3 cd247 s1pr1 gpr143 gpr83 gabrb3 cxcr3 cxcr4 h2-oa sell sema7a spn stab2 grin1 grin2a grin2b fcer2a grin2d grin3a gabrd fas cd3d h2-ob h2-q7 h2-q8 ifna1 ifna5 htr7 fcgr2b hcst thy1 cd3e ifna9 il2 il4 chrna5 chrna6 kiss1r mc2r gabrg1 tg tpo tshb icam1 cd3g mrgpra1 gabrg2 icam2 ifitm1 CLUSTER chrna7 mrgpra4 gabrg3 01 nmbr nmur1 chrna9 nmur2 npffr2 gabrq ciita npy1r ntsr1 gabrr1 icos chrnb1 ntsr2 ppyr1 gabrr2 rag2 CLUSTER cd209b cd22 cd244 slc18a3 cd274 icosl prokr2 glra1 cd40 chrnb2 02 slc6a11 slc6a13 igh-6 tnfrsf13b glra2 il2rb sstr1 sstr2 cd40lg glra3 cd28 il2rg tnfrsf13c chrnb3 sstr4 sstr5 cga zap70 glra4 cd86 il7r tacr1 tacr2 lck htr3b chrnb4 tacr3 tspo itga4 itgb7 cd48 cd5 tnfrsf18 itgax cd69 cd74 tnfrsf4 aicda kcnj3 klrg1 chrnd tnfrsf9 btk klrd1 cd79a ulbp1 B klrk1 cd79b tnfrsf17 ms4a1 cd83 cd4 vcam1 madcam1 tnfsf13b chrne chrng

Figure 6 (See legend on next page.) Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 13 of 19 http://arthritis-research.com/content/15/5/R174

(See figure on previous page.) Figure 6 Enrichments in the transcriptional landscape of the extracellular milieu during transition from subclinical to overt Sjögren’s syndrome. (A) Gene sets (GSs) enriched at 16 weeks of age mirror the establishment of a pathogenic immune reaction in the targeted tissues and, in addition, reflect partial normalization of prior neurotransmitter receptor gene-associated alterations. The layout parameters correspond precisely to the layout parameters of Figure 3. The reference node allows estimation of scaling and direct comparison of Figures 3, 5A, 6A and 7A and Additional file 1: Figure S2A. ACh, acetylcholine; GABA, γ-aminobutyric acid; Gly, glycine; MHCII, major histocompatibility complex class II; NK, natural killer. (B) Annotation of the Markov cluster algorithm clusters displayed in Figure 5A with their respective leading edge gene clouds. The layout parameters of Figure 5B correspond precisely to the layout parameters of Figure 4. The reference node allows estimation of scaling and direct comparison of Figures 4, 5B, 6B and 7B and Additional file 1: Figure S2B.

[38]. The latter represent the most probable source for the lineages at this stage may thereby involve the two LE gene concomitant increase in transcription of various INFα- B7 family members B and T lymphocyte attenuator (Btla) encoding genes (Infα-1, Infα-5, Infα-9 and Ifitm1) in the and Tnf receptor superfamily 18 (Tnfrsf18) (Figure 6B). SGs of C57BL/6.NOD-Aec1Aec2 mice (Figures 6A and 6B). With respect to late costimulatory signals, Cd40:Cd40lg The establishment of a NK cell population in the tar- and inducible T-cell costimulator (Icos):IcosL are the geted tissues is supported by several distinct LE members receptor-ligand pairs present in the LE of Cluster_01 (Cluster_01; Figure 6B). Cytotoxicity-triggering receptors (Figure 6B). Both these systems, together with LE-gene NKG2-D type II integral (Klrk1), Cd244 Il4, are critical for mounting effective TH2responses[38]. and its ligand encoded by UL16-binding protein 1 (Ulbp1) B-cell-specific genes (for example, immunoglobulin represent three key components of NK cells’ effector path- heavy constant μ (Igh-6), Cd79a, Cd79b, Cd19 and Cd22) way. In contrast, the NK cell receptor complex encoded by are highly represented in the LE gene cloud of Cluster_01 killer cell lectinlike receptor subfamily D member 1 (Klrd1) (Figure 6B). Increased transcription of Tnfrsf13C (that is, and G member 1 (Klrg1) exert a regulatory anticytotoxic Baffr), Tnfrsf17 (that is, Bcma)andTnfrsf13B (that is, Taci), effect [38,39]. The LE genes Cd244 and Cd48,inconjunc- together with their common ligand Tnfsf13b (that is, Baff), tion with Cd2 and intercellular adhesion molecule 2 as well as the activation-induced cytidine deaminase gene (Icam2), may further suggest regulation of CD8+ T cells by (Aicda) (Figure 6B and Additional file 2; LE gene list NK cells. Expression of major histocompatibility complex for KE_04672 (INTESTINAL IMMUNE NETWORK FOR (MHC) and MHC-related genes, however, were skewed to- IGA PRODUCTION); TAGS = 60%), further indicates ward upregulation of MHC class II (MHCII) and MHCII strong signaling for survival, proliferation and differenti- invariant chain (Cd74) expression (Figure 6B). ation of B cells in SGs marked by overt disease [38,40]. The chemokine receptor-ligand profile characterizes The second biological theme enriched by 16 weeks of emigration of multiple APC and lymphocyte populations age pertains to neurotransmission and marks a partial re- (Cxcr4:Cxcl12), as well as reinforced recruitment of versal of changes that occurred earlier in the disease T-helper type 1 (TH1) cells, NK cells and plasmacytoid course. Of the 78 LE genes defining enrichment at this dendritic cells (Cxcl3:Cxcl9/Cxcl10 and Ccr7:Ccl19) later stage (Figure 6B; Cluster_02), 40 were previously as- (Figure 6B) [38]. Immune cell homing may also be facili- sociated with depletion of GSs concerning neurotransmis- tated by increased expression of LE genes that encode sion at 8 weeks of age (Figure 5B; Cluster_03). Reinitiating mucosal vascular addressin 1 transcription are mainly ACh-, GABA- and Gly-gated (MAdCAM-1) and lymphocyte function–associated anti- ionotropic receptors coding genes (Additional file 2). gen 1 (Lfa-1), encoded by Itgal and Itgb2 and Icam1. Overlaps were also found for genes encoding receptors for ICAM1 and LFA-1 ligation is also critical for Cd28- dopamine (that is, Drd1a) and substance P (that is, Tacr1). dependent T-cell activation [38]. The pattern of LE genes Unique to enrichment at 16 weeks of age were genes en- encoding costimulatory molecules assigns importance to coding for metabotropic receptors specific for ACh (that both the activating Cd28-dependent pathway and the is, Chrm4) and somatostatin (that is, Sstr1, Sstr2, Sstr4, inhibitory cytotoxic T-lymphocyte antigen 4 (Ctla4)– Sstr5) (Figure 6B and Additional file 2) [36]. dependent pathway (Figure 6B; Cluster_01). Regarding the T-cell-associated central component of the immunological Transcriptional changes underlying themes being depleted synapse, T-cell receptor (TCR) accessory proteins (for during transition from subclinical to overt Sjögren’s example, Cd3), TCR coreceptors Cd4 and Cd8 and TCR- syndrome–like disease associated molecules (for example, Cd45 (Ptprc)) are also GSs (Figure 7A) and their LE genes (Figure 7B) depleted covered by the LE gene cloud of Cluster_01. The con- and downregulated, respectively, during this time period, comitant upregulation of Il2, Il2rb and Il2rg,aswellasthe predominantly signify the reversal of previous enrich- presence of Cd69, represent effects downstream of T-cell ments in FAs and cell-cell junction–associated GSs ob- activation [38]. Regulating activation of T-cell effector served at 8 weeks of age (Figures 3 and 4). Pairwise Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 14 of 19 http://arthritis-research.com/content/15/5/R174

1) Focal adhesion & cell-cell junctions (partial reversal)

REFERENCE GO_0016327 NODE (APICOLATERAL PLASMA MEMBRANE) GO_0016328 (LATERAL PLASMA MEMBRANE)

KE_04530 (TIGHT Cell-cell JUNCTION)

GO_0043296 adherens junctions (APICAL JUNCTION COMPLEX)

GO_0005913 Apicolateral plasma (CELL-CELL ADHERENS JUNCTION) membrane & tight junctions

GO_0005923 GO_0005911 (TIGHT (CELL-CELL JUNCTION) JUNCTION)

GO_0030055 (CELL-SUBSTRATE GO_0007044 JUNCTION) (CELL-SUBSTRATE JUNCTION ASSEMBLY) GO_0034329 (CELL GO_0005925 JUNCTION (FOCAL ASSEMBLY) ADHESION)

GO_0005924 (CELL-SUBSTRATE Focal adhesion ADHERENS JUNCTION) NC_SIGNALING EVENTS MEDIATED BY maturation FOCAL ADHESION KINASE

GO_0005901 BI_INTEGRIN (CAVEOLA) A PATHWAY

1) Focal adhesion & cell-cell junctions (partial reversal)

REFERENCE pik3ca ankrd23 anxa5 slc5a1 epb4.1 epb4.1l1 ptpn21 ldb1 itga6 itgb3 erbb2ip lin7c fhl2 epb4.1l2 arhgef2 itgb4 spna2 NODE pard3 flnb git1 exoc3 exoc4 b4galt1 cdc42bpa magi1 ssx2ip stx3 pxn pard6a jub hic2 ilk f11r rfwd2 magi3 sympk gnai1 gnai3 lims1 irf2 hras1 sdc1 cdh1 pard6g synpo ppp2ca ppp2cb marveld2 lpp 2900073g15rik actg1 col17a1 tiam1 igsf5 actn4 akt1 akt2 ppp2r1a ppp2r1b itgb1 sdc4 ppp2r2c ppp2r2d mdc1 dpp4 micall2 amotl1 dag1 prkcd prkce nexn jun tjp1 inadl syne2 prkch itgb5 b230120h23rik cask dsc2 gja1 gjb1 mpp5 tenc1 pak1 map2k1 tjp2 jam2 cdk4 prkcz map2k4 mapk1 cgn tgfb1i1 gjb2 mpp7 kras llgl2 jup parva mapk3 mapk8 prkd1 myo1c tjp3 cgnl1 tln2 ptk2 ocln pgm5 ppp1r12b ptprj ptprk cldn1 tns4 pkp3 plekha7 trip6 pvrl2 cldn10 shroom2 cldn23 vcl zfp384 CLUSTER cldn3 01 cldn4 atp1b3 ldlr cdh13 cln3 cldn7 dlc1 cav1 lipe arhgap26 ctnnb1 krt14 krt5 gnaq hdac6 dst arhgef11 cdh13 myof clic4 cldn8 hk1 igf1r insr ldlr ebag9 ctnnd1 lama3 arhgef7 lipe myof cln3 ptrf csnk2a1 crb3 ptrf slc27a1 enah cxadr lamb3 cald1 csnk2a2 csnk2b slc2a1 lamc1 dlc1 slc27a1 smpd2 epb4.1l5 epcam lamc2 cav1 rab13 tsc2 ctnna1 src fn1 gnpat erbb3 pdpk1 plec crkl itga1 sgsm3 gnaq slc2a1 cttn braf hdac6 smpd2 smad3 capn2 myh10 taok2 ccnd1 myh11 keap1 rac1 crk myh14 tesk2 arhgap6 hk1 src pkp1 dock1 myl12b raf1 tns1 kifc3 2310057j16rik grb7 myl9 rgnef cd151 tsc1 kirrel add3 pmp22 grlf1 nras rock2 cd9 mllt4 igf1r tsc2 bcr GO_0005901 capn1 mapk9 insr B capns1 nck2 pcdh1 tmem47 akap7 dlg5 rras2 map2k2 fzd6 atp1b3 (CAVEOLA) Figure 7 (See legend on next page.) Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 15 of 19 http://arthritis-research.com/content/15/5/R174

(See figure on previous page.) Figure 7 Depletions in the transcriptional landscape of the extracellular milieu during transition from subclinical to overt Sjögren’s syndrome. (A) Gene sets (GSs) depleted at 16 weeks of age signify, to a major extent, the partial reversal of enrichments in FAs and cell–cell junction-associated GSs observed at 8 weeks of age. The layout parameters of Figure 6A correspond precisely to the layout parameters of Figure 3. The reference node allows estimation of scaling and direct comparison of Figures 3, 5A, 6A and 7A and Additional file 1: Figure S2A. (B) Annotation of the Markov cluster algorithm clusters displayed in Figure 6A with their respective LE gene clouds. The layout parameters of Figure 6B correspond precisely to the layout parameters of Figure 4. The reference node allows estimation of scaling and direct comparison of Figures 4, 5B, 6B and 7B and Additional file 1: Figure S2B. comparison of the overlapping GSs revealed that, on aver- of these data sets still pose great challenges. This is also age, 54% of LE genes contributing to depletion at 16 weeks true with regard to delineating the underlying biological of age also contributed to these GSs’ prior enrichment at and chronological complex changes in biological states, 8 weeks of age. The highest percentage of LE members such as covert stages of autoimmunity in an organ subse- following this pattern was identified for NC_SIGNALING quently targeted by an autoimmune disease. Thus, the EVENTS MEDIATED BY FOCAL ADHESION KINASE, possibility of assessing all relationships among all compo- with 79%, and the lowest percentage was found for nents of a biological system simultaneously with an inte- GO_0043296 (APICAL JUNCTION COMPLEX), with grated and standardized concept such as the one 43% (Additional file 2). The LE genes not included in presented herein meets a clear demand [41]. these LE overlaps did not define additional biological With respect to the immune system–specific findings of themes, but instead contributed predominantly to the in- this study, considering the presence of IFN signatures in pa- creased average coverage of the EM-related GSs at tients with SS [11], enrichment of innate immune response 16 weeks (TAGS 45%) compared to 8 weeks of age (TAGS pathways at 8 weeks of age in C57BL/6.NOD-Aec1Aec2 38%) (Additional file 2). mice was anticipated. In addition, the molecular basis underlying the effector immune response at disease onset Discussion mimicked all major aspects of sialadenitis described in pa- Although the technology for generating global gene ex- tients with SS [9]. Further validating our findings is that pression profiles has matured, analysis and interpretation C57BL/6.NOD-Aec1Aec2 mice, during their spontaneous

25 * * 08wk enriched Aec2 20 **** 08wk depleted & * * ****** 12wk depleted

Aec1 15 * 16wk enriched 16wk depleted 10

per GS per per theme 5

% of LE genes in 0

2) Gap junctions1) Collagen I & III 1) Innate immunity (partial reversal) 1) Adaptive immunity 2) Insr / Igfr14) signalingCell-cell junctions &3) Neurotransmission 2) Neurotransmission

3) Focal adhesioncytoskeleton & motility organization 1) Focaljunctions adhesion (partial & cell-cell reversal) 1) Deceleration of ECM turnover Figure 8 Degree of dependence of the major biological themes on genes located in Aec1 and Aec2 (mean ± SEM). Proportion of leading edge (LE) genes located in the disease-causing congenic loci Aec1 and Aec2/gene set (GS)/major biological theme within the same disease phase were compared using one-way analysis of variance and Tukey’s post hoc test. P-values <0.05 were considered significant (*P < 0.05, **P < 0.01, ***P < 0.0001). Columns represent the means, and error bars represent the standard errors of the mean. ECM, extracellular matrix; Igfr1, insulin-like growth factor receptor 1; Insr, insulin receptor. Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 16 of 19 http://arthritis-research.com/content/15/5/R174

and slow development toward overt disease, displayed alter- herein represents a crucial feature which has not previ- ations in biological pathways that, if knocked out or overex- ously been reported in the context of this data set. (3) pressed from birth on a healthy genetic background, induce By applying this methodology, the data set could be aspects of SS. These models of SS include mice deficient for interpreted in significantly more detail, which subse- NF-κB feedback regulation (C57BL/6.IκBαM/M)andmice quently could be combined to present a more compre- transgenic for Baff [7]. hensive picture. Direct comparison of the results presented herein with The transcriptional landscape of the EM of tissues tar- the conclusions formulated subsequent to analyses using geted by autoimmunity described herein opens a novel conventional “top gene list” approaches [19,20] defines the and integrative perspective on the development of auto- added value of this systems biology–based methodology as immune diseases that might be of more general relevance follows. (1) Focusing on the transcriptional landscape re- (Figure 9) [42]. As a first step, it will be important to in- lated to and associated with the EM seemed adequate to vestigate how strongly, in other experimental models of map, in its entirety and in a standardized fashion, the al- autoimmunity, the LE genes differ. The chronological in- terations in the SG’s decision-making processes associated terrelationships and major biological themes identified with the emergence of autoimmunity in this model. Em- herein may be the same, however. This knowledge may phasizing the EM prevented the mapping of some of prove especially critical when aiming to delineate, on a previously documented downstream effects induced via systems level, the mechanisms of action and the targeted signals transmitted by the EM, however [19,20]. (2) The organ’s state subsequent to experimental immunomodula- early activation of the innate immune system described tory intervention [43].

40 no. o 400 f LE g 20 e nes in in nes

200 Aec1

0 & Aec2

no. of LE genes 0 -20

-200

4812 16 age in weeks

aberrant salivary gland physiology autoantibodies focal inflammation hyposalivation Immunity Focal adhesion & motility, ECM turnover cell-cell junctions & Insr / Igfr1 signaling cytoskeleton organization Neurotransmission Figure 9 Summary of the interrelationships between the major biological themes defining the alterations in the extracellular milieu during the emergence of Sjögren’s syndrome. Time points are aligned in correspondence with the chronology of the development of major features of Sjögren’s syndrome in C57BL/6.NOD-Aec1Aec2 mice [8]. Plotted on the y-axis is the cumulative number of leading edge (LE) genes (solid lines) associated with enrichment and depletion of the respective summarized biological theme over time. The right y-axis represents the same parameter, but for the LE genes specifically located in the disease-causing congenic loci Aec1 and Aec2 (dashed lines). The summarized biological theme Immunity comprises the themes Innate immunity (Figure 3) and Adaptive immunity (Figure 6). The summarized biological theme Focal adhesion and motility, cell–cell junctions and cytoskeleton organization comprises the themes Focal adhesion and motility (Figure 3), Gap junctions (Figure 5) and Focal adhesion and cell-cell junctions (Figure 7). The summarized biological theme Insr/Igfr1 signaling comprises the theme Insr/Igfr1 signaling (Figure 3). The summarized biological theme ECM turnover comprises the themes ECM turnover (Figure 5) and Collagen I and III (Additional file 1: Figure S2). The summarized biological theme Neurotransmission comprises the themes Neurotransmission in Figures 5 and 6. ECM, extracellular matrix. Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 17 of 19 http://arthritis-research.com/content/15/5/R174

The chronology of the etiopathology defined herein es- basis that has allowed us to formulate conclusions in tablishes several important points. First, long before overt agreement with the generality of our aim. In the future, experimental SS, susceptibility loci-dependent and par- assigning specific weights to the individual genes based tially transient alterations associated with the targeted tis- on their uniqueness or importance to a specific GS may sue’s homeostasis and integrity formed the basis for an further standardize and facilitate the final steps of data innate immune reaction. The latter, in contrast, was interpretation. In the meantime, it is important to pro- dependent predominantly on genes descending from vide additional user-friendly graphical layouts of the net- the asymptomatic C57BL/6 strain that served as a gene- works, such as the ones presented herein, to enable the tic background for the generation of C57BL/6.NOD- reader to scrutinize the authors’ detailed interpretation Aec1Aec2 mice. If a role of genes governing the SGs’ of the networks. homeostatic state at such an early stage of autoimmunity can be confirmed, these genes may indeed crucially con- Conclusions tribute to an individual’s risk of developing SS [5]. Second, By adhering to the principles of systems biology and the long-lasting, stable subclinical disease state may eluci- adapting bioinformatics-based methodologies and data date novel diagnostic strategies for identification of SS at visualization to suit our aims, this study has delineated a an earlier state and thereby enable timely immunomodula- novel perspective on the chronology and interplay be- tory treatment [44]. Third, major themes that defined this tween the SGs’ EM and the role of the innate and adaptive stable subclinical disease were abandoned concomitantly immune systems during the emergence of spontaneous, with the onset of overt disease. This permits speculation experimental SS (Figure 9). The timeline defined herein about whether these transient alterations may represent highlights the importance of genes governing the target processes initiated by the SGs to resolve environmental tissue’s homeostatic state in establishing a stable subclin- challenges or to compensate for developmental deficien- ical disease state long before the clinical manifestation of cies, primarily without involvement of the adaptive im- SS. Formulating conclusions in agreement with the gener- mune system. Fourth, LE gene patterns associated with ality of our aim was possible only after having developed costimulatory signals revealed both effector and regulatory and applied the integrated data analysis and data ligand-receptor pairs’ being present, indicating that ef- visualization pipeline, which is also presented herein. This fector as well as immunoregulatory processes govern the data-driven approach advances systematic and impartial onset of overt disease [45]. interpretation of global datasets on the background of Although global data sets are seldom adequate to define standardized, consensus-based, a priori-defined biological the role of a single gene or protein, the isolated study of knowledge. It is widely applicable to the fields of immun- individual components in turn is limited in terms of eluci- ology and rheumatology and will greatly facilitate analysis dating how properties of biological systems emerge as a of complex alterations in biological states on a systems result of coordinated interactions between its numerous level, such as changes induced as a consequence of experi- members and processes [41]. To take full advantage of the mental treatment interventions. unbiased nature of “-omics” data sets, our concept inte- grates data analysis by relying extensively on bioinformat- Additional files ics resources for compilation of consensus-based, a priori- defined biological knowledge with an interactive model for Additional file 1: Figure S1. Verification of the microarray data. Figure data interpretation based on networks computed entirely S2. Stabilization of the subclinical disease state between 8 and 12 weeks of age. Figures S3 to S7. Contributions of genes located in susceptibility from experimental data. Importantly, this concept is trans- regions Aec1 and Aec2 to enrichments and depletions associated with ferable to global data sets of any nature and achieves an transition from pre- to subclinical Sjögren’s syndrome (SS) and to onset important reduction in the number of arbitrary cutoffs set of overt SS, respectively. Figure S8. Interactome model based on the leading edge members associated with enrichments in the transcriptional at the stage of data analysis. It also diminishes significantly landscape of the extracellular milieu during transition from pre- to the amount of personal bias commonly introduced during subclinical Sjögren’s syndrome between 4 and 8 weeks of age. the process of data interpretation [46] and overcomes the Additional file 2: All node parameters, LE genes and LE genes confines of lists and matrices, which have clear limitations located in susceptibility regions Aec1 and Aec2 in tabular form. in conveying large amounts of complex data and interrela- Additional file 3: The networks displayed in Figures 3-7 as infinitely scalable and electronically searchable vector graphics. tionships [47]. Obviously, to base such mappings on additional di- Abbreviations mensions, such as global protein synthesis or posttran- ACh: Acetylcholine; ADAMTS: A disintegrin-like and metalloprotease with scriptional modification profiles, would significantly thrombospondin motifs; AICDA: Activation-induced cytidine deaminase gene; improve the validity of such analyses. They will become AKT: Protein kinase B; Atg: Autophagy-related protein; BTLA: B7 family member B- and T-lymphocyte attenuator; Capn: Calpain; CDH1: E-cadherin; more feasible technologically and economically in the fu- Ctla4: Cytotoxic T-lymphocyte antigen 4; ES: Enrichment score; FBN1: Fibrillin; ture [48]. In this study, we have computed a meaningful Fgfr1: Fibroblast growth factor receptor 1; GABA: γ-aminobutyric acid; Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 18 of 19 http://arthritis-research.com/content/15/5/R174

Gly: Glycine; Icam: Intercellular adhesion molecule; Icos: Inducible T-cell costi- 3. Rai E, Wakeland EK: Genetic predisposition to autoimmunity—What have mulator; IFIH1: Interferon-induced helicase C domain–containing protein 1; we learned? Semin Immunol 2011, 23:67–83. IFN: Interferon; Igfr1: Insulin-like growth factor receptor 1; IgH- 4. Barbalat R, Ewald SE, Mouchess ML, Barton GM: Nucleic acid recognition 6: Immunoglobulin heavy chain, type μ; IL: Interleukin; Insr: Insulin receptor; by the innate immune system. Annu Rev Immunol 2011, 29:185–214. IRF3: Interferon regulatory factor 3; ITG: Integrin; KLRD1: Killer cell lectinlike 5. Swamy M, Jamora C, Havran W, Hayday A: Epithelial decision makers: in receptor subfamily D member 1; KLRG1: Killer cell lectinlike receptor search of the ‘epimmunome’. Nat Immunol 2010, 11:656–665. subfamily G member 1; KLRK1: Cytotoxicity-triggering receptor; NKG2-D: Type 6. Kollias G, Papadaki P, Apparailly F, Vervoordeldonk MJ, Holmdahl R, II integral membrane protein; LAM: Laminin; LFA-1: Lymphocyte function– Baumans V, Desaintes C, Di Santo J, Distler J, Garside P, Hegen M, associated antigen 1; Ly96: Lymphocyte antigen 96; MadCAM1: Vascular Huizinga TWJ, Jüngel A, Klareskog L, McInnes I, Ragoussis I, Schett G, addressin cell adhesion molecule 1; MAPK8: Mitogen-activated protein kinase t’ Hart B, Tak PP, Toes R, van den Berg W, Wurst W, Gay S: Animal models 8; MET: Hepatocyte growth factor receptor; MMP: Matrix metalloproteinase; for arthritis: innovative tools for prevention and treatment. Ann Rheum MMTV: Mouse mammary tumor virus integration site family member; Dis 2011, 70:1357–1362. mTorC: Mammalian target of rapamycin complex; NDP: Norrie disease; NF- 7. Delaleu N, Nguyen CQ, Peck AB, Jonsson R: Sjögren’s syndrome: studying κB: κ-light-chain-enhancer of activated B cells; Ngf: Nerve growth factor; the disease in mice. Arthritis Res Ther 2011, 13:217. PI3K: Phosphoinositide 3-kinase; P2X: ATP-gated channels P2X purinoceptor; 8. Cha S, Nagashima H, Brown VB, Peck AB, Humphreys-Beher MG: Two NOD p75NTR: p75 neurotrophin receptor; Smad: Smad family member; Idd-associated intervals contribute synergistically to the development of Spp1: Osteopontin; STAT: Signal transducer and activator of transcription; autoimmune exocrinopathy (Sjögren’s syndrome) on a healthy murine TAGS: Percentage of gene hits before (for positive enrichment score) or after background. Arthritis Rheum 2002, 46:1390–1398. (for negative enrichment score) peak in running enrichment score, which 9. Delaleu N, Manoussakis MN, Moutsopoulos HM, Jonsson R: Etiology and indicates percentage of genes contributing to enrichment score; pathogenesis of Sjögren’s syndrome with special emphasis on the Tgfβ: Transforming growth factor β; TLR: Toll-like receptor; TNF: Tumor salivary glands.InSjögren’s Syndrome. 1st edition. Edited by Fox RI, Fox C. necrosis factor; TNFRSF: Tumor necrosis factor receptor superfamily; New York: Springer; 2011:243–268. TRAIL: Tumor necrosis factor–related apoptosis-inducing ligand; Ulbp1: UL16 10. Tobón GJ, Pers JO, Youinou P, Saraux A: B cell-targeted therapies in binding protein 1; Vcan: Versican; Wisp: Wingless-type MMTV Integration site Sjögren’s syndrome. Autoimmun Rev 2010, 9:224–228. family 1-inducible-signaling pathway protein; Wnt: Wingless-type MMTV 11. Båve U, Nordmark G, Lövgren T, Rönnelid J, Cajander S, Eloranta ML, Alm GV, Integration site family. Rönnblom L: Activation of the type I interferon system in primary Sjögren’s syndrome: a possible etiopathogenic mechanism. Arthritis Rheum 2005, Competing interests 52:1185–1195. The authors have no financial or nonfinancial competing interests to declare. 12. Low HZ, Witte T: Aspects of innate immunity in Sjögren’s syndrome. Arthritis Res Ther 2011, 13:218. 13. Ewert P, Aguilera S, Alliende C, Kwon YJ, Albornoz A, Molina C, Urzúa U, Authors’ contributions Quest AF, Olea N, Pérez P, Castro I, Barrera MJ, Romo R, Hermoso M, ND conceived and implemented the data analysis and data visualization Leyton C, González MJ: Disruption of structure in salivary pipeline, analyzed and interpreted the data, created all the figures and wrote glands from Sjögren’s syndrome patients is linked to proinflammatory the manuscript. CQN took part in designing the study, carried out the animal cytokine exposure. Arthritis Rheum 2010, 62:1280–1289. experiments as well as the remaining experiments in the laboratory, and was 14. Sequeira SJ, Larsen M, DeVine T: Extracellular matrix and growth factors in involved in drafting the manuscript. KMT conceived and successfully adapted salivary gland development. Front Oral Biol 2010, 14:48–77. the WordCloud plug-in and was involved in drafting the manuscript. RJ con- 15. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, tributed to the conception of the study, was involved in revising the manu- Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP: Gene set script and coordinated the project. ABP conceived the study, planned the enrichment analysis: a knowledge-based approach for interpreting animal experiments, participated in analyzing and interpreting the data and genome-wide expression profiles. Proc Natl Acad Sci U S A 2005, wrote the manuscript. All authors read and approved the final manuscript. 102:15545–15550. 16. Morris JH, Apeltsin L, Newman AM, Baumbach J, Wittkop T, Su G, Bader GD, Acknowledgements Ferrin TE: clusterMaker: a multi-algorithm clustering plugin for Cytoscape. We thank Dr B Delaleu-Justitz for careful revision of the manuscript. This BMC Bioinforma 2011, 12:436. study was funded by the Research Council of Norway (ND, KMT and RJ), the 17. Smoot ME, Ono K, Ruscheinski J, Wang PL, Ideker T: Cytoscape 2.8: new Bergen Medical Research Foundation (ND and RJ), the Broegelmann features for data integration and network visualization. Bioinformatics Foundation (ND and RJ), Helse Bergen (ND and RJ) and National Institute of 2011, 27:431–432. Dental and Craniofacial Research (NIDCR) grant R01 DE-014344 (to ABP), 18. NguyenCQ,SharmaA,SheJX,McIndoeRA,PeckAB:Differential National Institutes of Health grant AI0819529 (to ABP) and NIDCR R01 gene expressions in the lacrimal gland during development and DE-018958 (to CQN). onset of keratoconjunctivitis sicca in Sjögren’s syndrome (SJS)-like disease of the C57BL/6.NOD-Aec1Aec2 mouse. Exp Eye Res 2009, Author details 88:398–409. 1 Broegelmann Research Laboratory, Department of Clinical Science, University 19. NguyenCQ,SharmaA,LeeBH,SheJX,McIndoeRA,PeckAB:Differential gene of Bergen, Laboratory Building, Jonas Liesvei 65, 5021 Bergen, Norway. expression in the salivary gland during development and onset of 2 Department of Infectious Diseases and Pathology, College of Veterinary xerostomia in Sjögren’s syndrome-like disease of the C57BL/6.NOD-Aec1Aec2 Medicine, University of Florida, 2015 S.W. 16th Avenue, Gainesville, FL 32610, mouse. Arthritis Res Ther 2009, 11:R56. USA. 3Computational Biology Unit, Uni-Computing, UniResearch AS, Thormøh- 4 20. Horvath S, Nazmul-Hossain ANM, Pollard RPE, Kroese FGM, Vissink A, lensgate 55, 5008 Bergen, Norway. Department of Rheumatology, Haukeland Kallenberg CGM, Spijkervet FKL, Bootsma H, Michie SA, Gorr SU, Peck AB, University Hospital, Jonas Liesvei 65, 5021 Bergen, Norway. CaiC,ZhouH,WongDTW:Systems analysis of primary Sjögren’s syndrome pathogenesis in salivary glands identifies shared Received: 14 December 2012 Accepted: 11 October 2013 pathways in human and a mouse model. Arthritis Res Ther 2012, Published: 31 October 2013 14:R238. 21. http://www.sanger.ac.uk/resources/databases/pfam.html. References 22. Merico D, Isserlin R, Stueker O, Emili A, Bader GD: Enrichment map: a 1. Ziegler AG, Nepom GT: Prediction and pathogenesis in type 1 diabetes. network-based method for gene-set enrichment visualization and Immunity 2010, 32:468–478. interpretation. PLoS One 2010, 5:e13984. 2. Arbuckle MR, McClain MT, Rubertone MV, Scofield RH, Dennis GJ, James 23. Glez-Peña D, Gómez-López G, Pisano DG, Fdez-Riverola F: WhichGenes: a JA, Harley JB: Development of autoantibodies before the clinical web-based tool for gathering, building, storing and exporting gene sets onset of systemic lupus erythematosus. N Engl J Med 2003, with application in gene set enrichment analysis. Nucleic Acids Res 2009, 349:1526–1533. 37:W329–W334. Delaleu et al. Arthritis Research & Therapy 2013, 15:R174 Page 19 of 19 http://arthritis-research.com/content/15/5/R174

24. Oesper L, Merico D, Isserlin R, Bader GD: WordCloud: a Cytoscape plugin 47. Gehlenborg N, O’Donoghue SI, Baliga NS, Goesmann A, Hibbs MA, Kitano H, to create a visual semantic summary of networks. Source Code Biol Med Kohlbacher O, Neuweger H, Schneider R, Tenenbaum D, Gavin AC: Visualization 2011, 6:7. of omics data for systems biology. Nat Methods 2010, 7:S56–S68. 25. Vallabhapurapu S, Karin M: Regulation and function of NF-κB 48. Chen C, McGarvey PB, Huang H, Wu CH: Protein bioinformatics transcription factors in the immune system. Annu Rev Immunol 2009, infrastructure for the integration and analysis of multiple 27:693–733. high-throughput “omics” data. Adv Bioinformatics 2010, 2010:423589. 26. Gentry JJ, Barker PA, Carter BD: The p75 neurotrophin receptor: multiple interactors and numerous functions. Prog Brain Res 2004, doi:10.1186/ar4362 146:25–39. Cite this article as: Delaleu et al.: Transcriptional landscapes of emerging 27. Johnstone RW, Frew AJ, Smyth MJ: The TRAIL apoptotic pathway in autoimmunity: transient aberrations in the targeted tissue’s extracellular cancer onset, progression and therapy. Nat Rev Cancer 2008, 8:782–798. milieu precede immune responses in Sjögren’s syndrome. Arthritis 28. Conway K, Price P, Harding KG, Jiang WG: The molecular and clinical Research & Therapy 2013 15:R174. impact of hepatocyte growth factor, its receptor, activators, and inhibitors in wound healing. Wound Repair Regen 2006, 14:2–10. 29. Caron E, Ghosh S, Matsuoka Y, Ashton-Beaucage D, Therrien M, Lemieux S, Perreault C, Roux PP, Kitano H: A comprehensive map of the mTOR signaling network. Mol Syst Biol 2010, 6:453. 30. Alberts B, Johnson A, Lewis J, Raff M, Roberts K, Walter P: Molecular Biology of the Cell. 5th edition. New York: Garland Science; 2008. 31. Geiger B, Spatz JP, Bershadsky AD: Environmental sensing through focal adhesions. Nat Rev Mol Cell Biol 2009, 10:21–33. 32. Brooks PC, Klemke RL, Schon S, Lewis JM, Schwartz MA, Cheresh DA: Insulin-like growth factor receptor cooperates with integrin αv β5 to promote tumor cell dissemination in vivo. JClinInvest1997, 99:1390–1398. 33. Ivaska J, Heino J: Cooperation between integrins and growth factor receptors in signaling and endocytosis. Annu Rev Cell Dev Biol 2011, 27:291–320. 34. Shoulders MD, Raines RT: Collagen structure and stability. Annu Rev Biochem 2009, 78:929–958. 35. Georgiadis D, Yiotakis A: Specific targeting of metzincin family members with small-molecule inhibitors: progress toward a multifarious challenge. Bioorg Med Chem 2008, 16:8781–8794. 36. Purves D, Augustine GJ, Fitzpatrick D, Hall WC, LaMantia AS, White LE (Eds): Neuroscience. 5th edition. Sunderland, MA: Sinauer Associates; 2012. 37. Ren K, Dubner R: Interactions between the immune and nervous systems in pain. Nat Med 2010, 16:1267–1276. 38. Murphy K: Janeway’s Immunobiology. 8th edition. New York: Garland Science; 2012. 39. Vivier E, Tomasello E, Baratin M, Walzer T, Ugolini S: Functions of natural killer cells. Nat Immunol 2008, 9:503–510. 40. Mackay F, Schneider P: Cracking the BAFF code. Nat Rev Immunol 2009, 9:491–502. 41. van der Greef J, Martin S, Juhasz P, Adourian A, Plasterer T, Verheij ER, McBurney RN: The art and practice of systems biology in medicine: mapping patterns of relationships. JProteomeRes2007, 6:1540–1559. 42. Zhernakova A, van Diemen CC, Wijmenga C: Detecting shared pathogenesis from the shared genetics of immune-related diseases. Nat Rev Genet 2009, 10:43–55. 43. Campbell SJ, Gaulton A, Marshall J, Bichko D, Martin S, Brouwer C, Harland L: Visualizing the drug target landscape. Drug Discov Today 2010, 15:3–15. 44. Bach JF, Chatenoud L: A historical view from thirty eventful years of immunotherapy in autoimmune diabetes. Semin Immunol 2011, 23:174–181. 45. Samy ET, Wheeler KM, Roper RJ, Teuscher C, Tung KS: Cutting edge: autoimmune disease in day 3 thymectomized mice is actively controlled by endogenous disease-specific regulatory T cells. J Immunol 2008, 180:4366–4370. Submit your next manuscript to BioMed Central 46. Ideker T, Dutkowski J, Hood L: Boosting signal-to-noise in complex biology: and take full advantage of: prior knowledge is power. Cell 2011, 144:860–863. • Convenient online submission • Thorough peer review • No space constraints or color figure charges • Immediate publication on acceptance • Inclusion in PubMed, CAS, Scopus and Google Scholar • Research which is freely available for redistribution

Submit your manuscript at www.biomedcentral.com/submit