Disease Mechanisms in RheumatologyTools ... - Wiley Online Library

ARTHRITIS & RHEUMATISM Vol. 65, No. 6, June 2013, pp 1407–1417 DOI 10.1002/art.37909 © 2013, American College of Rheumatology

Arthritis & Rheumatism An Official Journal of the American College of Rheumatology www.arthritisrheum.org and wileyonlinelibrary.com

DISEASE MECHANISMS IN RHEUMATOLOGY—TOOLS AND PATHWAYS

Current Perspectives on Systems Immunology Approaches to Rheumatic Diseases Laurent Chiche,1 Noemie Jourde-Chiche,2 Virginia Pascual,3 and Damien Chaussabel4 and “hypothesis-free” assessment of genes and complex biologic networks or pathways implicated in specific conditions. Directly applied to patients, it has for instance revealed an interferon (IFN) signature in systemic lupus erythematosus (SLE), where classic genetic and in vitro “focused” approaches (Figure 1A) or in vivo animal studies had either failed or proved poorly translational. Second, transcriptomics can be used to guide or complement other biologic approaches, including proteomic or genomic studies. Finally, transcriptomics can provide the basis for translational clinical applications. Indeed, in comparison to real-time polymerase chain reaction (PCR) that measures gene expression of only a limited number of transcripts, whole transcriptome profiling has clearly become the method of choice for biomarker discovery in various fields including rheumatology (3–6). In contrast to sequence-based genetic approaches studying the association of rheumatic diseases with common polymorphisms or rare mutations, transcriptomic studies rely on the measurement of RNA (transcript) abundance that changes over time, especially under the influence of environmental factors or pathologic conditions. Robust and cost-effective highthroughput gene expression microarray platforms allow the simultaneous measurement of tens of thousands of transcripts in a given sample. Indeed, currently available platforms permit the measurement of up to 50,000 mRNA transcripts simultaneously. Significant efforts have been made by the scientific community to define the conditions required to obtain reliable high-quality data, but a number of technical and analytical challenges persist (7). Microarrays do not provide fully quantitative results, and changes in transcript abundance must be measured in reference to

Introduction In this review, we survey 3 areas of major interest to rheumatologists. These are systems approaches and rheumatic diseases, the current status of transcriptome profiling technologies, and the new focus on blood transcriptomics. Systems approaches are integrative approaches that rely on high-throughput techniques to capture the state of a system on a global scale. Systems immunology leverages these technologies to investigate complex immunologic processes. During the last decade, rheumatologists have evaluated the utility of systems approaches, in particular transcriptomics (the determination of the expression level of messenger RNAs [mRNAs] in a given cell population, also referred to as expression profiling), especially in the setting of autoimmune rheumatic diseases resulting from the complex combination of environmental, epigenetic, and genetic factors (1,2). First, this systems approach allows a global 1 Laurent Chiche, MD, PhD: Benaroya Research Institute, Seattle, Washington, and Aix-Marseille Universite´ and Ho ˆpital de la Conception, Assistance Publique Ho ˆpitaux de Marseille, Marseille, France; 2Noemie Jourde-Chiche, MD, PhD: Benaroya Research Institute, Seattle, Washington, Baylor Institute for Immunologic Research, Dallas, Texas, and Aix-Marseille Universite´ and Ho ˆpital de la Conception, Assistance Publique Ho ˆpitaux de Marseille, Marseille, France; 3Virginia Pascual, MD: Baylor Institute for Immunologic Research, Dallas, Texas; 4Damien Chaussabel, PhD: Benaroya Research Institute, Seattle, Washington. Dr. Pascual has received consulting fees, speaking fees, and/or honoraria from Genentech, Pfizer, Roche, Bristol-Myers Squibb, and Kyowa Hakko Kirin (less than $10,000 each) and research grants from Roche. Address correspondence to Damien Chaussabel, PhD, Systems Immunology Division, Benaroya Research Institute, 1201 Ninth Avenue, Seattle, WA 98101. E-mail: dchaussabel@benaroya research.org. Submitted for publication December 13, 2012; accepted in revised form February 14, 2013.

1407

1408

CHICHE ET AL

Figure 1. Blood transcriptome analysis and specific experimental designs. In contrast to more focused biologic approaches (A), microarray technology can be used on peripheral blood samples (B) to obtain whole transcriptome analysis as output. Whole blood (WB), peripheral blood mononuclear cells (PBMCs), or even isolated single-cell types (38) can be considered. A “reverse proteomic” or “reporter assay” approach (41) can also be applied, using PBMCs from healthy donors that are assessed with microarray after in vitro exposure to patients’ serum or plasma (C). Finally, microarray can also be used for the discovery of potential pharmacodynamic markers to monitor the efficacy of immune-targeted drugs (D): healthy donor PBMCs, human cell lines, or whole blood samples can be stimulated ex vivo with specific molecules, and healthy donor PBMCs can be exposed ex vivo to patients’ sera with or without various doses of the specific inhibitor of this same molecule (26).

control samples. Many parameters other than disease status may influence transcript abundance, including cell-subset composition of the studied tissue, age, sex, medications, or comorbidities. These sources of “biologic” variability are much more difficult to control than potential sources of “technical” variability for which mitigating strategies have been developed. Therefore, careful study design and selection of patients is paramount to the success of these studies. As recently pointed out by Tektonidou and Ward, however, fewer than half of studies of diagnostic biomarkers in the rheumatology field included groups that were agematched, sex-matched, or controlled for treatment (3).

Due to the high dimensionality of the data and potential for “overfitting” (false discovery), independent validation on separate data sets or through the reinterpretation of microarray data deposited in public repositories (GEO, European Bioinformatics Institute) is warranted, as recommended in the publication standards of a wide range of journals (8). Finally, and although it may seem paradoxical with such an unbiased approach, the functional interpretation of microarray results still relies mostly on “knowledge-driven” annotations/tools. New technologies have recently become available that may complement and in some cases replace microarray technology. An example that presents various

SYSTEMS IMMUNOLOGY APPROACHES TO RHEUMATIC DISEASES

advantages over microarray is RNA sequencing (9). First, RNA sequencing does not rely on the design of probes. Indeed, RNA is converted to a library of complementary DNA fragments that are sequenced, providing millions of short sequences or reads (length and number of reads vary depending on the platform used) that are then uniquely mapped on a reference genome. Second, in addition to transcript abundance, this approach provides information on transcriptome structure (splice variants), profiles of noncoding RNA species, and genetic polymorphisms. However, even if RNA sequencing is expected to rapidly become more costeffective, important challenges remain, especially with respect to storage and interpretation of such large dimension data. In addition, digital approaches are currently available, such as the one developed by Nanostring or the nanoliter-scale high-throughput quantitative PCR developed by Fluidigm (10). These platforms can profile hundreds of selected genes with high sensitivity, low cost, and fast turnaround times and therefore fit perfectly into the needs of translational applications (e.g., clinical trials). Whole transcriptome profiling has demonstrated clinical utility, especially in the field of oncology (4,5). The rheumatology community soon thereafter pursued the identification of biomarkers for diagnosis, assessment of disease activity, prediction of flares, and/or response to treatment in different diseases. Gene expression profiling was thus performed using various pathologic tissues: synovia in rheumatoid arthritis (RA) (11), kidney or skin in SLE (12), salivary glands in Sjo ¨gren’s syndrome (13), muscle in autoimmune myositis (14), and the like. These studies provided interesting clues to the pathogenesis of these complex conditions. However, they posed more challenges than studies performed with cancer tissues. First, inflammatory infiltrates are quite heterogeneous in cellular composition compared to the more “clonal” tumors. Also, tissue sampling is invasive, which limits access to samples and hampers longitudinal monitoring in active and inactive phases of diseases or after initiation of treatment. In this regard, blood transcriptomics appears well suited to monitor and predict disease progression in the context of large-scale longitudinal studies (6). Even though a concern has been that the blood may not totally reflect the pathologic process occurring in target organs, this approach has proven valid in systemic diseases such as SLE, RA, or spondyloarthropathy (1,2,11,12,15) as well as in more “organ-specific” autoimmune diseases such as multiple sclerosis (16,17). Both whole blood and peripheral blood mononuclear cells (PBMCs) have

1409

been used for transcriptome analysis (Figure 1B), but use of whole blood has become the method of choice for large-scale studies. First, processing is straightforward, reducing cost and technical variability. Moreover, the recent availability of dedicated tubes (PAXgene, Tempus) containing a solution that immediately lyses red blood cells has allowed RNA stabilization at the bedside from small blood volumes, which is particularly valuable in children. In addition, whole blood provides global access to the transcriptomic information carried by all blood cells, including granulocytes that are important components of the inflammatory response (see below). Here, instead of an exhaustive review of the field to date (1,2), we aim to provide a few examples that best illustrate the potential use of transcriptome profiling technology in the field of rheumatology. This will serve as a basis for the discussion of existing gaps and potential opportunities in the field. Profiling the blood of patients with rheumatic diseases Blood transcriptome profiling of SLE and other rheumatic conditions. The usefulness of blood transcriptomic analysis to identify potential pathogenic pathways underlying human rheumatic conditions is well illustrated by SLE studies. Indeed, several gene expression studies of peripheral blood cells identified an increased abundance of IFN-inducible genes or “type I interferon signature” in most SLE patients (18–27) (Table 1). Before the use of microarray, traditional genetic approaches had not identified candidate lupus susceptibility genes within the IFN pathway, and type I IFN production was not present in most murine SLE models. Indeed, even though the serum of SLE patients had been described to exert an “interferogenic activity” (28,29), not every SLE patient had detectable serum levels of type I IFN. In 2003, 2 independent teams identified a strong IFN signature in pediatric and adult SLE patients (18,19), and this was later confirmed by additional studies using microarrays and/or reverse transcriptase–PCR (Table 1). Subsequent genetic studies identified strong associations between SLE and type I IFN–related gene polymorphisms (i.e., IRF5, STAT4, TREX1) (30), and proteomics studies showed that the levels of a few IFN-regulated chemokines could be predictive of flares in SLE patients with quiescent disease (31). An IFN signature was later reported in blood, muscle, and skin from patients with autoimmune myositis (14,27,32) as well as in blood and salivary gland tissue from patients with Sjo ¨gren’s syndrome (13) and in

48/42

77/28

48/48

Cross-sectional

Kirou et al, Cross-sectional 2005 (20)

Feng et al, Cross-sectional 2006 (21)

Adult

Adult

Yao et al, Cross-sectional 95/24, 2009 (26) ⫹ longitudinal L: 27

Higgs et al, Cross-sectional 2011 (27)

Adult

Adult

Yes

No (all had active disease)

Yes

Yes

MA ⫹ PCR

Pulse GCs (n ⫽ 2), IS 15%, AM 24%# Pulse GCs (n ⫽ 8), MA ⫹ AM 63%, IS 38% PCR

MA

AM 85%, IS 59%

AM 68%, IS 39%

PCR

PCR ⫹ focus MA§ Pulse GCs (n ⫽ 11), MA AM 47%, IS 25%

AM 66%, IS 43%

–

Whole 22 DM, blood 20 PM, 28 SSc, 89 RA

Whole NA blood

Whole NA blood

–

–

–

PBMCs 199 with 7 – non-SLE diseases Whole NA Nephritis blood

Whole NA blood

Nephritis

Whole 14 RA, 8 Nephritis blood vasculitis

PCR

SLEDAI

NA

NA

NA

NA

NA

NA

NA

NA

Effect of pulse GCs

SLEDAI-2K, Higher NA anti-DNA, no risk of longitudinal flare correlation SELENANo higher NA SLEDAI, no risk of longitudinal flare correlation NA NA Dose effect of antiIFN␣ mAb SLEDAI (not NA NA BILAG index), compl

Anti-RBP and anti-DNA, compl, SLEDAI-2K‡ SELENASLEDAI, antiDNA, compl SLEDAI-2K, anti-DNA, compl SLEDAI

Effect of pulse GCs NA

5 (IFN score)

21

3 (IFN score)

5 (IFN score)

NA¶

31

5 (IFN score)

3 (IFN score)

14 (IFN score)

8

3

15

2

5

NA

15

5

2

7

7

0

2

0

0

NA

10

0

1

2

1

0

0

0

0

NA

3

0

0

0

0

No. of No. of No. of No. of genes in genes in genes in genes in Treatment score or module module module Activity measure Prognosis monitoring signature 1.2 3.4 5.12

Nephritis, CNS, NA, no. of SLE NA hematologic† criteria

–

Pulse GCs (n ⫽ 1), AM 50%

PBMCs NA

PBMCs 12 JCA

Severity

PBMCs 20 RA, 2 uveitis

MA

MA

Technology Samples

Pulse GCs (n ⫽ 4), PCR AM 45%, IS 39%

AM 52%, IS 29%

Pulse GCs (n ⫽ 9), IS 33%

Medications

Other diseases

* Characteristics of 10 representative gene expression studies conducted in systemic lupus erythematosus (SLE) patients between 2003 and 2011 are presented, including the number of genes used as an interferon (IFN) signature or score and the number of these genes belonging to the modular framework developed by Chaussabel et al (23) (see Figure 3). GCs ⫽ glucocorticoids; IS ⫽ immunosuppressants; MA ⫽ microarrays; PBMCs ⫽ peripheral blood mononuclear cells; JCA ⫽ juvenile chronic arthritis; SLEDAI ⫽ SLE Disease Activity Index; NA ⫽ not available; AM ⫽ antimalarials; CNS ⫽ central nervous system; PCR ⫽ polymerase chain reaction; RA ⫽ rheumatoid arthritis; anti-RBP ⫽ anti–RNA binding protein; compl ⫽ complement consumption; SLEDAI-2K ⫽ SLEDAI 2000; SELENA–SLEDAI ⫽ Safety of Estrogens in Lupus Erythematosus National Assessment version of the SLEDAI; L ⫽ longitudinal; mAb ⫽ monoclonal antibody; DM ⫽ dermatomyositis; PM ⫽ polymyositis; SSc ⫽ systemic sclerosis; BILAG ⫽ British Isles Lupus Assessment Group. † Based on historical (not currently active) clinical data. ‡ One SLE patient was followed up longitudinally and her IFN score paralleled disease activity. § The focus array comprised 423 oligonucleotide probes, representing 329 unique genes, selected from differentially expressed genes in SLE from a previous pan-genome expression study generated on Affymetrix U133 GeneChips and genes previously shown to be important in SLE susceptibility and pathogenesis. ¶ The modular score used corresponded to 11 modules (628 genes), only 1 of which was IFN-related (122 genes). # Exposure to antimalarials was allowed in the study, but exact exposure information was provided for only 41 patients.

262/24

Yes

Yes

No (all had relatively quiescent disease) Yes

Pediatric Yes

Adult

Adult

Adult

Adult

Pediatric Yes

Petri et al, Cross-sectional 66/27, 2009 (25) ⫹ longitudinal L: 11

Nikpour et Cross-sectional 269/0 al, 2008 (22) Chaussabel Cross-sectional 63/12, et al, 2008 ⫹ longitudinal L: 20 (23) Landolt et Cross-sectional 94/11, al, 2009 ⫹ longitudinal L: 27 (24)

30/9

Design

Includes No. patients with SLE active patients/ disease and no. patients with healthy inactive controls Population disease

Blood transcriptome studies in SLE*

Cross-sectional

Bennett et al, 2003 (18) Baechler et al, 2003 (19)

Author, year (ref.)

Table 1.

1410 CHICHE ET AL


PBMCs from a subgroup of patients with RA (33) and systemic sclerosis (34). The immune status of a patient may not be entirely reflected by blood profile obtained at the steady state. Indeed, the ex vivo microarray analysis of PBMCs obtained from patients before and after targeted immune-based interventions has also proven very valuable. Palucka et al showed that treatment with tumor necrosis factor (TNF) antagonists in patients with systemic-onset juvenile idiopathic arthritis (JIA) was associated with increased transcription of IFN-regulated genes in blood leukocytes (35). This simple study design unequivocally demonstrated the existence of crossregulatory mechanisms between TNF and type I IFN in human subjects in vivo. It also provided an explanation for the increased anti–double-stranded DNA (antidsDNA) antibody titers and reversible lupus-like syndrome observed in arthritis patients treated with TNF antagonists, and it illustrated how microarray could produce immunologic insight into patients undergoing therapeutic intervention. Subsequently, others demonstrated that TNF blockade in RA patients modulated the expression of IFN response gene activity, but in a heterogeneous manner (36). Indeed, in some RA patients, the treatment induced expression of type I IFN response genes, whereas in others, no effect or a small decrease was observed. Interestingly, RA patients who exhibited increased IFN-related gene expression after 1–2 months of antiTNF treatment showed poor clinical response at 16 weeks. Monitoring the IFN signature after TNF blockade might therefore serve as a therapeutic response–predictive biomarker. Because many promising clinical trials in rheumatic diseases failed to show treatment efficacy despite the existence of a subset of responder patients, the development of such predictive biomarkers has become a priority and should encourage the integration of transcriptomic data in the early steps of drug development and in clinical trial patient stratification. The potential role of IFN in the pathogenesis of SLE (see above) has led several pharmaceutical companies to develop drugs that target the type I IFN pathway. Thus, taking advantage of samples collected during a phase I trial evaluating various doses of an anti-IFN␣ monoclonal antibody, Yao et al studied how IFN signature could be used as a potential pharmacodynamic marker to monitor the efficacy of these drugs (i.e., evaluate inhibition of the molecular target) and to identify potential responders (26). A complex approach, combining data obtained from healthy donor whole blood samples stimulated ex vivo with each of 10 IFN␣

1411

subtypes or IFN␤, whole blood from SLE patients, and healthy donor PBMCs exposed ex vivo to SLE patient sera and inhibited by sifalimumab in a dose-dependent manner (Figures 1C and D), was used to obtain a panel of candidate biomarkers (containing also a few genes chosen from the literature). A specific and dose-dependent inhibition of overexpression of the type I IFN signature was observed in whole blood from treated SLE patients; unfortunately, however, no information on correlation with disease activity parameters was provided in this report and data were not made publicly available. The same blood IFN panel was recently shown to be correlated with disease activity in longitudinal followup of individual patients with autoimmune myositis (37). Signature deconvolution—profiling of purified leukocyte populations and other methods. Whole blood or PBMC microarray data interpretation may be complicated by 2 factors: first, changes in transcript abundance can be attributed to either transcriptional regulation and/or relative changes in cell abundance; second, transcriptional changes from a small proportion of cells can be “masked” by signals coming from more abundant circulating cells or from cells that have migrated from blood to tissues at the time of sampling. To determine if a change in transcript abundance can be attributed to transcriptional regulation or to relative changes in composition of leukocyte populations, 2 approaches have been used. The first relies on the sorting of different cell populations present in the blood that are then separately profiled (Figure 1B). In a recent study, McKinney et al used this approach to show that transcriptional profiling of purified CD8⫹ T cells (and not whole blood) allowed identification of 2 distinct subsets of patients related to long-term prognosis, both in antineutrophil cytoplasmic antibody–associated vasculitis and in SLE (38). Interestingly, these “prognostic” subsets were also identified among the healthy population, and measuring expression levels of only 3 genes (including protein tyrosine phosphatase N22 [PTPN22]) was sufficient to classify patients (38). However, isolation methods require extensive sample processing that may introduce technical bias (transcriptional changes, contamination by other cell populations) that may preclude translational applications. The second approach consists of deconvoluting whole blood transcriptional profiles “in silico.” Cellular composition or cell-specific levels of gene expression are deduced using statistical methods. This approach, which requires prior knowledge of the expression signature of each immune cell subset, has been notably used in SLE and probably merits further attention (39,40). The inter-

1412

pretation of the results coming from such an analysis tends to be more cumbersome intellectually, and the sensitivity of the blood transcriptome profiling assay remains a limiting factor. Reverse proteomic approach. Measuring global transcriptional responses in healthy donor PBMCs exposed to patient plasma or serum in vitro can represent an effective and unbiased means to identify inflammatory mediators and/or therapeutic targets (Figure 1C). Allantaz et al successfully used this “reverse proteomics” approach to investigate the pathogenesis of pediatric rheumatic diseases (41). They used microarrays to reveal that healthy control blood cells treated with sera from patients with systemic-onset JIA exhibited increased expression of genes belonging to the interleukin-1 (IL-1) pathway. Notably, these results served as a basis for the successful use of recombinant IL-1 receptor antagonist (anakinra) in several patients with systemic-onset JIA that was refractory to conventional therapy. Profiling the plasma proteome remains a daunting technical challenge (42), and this work demonstrates the value of utilizing blood leukocytes as sensors to detect and characterize immunogenic activity of patient plasma on a systems scale. Obstacles to translation. In addition to the identification of disease signatures that may improve our understanding of the underlying pathogenesis and identify potential therapeutic targets in autoimmune diseases, there is a clear need for biomarkers in this field. Indeed, in the very illustrative case of SLE, it is still not clear how this IFN signature relates to disease activity and/or outcomes, and relatively little progress has been made to translate the amount of data generated so far into new routine tests. Although an IFN signature is not specific to SLE, it could be used in combination with other criteria for the early diagnosis of SLE, especially in patients with no anti-DNA antibodies and/or with incomplete SLE (43). Second, gene expression biomarkers could be used to perform objective and easy assessment of disease activity and/or identify patients who are at risk of future flares (prognostic biomarkers). The current activity scores used in SLE rely on combined measurement of multiple clinical and laboratory variables that are not easy to implement in routine practice and/or include subjective items. Anti-dsDNA antibody titers are considered a marker of disease activity, but these antibodies are not present in all patients and titers do not always reflect clinical activity. Several cross-sectional studies have demonstrated that peripheral blood gene expression patterns differ between SLE patients with active and

CHICHE ET AL

those with inactive disease (Table 1), but the few longitudinal studies on IFN signature in SLE did not confirm the correlation of this signature, as an isolated marker, with disease activity (24,25). When combined with other signatures, however, the IFN signature correlated with validated measurements of disease activity in pediatric SLE patients (23). Finally, predictive biomarkers to identify patients who are most likely to benefit from a particular treatment (and/or to follow the response to this specific treatment) are needed. There are many reasons why biomarker discovery using transcriptomic analysis to date has not translated into clinical tests as well as hoped in diseases displaying exuberant signatures such as SLE. Disease heterogeneity is one of them. Treatments and comorbidities can also affect transcriptional signatures. Working with pediatric populations (with fewer comorbidities and unre-

Figure 2. Blood transcriptome and biomarker discovery in systemic lupus erythematosus (SLE). The horizontal dashed line corresponds to a cutoff above which disease is considered clinically active. In longitudinal studies, considering all available samples (i.e., from patients with active and inactive disease) together instead of focusing on samples from patients with quiescent disease could lead to risking the loss of relevant correlation. The sample corresponding to the clinically overt flare (S4) may be under the influence of therapies already initiated that could modify signatures. Ideally, blood samples should be obtained systematically and with similar and short intervals to devise, for instance, a prognosis score. Indeed, if sample S3 is not available, rapid molecular changes occurring just before the occurrence of the clinical flare may be missed. SLEDAI ⫽ SLE Disease Activity Index. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131.


lated medications) and/or newly diagnosed untreated patients is part of the solution, but findings in pediatric studies may not be fully representative of adult SLE (18). Biomarker discovery is challenging in many ways. Large patient cohorts and longitudinal sampling are needed to validate novel biomarkers (3). In longitudinal studies especially, identification of predictive biomarkers should rely on the analysis of samples obtained when the disease is considered “clinically” quiescent, instead of pooling these samples with those obtained at the time of clinically overt flare and risking the loss of relevant correlation (Figure 2). Indeed, samples corresponding to overt flares may reflect the effect of therapies already initiated and/or important modifications of cellular composition due to the migration of immune cells from blood to inflamed tissues (i.e., silencing of IFN signature of SLE with intravenous pulse steroids). Ideally, only blood samples obtained systematically (and at regular and short intervals) before a flare might be informative to devise, for instance, a prognostic IFN score (Figure 2). Finally, as blood profiling may miss important markers due to the complex and variable cellular composition across time, other complementary experimental designs may be considered (Figure 1).

1413

New challenges, new perspectives—systems rheumatology Molecular stratification of patients. Rheumatic diseases are clinically heterogeneous, most likely as a reflection of underlying biologic heterogeneity. Considering patient populations in terms of group averages may therefore prevent us from identifying pathologic processes and/or establishing the efficacy of novel treatments that might benefit only a patient subset. Analysis at the individual level allows unbiased stratification of patients (molecular stratification). In that setting, data dimension reduction can facilitate both visualization and functional interpretation of large-scale data sets. We have developed a data mining strategy for the specific purpose of analyzing blood transcriptional profiles. This approach consists of grouping genes with similar transcriptional patterns across multiple data sets, building a coclustering network (23). These gene sets, identified through an entirely data-driven process, are called “modules.” This modular transcriptional framework (or modular repertoire) allows reducing the number of variables by collapsing sets of coordinately expressed genes into a new transcriptional entity, the

Figure 3. Blood transcriptome and modular framework in systemic lupus erythematosus (SLE). Using the second generation of the modular framework, a disease fingerprint for SLE is provided (top left). Differences in expression levels between study groups (SLE patients versus healthy controls matched for age, sex, and ethnicity) are displayed for each module on a grid. Each position on the grid is assigned to a given module; a red spot indicates an increase in transcript abundance (overexpressed [over-xp]) and a blue spot indicates a decrease (underexpressed [under-xp]). The spot intensity is determined by the proportion of transcripts reaching significance for a given module. A posteriori, biologic interpretation has linked several modules to immune cells or pathways. For SLE patients, the fingerprint includes a strongly up-regulated interferon modular signature (M1.2, M3.4, and M5.12) (top right). NK ⫽ natural killer.

1414

CHICHE ET AL

Figure 4. Individual blood transcriptome fingerprints in systemic lupus erythematosus (SLE). The modular framework (see Figure 3) can be used to visualize individual fingerprints. As an example, the individual modular fingerprint of adult SLE patients with longitudinal followup (n ⫽ 29) is provided as a circus plot and highlights a previously underestimated heterogeneity within the interferon signature observed at the individual level (the following modules are represented from the outside to the inside on the circus plot: M5.12, M3.4, M1.2, M4.10, M4.15, M4.1, M3.6, M4.11, M5.15).

module. A visualization scheme for mapping global transcriptional changes for individual diseases on a modular basis has been developed (Figure 3). This modular framework can be used to visualize individual fingerprints. As an example, in SLE, investigators in recent studies using few IFN-related genes have come to the conclusion that expression of IFN-responsive genes is not a dynamic component of the SLE disease process, but rather a relatively stable characteristic of the individual and may reflect an inherent activation state of the IFN pathway (24,25). When using the second generation of our modular framework (which includes 3 IFNrelated modules), we observe heterogeneity within the IFN modular signature observed at the individual level (Figure 4), with some of the corresponding modules exhibiting more variability across time than others. This emphasizes the interest of unbiased data-driven approaches such as module repertoire analysis in supporting biomarker discovery. Data sharing and meta-analysis. Combining analyses from different studies (or meta-analysis) could be one of the short-term solutions to increase power and allow cross-validation of known signatures. However, several challenges need to be considered when integrat-

ing different microarray data sets for meta-analysis. Until recently, such approaches have been performed at the gene level, identifying commonly expressed genes between studies, with limitations due to different probe design between different platforms and the presence of batch effects. More recently, Arasappan et al reported a microarray meta-analysis based on the identification of pathways coordinately expressed in multiple data sets of SLE (44). They used as input blood gene expression profiles from 4 different studies including SLE patients. Once differentially expressed genes were identified for each data set, they used IPA to identify biologic pathways that were differentially expressed between SLE patients and controls and obtained 3 main biologic pathways that were consistently enriched in SLE patients, as well as a metasignature of 37 genes. This interesting approach warrants several comments. First, although the identification of signatures that are common to the different data sets of the same disease (SLE) increases the robustness of these metasignatures, this approach at the data set level is likely to miss other interesting signatures that would be present, for example, only in a subset of patients from one or more data sets. Indeed, the type I IFN signature repre-


sents only part of the transcriptional alterations observed in SLE (Figures 3 and 4). Genes expressed by granulocytes were also identified as up-regulated in some SLE patients (18,19,23,26). While it prompted the investigation of the role of neutrophils in the pathogenesis of SLE (45), this “neutrophil signature” was not retained in the meta-analysis. Such meta-analysis approaches will benefit from the increasing number of publicly available data sets on GEO (46) or other repositories, as long as efforts are made to properly annotate samples with demographic and clinical data. Finally, full access to various data sets should also favor gene expression comparative analysis between different autoimmune diseases as well as between rheumatic autoimmune conditions and nonautoimmune conditions (i.e., infections) (47). This comparison “across diseases” may allow the identification of overlap and specificity (unique signatures) and promote understanding of disease pathogenesis. There is an urgent need for tools allowing the integration of transcriptomic profiles with both laboratory (cytometric, proteomic, genomic) and clinical data. Notably, these tools should allow the exploration of the new meta(or aggregated) data sets obtained, especially at the individual level, either across various experiments (or assays) within the same data set (“condition-specific” discovery) and/or across various data sets (“acrosscondition” discovery). Systems transition—perspectives. Biomarkers derived from systems biology approaches could be particularly useful at the early stages of development of new targeted therapy to stratify patients and identify subgroups of patients likely to show the greatest clinical benefit. Large-scale gene expression profiling studies have been key to the translational success in the field of oncology (4,5). However, in the field of rheumatology, collaborative efforts between groups to increase sample size are still needed. Ideally, any therapeutic clinical trial in the field should include transcriptomic analysis. This was recently done in a few small studies evaluating anti-IFN␣ treatment in SLE patients (26) or IL-1 receptor antagonist treatment in patients with systemic-onset JIA (48). Performing such ancillary gene expression studies also has the advantage of providing longitudinal data in clinically well-characterized patients. Indeed, longitudinal studies of patients who experience changes in disease activity provide a more valid design than crosssectional comparisons, and fewer than half of recent studies in the field include such longitudinal designs (3). Interpretation of longitudinal microarray data is a com-

1415

plex statistical task that has been very recently developed for acute inflammatory conditions like trauma or sepsis and that could be applied to rheumatic diseases (49). However, limiting such collaborative approaches to clinical trials may be insufficient because of the difficulties in establishing such large trials and because patients enrolled in such trials may not be representative of every presentation and/or therapeutic management of heterogeneous diseases such as SLE. Therefore, an international public–private consortium dedicated to biomarker discovery in rheumatic diseases, based on the model of what has been done in the field of oncology, may be the ideal way to achieve the needed systems transition. Conclusions Systems biology approaches, especially blood transcriptome analysis, have been widely applied to various rheumatic diseases. Microarray technology has yielded insights into the pathogenesis of complex diseases such as SLE and identified potential therapeutic targets where other classic approaches had failed. In addition to the blood profiling approach, the study of specific cell subsets or of the transcriptional effects of patient serum factors on healthy blood cells has also been shown to be helpful for dissecting the pathogenesis of these diseases. Systems approaches have been shown to be well suited for investigating molecular heterogeneity, a fundamental aspect of most chronic inflammatory diseases. Efforts to develop the appropriate methodology for mining the large amounts of data already generated, as well as purposely designed large-scale profiling studies, should soon be leveraged for better translation to new biomarker discovery. Aside from ongoing technological improvements, large-scale collaborative studies to permit validation of data in multicenter settings seem to be pivotal for accelerating biomarker discovery in rheumatic diseases. ACKNOWLEDGMENTS We thank Anna Bjork, Kristen Dang, Mary Roy, and Elizabeth Whalen for their help with editing the text, figures, and table. We also thank the patients and physicians involved in the LU-PUCE cohort study funded by Appel d’Offres de Recherche Clinique–Assistance Publique Ho ˆpitaux de Marseille, the Association pour le De´veloppement des Recherches Biologiques et Me´dicales, and the Centre de Recherche en Ne´phrologie, Ho ˆpital de la Conception, Marseille. AUTHOR CONTRIBUTIONS All authors drafted the article, revised it critically for important intellectual content, and approved the final version to be published.

1416

CHICHE ET AL

REFERENCES 1. Bauer JW, Bilgic H, Baechler EC. Gene-expression profiling in rheumatic disease: tools and therapeutic potential. Nat Rev Rheumatol 2009;5:257–65. 2. Pascual V, Chaussabel D, Banchereau J. A genomic approach to human autoimmune diseases. Annu Rev Immunol 2010;28: 535–71. 3. Tektonidou MG, Ward MM. Validity of clinical associations of biomarkers in translational research studies: the case of systemic autoimmune diseases. Arthritis Res Ther 2010;12:R179. 4. Hummel M, Bentink S, Berger H, Klapper W, Wessendorf S, Barth TF, et al, for the Molecular Mechanisms in Malignant Lymphomas Network Project of the Deutsche Krebshilfe. A biologic definition of Burkitt’s lymphoma from transcriptional and genomic profiling. N Engl J Med 2006;354:2419–30. 5. Van de Vijver MJ, He YD, van ’t Veer LJ, Dai H, Hart AA, Voskuil DW, et al. A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 2002;347:1999–2009. 6. Pham MX, Teuteberg JJ, Kfoury AG, Starling RC, Deng MC, Cappola TP, et al, for the IMAGE Study Group. Gene-expression profiling for rejection surveillance after cardiac transplantation. N Engl J Med 2010;362:1890–900. 7. MAQC Consortium, Shi L, Reid LH, Jones WD, Shippy R, Warrington JA, Baker SC, et al. The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements. Nat Biotechnol 2006;24:1151–61. 8. Miame Brazma A, Hingamp P, Quackenbush J, Sherlock G, Spellman P, Stoeckert C, et al. Minimum information about a microarray experiment (MIAME)—toward standards for microarray data. Nat Genet 2001;29:365–71. 9. Malone JH, Oliver B. Microarrays, deep sequencing and the true measure of the transcriptome. BMC Biol 2011;9:34. 10. Jang JS, Simon VA, Feddersen RM, Rakhshan F, Schultz DA, Zschunke MA, et al. Quantitative miRNA expression analysis using fluidigm microfluidics dynamic arrays. BMC Genomics 2011;12:144. 11. Timmer TC, Baltus B, Vondenhoff M, Huizinga TW, Tak PP, Verweij CL, et al. Inflammation and ectopic lymphoid structures in rheumatoid arthritis synovial tissues dissected by genomics technology: identification of the interleukin-7 signaling pathway in tissues with lymphoid neogenesis. Arthritis Rheum 2007;56: 2492–502. 12. Berthier CC, Bethunaickan R, Gonzalez-Rivera T, Nair V, Ramanujam M, Zhang W, et al. Cross-species transcriptional network analysis defines shared inflammatory responses in murine and human lupus nephritis. J Immunol 2012;189:988–1001. 13. Emamian ES, Leon JM, Lessard CJ, Grandits M, Baechler EC, Gaffney PM, et al. Peripheral blood gene expression profiling in Sjo ¨gren’s syndrome. Genes Immun 2009;10:285–96. 14. Wong D, Kea B, Pesich R, Higgs BW, Zhu W, Brown P, et al. Interferon and biologic signatures in dermatomyositis skin: specificity and heterogeneity across diseases. PLoS One 2012;7:e29161. 15. Sharma SM, Choi D, Planck SR, Harrington CA, Austin CR, Lewis JA, et al. Insights in to the pathogenesis of axial spondyloarthropathy based on gene expression profiles. Arthritis Res Ther 2009;11:R168. 16. Hecker M, Paap BK, Goertsches RH, Kandulski O, Fatum C, Koczan D, et al. Reassessment of blood gene expression markers for the prognosis of relapsing-remitting multiple sclerosis. PLoS One 2011;6:e29648. 17. Ottoboni L, Keenan BT, Tamayo P, Kuchroo M, Mesirov JP, Buckle GJ, et al. An RNA profile identifies two subsets of multiple sclerosis patients differing in disease activity. Sci Transl Med 2012;4:153ra131. 18. Bennett L, Palucka AK, Arce E, Cantrell V, Borvak J, Banchereau

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

J, et al. Interferon and granulopoiesis signatures in systemic lupus erythematosus blood. J Exp Med 2003;197:711–23. Baechler EC, Batliwalla FM, Karypis G, Gaffney PM, Ortmann WA, Espe KJ, et al. Interferon-inducible gene expression signature in peripheral blood cells of patients with severe lupus. Proc Natl Acad Sci U S A 2003;100:2610–5. Kirou KA, Lee C, George S, Louca K, Peterson MG, Crow MK. Activation of the interferon-␣ pathway identifies a subgroup of systemic lupus erythematosus patients with distinct serologic features and active disease. Arthritis Rheum 2005;52:1491–503. Feng X, Wu H, Grossman JM, Hanvivadhanakul P, FitzGerald JD, Park GS, et al. Association of increased interferon-inducible gene expression with disease activity and lupus nephritis in patients with systemic lupus erythematosus. Arthritis Rheum 2006;54:2951–62. Nikpour M, Dempsey AA, Urowitz MB, Gladman DD, Barnes DA. Association of a gene expression profile from whole blood with disease activity in systemic lupus erythaematosus. Ann Rheum Dis 2008;67:1069–75. Chaussabel D, Quinn C, Shen J, Patel P, Glaser C, Baldwin N, et al. A modular analysis framework for blood genomics studies: application to systemic lupus erythematosus. Immunity 2008;29: 150–64. Landolt-Marticorena C, Bonventi G, Lubovich A, Ferguson C, Unnithan T, Su J, et al. Lack of association between the interferon-␣ signature and longitudinal changes in disease activity in systemic lupus erythematosus. Ann Rheum Dis 2009;68:1440–6. Petri M, Singh S, Tesfasyone H, Dedrick R, Fry K, Lal P, et al. Longitudinal expression of type I interferon responsive genes in systemic lupus erythematosus. Lupus 2009;18:980–9. Yao Y, Richman L, Higgs BW, Morehouse CA, de los Reyes M, Brohawn P, et al. Neutralization of interferon-␣/␤–inducible genes and downstream effect in a phase I trial of an anti–interferon-␣ monoclonal antibody in systemic lupus erythematosus. Arthritis Rheum 2009;60:1785–96. Higgs BW, Liu Z, White B, Zhu W, White WI, Morehouse C, et al. Patients with systemic lupus erythematosus, myositis, rheumatoid arthritis and scleroderma share activation of a common type I interferon pathway. Ann Rheum Dis 2011;70:2029–36. Vallin H, Blomberg S, Alm GV, Cederblad B, Ronnblom L. Patients with systemic lupus erythematosus (SLE) have a circulating inducer of interferon-alpha (IFN-␣) production acting on leucocytes resembling immature dendritic cells. Clin Exp Immunol 1999;115:196–202. Blanco P, Palucka AK, Gill M, Pascual V, Banchereau J. Induction of dendritic cell differentiation by IFN-␣ in systemic lupus erythematosus. Science 2001;294:1540–3. Lee-Kirsch MA, Gong M, Chowdhury D, Senenko L, Engel K, Lee YA, et al. Mutations in the gene encoding the 3⬘-5⬘ DNA exonuclease TREX1 are associated with systemic lupus erythematosus. Nat Genet 2007;39:1065–7. Bauer JW, Petri M, Batliwalla FM, Koeuth T, Wilson J, Slattery C, et al. Interferon-regulated chemokines as biomarkers of systemic lupus erythematosus disease activity: a validation study. Arthritis Rheum 2009;60:3098–107. Baechler EC, Bauer JW, Slattery CA, Ortmann WA, Espe KJ, Novitzke J, et al. An interferon signature in the peripheral blood of dermatomyositis patients is associated with disease activity. Mol Med 2007;13:59–68. Reynier F, Petit F, Paye M, Turrel-Davin F, Imbert PE, Hot A, et al. Importance of correlation between gene expression levels: application to the type I interferon signature in rheumatoid arthritis. PLoS One 2011;6:e24828. Assassi S, Mayes MD, Arnett FC, Gourh P, Agarwal SK, McNearney TA, et al. Systemic sclerosis and lupus: points in an interferonmediated continuum. Arthritis Rheum 2010;62:589–98. Palucka AK, Blanck JP, Bennett L, Pascual V, Banchereau J.


36.

37.

38. 39.

40. 41.

42.

Cross-regulation of TNF and IFN-␣ in autoimmune diseases. Proc Natl Acad Sci U S A 2005;102:3372–7. Van Baarsen LG, Wijbrandts CA, Rustenburg F, Cantaert T, van der Pouw Kraan TC, Baeten DL, et al. Regulation of IFN response gene activity during infliximab treatment in rheumatoid arthritis is associated with clinical response to treatment. Arthritis Res Ther 2010;12:R11. Greenberg SA, Higgs BW, Morehouse C, Walsh RJ, Won Kong S, Brohawn P, et al. Relationship between disease activity and type 1 interferon- and other cytokine-inducible gene expression in blood in dermatomyositis and polymyositis. Genes Immun 2012;13: 207–13. McKinney EF, Lyons PA, Carr EJ, Hollis JL, Jayne DR, Willcocks LC, et al. A CD8⫹ T cell transcription signature predicts prognosis in autoimmune disease. Nat Med 2010;16:586–91. Abbas AR, Wolslegel K, Seshasayee D, Modrusan Z, Clark HF. Deconvolution of blood microarray data identifies cellular activation patterns in systemic lupus erythematosus. PLoS One 2009;4: e6098. Shen-Orr SS, Tibshirani R, Khatri P, Bodian DL, Staedtler F, Perry NM, et al. Cell type-specific gene expression differences in complex tissues. Nat Methods 2010;7:287–9. Allantaz F, Chaussabel D, Stichweh D, Bennett L, Allman W, Mejias A, et al. Blood leukocyte microarrays to diagnose systemic onset juvenile idiopathic arthritis and follow the response to IL-1 blockade. J Exp Med 2007;204:2131–44. Tu C, Rudnick PA, Martinez MY, Cheek KL, Stein SE, Slebos RJ, et al. Depletion of abundant plasma proteins and limitations of plasma proteomics. J Proteome Res 2010;9:4982–91.

1417

43. Li QZ, Zhou J, Lian Y, Zhang B, Branch VK, Carr-Johnson F, et al. Interferon signature gene expression is correlated with autoantibody profiles in patients with incomplete lupus syndromes. Clin Exp Immunol 2010;159:281–91. 44. Arasappan D, Tong W, Mummaneni P, Fang H, Amur S. Metaanalysis of microarray data using a pathway-based approach identifies a 37-gene expression signature for systemic lupus erythematosus in human peripheral blood mononuclear cells. BMC Med 2011;9:65. 45. Garcia-Romo GS, Caielli S, Vega B, Connolly J, Allantaz F, Xu Z, et al. Netting neutrophils are major inducers of type I IFN production in pediatric systemic lupus erythematosus. Sci Transl Med 2011;3:73ra20. 46. Barrett T, Edgar R. Gene expression omnibus: microarray data storage, submission, retrieval, and analysis. Methods Enzymol 2006;411:352–69. 47. Nakaya HI, Gardner J, Poo YS, Major L, Pulendran B, Suhrbier A. Gene profiling of Chikungunya virus arthritis in a mouse model reveals significant overlap with rheumatoid arthritis. Arthritis Rheum 2012;64:3553–63. 48. Quartier P, Allantaz F, Cimaz R, Pillet P, Messiaen C, Bardin C, et al. A multicentre, randomised, double-blind, placebo-controlled trial with the interleukin-1 receptor antagonist anakinra in patients with systemic-onset juvenile idiopathic arthritis (ANAJIS trial). Ann Rheum Dis 2011;70:747–54. 49. Xiao W, Mindrinos MN, Seok J, Cuschieri J, Cuenca AG, Gao H, et al. A genomic storm in critically injured humans. J Exp Med 2011;208:2581–90.