Article Text

Download PDFPDF

Does gene expression analysis inform us in rheumatoid arthritis?
  1. T Häupl1,
  2. B Stuhlmüller1,
  3. A Grützkau2,
  4. A Radbruch2,
  5. G-R Burmester1
  1. 1
    Department of Rheumatology and Clinical Immunology, Charité University Medicine, Berlin, Germany
  2. 2
    German Arthritis Research Center, Berlin, Germany
  1. Correspondence to Dr G-R Burmester, Department of Rheumatology and Clinical Immunology, Charité University Medicine Berlin, Charitéplatz 1, 10117 Berlin, Germany; gerd.burmester{at}


Transcription profiling has become a standard technology in research. It is mainly applied in the search for biomarkers to improve diagnostic and prognostic classification, to quantify disease activity and to predict or indicate response to therapy. This review will focus on rheumatoid arthritis and discuss considerations for sample selection, prerequisites for functional interpretation of data and the current status of information deduced in the field of biomarkers for the various clinical questions. In the next few years, prediction of response to treatment is the most important aim of biomarker research. With the growing number of new biological agents, there is increasing pressure to identify molecular parameters that will not only guide the therapeutic decision but also help to define the most important targets for which new biological agents should be tested in clinical studies.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Microarrays for gene expression analysis are the most powerful technologies developed in recent years to generate a comprehensive molecular image of a biological system. Depending on the type of sample this image consists of the molecular profile of a single cell type or a mixture of several different cell populations. Furthermore, molecular processes of housekeeping, cell-specific functions and gene activation as a response to triggers may all influence and contribute to the expression signal of a gene.

This technology has been applied in the search for diagnostic patterns by comparing gene expression in a defined status of a disease with healthy controls, other diseases or another status of the same disease. The differences may reveal markers for diagnostic classification, disease activity, prognosis and response to treatment. Furthermore, transcription profiles may disclose pharmacological mechanisms and molecular side-effects of treatment. The molecular patterns may also contain functional signatures relevant for treatment and the prediction of response before the initiation of therapy.

Considerations for sample selection

The information generated by transcription profiling is depending on the types of samples analysed. Two principal aspects have to be considered. First, is the sample derived from the joint that is the main site of inflammation or is the sample from the blood and reflects systemic activation as part of the disease process. Second, does the sample consist of a mixture of different cell types or is it a purified subpopulation of cells?

All cells involved in the process of joint inflammation and destruction are present and the relevant molecular mechanisms of arthritis are active in this tissue. However, compared with blood, tissue is difficult to access. Furthermore, there are substantial cell-specific differences in gene activity between fibroblasts/macrophages and other infiltrating immune cells.1 Therefore, when comparing synovitis tissue with normal donor tissue the profile mainly reflects the infiltration of immune cells that are absent in normal tissue.

To identify molecular mechanisms of disease, it is important to focus on gene regulation. This is only achievable by comparing each cell type in the inflamed tissue with its normal donor cell type. Microdissection or digestion of the tissue with subsequent cell sorting are potential options to separate the different cell types. According to the authors’ own experience the disadvantage of microdissection is that cells of interest are only enriched. Tissue digestion and subsequent cell sorting reveals much higher purity but depends on the incubation of vital tissue and thus induces artificial gene regulation ex vivo.

In contrast to tissue, blood is a convenient sample. Nevertheless, there are similar aspects to discuss when using whole blood with variable mixtures of cells. There is even more concern that relevant transcripts of rare cell types will escape detection. Cell separation from blood, however, is much easier and can be performed with new technologies that have been shown to avoid artificial gene activation.2

Despite these benefits for accessibility and purification, cells in the blood may only reflect mechanisms of systemic activation. Many molecular processes involved in cell recruitment, amplification and perpetuation of inflammation as well as joint destruction may not be sufficiently displayed. Plasma and endothelium provide a multitude of mechanisms to suppress the uncontrolled exacerbation of inflammatory processes in diseases such as rheumatoid arthritis (RA). Blood may thus only partly reflect the molecular pathology of RA. To demonstrate the quantitative difference between blood and tissue, the number of genes differentially expressed between RA and normal donor samples are displayed in fig 1.

Figure 1

Differential gene expression in rheumatoid arthritis (RA) patients compared with normal donors (ND) in different types of samples. Compared with tissue, the number of probesets with increased expression in RA is much lower in whole blood or different cell types separated from blood. This difference was determined for a fraction of pairwise comparisons with increased expression in RA ranging from 50% to 100%.

Prerequisites for functional analysis

Current knowledge about gene function

Microarray hybridisation can generate huge datasets with expression levels for all currently known genes. To interpret such results and to classify according to function, additional information is needed. Tools have been developed to collate published knowledge about individual genes (eg, EntrezGene,3 GeneCards,4 the Kyoto encyclopedia of genes and genomes (KEGG),5 OMIM6) and to draw networks of molecular interaction (eg, Ingenuity Pathway Analysis, KEGG, PathwayAssist,7 PubGene8). The gene ontology initiative9 provides attributes on the “biological process”, the “cellular component” and the “molecular function” for each gene. All these resources are currently the most important toolboxes for functional interpretation of transcriptome data. Newer developments are stepwise integrating experimental datasets from transcription profiling using different algorithms and diagrams on the expression of each gene in various tissues of the human body.

Functional signatures as references for profile components

There is a crucial discrepancy between data previously generated to describe functional pathways of signalling and the current need to perform functional interpretation of transcriptome data. Unravelling the molecular interaction in signal transduction is an important step to identify genes needed to trigger a pathway process. However, focussing on these genes does not inform us whether the signalling process is active. In fact, the genes of signalling pathways are always transcribed if the cell has to be ready to receive.10 To recognise the full dimension of a signalling process with its effect on all known genes, genome-wide expression profiling will be necessary. In order to extend the current concept of functional analysis based on written information, knowledge from the profiling of defined conditions has to be included. Most relevant are reference signatures from different cell and tissues types as well as from different stimulation experiments.

Databases and new tools for integrative analysis

To generate these reference data for interpretation is one step. The other important step is the development of tools to store this information in a standardised pre-analysed format, to enable immediate access for comparison with new data, and to provide appropriate algorithms for qualitative and quantitative identification of reference profile components in transcriptomes derived from patients. Several databases exist, but do not yet cover all aspects needed. The National Centre for Biotechnology Information’s GEO11 and ArrayExpress12 are designed as repositories of raw data files with experimental information. Celsius13 is an advanced platform but requires much knowledge in bioinformatics and programming. BASE14 or MARS15 are data warehouse concepts open for a wide range of different types of microarrays and storage of comprehensive lists of experimental data but with limited public access. As a result of our own efforts, the SiPaGene database was generated16 and further developed by BioRetis ( providing pre-analysed data, multiple different options to select candidate genes, the generation of subgroup comparisons and multiple different group comparisons also with reference datasets. With a particular administration of rights to access, this database enables us to perform primary analysis and allows us to share data at various levels. Nevertheless, other tools needed for complex analysis are only partly available and often depend on expert knowledge and novel bioinformatic programming algorithms.

Patient selection and clinical parameters

A very important prerequisite is the appropriate selection of patients based on clinical parameters that fit to the question that is raised. Whole genome profiling suggests that the excessive analysis of all possible genes may help to compensate for feasibility considerations that have motivated us to focus on accessible and possibly simplified systems. Similar to the discussed aspects of sample types (whole blood or purified cells), combining samples with a different underlying pathomechanism may have a profound impact on the interpretability of data and might inhibit the identification of useful molecular markers. Investigation of patients with combination therapies is unfocussed if markers for the prediction of response to a defined main drug are requested.

Information generated by gene expression profiling in RA

Synovial tissue

Molecular profiling of RA synovitis has been started by van der Pouw Kraan et al.17 They identified differences between samples with one group indicative of an adaptive immune response and a second group suggestive of fibroblast dedifferentiation. The first group may be further subdivided by genes related to the classic pathway of complement activation. In a subsequent extended study, evidence was demonstrated for differences in the activation of the STAT-1 pathway between subgroups of RA tissues.18 Lindberg et al19 published evidence that substantial sampling effects may be found and that these were related to differences in the level of inflammation. However, appropriate collection from areas representing typical inflammation revealed patient-specific patterns and thus demonstrated the relevance of such information.

Nevertheless, variance in cellular composition seems to influence such profiles in an unforeseeable way. Principal concepts to solve this problem have been suggested and depend on reference signatures for comparison.1 Using targeted biopsies from the lining layer of the inflamed synovium, gene expression profiles also divided samples into two groups.20 Besides many other genes, one group with histologically most prominent lymphoid infiltration and palisading showed again the increased expression of STAT-1 as well as IFNγ receptor 2 and integrin B2. The other group with lower histological scores of inflammation showed increased expression, for example, of caspase 9, p53-induced gene 11 and IL-10 receptor B, suggesting that molecules involved in the control of inflammation are present in this group. Interestingly, the first group included all long-standing arthritides, whereas only early arthritides were in the second group. Most recently, another group summarised their findings in early and long-standing RA synovitis based on gene ontology. According to these observations, T-cell activation, endothelin-signalling pathway, hypoxia response and plasminogen-activating cascade were more prominent in early RA and cell cycle, cell surface receptor-mediated signal transduction, cell-cycle control, ligand-mediated signalling, apoptosis inhibition and granulocyte-mediated immunity in long-standing RA.21 A role for hypoxia-induced mechanisms was also suggested by the authors’ own studies when comparing RA with osteoarthritis synovium, hypoxic control tissue and peripheral blood.22 Huber et al23 also compared RA with osteoarthritis and normal donor synovium applying the functional information stored in KEGG. Compared with normal tissue, RA samples revealed more cytokine–cytokine receptor interactions, activation of the transforming growth factor beta (TGFβ) pathway and anti-apoptotic signalling. Compared with osteoarthritis synovium B-cell receptor and vascular endothelial growth factor signalling were increased in RA. Another aspect was discussed in the authors’ own studies when focussing on growth factors and bone morphogenetic proteins. As triggers of differentiation and repair, the decrease and change in histological distribution suggest that mechanisms of joint homeostasis are impaired,24 and that therapy may exert additional effects on this regulatory system.25

Synovial fibroblasts and infiltrating cells

Several groups have focussed on individual cell types of the inflamed synovial tissue. Pierer et al26 investigated the effect of TLR-2 stimulation in RA and osteoarthritis synovial fibroblasts and found differences in the induction of the chemokines GCP-2 (CXCL6), RANTES (CCL5) and MCP-2 (CCL8). Kasperkovitz et al27 demonstrated that profiles of fibroblast-like synoviocytes derived from patients with RA divide samples into two main groups and thus show the imprint of synovial tissue heterogeneity reported previously. Fibroblasts from highly inflamed synovium revealed increased expression of TGFβ/activin A-inducible gene profiles characteristic of myofibroblasts and involved in wound healing. In contrast, low inflammatory synovium-derived fibroblasts expressed elevated levels of growth factors (insulin-like growth factor 2/insulin-like growth factor binding protein 5). Constitutive upregulation of the TGFβ pathway in RA synovial fibroblasts was also described by Pohlers et al28 and was found to be associated with enhanced expression of matrix metalloproteinase (MMP) 11. Galligan et al29 identified several homeobox genes to be upregulated in RA compared with osteoarthritis synovial fibroblasts and could correlate disease activity with the expression of several genes so far unknown in RA. Timmer et al30 compared different types of tissues with follicular, aggregated or diffuse infiltrates. Clustering of genes revealed for specimens with follicular infiltrates the differential expression of several chemokines (CXCL13, CCL21, CXCL12, CCL19) and associated receptors (CCR7, CXCR4, CXCR5) as well as genes associated with Jak/STAT signalling, T-cell and B-cell-specific pathways and IL-7 signal transduction. Histological analysis identified that IL-7 was predominantly expressed by fibroblast-like synoviocytes, macrophages and in blood vessel endothelial cells. Extracellular IL-7 protein was found only in the area of B-cell follicles and thus may play an important role in lymphoid neogenesis in RA synovial tissue. Finally, Auer et al31 investigated neutrophils purified from the synovial fluid compartment with microarrays and identified these cells as a novel source of the T-cell-attracting chemokine CCL18 in RA joints.


Compared with investigations on synovial and pannus tissue little is known about the transcriptome of cartilage in RA, although it is the target tissue of destruction. Early studies by Elliott et al32 and Vincenti and Brinckerhoff33 investigated the role of IL-1β in a chondrosarcoma cell line. Gebauer et al34 compared transcriptomes of primary chondrocytes with chondrosarcoma cell lines and found only little similarity. Nevertheless, nuclear factor kappa B was a common transcriptional regulator of IL-1β induced genes, which included several MMP and IL-6 in both cell types and LIF and BMP2 only in primary chondrocytes. Many more studies on cartilage transcriptomes were performed in the field of osteoarthritis research.35 With a focus on RA, the authors’ own studies investigated the transcriptome of in-vitro engineered cartilage before and after exposure to the RA synovial fibroblast secretome.36 37 With several chemokines and cytokines released by the RA fibroblasts, this secretome induced in artifical cartilage markers of inflammation (adenosine A2A receptor, cyclooxygenase-2), the nuclear factor kappa B signalling pathway (Toll-like receptor 2, spermine synthase, receptor-interacting serine-threonine kinase 2), cytokines/chemokines and receptors (CXCL1–3, CXCL8, CCL20, CXCR4, IL-1β, IL-6), MMP-10, MMP-12) and suppressed matrix synthesis (cartilage oligomeric matrix protein, chondroitin sulfate proteoglycan 2).

Blood and isolated cells from blood

Maas et al38 compared peripheral blood mononuclear cells (PBMC) from various autoimmune diseases with vaccination-induced changes and reported a molecular pattern common to all autoimmune diseases investigated—RA, systemic lupus erythematosus (SLE), type I diabetes and multiple sclerosis. Subsequently, Olsen et al39 compared PBMC from early with long-standing RA and identified differences that suggested some overlap with the normal viral antigen immune response as well as with a subset of patients with SLE. van der Pouw Kraan et al40 also reported on a RA subgroup that presented with an interferon type I signature in PBMC, a signature that was previously reported to be a typical molecular characteristic in SLE.41 42 They also described a pathogen-response programme in PBMC reminiscent of that of a poxvirus infection in macaques.43 These observations suggest that different molecular mechanisms may be involved in RA, which may explain other reports and observations, most strikingly the differences in treatment response. However, profiling results may be more complex and a recent report by Junta et al44 suggests that immunogenetic, pathogenic and treatment features may also be reflected in PBMC transcriptomes. The authors’ own investigations on RA disease reactivation after pregnancy with PBMC included reference signatures of various cell types, and revealed that increased phagocyte and recurring lymphocyte gene activity were associated with a flare.45 Furthermore, there was a correlation especially with monocyte transcripts and clinical disease activity measures.46 Grützkau et al2 investigated gene expression profiles in monocytes of various autoimmune diseases and could identify discriminative patterns between RA, SLE, spondylarthropathies, osteoarthritis and normal donors. Interestingly, there was again the type I interferon pattern found in SLE monocytes. Compared with RA or spondylarthropathy, the number of type I interferon genes and expression intensity was much higher in SLE, suggesting that more detailed investigations are needed to interpret the various reports on type I interferon signatures. With a focus on lymphoblastoid B-cell lines, Haas et al47 demonstrated differential gene expression in RA disease-discordant monozygotic twins. The more than 700 genes including laeverin, 11beta-hydroxysteroid dehydrogenase type 2 and cysteine-rich, angiogenic inducer 61 were associated with apoptosis, angiogenesis, proteolysis and intracellular signalling (cascades). Investigating isolated B cells, Szodoray et al48 found the regulation of genes related to cell cycle, proliferation, apoptosis, autoimmunity, cytokine networks, angiogenesis and neuro-immune regulation. Moreover, various cytokines involved in B-cell activation and pathway modulation were found to be increased in the serum of these RA patients. Turrel-Davin et al49 investigated whole blood and identified FoxO3a as a prominent candidate overexpressed in RA. They could attribute this increase to the neutrophil population in the blood and T cells in the synovial tissue and suggest implications on cell cycle regulation and apoptosis. Another important compartment, the bone marrow, has been studied by Nakamura et al.50 They identified several candidate genes overexpressed in RA, one of them, AREG, and its signalling through epidermal growth factor receptor also increased in RA compared with osteoarthritis synoviocytes. Nevertheless, the interpretation of bone marrow expression profiling seems to be of high complexity considering the variety of cell types and differentiation processes in this compartment.51

Therapy and prediction of response

Over the past two decades, the elucidation of RA pathomechanisms has led to the discovery of different molecular targets for treatment. The new biological drugs enable highly specific interference with the action of these targets. However, clinical response to all these biological agents is heterogeneous and switching between biological drugs against different or even the same target may change responsiveness. The combination of biological agents with methotrexate was tested and found to improve the response rate, thus becoming the standard treatment strategy for today. In 1999, Bridges52 stated that predictive markers indicative for therapeutic response to the new biological drugs are an important aim for future research. The meta-analysis by Hyrich et al53 suggests that the definitions of response based on mostly clinical parameters may not even be sufficient to define the optimal timepoint for switching therapy. Therefore, expression profiling has been applied by several groups in the search for new biomarkers that may predict or indicate response for each patient individually. In principle, measuring the target itself or molecular patterns related to the activity of the targeted pathomechanism may provide appropriate information.

Lindberg et al54 investigated synovial tissue biopsies before and after therapy with infliximab. In anti-TNF responders TNF and MMP-3 were significantly upregulated indicating that TNF itself but also MMP-3, which can be induced by TNF,55 could be important biomarkers for anti-TNF treatment stratification. In general, anti-TNF responders seem to exhibit a higher level of cellularity and inflammatory activity in expression profiles before therapy.56 Unfortunately, synovial biopsies are inconvenient, and therefore several groups have investigated blood in the search for predictive biomarkers. Koczan et al57 investigated PBMC very early 72 h after the initiation of therapy with infliximab. They identified a panel of predictive candidate genes including cytokines and chemokines such as CCL4, IL-8 and IL-1β that discriminated between responders and non-responders after 3 months. Lequerre et al58 also studied PBMC and described a set of 20 genes that predicted response with a sensitivity of 90% and a specificity of 70%. These genes belonged to functions such as cell adhesion, inhibition of cell migration/invasion, intracellular or extracellular signalling and innate or adaptive immunity. Another study suggests that IFN-related genes may indicate differences in responsiveness, but also stated that these genes were expressed much lower in this subgroup of RA patients compared with SLE patients.59 In the authors’ own studies, we investigated profiles of purified monocytes from RA patients before and after treatment with adalimumab.60 Genes differentially expressed in RA compared with normal donors were also distinctive for response after anti-TNF therapy for at least 3 months, and clustered responders together with normal donors and non-responders with untreated RA. Besides correct classification after treatment, a comparison of responders with non-responders before treatment suggests that few markers may even predict response before initiation of treatment.

Another approach to identify potential biomarkers for the prediction of response may be the analysis of the activity profile of different drugs. We compared two synovial fibroblast-derived cell lines from RA and normal donors before and after treatment with non-steroidal anti-inflammatory drugs, disease-modifying antirheumatic drugs and glucocorticoids.37 The most prominent suppression of disease-related profiles was achieved with glucocorticoids, especially by suppressing proinflammatory cytokines/chemokines. In a second study, we investigated the effect of these different drugs on chondrocytes that were stimulated with supernatant of untreated and treated RA synovial fibroblasts.61 Again, proinflammatory changes induced by the secretome of the RA synovial fibroblast were best suppressed by glucocorticoids. Besides the suppression of cytokines and chemokines, glucocorticoids reverted RA-induced protease activation, matrix degradation and the inhibition of cartilage matrix synthesis. Disease-modifying antirheumatic drugs and especially non-steroidal anti-inflammatory drugs exhibited only moderate to minor effects with drug-specific expression patterns. Fitting such drug-specific activity profiles to individual disease-related profiles could provide another strategy for selecting the best treatment. In addition, these profiles may also help to predict, recognise or confirm drug side-effects observed in clinical studies.25


In summary, expression profiling in RA has explored many different aspects and conditions of this disease. The most obvious finding is that the molecular network is of much more complexity than initially suspected. Not single molecules or pathomechanisms but a multitude of inflammatory activities seem to be involved. With the growing set of defined conditions, comparative analyses are stepwise combining the puzzle of functional interactions. Reference profiles of cell types, cytokine and drug-induced gene regulation will play a key role in this effort to dissect the biological system. Current datasets give a first vision of what is a specific and what an overlapping gene activity, and thus will stepwise help to classify the profiles from clinical samples according to major and minor functional components. However, this process is just evolving and will need much more collaborative effort to generate the reference data needed to decipher the large network of inflammatory gene activation in the various rheumatic diseases.



  • Competing interests None.

  • Provenance and Peer review Not commissioned; externally peer reviewed.