Synovial phenotypes in rheumatoid arthritis correlate with response to biologic therapeutics
- Glynn DennisJr†1,
- Cécile TJ Holweg†2,
- Sarah K Kummerfeld†3,
- David F Choy2,
- A Francesca Setiadi2,
- Jason A Hackney3,
- Peter M Haverty3,
- Houston Gilbert4,
- Wei Yu Lin1,
- Lauri Diehl5,
- S Fischer6,
- An Song6,
- David Musselman7,
- Micki Klearman7,
- Cem Gabay8,
- Arthur Kavanaugh9,
- Judith Endres10,
- David A Fox10,
- Flavius Martin1, 11 and
- Michael J Townsend2Email author
© Dennis et al.; licensee BioMed Central Ltd. 2014
Received: 29 July 2013
Accepted: 25 February 2014
Published: 30 April 2014
Rheumatoid arthritis (RA) is a complex and clinically heterogeneous autoimmune disease. Currently, the relationship between pathogenic molecular drivers of disease in RA and therapeutic response is poorly understood.
We analyzed synovial tissue samples from two RA cohorts of 49 and 20 patients using a combination of global gene expression, histologic and cellular analyses, and analysis of gene expression data from two further publicly available RA cohorts. To identify candidate serum biomarkers that correspond to differential synovial biology and clinical response to targeted therapies, we performed pre-treatment biomarker analysis compared with therapeutic outcome at week 24 in serum samples from 198 patients from the ADACTA (ADalimumab ACTemrA) phase 4 trial of tocilizumab (anti-IL-6R) monotherapy versus adalimumab (anti-TNFα) monotherapy.
We documented evidence for four major phenotypes of RA synovium – lymphoid, myeloid, low inflammatory, and fibroid - each with distinct underlying gene expression signatures. We observed that baseline synovial myeloid, but not lymphoid, gene signature expression was higher in patients with good compared with poor European league against rheumatism (EULAR) clinical response to anti-TNFα therapy at week 16 (P =0.011). We observed that high baseline serum soluble intercellular adhesion molecule 1 (sICAM1), associated with the myeloid phenotype, and high serum C-X-C motif chemokine 13 (CXCL13), associated with the lymphoid phenotype, had differential relationships with clinical response to anti-TNFα compared with anti-IL6R treatment. sICAM1-high/CXCL13-low patients showed the highest week 24 American College of Rheumatology (ACR) 50 response rate to anti-TNFα treatment as compared with sICAM1-low/CXCL13-high patients (42% versus 13%, respectively, P =0.05) while anti-IL-6R patients showed the opposite relationship with these biomarker subgroups (ACR50 20% versus 69%, P =0.004).
These data demonstrate that underlying molecular and cellular heterogeneity in RA impacts clinical outcome to therapies targeting different biological pathways, with patients with the myeloid phenotype exhibiting the most robust response to anti-TNFα. These data suggest a path to identify and validate serum biomarkers that predict response to targeted therapies in rheumatoid arthritis and possibly other autoimmune diseases.
Rheumatoid arthritis (RA) is an autoimmune disease characterized by symmetrical joint involvement, inflammation, synovial lining hyperplasia, and formation of invasive granulation tissue or pannus. Progression of RA pathogenesis is associated with impaired joint function resulting from immune-mediated destruction of bone and cartilage [1–3]. Considerable patient-to-patient variation exists in the number of affected joints, the levels of autoantibody titers and serum cytokines, and the rate of joint destruction [4, 5]. Disease heterogeneity is further evident upon histological examination of synovial tissues, where a spectrum of cellular compositions are found, ranging from diffuse leukocytic infiltration to well-organized, lymphocyte-containing follicle-like structures .
Not surprisingly, RA is also heterogeneous in response to treatment. Although the development of targeted therapeutic strategies blocking TNF α, IL-6 receptor, T-cell co-stimulation blockade and B-cell depletion have provided meaningful clinical benefit to patients, a key unmet need in the management of RA is the prospective identification of patients who are likely to benefit from specific therapies. We hypothesized that a deeper understanding of the molecular basis of disease heterogeneity will lead to the discovery of predictive biomarkers able to identify individual patients who will benefit from a particular therapeutic strategy .
Insight into pathogenic molecular pathways of RA has emerged in recent years from genome-wide analysis of synovial tissue gene expression. Multiple studies have assessed molecular heterogeneity in RA tissue, but few findings have been validated with subsequent cohorts. Early studies [8, 9] revealed considerable molecular heterogeneity and proposed RA patient subgroups exhibiting gene expression patterns consistent with ongoing inflammation and adaptive immunity or, alternatively, little immune infiltrate and instead expressing sets of genes involved in extracellular matrix remodeling . Further, it has been observed that lymphoid follicle-containing synovial samples have increased expression of sets of genes involved in Janus kinase (JAK)/signal transducer and activator of transcription (STAT) signaling, and IL-7 signal transduction , suggesting that differences in gene expression patterns reflect differences in relative cellular composition of the RA joint.
Gene and protein expression studies of synovial tissue at baseline prior to initiating TNFα blockade have also generated different hypotheses to account for the differences between good and poor responders. In two studies, patients who responded to anti-TNFα treatment had transcription profiles enriched for inflammatory processes and TNFα protein expression [12, 13], whereas another report concluded that good responders actually had lower inflammatory processes and cell-surface markers such as the IL-7 receptor alpha chain . A large gene expression study of synovial tissues from 62 patients obtained prior to initiating anti-TNFα therapy identified very few transcripts that were different between good and poor responders . In the current study, we build on these observations by characterizing different molecular phenotypes of RA synovium - lymphoid, myeloid and fibroid - and used these to identify soluble biomarkers that predict differential treatment effects in RA patients.
Patients and synovial tissues
Synovial tissues were obtained from RA subjects undergoing arthroplasty and/or synovectomy of affected joints (University of Michigan, two sequential cohorts, n = 49 and n = 20). Written consent was obtained from patients, and the University of Michigan Institutional Review Board approved the study protocol. RA was diagnosed based upon the 1987 College of Rheumatology (ACR) criteria . Patients were treated using the standard of care for RA (non-steroidal anti-inflammatory drugs (NSAIDs) and disease-modifying anti-rheumatic drugs (DMARDs)) and some patients were also treated with biologics (adalimumab, etanercept, infliximab, anakinra and rituximab). Patients were diagnosed with RA at least three years before surgery and 70% of patients for whom data were available were rheumatoid factor (RF)-positive. Excised tissues were immediately snap-frozen in liquid nitrogen and stored at -80°C. Each tissue was used for both histology and RNA extraction. For cryo-sectioning, samples were brought briefly to -20°C, sectioned and immediately returned to -80°C to maintain RNA integrity. All tissues used for downstream studies were prospectively randomized during processing and sectioning, prior to expression analysis, to minimize technical batch effects in the data.
Frozen samples were weighed and homogenized in RLT buffer (Qiagen, Valencia, California, USA) + β-mercaptoethanol (10 μl/ml) at a concentration of 100 mg/ml. Prior to isolating RNA using an RNeasy minikit (Qiagen) with on-column DNase digestion, samples were digested with Proteinase K (Qiagen) for 10 minutes at 55°C.
Histopathology and immunohistochemistry
Stains were performed on 5-μm-thick frozen sections of human synovial tissue fixed in acetone. Some sections were stained with hematoxylin and eosin for histologic evaluation. Other sections were blocked in 10% serum for 30 minutes and stained for the detection of cells expressing the following lineage markers (CD20 - mouse anti-human clone L26, 5 μg/ml, Dako (Carpinteria, California, USA); CD3 - rabbit anti-human antibody SP7, 1:200 dilution, NeoMarkers (Fremont, California, USA) and CD68 - mouse anti-human clone KP-1, 2.5 μg/ml, Dako). All immunohistochemical stains were detected with species specific, biotinylated secondary antibodies and 3,3′-diaminobenzidine (DAB).
The protocols for preparation of cRNA and for array hybridization were followed as recommended by Affymetrix, Inc. (Santa Clara, CA, USA). Samples were hybridized to GeneChip® Human Genome U133 Plus 2.0 Arrays (Affymetrix, Inc.). Arrays were washed and stained in the Affymetrix Fluidics station and scanned on a GeneChip® scanner 3000. Expression signals were obtained using the Affymetrix GeneChip® operating system and analysis software.
Microarray data analyses
Microarray data for all samples are freely available for download [GEO:GSE48780] [17, 18]. Statistical analysis of microarray data was performed with the open-source tools available in the statistical programming environment R  and the Bioconductor project . Microarray data was normalized using the robust multichip average method (RMA) [21, 22]. This approach included three steps: background correction, quantile normalization and summarization. Following RMA processing, probe sets were filtered to exclude those that are believed to cross-hybridize or show other deficiencies according to the Affymetrix quality assessment classification (only A-class probes were included). In addition, probe sets without an Entrez ID-mapping were excluded. Microarray data were further filtered to a single probe set per gene. For genes with multiple probesets, only the probe set with the largest variance was used .
For the primary analysis of the University of Michigan samples, probe sets were further filtered retaining the top 40% most variable genes based on their SD across all samples . Probe sets were then centered and scaled. In order to identify groups of samples that showed similar expression profiles, we used agglomerative hierarchical clustering (Ward’s method, Euclidean distance on scaled and centered data). We divided the samples into groups based on the resulting clustering. The optimal number of groups was selected via two common metrics that quantify the tightness of clustering by considering the distance between samples within a group and the inter-group distance: mean silhouette width and k-nearest neighbor distances. We calculated these metrics for between three and eight groups, and both metrics indicated that separating the samples into five groups minimized the within-group sample distance and maximized the between-group distance. For testing cluster robustness, we used a re-sampling approach in which we randomly excluded five samples from the dataset, then selected the top 40% highest variance genes and performed clustering using the partitioning around medoids (PAM) algorithm with k = 5. The frequency with which a pair of samples was found in the same cluster in a given re-sampling was calculated for all pairs. Significantly over-represented pathways between the phenotypes were identified using the Database for Annotation, Visualization and Integrated Discovery (DAVID) tool . For each phenotype a set up-regulated and a separate set of downregulated genes was identified by comparing samples from that phenotype with all other samples and selecting genes that were differentially expressed at a false discovery rate (FDR) cutoff of 0.01. These differentially expressed sets were used as input to the DAVID tool using the default parameters recommended by the developers. Outputs from the DAVID analysis, including levels of genes from each process within the four synovial groups as defined by their t-statistic values and P-values, are available in the Additional files 1 and 2. The external dataset GSE21537 was downloaded from the GEO database, and was normalized and background-corrected using the variance stabilization and normalization (VSN) for microarray.
Gene set analysis
Pathway level analysis was carried out using gene set enrichment analysis (GSEA), using the Bioconductor GSEAlm package . Gene sets used in the analysis comprised the Molecular Signatures DataBase (MSigDB) from the Broad Institute , purified immune-cell type-specific gene expression , and a manually curated list of genes associated with angiogenesis processes. In addition, gene sets were defined based upon gene expression from microarray analysis of in vitro stimulated sorted blood monocytes (CD14+) that underwent classical activation (M1) with lipopolysaccharide (LPS) and IFNγ versus alternative activation (M2) with IL-4 and IL-13 for 24 hours, as well as in vitro stimulation of primary synovial fibroblasts from RA patients with TNFα or media-only control for 6 hours. All genes in each of the gene sets are listed in Additional file 3: Table S1. Summary gene-set scores were calculated using a quartile trimmed mean of the normalized probe-set values present in the gene set. Statistical significance of gene-set scores between the different synovial phenotypes was calculated using the t-test followed by Benjamini-Hochberg correction of P-values .
Group-specific genes for the myeloid, lymphoid and fibroid phenotypes were defined by identification of genes that were differentially expressed between each pair of groups using a moderated t-statistic (FDR <0.01), and then a list of genes was assembled for each group of the genes that were upregulated between that group and one or more others. Any gene that was differentially expressed between more than one pair of groups was discarded and the top 100 upregulated genes for each group were selected based on P-value ranking. Genes are listed in Additional file 3: Table S1. To assess relationships between the group-specific gene sets and response to anti-TNFα treatment, each group-specific gene set was mapped to the microarray expression dataset generated by  utilizing all available matching genes. Receiver-operating characteristic (ROC) analysis was performed using continuous gene-set scores compared against the European League Against Rheumatism (EULAR) good-versus-poor response criteria to anti-TNFα treatment, and area under the ROC curve (AUC) was determined for each gene set.
Serum biomarker assessments in the ADalimumab ACTemrA (ADACTA) clinical trial
Serum samples from 198 of the 326 patients in the ADACTA trial (ClinicalTrials.gov Identifier: NCT01119859) , where written consent had been given for exploratory biomarker analysis, were assessed for baseline pre-treatment levels of soluble intercellular adhesion molecule 1 (sICAM1) and C-X-C motif chemokine 13 (CXCL13) using customized electrochemiluminescence assays incorporating sample diluent blocking reagents to minimize interference from heterophilic antibodies. Biomarker subgroups were defined as low (below pretreatment median) or high (equal to or greater than pretreatment median) for each of the two markers. Relative treatment effectiveness (week-24 ACR50 criteria) of adalimumab compared with tocilizumab was assessed by logistic regression for each biomarker-defined subgroup. An odds ratio >1.0 and <1.0 than one correspond to favorable outcomes for adalimumab or tocilizumab respectively. Subpopulation treatment effect pattern plot (STEPP) analysis  was also performed on relative treatment effectiveness (week-24 ACR50 response) of adalimumab compared with tocilizumab for these two biomarkers. Assessment of statistical significance between subgroups was assessed using the Fisher exact test. ROC analysis was performed using continuous serum biomarker values compared against achievement of ACR50 response at 24 weeks for adalimumab or tocilizumab, and the AUC was determined.
Molecular phenotypes in RA synovium
To characterize putative phenotypes of RA according to their pathway composition we first identified subsets of genes that were specifically upregulated within each of the four clusters using a one-versus-all approach (see Methods). Each of the cluster-specific gene lists was then subjected to keyword over-representation analysis using DAVID. Immune response genes were abundant in both C1 (now termed the lymphoid phenotype) and C2 (myeloid phenotype), with the C1 lymphoid gene sets highly restricted to B and/or T lymphocyte activation and differentiation, immunoglobulin production, and antigen presentation together with enrichment of cytokine signaling including the Jak/STAT pathway and IL-17 signaling (Figure 1B). In contrast, the gene sets upregulated in the C2 myeloid group were also enriched for immune function, but were characterized by processes associated with chemotaxis, TNFα and IL-1β production, Toll-like receptor and nucleotide-binding oligomerization domain (NOD)-like receptor signaling, Fcγ-receptor-meditated phagocytosis, and proliferation of mononuclear cells. Cluster 3 (designated a low inflammatory phenotype) showed only enrichment for inflammatory response and wound response processes. The remaining C4 cluster, designated the fibroid phenotype, was enriched for genes associated with transforming growth factor (TGF) β signaling, bone morphogenetic protein (BMP) signaling together with associated Sma Mothers Against Decapentaplegic (SMAD) binding, as well as endocytosis and cell projection processes (Figure 1B) but lacked enrichment of any immune system processes. We further confirmed that the identified processes of interest were not solely driven by a small set of recurring genes, by directly comparing each gene set identified by the DAVID analysis with each other and observing that their overlap was generally low (Additional file 5: Figure S2). However, these analyses also suggested certain biological processes might reflect similar gene expression profiles occurring together in the same patients, for example, Toll-like receptor signaling, NOD-like receptor signaling, and Fc-γR-mediated phagocytosis occurred together primarily in the myeloid group, whereas processes such as antigen processing and presentation overlapped with both lymphoid group processes such as B and T cell activation and myeloid group processes such as FcγR-mediated phagocytosis and mononuclear cell proliferation, as might be expected based upon their connected immunological roles. Further, examination of genes that were specifically downregulated within each of the four clusters indicated the C4 fibroid cluster had significant downregulation of multiple immune-system processes associated with B cells, immunoglobulins, myeloid cells, innate immune response, including NOD-like receptor signaling, and chemotactic processes (Additional file 6: Figure S3A). In contrast, the C1 cluster had significant downregulation of TGFα and Wnt signaling together with processes associated with mesenchymal cell proliferation, proteolysis, cellular transport and RNA metabolism and processing, whereas both the C2 and C1 clusters had decreased representation of processes associated with transcription and splicing. As observed for the upregulated gene processes, the overlap between downregulated gene processes was also low (Additional file 6: Figure S3B).
Assessment of gene expression and gene sets in RA synovial clusters
We further assessed biological processes associated with the synovial phenotypes using experimentally derived gene-set modules representing a spectrum of hematopoietic lineage cells derived from specific expression in purified classically activated M1 monocytes, alternatively activated M2 monocytes, B cells, T cells, TNFα-stimulated synovial fibroblasts and angiogenesis-associated genes (see Methods and Additional file 3: Table S1 for a list of the module genes). The lymphoid phenotype was enriched specifically for B-cell modules (Figure 3B) whereas the myeloid phenotype was enriched for inflammatory M1 monocytes and TNFα-induced modules (Figure 3D, E). In contrast, T-cell genes were expressed similarly in both lymphoid and myeloid phenotypes (Figure 3C). The M2 monocyte module was expressed most highly in the low inflammatory phenotype (Figure 3F) while the angiogenesis module was highest in the fibroid phenotype and lowest in the lymphoid phenotype (Figure 3G). Application of the M1-monocyte and B-cell gene sets to two additional RA synovial datasets showed consistent differential expression patterns to those observed in the initial training dataset, further indicating that these molecular axes define a large proportion of the transcriptional heterogeneity found in the RA synovium (Additional file 7: Figure S4). Further, patients with lower levels of B cell and M1 monocytes had increased levels of fibroid subset genes consistent with the pattern seen in the training data set (Additional file 7: Figure S4B-D). Further, survey of tissue sections characterized by high or low levels of B lymphocytes determined by immunohistochemistry compared with the magnitude of a B-cell gene-set score demonstrated the correlation between histology and gene-set data (Additional file 8: Figure S5). These gene expression data support the notion that there are at least two inflammatory axes of disease in the RA synovium comprising activation of B cells and activation of inflammatory monocytes that are not completely overlapping, whereas other synovial tissues display a low inflammatory pauci-immune phenotype with potential angiogenic, osteoclast/osteoblast dysregulation and fibroblast activation processes in action. Consistent with lack of immune system involvement in the fibroid synovial phenotype, we observed that, for the patients who had available data on serological status, 100% of lymphoid- and myeloid-phenotype patients were RF-positive, 75% of the low inflammatory phenotype patients were RF-positive, and in contrast the fibroid phenotype patients were RF-negative.
Clinical response to targeted therapies
These results were further confirmed by additional analysis of this dataset using the previously utilized gene sets, which showed that the pretreatment biological process most strongly associated with good versus poor response to anti-TNFα therapy was classically M1 activated M1 monocytes (P = 0.006, Figure 4D), whereas in contrast neither the B-cell or T-cell gene sets showed no significant association with response (Figure 4E and F, P = 0.18 and P = 0.9 respectively). We further observed trends in association of pretreatment levels of M2 alternatively activated monocytes (P = 0.054, Additional file 9: Figure S6A) and TNFa-treated synovial fibroblasts (P = 0.08, Additional file 9: Figure S6B), whereas angiogenesis processes were significantly associated with good response (P = 0.018, Additional file 9: Figure S6C). In addition, we conducted ROC analysis of the gene sets versus EULAR response, and calculation of the AUC revealed that, consistent with the above findings, the myeloid and M1 classically activated monocyte gene sets produced the largest AUCs (0.65, Additional file 10: Figure S7A; and 0.77, Figure S7D respectively). These data indicate that application of predefined molecular synovial phenotypes, namely the myeloid phenotype and associated M1-activated monocytes, has the potential to enrich for responders to anti-TNFα therapy and that pretreatment levels of these biological processes were most strongly associated with anti-TNFα therapeutic outcome.
Derivation of serum biomarkers from differential synovial gene expression
We developed immunoassays to determine levels of circulating soluble ICAM1 (sICAM1) and CXCL13 in serum, and tested pretreatment samples from patients with active RA enrolled in the ADACTA trial (below). We observed that both serum biomarkers were significantly higher in disease compared with samples from non-disease control donors (Figure 5C, D) but importantly were only weakly correlated with each other (Spearman P <0.33, Figure 5E) suggesting they are reflective of different inflammatory immune processes.
sICAM1 and CXCL13 define RA subpopulations with differential clinical outcomes to adalimumab (anti-TNFα compared with tocilizumab (anti-IL-6R) therapy
We finally assessed whether baseline levels of sICAM1 and CXCL13 were differentially associated with subsequent treatment outcome to adalimumab compared with tocilizumab, as we hypothesized based upon the previous results that a population with elevated levels of a myeloid biomarker have elevated clinical response to anti-TNFα therapy but that elevation of a lymphoid marker would not. We utilized pretreatment samples from the ADACTA trial, a randomized, double blind, controlled phase-4 head to head study of tocilizumab (a humanized monoclonal antibody that binds to membrane-bound and soluble forms of the human IL-6 receptor) monotherapy, compared with adalimumab (a fully human monoclonal antibody against TNFα) monotherapy, in methotrexate-intolerant patients with active RA . This trial was notable as it allowed an initial assessment of biomarker-defined populations within the same trial against two different targeted therapies. As this was a post hoc exploratory analysis without pre-specified biomarker thresholds, we first assessed each biomarker individually using the median as a cutoff to define biomarker-low and biomarker-high subpopulations.
Summary of baseline biomarker-defined subgroup efficacy at 24 weeks in the ADACTA trial
Biomarker subset, number
ADA ACR20 (%)
ADA ACR50 (%)
ADA ACR70 (%)
ADA ∆DAS28-ESR (±SE)
ACR50 odds ratio ADA versus TCZ (95% CI)
Biomarker subset, number
TCZ ACR20 (%)
TCZ ACR50 (%)
TCZ ACR70 (%)
TCZ ∆DAS28-ESR (±SE)
ACR50 odds ratio TCZ vs. ADA (95% CI)
sICAM1 and CXCL13 biomarker populations were defined by cutoffs determined by the median values. We explored the heterogeneity of the relative treatment effect using alternative biomarker cutoffs using STEPP analysis. Assessment of individual biomarkers showed that increasing levels of sICAM1 were associated with increasing likelihood of ACR50 response to adalimumab versus tocilizumab (Additional file 11: Figure S8A) but increasing levels of CXCL13 were associated with decreasing ACR50 response to adalimumab versus tocilizumab (Additional file 11: Figure S8B). Further, examination of continuous levels of both biomarkers using two-dimensional STEPP analysis also showed the highest likelihood of ACR50 response to adalimumab versus tocilizumab in patients with the highest levels of sICAM1 but the lowest levels of CXCL13 (Additional file 11: Figure S8C), whereas conversely, the lowest likelihood of response to adalimumab versus tocilizumab was observed in the patient population with the lowest sICAM1 and highest CXCL13 levels. These data suggest that further differentiation of relative treatment effect may be observed using optimized cutoffs, as determined in a prospective study.
Finally, ROC analysis was performed to assess the predictive ability for ACR50 response of these two biomarkers on an individual patient basis. sICAM1 and CXCL13 showed only modest predictive ability for adalimumab or tocilizumab on an individual patient basis based upon their respective AUCs (0.57 and 0.6 respectively, Additional file 12: Figure S9A, D), whereas assessment of the two biomarkers in combination showed slight increases in the respective AUCs (Additional file 12: Figure S9C, D, E, F).
In totality, these data illustrate the concept that myeloid and lymphoid phenotype-derived circulating biomarkers can together define RA patient subpopulations that show differential clinical response to therapies directed at different targets, and that myeloid-dominant patient populations with high levels of sICAM1 and low levels of CXCL13 had the most robust response to anti-TNFα therapy.
In this report, we describe the presence of major cellular and molecular heterogeneity in RA synovial tissue characterized by two inflammatory phenotypes dominated by B cells and plasmablasts (lymphoid) and inflammatory macrophages (myeloid) as well as a low inflammatory pauci-immune phenotype, show that elevation of the myeloid, but not lymphoid axis, in synovial tissue is significantly associated with good clinical outcome to anti-TNFα therapy, and finally show that two systemic biomarkers chosen based on their differential tissue expression between the inflammatory phenotypes, CXCL13 for lymphoid and sICAM1 for myeloid, together define RA patient subpopulations with differential clinical response to anti-TNFα compared with anti-IL-6R therapies.
The concept that important heterogeneity exists in RA synovial tissue both at a histological as well as at a molecular level has been previously illustrated by several seminal studies [8, 10, 33], which showed differential presence of histological synovial aggregates and diffuse synovial inflammation, as well as differential gene expression, across RA synovial samples. The objective of the current study was to test the idea that heterogeneous RA synovial tissues can be assigned to subgroups that share common patterns of gene expression, have different associated systemic biomarkers, and that might respond differentially to therapy. Thus, we employed an analysis strategy that queried independently the questions of molecular heterogeneity and response heterogeneity. First, we assessed molecular heterogeneity of RA synovium independent of treatment response and validated proposed phenotypes using various molecular techniques and external patient cohorts. We next observed that core biological modules, as defined using pathway analysis, designated lymphoid (B cell- and plasmablast-dominated), myeloid (macrophage and NF-κB process dominated) and fibroid (comprising hyperplastic but pauci-immune tissues) could be surveyed across multiple RA patient synovial tissue cohorts to identify reproducible RA phenotypes. Importantly, the dominant biology associated with each gene expression-defined subset was consistent with histological and flow cytometry assessment of synovial tissue where the lymphoid subset was associated with presence of histological aggregates and, the myeloid subset with more diffuse immune infiltration while the fibroid subset had little immune infiltration and complete absence of aggregates. Further, survey of tissue sections characterized by high or low levels of B lymphocytes determined by immunohistochemistry correlated with the magnitude of a B cell gene-set score. We also observed the presence of a low inflammatory phenotype, indicating that synovial heterogeneity exists as a continuum of dysregulated biological processes rather than absolutely discrete subsets of disease. We did not observe differences in therapeutic usage (methotrexate, anti-TNFα agents, steroids) between patients with different synovial phenotypes where these data were available (data not shown). However, we did note that for the patients with data available, RF serological positivity was restricted to the lymphoid, myeloid and a majority of the low inflammatory phenotype patients. These data are consistent with previously observed gene expression heterogeneity in RA synovial tissue suggesting there are both inflammatory and non inflammatory synovial subgroups in RA. We further observed presence of patients with low or high inflammatory phenotypes based upon M1-activated monocytes, B cell and fibroid gene sets in two additional datasets, although the M1 and B cell gene sets were not as divergent as observed in the original training set. Reasons for this could include introduction of additional noise and loss of sensitivity due to the different platform used in the GSE21537 dataset resulting in loss of data due to missing or non-mapping probes as compared with the Affymetrix platform as well as differences in the patient populations, as there were higher levels of fibroid gene-set scores in both patient cohorts compared with the training dataset, meaning decreased representation of patients in the highly inflammatory subgroups.
Indeed, it has been clearly shown that patients with high levels of expression of inflammatory genes in the synovium have higher levels of systemic inflammation including C-reactive protein levels, ESRs and platelet counts as well as a shorter duration of disease as compared to patients with low synovial inflammation . Further, absence of significant synovial inflammation has been linked to decreased presence of anti-citrullinated protein antibodies . Consistent with this finding of a pauci-immune phenotype of RA, patients with lower levels of both synovial and systemic inflammation have been shown to have lower drug-response rates to both B-cell depletion therapy and anti-TNFα [36–38].
We then assessed whether the inflammatory biological modules would be differentially informative for predicting the outcome of response to anti-TNFα therapy through analysis of a large and well-defined external dataset. Strikingly, patients with high pretreatment expression of genes defined in the myeloid phenotype and M1 classically activated monocytes, but not high levels of lymphoid subset or B-cell genes, showed a greater 16-week good EULAR response to infliximab treatment. This is consistent with the observation that inflammatory M1 macrophages, a key lineage involved in production of TNFα, as well as expression of TNFα itself along with IL-1β and NF-κB-associated processes, are preferentially increased in the myeloid phenotype compared with all of the others. Further, other studies have consistently concluded that baseline levels of synovial macrophages and TNFα gene expression are correlated with response [13, 39], suggesting the presence of TNFα-secreting classically activated monocytes and macrophages are important for clinical outcome. However, the EULAR moderate responders had a wide range of values for both the myeloid and M1 genes, which suggest that other factors will contribute to determining treatment outcome with anti-TNFα agents. In contrast, a large histological study demonstrated that RA patients with high levels of synovial lymphoid neogenesis (LN), comprising highly organized B/T cell aggregates, demonstrated resistance to anti-TNFα therapy and good clinical outcome in these patients was accompanied with reversal of LN . Consistent with this, we observed that the presence of the lymphoid phenotype was not a predictor of response to anti-TNFα despite being associated with the presence of synovial inflammation and histological aggregates. In sum, these data suggest that simply the presence of inflammation alone is insufficient to predict clinical outcome to anti-TNFα treatment, and rather that sub-phenotypes of synovitis show differential clinical benefit with the lymphoid phenotype showing greater resistance to anti-TNFα as compared with the myeloid phenotype, perhaps due in part to the presence of other major processes driving synovitis including production of other inflammatory mediators, LN, and robust antigen presentation by autoreactive B cells. It is also noteworthy that we observed an association between pretreatment expression of genes associated with angiogenesis and clinical response to anti-TNFα, suggesting that the presence of synovial neoangiogenesis may also contribute to favorable outcome to blockade of TNFα.
Next, we hypothesized that the biological processes underlying the RA phenotypes might allow for rational serum protein biomarker selection to prospectively identify patient populations prior to starting a targeted therapy. As synovial tissue is not readily available for prospective assessment prior to initiation of therapy, systemic circulating biomarkers have greater potential utility although they will likely integrate the activity of specific biological pathways in multiple tissues, including the secondary lymphoid system, in addition to synovial tissue. We assessed candidates that were differentially expressed in the inflammatory lymphoid and myeloid subsets using a statistical ranking, and looked for markers that were strongly elevated in RA serum as compared with serum from non disease control donors. Two markers that fulfilled these criteria were soluble ICAM1 (myeloid) and CXCL13 (lymphoid). ICAM1, an adhesion molecule that binds to LFA-1, is a gene that is strongly regulated by NF-κB signaling and is upregulated on a variety of cell types in response to TNFα signaling including synovial fibroblasts and especially vascular endothelial cells, both of which are highly represented in the inflammatory rheumatoid synovium [41, 42]. sICAM1 is shed from the cell membrane by proteolytic cleavage. CXCL13 is a B cell chemoattractant that is highly expressed by follicular dendritic cells in secondary lymphoid tissue and ectopic germinal centers and is induced by LTα/LTβR signaling . Further, a recent report of a small synovial biopsy study of RA patients undergoing rituximab therapy showed a correlation between synovial tissue expression of CXCL13 and levels of CXCL13 protein in the serum (r = 0.6)  that suggests CXCL13 expression in the rheumatoid synovium is a major source of serum CXCL13. Synovial and serum levels of CXCL13 have also recently been linked with radiological joint destruction in RA patients , which argues that this gene, and by association the lymphoid synovial phenotype, is linked with progressive and destructive RA pathogenesis. In contrast, to our knowledge no reports have been made to date that have directly compared sICAM1 levels in serum with ICAM1 gene expression in synovial tissue, and we have not been able to conduct such an analysis in this study due to incomplete matching serum samples. Analysis of serum samples from the ADACTA adalimumab (anti-TNFα) compared with tocilizumab (anti-IL-6R) trial facilitated an assessment of these biomarkers in an inflammatory RA population that not only allowed a direct comparison of clinical response to different targeted therapies within one clinical study, but also avoided confounding effects of concomitant immunosuppression from background methotrexate as this study was conducted using both therapeutic agents as monotherapy . Consistent with our model of different inflammatory axes being present in RA, we noted that although both sICAM1 (myeloid) and CXCL13 (lymphoid) were significantly elevated in disease compared with control samples, they were only weakly correlated to each other. Further, we noted that patients with high pretreatment serum sICAM1 levels and decreased CXCL13 levels (high myeloid and low lymphoid activity) had increased ACR50 and ACR70 response rates and decreased DAS28-ESR scores to anti-TNFα therapy compared with anti-IL-6R therapy, whereas conversely, patients with high CXCL13 and decreased sICAM1 levels had preferential response to anti-IL-6R compared with anti-TNFα therapy. We did note differences in the magnitude of the differences between ACR50 response rates and changes in DAS28-ESR between the biomarker-defined populations in the tocilizumab arm where the changes in DAS28 were consistent but smaller than those observed for ACR50. These differences could not be accounted for by one component of the response instrument, for example, ESR or swollen-joint count, and are likely due more to differences in precision between the two instruments. These results are consistent with the previous data showing that patients with elevation of the myeloid inflammatory axis had robust responses to anti-TNFα drugs, and further emphasize that within an inflammatory RA population, there are patient subsets that subsequently have differential clinical outcomes to different targeted therapies.
What underlying biological basis could explain why blockade of the IL-6 pathway causes robust clinical responses in a different patient population to that responding to anti-TNFα blockade? Although IL-6 has long been appreciated as a key inflammatory cytokine important in the pathogenesis of RA as well as other inflammatory diseases , its biology and expression are not completely overlapping with that of TNFα. Our synovial tissue gene-expression data have shown that although TNFα is strongly associated with the myeloid phenotype and activity of classically activated myeloid cells and NF-κB pathway activity, IL-6, its receptors IL-6R and IL-6ST/gp130, and the key IL-6-associated TF, STAT3, are more broadly expressed across the lymphoid and low inflammatory synovial subsets (Figure 3A) and are not highly correlated with TNFα expression or restricted to the myeloid phenotype. Indeed, IL-6 can be induced in a variety of cell lineages exposed to multiple inflammatory stimuli in the joint including synovial fibroblasts themselves [32, 46]. Further, the IL-6/IL-6R pathway signals using the JAK/STAT pathway in contrast to the canonical NF-κB signaling predominantly utilized by TNFα  and plays a key role in inducing B cells to differentiate to antibody-secreting cells. Importantly, anti-IL-6R therapy has been shown to be effective in patients who are refractory to anti-TNFα therapies . Thus, it is conceivable that the IL-6/IL-6R pathway is highly involved with the driving synovitis in the B-cell-dominant lymphoid axis as well as potentially similarly important in driving synovitis in the low inflammatory subset, whereas in contrast, within the activated monocyte-dominated myeloid axis the TNFα pathway is dominant in driving synovitis such that blockade of IL-6 signaling is less effective. Whilst intriguing, and consistent with the biological hypotheses developed based upon our synovial tissue analyses, the findings described here represent only an initial testing of the sICAM1/CXCL13 biomarker hypothesis without a predefined cutoff for the analysis, hence our utilization of the median as the cutoff for this analysis, and the statistical power was limited by available patient numbers and multiple testing issues. Furthermore, analysis of these biomarkers on an individual patient basis using ROC analysis showed that they have only modest predictive ability for ACR50 outcome to adalimumab or tocilizumab at 24 weeks. Therefore, although the biomarkers described here demonstrate the presence of populations of RA patients with differential clinical response to targeted therapies, they do not presently have strong clinical utility for decision-making for individual patients. Improvement of individual patient predictive-ability might be achieved by incorporation of additional biomarkers into a predictive model that could be subjected to rigorous confirmatory studies in larger patient cohorts treated with anti-TNFα and anti-IL-6/IL-6R blocking agents including combination treatment with methotrexate, with incorporation of prespecified cutoff values in the analysis plan. Indeed, the two-dimensional STEPP analysis performed in this study suggested that altering the biomarker threshold cutoffs for both sICAM1 and CXCL13 could yield greater efficacy differentials for ACR50 response rates between adalimumab and tocilizumab than those achieved by using their respective medians.
Additional limitations of this study include limited availability of clinical data in the RA cohort used for the initial gene-signature discovery owing to the retrospective nature of interrogation of clinical chart data after sample collection from joint surgery, and a lack of consent for chart review in some cases. In particular, there were incomplete or missing data for serological autoantibody status for RF or anti-citrullinated protein antibodies. Also, the RA patient population studied for synovial gene expression represents late-stage disease where patients received joint surgery to correct deformity, replace joints, or manage pain. This study also does not address the presence and stability of synovial phenotypes longitudinally from early to late-stage disease and with respect to development of bone erosion. Finally, in the current study we have not applied an exhaustive investigation of all the potential serum biomarkers that may correlate with synovial subtypes, in part due to the desire to minimize multiple testing issues due to the limited number of anti-TNFα-treated patient samples available for biomarker analysis. These important questions are being addressed in a series of follow-up prospective studies.
Utilizing genome-wide expression analysis of synovial tissues from a large RA cohort, we have defined distinct molecular and cellular phenotypes that reflect the considerable heterogeneity present in the RA synovium. In particular, two distinct inflammatory axes emerge from this analysis: one dominated by B cells and the other dominated by inflammatory macrophages and NF-κB-activating cytokines, such as TNFα. It is important to point out that these cellular and molecular signatures, as well as the RA patients, represent a continuous rather than a discrete distribution, as is evident from the presence of lower inflammatory patients with intermediate molecular characteristics between these polar phenotypes. Analysis of respective gene-set modules and serum biomarkers suggest differential clinical response to anti-TNFα and anti-IL6R therapy is dependent in part on the presence of these inflammatory axes. A further subgroup of patients presented with a pauci-immune phenotype lacking major B cell or macrophage infiltration and may reflect a distinct subgroup of patients. These synovial phenotypes explain some of the underlying clinical and drug response heterogeneity in RA, and identifying and stratifying patients prospectively with respect to their synovial phenotype, for example by using blood biomarkers, may be important in making therapeutic decisions for targeting therapies. Such considerations are also likely to be very important for clinical trial design for new therapies to select patients prospectively for increased clinical response rates, and for the design of clinical studies to differentiate targeted therapies with different mechanisms of action.
Flavius Martin and Michael J Townsend co-directed the project.
American College of Rheumatology
area under the receiver-operating characteristic curve
bone morphogenetic protein
C-X-C motif chemokine 13
disease activity score (from 28 joints)
Database for Annotation Visualization and Integrated Discovery
disease-modifying anti-rheumatic drug
erythrocyte sedimentation rate
European League Against Rheumatism
fluorescence-activated cell sorting
false discovery rate
lower limit of quantification
Molecular Signatures DataBase
nuclear factor kappa-light-chain-enhancer of activated B cells
non-steroidal anti-inflammatory drug
partitioning around medoids
robust multichip average
soluble intercellular adhesion molecule 1
signal transducer and activator of transcription
subpopulation treatment effect pattern plot
tumor necrosis factor.
We thank Drs Andrew Urquhart, Kevin Chung and the orthopedic and operating room staff of the University of Michigan for the collection of synovial samples, Dr Zora Modrusan for support in generating microarray data, and Drs Timothy Behrens, John Monroe, Nicholas Lewin-Koh and Robert Gentleman for helpful discussions regarding the data analysis. We also thank Josefa Chuh, Marion Patrick, Christopher Motyl, and Peter White for technical assistance.
- Goronzy JJ, Weyand CM: Rheumatoid arthritis. Immunol Rev. 2005, 204: 55-73. 10.1111/j.0105-2896.2005.00245.x.View ArticlePubMedGoogle Scholar
- Lee DM, Weinblatt ME: Rheumatoid arthritis. Lancet. 2001, 358: 903-911. 10.1016/S0140-6736(01)06075-5.View ArticlePubMedGoogle Scholar
- Tak PP, Bresnihan B: The pathogenesis and prevention of joint damage in rheumatoid arthritis: advances from synovial biopsy and tissue analysis. Arthritis Rheum. 2000, 43: 2619-2633. 10.1002/1529-0131(200012)43:12<2619::AID-ANR1>3.0.CO;2-V.View ArticlePubMedGoogle Scholar
- Lindstrom TM, Robinson WH: Biomarkers for rheumatoid arthritis: making it personal. Scand J Clin Lab Invest Suppl. 2010, 242: 79-84.PubMed CentralView ArticlePubMedGoogle Scholar
- Scott DL, Wolfe F, Huizinga TW: Rheumatoid arthritis. Lancet. 2010, 376: 1094-1108. 10.1016/S0140-6736(10)60826-4.View ArticlePubMedGoogle Scholar
- Weyand CM, Goronzy JJ: Ectopic germinal center formation in rheumatoid synovitis. Ann NY Acad Sci. 2003, 987: 140-149. 10.1111/j.1749-6632.2003.tb06042.x.View ArticlePubMedGoogle Scholar
- Chan AC, Behrens TW: Personalizing medicine for autoimmune and inflammatory diseases. Nat Immunol. 2013, 14: 106-109. 10.1038/ni.2473.View ArticlePubMedGoogle Scholar
- van der Pouw Kraan TC, van Gaalen FA, Huizinga TW, Pieterman E, Breedveld FC, Verweij CL: Discovery of distinctive gene expression profiles in rheumatoid synovium using cDNA microarray technology: evidence for the existence of multiple pathways of tissue destruction and repair. Genes Immun. 2003, 4: 187-196. 10.1038/sj.gene.6363975.View ArticlePubMedGoogle Scholar
- van der Pouw Kraan TC, van Gaalen FA, Kasperkovitz PV, Verbeet NL, Smeets TJ, Kraan MC, Fero M, Tak PP, Huizinga TW, Pieterman E, Breedveld FC, Alizadeh AA, Verweij CL: Rheumatoid arthritis is a heterogeneous disease: evidence for differences in the activation of the STAT-1 pathway between rheumatoid tissues. Arthritis Rheum. 2003, 48: 2132-2145. 10.1002/art.11096.View ArticlePubMedGoogle Scholar
- van Baarsen LG, Bos CL, van der Pouw Kraan TC, Verweij CL: Transcription profiling of rheumatic diseases. Arthritis Res Ther. 2009, 11: 207-10.1186/ar2557.PubMed CentralView ArticlePubMedGoogle Scholar
- Timmer TC, Baltus B, Vondenhoff M, Huizinga TW, Tak PP, Verweij CL, Mebius RE, van der Pouw Kraan TC: 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: 2492-2502. 10.1002/art.22748.View ArticlePubMedGoogle Scholar
- van der Pouw Kraan TC, Wijbrandts CA, van Baarsen LG, Rustenburg F, Baggen JM, Verweij CL, Tak PP: Responsiveness to anti-tumour necrosis factor alpha therapy is related to pre-treatment tissue inflammation levels in rheumatoid arthritis patients. Ann Rheum Dis. 2008, 67: 563-566.View ArticlePubMedGoogle Scholar
- Wijbrandts CA, Dijkgraaf MG, Kraan MC, Vinkenoog M, Smeets TJ, Dinant H, Vos K, Lems WF, Wolbink GJ, Sijpkens D, Dijkmans BA, Tak PP: The clinical response to infliximab in rheumatoid arthritis is in part dependent on pretreatment tumour necrosis factor alpha expression in the synovium. Ann Rheum Dis. 2008, 67: 1139-1144. 10.1136/ard.2007.080440.PubMed CentralView ArticlePubMedGoogle Scholar
- Badot V, Galant C, Nzeusseu Toukap A, Theate I, Maudoux AL, Van den Eynde BJ, Durez P, Houssiau FA, Lauwerys BR: Gene expression profiling in the synovium identifies a predictive signature of absence of response to adalimumab therapy in rheumatoid arthritis. Arthritis Res Ther. 2009, 11: R57-10.1186/ar2678.PubMed CentralView ArticlePubMedGoogle Scholar
- Lindberg J, Wijbrandts CA, van Baarsen LG, Nader G, Klareskog L, Catrina A, Thurlings R, Vervoordeldonk M, Lundeberg J, Tak PP: The gene expression profile in the synovium as a predictor of the clinical response to infliximab treatment in rheumatoid arthritis. PLoS One. 2010, 5: e11310-10.1371/journal.pone.0011310.PubMed CentralView ArticlePubMedGoogle Scholar
- Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper NS, Healey LA, Kaplan SR, Liang MH, Luthra HS, Medsger TA, Mitchell DM, Neustadt DH, Pinals RS, Schaller JG, Sharp JT, Wilder RL, Hunder GG: The American rheumatism association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum. 1988, 31: 315-324. 10.1002/art.1780310302.View ArticlePubMedGoogle Scholar
- Barrett T, Troup DB, Wilhite SE, Ledoux P, Evangelista C, Kim IF, Tomashevsky M, Marshall KA, Phillippy KH, Sherman PM, Muertter RN, Holko M, Ayanbule O, Yefanov A, Soboleva A: NCBI GEO: archive for functional genomics data sets–10 years on. Nucleic Acids Res. 2011, 39: D1005-D1010. 10.1093/nar/gkq1184.PubMed CentralView ArticlePubMedGoogle Scholar
- Edgar R, Domrachev M, Lash AE: Gene expression omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res. 2002, 30: 207-210. 10.1093/nar/30.1.207.PubMed CentralView ArticlePubMedGoogle Scholar
- R Development Core Team: R: a language and environment for statistical computing. R Foundation for Statistical Computing. 2011, Vienna, Austria: R Foundation for Statistical Computing,http://cran.r-project.org. ISBN 3-900051-07-0,Google Scholar
- Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, Ellis B, Gautier L, Ge Y, Gentry J, Hornik K, Hothorn T, Huber W, Iacus S, Irizarry R, Leisch F, Li C, Maechler M, Rossini AJ, Sawitzki G, Smith C, Smyth G, Tierney L, Yang JY, Zhang J: Bioconductor: open software development for computational biology and bioinformatics. Genome Biol. 2004, 5: R80-10.1186/gb-2004-5-10-r80.PubMed CentralView ArticlePubMedGoogle Scholar
- Bolstad BM, Irizarry RA, Astrand M, Speed TP: A comparison of normalization methods for high density oligonucleotide array data based on variance and bias. Bioinformatics. 2003, 19: 185-193. 10.1093/bioinformatics/19.2.185.View ArticlePubMedGoogle Scholar
- Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, Speed TP: Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics. 2003, 4: 249-264. 10.1093/biostatistics/4.2.249.View ArticlePubMedGoogle Scholar
- Hackstadt AJ, Hess AM: Filtering for increased power for microarray data analysis. BMC Bioinformatics. 2009, 10: 11-10.1186/1471-2105-10-11.PubMed CentralView ArticlePubMedGoogle Scholar
- Bourgon R, Gentleman R, Huber W: Independent filtering increases detection power for high-throughput experiments. Proc Natl Acad Sci USA. 2010, 107: 9546-9551. 10.1073/pnas.0914005107.PubMed CentralView ArticlePubMedGoogle Scholar
- Dennis G, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, Lempicki RA: DAVID: Database for annotation, visualization, and integrated discovery. Genome Biol. 2003, 4: 3-10.1186/gb-2003-4-5-p3.View ArticleGoogle Scholar
- Oron AP, Jiang Z, Gentleman R: Gene set enrichment analysis using linear models and diagnostics. Bioinformatics. 2008, 24: 2586-2591. 10.1093/bioinformatics/btn465.PubMed CentralView ArticlePubMedGoogle Scholar
- Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, Mesirov JP: Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA. 2005, 102: 15545-15550. 10.1073/pnas.0506580102.PubMed CentralView ArticlePubMedGoogle Scholar
- Abbas AR, Baldwin D, Ma Y, Ouyang W, Gurney A, Martin F, Fong S, van Lookeren CM, Godowski P, Williams PM, Chan AC, Clark HF: Immune response in silico (IRIS): immune-specific genes identified from a compendium of microarray expression data. Genes Immun. 2005, 6: 319-331. 10.1038/sj.gene.6364173.View ArticlePubMedGoogle Scholar
- Hochberg Y, Benjamini Y: More powerful procedures for multiple significance testing. Stat Med. 1990, 9: 811-818. 10.1002/sim.4780090710.View ArticlePubMedGoogle Scholar
- Gabay C, Emery P, van Vollenhoven R, Dikranian A, Alten R, Pavelka K, Klearman M, Musselman D, Agarwal S, Green J, Kavanaugh A: Tocilizumab monotherapy versus adalimumab monotherapy for treatment of rheumatoid arthritis (ADACTA): a randomised, double-blind, controlled phase 4 trial. Lancet. 2013, 381: 1541-1550. 10.1016/S0140-6736(13)60250-0.View ArticlePubMedGoogle Scholar
- Lazar AA, Cole BF, Bonetti M, Gelber RD: Evaluation of treatment-effect heterogeneity using biomarkers measured on a continuous scale: subpopulation treatment effect pattern plot. J. Clin Oncol. 2010, 28: 4539-4544. 10.1200/JCO.2009.27.9182.PubMed CentralView ArticlePubMedGoogle Scholar
- Kishimoto T: Interleukin-6: from basic science to medicine–40 years in immunology. Annu Rev Immunol. 2005, 23: 1-21. 10.1146/annurev.immunol.23.021704.115806.View ArticlePubMedGoogle Scholar
- Weyand CM, Kang YM, Kurtin PJ, Goronzy JJ: The power of the third dimension: tissue architecture and autoimmunity in rheumatoid arthritis. Curr Opin Rheumatol. 2003, 15: 259-266. 10.1097/00002281-200305000-00013.View ArticlePubMedGoogle Scholar
- van Baarsen LG, Wijbrandts CA, Timmer TC, van der Pouw Kraan TC, Tak PP, Verweij CL: Synovial tissue heterogeneity in rheumatoid arthritis in relation to disease activity and biomarkers in peripheral blood. Arthritis Rheum. 2010, 62: 1602-1607.View ArticlePubMedGoogle Scholar
- van Oosterhout M, Bajema I, Levarht EW, Toes RE, Huizinga TW, van Laar JM: Differences in synovial tissue infiltrates between anti-cyclic citrullinated peptide-positive rheumatoid arthritis and anti-cyclic citrullinated peptide-negative rheumatoid arthritis. Arthritis Rheum. 2008, 58: 53-60. 10.1002/art.23148.View ArticlePubMedGoogle Scholar
- Hogan VE, Holweg CT, Choy DF, Kummerfeld SK, Hackney JA, Teng YK, Townsend MJ, van Laar JM: Pretreatment synovial transcriptional profile is associated with early and late clinical response in rheumatoid arthritis patients treated with rituximab. Ann Rheum Dis. 1888–1894, 2012: 71-Google Scholar
- Hueber W, Tomooka BH, Batliwalla F, Li W, Monach PA, Tibshirani RJ, Van Vollenhoven RF, Lampa J, Saito K, Tanaka Y, Genovese MC, Klareskog L, Gregersen PK, Robinson WH: Blood autoantibody and cytokine profiles predict response to anti-tumor necrosis factor therapy in rheumatoid arthritis. Arthritis Res Ther. 2009, 11: R76-10.1186/ar2706.PubMed CentralView ArticlePubMedGoogle Scholar
- Lal P, Su Z, Holweg CT, Silverman GJ, Schwartzman S, Kelman A, Read S, Spaniolo G, Monroe JG, Behrens TW, Townsend MJ: Inflammation and autoantibody markers identify rheumatoid arthritis patients with enhanced clinical benefit following rituximab treatment. Arthritis Rheum. 2011, 63: 3681-3691. 10.1002/art.30596.View ArticlePubMedGoogle Scholar
- Klaasen R, Thurlings RM, Wijbrandts CA, van Kuijk AW, Baeten D, Gerlag DM, Tak PP: The relationship between synovial lymphocyte aggregates and the clinical response to infliximab in rheumatoid arthritis: a prospective study. Arthritis Rheum. 2009, 60: 3217-3224. 10.1002/art.24913.View ArticlePubMedGoogle Scholar
- Canete JD, Celis R, Moll C, Izquierdo E, Marsal S, Sanmarti R, Palacin A, Lora D, de la Cruz J, Pablos JL: Clinical significance of synovial lymphoid neogenesis and its reversal after anti-tumour necrosis factor alpha therapy in rheumatoid arthritis. Ann Rheum Dis. 2009, 68: 751-756. 10.1136/ard.2008.089284.View ArticlePubMedGoogle Scholar
- Krenn V, Schedel J, Doring A, Huppertz HI, Gohlke F, Tony HP, Vollmers HP, Muller-Hermelink HK: Endothelial cells are the major source of sICAM-1 in rheumatoid synovial tissue. Rheumatol Int. 1997, 17: 17-27. 10.1007/PL00006846.View ArticlePubMedGoogle Scholar
- Witkowska AM, Borawska MH: Soluble intercellular adhesion molecule-1 (sICAM-1): an overview. Eur Cytokine Netw. 2004, 15: 91-98.PubMedGoogle Scholar
- Corsiero E, Bombardieri M, Manzo A, Bugatti S, Uguccioni M, Pitzalis C: Role of lymphoid chemokines in the development of functional ectopic lymphoid structures in rheumatic autoimmune diseases. Immunol Lett. 2012, 145: 62-67. 10.1016/j.imlet.2012.04.013.View ArticlePubMedGoogle Scholar
- Rosengren S, Wei N, Kalunian KC, Kavanaugh A, Boyle DL: CXCL13: a novel biomarker of B-cell return following rituximab treatment and synovitis in patients with rheumatoid arthritis. Rheumatol (Oxford). 2011, 50: 603-610. 10.1093/rheumatology/keq337.View ArticleGoogle Scholar
- Meeuwisse CM, van der Linden MP, Rullmann TA, Allaart CF, Nelissen R, Huizinga TW, Garritsen A, Toes RE, van Schaik R, van der Helm-van Mil AH: Identification of CXCL13 as a marker for rheumatoid arthritis outcome using an in silico model of the rheumatic joint. Arthritis Rheum. 2011, 63: 1265-1273. 10.1002/art.30273.View ArticlePubMedGoogle Scholar
- Choy E: Understanding the dynamics: pathways involved in the pathogenesis of rheumatoid arthritis. Rheumatol (Oxford). 2012, 51: v3-v11. 10.1093/rheumatology/kes113.View ArticleGoogle Scholar
- Chen G, Goeddel DV: TNF-R1 signaling: a beautiful pathway. Science. 2002, 296: 1634-1635. 10.1126/science.1071924.View ArticlePubMedGoogle Scholar
- Emery P, Keystone E, Tony HP, Cantagrel A, van Vollenhoven R, Sanchez A, Alecock E, Lee J, Kremer J: IL-6 receptor inhibition with tocilizumab improves treatment outcomes in patients with rheumatoid arthritis refractory to anti-tumour necrosis factor biologicals: results from a 24-week multicentre randomised placebo-controlled trial. Ann Rheum Dis. 2008, 67: 1516-1523. 10.1136/ard.2008.092932.PubMed CentralView ArticlePubMedGoogle Scholar
This article is published under license to 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 credited.