Synovial phenotypes in rheumatoid arthritis correlate with response to biologic therapeutics

Dennis et al Arthritis Research amp Therapy 2014 16R90httparthritis-researchcomcontent162R90

RESEARCH ARTICLE Open Access

Synovial phenotypes in rheumatoid arthritiscorrelate with response to biologic therapeuticsGlynn Dennis Jr1dagger Ceacutecile TJ Holweg2dagger Sarah K Kummerfeld3dagger David F Choy2 A Francesca Setiadi2Jason A Hackney3 Peter M Haverty3 Houston Gilbert4 Wei Yu Lin1 Lauri Diehl5 S Fischer6 An Song6David Musselman7 Micki Klearman7 Cem Gabay8 Arthur Kavanaugh9 Judith Endres10 David A Fox10Flavius Martin111 and Michael J Townsend2

Abstract

Introduction Rheumatoid arthritis (RA) is a complex and clinically heterogeneous autoimmune disease Currentlythe relationship between pathogenic molecular drivers of disease in RA and therapeutic response is poorlyunderstood

Methods We analyzed synovial tissue samples from two RA cohorts of 49 and 20 patients using a combination ofglobal gene expression histologic and cellular analyses and analysis of gene expression data from two furtherpublicly available RA cohorts To identify candidate serum biomarkers that correspond to differential synovialbiology and clinical response to targeted therapies we performed pre-treatment biomarker analysis compared withtherapeutic 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

Results We documented evidence for four major phenotypes of RA synovium ndash lymphoid myeloid lowinflammatory and fibroid - each with distinct underlying gene expression signatures We observed that baselinesynovial myeloid but not lymphoid gene signature expression was higher in patients with good compared withpoor European league against rheumatism (EULAR) clinical response to anti-TNFα therapy at week 16 (P =0011)We observed that high baseline serum soluble intercellular adhesion molecule 1 (sICAM1) associated with themyeloid phenotype and high serum C-X-C motif chemokine 13 (CXCL13) associated with the lymphoid phenotypehad differential relationships with clinical response to anti-TNFα compared with anti-IL6R treatment sICAM1-highCXCL13-low patients showed the highest week 24 American College of Rheumatology (ACR) 50 response rate toanti-TNFα treatment as compared with sICAM1-lowCXCL13-high patients (42 versus 13 respectively P =005)while anti-IL-6R patients showed the opposite relationship with these biomarker subgroups (ACR50 20 versus69 P =0004)

Conclusions These data demonstrate that underlying molecular and cellular heterogeneity in RA impacts clinicaloutcome to therapies targeting different biological pathways with patients with the myeloid phenotype exhibitingthe most robust response to anti-TNFα These data suggest a path to identify and validate serum biomarkers thatpredict response to targeted therapies in rheumatoid arthritis and possibly other autoimmune diseases

Trial registration ClinicalTrialsgov NCT01119859

Correspondence townsem1genecomdaggerEqual contributors2ITGR Diagnostics Discovery Genentech South San Francisco California USAFull list of author information is available at the end of the article

copy Dennis et al licensee BioMed Central Ltd This is an Open Access article distributed under the terms of the CreativeCommons Attribution License (httpcreativecommonsorglicensesby20) which permits unrestricted use distribution andreproduction in any medium provided the original work is properly cited

2014

Dennis et al Arthritis Research amp Therapy Page 2 of 182014 16R90httparthritis-researchcomcontent162R90

IntroductionRheumatoid arthritis (RA) is an autoimmune diseasecharacterized by symmetrical joint involvement inflam-mation synovial lining hyperplasia and formation of in-vasive granulation tissue or pannus Progression of RApathogenesis is associated with impaired joint functionresulting from immune-mediated destruction of boneand cartilage [1-3] Considerable patient-to-patient vari-ation exists in the number of affected joints the levels ofautoantibody titers and serum cytokines and the rate ofjoint destruction [45] Disease heterogeneity is furtherevident upon histological examination of synovial tissueswhere a spectrum of cellular compositions are found ran-ging from diffuse leukocytic infiltration to well-organizedlymphocyte-containing follicle-like structures [6]Not surprisingly RA is also heterogeneous in response

to treatment Although the development of targeted thera-peutic strategies blocking TNF α IL-6 receptor T-cell co-stimulation blockade and B-cell depletion have providedmeaningful clinical benefit to patients a key unmet needin the management of RA is the prospective identificationof patients who are likely to benefit from specific therap-ies We hypothesized that a deeper understanding of themolecular basis of disease heterogeneity will lead to thediscovery of predictive biomarkers able to identify individ-ual patients who will benefit from a particular therapeuticstrategy [7]Insight into pathogenic molecular pathways of RA has

emerged in recent years from genome-wide analysis of syn-ovial tissue gene expression Multiple studies have assessedmolecular heterogeneity in RA tissue but few findings havebeen validated with subsequent cohorts Early studies [89]revealed considerable molecular heterogeneity and pro-posed RA patient subgroups exhibiting gene expressionpatterns consistent with ongoing inflammation and adap-tive immunity or alternatively little immune infiltrateand instead expressing sets of genes involved in extra-cellular matrix remodeling [10] Further it has been ob-served that lymphoid follicle-containing synovial sampleshave increased expression of sets of genes involved inJanus kinase (JAK)signal transducer and activator of tran-scription (STAT) signaling and IL-7 signal transduction[11] suggesting that differences in gene expression pat-terns reflect differences in relative cellular composition ofthe RA jointGene and protein expression studies of synovial tissue

at baseline prior to initiating TNFα blockade have alsogenerated different hypotheses to account for the differ-ences between good and poor responders In two studiespatients who responded to anti-TNFα treatment had tran-scription profiles enriched for inflammatory processes andTNFα protein expression [1213] whereas another reportconcluded that good responders actually had lower in-flammatory processes and cell-surface markers such as

the IL-7 receptor alpha chain [14] A large gene expressionstudy of synovial tissues from 62 patients obtained priorto initiating anti-TNFα therapy identified very fewtranscripts that were different between good and poorresponders [15] In the current study we build on theseobservations by characterizing different molecular pheno-types of RA synovium - lymphoid myeloid and fibroid -and used these to identify soluble biomarkers that predictdifferential treatment effects in RA patients

MethodsPatients and synovial tissuesSynovial tissues were obtained from RA subjects under-going arthroplasty andor synovectomy of affected joints(University of Michigan two sequential cohorts n = 49and n = 20) Written consent was obtained from patientsand the University of Michigan Institutional Review Boardapproved the study protocol RA was diagnosed basedupon the 1987 College of Rheumatology (ACR) criteria[16] Patients were treated using the standard of care forRA (non-steroidal anti-inflammatory drugs (NSAIDs) anddisease-modifying anti-rheumatic drugs (DMARDs)) andsome patients were also treated with biologics (adalimu-mab etanercept infliximab anakinra and rituximab)Patients were diagnosed with RA at least three yearsbefore surgery and 70 of patients for whom data wereavailable were rheumatoid factor (RF)-positive Excised tis-sues were immediately snap-frozen in liquid nitrogen andstored at -80degC Each tissue was used for both histologyand RNA extraction For cryo-sectioning samples werebrought briefly to -20degC sectioned and immediatelyreturned to -80degC to maintain RNA integrity All tissuesused for downstream studies were prospectively random-ized during processing and sectioning prior to expressionanalysis to minimize technical batch effects in the data

RNA isolationFrozen samples were weighed and homogenized in RLTbuffer (Qiagen Valencia California USA) + β-mercap-toethanol (10 μlml) at a concentration of 100 mgmlPrior to isolating RNA using an RNeasy minikit (Qiagen)with on-column DNase digestion samples were digestedwith Proteinase K (Qiagen) for 10 minutes at 55degC

Histopathology and immunohistochemistryStains were performed on 5-μm-thick frozen sections of hu-man synovial tissue fixed in acetone Some sections werestained with hematoxylin and eosin for histologic evaluationOther sections were blocked in 10 serum for 30 minutesand stained for the detection of cells expressing the followinglineage markers (CD20 - mouse anti-human clone L26 5 μgml Dako (Carpinteria California USA) CD3 - rabbit anti-human antibody SP7 1200 dilution NeoMarkers (FremontCalifornia USA) and CD68 - mouse anti-human clone KP-1


25 μgml Dako) All immunohistochemical stains were de-tected with species specific biotinylated secondary antibodiesand 33prime-diaminobenzidine (DAB)

Microarray hybridizationThe protocols for preparation of cRNA and for arrayhybridization were followed as recommended by Affy-metrix Inc (Santa Clara CA USA) Samples were hybrid-ized to GeneChipreg Human Genome U133 Plus 20 Arrays(Affymetrix Inc) Arrays were washed and stained in theAffymetrix Fluidics station and scanned on a GeneChipregscanner 3000 Expression signals were obtained usingthe Affymetrix GeneChipreg operating system and ana-lysis software

Microarray data analysesMicroarray data for all samples are freely available fordownload [GEOGSE48780] [1718] Statistical analysisof microarray data was performed with the open-sourcetools available in the statistical programming environ-ment R [19] and the Bioconductor project [20] Micro-array data was normalized using the robust multichipaverage method (RMA) [2122] This approach includedthree steps background correction quantile normalizationand summarization Following RMA processing probesets were filtered to exclude those that are believed tocross-hybridize or show other deficiencies according tothe Affymetrix quality assessment classification (only A-class probes were included) In addition probe sets with-out an Entrez ID-mapping were excluded Microarray datawere further filtered to a single probe set per gene Forgenes with multiple probesets only the probe set with thelargest variance was used [23]For the primary analysis of the University of Michigan

samples probe sets were further filtered retaining thetop 40 most variable genes based on their SD across allsamples [24] Probe sets were then centered and scaledIn order to identify groups of samples that showed simi-lar expression profiles we used agglomerative hierarch-ical clustering (Wardrsquos method Euclidean distance onscaled and centered data) We divided the samples intogroups based on the resulting clustering The optimalnumber of groups was selected via two common metricsthat quantify the tightness of clustering by considering thedistance between samples within a group and the inter-group distance mean silhouette width and k-nearestneighbor distances We calculated these metrics for be-tween three and eight groups and both metrics indicatedthat separating the samples into five groups minimizedthe within-group sample distance and maximized thebetween-group distance For testing cluster robustness weused a re-sampling approach in which we randomly ex-cluded five samples from the dataset then selected the top40 highest variance genes and performed clustering

using the partitioning around medoids (PAM) algorithmwith k = 5 The frequency with which a pair of sampleswas found in the same cluster in a given re-sampling wascalculated for all pairs Significantly over-representedpathways between the phenotypes were identified usingthe Database for Annotation Visualization and IntegratedDiscovery (DAVID) tool [25] For each phenotype a setup-regulated and a separate set of downregulated geneswas identified by comparing samples from that phenotypewith all other samples and selecting genes that were differ-entially expressed at a false discovery rate (FDR) cutoff of001 These differentially expressed sets were used as inputto the DAVID tool using the default parameters recom-mended by the developers Outputs from the DAVID ana-lysis including levels of genes from each process withinthe four synovial groups as defined by their t-statisticvalues and P-values are available in the Additionalfiles 1 and 2 The external dataset GSE21537 was down-loaded from the GEO database and was normalized andbackground-corrected using the variance stabilization andnormalization (VSN) for microarray

Gene set analysisPathway level analysis was carried out using gene set en-richment analysis (GSEA) using the Bioconductor GSEAlmpackage [26] Gene sets used in the analysis comprised theMolecular Signatures DataBase (MSigDB) from the BroadInstitute [27] purified immune-cell type-specific gene ex-pression [28] and a manually curated list of genes associ-ated with angiogenesis processes In addition gene setswere defined based upon gene expression from microarrayanalysis of in vitro stimulated sorted blood monocytes(CD14+) that underwent classical activation (M1) withlipopolysaccharide (LPS) and IFNγ versus alternative acti-vation (M2) with IL-4 and IL-13 for 24 hours as well asin vitro stimulation of primary synovial fibroblasts fromRA patients with TNFα or media-only control for 6 hoursAll genes in each of the gene sets are listed in Additionalfile 3 Table S1 Summary gene-set scores were calculatedusing a quartile trimmed mean of the normalized probe-set values present in the gene set Statistical significance ofgene-set scores between the different synovial phenotypeswas calculated using the t-test followed by Benjamini-Hochberg correction of P-values [29]Group-specific genes for the myeloid lymphoid and fi-

broid phenotypes were defined by identification of genesthat were differentially expressed between each pair ofgroups using a moderated t-statistic (FDR lt001) andthen a list of genes was assembled for each group of thegenes that were upregulated between that group and oneor more others Any gene that was differentially expressedbetween more than one pair of groups was discarded andthe top 100 upregulated genes for each group were se-lected based on P-value ranking Genes are listed in


Additional file 3 Table S1 To assess relationships be-tween the group-specific gene sets and response toanti-TNFα treatment each group-specific gene set wasmapped to the microarray expression dataset generated by[15] utilizing all available matching genes Receiver-operating characteristic (ROC) analysis was performedusing continuous gene-set scores compared against theEuropean League Against Rheumatism (EULAR) good-versus-poor response criteria to anti-TNFα treatment andarea under the ROC curve (AUC) was determined foreach gene set

Serum biomarker assessments in the ADalimumabACTemrA (ADACTA) clinical trialSerum samples from 198 of the 326 patients in theADACTA trial (ClinicalTrialsgov Identifier NCT01119859)[30] where written consent had been given for exploratorybiomarker analysis were assessed for baseline pre-treatmentlevels of soluble intercellular adhesion molecule 1 (sICAM1)and C-X-C motif chemokine 13 (CXCL13) using custom-ized electrochemiluminescence assays incorporating sam-ple diluent blocking reagents to minimize interferencefrom heterophilic antibodies Biomarker subgroups weredefined as low (below pretreatment median) or high(equal to or greater than pretreatment median) for each ofthe two markers Relative treatment effectiveness (week-24 ACR50 criteria) of adalimumab compared with toci-lizumab was assessed by logistic regression for eachbiomarker-defined subgroup An odds ratio gt10 and lt10than one correspond to favorable outcomes for adalimu-mab or tocilizumab respectively Subpopulation treatmenteffect pattern plot (STEPP) analysis [31] was also performedon relative treatment effectiveness (week-24 ACR50 re-sponse) of adalimumab compared with tocilizumab forthese two biomarkers Assessment of statistical significancebetween subgroups was assessed using the Fisher exact testROC analysis was performed using continuous serum bio-marker values compared against achievement of ACR50 re-sponse at 24 weeks for adalimumab or tocilizumab and theAUC was determined

ResultsMolecular phenotypes in RA synoviumGene expression profiles of synovial tissues from 49 sub-jects with clinically diagnosed RA were subjected tounsupervised hierarchical clustering (HCL) in order toassess transcriptional heterogeneity and identify putativephenotypes of RA We identified five main clusters ofpatient samples (C1 to C5) (Figure 1A) These clusterswere visualized using principal components analysis ofthe scaled and centered data (Additional file 4 FigureS1A) and samples from clusters C1 to C4 showed differ-ences along principal components 1 and 2 whereas sam-ples from C5 were not well-separated in these two

projections We further assessed cluster robustness usingseveral additional statistical methods (discussed inAdditional file 4 Figure S1B and C) that further confirmedC5 was not well-separated and distinct from C4 We there-fore conducted all further analyses on clusters C1 to C4To characterize putative phenotypes of RA according

to their pathway composition we first identified sub-sets of genes that were specifically upregulated withineach of the four clusters using a one-versus-all ap-proach (see Methods) Each of the cluster-specific genelists was then subjected to keyword over-representationanalysis using DAVID Immune response genes wereabundant in both C1 (now termed the lymphoid pheno-type) and C2 (myeloid phenotype) with the C1 lymphoidgene sets highly restricted to B andor T lymphocyte acti-vation and differentiation immunoglobulin productionand antigen presentation together with enrichment ofcytokine signaling including the JakSTAT pathway andIL-17 signaling (Figure 1B) In contrast the gene sets up-regulated in the C2 myeloid group were also enriched forimmune function but were characterized by processes as-sociated with chemotaxis TNFα and IL-1β productionToll-like receptor and nucleotide-binding oligomerizationdomain (NOD)-like receptor signaling Fcγ-receptor-meditated phagocytosis and proliferation of mononuclearcells Cluster 3 (designated a low inflammatory phenotype)showed only enrichment for inflammatory response andwound response processes The remaining C4 clusterdesignated the fibroid phenotype was enriched for genesassociated with transforming growth factor (TGF) β sig-naling bone morphogenetic protein (BMP) signalingtogether with associated Sma Mothers Against Decapenta-plegic (SMAD) binding as well as endocytosis and cellprojection processes (Figure 1B) but lacked enrichment ofany immune system processes We further confirmed thatthe identified processes of interest were not solely drivenby a small set of recurring genes by directly comparingeach gene set identified by the DAVID analysis with eachother and observing that their overlap was generally low(Additional file 5 Figure S2) However these analyses alsosuggested certain biological processes might reflect similargene expression profiles occurring together in the samepatients for example Toll-like receptor signalingNOD-like receptor signaling and Fc-γR-mediatedphagocytosis occurred together primarily in the mye-loid group whereas processes such as antigen pro-cessing and presentation overlapped with both lymphoidgroup processes such as B and T cell activation and mye-loid group processes such as FcγR-mediated phagocytosisand mononuclear cell proliferation as might be ex-pected based upon their connected immunologicalroles Further examination of genes that were spe-cifically downregulated within each of the four clus-ters indicated the C4 fibroid cluster had significant

Groups C1 C2 C3 C4 C5

A

2

-2

Z s

core

B

C1 C2 C3 C4

-log10(adj p)

0 1 2 3 4 5 6 7

Immunoglobulin subtypeImmunoglobulin Vminusset domainlymphocyte activationcytokineminuscytokine receptor interactionNatural killer cell mediated cytotoxicityregulation of T cell activationcellular defense responseantigen processing and presentationB cellT cell receptor signalingT Helper cell surface moleculesIL-17 signalingJakminusSTAT signalingchemotaxisdefense responsepositive regulation of TNFresponse to woundingTollminuslike receptor signalingNODminuslike receptor signalingFcγ Receptorminusmediated phagocytosismononuclear cell proliferationpositive regulation of ILminus1β secretionregulation of cytokine productioninflammatory responseSMAD bindingTGFβ signalingBMP signalingenzyme-linked receptor protein signalingcell projectionendocytosis

Figure 1 Stratification of rheumatoid arthritis (RA) transcriptional heterogeneity into homogeneous molecular phenotypes(A) Two-dimensional hierarchical clustering of approximately 7000 probes (rows) representing quantile-normalized and scaled expression valuesof the top 40 most variable probe sets (variability assessed using SD) in 49 RA patients (columns) inferring five molecular subgroups of synovialtissues Patient-sample ordering and dendrogram based on agglomerative hierarchical clustering (Ward method) resulting tree used to selectpatient subgroups number of patient subgroups selected to maximize mean silhouette width and k-nearest neighbor distances (k = 5considered optimal) z-score-based color intensity scale for each probe in each sample is shown Patient samples clustering into five mainbranches are color-coded left to right (bottom of the heatmap) C1 = red (n = 8) C2 = purple (n = 14) C3 = gray (n = 16) C4 = green (n = 8)C5 = light blue (n = 3) (B) Heatmap depicting over-represented Database for Annotation Visualization and Integrated Discovery biologicalprocess categories for genes upregulated in the four largest synovial clusters Each column represents one cluster (C1 to C4) color-coordinatedas in panel A Each row corresponds to a biological process category Heatmap colors reflect log10 (adjusted P-value) from modified Fisher exact testfor categorical over-representation Annotation for each cluster based on the key biological processes is indicated BMP bone morphogenetic proteinTGF transforming growth factor SMAD Sma Mothers Against Decapentaplegic NOD nucleotide-binding oligomerization domain JAK-STAT Januskinase-signal transducer and activator of transcription


downregulation of multiple immune-system processesassociated with B cells immunoglobulins myeloidcells innate immune response including NOD-like re-ceptor signaling and chemotactic processes (Additionalfile 6 Figure S3A) In contrast the C1 cluster had sig-nificant downregulation of TGFα and Wnt signalingtogether with processes associated with mesenchymalcell proliferation proteolysis cellular transport andRNA metabolism and processing whereas both theC2 and C1 clusters had decreased representation ofprocesses associated with transcription and splicing Asobserved for the upregulated gene processes the overlap

between downregulated gene processes was also low(Additional file 6 Figure S3B)Next we assessed histological specimens derived from

the tissues used for microarray analysis for cellular com-position and the presence of cellular aggregates reflectiveof local B and T cell proliferation and lymphoid neogen-esis Representative tissue sections for each cluster werestained with cell-type-specific markers for T cells (CD3)and B cells (CD20) to assess the lymphocyte content ofsamples (Figure 2A) The results corroborated cellulardifferences observed in their respective gene-expressionprofiles Samples in the lymphoid cluster were enriched

CD3

CD20

B

CD

3

CD20

CD

45

CD90

Agg- Agg+ Agg- Agg+ Agg- Agg+ Agg- Agg+0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

A Lymphoid Myeloid Low Inflam Fibroid

C

Pat

ient

s (

)

Figure 2 Rheumatoid arthritis (RA) molecular phenotypesreflect cellular and biological differences (A)Immunohistochemical detection of T cells (CD3) and B cells (CD20)in synovial tissue sections Columns correspond to representativesections for each of the RA molecular phenotypes designated bycolor-coordinated bars on top Scales on images refer to a length of500 microns (B) Fluorescence activated cell-sorting analysis of freshsynovial tissue samples Cells were stained with CD3- and CD20- gatedby forward and side-scatter lymphocyte parameters and fluorescentintensities plotted in a scatter-plot with T cells (CD3) on the y-axis andB cells (CD20) on the x-axis (top panel) Contour-plots from the samepatients above showing macrophages (CD45+ lymphocyte-gateexclusion) along the y-axis and fibroblasts (CD90) along the x-axis(bottom panel) Samples are arranged left to right according to theirphenotype membership as in panel A (C) Bar plots of the percentagesof patient synovial tissues that contained non-aggregated (Agg-) oraggregated (Agg+) cellular infiltration as determined byimmunohistological assessment of CD3- and CD20-positive cells


for CD20-positive B cells whereas CD3-positive T cellswere present at varying levels in samples from all themajor clusters Using fluorescence-activated cell sorting(FACS) analysis of representative dissociated synoviocytesamples from each cluster (Figure 2B) we found fibro-blasts (CD45-CD90+) macrophages (CD45+CD90-) andT cells (CD3+) to varying degrees in all clusters whereasB cells (CD20+) were restricted to lymphoid and myeloid

clusters but were more abundant in lymphoid Furtherhistologic cellular aggregates reflecting proliferating B andT cells were abundant in lymphoid samples present butless abundant in myeloid and low inflammatory samplesand absent in the fibroid samples (Figure 2C)

Assessment of gene expression and gene sets in RAsynovial clustersTo further assess the underlying cellular and pathwayrepresentation of the identified RA synovial phenotypeswe examined the expression of genes with well-understoodbiological function that showed differential expressionacross the RA phenotypes (Figure 3A) The myeloidphenotype had the highest amongst the synovial sub-groups of levels of nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) pathway genesincluding TNFα IL-1β IL-1RA ICAM1 and MyD88the inflammatory chemokines CCL2 and IL-8 andgranulocyte and inflammatory macrophage lineage genessuch as S100A12 CD14 and OSCAR In contrast thelymphoid phenotype had the highest expression of B cell-and plasmablast-associated genes including CD19 CD20XBP1 immunoglobulin heavy and light chains CD38 andCXCL13 The fibroid phenotype had low or absent ex-pression of these genes and instead had elevation ofgenes associated with fibroblast and osteoclastosteoblastregulation such as FGF2 FGF9 BMP6 and TNFRSF11bosteoprotogerin In addition this phenotype had higher ex-pression of components of the Wnt and TGFβ pathwaysThe low inflammatory phenotype showed expression ofgenes associated with all of the previous phenotypes indi-cating this contains representation of all of the prior phe-notypes In addition expression of IL-6 the IL-6 receptorcomponents IL-6R and IL-6STgp130 and associated sig-naling component STAT3 was broadly observed across allphenotypes consistent with the multiple roles of the IL-6pathway in both lymphocyte and fibroblast biology [32]We further assessed biological processes associated with

the synovial phenotypes using experimentally derived gene-set modules representing a spectrum of hematopoieticlineage cells derived from specific expression in purifiedclassically activated M1 monocytes alternatively activatedM2 monocytes B cells T cells TNFα-stimulated synovialfibroblasts and angiogenesis-associated genes (see Methodsand Additional file 3 Table S1 for a list of the modulegenes) The lymphoid phenotype was enriched specificallyfor B-cell modules (Figure 3B) whereas the myeloidphenotype was enriched for inflammatory M1 monocytesand TNFα-induced modules (Figure 3D E) In contrastT-cell genes were expressed similarly in both lymphoidand myeloid phenotypes (Figure 3C) The M2 monocytemodule was expressed most highly in the low inflamma-tory phenotype (Figure 3F) while the angiogenesis modulewas highest in the fibroid phenotype and lowest in the

A LymphoidMyeloid diorbiFyrotammalfnIwoL

Sca

led

Exp

ress

ion

Scl

aed

B C

ell S

core

minus05

00

05

10

L M X F

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

L M X F

Sca

led

M2

Mon

ocyt

e S

core

minus08

minus06

minus04

minus02

00

02

04

L M X F

Sca

led

T C

ell S

core

minus05

00

05

10

15

L M X F

Sca

led

TN

Fα S

core

minus04

minus02

00

02

L M X F

Sca

led

Ang

ioge

nesi

s S

core

minus10

minus05

00

05

L M X F

B

F

C

E

D

G

minus3

minus2

minus1

0

1

2

3

TNFSF11IL6STSTAT3ANKHIL6DVL1TGFB2FZD8BMP6TNFRSF11BFGF2IL6RFGF9WNT9ADKK3CD7CD3DCXCL13SLAMF6CD19MS4A1IGJXBP1IGKCD38CD14CD300AOSCARMYD88S100A12NFKB1TNFCCL2IL8IL1BICAM1IL1RA

Figure 3 Distribution of biological process genes and gene sets across the synovial tissue phenotypes (A) Heatmap of expression ofselected genes in lymphoid (red) myeloid (purple) and fibroid (green) patient subgroups Patient-sample clusters are supervised by priorphenotype assignment and genes are distributed by unsupervised clustering (B-G) Distribution of biological processes for each synovialphenotype (L = lymphoid M =myeloid X = low inflammatory F = fibroid) was assessed using predefined gene sets to interrogate the respectivemicroarray datasets Gene sets reflecting B cells (B) T cells (C) M1 classically activated monocytes (D) genes induced by TNFα (E) M2alternatively activated monocytes (F) and angiogenesis (G) Each subgroup was compared to all other groups using the f-test and significantBenjamini-Hochberg-corrected P-values for a group compared with all other groups are indicated (P le005 P le001 P le0001) for subgroupswith positive t-statistic values


lymphoid phenotype (Figure 3G) Application of theM1-monocyte and B-cell gene sets to two additional RAsynovial datasets showed consistent differential expressionpatterns to those observed in the initial training datasetfurther indicating that these molecular axes define a largeproportion of the transcriptional heterogeneity found in

the RA synovium (Additional file 7 Figure S4) Furtherpatients with lower levels of B cell and M1 monocytes hadincreased levels of fibroid subset genes consistent withthe pattern seen in the training data set (Additionalfile 7 Figure S4B-D) Further survey of tissue sectionscharacterized by high or low levels of B lymphocytes


determined by immunohistochemistry compared with themagnitude of a B-cell gene-set score demonstratedthe correlation between histology and gene-set data(Additional file 8 Figure S5) These gene expressiondata support the notion that there are at least two in-flammatory axes of disease in the RA synovium compris-ing activation of B cells and activation of inflammatorymonocytes that are not completely overlapping whereasother synovial tissues display a low inflammatory pauci-immune phenotype with potential angiogenic osteoclastosteoblast dysregulation and fibroblast activation processesin action Consistent with lack of immune system involve-ment in the fibroid synovial phenotype we observed thatfor the patients who had available data on serological sta-tus 100 of lymphoid- and myeloid-phenotype patientswere RF-positive 75 of the low inflammatory phenotypepatients were RF-positive and in contrast the fibroidphenotype patients were RF-negative

Clinical response to targeted therapiesGiven the over-representation of myeloid and TNFα-associated gene expression in the myeloid phenotype wehypothesized that patients who displayed this inflamma-tory synovial phenotype would have the best clinical re-sponse to anti-TNFα treatment as compared with theinflammatory lymphoid phenotype To test the ability ofthese predefined synovial phenotypes to identify thera-peutic response to TNFα blockade we interrogated a pa-tient cohort synovial gene-expression dataset (GSE21537[15] a study that used the anti-TNFα agent infliximab)using pre-specified myeloid and lymphoid gene sets thatwere derived using an unbiased statistics-based approachfrom the training cohort data described in Figures 1 2and 3 (see Methods) The GSE21537 dataset used a dif-ferent non commercial microarray platform in contrastto the Affymetrix platform utilized for the training setwhich required the predefined phenotype gene sets to bemapped onto the GSE21537 microarray expression data-set Baseline gene-set scores were compared against pa-tient subgroups defined by their EULAR clinical response(good versus poor) to anti-TNFα treatment based uponimprovement in the disease activity score from 28 joints(DAS28) at 16 weeks Strikingly we observed that baselineexpression of the myeloid gene set was significantly higherin patients with good EULAR response compared to nonresponders (P = 0011 Figure 4A) In contrast the lymph-oid gene set despite also marking inflammatory synovialprocesses did not show association with clinical outcome(P = 026 Figure 4B) and the fibroid phenotype gene setwas also unaltered between good and poor responders(P gt05 Figure 4C)These results were further confirmed by additional ana-

lysis of this dataset using the previously utilized gene setswhich showed that the pretreatment biological process

most strongly associated with good versus poor responseto anti-TNFα therapy was classically M1 activated M1monocytes (P = 0006 Figure 4D) whereas in contrastneither the B-cell or T-cell gene sets showed no signifi-cant association with response (Figure 4E and F P = 018and P = 09 respectively) We further observed trendsin association of pretreatment levels of M2 alterna-tively activated monocytes (P = 0054 Additional file 9Figure S6A) and TNFa-treated synovial fibroblasts (P= 008Additional file 9 Figure S6B) whereas angiogenesis pro-cesses were significantly associated with good response(P = 0018 Additional file 9 Figure S6C) In addition weconducted ROC analysis of the gene sets versus EULARresponse and calculation of the AUC revealed that con-sistent with the above findings the myeloid and M1 clas-sically activated monocyte gene sets produced the largestAUCs (065 Additional file 10 Figure S7A and 077Figure S7D respectively) These data indicate that ap-plication of predefined molecular synovial phenotypesnamely the myeloid phenotype and associated M1-activated monocytes has the potential to enrich for re-sponders to anti-TNFα therapy and that pretreatmentlevels of these biological processes were most stronglyassociated with anti-TNFα therapeutic outcome

Derivation of serum biomarkers from differential synovialgene expressionGiven the observation that synovial heterogeneity affectstreatment outcome to anti-TNFα therapy we investigatedwhether we could identify differential gene expression inthe inflammatory synovial phenotypes that might bereflected as circulating biomarkers in peripheral bloodUsing the F-test on the original synovial gene-expressiondataset we identified genes that differed between the syn-ovial phenotypes and then identified genes that best dif-ferentiated one synovial phenotype compared with allothers using the pairwise t-test between all pairs of groups(P lt0001 multiple-hypothesis test correction using theBenjamini-Hochberg method) and further assessed genesencoding potential soluble biomarkers with a positivet-statistic value in each phenotype We focused on twobiomarkers ICAM1 differentially expressed in the mye-loid phenotype (Figure 5A) and CXCL13 enriched in thelymphoid phenotype (Figure 5B)We developed immunoassays to determine levels of

circulating soluble ICAM1 (sICAM1) and CXCL13 inserum and tested pretreatment samples from patientswith active RA enrolled in the ADACTA trial (below)We observed that both serum biomarkers were signifi-cantly higher in disease compared with samples from non-disease control donors (Figure 5C D) but importantly wereonly weakly correlated with each other (Spearman P lt033Figure 5E) suggesting they are reflective of different inflam-matory immune processes

Sca

led

B C

ell S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

Mye

loid

Sco

re

minus10

minus05

00

05

10

Good Moderate PoorS

cale

d Ly

mph

oid

Sco

re

minus10

minus05

00

05

10

15

Good Moderate Poor

Sca

led

Fib

roid

Sco

re

minus10

minus05

00

05

10

Good Moderate Poor

A B C

D E F

Sca

led

T C

ell

Sco

reminus10

minus05

00

05

10

Good Moderate Poor

Figure 4 Pretreatment magnitude of gene sets derived from the synovial myeloid phenotype and classically activated monocytescorrelates with clinical response to anti-TNFα (infliximab) therapy Analysis of synovial tissue microarray data from 62 rheumatoid arthritispatients in GSE21537 prior to initiation of infliximab (anti-TNFα therapy) Scores for gene sets for phenotypes defined from the Michigan cohorttraining data as well as gene sets derived from purified immune cell lineages (see Methods) were calculated from the GSE21537 data andcompared against anti-TNFα clinical outcome at 16 weeks as defined by European League Against Rheumatism (EULAR) response criteria asassigned in GSE21537 Scores versus EULAR response are plotted for the synovial myeloid phenotype (A) lymphoid phenotype (B) fibroidphenotype (C) as well as classically activated M1 monocytes (D) B cells (E) and T cells (F) Statistical significance for good compared with poorEULAR response for the level of each gene-set module was calculated based upon the t-statistic ( = P le005 P le001)


sICAM1 and CXCL13 define RA subpopulations withdifferential clinical outcomes to adalimumab (anti-TNFαcompared with tocilizumab (anti-IL-6R) therapyWe finally assessed whether baseline levels of sICAM1and CXCL13 were differentially associated with subsequenttreatment outcome to adalimumab compared with toci-lizumab as we hypothesized based upon the previous re-sults that a population with elevated levels of a myeloidbiomarker have elevated clinical response to anti-TNFαtherapy but that elevation of a lymphoid marker wouldnot We utilized pretreatment samples from the ADACTAtrial a randomized double blind controlled phase-4 headto head study of tocilizumab (a humanized monoclonalantibody that binds to membrane-bound and soluble formsof the human IL-6 receptor) monotherapy compared withadalimumab (a fully human monoclonal antibody againstTNFα) monotherapy in methotrexate-intolerant patientswith active RA [30] This trial was notable as it allowed aninitial assessment of biomarker-defined populations within

the same trial against two different targeted therapiesAs this was a post hoc exploratory analysis without pre-specified biomarker thresholds we first assessed each bio-marker individually using the median as a cutoff to definebiomarker-low and biomarker-high subpopulationsAn additional motivation to employ categorical analysis

of predictor variables stemmed from the presence of left-censored (below the lower limit of quantification (LLOQ))observations for baseline levels of CXCL13 where 96(19 of 198 samples) were observed to have values lowerthan the LLOQ and categorical analysis was used to ac-commodate left-censored data and avoided potential biasthat may result from imputation of left-censored data inparametric analyses We initially observed that there was adifferential relationship between clinical outcome to eachtherapy and baseline biomarker levels patient populationswith lower sICAM1 levels the myeloid phenotype bio-marker or higher CXCL13 levels the lymphoid phenotypemarker were associated with lower likelihood as defined

BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600

sICAM (ngml)NC RANC RA

plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F

Figure 5 Assessment of serum biomarkers extrapolated from lymphoid and myeloid synovial phenotype gene expression in thesynovial transcriptome training dataset Intercellular adhesion molecule 1 (ICAM1) (A) and C-X-C motif chemokine 13 (CXCL13) (B) genesare expressed at highest levels in the myeloid (M) and lymphoid (L) phenotypes respectively Array probes for each transcript were comparedacross all groups using the f-test and in both cases Benjamini-Hochberg-corrected P lt 0001 X = low inflammatory phenotype and F = fibroidphenotype Soluble (s)ICAM1 (C) and CXCL13 (D) are elevated in serum samples from rheumatoid arthritis (RA) patients (ADACTA trial) ascompared with normal control (NC) serum P-values derived from the Wilcoxon test are indicated (E) Serum sICAM1 and CXCL13 levels wereonly weakly correlated in RA (ρ lt 033 Spearman rank correlation coefficient)


by the odds ratio of week-24 ACR50 response to adalimu-mab compared with tocilizumab (Figure 6A) Given thesereciprocal associations we next looked at the two bio-markers in combination both using the biomarker medianvalues for each as cutoffs as well as continuous biomarkervalues These analyses further indicated that heteroge-neous treatment effects were present as the patient popu-lation with high sICAM1 but low CXCL13 had higherlikelihood of ACR50 response to adalimumab comparedwith tocilizumab whereas conversely there was a higherlikelihood of ACR50 response to tocilizumab comparedwith adalimumab in patients with high CXCL13 but lowsICAM1 (Figure 6B) Importantly the differences in rela-tive treatment effectiveness among biomarker-definedsubgroups were borne out by contrasting absolute ACRresponses among both treatment arms (Figure 6C D) asopposed to heterogeneous responses observed only in asingle treatment arm Assessing each drug treatment armseparately using week-24 ACR20 ACR50 and ACR70response-rates across biomarker median-defined patientsubgroups showed that sICAM1-highCXCL13-low pa-tients had the highest clinical responses from adalimumabtreatment (Figure 6C E) compared to the other patientsin the treatment arm (ACR20 Δ = 46 P = 0005 ACR50

Δ = 29 P = 005 and ACR70 Δ = 16 P-value not sig-nificant (Fisher exact test)) Conversely the sICAM1-lowCXCL13-high patients had the highest responses to toci-lizumab (Figure 6D E ACR20 Δ = 20 P-value not sig-nificant ACR50 Δ = 49 P = 0004 and ACR70 Δ = 45P = 0004 (Fisher exact test)) In addition the remainingbiomarker-defined subgroups (highhigh and lowlow) ex-hibited intermediate ACR50 response rates for both ther-apies (Figure 6E) These differences were also consistentin the trends for change in DAS28-erythrocyte sedimenta-tion rate (ESR) (plusmn standard error) at 24 weeks for ada-limumab (-23 plusmn 037 for sICAM1-highCXCL13-low patientscompared with -11 plusmn 033 for sICAM1-lowCXCL13-highpatients) and tocilizumab (-36 plusmn 032 for sICAM1-lowCXCL13-high patients compared with -32 plusmn 037 forsICAM1-highCXCL13-low patients) The biomarker-defined subgroup efficacy results for each therapyincluding odds ratios for ACR50 response are sum-marized in Table 1sICAM1 and CXCL13 biomarker populations were de-

fined by cutoffs determined by the median values Weexplored the heterogeneity of the relative treatment ef-fect using alternative biomarker cutoffs using STEPPanalysis Assessment of individual biomarkers showed

001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

ACR response rates () per biomarker subgroupACR20 ACR50 ACR70

0

20

40

60

80

AC

R50

Res

pons

e R

ate

()

Adalimumab (anti-TNFα)Tocilizumab (anti-IL-6R)

sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2

Figure 6 (See legend on next page)


(See figure on previous page)Figure 6 Lymphoid (C-X-C motif chemokine 13 (CXCL13)) and myeloid (soluble intercellular adhesion molecule 1 (sICAM1)) serumbiomarkers define rheumatoid arthritis patient subgroups with differential clinical response to anti-TNFα (adalimumab) compared withanti-IL-6R (tocilizumab) in the ADACTA trial Relative treatment effectiveness (week-24 American College of Rheumatology (ACR)50 response)of adalimumab compared with tocilizumab was assessed by logistic regression for (A) each individual biomarker and (B) biomarker combination-defined subgroups using their respective medians as cutoffs (see Methods) Relative treatment effectiveness for adalimumab versus tocilizumab isrepresented by odds ratio and 95 CI for ACR50 response Week-24 ACR20 (gray) ACR50 (green) and ACR70 (purple) response rates () perbiomarker-defined subgroup are represented by radial plot for adalimumab (C) and tocilizumab (D) treatment arms The direction of each radialline corresponds to a biomarker subgroup as follows sICAM1 low (bottom) and high (top) CXCL13 low (left) and high (right) Low and highdesignations refer to biomarker values above and below their respective medians Distance from radial plot center indicates response rateSummary of week-24 ACR50 response rates for sICAM1-highCXCL13-low sICAM1-highCXCL13-high sICAM1-lowCXCL13-low and sICAM1-lowCXCL13-high ADACTA RA patients (E) The treatment-effect deltas between sICAM1-highCXCL13-low and sICAM1-lowCXCL13-high patientgroups are indicated for both adalimumab and tocilizumab


that increasing levels of sICAM1 were associated withincreasing likelihood of ACR50 response to adalimumabversus tocilizumab (Additional file 11 Figure S8A) butincreasing levels of CXCL13 were associated with decreas-ing ACR50 response to adalimumab versus tocilizumab(Additional file 11 Figure S8B) Further examination of con-tinuous levels of both biomarkers using two-dimensionalSTEPP analysis also showed the highest likelihood ofACR50 response to adalimumab versus tocilizumab in pa-tients with the highest levels of sICAM1 but the lowestlevels of CXCL13 (Additional file 11 Figure S8C) whereasconversely the lowest likelihood of response to adalimu-mab versus tocilizumab was observed in the patient popu-lation with the lowest sICAM1 and highest CXCL13levels These data suggest that further differentiation ofrelative treatment effect may be observed using optimizedcutoffs as determined in a prospective studyFinally ROC analysis was performed to assess the pre-

dictive ability for ACR50 response of these two biomarkerson an individual patient basis sICAM1 and CXCL13showed only modest predictive ability for adalimumab ortocilizumab on an individual patient basis based upontheir respective AUCs (057 and 06 respectively Additionalfile 12 Figure S9A D) whereas assessment of the two

Table 1 Summary of baseline biomarker-defined subgroup ef

Biomarker subset number ADA ACR20 () ADA ACR50 () A

sICAM1highCXCL13low (26) 73 42

sICAM1lowCXCL13high (15) 27 13

sICAM1highCXCL13high (32) 50 28

sICAM1lowCXCL13low (33) 52 24

Biomarker subset number TCZ ACR20 () TCZ ACR50 () T





Data are shown for American College of Rheumatology (ACR) 20 50 and 70 responsedimentation rate (ESR) (plusmn standard error SE) and odds ratio with 95 CI for ACR

biomarkers in combination showed slight increases in therespective AUCs (Additional file 12 Figure S9C D E F)In totality these data illustrate the concept that mye-

loid and lymphoid phenotype-derived circulating bio-markers can together define RA patient subpopulationsthat show differential clinical response to therapies di-rected at different targets and that myeloid-dominantpatient populations with high levels of sICAM1 and lowlevels of CXCL13 had the most robust response to anti-TNFα therapy

DiscussionIn this report we describe the presence of major cellularand molecular heterogeneity in RA synovial tissue char-acterized by two inflammatory phenotypes dominatedby B cells and plasmablasts (lymphoid) and inflamma-tory macrophages (myeloid) as well as a low inflammatorypauci-immune phenotype show that elevation of the mye-loid but not lymphoid axis in synovial tissue is signifi-cantly associated with good clinical outcome to anti-TNFαtherapy and finally show that two systemic biomarkerschosen based on their differential tissue expression be-tween the inflammatory phenotypes CXCL13 for lymph-oid and sICAM1 for myeloid together define RA patient

ficacy at 24 weeks in the ADACTA trial

DA ACR70 () ADA ΔDAS28-ESR (plusmnSE) ACR50 odds ratio ADAversus TCZ (95 CI)

23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)

CZ ACR70 () TCZ ΔDAS28-ESR (plusmnSE) ACR50 odds ratio TCZvs ADA (95 CI)

7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)

se rates change in disease activity score in 28 joints (DAS28)-erythrocyte50 response ADA adalimumab (anti-TNFα) TCZ tocilizumab (anti-IL-6R)


subpopulations with differential clinical response to anti-TNFα compared with anti-IL-6R therapiesThe concept that important heterogeneity exists in RA

synovial tissue both at a histological as well as at a mo-lecular level has been previously illustrated by severalseminal studies [81033] which showed differential pres-ence of histological synovial aggregates and diffuse syn-ovial inflammation as well as differential gene expressionacross RA synovial samples The objective of the currentstudy was to test the idea that heterogeneous RA synovialtissues can be assigned to subgroups that share commonpatterns of gene expression have different associated sys-temic biomarkers and that might respond differentiallyto therapy Thus we employed an analysis strategy thatqueried independently the questions of molecular hetero-geneity and response heterogeneity First we assessedmolecular heterogeneity of RA synovium independentof treatment response and validated proposed pheno-types using various molecular techniques and externalpatient cohorts We next observed that core biologicalmodules as defined using pathway analysis designatedlymphoid (B cell- and plasmablast-dominated) myeloid(macrophage and NF-κB process dominated) and fibroid(comprising hyperplastic but pauci-immune tissues) couldbe surveyed across multiple RA patient synovial tissuecohorts to identify reproducible RA phenotypes Import-antly the dominant biology associated with each geneexpression-defined subset was consistent with histologicaland flow cytometry assessment of synovial tissue wherethe lymphoid subset was associated with presence of histo-logical aggregates and the myeloid subset with more dif-fuse immune infiltration while the fibroid subset had littleimmune infiltration and complete absence of aggregatesFurther survey of tissue sections characterized by highor low levels of B lymphocytes determined by immuno-histochemistry correlated with the magnitude of a B cellgene-set score We also observed the presence of a low in-flammatory phenotype indicating that synovial hetero-geneity exists as a continuum of dysregulated biologicalprocesses rather than absolutely discrete subsets of dis-ease We did not observe differences in therapeutic usage(methotrexate anti-TNFα agents steroids) between pa-tients with different synovial phenotypes where these datawere available (data not shown) However we did notethat for the patients with data available RF serologicalpositivity was restricted to the lymphoid myeloid and amajority of the low inflammatory phenotype patientsThese data are consistent with previously observed geneexpression heterogeneity in RA synovial tissue suggestingthere are both inflammatory and non inflammatory syn-ovial subgroups in RA We further observed presence ofpatients with low or high inflammatory phenotypes basedupon M1-activated monocytes B cell and fibroid gene setsin two additional datasets although the M1 and B cell

gene sets were not as divergent as observed in the originaltraining set Reasons for this could include introduction ofadditional noise and loss of sensitivity due to the differentplatform used in the GSE21537 dataset resulting in loss ofdata due to missing or non-mapping probes as comparedwith the Affymetrix platform as well as differences in thepatient populations as there were higher levels of fibroidgene-set scores in both patient cohorts compared with thetraining dataset meaning decreased representation of pa-tients in the highly inflammatory subgroupsIndeed it has been clearly shown that patients with high

levels of expression of inflammatory genes in the synoviumhave higher levels of systemic inflammation including C-reactive protein levels ESRs and platelet counts as well asa shorter duration of disease as compared to patients withlow synovial inflammation [34] Further absence of signifi-cant synovial inflammation has been linked to decreasedpresence of anti-citrullinated protein antibodies [35] Con-sistent with this finding of a pauci-immune phenotypeof RA patients with lower levels of both synovial andsystemic inflammation have been shown to have lowerdrug-response rates to both B-cell depletion therapy andanti-TNFα [36-38]We then assessed whether the inflammatory biological

modules would be differentially informative for predictingthe outcome of response to anti-TNFα therapy throughanalysis of a large and well-defined external dataset Strik-ingly patients with high pretreatment expression of genesdefined in the myeloid phenotype and M1 classically acti-vated monocytes but not high levels of lymphoid subsetor B-cell genes showed a greater 16-week good EULARresponse to infliximab treatment This is consistent withthe observation that inflammatory M1 macrophages akey lineage involved in production of TNFα as well asexpression of TNFα itself along with IL-1β and NF-κB-associated processes are preferentially increased in themyeloid phenotype compared with all of the others Fur-ther other studies have consistently concluded that baselinelevels of synovial macrophages and TNFα gene expressionare correlated with response [1339] suggesting the pres-ence of TNFα-secreting classically activated monocytesand macrophages are important for clinical outcomeHowever the EULAR moderate responders had a widerange of values for both the myeloid and M1 genes whichsuggest that other factors will contribute to determiningtreatment outcome with anti-TNFα agents In contrast alarge histological study demonstrated that RA patientswith high levels of synovial lymphoid neogenesis (LN)comprising highly organized BT cell aggregates demon-strated resistance to anti-TNFα therapy and good clinicaloutcome in these patients was accompanied with reversalof LN [40] Consistent with this we observed that thepresence of the lymphoid phenotype was not a predictorof response to anti-TNFα despite being associated with


the presence of synovial inflammation and histological ag-gregates In sum these data suggest that simply the pres-ence of inflammation alone is insufficient to predictclinical outcome to anti-TNFα treatment and rather thatsub-phenotypes of synovitis show differential clinicalbenefit with the lymphoid phenotype showing greater re-sistance to anti-TNFα as compared with the myeloidphenotype perhaps due in part to the presence of othermajor processes driving synovitis including production ofother inflammatory mediators LN and robust antigenpresentation by autoreactive B cells It is also noteworthythat we observed an association between pretreatment ex-pression of genes associated with angiogenesis and clinicalresponse to anti-TNFα suggesting that the presence ofsynovial neoangiogenesis may also contribute to favorableoutcome to blockade of TNFαNext we hypothesized that the biological processes

underlying the RA phenotypes might allow for rationalserum protein biomarker selection to prospectively iden-tify patient populations prior to starting a targeted therapyAs synovial tissue is not readily available for prospectiveassessment prior to initiation of therapy systemic circulat-ing biomarkers have greater potential utility although theywill likely integrate the activity of specific biological path-ways in multiple tissues including the secondary lymphoidsystem in addition to synovial tissue We assessed candi-dates that were differentially expressed in the inflamma-tory lymphoid and myeloid subsets using a statisticalranking and looked for markers that were strongly ele-vated in RA serum as compared with serum from nondisease control donors Two markers that fulfilled thesecriteria were soluble ICAM1 (myeloid) and CXCL13(lymphoid) ICAM1 an adhesion molecule that bindsto LFA-1 is a gene that is strongly regulated by NF-κB signaling and is upregulated on a variety of celltypes in response to TNFα signaling including synovialfibroblasts and especially vascular endothelial cells bothof which are highly represented in the inflammatoryrheumatoid synovium [4142] sICAM1 is shed fromthe cell membrane by proteolytic cleavage CXCL13 isa B cell chemoattractant that is highly expressed byfollicular dendritic cells in secondary lymphoid tissueand ectopic germinal centers and is induced by LTαLTβRsignaling [43] Further a recent report of a small synovialbiopsy study of RA patients undergoing rituximab therapyshowed a correlation between synovial tissue expressionof CXCL13 and levels of CXCL13 protein in the serum(r = 06) [44] that suggests CXCL13 expression in therheumatoid synovium is a major source of serum CXCL13Synovial and serum levels of CXCL13 have also recentlybeen linked with radiological joint destruction in RA pa-tients [45] which argues that this gene and by associationthe lymphoid synovial phenotype is linked with progres-sive and destructive RA pathogenesis In contrast to our

knowledge no reports have been made to date that havedirectly compared sICAM1 levels in serum with ICAM1gene expression in synovial tissue and we have not beenable to conduct such an analysis in this study due toincomplete matching serum samples Analysis of serumsamples from the ADACTA adalimumab (anti-TNFα)compared with tocilizumab (anti-IL-6R) trial facilitated anassessment of these biomarkers in an inflammatory RApopulation that not only allowed a direct comparison ofclinical response to different targeted therapies within oneclinical study but also avoided confounding effects of con-comitant immunosuppression from background metho-trexate as this study was conducted using both therapeuticagents as monotherapy [30] Consistent with our model ofdifferent inflammatory axes being present in RA we notedthat although both sICAM1 (myeloid) and CXCL13(lymphoid) were significantly elevated in disease comparedwith control samples they were only weakly correlated toeach other Further we noted that patients with high pre-treatment serum sICAM1 levels and decreased CXCL13levels (high myeloid and low lymphoid activity) had in-creased ACR50 and ACR70 response rates and decreasedDAS28-ESR scores to anti-TNFα therapy compared withanti-IL-6R therapy whereas conversely patients with highCXCL13 and decreased sICAM1 levels had preferential re-sponse to anti-IL-6R compared with anti-TNFα therapyWe did note differences in the magnitude of the differ-ences between ACR50 response rates and changes inDAS28-ESR between the biomarker-defined populations inthe tocilizumab arm where the changes in DAS28 wereconsistent but smaller than those observed for ACR50These differences could not be accounted for by one com-ponent of the response instrument for example ESR orswollen-joint count and are likely due more to differ-ences in precision between the two instruments Theseresults are consistent with the previous data showing thatpatients with elevation of the myeloid inflammatory axishad robust responses to anti-TNFα drugs and furtheremphasize that within an inflammatory RA populationthere are patient subsets that subsequently have differen-tial clinical outcomes to different targeted therapiesWhat underlying biological basis could explain why

blockade of the IL-6 pathway causes robust clinical re-sponses in a different patient population to that respond-ing to anti-TNFα blockade Although IL-6 has long beenappreciated as a key inflammatory cytokine important inthe pathogenesis of RA as well as other inflammatory dis-eases [32] its biology and expression are not completelyoverlapping with that of TNFα Our synovial tissue gene-expression data have shown that although TNFα isstrongly associated with the myeloid phenotype andactivity of classically activated myeloid cells and NF-κB pathway activity IL-6 its receptors IL-6R and IL-6STgp130 and the key IL-6-associated TF STAT3


are more broadly expressed across the lymphoid andlow inflammatory synovial subsets (Figure 3A) and are nothighly correlated with TNFα expression or restricted tothe myeloid phenotype Indeed IL-6 can be induced in avariety of cell lineages exposed to multiple inflammatorystimuli in the joint including synovial fibroblasts them-selves [3246] Further the IL-6IL-6R pathway signalsusing the JAKSTAT pathway in contrast to the canonicalNF-κB signaling predominantly utilized by TNFα [47] andplays a key role in inducing B cells to differentiate toantibody-secreting cells Importantly anti-IL-6R therapyhas been shown to be effective in patients who are refrac-tory to anti-TNFα therapies [48] Thus it is conceivablethat the IL-6IL-6R pathway is highly involved with thedriving synovitis in the B-cell-dominant lymphoid axis aswell as potentially similarly important in driving synovitisin the low inflammatory subset whereas in contrastwithin the activated monocyte-dominated myeloid axisthe TNFα pathway is dominant in driving synovitis suchthat blockade of IL-6 signaling is less effective Whilstintriguing and consistent with the biological hypothesesdeveloped based upon our synovial tissue analyses thefindings described here represent only an initial testing ofthe sICAM1CXCL13 biomarker hypothesis without apredefined cutoff for the analysis hence our utilization ofthe median as the cutoff for this analysis and the statis-tical power was limited by available patient numbers andmultiple testing issues Furthermore analysis of these bio-markers on an individual patient basis using ROC analysisshowed that they have only modest predictive abilityfor ACR50 outcome to adalimumab or tocilizumab at24 weeks Therefore although the biomarkers describedhere demonstrate the presence of populations of RA pa-tients with differential clinical response to targeted therap-ies they do not presently have strong clinical utility fordecision-making for individual patients Improvement ofindividual patient predictive-ability might be achieved byincorporation of additional biomarkers into a predictivemodel that could be subjected to rigorous confirmatorystudies in larger patient cohorts treated with anti-TNFαand anti-IL-6IL-6R blocking agents including combin-ation treatment with methotrexate with incorporation ofprespecified cutoff values in the analysis plan Indeed thetwo-dimensional STEPP analysis performed in this studysuggested that altering the biomarker threshold cutoffs forboth sICAM1 and CXCL13 could yield greater efficacydifferentials for ACR50 response rates between adalimu-mab and tocilizumab than those achieved by using theirrespective mediansAdditional limitations of this study include limited avail-

ability of clinical data in the RA cohort used for the initialgene-signature discovery owing to the retrospective natureof interrogation of clinical chart data after sample collec-tion from joint surgery and a lack of consent for chart

review in some cases In particular there were incompleteor missing data for serological autoantibody status for RFor anti-citrullinated protein antibodies Also the RA pa-tient population studied for synovial gene expression rep-resents late-stage disease where patients received jointsurgery to correct deformity replace joints or managepain This study also does not address the presence andstability of synovial phenotypes longitudinally from earlyto late-stage disease and with respect to development ofbone erosion Finally in the current study we have not ap-plied an exhaustive investigation of all the potential serumbiomarkers that may correlate with synovial subtypes inpart due to the desire to minimize multiple testing issuesdue to the limited number of anti-TNFα-treated patientsamples available for biomarker analysis These importantquestions are being addressed in a series of follow-up pro-spective studies

ConclusionsUtilizing genome-wide expression analysis of synovial tis-sues from a large RA cohort we have defined distinct mo-lecular and cellular phenotypes that reflect the considerableheterogeneity present in the RA synovium In particulartwo distinct inflammatory axes emerge from this analysisone dominated by B cells and the other dominated by in-flammatory macrophages and NF-κB-activating cytokinessuch as TNFα It is important to point out that these cellu-lar and molecular signatures as well as the RA patientsrepresent a continuous rather than a discrete distributionas is evident from the presence of lower inflammatory pa-tients with intermediate molecular characteristics betweenthese polar phenotypes Analysis of respective gene-setmodules and serum biomarkers suggest differential clinicalresponse to anti-TNFα and anti-IL6R therapy is dependentin part on the presence of these inflammatory axes A fur-ther subgroup of patients presented with a pauci-immunephenotype lacking major B cell or macrophage infiltrationand may reflect a distinct subgroup of patients These syn-ovial phenotypes explain some of the underlying clinicaland drug response heterogeneity in RA and identifying andstratifying patients prospectively with respect to their syn-ovial phenotype for example by using blood biomarkersmay be important in making therapeutic decisions for tar-geting therapies Such considerations are also likely to bevery important for clinical trial design for new therapies toselect patients prospectively for increased clinical responserates and for the design of clinical studies to differentiatetargeted therapies with different mechanisms of action

Additional files

Lists of the Database for Annotation Visualizationand Integrated Discovery (DAVID) biological processes genesrepresented within the upregulated genes in the synovial

Additional file 1


subgroups For each gene we report differential gene expressionbetween each group and all other samples We provide the t-statisticvalues (positive or negative) with associated P-values for each groupL = lymphoid M =myeloid X = low inflammatory F = fibroid

Lists of the Database for Annotation Visualizationand Integrated Discovery (DAVID) biological process genesrepresented within the downregulated genes in the synovialsubgroups For each gene we report differential gene expressionbetween each group and all other samples We provide the t-statisticvalues (positive or negative) with associated P-values for each groupL = lymphoid M =myeloid X = low inflammatory F = fibroid

Table S1 List of genes utilized in gene setenrichment analyses

Figure S1 Assessment of robustness of synovialgene expression heterogeneity (A) Principal component analysisshowing the first (x-axis) and second (y-axis) components of variationover approximately 7000 probes and 49 patients using the prcompR-function on quantile-normalized expression data Each patient tissue iscolor-coded according to the groupings in Figure 1A and groupingcircles have been added for visual clarity (B) Re-sampling analysis usingpartitioning around medoids (PAM) analysis of approximately 7000probes 49 patients and 5 predefined clusters of tissue samples (k = 5)Heatmap colors represent the frequency with which a pair of samplesare found in the same cluster and are represented as a percentageof the total number of samplings in which the pair was observed(C) Assessment of cluster robustness via determination of silhouettewidth of approximately 7000 clustered probes from the 49 patientsAverage silhouette widths for each of the five clusters are indicated

Figure S2 Assessment of overlap between biologicalprocess gene-sets utilized by the Database for Annotation Visualizationand Integrated Discovery (DAVID) pathway analysis tool for unregulatedgenes in each of the four synovial clusters defined in Figure 1A Theoverlap of genes shared by gene sets are illustrated using a heatmapwhere each value represents the proportion of genes from the categoryon the y-axis that are in common with the corresponding gene set onthe x axis (indicated by the color bar 0 = 0 1 = 100) The matrix is notsymmetrical because the size of the gene sets is not constant

Figure S3 (A) Heatmap visualization of processesenriched in downregulated genes in each of the four synovial clustersdefined in Figure 1A using the Database for Annotation Visualization andIntegrated Discovery (DAVID) pathway analysis tool Colors refer tostatistical significance of processes to each cluster (B) Assessment ofoverlap between biological process gene sets utilized by the DAVIDpathway analysis tool for downregulated genes in each of the foursynovial clusters defined in Figure 1A The overlap of genes shared bygene sets are illustrated using a heatmap where each value representsthe proportion of genes from the category on the y-axis that are incommon with the corresponding gene set on the x-axis (indicated bythe color bar 0 = 0 1 = 100) The matrix is not symmetrical becausethe size of the gene sets is not constant

Figure S4 B cell M1 classically activated monocyteand fibroid gene modules capture synovial tissue transcriptionalheterogeneity in additional rheumatoid arthritis (RA) patient cohorts(A) Scatter plot of the training cohort of 49 patient synovial samplesprojected in gene set space of the B cell (x-axis) and M1 monocyte(y-axis) biological modules Samples are colored according to theircluster assignments in Figure 1 (red = lymphoid purple =myeloidgreen = fibroid grey = low inflammatory) Filled circles indicate sampleswith histologic aggregates and empty circles indicate samples lackingaggregates Scatter plot of the same 49 RA patients projected in gene setspace of the B cell (x-axis) and M1 monocyte (y-axis) biological modulesand samples are also colored according to their respective fibroid geneset scores as indicated by the color bar (C) Scatter plot of 33 previouslyunanalyzed patient samples from a parallel Michigan RA cohort projectedin gene-set space of the B cell (x-axis) and M1 monocyte (y-axis)biological modules Samples are colored according to their respectivefibroid gene-set scores as indicated by the color bar (D) Scatter plot of a

Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7

publicly available cohort of 62 RA histologically characterized patients(GSE21537) projected in gene-set space of the B cell (x-axis) and M1monocyte (y-axis) biological modules Samples are colored according totheir respective fibroid gene-set scores as indicated by the color bar

Figure S5 CD20 Immunohistochemistry (IHC)correlates with B cell gene-set score in a replication rheumatoid arthritis(RA) patient cohort Representative CD20 IHC (brown staining) is shownfor synovial samples with a high or low B cell gene-set score with low(A B respectively) and high (C D respectively) magnification B cellgene-set scores were also plotted against CD20 IHC scores and theP-value for Spearman rank correlation coefficient is indicated (E)

Figure S6 Association of pretreatment synovialgene-set scores with good versus poor European League AgainstRheumatism (EULAR) response to anti-TNFα (infliximab) therapy at 16weeks in the GSE21537 synovial expression dataset Statistical significancefor good compared with poor response for the level of each gene-setmodule was calculated based upon the t-statistic Scaled gene-set scoresfor M2 alternatively activated monocytes (A) (P = 0054) TNFα-stimulatedfibroblast-like synoviocytes (B) (P = 008) and angiogenesis (C) (P = 002)marked with asterisk) are plotted against 16-week EULAR response

Figure S7 Receiver-operating-characteristic (ROC)curves to assess the ability of pretreatment synovial phenotypes definedby scaled gene-set scores to differentiate between good versus poorEuropean League Against Rheumatism (EULAR) response to anti-TNFα(infliximab) therapy at 16 weeks in the GSE21537 synovial expressiondataset ROC curves were generated for the myeloid (A) lymphoid(B) and fibroid (C) phenotypes and also for gene sets reflective of M1classically-activated monocytes (D) B cells (E) and T cells (F) Area underthe ROC curve (AUC) is indicated for each plot

Figure S8 Biomarker subpopulation treatmenteffect pattern plot (STEPP) analysis of the ADalimumab ACTemrA(ADACTA) trial Assessment of individual biomarkers compared withtreatment effect One-dimensional STEPP analysis of week-24 AmericanCollege of Rheumatology (ACR) 50 relative treatment effectiveness ofadalimumab compared with tocilizumab for the serum markers solubleintercellular adhesion molecule 1 (sICAM1) (A) and C-X-C motifchemokine 13 (CXCL13) (B) respectively in the ADACTA trial Week-24ACR50 odds ratios are shown in solid blue and 95 CIs as accompanyingdashed lines The x-axes correspond to the subgroup of subjects whosebaseline biomarker levels were within 20 percentiles below and abovethe indicated subpopulation median with actual values (pgml) inparentheses The dotted horizontal line indicates equivalent relativetreatment effect (C) Two-dimensional STEPP analysis for sICAM1 andCXCL13 Each cell of the heatmap corresponds to a subgroup of subjectswhose baseline biomarker levels were within 25 percentiles below andabove the indicated subpopulation median as defined by eachbiomarker Concentrations of each biomarker at the indicated percentageare in parentheses in plot margins Heatmap colors indicate odds ratio(95 CI in brackets) from logistic regression corresponding to outcomesfor adalimumab versus tocilizumab Counts of subjects in each treatmentarm for each subgroup are indicated as n = (tocilizumab)(adalimumab)

Figure S9 Receiver-operating-characteristic (ROC)curves to assess the ability of pretreatment C-X-C motif chemokine 13(CXCL13) and soluble intercellular adhesion molecule 1 (sICAM1) todifferentiate for clinical response in the ADalimumab ACTemrA (ADACTA)trial biomarker population ROC curves were generated for sICAM1 versusachievement of an American College of Rheumatology (ACR)50 responseat week 24 for adalimumab in all-comers (A) CXCL13-high (B) andCXCL13-low patient subsets (C) and for CXCL13 versus achievement ofan ACR50 response at week 24 for tocilizumab in all-comers (D)sICAM1-high (E) and sICAM1-low patient subsets (F) Biomarker high andlow designations were made using their respective medians as the cutoffArea under the ROC curve (AUC) is indicated for each plot

Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12

AbbreviationsACR American College of Rheumatology ADACTA ADalimumab ACTemrAAgg aggregated AUC area under the receiver-operating characteristic curveBMP bone morphogenetic protein CXCL13 C-X-C motif chemokine 13


DAB 33prime-diaminobenzidine DAS28 disease activity score (from 28 joints)DAVID Database for Annotation Visualization and Integrated DiscoveryDMARD disease-modifying anti-rheumatic drug ESR erythrocytesedimentation rate EULAR European League Against RheumatismFACS fluorescence-activated cell sorting FDR false discovery rateHCL hierarchical clustering IFN interferon IL interleukin JAK Janus kinaseLLOQ lower limit of quantification LN lymphoid neogenesisLPS lipopolysaccharide MSigDB Molecular Signatures DataBaseNF-κB nuclear factor kappa-light-chain-enhancer of activated B cells NS notsignificant NSAID non-steroidal anti-inflammatory drug PAM partitioningaround medoids RA rheumatoid arthritis RF rheumatoid factor RMA robustmultichip average ROC receiver-operating characteristic sICAM1 solubleintercellular adhesion molecule 1 STAT signal transducer and activator oftranscription STEPP subpopulation treatment effect pattern plot TNF tumornecrosis factor

Competing interestsThe studies described here were funded by GenentechF Hoffmann-LaRoche the current or former employer of GD CH SK DC AFS JH PH HGWYL LD SF AS DM MK FM and MT who participated in study design datacollection analysis and preparation of the manuscript

Authors contributionsGD data collection and analysis manuscript writing critical revision finalapproval of manuscript CH data collection and analysis manuscript writingcritical revision final approval of manuscript SK data analysis manuscriptwriting critical revision final approval of manuscript DC data analysis criticalrevision final approval of manuscript AFS data collection and analysiscritical revision final approval of manuscript WYL data collection andanalysis critical revision final approval of manuscript LD data analysiscritical revision final approval of manuscript SF data collection and analysiscritical revision final approval of manuscript AS data collection and analysiscritical revision final approval of manuscript JH data analysis critical revisionand final approval of manuscript PH data analysis critical revision and finalapproval of manuscript HG data analysis critical revision and final approvalof manuscript DM data collection critical revision and final approval ofmanuscript MK data collection critical revision and final approval ofmanuscript CG data collection critical revision and final approval ofmanuscript AK data collection critical revision and final approval ofmanuscript JE acquisition of samples data collection and final approval ofmanuscript DF acquisition of samples data collection and analysis criticalrevision final approval of manuscript FM study conception and design datacollection and analysis manuscript writing final approval of the manuscriptMT study conception and design data collection and analysis manuscriptwriting critical revision final approval of the manuscript All authors read andapproved the final manuscript

Authorsrsquo informationFlavius Martin and Michael J Townsend co-directed the project

AcknowledgementsWe thank Drs Andrew Urquhart Kevin Chung and the orthopedic andoperating room staff of the University of Michigan for the collection ofsynovial samples Dr Zora Modrusan for support in generating microarraydata and Drs Timothy Behrens John Monroe Nicholas Lewin-Koh andRobert Gentleman for helpful discussions regarding the data analysis Wealso thank Josefa Chuh Marion Patrick Christopher Motyl and Peter Whitefor technical assistance

Author details1Departments of Immunology Discovery Genentech South San FranciscoCalifornia USA 2ITGR Diagnostics Discovery Genentech South San FranciscoCalifornia USA 3Bioinformatics and Computational Biology GenentechSouth San Francisco California USA 4Non-clinical Biostatistics GenentechSouth San Francisco California USA 5Pathology Genentech South SanFrancisco California USA 6Bioanalytical Sciences Genentech South SanFrancisco California USA 7Product Development Genentech South SanFrancisco California USA 8University Hospital of Geneva GenevaSwitzerland 9University of California San Diego San Diego California USA10Rheumatic Disease Core Center and Division of RheumatologyDepartment of Internal Medicine University of Michigan Medical School Ann

Arbor Michigan USA 11Current address Inflammation Therapeutic AreaAmgen 1201 Amgen Court West Seattle Washington USA

Received 29 July 2013 Accepted 25 February 2014Published 30 Apr 2014

References1 Goronzy JJ Weyand CM Rheumatoid arthritis Immunol Rev 2005

20455ndash732 Lee DM Weinblatt ME Rheumatoid arthritis Lancet 2001 358903ndash9113 Tak PP Bresnihan B The pathogenesis and prevention of joint damage in

rheumatoid arthritis advances from synovial biopsy and tissue analysisArthritis Rheum 2000 432619ndash2633

4 Lindstrom TM Robinson WH Biomarkers for rheumatoid arthritis makingit personal Scand J Clin Lab Invest Suppl 2010 24279ndash84

5 Scott DL Wolfe F Huizinga TW Rheumatoid arthritis Lancet 20103761094ndash1108

6 Weyand CM Goronzy JJ Ectopic germinal center formation inrheumatoid synovitis Ann NY Acad Sci 2003 987140ndash149

7 Chan AC Behrens TW Personalizing medicine for autoimmune andinflammatory diseases Nat Immunol 2013 14106ndash109

8 van der Pouw Kraan TC van Gaalen FA Huizinga TW Pieterman EBreedveld FC Verweij CL Discovery of distinctive gene expression profilesin rheumatoid synovium using cDNA microarray technology evidencefor the existence of multiple pathways of tissue destruction and repairGenes Immun 2003 4187ndash196

9 van der Pouw Kraan TC van Gaalen FA Kasperkovitz PV Verbeet NL SmeetsTJ Kraan MC Fero M Tak PP Huizinga TW Pieterman E Breedveld FC AlizadehAA Verweij CL Rheumatoid arthritis is a heterogeneous disease evidencefor differences in the activation of the STAT-1 pathway betweenrheumatoid tissues Arthritis Rheum 2003 482132ndash2145

10 van Baarsen LG Bos CL van der Pouw Kraan TC Verweij CL Transcriptionprofiling of rheumatic diseases Arthritis Res Ther 2009 11207

11 Timmer TC Baltus B Vondenhoff M Huizinga TW Tak PP Verweij CLMebius RE van der Pouw Kraan TC Inflammation and ectopic lymphoidstructures in rheumatoid arthritis synovial tissues dissected by genomicstechnology identification of the interleukin-7 signaling pathway intissues with lymphoid neogenesis Arthritis Rheum 2007 562492ndash2502

12 van der Pouw Kraan TC Wijbrandts CA van Baarsen LG Rustenburg FBaggen JM Verweij CL Tak PP Responsiveness to anti-tumour necrosisfactor alpha therapy is related to pre-treatment tissue inflammationlevels in rheumatoid arthritis patients Ann Rheum Dis 2008 67563ndash566

13 Wijbrandts CA Dijkgraaf MG Kraan MC Vinkenoog M Smeets TJ Dinant HVos K Lems WF Wolbink GJ Sijpkens D Dijkmans BA Tak PP The clinicalresponse to infliximab in rheumatoid arthritis is in part dependent onpretreatment tumour necrosis factor alpha expression in the synoviumAnn Rheum Dis 2008 671139ndash1144

14 Badot V Galant C Nzeusseu Toukap A Theate I Maudoux AL Van denEynde BJ Durez P Houssiau FA Lauwerys BR Gene expression profiling inthe synovium identifies a predictive signature of absence of responseto adalimumab therapy in rheumatoid arthritis Arthritis Res Ther 200911R57

15 Lindberg J Wijbrandts CA van Baarsen LG Nader G Klareskog L Catrina AThurlings R Vervoordeldonk M Lundeberg J Tak PP The gene expressionprofile in the synovium as a predictor of the clinical response toinfliximab treatment in rheumatoid arthritis PLoS One 2010 5e11310

16 Arnett FC Edworthy SM Bloch DA McShane DJ Fries JF Cooper NS HealeyLA Kaplan SR Liang MH Luthra HS Medsger TA Jr Mitchell DM NeustadtDH Pinals RS Schaller JG Sharp JT Wilder RL Hunder GG The Americanrheumatism association 1987 revised criteria for the classification ofrheumatoid arthritis Arthritis Rheum 1988 31315ndash324

17 Barrett T Troup DB Wilhite SE Ledoux P Evangelista C Kim IFTomashevsky M Marshall KA Phillippy KH Sherman PM Muertter RN HolkoM Ayanbule O Yefanov A Soboleva A NCBI GEO archive for functionalgenomics data setsndash10 years on Nucleic Acids Res 2011 39D1005ndashD1010

18 Edgar R Domrachev M Lash AE Gene expression omnibus NCBI geneexpression and hybridization array data repository Nucleic Acids Res 200230207ndash210

19 R Development Core Team R a language and environment for statisticalcomputing In R Foundation for Statistical Computing Vienna Austria R


Foundation for Statistical Computing 2011 httpcranr-projectorgISBN 3-900051-07-0

20 Gentleman RC Carey VJ Bates DM Bolstad B Dettling M Dudoit S Ellis BGautier L Ge Y Gentry J Hornik K Hothorn T Huber W Iacus S Irizarry RLeisch F Li C Maechler M Rossini AJ Sawitzki G Smith C Smyth G TierneyL Yang JY Zhang J Bioconductor open software development forcomputational biology and bioinformatics Genome Biol 2004 5R80

21 Bolstad BM Irizarry RA Astrand M Speed TP A comparison ofnormalization methods for high density oligonucleotide array databased on variance and bias Bioinformatics 2003 19185ndash193

22 Irizarry RA Hobbs B Collin F Beazer-Barclay YD Antonellis KJ Scherf USpeed TP Exploration normalization and summaries of high densityoligonucleotide array probe level data Biostatistics 2003 4249ndash264

23 Hackstadt AJ Hess AM Filtering for increased power for microarray dataanalysis BMC Bioinformatics 2009 1011

24 Bourgon R Gentleman R Huber W Independent filtering increasesdetection power for high-throughput experiments Proc Natl Acad Sci USA2010 1079546ndash9551

25 Dennis G Jr Sherman BT Hosack DA Yang J Gao W Lane HC Lempicki RADAVID Database for annotation visualization and integrated discoveryGenome Biol 2003 43

26 Oron AP Jiang Z Gentleman R Gene set enrichment analysis using linearmodels and diagnostics Bioinformatics 2008 242586ndash2591

27 Subramanian A Tamayo P Mootha VK Mukherjee S Ebert BL Gillette MAPaulovich A Pomeroy SL Golub TR Lander ES Mesirov JP Gene setenrichment analysis a knowledge-based approach for interpretinggenome-wide expression profiles Proc Natl Acad Sci USA 200510215545ndash15550

28 Abbas AR Baldwin D Ma Y Ouyang W Gurney A Martin F Fong S vanLookeren CM Godowski P Williams PM Chan AC Clark HF Immune responsein silico (IRIS) immune-specific genes identified from a compendium ofmicroarray expression data Genes Immun 2005 6319ndash331

29 Hochberg Y Benjamini Y More powerful procedures for multiplesignificance testing Stat Med 1990 9811ndash818

30 Gabay C Emery P van Vollenhoven R Dikranian A Alten R Pavelka KKlearman M Musselman D Agarwal S Green J Kavanaugh A Tocilizumabmonotherapy versus adalimumab monotherapy for treatment ofrheumatoid arthritis (ADACTA) a randomised double-blind controlledphase 4 trial Lancet 2013 3811541ndash1550

31 Lazar AA Cole BF Bonetti M Gelber RD Evaluation of treatment-effectheterogeneity using biomarkers measured on a continuous scale sub-population treatment effect pattern plot J Clin Oncol 2010 284539ndash4544

32 Kishimoto T Interleukin-6 from basic science to medicinendash40 years inimmunology Annu Rev Immunol 2005 231ndash21

33 Weyand CM Kang YM Kurtin PJ Goronzy JJ The power of the thirddimension tissue architecture and autoimmunity in rheumatoid arthritisCurr Opin Rheumatol 2003 15259ndash266

34 van Baarsen LG Wijbrandts CA Timmer TC van der Pouw Kraan TC Tak PPVerweij CL Synovial tissue heterogeneity in rheumatoid arthritis inrelation to disease activity and biomarkers in peripheral bloodArthritis Rheum 2010 621602ndash1607

35 van Oosterhout M Bajema I Levarht EW Toes RE Huizinga TW van Laar JMDifferences in synovial tissue infiltrates between anti-cyclic citrullinatedpeptide-positive rheumatoid arthritis and anti-cyclic citrullinatedpeptide-negative rheumatoid arthritis Arthritis Rheum 2008 5853ndash60

36 Hogan VE Holweg CT Choy DF Kummerfeld SK Hackney JA Teng YKTownsend MJ van Laar JM Pretreatment synovial transcriptional profile isassociated with early and late clinical response in rheumatoid arthritispatients treated with rituximab Ann Rheum Dis 1888ndash1894 201271

37 Hueber W Tomooka BH Batliwalla F Li W Monach PA Tibshirani RJ VanVollenhoven RF Lampa J Saito K Tanaka Y Genovese MC Klareskog LGregersen PK Robinson WH Blood autoantibody and cytokine profilespredict response to anti-tumor necrosis factor therapy in rheumatoidarthritis Arthritis Res Ther 2009 11R76

38 Lal P Su Z Holweg CT Silverman GJ Schwartzman S Kelman A Read SSpaniolo G Monroe JG Behrens TW Townsend MJ Inflammation andautoantibody markers identify rheumatoid arthritis patients withenhanced clinical benefit following rituximab treatment Arthritis Rheum2011 633681ndash3691

39 Klaasen R Thurlings RM Wijbrandts CA van Kuijk AW Baeten D Gerlag DMTak PP The relationship between synovial lymphocyte aggregates and

the clinical response to infliximab in rheumatoid arthritis a prospectivestudy Arthritis Rheum 2009 603217ndash3224

40 Canete JD Celis R Moll C Izquierdo E Marsal S Sanmarti R Palacin A Lora Dde la Cruz J Pablos JL Clinical significance of synovial lymphoid neogenesisand its reversal after anti-tumour necrosis factor alpha therapy inrheumatoid arthritis Ann Rheum Dis 2009 68751ndash756

41 Krenn V Schedel J Doring A Huppertz HI Gohlke F Tony HP Vollmers HPMuller-Hermelink HK Endothelial cells are the major source of sICAM-1 inrheumatoid synovial tissue Rheumatol Int 1997 1717ndash27

42 Witkowska AM Borawska MH Soluble intercellular adhesion molecule-1(sICAM-1) an overview Eur Cytokine Netw 2004 1591ndash98

43 Corsiero E Bombardieri M Manzo A Bugatti S Uguccioni M Pitzalis C Roleof lymphoid chemokines in the development of functional ectopiclymphoid structures in rheumatic autoimmune diseases Immunol Lett2012 14562ndash67

44 Rosengren S Wei N Kalunian KC Kavanaugh A Boyle DL CXCL13 a novelbiomarker of B-cell return following rituximab treatment and synovitis inpatients with rheumatoid arthritis Rheumatol (Oxford) 2011 50603ndash610

45 Meeuwisse CM van der Linden MP Rullmann TA Allaart CF Nelissen RHuizinga TW Garritsen A Toes RE van Schaik R van der Helm-van Mil AHIdentification of CXCL13 as a marker for rheumatoid arthritis outcomeusing an in silico model of the rheumatic joint Arthritis Rheum 2011631265ndash1273

46 Choy E Understanding the dynamics pathways involved in thepathogenesis of rheumatoid arthritis Rheumatol (Oxford) 2012 51v3ndashv11

47 Chen G Goeddel DV TNF-R1 signaling a beautiful pathway Science 20022961634ndash1635

48 Emery P Keystone E Tony HP Cantagrel A van Vollenhoven R Sanchez AAlecock E Lee J Kremer J IL-6 receptor inhibition with tocilizumabimproves treatment outcomes in patients with rheumatoid arthritisrefractory to anti-tumour necrosis factor biologicals results froma 24-week multicentre randomised placebo-controlled trial Ann RheumDis 2008 671516ndash1523

Cite this article as Dennis et al Synovial phenotypes in rheumatoidarthritis correlate with response to biologic therapeutics ArthritisResearch amp Therapy

101186ar4555

2014 16R90

Submit your next manuscript to BioMed Centraland take full advantage of

bull Convenient online submission

bull Thorough peer review

bull No space constraints or color figure charges

bull Immediate publication on acceptance

bull Inclusion in PubMed CAS Scopus and Google Scholar

bull Research which is freely available for redistribution

Submit your manuscript at wwwbiomedcentralcomsubmit


IntroductionRheumatoid arthritis (RA) is an autoimmune diseasecharacterized by symmetrical joint involvement inflam-mation synovial lining hyperplasia and formation of in-vasive granulation tissue or pannus Progression of RApathogenesis is associated with impaired joint functionresulting from immune-mediated destruction of boneand cartilage [1-3] Considerable patient-to-patient vari-ation exists in the number of affected joints the levels ofautoantibody titers and serum cytokines and the rate ofjoint destruction [45] Disease heterogeneity is furtherevident upon histological examination of synovial tissueswhere a spectrum of cellular compositions are found ran-ging from diffuse leukocytic infiltration to well-organizedlymphocyte-containing follicle-like structures [6]Not surprisingly RA is also heterogeneous in response

to treatment Although the development of targeted thera-peutic strategies blocking TNF α IL-6 receptor T-cell co-stimulation blockade and B-cell depletion have providedmeaningful clinical benefit to patients a key unmet needin the management of RA is the prospective identificationof patients who are likely to benefit from specific therap-ies We hypothesized that a deeper understanding of themolecular basis of disease heterogeneity will lead to thediscovery of predictive biomarkers able to identify individ-ual patients who will benefit from a particular therapeuticstrategy [7]Insight into pathogenic molecular pathways of RA has

emerged in recent years from genome-wide analysis of syn-ovial tissue gene expression Multiple studies have assessedmolecular heterogeneity in RA tissue but few findings havebeen validated with subsequent cohorts Early studies [89]revealed considerable molecular heterogeneity and pro-posed RA patient subgroups exhibiting gene expressionpatterns consistent with ongoing inflammation and adap-tive immunity or alternatively little immune infiltrateand instead expressing sets of genes involved in extra-cellular matrix remodeling [10] Further it has been ob-served that lymphoid follicle-containing synovial sampleshave increased expression of sets of genes involved inJanus kinase (JAK)signal transducer and activator of tran-scription (STAT) signaling and IL-7 signal transduction[11] suggesting that differences in gene expression pat-terns reflect differences in relative cellular composition ofthe RA jointGene and protein expression studies of synovial tissue

at baseline prior to initiating TNFα blockade have alsogenerated different hypotheses to account for the differ-ences between good and poor responders In two studiespatients who responded to anti-TNFα treatment had tran-scription profiles enriched for inflammatory processes andTNFα protein expression [1213] whereas another reportconcluded that good responders actually had lower in-flammatory processes and cell-surface markers such as

the IL-7 receptor alpha chain [14] A large gene expressionstudy of synovial tissues from 62 patients obtained priorto initiating anti-TNFα therapy identified very fewtranscripts that were different between good and poorresponders [15] In the current study we build on theseobservations by characterizing different molecular pheno-types of RA synovium - lymphoid myeloid and fibroid -and used these to identify soluble biomarkers that predictdifferential treatment effects in RA patients

MethodsPatients and synovial tissuesSynovial tissues were obtained from RA subjects under-going arthroplasty andor synovectomy of affected joints(University of Michigan two sequential cohorts n = 49and n = 20) Written consent was obtained from patientsand the University of Michigan Institutional Review Boardapproved the study protocol RA was diagnosed basedupon the 1987 College of Rheumatology (ACR) criteria[16] Patients were treated using the standard of care forRA (non-steroidal anti-inflammatory drugs (NSAIDs) anddisease-modifying anti-rheumatic drugs (DMARDs)) andsome patients were also treated with biologics (adalimu-mab etanercept infliximab anakinra and rituximab)Patients were diagnosed with RA at least three yearsbefore surgery and 70 of patients for whom data wereavailable were rheumatoid factor (RF)-positive Excised tis-sues were immediately snap-frozen in liquid nitrogen andstored at -80degC Each tissue was used for both histologyand RNA extraction For cryo-sectioning samples werebrought briefly to -20degC sectioned and immediatelyreturned to -80degC to maintain RNA integrity All tissuesused for downstream studies were prospectively random-ized during processing and sectioning prior to expressionanalysis to minimize technical batch effects in the data

RNA isolationFrozen samples were weighed and homogenized in RLTbuffer (Qiagen Valencia California USA) + β-mercap-toethanol (10 μlml) at a concentration of 100 mgmlPrior to isolating RNA using an RNeasy minikit (Qiagen)with on-column DNase digestion samples were digestedwith Proteinase K (Qiagen) for 10 minutes at 55degC

Histopathology and immunohistochemistryStains were performed on 5-μm-thick frozen sections of hu-man synovial tissue fixed in acetone Some sections werestained with hematoxylin and eosin for histologic evaluationOther sections were blocked in 10 serum for 30 minutesand stained for the detection of cells expressing the followinglineage markers (CD20 - mouse anti-human clone L26 5 μgml Dako (Carpinteria California USA) CD3 - rabbit anti-human antibody SP7 1200 dilution NeoMarkers (FremontCalifornia USA) and CD68 - mouse anti-human clone KP-1
















A

2

-2

Z s

core

B

C1 C2 C3 C4

-log10(adj p)

0 1 2 3 4 5 6 7







CD3

CD20

B

CD

3

CD20

CD

45

CD90


20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100


C

Pat

ient

s (

)








Sca

led

Exp

ress

ion

Scl

aed

B C

ell S

core

minus05

00

05

10

L M X F

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

L M X F

Sca

led

M2

Mon

ocyt

e S

core

minus08

minus06

minus04

minus02

00

02

04

L M X F

Sca

led

T C

ell S

core

minus05

00

05

10

15

L M X F

Sca

led

TN

Fα S

core

minus04

minus02

00

02

L M X F

Sca

led

Ang

ioge

nesi

s S

core

minus10

minus05

00

05

L M X F

B

F

C

E

D

G

minus3

minus2

minus1

0

1

2

3













Sca

led

B C

ell S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

Mye

loid

Sco

re

minus10

minus05

00

05

10

Good Moderate PoorS

cale

d Ly

mph

oid

Sco

re

minus10

minus05

00

05

10

15

Good Moderate Poor

Sca

led

Fib

roid

Sco

re

minus10

minus05

00

05

10

Good Moderate Poor

A B C

D E F

Sca

led

T C

ell

Sco

reminus10

minus05

00

05

10

Good Moderate Poor






BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600


plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F






001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90
























A

2

-2

Z s

core

B

C1 C2 C3 C4

-log10(adj p)

0 1 2 3 4 5 6 7







CD3

CD20

B

CD

3

CD20

CD

45

CD90


20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100


C

Pat

ient

s (

)








Sca

led

Exp

ress

ion

Scl

aed

B C

ell S

core

minus05

00

05

10

L M X F

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

L M X F

Sca

led

M2

Mon

ocyt

e S

core

minus08

minus06

minus04

minus02

00

02

04

L M X F

Sca

led

T C

ell S

core

minus05

00

05

10

15

L M X F

Sca

led

TN

Fα S

core

minus04

minus02

00

02

L M X F

Sca

led

Ang

ioge

nesi

s S

core

minus10

minus05

00

05

L M X F

B

F

C

E

D

G

minus3

minus2

minus1

0

1

2

3













Sca

led

B C

ell S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

Mye

loid

Sco

re

minus10

minus05

00

05

10

Good Moderate PoorS

cale

d Ly

mph

oid

Sco

re

minus10

minus05

00

05

10

15

Good Moderate Poor

Sca

led

Fib

roid

Sco

re

minus10

minus05

00

05

10

Good Moderate Poor

A B C

D E F

Sca

led

T C

ell

Sco

reminus10

minus05

00

05

10

Good Moderate Poor






BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600


plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F






001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90
















A

2

-2

Z s

core

B

C1 C2 C3 C4

-log10(adj p)

0 1 2 3 4 5 6 7







CD3

CD20

B

CD

3

CD20

CD

45

CD90


20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100


C

Pat

ient

s (

)








Sca

led

Exp

ress

ion

Scl

aed

B C

ell S

core

minus05

00

05

10

L M X F

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

L M X F

Sca

led

M2

Mon

ocyt

e S

core

minus08

minus06

minus04

minus02

00

02

04

L M X F

Sca

led

T C

ell S

core

minus05

00

05

10

15

L M X F

Sca

led

TN

Fα S

core

minus04

minus02

00

02

L M X F

Sca

led

Ang

ioge

nesi

s S

core

minus10

minus05

00

05

L M X F

B

F

C

E

D

G

minus3

minus2

minus1

0

1

2

3













Sca

led

B C

ell S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

Mye

loid

Sco

re

minus10

minus05

00

05

10

Good Moderate PoorS

cale

d Ly

mph

oid

Sco

re

minus10

minus05

00

05

10

15

Good Moderate Poor

Sca

led

Fib

roid

Sco

re

minus10

minus05

00

05

10

Good Moderate Poor

A B C

D E F

Sca

led

T C

ell

Sco

reminus10

minus05

00

05

10

Good Moderate Poor






BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600


plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F






001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90










A

2

-2

Z s

core

B

C1 C2 C3 C4

-log10(adj p)

0 1 2 3 4 5 6 7







CD3

CD20

B

CD

3

CD20

CD

45

CD90


20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100


C

Pat

ient

s (

)








Sca

led

Exp

ress

ion

Scl

aed

B C

ell S

core

minus05

00

05

10

L M X F

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

L M X F

Sca

led

M2

Mon

ocyt

e S

core

minus08

minus06

minus04

minus02

00

02

04

L M X F

Sca

led

T C

ell S

core

minus05

00

05

10

15

L M X F

Sca

led

TN

Fα S

core

minus04

minus02

00

02

L M X F

Sca

led

Ang

ioge

nesi

s S

core

minus10

minus05

00

05

L M X F

B

F

C

E

D

G

minus3

minus2

minus1

0

1

2

3













Sca

led

B C

ell S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

Mye

loid

Sco

re

minus10

minus05

00

05

10

Good Moderate PoorS

cale

d Ly

mph

oid

Sco

re

minus10

minus05

00

05

10

15

Good Moderate Poor

Sca

led

Fib

roid

Sco

re

minus10

minus05

00

05

10

Good Moderate Poor

A B C

D E F

Sca

led

T C

ell

Sco

reminus10

minus05

00

05

10

Good Moderate Poor






BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600


plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F






001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90









CD3

CD20

B

CD

3

CD20

CD

45

CD90


20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100

0

20

40

60

80

100


C

Pat

ient

s (

)








Sca

led

Exp

ress

ion

Scl

aed

B C

ell S

core

minus05

00

05

10

L M X F

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

L M X F

Sca

led

M2

Mon

ocyt

e S

core

minus08

minus06

minus04

minus02

00

02

04

L M X F

Sca

led

T C

ell S

core

minus05

00

05

10

15

L M X F

Sca

led

TN

Fα S

core

minus04

minus02

00

02

L M X F

Sca

led

Ang

ioge

nesi

s S

core

minus10

minus05

00

05

L M X F

B

F

C

E

D

G

minus3

minus2

minus1

0

1

2

3













Sca

led

B C

ell S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

Mye

loid

Sco

re

minus10

minus05

00

05

10

Good Moderate PoorS

cale

d Ly

mph

oid

Sco

re

minus10

minus05

00

05

10

15

Good Moderate Poor

Sca

led

Fib

roid

Sco

re

minus10

minus05

00

05

10

Good Moderate Poor

A B C

D E F

Sca

led

T C

ell

Sco

reminus10

minus05

00

05

10

Good Moderate Poor






BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600


plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F






001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90










Sca

led

Exp

ress

ion

Scl

aed

B C

ell S

core

minus05

00

05

10

L M X F

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

L M X F

Sca

led

M2

Mon

ocyt

e S

core

minus08

minus06

minus04

minus02

00

02

04

L M X F

Sca

led

T C

ell S

core

minus05

00

05

10

15

L M X F

Sca

led

TN

Fα S

core

minus04

minus02

00

02

L M X F

Sca

led

Ang

ioge

nesi

s S

core

minus10

minus05

00

05

L M X F

B

F

C

E

D

G

minus3

minus2

minus1

0

1

2

3













Sca

led

B C

ell S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

Mye

loid

Sco

re

minus10

minus05

00

05

10

Good Moderate PoorS

cale

d Ly

mph

oid

Sco

re

minus10

minus05

00

05

10

15

Good Moderate Poor

Sca

led

Fib

roid

Sco

re

minus10

minus05

00

05

10

Good Moderate Poor

A B C

D E F

Sca

led

T C

ell

Sco

reminus10

minus05

00

05

10

Good Moderate Poor






BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600


plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F






001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90
















Sca

led

B C

ell S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

Mye

loid

Sco

re

minus10

minus05

00

05

10

Good Moderate PoorS

cale

d Ly

mph

oid

Sco

re

minus10

minus05

00

05

10

15

Good Moderate Poor

Sca

led

Fib

roid

Sco

re

minus10

minus05

00

05

10

Good Moderate Poor

A B C

D E F

Sca

led

T C

ell

Sco

reminus10

minus05

00

05

10

Good Moderate Poor






BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600


plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F






001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90









Sca

led

B C

ell S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

M1

Mon

ocyt

e S

core

minus05

00

05

10

Good Moderate Poor

Sca

led

Mye

loid

Sco

re

minus10

minus05

00

05

10

Good Moderate PoorS

cale

d Ly

mph

oid

Sco

re

minus10

minus05

00

05

10

15

Good Moderate Poor

Sca

led

Fib

roid

Sco

re

minus10

minus05

00

05

10

Good Moderate Poor

A B C

D E F

Sca

led

T C

ell

Sco

reminus10

minus05

00

05

10

Good Moderate Poor






BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600


plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F






001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90









BA

C

0

100

200

300

400

500C

XC

L13

(pg

ml)

0

100

200

300

400

500

600

sIC

AM

(ng

ml)

0

100

200

300

CX

CL1

3 (p

gm

l)

0 100 200 300 400 500 600


plt00001plt00001

CX

CL1

3(L

og2

Pro

be In

tens

ity)

ICA

M1

(Log

2 P

robe

Inte

nsity

)

L M X F

D E

85

9

95

10

105

11

115

12

4

5

6

7

8

9

10

11

12

13

14

15

L M X F






001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90









001 005 01 05 1 5 10

odds ratio (95 CI)

CXCL13

low

high

low

high

sICAM1 low

sICAM1 high

62=n51=n

n=26n=25

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high

n=26 n=32

n=15n=33

0 20 40 60 80

CXCL13

sIC

AM

1

hgihwol

low

high


0

20

40

60

80

AC

R50

Res

pons

e R

ate

()


sICAM1CXCL13

HighLow

HighHigh

LowHigh

LowLow

13

69

24

44

28

4242

20

Δ = 29

Δ = 49

A

C D

E

odds ratio (95 CI)

sICAM1 low

sICAM1 high

odds ratio (95 CI)

CXCL13 low

CXCL13 high

B

05 1 15 2

05 1 15 2
























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90






























23 minus23 (plusmn037) 293 (07-152)

7 minus11 (plusmn033) 007 (0009-03)

19 minus21 (plusmn031) 053 (017-16)

18 minus21 (plusmn032) 041 (013-12)


7 minus32 (plusmn037) 034 (007-14)

50 minus36 (plusmn032) 146 (31-1089)

31 minus32 (plusmn037) 19 (063-573)

24 minus29 (plusmn036) 25 (08-78)


















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90

























Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90



















Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90














Additional files


Additional file 1









Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90

















Additional file 2

Additional file 3

Additional file 4

Additional file 5

Additional file 6

Additional file 7







Additional file 8

Additional file 9

Additional file 10

Additional file 11

Additional file 12































































101186ar4555

2014 16R90






































































101186ar4555

2014 16R90










































101186ar4555

2014 16R90









Date post:	11-Nov-2023
Category:	Documents
Upload:	independent
View:	1 times
Download:	0 times

Synovial phenotypes in rheumatoid arthritis correlate with response to biologic therapeutics

Documents