T Cell–Dependent Affinity Maturation and Innate ImmunePathways Differentially Drive Autoreactive B Cell Responses in

Rheumatoid Arthritis

Daniel R. Lu,1 Andrew N. McDavid,2 Sarah Kongpachith,1 Nithya Lingampalli,1

Jacob Glanville,3 Chia-Hsin Ju,1 Raphael Gottardo,4 and William H. Robinson 1

Objective. Rheumatoid arthritis (RA) is character-ized by the activation of B cells that produce anti–citrullinated protein antibodies (ACPAs) and rheumatoidfactors (RFs), but the mechanisms by which tolerance isbroken in these B cells remain incompletely understood.We undertook this study to investigate whether ACPA+and RF+ B cells break tolerance through distinct molecu-lar mechanisms.

Methods. We developed antigen–tetramers to iso-late ACPA+ and RF+ B cells and performed single-cellRNA sequencing on 2,349 B cells from 6 RA patients and1 healthy donor to analyze their immunoglobulin reper-toires and transcriptional programs. Prominentimmunoglobulins were expressed as monoclonal antibod-ies and tested for autoantigen reactivity.

Results. ACPA+ and RF+ B cells were enriched inthe peripheral blood of RA patients relative to healthycontrols. Characterization of patient-derived monoclonalantibodies confirmed ACPA and RF targeting of tetra-mer-specific B cells at both antigen-inexperienced andaffinity-matured B cell stages. ACPA+ B cells used moreclass-switched isotypes and exhibited more somatic

hypermutations relative to RF+ B cells, and these differ-ences were accompanied by down-regulation of CD72 andup-regulation of genes that promote class-switching andT cell–dependent responses. In contrast, RF+ B cellsexpressed transcriptional programs that stimulate rapidmemory reactivation through multiple innate immunepathways. Coexpression analysis revealed that ACPA+and RF+ B cell–enriched genes belong to distinct tran-scriptional regulatory networks.

Conclusion. Our findings suggest that ACPA+ andRF+ B cells are imprinted with distinct transcriptionalprograms, which suggests that these autoantibodies asso-ciated with increased inflammation in RA arise from 2different molecular mechanisms.

Rheumatoid arthritis (RA) is characterized bychronic synovitis and joint destruction. Autoantibodiesare a hallmark of RA and include anti–citrullinatedprotein antibodies (ACPAs) and rheumatoid factors(RFs) (1). Recent data suggest that ACPA and RFautoantibodies uniquely contribute to systemic inflam-mation and are associated with increased disease activ-ity scores in RA (2). Furthermore, B cells, whichproduce autoantibodies and cytokines and provide Tcell help, contribute to pathology as evidenced by theefficacy of B cell–depleting therapies in RA (3).

B cell activation requires coordination of manycell-extrinsic and -intrinsic factors (1), and our understand-ing of how B cell tolerance is broken in RA remains lim-ited. B cell activation is regulated by B cell receptor(BCR) signaling and costimulatory signals, and dysregu-lated signaling events can promote the survival and differ-entiation of autoreactive B cells that would otherwise bedeleted or tolerized. RFs, which can be present in otherautoimmune diseases and chronic inflammation states,can also develop following acute microbial infections (4,5).RF generation has been recapitulated through challenges

Supported by the NIH (National Institute of Arthritis andMusculoskeletal and Skin Diseases grants R01-AR-063676, U19-AI-11049103, U01-AI-101981, and UH2-AR-067676).

1Daniel R. Lu, PhD, Sarah Kongpachith, PhD, NithyaLingampalli, BS, Chia-Hsin Ju, PhD, William H. Robinson, MD,PhD: Stanford University, Stanford, California, and VA Palo AltoHealth Care System, Palo Alto, California; 2Andrew N. McDavid,PhD: University of Rochester Medical Center, Rochester, New York;3Jacob Glanville, PhD: Stanford University, Stanford, California;4Raphael Gottardo, PhD: Fred Hutchinson Cancer Research Center,Seattle, Washington.

Drs. Lu and McDavid contributed equally to this work.Dr. Robinson has received consulting fees from Atreca, Inc.

(more than $10,000), owns stock or stock options in Atreca, Inc., andserves on the board of directors of Atreca, Inc.

Address correspondence to William H. Robinson, MD, PhD,CCSR 4135, 269 Campus Drive, Stanford, CA 94305. E-mail:wrobins@stanford.edu.

Submitted for publication January 18, 2018; accepted inrevised form May 29, 2018.

1732

ARTHRITIS & RHEUMATOLOGYVol. 70, No. 11, November 2018, pp 1732–1744DOI 10.1002/art.40578© 2018, American College of Rheumatology

with lipopolysaccharide or double-stranded DNA in mice,providing evidence that RF+ B cells become activated indiverse states of inflammation (6,7). In contrast, ACPAsare highly specific for RA and promote synovial inflamma-tion and joint destruction in mouse models (8,9). Indeed,ACPA+ B cells continuously undergo antigen-driven acti-vation throughout disease, as evidenced by the continualregeneration of ACPA+ plasmablasts in patients (10).Based on these observations, we reasoned that the devel-opment and persistence of RFs and ACPAs may involvedifferential regulation of B cells at different stages of dif-ferentiation and may arise through the loss of tolerancemediated by distinct mechanisms.

In the present study, we developed antigen–tetramer staining reagents to isolate RF+ and ACPA+ Bcells from seropositive RA patients and used single-cellRNA sequencing (RNA-Seq) to simultaneously recoverthe paired BCR variable region and the transcriptionalprofiles of individual B cells. By using a subset of B cellswith assessed surface markers and sequenced immuno-globulins, we developed a computational B cell classifier(“BCellNet”) that used the single-cell RNA-Seq profile toorganize B cells by subtype and characterize the primary,secondary, and antibody-producing stages of B cell differ-entiation. We applied these new methods to RA and iden-tified distinct activation programs that distinguish RF+ Bcell responses from ACPA+ B cell responses. Our findingssuggest that tolerance mediated by 2 different molecularmechanisms must be broken to activate ACPA+ and RF+B cells to produce autoantibodies and promote synovitisin RA.

PATIENTS AND METHODS

Study design. All samples were collected after obtaininginformed consent and according to human subject protocolsapproved by the Investigational Review Board at StanfordUniversity (see Supplementary Table 1, available on the Arthritis& Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract). Peripheral blood was obtained from 6individuals with RA who met the American College of Rheuma-tology 1987 revised classification criteria (11), were seropositivefor RF and/or anti–cyclic citrullinated peptide (anti-CCP), anddid not receive B cell–depleting therapy. Five age-matchedhealthy donors who were negative for HIVand tuberculosis wererecruited from the Stanford Blood Center.

Autoreactive B cell isolation.RF tetramers were preparedas described in Supplementary Methods, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract. Fourteen citrullinatedpeptides (400 lM) of highly prevalent citrullinated proteinantigens (12) were selected for citrullinated protein tetramerpreparation as described in Supplementary Methods. LiveCD3�CD14�CD19+ B cells were sorted using a FACSAria IIflow cytometer (Becton Dickinson) at a single-cell purity of>99%. B cells were considered ACPA+RF� or RF+ACPA� if

they stained double positive for the autoantigen of interest anddouble negative for the other tetramerized antigen.

Library preparation and next-generation sequencing.Template-switching reverse transcription of messenger RNAwas primed using the STRT-oligo (13), and complementaryDNAs (cDNAs) were barcoded using template-switched oligoscontaining cell and molecular identifiers (see SupplementaryTable 2, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract). Barcoded cDNA libraries were pooled and amplifiedusing PCR1-FWD and PCR1-REV primers and then soni-cated. We purified 50 fragments of cDNA molecules usingMyOne C1 Streptavidin Dynabeads (Life Technologies) anduncoupled them by restriction digestion. Libraries were con-structed using TruSeq-V2 kits (Illumina) and sequenced using aHiSeq 2000 Sequencing System (2 9 100 bp read length; Illu-mina). Details are available in Supplementary Methods, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract.

Single-cell RNA-Seq pipeline. Custom R scriptsextracted cell and molecule barcodes from raw sequencingreads. Reads were aligned to the human genome (Genome Ref-erence Consortium Human Build 38) (14) using Bowtie2 soft-ware (version 2.1.0) (15). Duplicate reads generated bypolymerase chain reaction were removed, and final expressionwas calculated using RSEM software (version 1.2.25) (16). TheBCellNet subtype classification was trained as described in Sup-plementary Methods (http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract), allowing us to estimate a classification func-tion that provided the most likely cell type for any cell with asingle-cell RNA-Seq profile. De novo assembly of immunoglob-ulin VH and VL regions, classification of germline V, D, and Jregions, and identification of third complementarity-determiningregion (CDR3) clusters were performed as described in Supple-mentary Methods.

Recombinant monoclonal antibody testing. A prelimi-nary list of candidate antibodies was first generated from autore-active CDR3 clusters. The top VH–VL pairings from eachCDR3 cluster with complete V-region sequence reconstitutionwere expressed as human IgG1 antibodies with IgL2 or IgK lightchains (LakePharma). RF reactivity was determined by generat-ing F(ab0)2 fragments and probing plates coated with HumanIgG, Fc fragment (Jackson ImmunoResearch). CCP binding wasdetermined using a CCP3 enzyme-linked immunosorbent assay(ELISA; Inova Diagnostics). The threshold for positivity was 3SD above the mean of negative control monoclonal antibodies.Autoantigen microarrays were printed and probed as previouslydescribed (9). Details are available in Supplementary Methods,http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract.

Differential expression testing. We tested differentialexpression through a series of linear models including a subset-stratified model, which accounts for the probability of cellsbelonging to the plasmablast (antibody-secreting cell [ASC]),double-negative (DN), memory, and naive subtypes, and a glo-bal model, which does not account for subtype. Further expla-nation of the 2 models is in Supplementary Methods, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract.

Both models were fit in MAST (17) version 0.933 usinga 2-step procedure to select genes that showed consistent differ-ential expression across patients. Genes were considered to bedifferentially expressed if they 1) were marginally differentiallyexpressed at a false discovery rate (FDR) of 10% and 2) had astandardized effect size of >2. The standardized effect size selectsgenes most likely to substantially affect transcription in the

MOLECULAR MECHANISMS OF AUTOREACTIVE B CELL RESPONSES IN RA 1733

population of RA patients. Both models also returned a partialresidual, which is the logarithm of expression after correction fornuisance factors. Gene functional categories were defined byorganizing the differentially expressed genes (DEGs) and 33genes of interest into 16 potentially overlapping groups. Partialresiduals were used to calculate functional scores for each cell.Details are available in Supplementary Methods, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract.

Coexpression analysis. We tested 487 genes of interestfor coexpression after stratifying by subtype and removing mar-ginal coexpression due to subtype, tetramer specificity, patient,and CDR effects. Pearson correlation matrices were calculatedfor each pair of genes, and the averages of the matrices were com-pared within the 16 gene sets. Gene pairs were considered differ-entially expressed if their average correlation differed from thebackground correlation coefficient at an FDR of <5%. Detailsare available in Supplementary Methods, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract.

Data and materials availability. Raw sequencing datahave been deposited in NCBI SRA and are available uponrequest from the corresponding author. Processed data areavailable at https://github.com/amcdavid/RA-BCell-Response.

RESULTS

Elevation of naive, memory, and antibody-secretingautoreactive B cells in blood of RA patients. In this study,we sampled blood B cells from seropositive RA patientsusing flow cytometry to examine the frequency of ACPA+and RF+ B cells as well as to isolate these cells for single-cell RNA-Seq analysis (Figure 1A). To isolate RF+ andACPA+ B cells, biotinylated human IgG Fc as well as 14citrullinated peptides previously demonstrated to be tar-geted by RA ACPAs were individually coupled to 2 sets ofstreptavidin-bound fluorophores (Figure 1B). Autoantigenlabeling with dual fluorophores was previously demon-strated to reduce the false-positive detection of fluo-rophore-reactive B cells, and tetramerization enabled thedetection of B cells expressing BCRs with moderate affinityfor autoantigen epitopes (10,18). Analysis of B cell fre-quencies revealed that RF+ and ACPA+ cells were virtuallyundetectable in healthy controls (<0.01%) and rare inblood from RA patients, comprising on average 0.25% and0.14%, respectively, of sampled B cells (Figure 1C). Thissmall enrichment of RF+ and ACPA+ cells in RA patientswas significant compared to healthy controls. While the fre-quency of RF+ and ACPA+ B cell subsets varied betweenpatients, autoreactive B cells from the CD27� (naive, non–affinity matured, and DN memory) and CD27+ (classicalmemory) compartments were present in all individuals,consistent with previous observations of defective centraland peripheral tolerance in RA (19,20). The presence ofRF+ and ACPA+ ASCs in blood of RA patients revealsthat they have active ongoing humoral autoimmunity.

We then performed single-cell RNA-Seq on sortedACPA+, RF+, and tetramer-negative B cells from RA

patients and 1 healthy donor, using a modified STRT-Seqmethod that enables single molecule quantification andcomputational reconstitution of the paired BCR for eachB cell (Figure 1A; also see Supplementary Figure 1,http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract).The population of tetramer-negative cells provides abaseline to compare characteristics of the autoreactivepopulations. After aligning and deduplicating reads, werecovered an average of 4.8 9 104 molecules per cell withno substantial differences between experimental groups(see Supplementary Figure 2, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract).

BCellNet predicts peripheral B cell subsets usingsingle-cell RNA-Seq profiles. To evaluate the contributionof various B cell subsets to RA pathology, we organized Bcells by subtype using a B cell transcriptomic classifier wedeveloped, termed BCellNet. Predicting subtype with thistranscriptomic classification scheme allows us to considera continuum of differentiation states (21) and assign asubset probability score for each cell, improving upon thequantized subset identifications available in flow cytomet-ric gating. In total, 159 genes (<2.5%) served as predictors(see Supplementary Table 3, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract). Some, such asMS4A1 and TNFRSF13B, were previously described inbulk-level studies (22), while others, such as SEC61G andEIF5A, were novel (see Supplementary Figure 3, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract).

Applying BCellNet to heterogeneous B cells iso-lated from RA samples demonstrated positive correlationsbetween subtype fractions assessed by flow cytometry andthe classifier (Spearman’s q = 0.56–0.89), indicating thatthe classifier recapitulates flow cytometry–identified popu-lations while also parsing subtype-dependent gene effects(see Supplementary Figure 4, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract). There were certaindiscrepancies between the flow fractions and classifierfractions, likely due to sampling variability in the flow sort-ing and classification error from BCellNet. However, wedo not believe that these discrepancies significantly alteredB cell subtype classification (see Supplementary Methods,http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract).In light of this, rather than using quantized assignments tothe most likely subtype, we used the continuously valuedposterior probabilities of subtype membership as stratify-ing covariates for subsequent analysis (see SupplementaryTable 4, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract).

We observed a general correlation between sero-logic autoantibody levels and the recovery of ACPA+ andRF+ B cells, despite minor variances in the number of cellssorted due to stochastic selection during fluorescence-

1734 LU ET AL

Disease Sample PBMCs

Antigen-specificB-cell Labeling

FACS Single-cellLibrary Prep

Deep Sequencing

mAbExpression and Antigen Discovery

Single-cell V(D)JReconstitution

Transcriptomic Analysis

Single cellClassifier

BA

C

R26 R48 R01 R59 R72 HD1HD2

HD3HD4

HD50

1000

2000

3000

4000

5000ACPA+ B cells

CD27-

CD27+

ASC

HealthyRA

R04 R26 R01 R59 R72 HD1HD2

HD3HD4

HD50

1000

2000

3000

4000

5000RF+ B cells

Cel

ls p

er 1

06 B c

ells

CD27-

CD27+

ASC

HealthyRA

CD19+ RF+

CD3+

CD19+ CIT+

RF-FITC CIT-APC

CIT-APC

FSC-A

SS

C-A

SSC-A

FSC

-W

CD19

Live

/Dea

d

CD19

CD

3/C

D14

CIT

-PE

RF-

Ale

xa75

0

RF-FITC

RF-

Ale

xa75

0

CIT

-PE

CIT-APCRF-FITC

RF-

Ale

xa75

0

CIT

-PE

29.9 99.2 81.2

52.4 24.6

61.024.0

00

Subset-basedAnalysis

D

Cel

ls p

er 1

06 B c

ells 0

50

100

150

Tet−

AC

PA RF

Tet−

AC

PA RF

Tet−

AC

PA RF

Tet−

AC

PA RF

Tet−

AC

PA RF

Tet−

AC

PA RF

Tet−

AC

PA RF

Cou

nt

SubtypeDNMemoryNaiveASC

HD5 R01 R26 R04 R48 R59 R72*

*

Figure 1. Interrogation of autoreactive B cell subsets through tetramer sorting and transcriptomic classification of single cells. A, Schematic of theexperimental workflow for the tetramer-based isolation, single-cell RNA sequencing, and simultaneous immunoglobulin repertoire and transcriptomicanalyses of B cells. B, Representative flow cytometry gating approach for the identification of B cells producing rheumatoid factors (RFs) or anti–citrullinated protein antibodies (ACPAs). C, Comparison of RF+ and ACPA+ B cell subtype frequencies in peripheral blood between rheumatoidarthritis (RA) patients and healthy donors (HDs). A minimum of 15,000 B cells were analyzed for each sample. * = P < 0.05 versus healthy donors.D, B cell subtype distribution using the BCellNet classifier for the recovered 2,349 single cells from 7 individuals. The ACPA+ and RF+ B cells fromsubject R01 produced poor-quality yields during library preparation and were excluded from analysis. No ACPA+ cells were recovered from subjectsR04 and HD5, and no RF+ cells were recovered from subjects R48 and HD5. PBMCs = peripheral blood mononuclear cells; FACS = fluorescence-activated cell sorting; mAb = monoclonal antibody; CIT+ = citrullinated antigen positive; PE = phycoerythrin; APC = allophycocyanin; FITC =fluorescein isothiocyanate; ASC = antibody-secreting cell; Tet� = tetramer negative; DN = double negative.

MOLECULAR MECHANISMS OF AUTOREACTIVE B CELL RESPONSES IN RA 1735

B

Citrullinated proteinTetramer-negative

A

0.1 1 10 1000.0

0.2

0.4

0.6

0.8

Concentration (µg/ml)

A (4

50 n

m)

0.1 1 10 1000.0

0.2

0.4

0.6

0.8

Concentration (µg/ml)

A (4

50 n

m)

0.1 1 10 1000.0

0.2

0.4

0.6

0.8

Concentration (µg/ml)

A (4

50 n

m)

0.1 1 10 1000.0

0.5

1.0

1.5

2.0

Concentration (µg/ml)

A (4

50 n

m)

0.1 1 10 1000.0

0.5

1.0

1.5

2.0

Concentration (µg/ml)

A (4

50 n

m)

0.1 1 10 1000.0

0.5

1.0

1.5

2.0

Concentration (µg/ml)

A (4

50 n

m)

D 1000 100 50 25 15 10 5 1

MFIACPA Tet-ctrl

Naive ASCC

Naive ASC

Cyc

lic C

itrul

linat

edP

eptid

e 3

Rhe

umat

oid

Fact

or

Memory

Memory

R01

Tet

-

R26

Tet

-

R26

AC

PA

R26

RF

R48

Tet

-

R48

AC

PA

R04

Tet

-

R04

RF

R59

Tet

-

R59

AC

PA

R59

RF

R72

Tet

-

R72

AC

PA

R72

RF

HD

5 Te

t-

IGHV5−51 − IGLV3−25IGHV5−51 − IGKV2−28IGHV4−61 − IGKV2−28IGHV4−59 − IGKV4−1IGHV4−59 − IGKV3−15IGHV4−59 − IGKV1−5IGHV4−59 − IGKV1−39IGHV4−4 − IGKV1−5IGHV4−39 − IGKV4−1IGHV4−39 − IGKV3−11IGHV4−39 − IGKV2−28IGHV4−34 − IGKV2−28IGHV3−NL1 − IGKV3−15IGHV3−7 − IGKV4−1IGHV3−7 − IGKV3−20IGHV3−7 − IGKV3−15IGHV3−7 − IGKV1−39IGHV3−53/66 − IGKV3−20IGHV3−53 − IGKV3−20IGHV3−53 − IGKV3−11IGHV3−33 − IGKV2−28IGHV3−30/33rn − IGKV4−1IGHV3−30/33rn − IGKV3−15IGHV3−30/33rn − IGKV3−11IGHV3−30/33rn − IGKV2−28IGHV3−30/33rn − IGKV1−5IGHV3−30 − IGKV4−1IGHV3−30 − IGKV3−20IGHV3−30 − IGKV3−15IGHV3−30 − IGKV3−11IGHV3−30 − IGKV2−28IGHV3−30 − IGKV1−5IGHV3−30 − IGKV1−39IGHV3−23 − IGKV4−1IGHV3−23 − IGKV3−20IGHV3−23 − IGKV3−15IGHV3−23 − IGKV3−11IGHV3−23 − IGKV2−28IGHV3−23 − IGKV1−5IGHV3−23 − IGKV1−39IGHV3−23 − IGKV1−27IGHV3−23 − IGKV1−13IGHV3−15 − IGKV4−1IGHV3−15 − IGKV3−15IGHV3−15 − IGKV3−11IGHV3−15 − IGKV1−5IGHV1−69 − IGKV4−1IGHV1−46 − IGKV3−20IGHV1−46 − IGKV3−15IGHV1−46 − IGKV3−11IGHV1−2 − IGKV4−1IGHV1−2 − IGKV3−15IGHV1−2 − IGKV3−11IGHV1−2 − IGKV1−5IGHV1−18 − IGKV4−1IGHV1−18 − IGKV3−20IGHV1−18 − IGKV3−15IGHV1−18 − IGKV3−11IGHV1−18 − IGKV2−28

IGH

V1IG

HV3

IGH

V4IG

HV5

Tet-

CIT

+

RF+

RF+

RF+

RF+

Tet-

Tet-

Tet-

Tet-

Tet-

Tet-

CIT

+

CIT

+

CIT

+

020 Sequence count

RF mAbACPA mAbTet- mAb

H2B

mA

b

Flu

26

RA

30

26c

370

48c

208

48c

225

48c

241

48c

243

48c

297

48c

312

59c

226

59c

271

59c

367

26c

300

26c

375

26c

395

48c

212

48c

321

48c

285

48c

289

26c

384

59c

219

72c

323

72c

379

72c

392

48c

288

48c

280

48c

255

48c

227

26c

288

26c

363

72c

361

26c

237

26a

344

26a

369

HD

5a28

9 2

6a38

9 0

4a24

1 5

9a28

1

H2B cit H2B ApoA1 231-248 cit1 Apo A1 cit Apo A1

a-enolase 2-21 cit2 a-enolase 414-433 a-enolase 414-433 cit2

Biglycan 247-266 cit3

hFibA 616-635 cit3 small cyc

H2B/a 62-81 cit2

a-enolase 2-21 Filaggrin 48-65 cit2 v1 cyclic

hFibA 172-191 cit4

Fibromodulin 345-364 cit2

hFibA 172-191

hFibA 211-230 cit2 hFibA 211-230

hFibA 616-635

Fibromodulin 345-364 Fibromodulin 186-205 cit Fibromodulin 186-205 Tenascin-C 122-141 cit2 Tenascin-C 122-141 Keratin, cit Keratin Vim 181-200 cit3 Vim 181-200

H2B/a 62-81 Lumican 198-217 cit Lumican 198-217

Biglycan 247-266 hFibA 541-560 HCgp39 7-22

Rheumatoid factor

Figure 2. B cells positive for rheumatoid factor (RF) and anti–citrullinated protein antibody (ACPA) exhibit distinct immunoglobulin gene usageand third complementarity-determining region (CDR3) convergence. A, Heatmap showing the most frequently used VH–VL pairs for the citrulli-nated antigen–positive (CIT+), RF+, and tetramer-negative (Tet�) populations in each patient. Column labels denote antigen-specific populationsarranged by rheumatoid arthritis (RA) patients and 1 healthy donor (HD). Row labels denote VH–VL pairs arranged by VH gene family. Redboxes identify prominent VH–VL pairs from which recombinant antibodies were expressed. B, Greedy, agglomerative clustering of single B cellswith heavy chain CDR3 and light chain CDR3 sequences sharing >80% homology. Each cell, colored by antigen specificity, is a node. Gray edgesconnect cells belonging to the same cluster. Prominent clusters from which recombinant antibodies were expressed are highlighted in blue. C, Rep-resentative results of enzyme-linked immunosorbent assay (n = 3 replicates) against RF and cyclic citrullinated peptide 3 for all recombinant anti-bodies, measured using absorbance values and plotted by B cell subtype as determined by BCellNet. The dashed green line denotes a positivecontrol monoclonal antibody (mAb). The horizontal line denotes the threshold for positive reactivity. D, Top citrullinated autoantigen microarrayhits for expressed monoclonal antibodies. Column labels denote expressed monoclonal antibodies. Flu 26 represents a control patient who receivedthe influenza vaccine 7 days prior to blood collection. Row labels denote both citrullinated and native peptides tested. Binding intensity is shownfor both citrullinated peptides and available corresponding noncitrullinated antigens. ASC = antibody-secreting cell; H2B = H2B histone; Apo A-I= apolipoprotein A-I; hFib A = human fibrinogen alpha; HC gp-39 = human cartilage gp-39; MFI = mean fluorescence intensity.

1736 LU ET AL

activated cell sorting. We also observed an enrichment inthe classification of memory cells through transcriptomeanalysis (Figure 1D). The BCellNet classifier accountedfor much of the variability inherent to single-cell RNA-Seq.The remaining bias in classification through this methodmost likely did not have a significant impact on our furtheranalyses of these cells (see Supplementary Methods, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract).

BCR analysis reveals CDR3 convergence andautoantigen targets. Next, we recovered paired BCR vari-able-region genes expressed by individual B cells, enablingintegrated analysis of affinity maturation between autoreac-tive populations and functional characterization of recombi-nant antibodies expressed from reconstituted VH/VL

sequences (see Supplementary Figure 5 and SupplementaryTable 5, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract). To investigate how affinity maturation shapes Bcell responses against the autoantigens, we clustered RF+,ACPA+, and tetramer-negative B cells sharing >80% Leven-shtein similarities in their CDR3 sequences (Figure 2B; alsosee Supplementary Figure 6, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract). This represented the moststringent threshold that preserved cluster membershipamong cells that exhibit identical germline genes andCDR3 lengths and that therefore putatively derive from thesame progenitor (23) (see Supplementary Table 6, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract).

To determine if the size of the CDR3 clusters canmeasure the magnitude of convergent evolution, we com-pared the number of cells per tetramer group in each

CDR3 cluster to calculate the cluster multiplicity. If eachcell from a tetramer group clustered separately, then themultiplicity would be 1, while if all N cells were in a singlecluster, the multiplicity would be N. ACPA+ B cell multi-plicities were significantly larger than tetramer-negative Bcell multiplicities in 2 of 4 patients (multiplicity differ-ences 1.3 and 0.3 cells/cluster; P < 0.002 under hyper-geometric sampling) and significantly larger than RF+ Bcell multiplicities in 1 of 3 patients (multiplicity difference0.17 cells/cluster; P < 0.04). No significant differenceswere observed between RF+ B cell multiplicities and tet-ramer-negative B cell multiplicities. Hence, the elevatedfrequency of CDR3 convergence suggests that ACPA+ Bcells undergo extensive affinity maturation relative to tet-ramer-negative and RF+ B cell populations.

We then expressed 42 representative RF andACPA BCRs from the most prominent VH–VL pairingsand CDR3 clusters as recombinant monoclonal antibod-ies to identify their autoantigen targets and examinewhether autoreactivity is present in antigen-inexperiencedand -experienced B cell stages (Figures 2A and B; alsosee Supplementary Table 7, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract). Eleven of the 12antibodies cloned from RF-sorted cells bound to humanIgG Fc, while only 1 of the 30 antibodies from ACPA-sorted cells bound to IgG Fc, and none of the antibodiesfrom tetramer-negative B cells bound to IgG Fc (Fig-ure 2C). We also found that 13 of the 30 monoclonal anti-bodies from citrullinated antigen–sorted cells bound toCCP3, a citrullinated peptide mimic used to diagnose RA.

A DCB

Tet- ACPA RF0

25

50

75

100

UnswitchedClass-switched

Antigen-ExperiencedB cells

(Memory + DN + ASC)Antigen-Experienced

B cells (Memory + DN + ASC)

Antigen-ExperiencedB cells

(Memory + DN + ASC)

***

** n.s.***

n.s. ***

ACPARF

Tet-

R01R04R26R48R59R72

AA

Mut

atio

n R

ate

0.090.100.110.120.130.140.150.16

Tet- ACPA RF

Cla

ss-S

witc

hed

Rat

e

0.10.20.30.40.50.60.70.8

ACPA Tet- RFTet-

ACPA RFTet- RF

Tet- RFTet- RF

Tet- RF0.0

0.1

0.2

0.3

0.4

0.5

AA

Mut

atio

n R

ate

IgG/ADN

IgG/A/EMemory

IgM/DMemory

IgMASC

IgG/AASC

% o

f pop

ulat

ion

ACPAACPA

ACPAACPA

Figure 3. Antigen-experienced B cells positive for rheumatoid factor (RF) and anti–citrullinated protein antibody (ACPA) differ in class-switchand somatic hypermutation rates. A, Proportion of antigen-experienced B cells (i.e., unswitched [IgM/D] and class-switched [IgG/A/E] memory,double-negative [DN] memory, and antibody-secreting cells [ASCs]) for each antigen specificity. B, Class-switch rates for antigen-experiencedACPA+, RF+, and tetramer-negative (Tet�) B cells in rheumatoid arthritis patients, adjusting for subset differences and patient variability using ageneralized linear mixed model. C, Somatic mutation rates for antigen-experienced ACPA+, RF+, and tetramer-negative B cells by subtype. D,Somatic mutation rates for immunoglobulins expressed by ACPA+, RF+, and tetramer-negative B cells, adjusting for subset differences and patient-to-patient variability using a linear mixed model. In B and D, values are the mean and 95% confidence interval. In C, values are the mean � SEM.* = P < 0.05; ** = P < 0.01; *** = P < 0.001. NS = not significant; AA = amino acid.

MOLECULAR MECHANISMS OF AUTOREACTIVE B CELL RESPONSES IN RA 1737

None of the antibodies from tetramer-negative B cellsrecognized CCP3, and only 2 of the 12 antibodies fromRF-sorted cells bound CCP3 (Figure 2C). Using theCCP3 ELISA and a protein microarray coated with >300RA autoantigens, we identified citrullinated antigen hitsfor 25 of the 30 antibodies from citrullinated antigen–sorted cells (Figure 2D). We do note that several antibod-ies cloned from ACPA+ B cells were cross-reactive withnoncitrullinated antigens, consistent with previous reports(9). This confirms that the RF–tetramer and ACPA–tet-ramer reagents enabled isolation of RF+ and ACPA+ Bcells. Notably, RF+ and ACPA+ B cells are in all subsets,including non–affinity matured cells, implicating dysfunc-tional central and peripheral tolerance in RA.

RF+ and ACPA+ B cells differ in class-switch andsomatic mutation frequencies. We next investigatedwhether the duration of affinity maturation differsbetween ACPA+ and RF+ B cells, as measured throughclass-switch recombination (CSR) rates and accumulatedsomatic mutations on the BCR. Because CSR and muta-tion rates vary between subtypes, which themselves vary byspecificity, we used a mixed-effects logistic regression tomeasure the unique effect of antigen on CSR and muta-tion rates adjusted for subtype. We found that ACPA+ Bcells had >3.8-fold greater odds of being class-switchedthan tetramer-negative B cells and 3.7-fold greater oddscompared to RF+ B cells, but there was no differencebetween tetramer-negative and RF+ B cells (Figures 3A

B

0.0

2.5

5.0

7.5

10.0

−log

(FD

R)

0

5

10

15

C

DN

Mem

ory

Nai

veA

SC

All differentially expressed transcripts

All

sing

le c

ells

AZ-score

3210−1

SpecificityTet-ACPARF

pSubtype10.80.60.40.2

SubjectHD5R01R26R04R48R59R72

SHM20−2−4−6

CDRate80006000400020000

−log

(FD

R)

−2.5 0.0 2.5Function Score (vs Tet-)

−5.0 −2.5 0.0 2.5 5.0Function Score (vs Tet-)

Antibody ResponseAntibody Response

B-cell Activation

Cytoskeletal

Germinal CenterInnate Immune

Metabolism

Transcription

Antibody ResponseAntigen Processing

B-cell Activation

Cytoskeletal

Immunosuppression

Innate Immune

Proliferation

Protein Processing

ACPA vs Tet- Among Subtypes

RF vs Tet- Among Subtypes

SubtypeDNMemoryNaiveASC

SubtypeDNMemoryNaiveASC

Q=0.05

Q=0.001

Q=0.05

Q=0.001

Spe

cific

itypS

ubty

peS

ubje

ctS

HM

CD

Rat

e

Figure 4. Aberrant primary and secondary autoreactive B cell gene expression signatures in rheumatoid arthritis (RA). A, Subset-stratified heatmapshowing the Z-scaled partial residual expression of the differentially expressed genes for all 2,349 B cells from 6 RA patients and 1 healthy donor(HD). Of the 454 differentially expressed genes, 121 are known to be immune related. B and C, Volcano plots showing functional modules differen-tially expressed between anti–citrullinated protein antibody–positive (ACPA+) B cells and tetramer-negative (Tet�) B cells (B) and between rheuma-toid factor–positive (RF+) B cells and tetramer-negative B cells (C) for all B cell subtypes. Modules consist of the average partial residuals from genesthat share cellular functions. Horizontal dashed lines denote false discovery rate (FDR)–adjusted P value (q value) cutoffs. ASC = antibody-secretingcell; DN = double negative; pSubtype = probability of subtype; SHM = somatic hypermutation; CDRate = cellular detection rate.

1738 LU ET AL

and B). We also found that RF+ B cells were 3% lessmutated from germline than tetramer-negative B cells,while ACPA+ B cells had 1% higher mutation rates thantetramer-negative B cells, although the difference wasnot significant (Figures 3C and D). Thus, ACPA+ B cellsundergo more isotype class-switching, have higher muta-tion rates, and show evidence of increased CDR3 con-vergence, while RF+ B cells tend to use unswitchedisotypes and have fewer mutations.

Functional differential regulation in the primaryand secondary RF+ and ACPA+ B cell responses. To elu-cidate the transcriptional profiles underlying the BCR dif-ferences between RF+ and ACPA+ B cell responses, weapplied a hurdle model to identify 454 DEGs betweenRF+, ACPA+, and tetramer-negative B cell populationsfor all patients in global (i.e., regressing out the effect ofsubset on gene expression) and subset-stratified compar-isons (Figure 4A; also see Supplementary Tables 8 and 9,

AB−cell Activation (25)

Naive Memory DN ASC

* * ** *

−5

0

5

LAPTM5KMT2DJAK3CD84EGFJAM2GDE1ITGB2TNFRSF1BMEF2CRELSELLFAM129CHERC3ASRGL1SOX5NAMPTCYFIP2BACH1BACH2PIGMDBN1SOX11

Nai

ve

Mem

ory

ACPA−RF Z1050−5−10

B−cell ActivationGenes

Antibody ResponseGenes

C

B

D

Antibody Response (15)

−5

0

5

Naive Memory DN ASC

* * * *

Func

tion

Sco

re

Func

tion

Sco

re

ACPA−RF Z1050−5−10

EXOC4JAK3MSH5BEND7HSP90AA1KMT2DCD72TBRG4CXCR4FCER2GNG7CD320CCND2TRAM2ETV6

Nai

ve

Mem

ory

Tet−

ACPA

RF

Tet−

ACPA

RF

Figure 5. Anti–citrullinated protein antibody–positive (ACPA+) B cells and rheumatoid factor–positive (RF+) B cells exhibit distinct transcrip-tional programs that promote antibody class-switching and B cell activation. A, Subset-stratified plot of average partial residual expression in theantibody response module (15 genes). * = P < 0.05. B, Subset-stratified plot of average partial residual expression in the B cell activation module(25 genes). * = P < 0.05. C, Z score testing for differential expression between ACPA+ and RF+ B cells in antibody response genes. D, Z scoretesting for differential expression between ACPA+ and RF+ B cells in B cell activation genes. Tet� = tetramer negative; DN = double negative;ASC = antibody-secreting cell.

MOLECULAR MECHANISMS OF AUTOREACTIVE B CELL RESPONSES IN RA 1739

http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract).Performing both global and subset-stratified analysesallowed us to identify a greater amount of DEGs thanwould be possible using either method alone. Both genesets were relevant to identifying autoreactive B cell signa-tures. To examine the cumulative functional impact of theDEGs, we organized DEGs into modules based upon pre-viouslypublished literature. The partial residual for each gene,which represents residual expression attributable to antigenspecificity, was used to calculate a module score forACPA+, RF+, and tetramer-negative cells for each sub-type (see Supplementary Figure 7, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract), pinpointing thedevelopmental stage(s) at which differential autoimmuneregulation occurs. The most significantly different mod-ules were concentrated within the naive and memorycompartments (Figures 4B and C).

ACPA+ B cell primary responses initiate class-switch programs and down-regulate CD72. In the naiveACPA+ B cell population, module scores for antibodyresponse and B cell activation were elevated relative tothe RF+ and tetramer-negative naive B cell populations(Figures 5A and B). Within these modules, the genesMSH5, JAK3, and EXOC4, which all promote antibodyclass-switched responses (24–26), were highly elevated inthe naive ACPA+ B cell compartment and sustained inmemory ACPA+ B cells (Figure 5C). Additionally, CD72,which inhibits the differentiation of resting B cells, wasdown-regulated in naive and memory ACPA+ populationsbut not in RF+ cells (27). In the B cell activation module,we detected naive ACPA+ B cell elevation of the adhe-sion molecule genes CD84, JAM2, and ITGB2, whichprolong B cell–T cell interactions and thereby promoteaffinity maturation (28,29) (Figure 5D). Taken together,these signatures suggest that ACPA+ B cell responsesare programmed for class-switching through T cell helpand facilitated by CD72 down-regulation in the naivestage.

RF+ B cells express transcriptional programsassociated with IgM responses and rapid recall re-sponses.Within the B cell activation module, the transcrip-tion factor SOX11, which promotes highly proliferativeIgM responses and suppresses germinal center transcrip-tion factor BCL6, and DBN1, a SOX11-regulated gene,were elevated in memory RF+ B cells relative to memoryACPA+ B cells and tetramer-negative B cells (30,31)(Figure 5D). The transcriptional regulator BACH2 wasalso significantly up-regulated in memory RF+ B cellsrelative to memory ACPA+ B cells, which is a feature ofunswitched B cell responses (32,33) (Figure 5D). Theselective up-regulation of SOX11 and BACH2 in memory

RF+ B cells provides an explanation for their low class-switch and mutation rates.

Differences in B cell costimulation drive ACPAgeneration and RF versatility to multiple innate activa-tion pathways. Innate immune mediators provide costim-ulatory signals that can augment B cell responses (34). Inthe memory compartment, RF+ and ACPA+ cells exhib-ited differentially expressed subsets of genes despite nothaving higher average module scores relative to tetramer-negative B cells (Figures 6A and B). RF+ memory B cellsexhibited altered expression of signature genes involvedin myeloid differentiation factor 88 (MyD88) signaling(LRRFIP1), interferon (IFN) signaling (IFI6, PSMB10),and DNAX-activation protein 12 (DAP-12) signaling(SIGLEC14), none of which were enriched in ACPA+memory B cells (Figure 6B). In contrast, ACPA+ memoryB cells showed coordinated up-regulation of positive regu-lators of NF-jB signaling (REL and HSP90AA1) (Fig-ure 6B), which is critical for promoting germinal centerresponses and survival (35).

Distinct antibody response and innate immunetranscriptional regulatory networks characterize ACPA+and RF+ B cells. We analyzed single-cell gene coexpres-sion to identify transcriptional regulatory networks withingenes differentially expressed by ACPA+ and RF+ B cellsand to provide insights into the mechanisms underlyingthe loss of tolerance in ACPA+ B cells as compared toRF+ B cells (see Supplementary Table 10, http://onlinelibrary.wiley.com/doi/10.1002/art.40578/abstract). In boththe primary response and memory compartments, ACPA+B cells coexpressed MSH5, CD84, and JAK3 (Figures 6Cand D), revealing a coregulated transcriptional networkspecific to ACPA development. We also investigatedwhether upstream effectors of JAK3 signaling were coex-pressed with JAK3. IL21R, which receives signals from fol-licular helper T cells, was coexpressed with the MSH5/CD84/JAK3 network in ACPA+ memory B cells (36,37)(Figure 6D). Coexpression of these genes suggests thatJAK3 links germinal center signals with class-switch func-tions in both ACPA+ B cell primary and secondaryresponses. Thus, antigen-experienced ACPA+ B cells exhi-bit a transcriptional program characterized by a robust ger-minal center response.

We found evidence that RF+ B cells coexpressedgenes belonging to multiple B cell response pathways.The innate immune genes PSMB10, LRRFIP1, IFI6,and SIGLEC14 were all coexpressed, implying thatindividual memory RF+ B cells exhibit coordinated up-regulation of genes (Figure 6D). Further, we foundthat these genes were all significantly coexpressed withPRDM1 and CCND2. The coexpression among theseinnate immune genes, SOX11, and markers of ASC

1740 LU ET AL

C

D

Enriched in RFEnriched in ACPANot enriched

Coexpression (Z)

-10 10

JAK3

MSH5

DBN1

GDE1

MYOG

PIGM

SOX11

LAPTM5

REL

CCND2

PRDM1

LRRFIP1

IFI6

PSMB10

TNFRSF1B

IL17RD

TRAF6

SIGLEC14

CD84

AICDAIL21R

IL21R

JAK3MSH5

DBN1MYOG

SOX11

IL17RD

TRAF6

PIGM

LRRFIP1

CD84

AICDA

LAPTM5

GDE1

Memory

Naive

Enriched in RFEnriched in ACPANot enriched

Coexpression (Z)

-10 10

P2RY2EFNA1TMEM181HSP90AA1TCL1AIL17RDTLR7PRKRIRADAM10PSMB10LRRFIP1HSP90AB1ARRB2UBE2V1SARNPMEGF9F5BPIFB1TRIM56RELVSIG10SAMD9TRIM73MGST3PTGER4TXLNGNKG7MLLT4ITGB2P2RY10F2RL2SRP72PYGLBRCA2TNFRSF1BARHGEF10SIGLEC14IL18RAPIFI6MUC5BUBE2D1ITGB3BPGZMHTNFAIP2CD2GZMBGRN

Nai

ve

Mem

ory

ACPA−RF Z1050−5−10

Innate ImmuneGenes

A

B

−5

0

5

Tet−

ACPA

RF

Innate Immune (47)

Naive Memory DN ASC

* * *

Func

tion

Sco

re

Figure 6. Innate immune mediators are enriched in rheumatoid factor–positive (RF+) B cells and comprise coexpression networks distinct fromthose of genes enriched in anti–citrullinated protein antibody–positive (ACPA+) B cells. A, Subset-stratified plot of average partial residual expres-sion in the innate immune module (47 genes). * = P < 0.05. B, Z score testing for differential expression between ACPA+ and RF+ B cells ininnate immune genes. C and D, Coexpression networks of selected module genes, stratified by subset. Genes enriched either in ACPA+ or in RF+B cells have differential expression Z scores >1 relative to the corresponding RF+ or ACPA+ B cell subsets, respectively, and expression greaterthan or equal to that in the tetramer-negative (Tet�) population. The color and weight of the edges indicate the Z score of the Pearson correla-tion coefficient for pairs of genes. Correlations with false discovery rate–adjusted P values greater than 0.05 are omitted. C, Network displayingcoexpression of B cell activation and innate immune genes involved in either ACPA+ or RF+ naive B cell responses. D, Network displaying coex-pression of B cell activation and innate immune genes involved in either ACPA+ or RF+ memory B cell responses. DN = double negative; ASC =antibody-secreting cell.

MOLECULAR MECHANISMS OF AUTOREACTIVE B CELL RESPONSES IN RA 1741

differentiation suggests that innate immune genes up-regulated in RF+ memory B cells help prime rapidrecall responses. Therefore, in contrast to ACPA+ Bcells, RF+ B cells up-regulate transcriptional programsthat are involved in innate immune response pathwaysand that facilitate rapid recall responses.

DISCUSSION

Accumulating clinical evidence implicates a patho-genic role of B cells in RA (3,38,39), and therapeuticmodalities that specifically target dysregulated pathwaysin autoreactive B cell development will be critical for sus-tained remission. In the present study, we investigated thedifferential regulation of B cell tolerance in RA usingantigen–tetramers to isolate disease-relevant B cells, andwe used single-cell RNA-Seq to define RF- and ACPA-specific immunoglobulin usage and transcriptional pro-grams. Our findings demonstrate that ACPA+ and RF+ Bcells are imprinted with distinct transcriptional programs,suggesting that different molecular mechanisms mediatethe loss of tolerance in ACPA+ B cells versus RF+ B cells.

While self-reactive B cell precursors leak into theperiphery at a low frequency in healthy individuals (40),the nearly exclusive detection of CD27�IgM+IgD+ACPA+B cells and the elevation of CD27�IgM+IgD+RF+ B cellsin RA patients indicate that ACPA+/RF+ B cell precursorsare predominantly deleted in healthy individuals but notin RA patients. This suggests a defect of central toler-ance in RA that is likely due to a combination of geneticand environmental factors. Defective regulation throughpolymorphisms in BCR signaling genes as well as alteredreceptor editing have been previously reported and maypermit the survival of autoreactive immature B cells thatare normally removed (41). Systemic inflammatory medi-ators can also promote the survival of autoreactive Bcells by overwhelming apoptotic or receptor-editing pro-cesses (42). While our data indicate that differential reg-ulation of ACPA+ B cells begins at the naive stage,future studies will be needed to decipher whether thisoccurs in the bone marrow or in the periphery.

The development of ACPAs appears to stemfrom multiple defective tolerance mechanisms that failto restrain ACPA+ B cell activation and from promo-tion by autoreactive T cell help. In support of this, astudy by Makrygiannakis and colleagues demonstratedthat citrullinated protein levels are increased in bothRA and non-RA inflamed tissues, yet only RA patientspersistently develop ACPAs (43). We found that CD72,an inhibitory coreceptor, was down-regulated in naiveACPA+ B cells relative to RF+ and tetramer-negativeB cells. This may enable transduction of activating B

cell signals that override tolerogenic signals received byACPA+ B cells (44). In addition, expression of thecoreceptor CD21 was down-regulated between naiveACPA+ and tetramer-negative B cells, although thiseffect was not significant in our study.

Once activated, ACPA+ B cells undergo extensiveaffinity maturation as evidenced by high mutation rates,class-switching rates, and CDR convergence that areassociated with expression of robust transcriptional pro-grams reminiscent of those in germinal centers. The ele-vation of JAK3 in ACPA+ B cells suggests dependenceon the IL21R/JAK3 pathways for ACPA production andprovides a potential explanation for why tofacitinib,which targets JAK3, results in lower remission rates inACPA+ patients (45). Further emphasizing the impor-tance of T cell help, we found an up-regulation in genesthat prolong B cell–T cell interactions (CD84, JAM2,and ITGB2) and in molecules that promote antibodyclass-switching (MSH5). While these transcriptional sig-natures suggest that affinity maturation of ACPA+ Bcells occurs in germinal centers, future studies will benecessary to determine whether this occurs in conven-tional germinal center reactions, through ectopic T cellhelp, or through other mechanisms.

Identification of shared transcriptional programsfor ACPA+ B cells that target diverse citrullinated pro-teins suggests that exposure to a spectrum of citrullinatedproteins rather than to a specific citrullinated epitopedrives the ACPA response, consistent with a previousstudy that examined ACPA formation during relapse (38).Hence, we propose that differential B cell imprinting,enhanced inflammation, and citrullinated protein avail-ability combinatorially create an environment conduciveto ACPA+ B cell activation and autoantibody production.

In contrast to the ACPA+ B cell response, theRF+ B cell response is characterized by transcriptionalprograms associated with broad innate immune activa-tion, rapid recall responses, and ASC differentiation.Unlike the extensive affinity maturation driving theACPA response, the RF response is characterized by Bcells that express IgM and exhibit lower mutation rates.The bias toward the IgM isotype, which can bind up to10 IgG molecules in its pentameric form, enables RFresponses to aggregate secreted antibodies and immunecomplexes. The low mutation rates suggest that certainV(D)J combinations may be intrinsically reactive withIgG Fc. While T cell help causes ACPA+ B cells toundergo prolonged affinity maturation, RF+ B cells candifferentiate into memory B cells or ASCs with minimalsomatic hypermutation.

Our data indicate that RF+ B cells mount rapidsecondary responses with costimulation from multiple

1742 LU ET AL

innate immune pathways (e.g., MyD88, IFN, DAP-12),which suggests that RF+ B cells are primed to expand inresponse to diverse inflammatory stimuli. Up-regulationof these signatures in RF+ memory B cells correlates withmarkers of ASCs, implying that RF memory B cell differ-entiation generates a reservoir of cells that can mountrapid recall responses.

The etiology of RF production in multiple inflam-matory contexts remains an area of active investigation.Given the presence of RF in other inflammatory diseasesand its relatively lower specificity for diagnosing RA (4,5),our results suggest that RF could represent a response toresolve antibody-mediated inflammation by facilitatingthe clearance of antibodies and immune complexes (46).Thus, RF elevations in RA may represent an expandedpool of naturally occurring B cells for the purpose ofengulfing antigen–antibody complexes to enable antigenpresentation to Tcells. Future studies comparing the geneexpression of RF+ B cells between individuals with andthose without RA will be important to provide furtherinsights into whether RF in fact results from dysregulatedB cell tolerance.

Our findings reveal that ACPA+ and RF+ Bcells are imprinted with distinct transcriptional pro-grams at different stages of development, with RF+ Bcells characterized by activation of innate immune acti-vation pathways, while ACPA+ B cells are character-ized by loss of tolerance in the primary responsefollowed by robust affinity-matured responses. Futurestudies will further define how ACPA+ and RF+ Bcells promote tissue damage in RA, as well as theimpact of targeted therapies on ACPA+ and RF+ Bcell activation.

ACKNOWLEDGMENTS

We thank S. Elliott, A. Gomez, L. Blum, M. S. Bloom,and Z. Weng for insightful discussions and input.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising itcritically for important intellectual content, and all authors approvedthe final version to be published. Dr. Robinson had full access to allof the data in the study and takes responsibility for the integrity ofthe data and the accuracy of the data analysis.Study conception and design. Lu, McDavid, Gottardo, Robinson.Acquisition of data. Lu, Kongpachith, Lingampalli, Ju.Analysis and interpretation of data. Lu, McDavid, Kongpachith,Lingampalli, Glanville, Ju, Gottardo, Robinson.

REFERENCES

1. McInnes IB, Schett G. The pathogenesis of rheumatoid arthritis.N Engl J Med 2011;365:2205–19.

2. Sokolove J, Johnson DS, Lahey LJ, Wagner CA, Cheng D, ThieleGM, et al. Rheumatoid factor as a potentiator of anti-citrullinatedprotein antibody–mediated inflammation in rheumatoid arthritis.Arthritis Rheumatol 2014;66:813–21.

3. Edwards JC, Szczepanski L, Szechinski J, Filipowicz-SosnowskaA, Emery P, Close DR, et al. Efficacy of B-cell-targeted therapywith rituximab in patients with rheumatoid arthritis. N Engl JMed 2004;350:2572–81.

4. Witte T, Hartung K, Sachse C, Matthias T, Fricke M, Kalden JR,et al, SLE Study Group. Rheumatoid factors in systemic lupuserythematosus: association with clinical and laboratory parame-ters. Rheumatol Int 2000;19:107–11.

5. Salonen EM, Vaheri A, Suni J, Wager O. Rheumatoid factor inacute viral infections: interference with determination of IgM,IgG, and IgA antibodies in an enzyme immunoassay. J Infect Dis1980;142:250–5.

6. Izui S, Eisenberg RA, Dixon FJ. IgM rheumatoid factors in miceinjected with bacterial lipopolysaccharides. J Immunol 1979;122:2096–102.

7. Kono DH, Haraldsson MK, Lawson BR, Pollard KM, Koh YT,Du X, et al. Endosomal TLR signaling is required for anti-nucleic acid and rheumatoid factor autoantibodies in lupus. ProcNatl Acad Sci U S A 2009;106:12061–6.

8. Kuhn KA, Kulik L, Tomooka B, Braschler KJ, Arend WP, RobinsonWH, et al. Antibodies against citrullinated proteins enhance tissueinjury in experimental autoimmune arthritis. J Clin Invest 2006;116:961–73.

9. Tan YC, Kongpachith S, Blum LK, Ju CH, Lahey LJ, Lu DR, et al.Barcode-enabled sequencing of plasmablast antibody repertoires inrheumatoid arthritis. Arthritis Rheumatol 2014;66:2706–15.

10. Kerkman PF, Fabre E, van der Voort EI, Zaldumbide A, RomboutsY, Rispens T, et al. Identification and characterisation of citrulli-nated antigen-specific B cells in peripheral blood of patients withrheumatoid arthritis. Ann Rheum Dis 2016;75:1170–6.

11. Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF,Cooper NS, et al. The American Rheumatism Association 1987revised criteria for the classification of rheumatoid arthritis.Arthritis Rheum 1988;31:315–24.

12. Sokolove J, Bromberg R, Deane KD, Lahey LJ, Derber LA,Chandra PE, et al. Autoantibody epitope spreading in the pre-clinical phase predicts progression to rheumatoid arthritis. PloSOne 2012;7:e35296.

13. Islam S, Kj€allquist U, Moliner A, Zajac P, Fan JB, L€onnerberg P,et al. Highly multiplexed and strand-specific single-cell RNA 50end sequencing. Nat Protoc 2012;7:813–28.

14. Kent WJ, Sugnet CW, Furey TS, Roskin KM, Pringle TH, ZahlerAM, et al. The Human Genome Browser at UCSC. Genome Res2002;12:996–1006.

15. Langmead B, Salzberg SL. Fast gapped-read alignment withBowtie 2. Nat Methods 2012;9:357–9.

16. Li B, Dewey CN. RSEM: accurate transcript quantification fromRNA-Seq data with or without a reference genome. BMC Bioin-formatics 2011;12:323.

17. Finak G, McDavid A, Yajima M, Deng J, Gersuk V, Shalek AK,et al. MAST: a flexible statistical framework for assessing tran-scriptional changes and characterizing heterogeneity in single-cellRNA sequencing data. Genome Biol 2015;16:278.

18. Doorenspleet ME, Klarenbeek PL, de Hair MJ, van Schaik BD,Esveldt RE, van Kampen AH, et al. Rheumatoid arthritis sy-novial tissue harbours dominant B-cell and plasma-cell clones asso-ciated with autoreactivity. Ann Rheum Dis 2014;73:756–62.

19. Samuels J, Ng YS, Coupillaud C, Paget D, Meffre E. Impairedearly B cell tolerance in patients with rheumatoid arthritis. J ExpMed 2005;201:1659–67.

20. Menard L, Samuels J, Ng YS, Meffre E. Inflammation-independentdefective early B cell tolerance checkpoints in rheumatoid arthritis.Arthritis Rheum 2011;63:1237–45.

21. Bendall SC, Davis KL, Amir ED, Tadmor MD, Simonds EF,Chen TJ, et al. Single-cell trajectory detection uncovers

MOLECULAR MECHANISMS OF AUTOREACTIVE B CELL RESPONSES IN RA 1743

progression and regulatory coordination in human B cell develop-ment. Cell 2014;157:714–25.

22. Kaminski DA, Wei C, Qian Y, Rosenberg AF, Sanz I. Advancesin human B cell phenotypic profiling. Front Immunol 2012;3:302.

23. Huang Y, Niu B, Gao Y, Fu L, Li W. CD-HIT Suite: a web serverfor clustering and comparing biological sequences. Bioinformatics2010;26:680–2.

24. Sekine H, Ferreira RC, Pan-Hammarstr€om Q, Graham RR,Ziemba B, de Vries SS, et al. Role for Msh5 in the regulation ofIg class switch recombination. Proc Natl Acad Sci U S A 2007;104:7193–8.

25. O’Shea JJ, Gadina M, Schreiber RD. Cytokine signaling in 2002:new surprises in the Jak/Stat pathway. Cell 2002;109 Suppl:S121–31.

26. Jabara HH, Buckley RH, Roberts JL, Lefranc G, Loiselet J,Khalil G, et al. Role of JAK3 in CD40-mediated signaling. Blood1998;92:2435–40.

27. Yamazaki T, Nagumo H, Hayashi T, Sugane K, Agematsu K.CD72-mediated suppression of human naive B cell differentiationby down-regulating X-box binding protein 1. Eur J Immunol2005;35:2325–34.

28. Postigo AA, Corb�ı AL, S�anchez-Madrid F, de Land�azuri MO.Regulated expression and function of CD11c/CD18 integrin onhuman B lymphocytes: relation between attachment to fibrinogenand triggering of proliferation through CD11c/CD18. J Exp Med1991;174:1313–22.

29. Wong EB, Soni C, Chan AY, Domeier PP, Shwetank, Abraham T,et al. B cell-intrinsic CD84 and Ly108 maintain germinal centerB cell tolerance. J Immunol 2015;194:4130–43.

30. Wang X, Bj€orklund S, Wasik AM, Grandien A, Andersson P,Kimby E, et al. Gene expression profiling and chromatinimmunoprecipitation identify DBN1, SETMAR and HIG2 asdirect targets of SOX11 in mantle cell lymphoma. PLoS One2010;5:e14085.

31. Larbolette O, Wollscheid B, Schweikert J, Nielsen PJ, WienandsJ. SH3P7 is a cytoskeleton adapter protein and is coupled to sig-nal transduction from lymphocyte antigen receptors. Mol CellBiol 1999;19:1539–46.

32. Kometani K, Nakagawa R, Shinnakasu R, Kaji T, Rybouchkin A,Moriyama S, et al. Repression of the transcription factor Bach2contributes to predisposition of IgG1 memory B cells towardplasma cell differentiation. Immunity 2013;39:136–47.

33. Seifert M, Przekopowitz M, Taudien S, Lollies A, Ronge V, DreesB, et al. Functional capacities of human IgM memory B cells inearly inflammatory responses and secondary germinal center reac-tions. Proc Natl Acad Sci U S A 2015;112:E546–55.

34. Leadbetter EA, Rifkin IR, Hohlbaum AM, Beaudette BC,Shlomchik MJ, Marshak-Rothstein A. Chromatin-IgG complexes

activate B cells by dual engagement of IgM and Toll-like recep-tors. Nature 2002;416:603–7.

35. Gerondakis S, Siebenlist U. Roles of the NF-jB pathway in lym-phocyte development and function. Cold Spring Harb PerspectBiol 2010;2:a000182.

36. Liu R, Wu Q, Su D, Che N, Chen H, Geng L, et al. A regulatoryeffect of IL-21 on T follicular helper-like cell and B cell inrheumatoid arthritis. Arthritis Res Ther 2012;14:R255.

37. Ouchida R, Kurosaki T, Wang JY. A role for lysosomal-associatedprotein transmembrane 5 in the negative regulation of surface B cellreceptor levels and B cell activation. J Immunol 2010;185:294–301.

38. Cambridge G, Leandro MJ, Lahey LJ, Fairhead T, Robinson WH,Sokolove J. B cell depletion with rituximab in patients with rheuma-toid arthritis: multiplex bead array reveals the kinetics of IgG andIgA antibodies to citrullinated antigens. J Autoimmun 2016;70:22–30.

39. Thurlings RM, Teng O, Vos K, Gerlag DM, Aarden L, StapelSO, et al. Clinical response, pharmacokinetics, development ofhuman anti-chimaeric antibodies, and synovial tissue response torituximab treatment in patients with rheumatoid arthritis. AnnRheum Dis 2010;69:409–12.

40. Wardemann H, Yurasov S, Schaefer A, Young JW, Meffre E,Nussenzweig MC. Predominant autoantibody production by earlyhuman B cell precursors. Science 2003;301:1374–7.

41. Begovich AB, Carlton VE, Honigberg LA, Schrodi SJ, ChokkalingamAP, Alexander HC, et al. A missense single-nucleotide polymorphismin a gene encoding a protein tyrosine phosphatase (PTPN22) is asso-ciated with rheumatoid arthritis. Am J Hum Genet 2004;75:330–7.

42. Nagaoka H, Takahashi Y, Hayashi R, Nakamura T, Ishii K,Matsuda J, et al. Ras mediates effector pathways responsible forpre-B cell survival, which is essential for the developmental pro-gression to the late pre-B cell stage. J Exp Med 2000;192:171–82.

43. Makrygiannakis D, af Klint E, Lundberg IE, L€ofberg R, UlfgrenAK, Klareskog L, et al. Citrullination is an inflammation-depen-dent process. Ann Rheum Dis 2006;65:1219–22

44. Akatsu C, Shinagawa K, Numoto N, Liu Z, Ucar AK, Aslam M,et al. CD72 negatively regulates B lymphocyte responses to thelupus-related endogenous toll-like receptor 7 ligand Sm/RNP.J Exp Med 2016;213:2691–706.

45. Bird P, Hall S, Nash P, Connell CA, Kwok K, Witcombe D, et al.Tofacitinib: treatment outcomes in seropositive versus seronega-tive patients in a Phase 3 RA Population [abstract]. ArthritisRheumatol 2016;68 Suppl 10. URL: http://acrabstracts.org/abstract/tofacitinib-treatment-outcomes-in-seropositive-versus-seronegative-patients-in-a-phase-3-ra-population/.

46. Roosnek E, Lanzavecchia A. Efficient and selective presentationof antigen-antibody complexes by rheumatoid factor B cells.J Exp Med 1991;173:487–9.

1744 LU ET AL