Inflammation and autoantibody markers identify rheumatoid arthritis patients with enhanced clinical benefit following rituximab treatment

Authors

  • Preeti Lal,

    1. Genentech, South San Francisco, California
    Search for more papers by this author
    • Drs. Lal, Su, Holweg, Kelman, Read, Spaniolo, Monroe, Behrens, and Townsend own stock or stock options in Genentech/Roche.

    • Drs. Lal, Su, and Townsend have submitted a patent application(s) for the predictive biomarker profile of rituximab benefit in rheumatoid arthritis.

  • Zheng Su,

    1. Genentech, South San Francisco, California
    Search for more papers by this author
    • Drs. Lal, Su, Holweg, Kelman, Read, Spaniolo, Monroe, Behrens, and Townsend own stock or stock options in Genentech/Roche.

    • Drs. Lal, Su, and Townsend have submitted a patent application(s) for the predictive biomarker profile of rituximab benefit in rheumatoid arthritis.

  • Cecile T. J. Holweg,

    1. Genentech, South San Francisco, California
    Search for more papers by this author
    • Drs. Lal, Su, Holweg, Kelman, Read, Spaniolo, Monroe, Behrens, and Townsend own stock or stock options in Genentech/Roche.

  • Gregg J. Silverman,

    1. New York University School of Medicine, New York, New York
    Search for more papers by this author
    • Dr. Silverman has received honoraria from Roche/Genentech (more than $10,000).

  • Sergio Schwartzman,

    1. Weill Cornell Medical College and New York Presbyterian Hospital, New York, New York
    Search for more papers by this author
    • Dr. Schwartzman has received consulting fees, speaking fees, and/or honoraria from Abbott (less than $10,000) and from UCB, Genentech/Roche, Centocor/Johnson & Johnson, and Pfizer (more than $10,000 each).

  • Ariella Kelman,

    1. Genentech, South San Francisco, California
    Search for more papers by this author
    • Drs. Lal, Su, Holweg, Kelman, Read, Spaniolo, Monroe, Behrens, and Townsend own stock or stock options in Genentech/Roche.

  • Simon Read,

    1. Roche Products Ltd., Welwyn Garden City, UK
    Search for more papers by this author
    • Drs. Lal, Su, Holweg, Kelman, Read, Spaniolo, Monroe, Behrens, and Townsend own stock or stock options in Genentech/Roche.

  • Greg Spaniolo,

    1. Genentech, South San Francisco, California
    Search for more papers by this author
    • Drs. Lal, Su, Holweg, Kelman, Read, Spaniolo, Monroe, Behrens, and Townsend own stock or stock options in Genentech/Roche.

  • John G. Monroe,

    1. Genentech, South San Francisco, California
    Search for more papers by this author
    • Drs. Lal, Su, Holweg, Kelman, Read, Spaniolo, Monroe, Behrens, and Townsend own stock or stock options in Genentech/Roche.

  • Timothy W. Behrens,

    1. Genentech, South San Francisco, California
    Search for more papers by this author
    • Drs. Lal, Su, Holweg, Kelman, Read, Spaniolo, Monroe, Behrens, and Townsend own stock or stock options in Genentech/Roche.

  • Michael J. Townsend

    Corresponding author
    1. Genentech, South San Francisco, California
    • Immunology Tissue Growth Repair (ITGR) Biomarker Discovery Group, Genentech, Inc., 1 DNA Way, South San Francisco, CA 94080
    Search for more papers by this author
    • Drs. Lal, Su, Holweg, Kelman, Read, Spaniolo, Monroe, Behrens, and Townsend own stock or stock options in Genentech/Roche.

    • Drs. Lal, Su, and Townsend have submitted a patent application(s) for the predictive biomarker profile of rituximab benefit in rheumatoid arthritis.


  • ClinicalTrials.gov identifiers: NCT00468546 (REFLEX [Randomized Evaluation of Long-Term Efficacy of Rituximab in Rheumatoid Arthritis]) and NCT00299130 (SERENE [Study Evaluating Rituximab's Efficacy in Methotrexate Inadequate Responders]).

Abstract

Objective

Rituximab significantly improves the signs and symptoms of rheumatoid arthritis (RA) and slows the progression of joint damage. The aim of this study was to identify clinical characteristics and biomarkers that identify patients with RA in whom the clinical benefit of rituximab may be enhanced.

Methods

The study group comprised 1,008 RA patients from 2 independent randomized placebo-controlled phase III clinical trials (REFLEX [Randomized Evaluation of Long-Term Efficacy of Rituximab in Rheumatoid Arthritis] and SERENE [Study Evaluating Rituximab's Efficacy in Methotrexate Inadequate Responders]). A novel threshold selection method was used to identify baseline candidate biomarkers present in at least 20% of patients that enriched for placebo-corrected American College of Rheumatology 50% improvement (ACR50 response; a high clinical efficacy bar) at week 24 after the first course of rituximab.

Results

The presence of IgM rheumatoid factor (IgM-RF), IgG-RF, IgA-RF, and IgG anti–cyclic citrullinated peptide (anti-CCP) antibodies together with an elevated C-reactive protein (CRP) level were associated with enhanced placebo-corrected ACR50 response rates in the REFLEX patients with RA who had an inadequate response to anti–tumor necrosis factor therapies. These findings were independently replicated using samples from patients in SERENE who had an inadequate response to disease-modifying antirheumatic drug treatment. The combination of an elevated baseline CRP level together with an elevated level of any RF isotype and/or IgG anti-CCP antibodies was further associated with an enhanced benefit to rituximab.

Conclusion

The presence of any RF isotype and/or IgG anti-CCP autoantibodies together with an elevated CRP level identifies a subgroup of patients with RA in whom the benefit of rituximab treatment may be enhanced. Although the clinical benefit of rituximab was greater in the biomarker-positive population compared with the biomarker-negative population, the clinical benefit of rituximab compared with placebo was also clinically meaningful in the biomarker-negative population.

Rheumatoid arthritis (RA) is a chronic, disabling autoimmune disease that involves inflammation of the synovial lining characterized by infiltration of macrophages, B cells, and T cells and the production of proinflammatory cytokines (1–3). Consequently, bone and cartilage destruction may lead to pain, immobility, and eventual joint deformity. Treatment options include corticosteroids, disease-modifying antirheumatic drugs such as methotrexate, biologic agents that target proinflammatory cytokines such as tumor necrosis factor α (TNFα), interleukin-1β (IL-1β), and IL-6, blockade of costimulation via CTLA-4/Fc, and antibody-mediated depletion of CD20+ B cells (4–9).

Although the etiology of RA remains unknown, the potential importance of B cells in the immunopathogenesis of RA has been demonstrated in randomized, placebo-controlled studies by successfully treating patients with B cell–depleting therapy (10–12). Rituximab, a chimeric anti-CD20 monoclonal antibody, specifically depletes CD20+ B cells, including subpopulations such as pre-B, naive, and memory B cells, via mechanisms that include antibody-dependent cellular cytotoxicity and complement-mediated lysis (13). Because CD20 is not expressed on plasmablasts and plasma cells, these cells are not directly depleted by rituximab.

Placebo-controlled randomized trials of rituximab have demonstrated significant clinical efficacy in patients with RA (11, 12, 14). In the REFLEX (Randomized Evaluation of Long-Term Efficacy of Rituximab in Rheumatoid Arthritis) and SERENE (Study Evaluating Rituximab's Efficacy in Methotrexate Inadequate Responders) phase III trials, rituximab treatment resulted in significant clinical improvement according to the American College of Rheumatology 50% improvement criteria (ACR50) (15) compared with placebo (12, 14). Rituximab treatment results in B cell depletion in blood as well as in the bone marrow and synovium (10, 16–19). However, despite rapid and persistent depletion of circulating peripheral B cells for an average of 4–6 months (20, 21), not all patients show clinically meaningful benefit. It is therefore desirable to better understand the population of patients with RA who experience maximal clinical benefit from rituximab treatment via the identification of biomarker profiles that prospectively define patient subgroups.

To date, no robust placebo-controlled analyses aimed at identifying biomarkers in specific populations that predict enhanced benefit of rituximab have been reported. We conducted a post hoc analysis to identify clinical characteristics and biomarkers that are predictive of enhanced benefit from rituximab, using data from 2 placebo-controlled phase III trials of rituximab, REFLEX and SERENE (12, 14), which were used as training and testing sets for discovery and validation.

PATIENTS AND METHODS

Patients.

Patients with active RA were enrolled in 2 randomized, placebo-controlled, double-blind, international phase III trials (REFLEX and SERENE) to evaluate the safety and efficacy of rituximab plus methotrexate compared with placebo plus methotrexate (Figure 1A). The demographic and clinical characteristics of these patients have been previously described (12, 14).

Figure 1.

Study designs of the REFLEX (Randomized Evaluation of Long-Term Efficacy of Rituximab in Rheumatoid Arthritis) and SERENE (Study Evaluating Rituximab's Efficacy in Methotrexate Inadequate Responders) clinical trials, and identification of optimal biomarker thresholds in REFLEX. A, The study arms, treatment time points, and primary efficacy end points for REFLEX (tumor necrosis factor inadequate response [TNF-IR]) and SERENE (methotrexate inadequate response [MTX-IR]) are illustrated. B, For each continuous baseline biomarker and American College of Rheumatology 50% improvement (ACR50) outcome measure at week 24 from REFLEX, a plot was generated of the subgroup efficacy difference versus the corresponding biomarker (C-reactive protein [CRP]) threshold value for percentiles between 20% and 80% of the range of values for the biomarker. C, Placebo-corrected efficacy differences for each biomarker-defined subgroup were calculated by subtracting the placebo response rate from the rituximab response rate, as illustrated.

In both trials, patients were randomized into rituximab treatment and placebo groups and were stratified according to rheumatoid factor (RF) status. For REFLEX, the intent-to-treat population comprised 499 of 517 randomized patients (201 assigned to the placebo group and 298 assigned to receive rituximab 1,000 mg) who previously experienced an inadequate response to TNF inhibitors. In SERENE (patients who previously had an inadequate response to methotrexate), all 509 patients (172 assigned to placebo, 167 assigned to rituximab 500 mg, and 170 assigned to rituximab 1,000 mg) who received a first course were evaluable.

In both studies, patients were ages 18–80 years, had a diagnosis of active RA according to the ACR 1987 revised criteria (22) at least 6 months prior to enrollment, a swollen joint count in 66 joints of ≥8, and a tender joint count in 68 joints of ≥8 at the time of screening and at baseline. In addition, patients in REFLEX had either an erythrocyte sedimentation rate (ESR) of ≥28 mm/hour or a C-reactive protein (CRP) level of ≥1.5 mg/dl, and patients in SERENE had either an ESR of ≥28 mm/hour or a CRP level of ≥0.6 mg/dl at the time of screening. Prior to screening, patients received methotrexate at a dosage of 10–25 mg/week for at least 12 weeks, with a stable dose for the last 4 weeks. Additional requirements for REFLEX included a washout period from TNF inhibitors (≥4 weeks for etanercept and adalimumab or ≥8 weeks for infliximab) and radiographic evidence of erosion in at least 1 joint. Participants in both studies provided informed written consent.

Treatment with rituximab or placebo was administered by intravenous (IV) infusion on days 1 and 15 with concomitant methotrexate (10–25 mg/week as prescribed by the treating physician). All patients received a 100-mg IV infusion of methylprednisolone prior to each infusion and also received folate at a dosage of ≥5 mg/week.

Serum samples obtained for regular clinical measurements were available for 499 patients from REFLEX and for 501 patients from SERENE. Optional baseline serum samples for biomarker research purposes were collected before the first infusion (pretreatment) in REFLEX (n = 339) and SERENE (n = 498). Samples were aliquoted to avoid repeated freeze–thaw cycles and stored at −80°C until used.

Clinical response.

For the main biomarker analyses, a response to rituximab was defined as 50% improvement (an ACR50 response) at week 24. The ACR50 response was chosen because it is a relatively high-bar clinical end point, and because the number of ACR50 responders was sufficient to allow for robust analysis of baseline biomarkers. Additional response measures, including the Disease Activity Score in 28 joints using the erythrocyte sedimentation rate (DAS28- ESR) (23) and the European League Against Rheumatism (EULAR) response (24) at week 24, were also examined. Patients who withdrew prematurely from the study or who received rescue therapy before week 24 were considered to be ACR50 and EULAR nonresponders at week 24, and their DAS28-ESR values were imputed using the last observation carried forward.

Biomarker measurements.

A panel of potential biomarkers, consisting of antibodies (total RF, IgM-RF, IgG-RF, IgA-RF, and IgG anti–cyclic citrullinated peptide [anti-CCP]), inflammation markers (CRP and serum amyloid A), a T cell activation marker (soluble CD25 [sCD25]), cytokines, and chemokines (TNFα, interferon-γ, IL-1β, IL-2, IL-6, IL-8, IL-10, IL-12p70, and granulocyte–macrophage colony-stimulating factor), were measured in baseline serum samples. CRP, ESR, total RF, and RF isotypes were measured as part of the clinical data collection process during the trials, using standard assays. Specifically, RF isotypes were measured using TheraTest EL-RF/3, which has received 510(k) clearance from the US Food and Drug Administration as an in vitro diagnostic assay. All other measurements were performed using commercially available kits according to the manufacturers' instructions, with the modifications described below. IgG anti-CCP antibodies were measured using the QUANTA Lite CCP3 IgG enzyme-linked immunosorbent assay (ELISA; INOVA Diagnostics). Soluble CD25 was measured using a BD OptEIA kit (BD Biosciences). Immunoglobulin-inhibiting reagent (2 mg/ml; Bioreclamation) was added to the samples, followed by incubation for 30 minutes at room temperature to block heterophilic antibodies before serum was added to the ELISA. All other analytes were measured using multiplex assays (Proinflammatory 9-Plex Ultra-Sensitive, Vascular Injury 4-Plex Panel II, and Growth Factor 4-Plex Panel I) on the Meso Scale Discovery (MSD) Platform. Blocking reagents for heterophilic antibodies were included in the blocking buffer of all MSD assays. The efficacy of blocking heterophilic antibodies has previously been determined and was described by DeForge et al (25).

Statistical analysis.

A novel threshold selection statistical method was used to identify candidate biomarker subgroups that represented at least 20% of patients from the REFLEX trial and enriched for placebo-corrected ACR50 responses (ACR50 response for the rituximab-plus-methotrexate group minus ACR50 response for the placebo-plus-methotrexate group) at week 24 after the first course of rituximab. To identify subgroups for whom the clinical benefit of rituximab might be increased, the study population from REFLEX was stratified according to baseline clinical characteristics and serologic biomarker measurements (in patients for whom serum samples were available). The baseline characteristics of the patient subgroups that had available biomarker serum samples were comparable with those of the overall patient group in the clinical trial. For surveys of each continuous biomarker (where a range of discrete values was possible) and the outcome measure ACR50 response at week 24, a plot was generated presenting subgroup efficacy differentials versus a range of potential threshold values (20th–80th biomarker percentiles in 5-percentile increments) to control bias.

The threshold giving the largest efficacy differential was then identified. For this threshold, a permutation test was used to address statistical significance. For each permutation, biomarker values were permuted, and both treatment assignment and the outcome measure were fixed. The largest efficacy differential was computed for the permutated data set, which was compared with the largest efficacy differential observed for the original data. Permutation P values were based on 2,000 permutations. A 95% confidence interval for the largest efficacy differential was calculated. The 4 biomarkers with the highest efficacy differentials (CRP, IgG anti-CCP, IgA-RF, and sCD25) that identified subgroups of rituximab-treated patients in the REFLEX trial with the potential for an enhanced clinical benefit were then prioritized and further investigated in the SERENE data set. In addition, 5 of the 2-biomarker (bivariate) combinations used in these trials were also studied. The sixth combination, IgG anti-CCP and IgA-RF, was not considered due to a high correlation between the 2 markers. Because the CRP level is one of the components of the ACR50 criteria, the DAS-ESR was prioritized for testing in the SERENE data set.

According to a recent report, no significant differences in either clinical or safety outcomes were apparent between the rituximab doses used in SERENE (500 mg and 1,000 mg) (14). Because ACR50 response rates were similar between the SERENE patients receiving rituximab 500 mg (26.3%; n = 167) and those receiving rituximab 1,000 mg (25.9%; n = 170) at 24 weeks, and because pharmacodynamic properties were comparable between the 2 doses (11), we combined both doses as a single treatment group for this analysis. Each biomarker was analyzed as described for REFLEX. Each of 5 bivariate subgroup candidates was constructed using an “and” rule and applying the same direction (i.e., high or low) that defined the best subgroup for the individual biomarkers. Bivariate subgroup candidates were formed by comparing patients with both elevated CRP levels and elevated IgA-RF titers and patients who were not in this subgroup; subgroups were determined by the 30th, 40th, 50th, 60th, and 70th percentiles of either the CRP or IgA-RF level, with the constraint that at least 20% of the patients were in each subgroup, in order to control bias. Exploratory analyses of other nonprioritized baseline biomarkers (e.g., IgM-RF, IgG-RF, and IgG anti-CCP) and combinations of biomarkers were also performed.

RESULTS

The REFLEX and SERENE studies were pivotal randomized, double-blind, placebo-controlled trials in patients with RA who experienced an inadequate response to TNF inhibitors and methotrexate, respectively (Figure 1A). In REFLEX, the overall ACR50 response rates at week 24 were 5% and 27% for the placebo and rituximab groups, respectively (P < 0.0001). In SERENE, the 24-week ACR50 response rates were 9% and 26% for the placebo and combined rituximab groups, respectively (P < 0.0001).

Discovery analysis of 28 biomarkers and baseline characteristics of clinical relevance was performed using the ACR50 response at week 24 as the clinical response criterion (Table 1). In the REFLEX data set, the analysis resulted in prioritization of candidate biomarkers stratified according to the highest placebo-corrected difference in the ACR50 response between rituximab-treated and placebo-treated patients at week 24 (Table 2). Placebo-corrected differences (Δhigh − Δlow) were calculated as shown in Figures 1B and C. In this example, REFLEX patients were stratified according to CRP levels ranging from 1.0 mg/dl to 5.8 mg/dl, and the optimal CRP level threshold was determined to be 3.9 mg/dl (corresponding to the 70th percentile for patients in REFLEX) (Figure 1B). Calculation of the ACR50 response rate difference of 26% between the 2 subgroups defined by a CRP level >3.9 mg/dl is shown in Figure 1C.

Table 1. Analyses of 28 baseline clinical characteristics and serologic markers in the REFLEX study*
MarkerSubgroup with better efficacyPlacebo-corrected ACR50 efficacy difference, %Size of biomarker-positive subgroup, %95% CI
  • *

    Prioritized biomarkers that were further evaluated using the SERENE (Study Evaluating Rituximab's Efficacy in Methotrexate Inadequate Responders) data set are C-reactive protein (CRP), IgA rheumatoid factor (IgA-RF), IgG anti–cyclic citrullinated peptide (anti-CCP), and soluble CD25. The 95% confidence intervals (95% CIs) are unadjusted for multiple comparisons. REFLEX = Randomized Evaluation of Long-term Efficacy of Rituximab in Rheumatoid Arthritis (RA); ACR50 = American College of Rheumatology 50% improvement response; DAS28-ESR = Disease Activity Score in 28 joints using the erythrocyte sedimentation rate; IL-1β = interleukin-1β; TNFα = tumor necrosis factor α; IFNγ = interferon-γ; GM-CSF = granulocyte–macrophage colony-stimulating factor.

  • Only anti-CCP–positive patients (>20 units) were considered.

Clinical characteristics    
 Age, years≤6113770, 26
 SexFemale381−12, 18
 Body weight, kg≤6115200, 29
 Body mass index≤22.714201, 27
 RA duration, years≤5.8825−5, 22
 Total Sharp score>15.7880−8, 23
 No. of swollen joints>33518−8, 19
 No. of tender joints>21875−7, 21
 ESR, mm/hour>3621599, 33
 CRP, mg/dl>3.9263014, 39
 DAS28-ESR>7.41030−3, 21
Autoantibodies    
 IgA-RF, units/ml>25288014, 42
 IgM-RF, units/ml>1,26619205, 32
 IgG-RF, units/ml>1917455, 29
 Total Ig RF, units/ml>2719756, 33
 IgG anti-CCP, units≤87826515, 48
 IgM anti-CCP, units≤297.51680−7, 37
Cytokines/chemokines/cells    
 IL-1β, pg/ml>1.61719−3, 39
 IL-2, pg/ml≤1.7671−14, 26
 IL-6, pg/ml>9.9305012, 47
 IL-8, pg/ml≤36.91240−6, 31
 IL-10, pg/ml≤2.91120−12, 36
 IL-12p70, pg/ml≤7.4865−10, 27
 Soluble CD25, pg/ml≤3,923293510, 47
 TNFα, pg/ml≤10.2820−11, 29
 IFNγ, pg/ml>6.71740−3, 36
 GM-CSF, pg/ml>21138−9, 31
 CD19, cells/μl>125860−4, 20
Table 2. Summary of efficacy differentials in the SERENE verification analyses based on optimal thresholds derived from the REFLEX study discovery analysis*
BiomarkerSubgroup with better efficacyPlacebo-corrected ACR50 efficacy difference, %Size of biomarker-positive subgroup, %
REFLEXSERENEREFLEXSERENE
  • *

    SERENE = Study Evaluating Rituximab's Efficacy in Methotrexate Inadequate Responders; REFLEX = Randomized Evaluation of Long-term Efficacy of Rituximab in Rheumatoid Arthritis; ACR50 = American College of Rheumatology 50% improvement response; CRP = C-reactive protein; IgA-RF = IgA rheumatoid factor.

  • Only anti–cyclic citrullinated peptide (anti-CCP)–positive patients (>20 units) were considered.

IgG anti-CCP, units≤87826−75139
CRP, mg/dl>3.926163016
Soluble CD25, pg/ml≤3,9232933571
IgA-RF, units/ml>2528138080

The candidate biomarkers demonstrating the highest differences (e.g., Δhigh − Δlow >25%) between subgroups included higher concentrations (>3.9 mg/dl) of CRP, higher titers (>25 units/ml) of IgA-RF, lower concentrations (≤3,923 pg/ml) of sCD25, and intermediate titers (≤878 units) of IgG anti-CCP antibodies (Figures 2A–D). For IgG anti-CCP, only patients with antibody positivity (>20 units) were considered. For all biomarker cutoff thresholds, at least 20% of patients were represented. Candidate biomarkers and their thresholds, as determined from REFLEX, were subsequently tested as single markers using data from SERENE. These analyses indicated that stratification by CRP level or IgA-RF cutoff values resulted in robust efficacy differentials (Δhigh − Δlow >10%) in both studies (Table 2). In contrast, titers of IgG anti-CCP antibodies ≤878 units and sCD25 concentrations ≤3,923 pg/ml did not reproducibly show enhanced benefit in SERENE (Δhigh − Δlow values of −7% and 3%, respectively).

Figure 2.

Exploratory subgroup analyses of prioritized biomarkers in REFLEX and SERENE. ACR50 response rates at week 24 of REFLEX for biomarker-defined subgroups were based on the following threshold values: CRP >3.9 mg/dl (A), IgA rheumatoid factor (IgA-RF) titers >25 units/ml (B), soluble CD25 (sCD25) concentrations ≤3,923 pg/ml (C), and IgG anti–cyclic citrullinated peptide (anti-CCP) titers 20–878 units (D). The subgroup with IgG anti-CCP titers of 20–878 units includes 51% of patients in the study population. ∗ = Only IgG anti-CCP–positive patients (>20 units) were considered. See Figure 1 for other definitions.

Specific biomarkers and their combinations identified from REFLEX were next tested prospectively using the SERENE data set. We again tested percentiles of the range of biomarker values between 20% and 80% to ensure a minimum biomarker-defined population size of 20% and hence capture a sizable proportion of the SERENE population that had an inadequate response to DMARDs. Of the 10 prioritized biomarkers, combinations, and thresholds studied, only one subgroup showed statistical evidence of enhanced clinical benefit using the SERENE data set (additional information is available from the corresponding author). The biomarker combination of an elevated CRP level (>2.9 mg/dl) and an elevated IgA-RF level (>25 units/ml) provided a placebo-corrected ACR50 efficacy differential (Δhigh − Δlow) of 19% (unadjusted permutation P = 0.02). As shown in Figure 3A, 22% of patients were included in this combination subgroup (biomarker-positive) with enhanced clinical benefit. Notably, this relationship was also preserved when only seropositive patients were considered. After a threshold sensitivity assessment of the CRP value percentile in this population, a subset of seropositive patients with a CRP level of >3.7 mg/dl (20% of patients in the trial) had a placebo-corrected ACR50 response rate of 36% versus 16% for the remainder of seropositive patients with a CRP level of ≤3.7 mg/dl. Furthermore, a similar relationship was observed when the 500-mg and 1,000-mg dose arms were considered separately (data not shown).

Figure 3.

Analysis of CRP and autoantibody combinations in SERENE (A–D) and REFLEX (E–H). Week 24 ACR50 response rates were calculated for the combination biomarker of elevated CRP level (>2.9 mg/dl) and seropositivity for IgA rheumatoid factor (IgA-RF) >25 units/ml (A and E), IgG-RF titer >20 units/ml (B and F), IgM-RF titer >20 units/ml (C and G), or IgG anti–cyclic citrullinated peptide (anti-CCP) titer >20 units (D and H) versus all other patients in the trials. See Figure 1 for other definitions.

To determine whether enhanced benefit with rituximab was observed with other clinical outcome measures, the ACR20 and EULAR responses were examined in biomarker-positive and biomarker-negative combination subgroups from the SERENE data set (Table 3). The SERENE patient population stratified by the combination of elevated CRP levels and elevated IgA-RF titers (representing 22% of the population) demonstrated increased ACR20 efficacy differentials, greater improvement in EULAR good/moderate responses, and a greater decrease in the mean DAS28-ESR change from baseline compared with the biomarker-negative combination population.

Table 3. CRP and IgA-RF biomarker combinations and various clinical outcome measures at week 24 in SERENE*
Efficacy measureBiomarker-positive subgroupBiomarker-negative subgroup
  • *

    Except where indicated otherwise, values are the percent (95% confidence interval [not adjusted for multiple comparisons]). Biomarker positivity was defined as a C-reactive protein (CRP) level of >2.9 mg/dl and an IgA rheumatoid factor (IgA-RF) level of >25 units/ml; 22% of the patients were biomarker positive. Biomarker negativity was defined as a CRP level of ≤2.9 mg/dl or an IgA-RF level of ≤25 units/ml; 78% of the patients were biomarker negative. SERENE = Study Evaluating Rituximab's Efficacy in Methotrexate Inadequate Responders; Δ ACR50 = the difference between rituximab-treated and placebo-treated patients (Δ) for the American College of Rheumatology criteria for 50% improvement. EULAR = European League Against Rheumatism; DAS28-ESR = Disease Activity Score in 28 joints using the erythrocyte sedimentation rate.

Δ ACR5034 (20, 48)13 (6, 20)
Δ ACR2044 (28, 60)26 (16, 36)
Δ EULAR good/moderate response55 (39, 71)25 (15, 35)
Δ mean DAS28-ESR from baseline−1.54 (−2.00, −1.08)−0.82 (−1.11, −0.53)

To assess whether the identified biomarker combination extended to other autoantibody isotypes of RF and/or anti-CCP, patients were stratified by the CRP threshold of 2.9 mg/dl obtained from the SERENE analysis together with the threshold cutoff values for seropositivity for IgM-RF, IgG-RF, IgA-RF, or IgG anti-CCP antibodies. The combination of a CRP level of >2.9 mg/dl with autoantibody positivity (>25 units/ml for IgA-RF and >20 units/ml for IgG-RF and IgM-RF and >20 units for IgG anti-CCP) identified a subgroup in the SERENE data set with greater placebo-corrected ACR50 response rates at week 24 (Figures 3A–D). Furthermore, the same analysis applied to the REFLEX data set showed similar trends (Figures 3E–H). Confidence intervals for the placebo-corrected efficacy differences using these CRP/autoantibody combinations were overlapping in all cases, indicating statistically similar results among the different autoantibody classes.

Finally, we tested whether the biomarker profile with the higher CRP threshold cutoff of 3.9 mg/dl, as established in REFLEX, together with an IgA-RF threshold cutoff of >25 units/ml, also had a consistent relationship with clinical benefit in SERENE. This set of threshold cutoff values defined a SERENE population of just 14% but demonstrated a placebo-corrected ACR50 response rate of 36% versus a 15% ACR50 response rate for biomarker profile–negative patients (additional information is available from the corresponding author). Thus, the enhanced clinical benefit we observed using this combination biomarker profile continued to be maintained in the SERENE trial when using the REFLEX-defined CRP threshold value. Therefore, the combination of elevated CRP levels and RF/anti-CCP antibody positivity prospectively defined an RA subgroup with greater placebo-corrected ACR50 efficacy at week 24 compared with patients not fulfilling these requirements.

DISCUSSION

Selective B cell depletion via rituximab-mediated targeting of CD20+ cells is an effective therapy for patients with RA, but not all patients have an adequate response to treatment (11, 12, 14). To identify populations of patients who are more likely to achieve a robust clinical benefit from rituximab treatment, a systematic analysis of serologic biomarkers in a discovery cohort of patients from the REFLEX trial was performed, and candidate biomarkers were then tested prospectively in a verification cohort from the SERENE trial. This methodology facilitated assessment of the placebo-controlled effect of rituximab in biomarker- and biomarker combination–stratified RA patient populations using high-hurdle clinical efficacy end points. The elevation of markers reflecting inflammation and seropositivity for prototypical RA autoantibodies was associated with robust efficacy of rituximab, and a combination profile of 2 biomarkers, CRP and IgA-RF, provided optimized stratification of patients into a subgroup in which the clinical benefit of rituximab was enhanced.

These findings are consistent with prior observations within smaller patient cohorts that rituximab-treated RA patients with baseline seropositivity for RF and/or anti-CCP antibodies had significantly increased odds of achieving ACR50 responses at week 24 compared with seronegative patients (P = 0.0096) (26–28). Indeed, the assessment of baseline RF and anti-CCP antibody status in REFLEX and SERENE also showed evidence for enrichment of clinical benefit when considering antibodies alone as biomarkers predictive of clinical benefit (Table 1). However, the inclusion of a biomarker of systemic inflammation, CRP, together with IgA-RF or other RF isotypes and IgG anti-CCP antibodies, further enriched a smaller subpopulation of patients with the potential for increased clinical benefit in both clinical studies (Figure 3). The biomarker discovery and verification exercise we undertook in this study aimed to identify the best predictive biomarker profile, with less consideration given to the size of the resulting subpopulation. This trade-off between maximal treatment benefit versus biomarker-positive population size must be kept in mind during the design of future clinical trials using predictive biomarkers as well as in making clinical decisions as to whether to treat patients based on a biomarker profile.

By what mechanisms might rituximab therapy fail to provide benefit in patients with RA? One such mechanism might be incomplete depletion of pathogenic B cells. The assessment of peripheral blood samples before and after rituximab treatment showed a rapid reduction (>95%) of detectable circulating B cells (10, 29). Most patients subsequently demonstrated sustained circulating B cell depletion for at least 4–6 months, together with a decrease in circulating CD19+CD20− plasmablasts, presumably due to depletion of activated B cell precursors (16, 21). However, there is currently no marker specific for pathogenic B cells. A lack of sustained clinical efficacy has been explained in some patients by rapid reconstitution of the B cell compartment, as demonstrated by the early reemergence of memory B cells in peripheral blood (20). In addition, a larger memory B cell population at the time of repopulation has been associated with a group of patients who did not experience a response to B cell–depleting therapy (30).

A further mechanism of treatment resistance may be explained by a lack of depletion of tissue-resident pathogenic B cells by rituximab. Indeed, persistence of autoantibodies and plasma cells has been observed in RA patients after anti-CD20 treatment and is correlated with poor clinical efficacy (31–33). Serial synovial biopsy specimens have also shown inconsistent depletion of B cells and plasma cells despite >95% depletion of blood B cell lineages (16–18), and rituximab has been proposed to affect the recirculation of pathogenic autoreactive B cells to the synovial tissues (34). It is noteworthy that high baseline levels of autoantibodies directed against anti–citrullinated protein antibodies (ACPAs), which are correlated with high levels of CD79α+ but not CD20+ cells in the synovial lining, have been shown to be associated with a moderate or poor response to rituximab treatment (35). This finding, which presumably reflects a rituximab-resistant plasmablast population in synovial tissue, is consistent with our observation that high baseline titers of anti-CCP antibodies in the REFLEX trial are associated with lower clinical benefit (Table 1). However, this observation was not replicated when the SERENE trial was evaluated (additional information is available from the corresponding author) and was also not reflected in the titers of any of the RF isotypes. It is possible that patients in REFLEX, including those with active RA refractory to anti-TNFα blockade, have an increased synovial tissue load of ACPA-producing CD20− plasmablasts as compared with the SERENE patients who on average had a shorter duration of disease.

A second mechanism that might underpin clinical responsiveness to rituximab is heterogeneity of the underlying disease pathophysiology in RA. Subpopulations of RA patients with distinct serum cytokine and chemokine profiles, autoantibody reactivities, and patterns of gene expression in synovial tissues have been described (36–38). Importantly, patients presenting with elevated synovial tissue inflammation, including infiltrating B cells, have recently been shown to have elevated levels of general inflammation markers as well as higher titers of autoantibodies (39). The markers we observed to be most associated with enhanced benefit from rituximab—CRP and RF/anti-CCP autoantibodies—have been linked to specific inflammatory and immune pathways in RA synovium. CRP is a widely used marker of inflammation that represents systemic activity of inflammatory cytokines (particularly IL-6) in liver hepatocytes (40, 41). Indeed, baseline CRP levels correlated significantly with baseline IL-6 levels in both REFLEX and SERENE (ρ = 0.60 and P < 0.001 and ρ = 0.59 and P < 0.001, respectively) (data not shown). It is also noteworthy that the optimal threshold cutoff value in REFLEX was higher than that in SERENE. This may reflect the population of patients with very clinically active RA refractive to anti-TNFα therapy in REFLEX. Furthermore, both IgA-RF and IgG anti-CCP are prognostic markers for severe and aggressive disease as well as biomarkers of autoreactive B cell activity (42–46). The biomarker combination profile described here is consistent with a potential mechanism of action of rituximab in the depletion of autoreactive B cell lineages that promote synovitis.

The current study was strengthened by the availability of 2 large pivotal clinical trials, including >1,000 RA patients, for analysis. The ability to conduct a discovery analysis in one clinical trial and then prospectively test promising biomarkers in a second independent trial supports the robustness of the analysis and reinforces the reproducibility of the biomarker analysis. Furthermore, the availability of a placebo arm in both trials uniquely enabled an assessment of whether the biomarkers were solely predictive of clinical benefit from rituximab therapy or were also prognostic of the overall disease course in a defined patient population, independent of specific therapeutic intervention.

A limitation of this analysis centers on the distinct enrollment criteria and the patient populations examined. REFLEX comprised a population of patients who previously had an inadequate clinical response to TNF inhibitors, while SERENE enrolled biologic agent–naive patients who had an inadequate response to methotrexate. The mean disease duration of REFLEX patients was longer than that of SERENE patients (12, 14). Furthermore, the entry criteria regarding baseline CRP levels for the 2 studies were different, resulting in a higher proportion of REFLEX patients with a CRP level of >1 mg/dl (79%) compared with SERENE patients (48%). Indeed, this is reflected in the difference in optimal CRP cutoff values identified using the threshold sensitivity method, where the REFLEX cutoff value is higher than the SERENE cutoff value (3.9 mg/dl versus 2.9 mg/dl). Therefore, the 2 patient populations may have some intrinsic biologic differences. On the other hand, there were many important similarities between the 2 cohorts studied, including average age, sex, race, number of tender and swollen joints, baseline DAS28, and response rates to rituximab (additional information is available from the corresponding author). We also observed that use of the REFLEX-defined CRP cutoff value as part of the assessment of the combination biomarker profile in SERENE continued to be consistent for enriching clinical benefit but defined a relatively small subpopulation (14%) of patients within SERENE.

In conclusion, we used a novel approach to systematically assess baseline clinical characteristics of and serologic biomarkers in RA patients from 2 large clinical trials, in order to identify subgroups for whom the potential of experiencing a benefit of rituximab is enhanced. A biomarker profile consisting of the combination of elevated baseline CRP levels and elevated baseline autoantibody titers identified such a subgroup of patients. Prospective validation of this biomarker profile is warranted to determine the optimal threshold for this combination biomarker in RA patients representative of those seen in current clinical practice.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Townsend had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Lal, Su, Kelman, Monroe, Behrens, Townsend.

Acquisition of data. Su, Holweg, Townsend.

Analysis and interpretation of data. Lal, Su, Holweg, Silverman, Schwartzman, Kelman, Read, Spaniolo, Monroe, Behrens, Townsend.

ROLE OF THE STUDY SPONSORS

Genentech, Roche, and Biogen Idec designed and ran the clinical trials and collected the data. Genentech supported generation and analysis of the biomarker data and drafting of the manuscript. Genentech and Roche supported submission of the manuscript for publication and approved the content of the submitted manuscript. Publication of this manuscript was not contingent on approval by the study sponsors.

Acknowledgements

We thank the patients and investigators who participated in the REFLEX and SERENE trials. We also thank Keith del Villar (Genentech) for editorial assistance.

Ancillary