Biomarkers of Inflammation and Development of Rheumatoid Arthritis in Women From Two Prospective Cohort Studies

Abstract

Objective

To examine the association of biomarkers of inflammation with preclinical rheumatoid arthritis.

Methods

A nested case-control study was performed using samples from 2 large, prospectively studied cohorts of women (the Women’s Health Study [WHS] and the Nurses’ Health Study [NHS]). Blood samples predating the first RA symptom were selected for incident RA cases and for 3 controls per case, matched on age, menopausal status, postmenopausal hormone use, and day, time and fasting status at blood draw. Plasma was tested for IL-6, soluble TNF receptor 2 (sTNFR2), as a proxy for TNF-α level, and high sensitivity CRP. Relationships between biomarkers and RA were assessed using conditional logistic regression models, adjusting for age, body mass index, smoking, ethnicity, and reproductive factors.

Results

Mean time to RA after blood draw was 5.2 years (range = 0.3–12 years) in 93 incident cases in NHS and 77 incident cases in WHS. Median IL-6 (p=0.03), and sTNFR2 (p=0.003) levels were significantly higher in preclinical RA cases compared to matched controls in NHS but not in WHS. Pooled analysis in NHS and WHS demonstrated significant association for sTNFR2 with RA [RR 2.0 (1.1–3.6), p, trend 0.004], a modest association for IL-6 [RR 1.4 (0.8–2.5), p, trend 0.06].

Conclusion

Soluble TNFR2, an inflammatory biomarker typically associated with active RA was elevated up to 12 years prior to RA symptoms. sTNFR2 was positively associated with incident RA in these nested case control studies. Studies with repeated assessments of biomarkers prior to RA may provide further insight the timing of inflammatory biomarker elevation in preclinical RA.

Rheumatoid arthritis (RA) is the most common autoimmune inflammatory arthritis, affecting approximately 1% of the population. Its etiology is unknown but it is presumed to be an immunologic disease with contributing genetic^1–5 and environmental factors, such as cigarette smoking ^6–15 and reproductive factors in women.^16–18 A growing body of evidence suggests that there are three phases to the development of RA, an asymptomatic period of genetic risk, a preclinical phase in which RA-related autoantibodies can be detected ^{19, 20}, and a clinical phase with acute signs and symptoms of inflammatory arthritis.²¹ Similar phases of development have been proposed in other autoimmune diseases such as type 1 diabetes and systemic lupus erythematosus.^{22, 23}

In acute and chronic inflammation, cytokines are instrumental in regulating the magnitude and duration of the inflammatory response. Tumor necrosis factor α (TNFα), and interleukin-6 (IL-6) are pleiotropic cytokines produced predominantly by macrophages, that initiate the T cell and synovial proliferation, and are responsible for joint destruction in RA²⁴; and levels are elevated in the serum and the joints during active RA.^25–28 Since TNFα degrades rapidly in stored samples²⁹, we studied soluble TNF Receptor 2 (sTNFR2) levels. Soluble TNFR2 expression parallels TNFα levels and is a surrogate marker for inflammation. ^{25, 26}. We aimed to study biomarkers of inflammatory and immune activity, including high sensitivity C-reactive protein (hs-CRP), interleukin-6 (IL-6) and soluble tumor necrosis factor receptor 2 (sTNFR2) during the pre-clinical phase of RA among three large female cohorts of US nurses, followed prospectively for up to 12 years after blood collection, with extensive epidemiologic data on risk factors for RA. We hypothesized that women who developed RA would have evidence of immune activation prior to the first symptoms of RA, compared to women who did not develop signs and symptoms of RA.

Methods

The Nurses’ Health Study cohort was established in 1976 and enrolled 121,700 US female registered nurses, ages 30–55 years. The NHSII cohort was established in 1989, and enrolled 116, 609 female registered nurses, ages 25–42 years. The WHS was a randomized, double-blind, placebo-controlled trial designed to evaluate the benefits and risks of low-dose aspirin and vitamin E in the primary prevention of cardiovascular disease and cancer among 39,876 female health professionals, aged 45 years and older. ^30–32 All women completed an initial questionnaire and have been followed biennially in NHS and annually in WHS by questionnaire to update exposures and disease diagnoses.

From 1989 through 1990, 32,826 NHS participants (ages 43 to 70 years) provided plasma samples in heparinized tubes.³³ Women arranged to have their blood drawn and shipped with an icepack by overnight carrier. Upon arrival, the blood samples were centrifuged and blood was aliquotted into plasma, WBC, and RBC components. Samples have been stored in liquid nitrogen freezers with an electronic alarm system since collection.

From 1996 through 1999, 29,611 NHSII participants (ages 32–51 years) provided blood samples. From 1992 through 1995, 28,133 in the Women’s Health Study provided plasma samples in EDTA tubes. Collection and storage procedures for NHSII and WHS were similar to those described for the NHS above. All aspects of this study were approved by the Partners’ HealthCare Institutional Review Board.

Identification of Rheumatoid Arthritis

Rheumatoid Arthritis (RA) case identification for the NHS and WHS cohorts was 2-stage procedure.¹⁸. The Connective Tissue Disease Screening Questionnaire (CSQ) was first mailed to subjects who reported a physician diagnosis of RA with a 77% response rate in NHS and 72% response rate in WHS.^{34, 35} Medical records of those who screened positively for RA or other connective tissue diseases (CTD) including systemic lupus erythematosus, mixed connective tissue disease, scleroderma, polymyositis, dermatomyositis, or Sjogren’s syndrome³⁴ were requested and received from 96% of subjects. Two board-certified rheumatologists trained in chart abstraction independently conduct a medical record review blinded to the second reviewer’s result, examining the charts for the American College of Rheumatology (ACR) classification criteria for RA³⁶, date of first RA symptom evidence of RA-specific medication treatment, the treating physician’s diagnosis. The specificity of CTD detection using a staged series design is very high, reducing misclassification of healthy subjects.³⁷ The reviewers met to discuss and resolve discrepancies and determine a consensus diagnosis. Subjects with four of the seven ACR criteria for RA and reviewer consensus were considered to have definite RA. For this nested case-control study, we also included a small number of subjects (N= 5 cases in NHS and 3 cases in WHS) with three documented ACR criteria for RA, a diagnosis of RA by their physician, and reviewer consensus on the diagnosis of RA. The presence of serum rheumatoid factor (RF) at diagnosis was obtained from the medical records as documented at the time of diagnosis.

Selection of Cases and Controls

Eligible cases included all incident RA cases with a stored blood sample whose date of blood draw preceded the date of the first RA symptom documented in the medical record by at least three months. The control group included subjects with a stored blood sample, excluding those with self-reported RA not confirmed by rheumatologist review and those with other self-reported CTDs. No control developed RA or other rheumatic disease during the 12 year follow-up period. For both cases and controls, we excluded women who reported any cancer (except non-melanoma skin cancer) at baseline or during follow-up, as cancer and its treatment can affect biomarker levels. Three controls for each confirmed RA case were randomly chosen from subjects with stored blood, matching on cohort, birth year (± 1 year), race/ethnicity, time of day, fasting status, menopausal status and postmenopausal hormone use at date of blood draw. For premenopausal women in NHSII, we also matched on timing of blood sample in the menstrual cycle. For WHS, we also matched on time since randomization.

Information on potential confounding variables

All exposure information was self-reported on the mailed questionnaires administered every two years since 1976 in NHS and 1989 in NHSII, and annually since 1992 in WHS. Cigarette smoking is the strongest environmental risk factor for RA^6–15 and is associated with biomarker levels, thus meeting the definition of a classic confounder. We adjusted for smoking coded as never, past, current < 15 cigarettes/day, and current ≥ 15 cigarettes/day. Reproductive covariates were chosen based on our past findings of associations between reproductive factors and the risk of RA in the NHS.¹⁸ Age at menarche (<12 vs. ≥ 12), regularity of menses between the ages 20 to 35 (regular, irregular), parity and duration of breastfeeding (nulliparous, parous and no breastfeeding, parous and 1–12 months, parous and ≥12 months), menopausal status and postmenopausal hormone use (premenopausal, postmenopausal and never use, postmenopausal and past use, postmenopausal and current use, dubious menopausal status), and body mass index (BMI) as a continuous variable were included as potential confounders of the biomarker and RA risk relationship. For WHS analyses, the same potential confounders were included except breastfeeding since data was not available for this variable. For each control, a reference date corresponding to RA onset date in the matched case was assigned. Covariate data from all questionnaires were selected from the questionnaire preceding the reference date.

Laboratory Assays

The laboratory selected for this study has high assay precision and runs internal positive and negative quality control (QC) samples daily. The laboratory has undergone rigorous blinded pilot testing with aliquots from NHS QC specimens in which aliquots were divided into two blinded samples. Coefficients of variation ranged from 0.07–6.1% for IL-6, from 4.9–11.6% for sTNFR2, and 1.6–3.0% for CRP. For study samples, all assays were blinded to case/control status. Samples were labeled by number only, and matched case-control pairs handled together identically, shipped in the same batch, and assayed in the same run. The order within each case-control pair was random. Aliquots from pooled quality control specimens, indistinguishable from study specimens, were interspersed randomly among case-control samples to monitor QC. IL-6 was measured by an ultra-sensitive quantitative sandwich enzyme immunoassay from R&D Systems. IL-6 is quantified in pg/ml and assay sensitivity is 0.94 pg/mL. TNFα cannot be reliably measured in stored plasma as it degrades rapidly. ²⁹. Thus, for this study, we measured soluble TNFR2 as a proxy for TNFα since blood samples were shipped to our lab and arrived within 24 hours and most of the TNFα had likely degraded. Soluble TNFR2 stability was assessed in 17 fresh blood samples, at baseline, and after a delay in shipping of 24 hours and 36 hours, and the intraclass correlation coefficient (ICC) was >75% for the comparison of 0 to 36 hours.³⁸ Soluble TNFR2 was measured by a quantitative sandwich ELISA assay from R & D Systems. Day-to-day assay variabilities at concentrations of 54.8, 252 and 356 pg/mL are 8.8, 3.7 and 5.8%, respectively in our laboratory.

In addition, because of the potential for interference by rheumatoid factor (RF) in the ELISA assays used for IL-6 and sTNFR2 measurements³⁹, we conducted an additional pilot study. We selected 16 samples from RA patients with a range of RF titers (0–680). RF was depleted from these samples by affinity absorption with human IgG conjugated sepharose (IgG Sepharose 6 fast flow, GE Healthcare). Sera was diluted 1:1 with Tris-buffered saline and incubated with an equal volume of IgG-sepharose for over 4 hours at 4 C. As control, diluted sera were exposed to unconjugated sepharose under identical conditions. Complete depletion of RF was confirmed by nephelometry.⁴⁰. A subset of high-titer RF samples required two rounds of affinity absorption. Following RF depletion, 32 samples, 16 pre-depletion, and 16 post-depletion were assayed for IL-6 and sTNFR2 by the laboratory, with technicians blinded to sample status. Intraclass Correlation Coefficients (ICC) were calculated as a measure of how well the pre-depletion and post-depletion biomarker levels agreed in the entire dataset and in the RF+ subset (an ICC close to 1.0 indicates good agreement). ICCs were 0.99 for sTNFR2 and 0.88 for IL-6. Comparing 12 samples that were initially RF+, to post-depletion results, ICCs were 0.98 for TNRF2 and 0.93 for IL-6 indicating excellent reproducibility.

In NHS samples, high sensitivity CRP (hs-CRP) in mg/dl was measured by high sensitivity latex-enhanced immunonephelometric assay on a BN II analyzer (Dade Behring, Newark, Delaware), with a coefficient of variation of < 5% in our laboratory. Plasma from NHS samples was tested for anti-CCP antibodies with the second generation DIASTAT™ ELISA assay [Axis-Shield Diagnostics Limited, Dundee, UK] with a positive anti-CCP defined as > 5 U/ml. We tested 93 pre-clinical RA samples and 279 controls for anti-CCP. Among 67 samples that were anti-CCP negative in the pre-clinical period, there were 43 samples available from a second blood draw in 2001 (after the first symptom of RA for 38 subjects, and within one year before first symptom for 5 subjects). There were 24 subjects who were anti-CCP negative, who did not give a second blood sample. Anti-CCP assays were not performed in WHS due to limited availability of samples.

Statistical analysis

We calculated means with standard deviation and medians with range for each biomarker, IL-6, sTNFR2, hs-CRP, and anti-CCP. For any biomarker that was not normally distributed, we performed log transformation. Odds ratios and 95% confidence intervals (CI) were determined using conditional logistic regression adjusted for potential confounders comparing quartiles of biomarkers (log IL-6, sTNFR2, and log hs-CRP with cutpoints based on the control distribution within each study. The odds ratio appropriately estimates the relative risk (RR)⁴¹, therefore we henceforth use the term RR. Trends were calculated using continuous biomarker values (log transformed for IL-6 and hs-CRP) and calculating the Wald statistic. High sensitivity CRP was studied previously in WHS as a predictor of RA and was found not to be significantly associated with RA risk⁴², thus only data from IL-6 and sTNFR2 are presented for WHS. For IL-6 and sTNFR2 biomarkers, we pooled data from NHS and WHS preserving the matching, keeping cohort specific quartile cutoffs, and adjusting for covariates that were available in both datasets. SAS version 9.1 was used for all analyses. [SAS Institute, Cary NC]

To compare frequency of abnormal biomarkers in each time period between blood draw and RA onset (0.3–4, 4–8, 8–12 years), thresholds to define high vs. low biomarker levels were determined based on the top quartile vs. the bottom 3 quartiles among the control group for NHS and WHS cohorts separately.

We conducted stratified analyses using unconditional polytomous logistic regression (PLR) among NHS and WHS samples to determine if the relative risks across time interval groups differed, comparing a model in which the association between each biomarker and RA was held constant across time interval groups, to one allowing the association to vary, using the likelihood ratio test.⁴³ Polytomous logistic regression has been used in prior biomarker analyses in NHS⁴⁴ and other case-control studies of RA risk factors.⁴⁵ Thus, although we had only one blood sample prior to the first symptom of RA, we could employ this method to compare values participants within each of these three time intervals before RA, comparing cases and controls, and biomarker abnormalities. We also used polytomous logistic regression to assess the relationship between three biomarkers (log IL-6, log hs-CRP, sTNFR2) and relative risk of anti-CCP positive and anti-CCP negative RA phenotypes in the NHS only based on anti-CCP antibody status which was measured in stored blood samples. Anti-CCP status was based on a positive anti-CCP (>5 U/mL) from first blood samples collected prior to RA symptoms (N = 93 incident blood samples) or among available samples from subjects who had provided a second blood sample in 2000–2001 and were anti-CCP negative prior to RA (N = 43 total samples, 38 prevalent, 5 incident to RA symptoms). In separate models, we assessed the same relationships with RF status in NHS and WHS, based on laboratory data from the medical record review. Because of the potential for misclassification in RF and anti-CCP variables, but assuming that seropositivity on either assay captures a more severe phenotype, we have also performed an analysis stratified by seropositive (RF or anti-CCP) vs. seronegative. All stratified models were adjusted for matching factors, cigarette smoking, and body mass index.

Results

In the NHS cohorts, we confirmed 93 incident RA cases and in WHS we confirmed 77 incident RA cases with stored blood drawn prior to the first RA symptoms. Mean time to RA symptoms after blood draw was 5.1 years (range = 0.4 – 12 years) in NHS and 5.3 years (0.3 – 11 years) in 77 incident cases in WHS (Table 1). Twenty-one percent of cases developed RA 8–12 years after blood draw, 42% between 4–8 years, and 38% within the first four years (0.3–4 years before first symptom of RA). In NHS, forty-nine (53%) cases were RF positive by medical record review of records from the time of diagnosis; and 29 (28%) were anti-CCP positive prior to RA onset. Among those who were ever seropositive, 13 (23%) had radiographic changes and 10 (18%) had rheumatoid nodules at RA diagnosis compared with 7 (19%) with radiographic changes and 1 (3%) with rheumatoid nodules of those who were seronegative.

Table 1.

Characteristics of pre-clinical RA cases and matched controls at time of blood draw in the Nurses’ Health Studies and Women’s Health Study

		NHS		WHS
		RA Cases (n=93)	Matched Controls (n=279)	RA Cases (n=77)	Matched Controls (n=227)
Caucasian, n (%)		92 (99)	275 (99)	74 (96)	217 (96)
Age at blood draw, mean ± SD		54.6 ± 8.2	54.6 ± 8.1	55.4 ± 7.6	55.2 ± 7.4
Age at RA diagnosis, mean ± SD		60.2 ± 9.7	N/A	61.3 ± 7.6	N/A
Time to RA onset, years, mean ± SD		5.1 ± 3.5	N/A	5.3 ± 2.5	N/A
RF Positive*		49 (53)	N/A	46 (60)	N/A
Anti-CCP Positive (preclinical) †		26 (28)	0	N/A	N/A
Anti-CCP Positive (ever) §		39 (42)	N/A	N/A	N/A
Seropositive		57 (61%)	N/A	N/A	NA
Radiographic changes €		49(53%)	N/A	16 (21%)	N/A
Smoking
Never smoker, n (%)		44 (47)	124 (44)	28 (36)	107 (47)
Past Smoker, n (%)		45 (48)	126 (45)	42 (55)	96 (42)
Current Smoker < 15 /day		1 (1)	11 (4)	3 (4)	8 (4)
Current Smoker ≥ 15/day		3 (3)	18 (6)	4 (5)	16 (7)
Parity and Breastfeeding
Nulliparous		4 (4)	25 (9)	10 (13)	26 (11)
Parous / No Breastfeeding		30 (32)	69 (25)
Parous / ≤1 year Breastfeeding		46 (49)	130 (47)	67‡ (87)	201 (89)
Parous / >1 year Breastfeeding		13 (14)	55 (20)

NHS= Nurses’ Health Study, WHS= Women’s Health Study

^* Rheumatoid Factor positive at diagnosis, according to laboratory tests documented in the medical record

^† Anti-cyclic citrullinated peptide antibodies (anti-CCP) > 5 U/ml (positive) according to laboratory testing performed for NHS in samples collected prior to onset; anti-CCP not tested in WHS

^§ Anti-CCP measured in subjects who were anti-CCP negative in the pre-clinical time period, and provided a second blood sample (43 out of 67 anti-CCP negative subjects had available samples)

^€ Erosions or periarticular osteopenia consistent with RA

^‡ Breastfeeding information not available for WHS

In WHS, 46 (60%) were RF positive by chart review. The ethnic distribution was similar >95% Caucasian in both studies. Distribution of other potential confounders across case control groups was similar in NHS and WHS.

Median IL-6 (p=0.03), sTNFR2 (p=0.003), and anti-CCP (p<0.0001), but not median hs-CRP (p=0.09) levels, were significantly higher in preclinical RA cases compared to matched controls (Table 2) in the NHS. Mean log IL-6 (p=0.02), but not mean log hs-CRP levels (p=0.08) were significantly elevated in preclinical RA cases compared to matched controls. In WHS median IL-6 and TNFRII mean and median levels were not significantly different in cases compared to controls.

Table 2.

Inflammatory biomarkers in preclinical RA compared with matched controls in the Nurses’ Health Studies and Women’s Health Study

NHS	Cases N=93		Controls N= 279		P-value*
Biomarker	Median	Range	Median	Range
NHS
IL-6, pg.ml	1.35	(0.37 – 15.09)	1.17	(0.30 – 13.45)	0.03
sTNFR2, pg/ml	2164	(1306 – 5528)	1989	(1166 – 4995)	0.003
hs-CRP, mg/dl	1.83	(0.11 – 27.78)	1.30	(0.08 – 37.89)	0.09
Anti- CCP, U/ml	3.0	(0 - >100)	2.0	(0 – 4.0)	<0.0001
WHS	Cases N=77		Controls N= 227		P-value*
Biomarker	Median	Range	Median	Range
IL-6, pg/ml	1.61	(0.38 – 19.42)	1.39	(0.44 – 18.65)	0.41
sTNFR2, pg.ml	2203	(1159 – 4208)	2109	(1144 – 6555)	0.45

^*Wilcoxon rank sum

In the NHS, comparing the highest to the lowest quartile in multivariable adjusted models, two biomarkers were associated with an increased risk of RA: IL-6 [RR 1.6 (0.7 – 3.3), p trend = 0.03] and sTNFR2, [RR 2.9 (1.2 – 7.1), p trend = 0.01]. CRP was not significantly associated with RA, [RR 1.5 (0.7 – 3.6), p trend = 0.26] (Table 3). For comparison, in a prior analysis of preclinical anti-CCP results in the same blood samples, anti-CCP was associated with a substantial increased relative risk of RA, [RR 11.2 (95% CI 4.7 – 26.9].⁴⁶ The IL-6 association was not replicated in WHS, however, and in pooled adjusted analysis for NHS and WHS, P for trend was 0.06. Soluble TNFR2 was modestly associated with RA [RR 1.5 (0.7–3.4), p-trend 0.18] in WHS. However, in pooled analysis for NHS and WHS, sTNFR2 was significantly associated with RA, [RR 2.0 (1.1–3.6), p-trend=0.004], after adjusting for potential confounders.

Table 3.

Relative risk and 95% confidence interval for association of quartiles of IL6, sTNFR2, and CRP measured before RA onset with RA in the Nurses’ Health Study (1989–2002) and the Nurses’ Health Study II (1993–2003) and Women’s Health Study (1992–2002)

		NHS
IL-6*					P-trend
N	88	92	86	106
Median in controls	−0.426	0.017	0.332	0.943
Cases/controls	19 / 69	21 / 71	17 / 69	36 / 70
Unadjusted	1.0	1.1 (0.5 – 2.1)	0.9 (0.4 – 1.9)	2.0 (1.0 – 3.9)	0.01
Adjusted‡	1.0	0.8 (0.4 – 1.7)	0.8 (0.3 – 1.7)	1.6 (0.7 – 3.3)	0.03
sTNFR2
N	81	90	100	101
Median in controls	1551.2	1835	2143.8	2624.3
Cases/controls	12 / 69	20 / 70	30 / 70	31 / 70
Unadjusted	1.0	2.0 (0.8 – 4.6)	3.0 (1.3 – 6.9)	3.1 (1.3 – 7.3)	0.004
Adjusted‡	1.0	2.0 (0.8 – 4.8)	3.1 (1.3 – 7.5)	2.9 (1.2 – 7.1)	0.01
hs-CRP*
N	85	92	98	97
Median in controls	−1.407	−0.18	0.737	1.756
Cases/controls	15 / 70	22 / 70	29 / 69	27 / 70
Unadjusted	1.0	1.4 (0.7 – 2.9)	2.0 (1.0 – 4.2)	1.9 (0.9 – 4.0)	0.06
Adjusted‡	1.0	1.3 (0.6 – 2.8)	1.7 (0.8 – 3.6)	1.5 (0.7 – 3.6)	0.26
		WHS
IL-6*
N	76	72	77	79
Median in controls	0.270	0.162	0.545	1.139
Cases/controls	19 / 57	16 / 56	20 / 57	22 / 57
Unadjusted	1.0	0.9 (0.4 – 1.9)	1.1 (0.5 – 2.2)	1.1 (0.5 – 2.4)	0.51
Adjusted†	1.0	0.8 (0.4 – 1.8)	0.9 (0.4 – 2.0)	1.1 (0.5 – 2.6)	0.59
sTNFR2
N	74	73	75	82
Median in controls	1685.5	1961.7	2268.9	2757.7
Cases/controls	18 / 57	15 / 57	19 / 57	25 / 57
Unadjusted	1.0	0.8 (0.4 – 1.8)	1.2 (0.5 – 2.6)	1.4 (0.7 – 3.1)	0.21
Adjusted†	1.0	0.8 (0.4 – 1.8)	1.2 (0.5 – 2.6)	1.5 (0.7 – 3.4)	0.16
		NHS and WHS
IL-6*
N	164	164	163	185
Median in controls	−0.367	0.072	0.409	1.049
Cases/controls	38 / 126	37 / 127	37 / 126	58 / 127
Unadjusted	1.0	1.0 (0.6 – 1.6)	1.0 (0.6 – 1.7)	1.6 (0.9 – 2.5)	0.02
Adjusted†	1.0	0.9 (0.5 – 1.5)	0.9 (0.5 – 1.6)	1.4 (0.8 – 2.5)	0.06
sTNFR2
N	155	163	175	183
Median in controls	1610.4	1905.8	2204.2	2689.3
Cases/controls	30 / 125	35 / 128	49 / 127	56 / 127
Unadjusted	1.0	1.2 (0.7 – 2.1)	1.8 (1.0 – 3.2)	2.0 (1.2 – 3.5)	0.002
Adjusted†	1.0	1.2 (0.7 – 2.1)	1.8 (1.0 – 3.2)	2.0 (1.1 – 3.6)	0.004

^*log IL-6, log CRP as not normally distributed

^‡Adjusted: Conditional logistic regression conditioned on matching factors, adjusted for age, ethnicity, body mass index, cigarette smoking (never, past, current <15 cigarettes/day,≥15 cigarettes/day), parity, and breastfeeding

^†Adjusted: Conditional logistic regression conditioned on matching factors, adjusted for age, ethnicity, body mass index, cigarette smoking, (never, past, current <15 cigarettes/day,≥15 cigarettes/day), parity

Spearman correlations between the four biomarkers demonstrated strong associations between IL-6, sTNFR2, and hs-CRP (data not shown). There were no significant correlations between anti-CCP as a continuous variable and these three biomarkers. In analysis stratified by case status, the correlations were slightly stronger for IL-6 and sTNFR2 with hs-CRP among the preclinical RA cases (range r=0.40–0.45, p<0.0001) than among the controls (range r=0.27–0.47, p<0.0001). Because of the strong correlation among three biomarkers, IL-6, sTNFR2, and hs-CRP, we performed additional multivariable models for each biomarker in NHS, adjusting for the same covariates plus the other two biomarkers. In these analyses, the only biomarker that was independently associated with RA was sTNFR2, [RR 3.1 (95%CI 1.2–8.0), p-trend = 0.02] (results not shown).

Using PLR models for stratified analysis of the time interval between blood draw and RA onset, in the shortest time interval there was a significant trend for IL-6 and relative risk of RA (p=0.003) (Table 4). For sTNFR2, in both 0.3–4 year and 4–8 year intervals, there was a significant trend for sTNFR2 and relative risk of RA (P= 0.01, 0.05, respectively). HsCRP demonstrated no significant associations.

Table 4.

Relative risk and 95% confidence interval for association of quartiles of IL-6, sTNFR2, and hs-CRP with RA in the Nurses’ Health Study, Nurses’ Health Study II, and Women’s Health Study stratified by time interval between blood draw and RA onset

		Quartiles of IL-6			P-value
Time between blood draw and RA onset, years*	Q1	Q2	Q3	Q4	Trend†	Heterogeneity††
0.3–4 / 4–8 / 8–12 / Controls, n	13 / 19 / 6 / 125	12 / 15 / 9 / 126	12 / 20 / 5 / 125	26 / 17 / 15 / 127
<4 years	1.0	0.9 (0.4 – 2.1)	0.9 (0.4 – 2.0)	1.9 (0.9 – 3.9)	0.003	0.07
4–8 years	1.0	0.8 (0.4 – 1.6)	1.0 (0.5 – 2.0)	0.8 (0.4 – 1.7)	0.97
8–12 years	1.0	1.5 (0.5 – 4.3)	0.8 (0.2 – 2.7)	2.4 (0.9 – 6.4)	0.29
		Quartiles of sTNFR2			P-value
Time between blood draw and RA onset, years*	Q1	Q2	Q3	Q4	Trend†	Heterogeneity††
0.3–4 / 4–8 / 8–12 / Controls, n	11 / 11 / 8 / 125	14 / 13 / 7 / 127	14 / 26 / 9 / 126	24 / 21 / 10 / 125
<4 years	1.0	1.3 (0.5 – 2.9)	1.2 (0.5 – 2.9)	2.1 (1.0 – 4.6)	0.01	0.77
4–8 years	1.0	1.2 (0.5 – 2.7)	2.3 (1.1 – 4.9)	1.9 (0.9 – 4.1)	0.05
8–12 years	1.0	0.9 (0.3 – 2.4)	1.2 (0.5 – 3.2)	1.2 (0.5 – 3.2)	0.51
		Quartiles of CRP			P-value
Time between blood draw and RA onset, years*	Q1	Q2	Q3	Q4	Trend†	Heterogeneity††
0.3–4 / 4–8 / 8–12 / Controls, n	7 / 6 / 2 / 70	12 / 5 / 4 / 69	7 / 13 / 9 / 69	12 / 7 / 8 / 70
<4 years	1.0	1.8 (0.7 – 5.0)	1.0 (0.3 – 3.2)	1.7 (0.6 – 5.0)	0.43	0.73
4–8 years	1.0	0.9 (0.3 – 3.1)	2.2 (0.8 – 6.3)	1.2 (0.3 – 3.9)	0.39
8–12 years	1.0	2.1 (0.4 – 12.1)	4.6 (0.9 – 22.4)	4.0 (0.8 – 20.5)	0.13

All models adjusted for matching factors, cigarette smoking, and body mass index

^†P for trend calculated using continuous biomarker and the Wald test

^††P for heterogeneity determined using polytomous logistic regression and the likelihood ratio test, comparing a model constraining RRs to be the same across case groups versus a model allowing the RRs to differ by case groups.

Using PLR analyses, we stratified by RA autoantibody phenotypes for anti-CCP (NHS only) and RF status (NHS and WHS). There was no association between IL-6 and anti-CCP positive RA or anti-CCP negative RA in PLR models adjusted for matching factors and potential confounders (body mass index and cigarette smoking) (Table 5). For sTNFR2, the PLR models demonstrated a strong trend between sTNFR2 quartiles and anti-CCP negative RA in a fully adjusted model (P, trend < 0.001), but not for anti-CCP positive RA. There was a strong trend between sTNFR2 quartiles and RF negative RA, P, trend p=0.003. However, an association was not seen for RF positive RA. However, in PLR analyses stratified by ever seropositive (RF status from the chart and anti-CCP status from the first or second blood draw), there was no significant association between sTNFR2 and seronegative RA. For hs-CRP, there were no significant associations with RA phenotypes (data not shown).

Table 5.

Relative risk and 95% confidence interval for association of quartiles of IL-6 measured before RA onset with RA in the Nurses’ Health Study, Nurses’ Health Study II, and Women’s Health Study stratified by RA phenotypes

		Quartiles of IL-6			P-value
Anti- CCP phenotype‡	Q1	Q2	Q3	Q4	Trend†	Heterogeneity††
+ / − / missing / Controls, n	9 / 5 / 5 / 68	13 / 4 / 3 / 71	4 / 8 / 5 / 69	13 / 13 / 10 / 70		0.95
Anti-CCP +	1.0	1.4 (0.5 – 3.4)	0.4 (0.1 – 1.5)	1.3 (0.5 – 3.4)	0.08
Anti-CCP −	1.0	0.8 ( 0.2 – 2.9)	1.6 (0.5 – 5.2)	2.4 (0.8 – 7.2)	0.23
Anti-CCP missing	1.0	0.6 (0.1 – 2.5)	1.0 (0.3 – 3.6)	1.8 (0.6 – 5.8)	0.15
RF phenotype:
RF+ / RF− / Controls, n	20 / 18 / 125	22 / 14 / 126	23 / 14 / 125	29 / 29 / 127
RF+ RA	1.0	1.1 (0.6 – 2.1)	1.1 (0.6 – 2.1)	1.4 (0.7 – 2.6)	0.14	0.98
RF− RA	1.0	0.8 (0.4 – 1.6)	0.7 (0.4 – 1.6)	1.5 (0.8 – 3.0)	0.10
Seropositive phenotype:
Sero+/Sero−/Controls	13 / 6 / 68	14 / 6 / 71	9 / 8 / 69	20 / 16 / 70
Sero+ RA	1.0	1.0 (0.4 – 2.3)	0.7 (0.3 – 1.8)	1.4 (0.6 – 3.2)	0.07	>0.99
Sero− RA	1.0	0.9 (0.3 – 3.1)	1.3 (0.4 – 4.1)	2.4 (0.9 – 6.8)	0.11
		Quartiles of sTNFR2			P-value
Anti- CCP phenotype‡	Q1	Q2	Q3	Q4	Trend†	Heterogeneity††
+ / − / missing / Controls, n	5 / 4 / 3 / 69	10 / 6 / 4 / 70	15 / 7 / 8 / 70	9 / 13 / 8 / 69		0.02
Anti-CCP +	1.0	2.0 (0.7 – 6.3)	3.1 (1.0 – 9.1)	1.8 (0.6 – 5.8)	0.74
Anti-CCP −	1.0	1.5 (0.4 – 5.7)	1.8 (0.5 – 6.5)	3.2 (1.0 – 10.7)	0.001
Anti-CCP missing	1.0	1.4 (0.3 – 6.3)	2.7 (0.7 10.9)	2.6 (0.7 – 10.7)	0.32
RF phenotype:
RF+ / RF− / Controls, n	17 / 13 / 125	19 / 16 / 127	31 / 18 / 126	27 / 28 / 125
RF+ RA	1.0	1.0 (0.5 – 2.1)	1.8 (1.0 – 3.5)	1.6 (0.8 – 3.0)	0.32	0.17
RF− RA	1.0	1.2 (0.6 – 2.6)	1.4 (0.6 – 2.9)	2.1 (1.0 – 4.3)	0.003
Seropositive phenotype
Sero+/Sero-/Controls	7 / 5 / 69	14 / 6 / 70	19 / 11 / 70	16 / 14 / 69
Sero+ RA	1.0	2.0 (0.8 – 5.4)	2.8 (1.1 – 7.2)	2.3 (0.8 – 6.1)	0.32	0.13
Sero− RA	1.0	1.2 (0.4 – 4.2)	2.3 (0.7 – 6.9)	2.8 (0.9 – 8.4)	>0.99

All models adjusted for matching factors, cigarette smoking, and body mass index

^‡Anti-CCP phenotype and Seropositive phenotype analyses performed in NHS only: anti-CCP positive define as positive at either blood draw, anti-CCP negative define as negative at both blood draws, anti-CCP missing defined as negative at first blood draw and missing at second blood draw

^†P for trend calculated using continuous biomarker (log transformed) and the Wald test

^††P for heterogeneity determined using polytomous logistic regression and the likelihood ratio test, comparing a model constraining RRs to be the same across case groups versus a model allowing the RRs to differ by case groups

We dichotomized the three biomarker levels based on the top quartile compared to the bottom 3 quartiles. Of note, the CRP cut-point of 3.14 for the top quartile is quite similar to the level of 3.0 that is recommended for cardiovascular risk stratification.⁴⁷ We compared biomarker abnormalities according to time before RA (Figure 1). Among those with blood drawn within 4 years of RA symptoms, a similar proportion were anti-CCP positive (46%) as compared to high IL-6 (41%), sTNFR2 (39%) and hs-CRP (31%). Among those with blood drawn 8–12 years before RA symptoms, the proportion with anti-CCP positive (4%) was lower than the proportions with high biomarkers, IL-6 (43%), sTNFR2 (29%) or hs-CRP (35%).

Figure 1.

Prevalence of elevated preclinical biomarker levels for RA cases diagnosed 0.3 – 4 years, 4–8 years, and 8–12 years after blood collection in the Nurses’ Health Study (1989–2002), Nurses’ Health Study II (1993–2003), and Women’s Health Study (1992–2002). IL-6 and sTNFR2 include WHS and NHS data, anti-CCP and hs-CRP include only NHS data. Thresholds were anti-CCP > 5 U/ml, IL-6 ≥ 2.21 mg/dl, sTNFR2 ≥ 2442.1 mg/dl and hsCRP ≥ 3.14 mg/dl.

Discussion

Using stored blood samples from women in the Nurses’ Health Study cohorts and Women’s Health Study trial, we have examined relationships between plasma inflammatory biomarkers and subsequent risk of RA and have found evidence for immune activation in preclinical RA cases compared to matched controls. Circulating levels of IL-6 and sTNFR2 were higher in women with preclinical RA cases in the NHS than in controls, with blood collected up to 12 years prior to RA symptoms, but were not significantly higher in WHS. The lack of association between biomarkers and RA in the WHS study may be due to the less severe RA as demonstrated by the lower frequency of radiographic changes or to limited power in this cohort. It is also possible that associations seen in NHS are false positives. Ultimately replication of these investigations in other cohorts over time will be the best indication of the validity of our results. With 170 incident RA cases in the pooled models, we observed a modest 40% increased risk comparing the top versus the bottom quartile of IL-6 concentrations, and a significant 100% increased RA risk comparing the top versus the bottom quartile of sTNFR2 concentrations. We were unable to demonstrate any significant association for hs-CRP as has been demonstrated in some studies,^{48, 49} but not others. We previous published a study of hs-CRP and RA risk in the WHS that showed no significant association.^{42, 50}

The levels of inflammatory biomarkers measured during the preclinical period in this study are much lower than levels typically seen with active RA^{27, 51–54} suggesting that few subjects had active synovitis at the time of blood sampling. However, even modest elevations in these biomarkers were predictive in time intervals up to 8 years before symptoms. Analyses stratified by time intervals between blood sampling and RA onset suggested that IL-6 was associated with RA only in the shortest time interval (< 4 years), however, sTNFR2 was associated with RA both the in short interval and within 4–8 years of RA onset. However, this study design with a single preclinical blood sample does not allow us to delineate the time course of biomarker changes in the preclinical phase of RA in individual patients.

The three biomarkers (IL-6, sTNFR2, and hs-CRP) were strongly correlated, but showed no significant correlation with anti-CCP levels. Elevation in the three biomarkers in the shortest time interval before RA onset was seen in about 40% of the preclinical RA cases, and similar to the proportion with positive anti-CCP. In the longest time interval (8–12 years) the proportion of positive anti-CCP was lower than other biomarkers. This data suggests that anti-CCP antibody production is a separate phenomenon from that of cytokine production.

The landmark finding of preclinical autoantibodies in RA^{19, 20} has revolutionized thinking about the pathogenesis of this autoimmune disease.²¹ Studies suggest a complex series of events in which a genetically susceptible host is exposed to environmental risk factors that trigger autoreactivity with autoantibody production, and a second event or exposure that might trigger the onset of clinical symptoms. The interaction between genes and environmental factors such as cigarette smoking within the subtype of anti-CCP positive RA⁵⁵ provides clues to disease pathogenesis, but much of the complexity of RA etiopathogenesis has yet to be delineated. Whether smoking or other environmental factors precipitate autoreactivity is not clear. Our findings suggest that during the preclinical phase of autoantibody production, there is immune reactivity, with production of pro-inflammatory cytokines that are typically seen in symptomatic RA, namely IL-6 and TNF. The findings of a associations between sTNFR2 and incident RA among RF negative and anti-CCP negative phenotypes separately, was not consistent with findings of an analysis that combined subjects that were seronegative for both tests, suggesting that the individual results could be due to misclassification of cases that were seronegative on one assay but not the other.

Other autoimmune diseases such as Type 1 diabetes ^56–58 and systemic lupus erythematosus (SLE)²³ share these hypothesized stages of disease development: genetic susceptibility, a preclinical phase with autoantibody production, followed by a symptomatic clinical phase. Targeted therapies to prevent such diseases during the preclinical phase are being developed for Type 1 DM ⁵⁷, but have yet to be developed for either SLE or RA.

In a prospective cohort with repeated blood samples from blood bank donors in the Netherlands, subjects with preclinical RA, but not controls, were found to have a statistically significant increase over time in mean hs-CRP level , with the highest levels observed at the time of symptom onset.^{20, 59} Similar findings were demonstrated for another acute phase reactant, secretory phospholipase A₂. Time lag analysis did not show a clear pattern of whether serologic abnormalities preceded or were simultaneous with acute phase reactant elevations. Our study was limited by a blood draw at only a single time point prior to RA onset, and thus is not comparable to this study. Two published studies, utilizing a single time point hs-CRP, did not demonstrate any association between hs-CRP levels and risk of RA in 90 preclinical RA cases an average of 6.6 years prior to diagnosis⁴², or between CRP obtained in 124 preclinical RA cases up to 20 years before onset.^{42, 50}

This study has several limitations. First, we used dataset-specific quartile cut points in the analysis, precluding the ability to relate absolute levels to risk. We do not have long-term stability data from repeated blood draws among women in these studies, and thus cannot determine whether a single sample measures the average long-term level of any biomarker. Data on biomarker reproducibility from a male cohort study with blood collected 4 years apart demonstrated intraclass correlation coefficients of 0.47 for IL-6 and 0.78 for sTNFR2.⁶⁰ The lower stability of IL-6 levels over time suggests that this could be responsible for lack of significant association with RA in this study. Further, the lack of repeated blood samples prior to RA symptoms limits our ability to examine patterns of biomarker levels over time in any single subject. RA diagnosis was validated by medical record review for ACR criteria. We included a small number of cases with a clinical diagnosis of RA, in which two reviewers agreed with the diagnosis, despite record documentation of three ACR criteria. Sensitivity analysis excluding these cases had similar results to the primary analysis. The date of first symptom was documented by medical record review, which could be inaccurate and it is not possible to determine definitively that the subjects were asymptomatic at the time of blood sample collection. There was potential for misclassification of antibody status due to use of chart data from the time of diagnosis and blood samples collected at variable time points relative to RA diagnosis. This study was limited by small sample size of 170 incident RA cases with stored blood samples, thus power may be limited to detect small to moderate associations and may explain the lack of associations with seropositive phenotypes of RA in the stratified analyses. The annual incidence of RA in the cohorts is 30 per 100,000 in NHS and 27.1/100,000 in WHS, which is similar to one population based study⁶¹ but lower than rates reported by other studies^{62, 63}, possibly due to inability to obtain responses from all self-reports, missing information on medical record reviews, or a “healthy participant” effect. The Nurses’ Health Studies and Women’s Health Study are comprised of largely well-educated, Caucasian women and these findings should be replicated in more diverse cohorts. Another limitation is that we were unable to adjust for confounding due to infections and other inflammatory conditions, although we did exclude subjects who self-reported SLE or other CTDs prior to blood sample collection, and some reproductive variables that were not available in WHS. Finally, the possibility that RF could interfere with ELISA assays has been raised by other investigators ³⁹, and we found no evidence of interference by RF with the assays for IL-6 or sTNFR2 used in this study.

Our findings of elevated sTNFR2 levels, a surrogate soluble receptor that this study uses to assess potential TNFα levels, among preclinical RA patients with no documented arthritis symptoms supports the hypothesis of three phases in the development of RA: genetic susceptibility, preclinical autoimmunity with immune activation, followed by clinical symptoms. These results could have implications for screening for inflammatory biomarkers for RA risk that could be used for risk counseling, or targeted interventions to prevent RA. Further studies with repeated blood draws in asymptomatic individuals, along with repeated assessments of environmental factors, may elucidate the pathway by which immune activation progresses to symptomatic RA.