Autoantibodies to Malondialdehyde-Acetaldehyde Are Detected Prior to Rheumatoid Arthritis Diagnosis and After Other Disease Specific Autoantibodies

Abstract

Objectives:

To examine serum autoantibodies to malondialdehyde-acetaldehyde (MAA) prior to rheumatoid arthritis (RA) diagnosis.

Methods:

Anti-MAA antibody isotypes, anti-CCP2, and RF-IgM were measured in pre- and post-RA diagnosis samples (n=214 cases, 210 controls). The timing of elevations in autoantibody concentrations relative to RA diagnosis was explored using separate mixed-models for each antibody and/or isotype. Associations between pre-diagnosis autoantibody concentrations in cases were examined using mixed effects linear regression models.

Results:

IgG (log2 difference 0.34) and IgA (log2 difference 0.43) anti-MAA antibody concentrations in cases diverged from controls 3.0 and 2.3 years, respectively, pre-RA diagnosis (p <0.05). There was no evidence of case-control divergence for IgM anti-MAA antibody. Anti-CCP2 and RF-IgM concentrations diverged between cases and controls beginning 17.6 and 7.2 years, respectively, prior to RA diagnosis. All three anti-MAA antibody isotypes were significantly associated with anti-CCP2 antibody and RF concentrations pre-diagnosis (β = 0.22–0.27 for RF-IgM; β = 0.44–0.93 for anti-CCP2; p <0.001).

Conclusions:

IgG and IgA anti-MAA autoantibodies are elevated pre-RA diagnosis but appear later in the pre-clinical course than anti-CCP2 or RF. These results suggest that MAA formation and anti-MAA immune responses could have a role in the transition from subclinical autoimmunity to clinically apparent arthritis.

Although demonstrating limited disease specificity (1), reports suggest that malondialdehyde-acetaldehyde (MAA) protein adducts and anti-MAA immune responses could play a pathogenic role in rheumatoid arthritis (RA). We have demonstrated, for instance, that MAA-modified proteins are enriched in RA-affected synovium and lung tissues where they co-localize with citrullinated antigen (2, 3). In addition, circulating levels of anti-MAA antibodies are increased in both seropositive and seronegative RA patients compared to controls and are associated with serum concentrations of both rheumatoid factor (RF) and anti-citrullinated protein antibody (ACPA) (4). To date, there has been no investigation examining whether anti-MAA antibodies precede disease onset, results that would further implicate MAA and related immune responses in RA development.

The purpose of this study was to evaluate serum anti-MAA antibody in individuals pre- and post-RA diagnosis using samples collected longitudinally from a cohort of active-duty military personnel with and without RA.

Methods

Study participants.

Serum samples from the U.S. Department of Defense Serum Repository (5) were obtained from individuals prior to and after RA diagnosis. A screen was performed of the military’s electronic medical record for active duty personnel with ≥2 RA diagnostic codes (≥1 from a rheumatologist). Medical records were reviewed for satisfaction of the 1987 American College of Rheumatology classification criteria (6) and date of diagnosis. Up to four samples per case were retrieved. These included, when available, three pre- and one post-diagnosis sample. A control was identified for each case, excluding individuals with previous RA or other forms of inflammatory arthritis. Controls were matched to cases based on age, sex, race, region of enlistment, and duration of sample storage. Prior to serologic testing, four controls were excluded as corresponding medical records were not available, leaving an analytic dataset of 214 cases and 210 controls.

Clinical data were abstracted and included: race/ethnicity, comorbidities, smoking status, and among cases, date of diagnosis/onset, the use of post-diagnostic disease-modifying anti-rheumatic drugs (DMARDs), and presence of radiographic erosions.

Autoantibody assays.

In brief, anti-citrullinated protein antibody (ACPA) was measured using a commercially available second-generation anti-CCP2 ELISA (Diastat, Axis-Shield Diagnostics, Dundee, Scotland; positivity ≥5 U/ml). IgM-rheumatoid factor (RF) was quantified (IU/ml) using ELISA (Inova Diagnostics, San Diego, CA; positivity >18.9 IU/ml). Anti-MAA antibodies (IgA, IgM, and IgG; ng/ml) were quantified using ELISA as previously reported (3, 4). Briefly, human serum albumin (HSA) and MAA-modified HSA (HSA-MAA) were coated on ELISA plates. Serum was evaluated for reactivity and anti-MAA antibody concentrations were determined by subtraction (HSA-MAA reactivity minus HSA reactivity). As no prior threshold to define anti-MAA seropositivity has been defined, we arbitrarily established positivity using ≥90^th percentile in a random selection of 156 controls (using only the remaining 54 controls in case-control comparisons examining the frequency of anti-MAA positivity).

Statistical analysis.

Participant characteristics were compared using chi-square or t-tests. Case-control differences in autoantibody concentrations at each time point were examined using a Wilcoxon rank-sum test. To examine possible effect modification between ACPA and anti-MAA pre-diagnosis, differences in anti-MAA antibody concentrations were examined in analyses stratified by ACPA status for each given time point. Anti-MAA antibody concentrations were log2 transformed for further analyses. As recently reported for ACPA and RF (7), the relative timing of elevations in anti-MAA antibody concentrations were explored using separate mixed-models for each isotype with the isotype as the dependent variable, RA status as the independent variable, random subject intercepts and slopes assumed to have multivariate normal distributions, and time as a continuous covariate modeled using cubic B-splines that were allowed to vary by group with internal knots placed at tertiles and boundary knots placed at extremes (8). Cholesky decomposition was used to constrain covariance matrix of the random effects to be positive-definite. Autoantibody concentrations were compared at one-month intervals to identify the time at which levels first differed significantly (p <0.05) between cases and controls. Comparisons were adjusted for multiple testing using a stepdown Holm-simulated method to control the family-wise type I error rate.

Associations between pre-diagnosis autoantibody concentrations in individuals developing RA were examined using mixed-effects linear regression models with either anti-CCP2 or RF-IgM as the dependent variable, a fixed effect for each of the anti-MAA antibodies in turn, and random subject intercepts.

Analyses were completed using SAS v9.4 (SAS Institute, Cary, NC).

The study protocol was approved by the Institutional Review Boards at the DoDSR, Walter Reed National Military Medical Center, and the University of Colorado.

Results

Participant characteristics are summarized in Table 1. Cases and controls were similar, although cases trended towards a higher frequency of smoking (32% vs. 23%, p = 0.06) and had higher frequencies of seropositivity to RF-IgM and anti-CCP2.

Table 1:

Study Participant Characteristics and Autoantibody Values

Characteristic	RA Cases n=214	Controls n=210	p-value
Age at time of diagnosis, mean (SD)	36.8 (7.9)	36.8 (7.9)	0.99
Female, %	48	48	0.99
Non-Hispanic White, %	58	55	0.93
Ever Smoker*, %	32	23	0.06
Anti-CCP2, U/ml, median (IQ range)**
Immediate / closest pre-diagnosis sample	59 (1, 216)	0.3 (0.1, 1.0)	<0.001
Post diagnosis sample	51 (3, 204)	0.4 (0.1, 0.9)	<0.001
RF-IgM, IU/ml, median (IQ range)**
Immediate / closest pre-diagnosis sample	29 (8, 105)	4 (2, 7)	<0.001
Post diagnosis sample	29 (8, 105)	4 (2, 8)	<0.001
Anti-MAA IgA, AU/ml, median (IQ range)**
Immediate / closest pre-diagnosis sample	124 (74, 247)	91 (53, 165)	<0.001
Post diagnosis sample	119 (63, 202)	93 (60, 204)	0.18
Anti-MAA IgM, AU/ml, median (IQ range)**
Immediate / closest pre-diagnosis sample	359 (217, 584)	360 (191, 540)	0.49
Post diagnosis sample	366 (227, 563)	368 (211, 557)	0.74
Anti-MAA IgG , AU/ml, median (IQ range)**
Immediate / closest pre-diagnosis sample	505 (301, 877)	348 (206, 568)	<0.001
Post diagnosis sample	438 (251, 774)	306 (207, 574)	0.002
Follow-up time post-RA diagnosis (cases) / from index date (controls), mean (SD)	6.9 (3.6)	7.2 (4.4)	0.54
Post-diagnosis RA medications (Ever Used), %
Methotrexate	87	-	-
Anti-TNF inhibitor	73	-	-
Radiographic erosions, %	45	-	-
Number of samples tested, no. (%)			-
2	0 (0)	1 (0)
3	3 (1)	102 (49)
4	211 (99)	107 (51)
Span of pre-RA samples in years, mean (SD)	−5.1 (5.7)	-	-
Span, oldest to newest sample, years, mean (SD)	12.8 (5.6)	12.3 (5.4)	0.39

^*Data missing regarding ‘ever smoking’ in 5 cases and 89 controls (imputed as never smokers); when analysis limited to non-missing data, ever smoking observed in 33% of cases and 41% of controls (p=0.15)

^**Immediate pre-diagnosis samples available for 212 cases and 207 controls; post-diagnosis samples available for 214 cases and 109 controls

Pre- and post-diagnosis trends in anti-CCP2 antibody, RF-IgM, and anti-MAA isotype antibody concentrations are shown in Figure 1. IgG anti-MAA antibody concentrations in cases diverged significantly from controls 3.0 years pre-diagnosis whereas IgA anti-MAA antibody became divergent 2.3 years before diagnosis (p <0.05 for both). There was no evidence of IgM anti-MAA antibody divergence over follow-up. Using the same approach, anti-CCP2 antibody and RF-IgM concentrations diverged significantly between cases and controls beginning 17.6 and 7.2 years, respectively, pre-diagnosis. In analyses limited to pre-diagnosis samples, all three anti-MAA antibody isotypes (IgA, IgM and IgG) were significantly associated with anti-CCP2 antibody and RF concentration (p <0.001) (Table 2).

Figure 1:

A. Pre-diagnostic trends in serum autoantibody concentrations (RA cases shown with solid lines vs. controls shown with dashed lines). Case-control differences in IgG anti-MAA antibody observed 3.0 years pre-diagnosis (a) whereas differences in IgA anti-MAA emerged 2.3 years (b) before RA diagnosis (p <0.05 for both); no evidence of case-control divergence for IgM anti-MAA antibody observed (not shown); RF-IgM and anti-CCP2 concentrations were higher in cases than controls beginning 7.2 (c) and 17.6 years (d), respectively, prior to RA diagnosis (note: for brevity, the graph is truncated at 12 years pre-diagnosis). B. Pre-diagnostic trend for IgG anti-MAA with confidence intervals. C. Pre-diagnostic trends for IgA anti-MAA with confidence intervals.

Table 2:

Associations of pre-diagnosis anti-MAA antibody concentrations with anti-CCP2 and IgM-RF concentrations in individuals developing RA (n=214)^*

	Anti-CCP2		IgM-RF
	β Coefficient (95% CI)	p-value	β Coefficient (95% CI)	p-value
IgA Anti-MAA	0.68 (0.47 to 0.90)	<0.001	0.29 (0.19 to 0.39)	<0.001
IgM Anti-MAA	0.42 (0.23 to 0.61)	<0.001	0.21 (0.12 to 0.29)	<0.001
IgG Anti-MAA	0.95 (0.72 to 1.17)	<0.001	0.32 (0.21 to 0.43)	<0.001

^*Associations examined using mixed effects linear regression with anti-CCP2 or IgM-RF as the dependent variable, each anti-MAA isotype as the independent variable in turn, and random subject intercepts for individual RA patients; autoantibodies log (base-2) transformed for analysis

Autoantibody concentrations and the proportion positive over time are shown in Supplemental Tables 1-2. Case-control differences in pre-diagnosis anti-MAA antibody concentrations limited to ACPA negative and ACPA positive cases for a given time point are shown in Supplemental Table 3. In analyses limited to ACPA negative cases and controls, differences in IgA anti-MAA antibody were similar in magnitude to overall results using the closest pre-diagnosis sample, although this difference did not achieve statistical significance and other differences were attenuated. In contrast differences in anti-MAA antibody were amplified in analyses limited to ACPA positive individuals and controls. Specifically, in this analysis, significant differences in IgG anti-MAA antibody were apparent at all sample time points (p <0.001) including the earliest sample. Using an arbitrary threshold of ≥90^th percentile of randomly selected controls (n = 156) to define ‘positivity’, anti-MAA antibody positivity in cases ranged between 7% (IgM) and 26% (IgG) in the pre-diagnosis sample most proximate to disease onset. Although positive IgA and IgG anti-MAA were observed in pre-diagnosis samples 2- to 4-fold more common in cases than controls, these differences did not reach statistical significance, in part owing to the small number of controls used in the comparison (n = 54).

Discussion

In this study, we have shown for the first time that IgG and IgA anti-MAA autoantibodies are elevated pre-RA diagnosis. Importantly, this work, coupled with prior efforts by our group (7), demonstrate that anti-MAA antibodies appear later in the pre-clinical course than ACPA or RF. Moreover, differences in pre-diagnosis concentrations of IgG and IgA anti-MAA antibodies were more pronounced in ACPA positive than ACPA negative cases. These data are consistent with prior reports from our group suggesting that MAA modifications, which colocalize with citrullinated proteins in RA effected tissues (2–4), might act to enhance RA-specific autoimmunity. Specifically, we have previously shown that the generation of ACPA and anti-citrulline T cell responses are markedly enhanced in animals immunized with co-modified proteins compared to animals immunized with only citrullinated antigen (9).

The early divergence of ACPA and RF, detectable years before the articular manifestations in RA, has fostered speculation that these autoantibodies might initially be produced at extra-articular mucosal sites (10). Support for a role of mucosal tissues in the evolution of RA-related autoimmunity include contributions of mucosal irritants (cigarette smoking, silica) in disease risk, the formation of ectopic lymphoid tissues in RA tissues, and the identification IgA autoantibody isotypes both pre- and post-diagnosis.

Whether anti-MAA antibodies might be produced at extra-articular sites is unknown. Previous work has demonstrated that anti-MAA antibodies are enriched in synovia from RA patients compared to paired sera. This contrasts with ACPA and RF, both of which were found in higher concentrations in the circulation than in paired synovial fluid (3). Moreover, MAA adducts co-localize in synovial tissues with B cells. Taken together, these observations suggest that anti-MAA antibodies are produced locally in RA-impacted joints. Thus, it is possible that the differential detection of anti-MAA antibody during the pre-diagnosis period in the current study, occurring much more proximate to disease onset, could serve as a marker of early articular inflammation. If true, local production (vs. systemic production) could explain the much smaller effect sizes observed for anti-MAA antibody divergence compared to ACPA or RF as well as the limited sensitivity observed in this context.

In addition to its detection pre-diagnosis prior to arthritis onset, there are additional lines of evidence to suggest that anti-MAA autoantibodies could also be produced at extra-articular mucosal sites. We have demonstrated, for instance, that MAA adducts are preferentially expressed in RA derived lung tissues where they co-localize not only with autoreactive B cells, but also with citrullinated antigens (2, 3). The similar times of divergence of IgG and IgA anti-MAA pre-RA diagnosis could indicate that mucosal processes are playing a role in the triggering or propagation of these immune responses. Although we observed trends suggesting higher IgM anti-MAA in cases vs. controls, this isotype may be common outside the context of RA (1). At the same time, its association with ACPA and RF observed in this study paralleled that for IgA and IgG anti-MAA, suggesting that similar processes may be driving expansion of these autoantibodies.

There are limitations to this study. The focus on active-duty military could impact generalizability, a possibility reflected in the high proportion of men (52%) and young age of RA onset (mean 37 years) observed. This latter point may be particularly important as a young age of onset may reflect more severe RA, which might explain the high frequency of methotrexate and biologic use among patients post-diagnosis. We lacked detailed information on comorbid conditions such as RA-related lung disease, prohibiting meaningful analyses that could inform possible extra-articular involvement as a ‘source’ of autoantibody production. The number of pre-diagnosis samples available was also limited to ~3 samples per case and the majority of these (~67%) were from the period spanning disease onset to 5 years pre-diagnosis, limiting overall precision of the estimates generated. This fact, along with a limited sample size, prohibits our ability to adequately examine effect modifications between autoantibody responses that could be present. Given inherent delays in the provision of clinical diagnoses, we have likely overestimated the time elapsed between autoantibody elevations and true “arthritis onset.” While highly relevant to the interpretation of the overall results, such an overestimation would not impact the relative timing differences between the different autoantibodies examined. Finally, the anti-MAA assay used in this study (and others from our group) have used modified human serum albumin. Whether assays incorporating more RA-relevant antigen (e.g. vimentin, enolase, collagen or others) could improve assay performance (e.g. sensitivity or specificity) in this setting remains unknown.

In conclusion, we found that IgA and IgG anti-MAA antibody isotypes differentiated individuals bound to develop RA from controls on average 2 to 3 years prior to diagnosis. Moreover, anti-MAA antibody concentrations measured prior to RA diagnosis are strongly associated with RF and ACPA. These novel autoantibodies, however, diverged much later in the pre-clinical course than either IgM-RF or anti-CCP2 antibody. These results suggest that MAA adduct formation and related immune responses could be involved in the transition from subclinical autoimmunity to clinically apparent RA.