Validation of a multiplex chip-based assay for the detection of autoantibodies against citrullinated peptides

With the discovery of anti-citrullinated protein/peptide antibodies (ACPAs), the interest in autoantibodies has increased during the last decade, from both a diagnostic and a prognostic RA-perspective. In the former American College of Rheumatology (ACR) 1987 classification criteria for rheumatoid arthritis (RA) [1], the presence of rheumatoid factor (RF) accounted for one of seven criteria, of which four should be met for an RA diagnosis. With the introduction of the new 2010 RA classification criteria [2], the impact of autoantibody serology has accordingly increased, and can now contribute to half of the points needed to classify a patient as having RA.

Commercial ACPA tests generally aim to identify collectively as many antibodies against citrullinated epitopes as possible. However, on the peptide level, the ACPA response in RA patients has been shown to be heterogeneous, as different RA patients show reactivity against different citrullinated peptides [3–8]. Although some studies have investigated the impact of having simultaneous ACPA reactivity to different citrullinated peptides (see, for example, [6–10]), such studies have hitherto been performed with multiple parallel enzyme-linked immunosorbent assay (ELISA) tests, an approach that is laborious and can demand sizeable volumes of scarce serum samples (for example, from historical cohorts). Such studies of multiple detailed ACPA specificities have proven informative concerning both the risk for RA development in the context of risk genes [8, 11], and the development of risk of arthritis in healthy individuals [6] as well as in arthralgia patients [7].

Most studies on ACPA fine specificity have so far focused on individual antibody responses to epitopes on three citrullinated autoantigens identified in rheumatoid joints: fibrin/fibrinogen [12, 13], vimentin [14], and α-enolase [15, 16], as well as the skin protein filaggrin, which was used in the early RA-specific tests, before the discovery of the nature of the ACPA response [17, 18]. A smaller number of studies have also investigated the response to epitopes on the cartilage-specific type II collagen (CII), another protein that has been found to be citrullinated in RA joints [19]. This protein poses certain demands on the assay used, as the native, noncitrullinated, triple-helical CII molecule in itself is an autoantigen (anti-collagen II antibodies AC2A), with conformational epitopes that differ from the epitopes in the citrullinated counterpart [20]. The murine counterparts of both ACPA and AC2A to the same epitopes have been crystallized and found to be distinct [20, 21]. Practical limitations have probably limited the number of ACPA epitopes investigated in parallel in these earlier specificity studies.

Most commercial ACPA tests investigate only the reactivity against citrullinated antigens, and only very few tests take into account the possibility of reactivity against the protein/peptide backbone by simultaneous investigation of reactivity against the arginine-containing noncitrullinated counterpart. Although the ACPA response in RA has been shown to be citrulline specific, a number of studies have shown false ACPA reactivity with concomitant reactivity to arginine-containing control antigens in inflammatory diseases like pulmonary tuberculosis [22, 23], hepatitis C infection with cryoglobulinemia [24], and autoimmune hepatitis [25]. During our work with establishment of different ACPA ELISAs, we also found that certain peptide backbones yield increased ELISA reactivity, irrespective of whether the peptide is investigated in its citrullinated or in its native arginine-containing form. Altogether, these findings argue for the usefulness of including arginine-containing control antigens when investigating multiple citrullinated antigens or unknown patient groups. Although such control wells are normally excluded from commercial ELISAs, to increase cost efficiency, they can easily be added to miniaturized systems in a microarrayed fashion. This was recently done in a microarray, allowing the simultaneous evaluation of up to 10 antigens per chip [26].

In this article, we describe the first version of a new microarrayed platform allowing the simultaneous investigation of more than 100 specific antibodies. In the present form, the platform is used for investigation of the ACPA response against 11 previously described citrullinated peptides and their arginine-containing equivalents, as well as one novel peptide, found to be citrullinated in RA synovial membrane and not described before. The peptides represent epitopes from five human autoantigens (fibrinogen, α-enolase, vimentin, collagen type II, and filaggrin), including multiple epitopes from the four antigens (fibrinogen, α-enolase, vimentin, and collagen type II) that have been found to be citrullinated in RA joints. We show that the heterogeneous ACPA response can be investigated with this technique in a serum-saving way. We foresee that this assay can be used in studies on RA patients in both prognostic and diagnostic contexts.

In this report, we describe a microarray in which specific reactivities against multiple citrullinated peptides can be investigated in parallel. Notwithstanding the microarray approach, each one of the individual autoantibody reactivities showed good correlation to results obtained with parallel ELISAs for the individual peptides. With a large validation cohort of RA patients and controls, followed by ROC curve analyses, we could show that the individual analyses showed high diagnostic specificity. At a defined specificity level (98%), the individual peptides showed marked differences in diagnostic performance, with some individual peptides performing similarly to the anti-CCP2 assay.

When the RA patients were split into those positive and negative for anti-CCP2, a large difference was shown in the number of peptide reactivities detected in the two groups. This difference was particularly pronounced concerning individuals with multiple reactivities. However, a few patients also in the anti-CCP2-negative group showed multiple reactivities against the citrullinated peptides in the multiplex assay. This indicates that the microarray might be useful for analyses of residual ACPA reactivities that are not covered by the conventional ACPA tests like anti-CCP2. The value of this assay to generate hypotheses in a diagnostic setting remains to be evaluated by using larger sets of sera from different cohorts of RA patients and disease controls (patients with infections and other noninfectious inflammatory diseases). Of special interest will be the use of the assay to estimate the risk of RA development in patients with early undifferentiated arthritis [6] and in other individuals at risk for RA. In such future studies, it will also be possible to evaluate further the eventual diagnostic value of the individual peptides included in the present multiplex assay.

In the current design, the microarray uses very small quantities of serum (< 10 μl). This is an advantage when available serum volumes are small and when replacement samples are not available (for example, when samples have been retrieved from historic biobanks, containing samples taken from individuals at a time point before the development of RA). We are currently performing such studies on the development of individual ACPA responses in sera from RA patients retrieved from historic biobanks from the years preceding the development of RA. In such studies, an imperative demand exists to use small serum quantities very efficiently.

Certain sera yielded high background reactivities that might obscure the readout for the individual antigens. However, this is not unique to this specific microarray. It is well known, among developers of addressable laser bead immunoassays (ALBIA) and ELISA assays, that certain sera show unspecific binding. Given the setup of most commercial ELISAs, in which each serum is investigated only in wells coated with the antigen in question but without serum-specific control wells, this unspecific binding will often pass unperceived. Because we can readily assess the overall nonspecific binding in the used assay, we have excluded sera with apparent unspecific binding, but have also noticed that inclusion of such sera had very little effect on the performance at group level of the citrullinated peptides used in this investigation.

We defined the cutoff levels as the 98^th percentile for the healthy controls, implying that 2% of the healthy controls showed a positive response to each individual peptide. The discriminatory property of each individual citrullinated peptide in this multiplex assay is thus consistent with that of the anti-CCP2 assay when comparing RA patients and healthy controls. For stochastic reasons, this results in a higher number of healthy controls having a positive signal for any of the 12 analyzed peptides. These positive signals were, however, mostly only slightly above the threshold, and in only a very few of the healthy controls was multiple reactivity seen, as was the case in most anti-CCP2-positive RA patients.

Our initial pilot studies indicated a gain in diagnostic sensitivity when subtracting the fluorescence intensities from the arginine control peptides. However, in this validation, by using a large set of RA patients and controls, this was not a universal finding, and similar numbers of peptides showed decrease or increase in diagnostic performance after correction for reactivity to control peptides. We conclude that the issue of arginine subtraction still is an open question, and that the most-suitable approach has to be decided for each antigen/peptide. This is also reflected by the different approaches used in currently available commercial ACPA tests, in which some assays, like the anti-Sa assay from Euroimmun, include sample-specific control wells with noncitrullinated antigen, whereas most assays, like the anti-CCP2 test, do not.

In a recent study by Chandra et al. [26], another multiplex approach was described for the detection of ACPAs, by using a chip design with up to 10 antigens per chip. Contrary to our study, biotinylated peptides were used consistently by Chandra et al., and a different setup of peptides was investigated. Thus, several features differ between the two multiplex approaches, in which our method may have certain advantages in allowing more antigens to be investigated in parallel (up to 170 per chip). However, a straight head-to-head comparison of the two approaches to multiplexing will be needed to harmonize the interpretation of ACPA multiplex assaying.

As in our present investigation, the authors of the previous report [26] used no standard curve for normalization of the responses against citrullinated peptides. Given the many different reactivities that are investigated simultaneously, the construction of such a standard is a complex task, and its resolution is not included in the present first methods description of the current platform. However, this issue also addresses future development of our multiplex platform.

A particular issue in any effort to measure antibodies against several different epitopes on citrullinated autoantigens is the degree of cross-reactivity between these different epitopes. Previous ELISA-based investigations have addressed this issue by peptide-absorption experiments and concluded that a large fraction of the antibodies against epitopes included in our assay are not cross-reactive, whereas a smaller fraction of antibodies cross-react between different epitopes [7, 8, 33]. A further complication is that half of the peptides used in the present study, as well as the seven citrullinated peptides described in the study by Chandra et al. [26], have multiple citrullination sites. As shown in Table 2 with the 98% specificity described for anti-CCP2 [32], reactivities against several of the multicitrullinated peptides show a similar degree of positivity as the anti-CCP2 assay does in the same patient group, suggesting the presence of multiple antibody-binding sites. Thus, we cannot rule out the possibility that ACPAs with different fine specificities bind to different epitopes on these particular peptides [31]. Further detailed studies using additional peptides and different absorption approaches as done, for example, in [8], will be needed to resolve these issues.

RA subgroups with different clinical prognosis are characterized by different autoantibody profiles. Besides the ACPA (and RF)-associated RA phenotype associated with poor long-term prognosis, we have, for example, defined an acute-onset RA phenotype associated with high levels of antibodies against native human CII [34, 35], whereas others have described a mild RA phenotype associated with antibodies against the RA33 antigen [36]. Our aim is to use the microarray for the simultaneous investigation of such autoantibody-defined RA phenotypes in parallel.

Our long-term aim for the currently described multiplex assay platform is further to expand the number of peptides and protein fragments included, and further to validate the current and future antibody targets in additional clinical cohorts. Such extensions of numbers of targets will be feasible, as abundant positions are left for antigen spotting.

We have developed a microarray for the simultaneous measurement of multiple specific ACPA responses. The individual analytes have been quantitatively validated against specific ELISAs and show a high specificity comparable to that of the anti-CCP2 assay. We foresee that this assay will have a broad range of applications in studies of prediction, diagnosis, and prognosis of arthritis, as well as in studies of specific ACPA responses before and during conventional and experimental therapies in RA.