Rheumatoid factor, not antibodies against citrullinated proteins, is associated with baseline disease activity in rheumatoid arthritis clinical trials

Disease activity is the major determinant of impairment of physical function and progression of joint damage [21, 22]. Recognizing elements that influence disease activity, therefore, is of particular importance. Autoantibodies have been suggested to increase disease activity by virtue of immune complex formation with or without complement activation through binding to Fc or complement receptors (or both) and subsequent increases in production of pro-inflammatory cytokines [12, 23]. In the present study, we have shown that baseline disease activity of patients with RA in clinical trials is positively associated with the presence of RF but not ACPA. This conclusion is derived from cross-sectional analyses of clinical trial patients; all of these patients had active disease and came from many centres in many countries and regions. They therefore constitute one of the broadest representations of RA patients in which such associations have been studied.

Interestingly, aside from the lower levels of composite measures of RA disease activity, RF⁻/ACPA⁺ patients had significantly lower levels across most of the individual clinical and laboratory disease activity variables and physical function compared not only with double-positive patients but also with those who were RF⁺/ACPA⁻. In some instances, ACPA⁺/RF⁻ patients even had lower scores than double-negative patients.

On the other hand, similar levels of disease activity were observed in RF⁺/ACPA⁻ compared with double-positive patients, despite the mentioned higher level of RF in the double-positive group. This may indicate either that the presence of RF, rather than its level, is the driving force for disease activity or alternatively that there is an effect of RF levels plus some “protective” effect of ACPA reducing higher disease activity due to higher RF levels in double-positives, therefore causing an apparent similarity of disease activity in these two groups. The latter hypothesis is clearly supported by two facts, the first of which is the comparison of ACPA⁺ and ACPA⁻ patients who are seronegative for RF. Here, in the absence of RF, the presence of ACPAs is also associated with lower levels of disease activity; the second fact is the matched analyses, which controlled for the level of the respective other autoantibody.

The important role of RF, and not of ACPA, in relation with higher levels of disease activity, was also implied in another recent study, in which the authors have revealed that RF⁺ patients regardless of ACPA status compared with RF⁻/ACPA⁺ patients exhibited not only higher disease activity but also higher levels of pro-inflammatory cytokines [24]. However, in that study, it was also hypothesized that ACPAs have a synergistic effect in the presence of RF regarding disease activity, a finding which can be neither supported nor ruled out by the present study.

Based on the epidemiological findings of our study, some translational explanations can be used to understand the clear relationship of RF with disease activity as well as its absence for ACPAs. Firstly, as assessed in routine practice and clinical trials, RF is of IgM isotype whereas ACPAs are mostly IgG antibodies; IgM isotype activates complement to a much larger extent than IgG [25], thus potentially leading to a stronger secondary inflammatory response via inflammatory complement breakdown products or complement-receptor-mediated macrophage activation or both [26–28]. In this setting, ACPAs might preferentially activate inhibitory Fcγ-receptors and thus mitigate the inflammatory response [29]. At the same time, however, it is conceivable that IgM ACPAs also exist but that current tests are able to detect IgG only.

Secondly, ACPAs likely represent products of long-lived memory plasma cells [30], whereas RF may be at least partly produced by a particular subset of B lymphocytes, B-1 cells; the former may have less plasticity and responsiveness to variations of the disease process or might be less involved in cell-cell interactions in RA than the latter [31, 32]. However, it should be noted that the existence of B-1 cells in humans has been a matter of some debate, although evidence of their existence is increasing [33–35] and they have been related especially to RF production in humans long ago [36]. However, here again epidemiological findings support their existence, as RF levels may change substantially in patients responding to therapies, but ACPA levels do not [37, 38]. This notion is reinforced by the high correlation of the presence of RF with comorbidity and cardiovascular death [39, 40]. Also, RF has been shown to induce pro-inflammatory cytokines like TNF [41]. All of these findings together support concepts that RF, and not ACPAs, is the main serological inflammatory player but that ACPAs on their own not only may not fuel RA disease activity but even may mitigate it. This finding should be clearly distinguished from the well-established association of ACPAs with radiographic damage [4]. Moreover, ACPAs have been suggested to have a direct effect on osteoclastogenesis, and therefore their association with joint damage may be independent of disease activity [42].

The broad overlap of RF and ACPAs has been a practical limitation for any study investigating differential effects of each of these antibodies and is also a limitation of our study. Indeed, the small sample size of single seropositive patients in the stratified comparisons may put the analysis at risk for incorrectly failing to reject the null hypothesis of no association of antibodies with disease activity. Since our results were statistically significant, a potentially limited sample size is obviously not relevant. In addition, we took an unprecedented approach to account for this co-linearity of RF and ACPAs by the above-matched analysis. Finally, the initial results obtained in one trial dataset were confirmed in a totally unrelated second large clinical trial database.

Since the data are based on clinical trials, another potential limitation may be selection bias related to the typical inclusion and exclusion criteria. This selection bias might affect the generalisability of the results. Particularly, results cannot be generalised to patients who have very little or no disease activity but in whom these associative analyses with disease activity would not be possible for lack of active disease. In all trials, some serological criteria and some minimal disease activity requirements applied. However, these two were always independent and never conditional. For the IMAGE trial, RF seronegativity at baseline was allowed only if there was evidence of erosive disease. Since ACPAs were not among the inclusion criteria in IMAGE, the prevalence of the autoantibodies (sizes of the four groups) is not necessarily reflective of a general RA population. We can nevertheless make the points for the validity of the results in the IMAGE trial by using the following arguments: firstly, in golimumab trials, ACPAs or RF (or both) were required, thus confirming the data of the IMAGE trial in another trial with different serological inclusion requirements; secondly, patients negative for RF were required to have elevated CRP and joint damage, thus biasing ACPA⁺/RF⁻ and ACPA⁻/RF⁻ patients toward higher rather than lower disease activity compared with the respective RF⁺ groups. The opposite was observed. That this was the reverse with the ACPA⁺/RF⁻ population further supports our conclusion, while the comparably high activity seen in double-negative patients may indeed be due to the—independent—prerequisite to have higher levels of disease activity. Notably, compared with double-positive patients, single-positive patients had lower levels of that respective autoantibody (about 50 % less), suggesting that the presence of both autoantibodies may be related to a higher autoimmune thrust than the presence of just one. Also, disease activity of RA is inherently fluctuant and, therefore, assessing it at a single time point, as analysed here, may not provide the full picture; however, the large group of patients studied here inherently has an averaging effect and, in addition, patients in clinical trials have to present with active disease at a pre-screening visit as well as the baseline visit, ensuring some degree of persistence in disease activity. Furthermore, patients in clinical trials come from many centres throughout many regions, thus reducing the potential selection bias that may be inherent to studies of patients coming from a single or just few centres. Finally, many of the potential differences were accounted for by the additional matched analysis.

The cross-sectional character of our study may be regarded as a limitation because antibody levels may change over time, particularly under the influence of effective therapy, although ACPA levels fluctuate less than RF levels [38]. However, the purpose here was to look at associations with autoantibody status and not with changes during therapy. We intentionally focused only on baseline data in order to prevent confounding by treatment. Also, in a separate analysis, glucocorticoid use at baseline did not have an influence on the observed results. In addition, the similarity of the results in the different trial populations, capturing MTX-naïve patients with early RA on the one hand (rituximab database) and patients across all lengths of disease duration and prior drug experience (golimumab database) on the other, was highly confirmatory and clearly adds to the generalisability of our findings. It is nevertheless important to bear in mind that, although the present data came from two large independent databases, analyses from additional databases in independent studies will be needed to fully confirm our conclusions.