The risk of systemic lupus erythematosus associated with Epstein–Barr virus infection: a systematic review and meta-analysis

Systemic lupus erythematosus is an autoimmune disease without clear pathogenesis. This disease is characterized by polyclonal B cell activation and altered T cell function with the presence of multiple autoantibodies and impaired cell-mediated immunity. It is believed that both genetic and environmental factors contribute to disease development [1]. Bacteria and virus infections are major important environmental factors that may be initiated and involved in the pathogenesis of SLE. The Epstein–Barr virus (EBV) is of particular interest. It has been reported that Epstein–Barr nuclear antigen-1 (EBNA-1) has a high degree of homology with some proteins that cause an autoimmune humoral response [2, 3]. This molecular mimicry may play an essential role in the induction of antibodies by EBV infection in SLE patients. Recently, Harley et al. [4] reported that nearly half of SLE risk loci are occupied by the EBNA-2 protein and many co-cluster with transcription factors, providing an essential new perspective on the mechanism of SLE pathogenesis.

The first positive association between EBV infection and SLE was found in 1971 [5]. Since then, many researchers have used various angles to investigate the possibility of this link. However, previous studies have failed to detect a consistent association. The first and only systematic review that updated on the association between SLE and sero-positivity for different EBV antibodies was that of Hanlon et al. [6]. These authors found a statistically significant higher sero-prevalence of viral capsid antigen (VCA) IgG but not EBNA1 in cases compared with controls. Meta-analyses for early antigen (EA)/D IgG and VCA IgA also significantly demonstrated higher ORs in cases compared with controls.

Many studies have since been published and may help in more fully estimating the association. In addition, some authors postulated that the increase in antibodies in SLE was brought about by generalized immune hyper-reactivity in lupus rather than by any specific property of the EBV. It was later thought that the best way to clarify this question would be at the DNA level [7‐10]. Therefore, we undertook this meta-analysis with a broader, more comprehensive search strategy with no language restriction including both the sero-prevalence of antibodies and the EBV DNA load.

We conducted a systematic literature search of MEDLINE (source: PubMed, January 1, 1966 to June 20, 2018) and EMBASE (January 1, 1974 to June 20, 2018) using text and keywords in combination as MeSH terms and text words (Additional file: Table S1). We searched articles published without language restrictions and scrutinized the references from these studies to identify other relevant studies.

A total of 2000 references were retrieved using our initial search algorithm, and of these, 33 studies were ultimately included (Additional file: Figure S1). One study was in Turkish [15], one in Chinese [16], one in French [17], and the others were in English. All of the reports were case–control studies published between 1988 and 2018. There were two pediatric studies and 31 adult studies. The characteristics of the 33 eligible studies are listed in Table 1.

The features of the participants in the included studies are summarized in Table 1. There were 2814 cases and 4048 controls. The participants were almost all female with an average age of 37.5 years. The median sample sizes of the cases and controls were 85 and 123, respectively. Most of the studies specified using 1982 or 1997 American College of Rheumatology (ACR) criteria for SLE diagnoses (29 of 33 studies) for cases. The controls included healthy and non-healthy participants with the majority of the studies recruiting healthy controls. Only eight studies recruited samples from the general community. Most studies recruited hospital controls or did not state the source. There were four studies that recruited controls from patients’ relatives.

Our review has again found an association between EBV sero-positivity and SLE based on VCA antibody (IgG, IgA, IgM), EBNA IgA, and EA antibody (IgG, IgA, IgM) testing. We did not observe evidence of differences in the sero-prevalence of EBNA IgG, which is indicative of latent infection. This analysis shows a significant association between the EBV DNA-positive rate and SLE (OR: 3.86, 95% CI 1.52–9.83, p = 0.005). Furthermore, meta-regression demonstrates that the average age of the participants negatively correlated with the association between DNA positivity and SLE (p = 0.004). To our knowledge, this systemic review is the first attempt to combine such estimates of the association between SLE and EBV DNA positivity.

Hanlon et al. [6] included 25 studies in their meta-analysis, but four of the studies did not specify the antigen for the tested antibody. Therefore, only 21 of these studies were used for analysis. In our review, 12 additional studies were added for analysis. In addition to increasing the total number of cases, the average sample size also increased. As a matter of fact, our results even more precisely verify Hanlon’s findings.

There was considerable heterogeneity between studies, and we examined different factors that might have been influential. Although none of the subgroup analyses reached statistical significance. Slightly higher OR values were observed for the pediatric studies with a wide confidence interval (0.08–187.98) compared with adult studies. James et al. [8] found that EBV infection increases SLE risk by as much as 50-fold in children, while Tsai et al. [10] found no difference between SLE patients and healthy controls. This discrepancy may be due to the small number of pediatric studies (only two) or only chance.

We could not find any difference in the combined OR of the studies using more recent criteria (1997 ACR criteria for SLE) compared with older criteria (1982 ACR criteria for SLE) or failure to report criteria (which may reflect a lack of reporting). Studies matched for age, sex, or living place did not have different ORs for EBV IgG sero-positivity. Other potential vital confounders include ethnicity and socioeconomic status. We did not test the effects of controlling for these factors because few studies have specifically mentioned controlling for them. Parks et al. [29] reported a racial difference in the association between EBV IgA and SLE where African-Americans were more likely to have a history of EBV infection. Comparing the ORs of studies that used IFA to detect EBV antibodies vs. those that used ELISA revealed a similar result. However, Almohmeed et al. [45] demonstrated that using IFA to test VCA and EA IgG had twice the OR compared with ELISA for MS. Because there is a lack of detailed studies analyzing the two testing methods for SLE, we could not explain the discrepancy between these two autoimmune diseases. Other elements that may contribute to differences in estimates of ORs include the source and selection of patients and controls. Few studies have clearly described their methods for selecting cases (random or consecutive). Ideally, society would be the best source for controls. However, only eight studies have been deemed to have appropriate community controls [16, 18, 29, 33, 40‐42, 44]. There were six studies that recruited bone marrow or blood donors [15, 17, 22, 28, 30, 31], which may be considered inappropriate community controls because such donors were selected with unhealthy risk behavior. There were four studies that recruited relatives of patients [8, 23, 35, 36], which are also inappropriate controls as relatives may share the same susceptible genetic factors as patients. Sixteen studies have recruited only seemingly healthy controls, and we observed that nine of these studies found no statistically significant higher EBV IgG sero-prevalence in cases compared with controls [19, 26‐29, 34, 37, 41, 42].

The pooled OR for EBV DNA and SLE was 3.86, which is consistent with sero-antibody prevalence. Yu et al. [39] reported a high OR for an SLE patient that was significantly different from others because the controls had low positivity in serum samples. Although three of seven studies failed to find a significant difference in the DNA-positivity rate and SLE [9, 37, 44], and one found a high DNA load in SLE compared with controls [44]. The negative result may be due to chance or the small sizes of these studies. Yu et al. [39] reported that the EBV DNA-positive rate declines with age for controls but not for SLE patients. Although our subgroup analysis did not find a significant difference between age-matched and unmatched groups, meta-regression demonstrated that the OR for the DNA-positive rate declines with average age. This may be due to the fact that young SLE patients tend to be more infected by EBV, which is consistent with James et al. [8].

EBV subclinically infects a majority of individuals worldwide and generates multiple antibodies in the serum. The most common antibody is VCA IgG, which represents exposure to EBV and lasts lifelong. Our studies found that the overall prevalence of VCA IgG in SLE patients and controls was 95.0 and 90.8%, respectively, and the OR reached a significant difference. In addition, many studies have revealed a higher titer in SLE patients compared with controls [18, 23, 33‐35]. Otherwise, the best way to specify the relationship between SLE and EBV infection is at the DNA level, which also had a higher OR in our study. A conclusion may be drawn that prior EBV infection is an essential factor for the development of SLE. Many studies had found that some humoral autoimmunity in SLE arises from molecular mimicry between EBNA-1 and lupus auto-antigens, which provide further evidence for suspecting an etiologic role for EBV in SLE. However, our study failed to find a higher prevalence of EBNA IgG in SLE patients, which is difficult to explain. Furthermore, a prospective study with a large sample size should be considered to solve this enigma. Antibodies directed against EA/D are generally known as an indication of lytic replication. The sero-prevalence of these antibodies is significantly higher for SLE patients, particularly for IgA, compared with controls, and the ORs also achieved statistical significance. The majority of the included studies have been consistent on this issue. Draborg et al. [33] thought that a possible mechanism for this phenomenon as a specific intrinsic defect in the immune systems of SLE patients is independent of immunosuppressive medication therapy.

The strengths of this meta-analysis include its comprehensive search strategy and absence of language restrictions; however, limitations must also be considered. First, the quality of individual studies was not always ideal, as demonstrated by a general lack of information on the recruitment of consecutive patients and selection controls for all studies. Second, there is heterogeneity in the ORs across studies, and subgroup analysis did not reveal the source. This heterogeneity may be derived from other confounders, such as ethnicity and laboratory measurements. Third, some of the funnel plot analyses showed asymmetry that suggested the possibility of publication bias. The trim and fill sensitivity analysis did not modify the general results, suggesting that association is not an artifact of unpublished negative studies. Nevertheless, that possibility is not entirely excluded by this method. Fourth, similar to all meta-analyses, our study has the limitation of being a retrospective analysis; thus, further prospective cohort studies are warranted to confirm these findings.

In conclusion, our review supports the hypothesis that prior EBV infection is an essential factor for the development of SLE as indicated by a higher positivity of VCA IgG and EBV DNA. These findings also suggest abnormal (EA, IgA, EBV DNA) humoral immune responses to EBV in the context of SLE. However, the studies included in our meta-analysis are heterogeneous and have small sample sizes. At the same time, many studies did not match for age and gender, only a few matched for race and other confounding factors, and descriptions of recruitment and laboratory testing were not specified in most of the papers. Moreover, the role of publication bias could not be excluded. Large prospective studies are needed to determine the relationship between SLE and infection before we could draw a causal relationship between the two.