Meta-analysis confirms a role for deletion in FCGR3B in autoimmune phenotypes

Although deletion in the low-affinity IgG receptor gene FCGR3B has repeatedly been implicated in systemic autoimmune disease, the role of FCGR3B copy number variation (CNV) in autoimmunity still remains unclear. Factors such as study size, ethnicity, specific disease phenotype and experimental methodology may explain these conflicting results. Here we aimed at using meta-analysis to assess the role for FCGR3B CNV in autoimmunity. We excluded studies using SybrGreen-based genotyping and found strong evidence for association between low (<2) FCGR3B CN and systemic lupus erythematosus [OR = 1.59 (1.32–1.92), P_meta=9.1 × 10⁻⁷], but not for rheumatoid arthritis [OR = 1.36 (0.89–2.06), P= 0.15]. However, a combined autoimmune phenotype analysis supports the deletion of FCGR3B as a risk factor for non-organ-specific autoimmunity [OR = 1.44 (1.28–1.62), P_meta= 2.9 × 10⁻⁹]. This meta-analysis implicates the clearance of immune complex in the etiology of non-organ-specific autoimmune disease.

INTRODUCTION

Low-affinity Fc receptors (FcγR) play a pivotal role in the initiation and regulation of the antibody-mediated immune response, linking humoral and cellular immunity. These receptors recognize the constant domain of IgG and are involved in the mobilization of macrophages, natural killer T-cells and neutrophils to areas of immune complex deposition. They also play an important role in the recognition and clearance of immune complexes, and as modulators of B-cell activity [see (1) for review]. The low-affinity Fc receptor FcγR gene family is clustered on chromosome 1q22, and consists of four activating receptors (FcγRIIA, FcγRIIC, FcγRIIIA and FcγRIIIB) and one inhibitory receptor (FcγRIIB) that have arisen through gene duplication and non-homologous recombination (2). Although a number of single-nucleotide polymorphisms (SNPs) in FcγR genes have been associated with a variety of autoimmune disorders, the role of copy-number variation (CNV) in disease susceptibility has only recently begun to be investigated (2). FCGR2C, FCGR3A and FCGR3B all exhibit CNV (3) and, of these, a role for FCGR3B in risk of autoimmunity has been the most intensively investigated to date. Association between FCGR3B deletion and a number of autoimmune disorders has been reported in European populations, including systemic lupus erythematosus (SLE) (4–6), rheumatoid arthritis (RA) (7) and in Japanese for ulcerative colitis (UC) (8).

The risk associated with low FCGR3B CN (0–1 alleles) and/or the protective effect of high (three or more alleles) CN has not been consistently replicated. For example both low (5) and high (6) FCGR3B CN have been reported as a risk factor for ANCA-associated vasculitis (AAV), whilst a separate study reports no association between FCGR3B CN and AAV (9). Both duplications and deletions have been associated with SLE and Sjogren's syndrome (10), and high CN has been reported as a risk factor for RA (11) whereas a separate study found association between low CN and RA (7). There are a number of potential explanations for these discrepancies: in addition to factors such as cohort size and resultant power, population-specific effects and ascertainment bias, differences in both detection method and analytic approach, and the difficulty in designing FCGR3B-specific probes and primers owing to strong sequence identity with FCGR3A make it challenging to compare results between studies.

In the case of the traditionally regarded organ-specific autoimmune diseases, there is no evidence for association between FCGR3B CN and organ-specific Addison's and Graves' disease (5), suggesting that FCGR3B gene dosage may modulate the risk of systemic but not organ-specific autoimmune disease, although it has been reported that high CN is a risk factor for idiopathic pulmonary fibrosis (12). It has consequently been hypothesized that low FCGR3B CN may predispose to diseases characterized by immune complex deposition (13).

To obtain a clearer understanding of the role of FCGR3B CN in autoimmune disease, we have undertaken a systematic review of all the studies published up to August 2011 and carried out meta-analyses of the results.

RESULTS

A total of 16 studies were identified in our review of the literature, with three initial reports (4,6,13) subsequently incorporated in later, more comprehensive papers (3,5,9). Summary data from these studies are presented in the Supplementary Data, both when CN was measured as a discrete (Supplementary Data) and/or continuous variable (Supplementary Data).

At least four different methodologies have been used to determine FCGR3B CN, and a number of different regions of the gene have been targeted (see Supplementary Data). In addition, it is likely that there are population-specific differences in the frequency of deletions and duplication events. To investigate the extent to which differences in ethnicity and experimental methodology influence apparent CN, we compared the distribution of CN between studies. The frequency of single-copy individuals in control populations varied depending on the detection method used (Table 1). Notably, the frequency of CN = 2 was lowest with SybrGreen in both European (61–70%) and Asian (39%) sample sets compared with other methods (81–88 and 59–76%, respectively). Overall, the CN distribution as determined by SybrGreen was significantly different from that measured by other detection methods; in European controls, for example, P= 0.018 for Taqman, 0.009 for paralog ratio test (PRT) and 0.014 for multiplex ligand-dependent probe amplification (MLPA), whilst there was no significant difference between Taqman, PRT or MLPA (Fig. 1).

Table 1.

Percentage of individuals in each FCGR3B CN class in control populations

Study	Population	n	Detection method	CN < 2	CN = 2	CN > 2
Fanciulli (5)	European	1052	SybrGreen, rounded CN	18.9	60.9	22.2
Bourzanos (12)		221	SybrGreen, post-hoc CN assignment	11.3	69.7	19.0
McKinney (7)		1309	Taqman, post-hoc CN assignment	6.0	82.0	12.0
Hollox (26)		189	PRT/REDVR	4.2	87.8	7.9
Morris (17)		365	PRT/REDVR	7.1	81.1	11.8
Neiderer (9)		1817	PRT	7.8	81.7	10.6
Marques (11)		304	MLPA	6.7	83.9	9.4
Breunis (13)		129	MLPA	7.6	82.2	10.9
Mamtani (10)		409	Taqman, rounded CN	∼10	∼82	∼8
Yang (19)	Asiana	173	SybrGreen, Rounded CN	10.4	39.3	49.7
Zhou (18)		146	Taqman, post hoc CN assignment	19.2	64.4	15.5
Asano (8)		2057	Taqman, post hoc CN assignment	7.0	71.1	21.9
Hollox (26)		64	PRT/REDVR	7.8	59.3	32.8
Niederer (9)		1874	PRT	10.0	76.1	13.9
Hollox (26)	Africanb	35	PRT/REDVR	14.3	85.7	0
Molokhia (16)		450	PRT/REDVR	16.7	76.0	7.3
Niederer (9)		152	PRT	17.2	75.2	6.6

^aMainland Chinese, Hong Kong Chinese, Japanese, Vietnamese.

^bYoruban, Afro-Caribbean, Kenyan.

Figure 1.

Comparison of different methodologies on FCGR3B CN distribution in healthy populations. The average frequency of FCGR3B deletions and duplications was calculated for SybrGreen, Taqman, PRT and MLPA in both European (A) and Asian (B) populations using data from published studies. Three-way contingency tables were used to compare the results between methods. The P-values for these pairwise comparisons are presented to the right of the CN distribution graphs.

Comparison of our data obtained using SybrGreen and Taqman PCR shows that while Taqman produces results that cluster around integer values, this was not the case with SybrGreen (Fig. 2). We also compared each method with PRT (Supplementary Data); there was 39% correspondence between CN as determined by SybrGreen and PRT, compared with 93% between Taqman and PRT. This is consistent with other reports (14). For these reasons, studies using SybrGreen detection were omitted from meta-analysis, as were Taqman assays where continuous data were rounded to the nearest integer. Because DNA degradation influences assay efficiency (15), cohorts collected and stored under different conditions may differ in mean and/or variance, therefore studies examining FCGR3B CN as a continuous variable only were also excluded.

Figure 2.

Comparison of FCGR3B CN determined by SybrGreen and Taqman qPCR. FCGR3B CN was calculated for the same set of samples using SybrGreen and Taqman qPCR. Both assays gave median values close to 2 (2.09 for Taqman and 2.47 for SybrGreen) and there was some correlation between SybrGreen and Taqman measurements for individual samples (r = 0.68). Integer copy-number values assigned by cluster analysis on data obtained from Taqman qPCR were subsequently verified by PRT/REDVR (7) (Supplementary Data).

Eight studies fulfilled our inclusion criteria for meta-analysis (7–9,11,13,16–18). The control numbers totalled 7460 and case numbers totalled 5595, with cases including RA (37.9%), SLE (32.9%), UC (13.4%), AAV (10.1%), Kawasaki disease (2.8%), idiopathic thrombocytopenic purpura (2.1%) and anti-glomerular basement membrane antibody disease (0.8%). There was no significant heterogeneity between populations in any of the meta-analyses (Breslow-Day P-values ranged between 0.19 and 0.60) (Fig. 3). There was no significant association between high (>2) CN disease and disease (OR = 0.94 (0.82–1.08), P= 0.39), but low (<2) CN of FCGR3B was a significant risk factor [OR = 1.43 (1.27–1.62), P= 1.1 × 10⁻⁸] compared with CN = 2. Overall, low (<2) FCGR3B CN was highly associated with autoimmunity [OR = 1.44 (1.28–1.62), P_meta= 2.9 × 10⁻⁹] when compared with CN ≥ 2 (Fig. 3). There was strong evidence for association between low (<2) FCGR3B CN and SLE [OR = 1.59 (1.32–1.92), P_meta=9.11 × 10⁻⁷], but not for RA [OR = 1.36 (0.89–2.06), P= 0.15] (Supplementary Data). We also meta-analysed only those studies using PRT or MLPA assays; for CN< 2 compared with CN ≥ 2, OR = 1.33 (1.11–1.60), P= 0.002 under a random effects model (P_het= 0.01).

Figure 3.

Meta-analysis of association between FCGR3B deletion and autoimmune disease. Meta-analysis of the risk of carrying <2 [(A) OR = 1.43 (1.27–1.62), P= 1.1 × 10⁻⁸] and >2 [(B) OR = 0.94 (0.82–1.08), P= 0.39] copies of FCGR3B relative to the most common gene dosage (two copies). (C) Meta-analysis of the risk of carrying <2 copies compared with carrying ≥2 copies [OR = 1.44 (1.28–1.62) P= 2.9 × 10⁻⁹]. AAV, ANCA-associated vasculitis; a-GMB, anti-glomerular basement membrane disease; KD, Kawasaki disease; LN, lupus nephritis; RA, rheumatoid arthritis; SLE, systemic lupus erythematosus; UC, ulcerative colitis.

DISCUSSION

The results of this meta-analysis show that the FCGR3B deletion contributes to autoimmune disease at a P< 5 × 10⁻⁸ genome-wide level of significance (OR = 1.44, P= 2.9 × 10⁻⁹). The phenotypes meta-analysed were dominated by SLE and RA (∼70%) with association stronger in SLE (OR = 1.59) than in RA (OR = 1.36). However, it is notable that all four SLE studies and two of the three RA studies had consistent ORs for CN < 2 of 1.31–1.69 with one RA study reporting OR = 0.89 (11). Our analysis is robust as we excluded data generated using SyberGreen technology and any Taqman data generated by rounding to the nearest integer (3,5,10,12,19) or analysed only as a continuous variable (20). When the ‘gold-standard’ methods of PRT/MLPA (21) only were analysed, significant evidence was also obtained (OR = 1.33, P= 0.002). Despite some caveats discussed later, this meta-analysis suggests that low FCGR3B CN is a common risk factor for a number of non-organ-specific autoimmune diseases.

There are a number of possible explanations of how reduced FcγRIIIB levels could increase the risk of autoimmune disease. First, FcγRIIIB can bind intravascular immune complexes in the absence of any other inflammatory mediators but fails to generate reactive oxygen species, whereas FcγRIIA is responsible for ROS-mediated tissue damage (22). Therefore, the reduced adherence and phagocytic capabilities of polymorphonuclear neutrophils (PMN) in the case of low FCGR3B CN may delay clearance of immune complexes during episodes of transient inflammation and allowing the development of an inflammatory milieu where FcγRIIA activity predominates. Secondly, changes in FCGR3B CN may alter PMN function indirectly, by shifting the balance of activating receptors and potentially ‘unmasking’ the effects of functional polymorphisms in other Fcγ receptors. Furthermore, PMN are an important source of proinflammatory chemokines, secretion of which is partly mediated through FcγRIIA and IIIA (23). A third possibility is that decreased clearance results in the deposition of immune complex in tissues, a characteristic feature of SLE (24). This is supported by functional studies that suggest the ability of neutrophils to generate reactive oxygen species is also unaffected by low FCGR3B CN (6). Nor is there any evidence for association of low FCGR3B CN with either malaria or bacterial sepsis (9), implying cytotoxic capacity and/or ability to clear opsonized bacteria of PMN is unaffected. These results suggest that decreased capacity of neutrophils to clear circulating immune complex rather than alterations in cytotoxic activity is the etiological effect.

Given the complexity of the FCGR locus, and the heterogeneity of CNV observed (3), it is possible that the apparent effect of FCGR3B deletion is due to altered expression of other CN variable genes in linkage disequilibrium with FCGR3B CN [depending on the breakpoint, the deletion also encompasses FCGR3A as well as FCGR2C and HSP7 (3)] and may differ between populations (9). However, the correlation between low FCGR3B CN, reduced neutrophil expression and soluble Fcγ3B levels suggest that the primary effect is due to this receptor (6). Although variation in FCGR2C has been associated with idiopathic thrombocytopenic purpura, it is the presence of an open-reading frame in the usually silent gene rather than absolute CN that is thought to be the etiological variation (13), so CNV in FCGR3B is unlikely to be a proxy marker for a functional effect caused by FCGR2C.

Another point highlighted by this study is the diversity of measurement techniques used to measure FCGR3B CN, some of which have technical deficiencies. In the 12 different studies reporting FCGR3B CN as an integer variable, 7 different CN assignment techniques have been used; (i) qPCR using SybrGreen (5,12,19); Taqman assays with either (ii) arbitrary rounding of fractional values to the nearest whole number (5,10,19) (iii) post hoc assignment based on some form of frequency analysis (CNVtools) (7,12), ABCopyCaller (18) or (iv) ΔΔCT using samples with a median ΔCt as the calibrator (8); (v) PRT with restriction enzyme digest variant ratio determination (16,17,25); (vi) multiple probe PRT (9) and (vii) MLPA (3,11). This complicates the attempts to compare studies or draw conclusions about the etiological effects of gene dosage, and is also a problem common to studies into other CN variable genes (14,26). This could be resolved by making raw data available for other researchers to inspect and/or a consensus as to which method(s) are considered robust enough to measure and report CN variation, yet be sufficiently high-throughput to enable typing of thousands of samples. A technically simple solution would be the genotyping of genetic variants that tag CN variation; however, genotyping of CN at FCGR3B suggests that no tag SNPs exist in European populations (27), although this analysis would have been compromised because the probes used would not have been able to distinguish between FCGR3A and FCGR3B, with the results reflecting the combined CN of the two genes. A separate study using gold-standard methods of CN determination also showed that no suitable tag SNPs existed in the International HapMap Phase II data (25).

Although qPCR has been a popular method for measuring CN, any differences in amplification efficiency or experimental conditions that alter the ratio between the test and reference genes will change the apparent CN (21). SybrGreen requires that test and control genes be assayed separately, increasing the potential for this type of experimental error. It has previously been shown that data obtained by this method fail to cluster around integer values (14), nor is the CN determined by SybrGreen consistent with that obtained by PRT (Supplementary Data). The significant differences in FCGR3B distribution between studies using SybrGreen and those utilizing superior assay techniques further underlines the shortcomings of SybrGreen-based qPCR to measure CN. Because test and reference genes are measured in the same reaction in Taqman qPCR, this source of error is minimized; however, DNA quality also influences assay efficiency, and can introduce false positives and batch effects when comparing sample sets collected and stored under different conditions (15,28). In addition, CN values are (rarely) integer values, and arbitrary ‘binning’ or rounding of samples to whole numbers can introduce false associations if the distribution of two populations differs in mean and/or variance (28), so that analyses of qPCR data need to take such measurement errors into account (14). For these reasons, both SybrGreen and Taqman studies with CN assigned by arbitrary binning were excluded from meta-analysis, as well as those where only continuous data were analysed. Although MLPA gives close to integer values for CN, avoiding the clustering problems seen with qPCR, there is the potential for unidentified sequence variations within the targeted region to give false results, and the use of multiple probes is probably warranted, with validation by an alternative method when inconsistent results are obtained. The PRT/REDVR method has been well validated, and is probably the most dependable assay for FCGR3B CN to date. It also has the advantage of determining NA type as well as gene dosage, although it is less suitable for high-throughput analysis.

An additional complication in designing accurate assays for determining FCGR3B CN is the high sequence identity between this gene and FCGR3A, with 97% identity between the two genomic sequences in the GRCh37.p5 primary assembly. A number of different regions within FCGR3B have been targeted in quantitative PCR-based methods and MLPA (Supplementary Data), not all of which are specific for FCGR3B, and there is the potential for previously unidentified sequence variations to be a confounding factor. One paper reports different results depending upon which of three MLPA probes were used (11), and although sequencing revealed an insertion/deletion polymorphism within the binding site of one probe, the reason for differences between the second two probes was not identified, although there is an SNP in the binding site for one (rs34015117) that is, polymorphic in African and Asian sample sets (minor allele frequency 10 and 5%, respectively) but monomorphic in Europeans. Similarly, two SNPs at the 3′ end of the forward primer used in a previous European RA study (7) have been identified by the 1000 Genomes project (rs115802971 and rs115130696) in individuals of African descent where the minor allele corresponds to the FCGR3A sequence. Although an individual carrying these minor alleles would falsely be determined to have a deletion, the allele frequencies in European populations is very low, with the minor allele of these SNPs observed in only one of the first 380 Europeans sequenced in the 1000 Genomes project. In both cases (7,10), the assays were validated by a second technique, and gave a deletion frequency in controls that was consistent with that reported by other studies. The unusually high carriage of a single copy of FCGR3B in the Chinese population (18) is difficult to explain. The assay used in this study was developed by Applied Biosystems, and although the exact sequence of probe and primers is not available, it targets a region in intron 3 near/within the region of an MLPA probe used by Asano et al. (8), which gives a deletion frequency of 7% in a Japanese cohort, similar to the deletion frequency obtained in a European cohort by the probe used by McKinney et al. (7). Careful consideration of assay techniques and a need for validation of results will be important for future studies, including direct comparison of the different Taqman probes used.

In conclusion, meta-analysis of studies using the more robust CN measurement technologies and more rigorous analysis provides strong evidence supporting a role of FCGR3B deletion in the aetiology of non-organ-specific autoimmune diseases, particularly SLE.

MATERIALS AND METHODS

Identification of eligible studies and data extraction

We considered all studies examining the association between FCGR3B CN and autoimmune disease. A literature search for papers investigating CNV in FCGR3B was done on PubMed, Google Scholar and ISI Web of Knowledge using Medical Subject Heading (MeSH) terms and/or text words ‘FCGR’, ‘Fc gamma receptor’, ‘FCGR3B’, ‘Fcg receptor 3B’, ‘copy number’, ‘copy number variation’, ‘copy number polymorphism’, ‘CN’, ‘CNV’, ‘CNP’. Although there was no restriction placed on language, race or geographic area, analysis was restricted to data contained within a published peer-reviewed paper. Where information on assay primers and the numbers of samples in each CN category (<2, 2 and >2) was not presented, these data were requested from the corresponding author.

A study was included in the meta-analysis if it was (i) published up to August 2011, (ii) it was original data (iii) discrete CN data were available and (iv) it was obtained using a robust assay method [PRT/REDVR, MLPA or qTaqman PCR with CN assigned by probability analysis (using programmes such as ABI Copy Caller or CNVtools) and/or visual assessment of histograms combined with experimental validation of deletions by NA genotyping]. Results obtained by SybrGreen qPCR were excluded because test and control genes are assayed separately and thus more prone to experimental error and variability than other methods in which test and control genes are measured in the same reaction. Differences in DNA quality can also influence the apparent copy number as determined by qPCR, introducing potential batch effects that can be exacerbated by arbitrary rounding of non-integer values to whole numbers. For this reason, Taqman qPCR with CN assigned by rounding to the nearest integer were also excluded.

Meta-analysis

Data from the individual studies was combined and an overall odds ratio for carrying <2 copies of FCGR3B relative to ≥2, or <2 relative to 2, or >2 relative to 2 was calculated using Mantel–Haenszel methods under a random effects model. STATA 7.0 software was used for all statistical analysis. The data in Supplementary Data were generated by logistic regression analysis.

Genotyping of FCGR3B using SybrGreen

Ninety-six healthy New Zealand European samples were genotyped using the SybrGreen technology previously described (Fig. 2) (4). These are the same samples genotyped in McKinney et al. (7) by Taqman and included the 72, in which CN was verified by PRT. Samples were aliquoted by robot to minimize experimental error and every reaction carried out in duplicate. CN was calculated using the standard curve method using a reference sample with CN = 2 as determined by PRT. Because there was no clear clustering visible in SybrGreen data, CN was assigned by rounding to the nearest whole number. For Taqman data, clear clustering was visible around CN = 1 and CN = 2, and integer values assigned using CNVtools (28).

SUPPLEMENTARY MATERIAL

Supplementary Data.

NOTE ADDED IN PROOF

Very recently two studies which met our inclusion criteria have tested for association of FCGR3B CN with RA have been published (29,30). Inclusion of these studies provides significant evidence for a role of deletion in FCGR3B in RA [OR = 1.54 (1.22–1.94), P = 3.0 × 10⁻⁴], and strengthened evidence for association with autoimmunity [OR = 1.46 (1.31–1.64), P < 1 × 10⁻⁹].

FUNDING

This work was supported by Arthritis New Zealand; and the Health Research Council of New Zealand (grant number 08/075).

ACKNOWLEDGEMENTS

Data were shared by Stylianos Bournazos, Manuella Fanciulli, Yang Hau, Michiaki Kubo, Wanling Lau, David Morris, Heather Niederer, Ken Smith and Timothy Vyse. Their assistance and co-operation are greatly appreciated.

Conflict of Interest statement. None declared.