Introduction

Rheumatoid arthritis (RA) is a chronic systemic disorder with autoimmune traits in which polyarticular synovitis is particularly prominent. [1] Its origin remains elusive but there is evidence, that both genetic and environmental triggers are implicated in the pathogenesis and growing evidence suggests that RA consists of at least 2 different subsets characterized by the presence or absence of antibodies to citrullinated protein antigen (ACPA). [2] Newly published genome-wide association studies (GWAS) have demonstrated several DNA sequence variations associated with RA. [3] In a recent GWA metaanalysis it was estimated that current genetic discoveries account for around 16% of the disease variance, [4] although more than half of the liability to RA was considered to be genetic, [5] regardless of autoantibody status. [6] The most cited heritability estimate at around 60% is derived from 2 previous twin studies. The population based Finnish Study by Aho et al. was based on record linkage between the Finnish Twin Register and the Sickness Insurance Register but without diagnostic validation. [7] The study by McGregor et al. from the UK was based on volunteer RA subjects recruited from various sources, thus implying a risk of selection bias. [8] The probandwise and the pairwise concordance estimate in the Finnish study was 22.0% and 12.3% in monozygotic (MZ) twins and 6.7% and 3.5% in dizygotic same sexed (DZss) twins. The pairwise concordance estimate was 15.4% in MZ twins and 3.6% in same- and opposite sexed dizygotic (DZss/os) twins in the UK study.

In 2002 we published a probandwise concordance estimate at 0% in MZ and 8.8% in DZss/os twins suggesting a significant role for non-genetic effector mechanisms in the causation of RA. [9] This study was based on a mail survey in 1994 to unselected twin birth cohorts including the decades1921 to 1930 and 1953 to 1982. Meanwhile, The Danish Twin Register has been expanded to include also the 1931–1952 cohorts. [10] Our purpose was, based on this large and ethnically homogenous population of RA twins, to provide an optimized heritability estimate on RA and to study the recurrence risk of the disease.

Methods

Results

A total of 56,707 twin individuals were enrolled of which 45,280 responded yielding a response rate at 80%. In the 1994 and the 2002 surveys the response rates were 81.4% and 75% respectively. The responders represented 27,671 twin pairs. The response rate was higher among women (82%) than men (74%). Concordant responses were more frequent in MZ twin pairs (70%) than among DZ same sex pairs (60%) and DZ opposite sex pairs (54%). The final study base comprised 26,470 twin pairs of which 16,852 (63.7%) were intact pairs. Subsequently, when the 479 twins reporting RA were traced, 42 (9%) were non-responders to follow-up, had secret addresses and telephone numbers, 42 (9%) had died, 4 (8 ‰) had emigrated. Among the twins traced, 223 (47%) did not have RA based on telephone interview and/or medical records. Osteoarthritis (19%) was the most prevalent cause of exclusion. Among the remaining twins 97% accepted to participate in the clinical examination and a total of 173 (incl. 5 secondary cases) fulfilled the modified ARA 87 criteria [12] of which 11 had competing diagnoses and were excluded. Table 1 presents the characteristics of the remaining 162 RA twins. There were no significant differences between MZ and DZ twins in any of the listed variables.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic and clinical data on twins with rheumatoid arthritis according to zygosity.

https://doi.org/10.1371/journal.pone.0057304.t001

The box plot shows the distribution of discordance time in each zygosity group (Fig. 1). There was no difference in mean discordance time between any of the zygosity groups. The point prevalence of RA was 0.20% (95% CI 0.14% –0.27%) in men, 0.54% (95% CI 0.44% –0.64%) in women and 0.37% (95% CI 0.31% –0.43%) in men and women, age- and sex adjusted to the 2004 Danish population. The increased relative risk of attracting RA conditioned on having an affected cotwin compared to the background population risk was 24.6–35.4 in MZ twins and 17.3–31.6 in DZss twins.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Discordance time (years) in RA affected twin pairs according to zygosity.

MZ denotes monozygotic, DZss dizygotic same sexed, and DZos dizygotic opposite sexed. From the bottom up the smallest observation, lower, median, upper quartile, and the largest observation is shown. There is one outlier among the DZos pairs.

https://doi.org/10.1371/journal.pone.0057304.g001

Record Linkage

The record linkage identified 86 twins with RA of which 69 had also been identified in the 2002 survey, thereby reaching an ascertainment probability at 80.2%. Two additional possible DZ concordant pairs were identified and one twin from each of these pairs reported RA but did not want to participate in the clinical examination. One further MZ pair was concordant. This twin pair was not eligible in the 1994 survey, one twin had deceased before the 2002 survey and the other did not respond.

Record linkage between twins born 1920 to 1982 and the Danish Register of Causes of Death, identified 10 twins where RA was stated in the death certificates. According this registry none of these pairs were concordant for RA.

Table 2 presents the concordances rates by zygosity and ACPA status. Adjustment for sex had only minor effects on the estimates. There was no significant difference in any of the concordance rates between MZ and DZss or between DZss and DZos. The probandwise concordance estimate was 9.1% in MZ twins, 6.4% in DZ same-sexed twins and 2.2% in opposite-sexed twins. All the seven RA concordant twin pairs were also concordant for the shared epitope except one DZss pair. Restricting the analysis to ACPA positive RA cases tended to increase the risk in MZ twins more than in DZss twins.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Tetrachoric correlations and concordance rates of rheumatoid arthritis in Danish twins according to zygosity and ACPA status.

https://doi.org/10.1371/journal.pone.0057304.t002

Discussion

This study shows, that there is a considerable familial aggregation of RA as reflected by recurrence risks at 9,5–13,1 in MZ co-twins and at 6,4–11,7 in DZss co-twins as opposed to a background population risk at only 0,37%. Furthermore, since the majority of MZ twins are discordant for RA, potential triggers in the environment should be considered emphasized by the small differences in concordance and tetrachoric correlations between MZ and DZss twin pairs. In addition, our estimate of RA heritability only amounts to 12% while shared and non-shared environmental effects account for 50 and 38% respectively. By restricting the analysis to ACPA positive RA, we observed only a marginal increase in the difference of RA concordance and correlation between MZ and DZss.

Our estimates are somewhat at variance with the most cited figures on the genetic contribution to RA based on twin studies. [5], [7], [8] The UK study was based on recruitment of twins from rheumatologists and through nationwide multimedia campaigns. This implies that it was not possible to calculate prevalence or ascertainment probability. As pointed out previously, and confirmed in this study, there is a risk of over ascertainment of female twins, MZ twins and concordant pairs. [25], [26] The risk of selection bias of MZ concordant pairs in the UK study is emphasized by the fact that the number of concordant pairs was smaller than the total number expected to be concordant just by coincidence in the UK, particularly regarding DZ pairs. [9].

The Finnish population based study was a record linkage study of the Finnish Twin Cohort and the Sickness Insurance Register. The diagnosis relied on documents from the Insurance Register submitted by the attending physician, [7] and the twins were not approached and examined. Thus, no clinical characteristics were presented to compare the zygosity groups or the representativeness of RA twins and the diagnosis was not verified according to classification criteria. Besides, there was a risk of inflated concordance rates because the eligibility for free drugs could be proposed and approved on a less stringent basis, if, for example, the fact that the cotwin had RA was used as a criterion in support of the diagnosis.

The disparities between previously published recurrence risk ratios of RA among first degree relatives probably reflect differences in ascertainment probability, epidemiologic measures of occurrence, clinical characteristics and follow-up time. [27] By contrast, in population based studies like the present, it is in the order of 0.9 to 2.4 [28] and among siblings down to 1.1 [29] In a recent Swedish register linkage study the sibling recurrence risk was up to 4.6 whereas the twin recurrence risk in a combination of both MZ and same sexed DZ twins was 6.5 and thus in accordance with our estimate. [30] Taken together, these figures indicate that the recurrence risk in twins is higher than in ordinary siblings. By contrast, the recurrence risk in spouses is low, zero to 1.2, [29], [30] and therefore does not support that shared environment in late life is involved in RA. Since our data suggest that the majority of environmental impact is attributable to shared factors, it indicates that exposures in early life may contribute to RA development. Birth weight [31], smoking [32] and maternal MHC gene composition [33] are examples of risk factors in utero; breast-feeding in perinatal life [34] and exposure to pets are risk factors in the prepubertal period of life. [35] There is only week evidence to suggest an effect of early life infections. Infection during the first year of life only showed a borderline risk of RA in later life and Maternal infections during pregnancy were unrelated to the risk of RA. [36].

Recently, the impact of epigenetic changes on expression of silencing genes has attracted considerable attention. Thus, growing evidence from animal and human studies suggests, that stochastic epigenetic alterations - rather than the non-shared environment - may account for the discordance between monozygotic twins. [37]–[39] The potential involvement of stochastic events reduces the possibility to elucidate potential environmental triggers and will tend to diminish our estimate of unique environmental effectors.

There is recent evidence that ACPAs are more closely associated with the shared epitope than with RA itself [40] and that smoking, the presently most well documented environmental risk factor, is only associated with ACPA positive RA, [41], [42] emphasizing a strong gene-environment interaction in this disease subset. The present data, and our recent reports on ACPA in twins, suggest a higher genetic contribution in the ACPA positive vs. the ACPA negative RA subset and that there exist a genetic contribution to the production of ACPA beyond the shared epitope and the PTPN-22 polymorphism. [43], [44] A reevaluation of the UK twin data presented almost identical heritability estimates between ACPA positive and ACPA negative RA. [6].

Although we have involved a large cohort of twins in our surveys the estimate of heritability based on structural equation modeling is hampered by imprecision due to the binary phenotype of RA, the low prevalence and the low concordance rates. This has been demonstrated in simulation experiments testing the statistical power of the classical twin studies to resolve sources of familial resemblance of binary outcomes. [45] However, the relatively high recurrence risk among cotwins compared to background population risk and the close recurrence risks between MZ and DZss cotwins, indicate that shared environment is etiologically important. Based on the already established association between RA and its two major genetic risk factors, HLA and PTPN22 polymorphisms in Caucasians, [24] a genetic component should be considered in the model. Provided that the recurrence risk in DZ twins is higher than in ordinary siblings it seems that long term effects of early shared environment are important. This component has not been taken into account in previous heritability estimates on RA. [5], [6] Reducing a full ACE model to an AE model, thereby neglecting the effect of shared environment, will inevitably increase the additive genetic component as well as the precision of the remaining components. This may contribute to the previously reported much higher heritability estimates at 60%.

The twins participating in the current study constitute an unselected sample of birth cohorts enrolled in the Danish Twin Registry. It has previously been demonstrated that Danish twins have increased mortality during the first year of life due to prematurity, but that mortality and health trajectories in adulthood are similar to singletons. However, it is well documented that severe RA is associated with excess mortality [46]–[48] and that severity is associated with genetic susceptibility. [49]–[52] Thus, in this historical cohort study there might be a selection bias towards twins with milder disease and loss to follow up. But the clinical characteristics of our RA twins seem to be comparable to RA cases recruited from clinical settings with regard to extraarticular manifestations and auto-antibody profile and we have probably underestimated erosive disease since we did not require an updated X-ray status. The record linkage with the Danish Register of Causes of Death did not reveal any additional concordant pairs.

Generally, twin studies should report probandwise concordance. [18] Nonetheless, our estimate on the probandwise concordance rate is probably too low because all the secondary cases might have been registered as probands since they developed RA after the postal survey but before clinical examination. The casewise concordance rates in our study may therefore represent a more just estimate of the recurrence risk in twins.

There were 36.3% incomplete participating pairs who did not have the opportunity to be doubly ascertained. However, none of the RA affected twins in these pairs reported RA on behalf of the cotwin, and none of the cotwins were identified in the Danish National Patient Registry. As the proportion of intact pairs was higher for MZ than DZ twins, any inherent bias of this kind would lead to an overestimate of MZ concordance compared to DZ concordance.

The estimated ascertainment probability of RA among the responders was 80%. Since we only found one concordant MZ and two concordant DZ pairs among the RA non-reporters in the Danish National Patient Registry and none in the Danish national registry of deaths, there is no reason to believe that complete ascertainment would seriously distort the observed proportion of concordant pairs.

Cross sectional studies do not reflect the lifetime risk of RA for cotwins. However, this may not imply a significant source of bias, provided that the mean and distribution of discordance time do not differ between MZ and DZ twins. The mean discordance time in our study amounted to almost 20 years in all zygosity groups and with a comparable distribution between MZ and DZ twin pairs.

There is a higher response rate among MZ than DZ twins in volunteer based twin studies [25] and we have therefore paid major attention to acquire high and equal ascertainment probabilities between the zygosity groups. Thus, all twins invited to participate were offered a visit at home or at work. This option probably contributes to the 97% participation rate in the clinical examination. As our study shows MZ twins are more likely to be concordant for response than DZ twins, which may lead to selection bias towards overrepresentation of MZ concordant pairs. It is unlikely that the low concordance rates were due to observer bias as none of the healthy cotwins showed any signs of arthritis and disease classification was supported by medical records and blood samples. Finally, there is growing evidence for shared genetic loci in separate autoimmune diseases including RA, insulin dependent diabetes mellitus, celiac disease and inflammatory bowel disease. [53]–[56] Unfortunately, there is no validated information on other autoimmune diseases in our sample of RA affected twin pairs.

The study base consisted of unselected twin birth cohorts which enabled us to calculate an RA prevalence at 0.37%. The most recent estimate of the prevalence in Danish singletons was 0.35. [57] Both studies had an estimated ascertainment probability of 80%, used the same classification criteria and both populations consisted entirely of Danish Caucasians. Furthermore, both estimates were age and sex adjusted to the 1994 Danish population. These almost identical prevalence estimates in twins and singletons based on homogenous populations of Caucasian origin support the external validity of our study.

In conclusion this study emphasizes that family factors are important for the development of RA. Although genetic effectors are important, shared and non-shared environmental triggers and/or epigenetic stochastic events seem to be even more significant. However, it should be borne in mind, that the genetic and non-genetic components may not be the same across disease subsets.