Hum Genet (1998) 103:475–482
© Springer-Verlag 1998
ORIGINAL INVESTIGATION
Jing-Ping Lin · Joseph M. Cash · Sharon Z. Doyle Sandra Peden · Keith Kanik · Chris I. Amos Sherri J. Bale · Ronald L. Wilder
Familial clustering of rheumatoid arthritis with other autoimmune diseases Received: 12 January 1998 / Accepted: 10 June 1998
Abstract Previous studies have shown that rheumatoid arthritis aggregates within families. However, no formal genetic analysis of rheumatoid arthritis in pedigrees together with other autoimmune diseases has been reported. We hypothesized that there are genetic factors in common in rheumatoid arthritis and other autoimmune diseases. Results of odds-ratio regression and complex segregation analysis in a sample of 43 Caucasian pedigrees ascertained through a rheumatoid arthritis proband or matched control proband, revealed a very strong genetic influence on the occurrence of both rheumatoid arthritis and other autoimmune diseases. In an analysis of rheumatoid arthritis alone, only one inter-class measure, parent–sibling, resulted in positive evidence of aggregation. However, three inter-class measures (parent–sibling, sibling–offspring, and parent–offspring pairs) showed significant evidence of familial aggregation with odds-ratio regression analysis of rheumatoid arthritis together with all other autoimmune diseases. Segregation analysis of rheumatoid arthritis alone revealed that the mixed model, including both polygenic and major gene components, was the most parsimonious. Similarly, segregation analysis of rheumatoid arthritis together with other autoimmune diseases revealed that a mixed model fitted the data significantly better than either major gene or polygenic models. These results were consistent with a previous
J.-P. Lin · S.Z. Doyle · S. Peden · S.J. Bale (✉) Genetic Studies Section, Laboratory of Skin Biology, Bldg 6, Rm. 429, 6 Center Drive MSC 2757, Bethesda, MD 20892-2757, USA Tel.: +1-301-4022679, Fax: +1-301-4022724 J.M. Cash Department of Rheumatic and Immunologic Diseases, The Cleveland Clinic Foundation, Cleveland, OH 44195, USA K. Kanik · R.L. Wilder Inflammatory Joint Diseases Section, Arthritis and Rheumatism Branch, National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Institutes of Health, Bethesda, MD 20892, USA C.I. Amos M.D. Anderson Cancer Center, Houston, TX 77030, USA
study which concluded that several genes, including one with a major effect, is responsible for rheumatoid arthritis in families. Our data showed that this conclusion also held when the phenotype was defined as rheumatoid arthritis and/or other autoimmune diseases, suggesting that several major autoimmune diseases result from pleiotropic effects of a single major gene on a polygenic background.
Introduction Rheumatoid arthritis (RA) is a chronic inflammatory disease typically affecting, symmetrically, the small joints of hands and feet. The prevalence of RA is about 1% in the Caucasian population, and the female to male ratio is 3:1 (Hochberg and Spector 1990). Numerous studies demonstrate that genetic factors have an important influence on an individual’s susceptibility to develop RA. Familial clustering of cases is relatively common, particularly when the proband is severely affected (Deighton and Walker 1991). Twin studies also support the hypothesis of genetic involvement in RA. Monozygotic twins have a significantly higher risk of developing disease than the dizygotic co-twin of an affected proband (Lawrence 1970). One segregation analysis of RA in families suggested that a highly penetrant, recessive major gene is the most parsimonious explanation of genetic risk (Lynn et al. 1995). In another study, a dominant susceptibility gene with incomplete penetrance is implicated (Yamashita et al. 1986). However, using a risk ratio theory developed by Risch (1990), Rigby et al. (1993) supported at least a two-locus model of inheritance for RA. Yet another investigation proposed that RA is due to polygenic inheritance (Pritchard 1994). Laboratory investigations of RA genetic susceptibility have focused mainly on the Class II MHC genes. Sibs with RA show significant deviation from random sharing at the HLA locus (Rigby 1992). Many studies suggest that at least some of the genetic risk for RA can be explained on the basis of a shared epitope of HLA DR molecules (Gregerson et al. 1987). Significant associations have been well docu-
476
mented between RA occurrence and HLA-DRB1 DR4 alleles (*0401, *0404, and *0408), DR1 alleles (*0101 and *0102), a DR10 allele (*1001), and a DR6 allele (*1402). All of these alleles have a similar or identical amino acid sequence in position 70–74 of the third hypervariable region of the DRB1 molecule, and this forms the basis of the shared epitope hypothesis. HLA-DRB1 genes also appear to influence disease severity in RA (Weyand et al. 1992). Some investigations support a dose effect of MHC influence on RA, with homozygosity for the shared epitope not only conferring a higher risk of developing RA, but also of suffering more aggressive joint disease. Patients with both DR4 and DR1 had a significantly higher frequency of extraarticular features compared with those with only one of these HLA alleles (Agrawal et al. 1995). In contrast, another study proposed that HLA RA-related subtypes were associated with susceptibility only, and not with disease severity (Suarez-Almazor et al. 1995). It is clear that the shared epitope theory does not fully explain the heritability of RA. A significant minority of RA patients develop typical disease without inheriting even one copy of the shared epitope. The prevalence of RA is approximately 1%, while the prevalence of persons with the shared epitope is about 50%. This means that only 2% of individuals with the shared epitope have RA. This is well below the identical twin concordance rate of 12–32% (Deighton and Walker 1991) and suggests the involvement of other susceptibility genes. The unified-shared-epitope hypothesis of RA susceptibility was rejected by using the marker association-segregation χ2 method (Dizier et al. 1993). Further, likelihood analysis demonstrated that an HLA-linked RA susceptibility locus accounts for approximately 50% of the liability to familial RA, implying 40% of RA cases (Hasstedt et al. 1994) are due to HLA-linked susceptibility loci. Mathematical modeling has also suggested that non-MHC genes may have a major role in the susceptibility of RA (Pritchard 1994). Grennan and Sanders (1988) calculated that the relevance of HLA status to developing RA is possibly only 20% of the total genetic influence. All these results strongly suggest that the genetic influence on development of RA is complex. Although RA shows evidence of familial aggregation, in one study only one-third of patients with RA were aware of at least one relative with the same disease (Pritchard 1994). Familial aggregation analysis demonstrated that a low proportion of RA proband families has significant excess risk of RA (Kwoh et al. 1996). However, there are several studies showing increased aggregation of other autoimmune diseases (OAID) and the presence of autoantibodies in the relatives of RA probands (Mulhern et al. 1966; Sels et al. 1997; Taneja et al. 1993; Teitsson et al. 1985; Thomas et al. 1983; Walker et al. 1986). The incidence of autoimmune thyroid disease (ATD), insulin-dependent diabetes mellitus (IDDM) and systemic lupus erythematosus (SLE) has been reported to be significantly higher than expected in the family members of RA probands (Taneja et al. 1993). These observations suggest that both RA and OAID may have a common, and genetically influenced, etiology.
To test the hypothesis that RA and OAID susceptibilities are under the influence of some common genetic component, we obtained family histories from 29 RA probands and 14 matched control probands. Family members were then contacted to obtain detailed medical histories and confirm diagnoses of RA or OAID. Mathematical techniques including odds-ratio regression and segregation analysis were applied to the data. We analyzed the results in two ways; the first considering only individuals in a family with RA as being affected, and then again considering as affected any individual with either RA or OAID.
Subjects and methods Study population We selected 30 RA case probands consecutively from a registry of patients enrolled in ongoing studies at the National Institutes of Health (NIH) designed to study the pathogenesis or therapy of RA. Family history of RA or OAID was not a criterion for entry into any protocol. Twenty-nine of the cases selected agreed to participate in this family study. All fulfilled the 1987 American College of Rheumatology (Arnett et al. 1988) criteria for diagnosis of RA, were seropositive for rheumatoid factor, and could be classified as severely affected. Each proband was asked to identify a friend of the same gender whose age was within 5 years of the proband’s age and who did not have RA to serve as a control proband (case-control design). Fifteen RA probands either did not identify a control (n=1) or the control refused to participate (n=14), reducing the total number of control probands to 14. All 14 participating control probands were confirmed not to have RA by a rheumatologist using standard criteria for evaluation. After providing informed consent, RA cases and control probands, their spouses, and first-degree relatives were interviewed by a research nurse (S.Z.D., S.P.) or a rheumatology fellow (J.M.C., K.K.). Interviews were conducted either in person or by telephone. Each individual completed a comprehensive 24-page questionnaire designed to capture all relevant information on demography, family history, previous diagnoses, medical history, laboratory, and other diagnostic data. All positive histories of possible autoimmune diseases including ATD, IDDM, ankylosing spondylitis (AS), myasthenia gravis (MG), rheumatic fever (RF), Sjogren’s syndrome, SLE, and multiple sclerosis were confirmed by direct contact with the individual’s personal physician and/or review of medical records. All questionnaire and medical information was coded and reviewed by a board-certified rheumatologist who was blind to the case/control status as well as the family history of the subject. Diagnostic categories were assigned in accord with standard medical criteria. Probands or family members who could neither be contacted directly nor interviewed by phone were excluded from analysis, unless at least two other family members reported the same affection status for them. Statistical methods Familial aggregation evaluated using odds-ratio regression models We used the logistic odds-ratio regression method developed by Liang and Beaty (1991) to evaluate familial aggregation of RA only and RA and/or OAID in our sample. Let n be the size of a single family and k the total number of families studied. Let Yj denote the binary outcome of the jth individual and we define log
Pr(Yj = 1) = β 0 + β1x1 j + ... + β p x pj = β t x Pr(Yj = 0)
(1)
477 Table 1 The age and gender distribution of rheumatoid arthritis (RA) in the first-degree relatives of RA proband and control proband pedigrees Age (years)
RA proband pedigree Male
0–19 20–39 40–59 60– Total
Control proband pedigree Female
Male
n
RA
n
RA
n
RA
n
RA
5 29 36 39 109
0 2 5 5 12
6 37 23 43 109
0 10 2 22 34
3 18 10 16 47
0 1 0 0 1
1 17 12 21 51
0 0 0 1 1
where xj is a set of p covariates for the jth individual and β is a set of corresponding parameters. The covariates are individual-specific effects and included age, gender, and the probands’ status in our study. In addition, this modeling scheme also permits allowance for the non-independence of family members by incorporating a familial odds ratio. The measure of familial aggregation is the odds ratio between the jth and kth members of the family: Pr(Yj = 1, Yk = 1) Pr(Yj = 0, Yk = 0) OR jk = (2) Pr(Yj = 1, Yk = 0) Pr(Yj = 0, Yk = 1) If the odds ratio is equal to 1.0, there is no association between individuals. An odds ratio of 1.0 for each pair of relatives indicates that there is no familial or genetic contribution to liability for the disease. An odds ratio significantly greater than 1.0 indicates familial aggregation of the disease. The familial odds ratio can be made class dependent by expressing the log odds ratio as a function of the class. log ORjk = γz
Female
25–34, 35–44, 45–54, 55–64, 65–74, >74 years (Linos et al. 1980). Pairwise comparison of models was achieved using likelihood ratio tests. Ascertainment correction was made by dividing the likelihood of each pedigree by the likelihood of the proband (Thompson and Cannings 1979). Asymptotically, minus twice the difference between ln likelihoods in two models is distributed χ2 with degrees of freedom equal to the difference in the number of parameters between the models. Akaike’s information criterion (AIC) (Akaike 1974) was also used to compare models. AIC is defined as minus twice the ln likelihood of the model plus twice the number of free parameters in the model. If the difference between any two AIC measurements for two models is greater than 1 or 2, that difference is considered to be significant (Sakamoto et al. 1986). A model with a small value of AIC is deemed to provide a better fit to the data relative to a model with a large AIC value. The model with the lowest AIC is considered to be the most parsimonious. In cases where two models are not hierarchical, the 02 test can not be used, although the AIC is still valid.
(3)
where z indicates family-specific class variables and reflects the relationship between each pair of relatives. Thus, for families including the proband’s siblings, parents, and offspring, we may estimate the following log ORjk: γss, γpp, γoo, γsp, γso, and γpo depending on whether the (j,k) pair are siblings, parents, offspring, sibling–parent, sibling–offspring, or parent–offspring pairs with respect to each family member’s relationship to the proband. The parameter estimation was carried out using a statistical program called EGEE (extended generalized estimating equations; Qaqish and Liang 1992). We used two different models to calculate the odds ratio: Model 1. OR(eγ) is the same for any pair of relatives in a family. Even though the relationship between a parent–parent pair is not genetically similar to other pairs, the model assumes all pairs have identical odds ratios. There are five parameters in this model: the intercept, three covariates, and a common γ (log OR) between all pairs of family members, regardless of the relationship to the proband. Model 2. The odds ratio is dependent on the relationship between family members. There are ten parameters in this model. In addition to the intercept and three covariates in Model 1, there are six log OR between family members: γss, γpp, γoo, γsp, γso, and γpo. Three covariates are considered in each model: age, sex, a dummy variable for the prostatus, and the intercept. Each model was investigated for the two different outcomes: RA only and RA and/or OAID. Segregation analysis Segregation analysis was carried out on the RA and control probands and their relatives using PAP (Hasstedt 1994) and the maxima obtained with GEMINI (Lalouel 1979). The parameters of the model include the major gene allele frequency (q), dominance which equals (µ2–µ1)/(µ3–µ1), where µI is the mean of genotype I, displacement which equals (µ3–µ1)/s where s represents the withingenotype standard deviation, and heritability, which represents the proportion of the total phenotypic variance due to the polygenic effect (polygenic variance)/(within-genotype variance). Sixteen liability classes were defined according to age- and gender-specific RA prevalence in the general population (eight age groups) 0–14, 15–24,
Results There were 29 RA proband pedigrees and 14 control proband pedigrees with 218 and 98 first-degree relatives, respectively. There was no significant difference in age distribution between the first-degree relatives of RA and control probands (P=0.84). Among the 218 first-degree relatives of RA probands, 46 had RA, whereas there were only 2 RA cases among the 98 first-degree relatives of control pedigrees. The age and gender distribution of RA in the first-degree relatives (not including probands) is shown in Table 1. Among the 218 first-degree relatives in RA proband pedigrees, 29 had OAID. There were 16 with ATD, 5 IDDM, 5 with history of RF, 1 with AS, 1 with MG, and 1 with both ATD and IDDM. These data are shown in Table 2. Among the 29 who had OAID, there were 8 individuals with both RA and ATD, all of whom were females over 59 years of age. The total number of individuals with RA and/or OAID was 67. Among the 98 control relatives, there were 2 with ATD and 2 with IDDM. One of the individuals with IDDM also had RA. We compared self-reported disease status to that reported by other family members. Of 96 reports by individuals of no RA in a family member, 81(84%) were accurate. Of 113 reports by a family member that a relative had RA, 112 reports were accurate. Therefore, nearly all inaccuracies involved under-reporting (i.e., false negative) rather than false-positive reports.
478 Table 2 The age and gender distribution of other autoimmune diseases (OAID) in the first-degree relatives of RA proband pedigrees (ATD autoimmune thyroid disease, IDDM insulin-dependent diabetes mellitus, RF rheumatic fever, AS ankylosing spondylitis, MG myasthenia gravis) Age (years)
0–19 20–39 40–59 60– Total
Male
Female
n
ATD
IDDM
RF
AS
n
ATD
IDDM
RF
MG
5 29 36 39 109
0 0 0 1 1
1 1 0 0 2
0 0 3 1 4
0 0 0 1 1
6 37 23 43 109
0 2 2 12 16
0 2 0 2 4
0 0 1 0 1
0 0 1 0 1
Table 3 Frequency of RA only and RA and/or OAID in different relative classes of RA probands and control probands
Disease frequency (%)
RA proband pedigree Parents Siblings Offspring Total Control proband pedigree Parents Siblings Offspring Total
Familial aggregation studies Table 3 shows the number of each class of first-degree relatives and the frequency of RA only, and RA and/or OAID in each group. The logistic regression model incorporated three individual effect variables: whether or not the individual was from a RA proband family, age, and gender. Each of these variables showed a significant effect on the occurrence of RA and OAID. When taking RA only into account, the probability of an individual from the RA proband pedigree being affected was 15.5 times (95% CI of OR=2.0–121.5) that of an individual from the control proband pedigree, after adjustment for age and gender. Females were at 3.3 times (95% CI of OR=1.4–7.7) higher risk of being affected than males. The probability of being affected increased with increasing age, with the odds ratio increasing 0.02 per year. When taking RA and/or OAID into account, the probability of being affected was 11.4 times (95% CI of OR=2.8–46.5) higher in relatives of RA probands than in control probands. The gender effect was 3.4 times (95% CI of OR=1.7–6.6) higher in females than males and the age effect showed 0.02 increased risk per year. These results are shown in Table 4. Two different models of familial aggregation for the two outcomes were fitted separately, and the results are presented in Table 4. In Model 1, a common risk for RA or RA/OAID was assumed, while in Model 2, the risk was allowed to be dependent on the class of relative. When RA only was evaluated, there was no significant evidence of fa-
n
RA only
RA and/or OAID
58 80 78 216
14(24.1) 21(26.3) 11(14.1) 46(21.3)
22(37.9) 28(35.0) 17(21.8) 67(31.0)
28 30 40 98
0(0) 1(3.3) 1(2.5) 2(2.0)
0(0) 4(13.3) 1(2.5) 5(5.1)
milial aggregation under Model 1. In Model 2, one interclass measure (ORsp) showed significant familial aggregation. When RA and/or OAID were taken into account, in Model 1 there was significant evidence of familial aggregation. In Model 2, three inter-class measures (Orsp,ORso,and ORpo) showed significant evidence of familial aggregation. Segregation analysis Table 5 presents the parameter estimates and likelihoods for a hierarchy of models using segregation analysis of RA only. When the sporadic model was compared with the polygenic model, the sporadic model was strongly rejected (χ2=199.8, df=1, P<0.001). A model with a major gene effect fited the data better than the polygenic model (AIC=218.2 for the major gene model versus AIC=225.4 for the polygenic model). A mixed model with a major gene and polygenic effect yielded an improvement in the likelihood of the data, though with the lowest AIC (216.9) of all models tested, it was not significantly different from the major gene model (χ2=3.3, df=1, P=0.069). The mixed model was, however, the most parsimonious. Under this model, the estimated major gene allele frequency was 0.027. The dominance was 0.9 and 1.0 for males and females, respectively, and displacement was 16.3 and 3.0. According to these parameters, the affection probabilities by genotype, gender, and age group were calculated and are listed in Table 5. The estimate of heritability was 1. Table 6 presents the results of segregation analysis for RA and/or OAID. Initially, a sporadic model was fitted to
479 Table 4 Odds ratio (OR) and logistic regression adjustment for RA only and RA and/or OAID among 314 relatives of 29 RA and 14 control probands
Explanatory variable RA only Proband status Sex Age ORss ORpp ORoo ORsp ORso ORpo RA and/or OAID Proband status Sex Age ORss ORpp ORoo ORsp ORso ORpo a
Model 1: common risk
Model 2: different risk
15.5(2.0–121.5) 3.3(1.4–7.7) 1.02(1.02–1.04)
15.6(1.9–129.0)a 3.9(1.6–9.5)a 1.03(1.01–1.05)a 2.6(0.8–8.5) 1.0(0.4–2.5) 1.7(0.6–5.0) 4.5(2.1–9.8)a 1.4(0.5–4.1) 1.0(0.4–2.5)
1.8(0.7–4.4)
11.9(3.1–46.5)a 3.7(1.8–7.7)a 1.03(1.01–1.05)a 1.4(0.6–3.3) 1.3(0.2–6.5) 1.6(0.4–5.6) 3.3(1.6–6.6)a 2.2(1.03–4.6)a 2.2(1.03–4.6)a
11.4(2.8–46.5) 3.4(1.7–6.6) 1.02(1–1.04)
2.2(1.1–4.3)a
Significant OR
Table 5 Complex segregation analysis of RA only (AIC Akaike’s information criterion, NP number of parameters) Parameter
Sporadic
Polygene
Majorgene
Mixed
Allele frequency
–
–
0.012
0.027
Dominance Female Male
– –
– –
0.85 0.57
1.00 0.90
Displacement Female Male
– –
– –
12.80 4.60
3.03 16.32
h2
–
1.0
–
1.0
–2lnL
423.2
223.4
208.2
204.9
NP
0
1
5
6
AIC (–2lnL+2NP)
423.2
225.4
218.2
216.9
the data. A model with polygenic effect fitted the data significantly better than the sporadic model (χ2=322.1, df=1, P<0.001). A model with a major gene effect yielded a marked improvement in the likelihood of the data (AIC=329.8 for the major gene model compared to AIC=339.0 for the polygenic model). The major gene model was, however, rejected in favor of a mixed model that in-
Affection probability by genotype in mixed model Females Age 0– 15– 25– 35– 45– 55– 65– 75–
AA 0.00000 0.00000 0.00000 0.00002 0.00016 0.00069 0.00304 0.00304
Aa 0.01844 0.04609 0.07924 0.14717 0.28313 0.43072 0.61103 0.61103
aa 0.01844 0.04609 0.07924 0.14717 0.28313 0.43072 0.61103 0.61103
Males Age 0– 15– 25– 35– 45– 55– 65– 75–
AA 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000 0.00000
Aa 0.00094 0.03176 0.03176 0.05823 0.13064 0.20410 0.33365 0.33365
aa 0.06590 0.39981 0.39981 0.51270 0.68377 0.78070 0.87936 0.87936
cludes a polygenic effect (χ2=6.6, df=1, P=0.01), and the AIC for this model was the lowest (323.4). Therefore, the mixed model was the most parsimonious model for RA and/or OAID. In the mixed model, the major gene allele frequency was estimated to be 0.011. The dominance estimates were 0.31 and 0.90 for males and females, respectively, and displacement was 6.1 and 13.0. The affection
480 Table 6 Complex segregation analysis of RA and/or OAID Parameter
Sporadic
Polygene
Allele frequency –
Majorgene
Mixed
–0.012
0.011
Dominance Female Male
– –
– –
0.78 0.49
0.90 0.31
Displacement Female Male h2
– – –
– – 1.0
13.00 6.23 –
13.0 6.1 1.0
–2lnL
659.1
337.0
319.8
311.4
NP
0
1
5
6
AIC (-2lnL+2NP)
659.1
339.0
329.8
323.4
probabilities by genotype, gender, and age group from the mixed model are listed in Table 6. Heritability was again estimated at 1.0.
Discussion Results of numerous investigations into RA have identified the involvement of a genetic component in the etiology of the disease. Several models of inheritance have been postulated including single locus autosomal recessive, autosomal dominant, two locus, and polygenic inheritance (Lynn et al. 1995; Pritchard 1994; Rigby et al. 1993; Yamashita et al. 1986). Positive associations between RA genetic susceptibility and Class II MHC genes have been observed, suggesting that at least some of the genetic risk for RA can be explained by mutations at the HLA locus (Weyand et al. 1992). However, linkage studies demonstrated that the inferred HLA-linked RA susceptibility locus accounts for only about 50% of familial RA, and thus, only one-fifth of RA in the population (Hasstedt et al. 1994). All of the above indicate the etiological complexity of this disorder. Although RA shows evidence of familial aggregation, two-thirds of RA cases do not report a positive family history (Pritchard 1994). However, several studies have noted evidence for increased susceptibility to OAID in the relatives of RA probands (Bias et al. 1986; Lippman et al. 1982; Mulhern et al. 1966; Taneja et al. 1993; Teitsson et al. 1985; Thomas et al. 1983; Walker et al. 1986). HLA locus associations identified in RA have been shown in SLE, ATD, IDDM, and multiple sclerosis as well (Taneja et al. 1993; Thomson 1995). The association with HLA-DR4 is certain-
Affection probability by genotype in mixed model Females Age 0– 15– 25– 35– 45– 55– 65– 75– Males Age 0– 15– 25– 35– 45– 55– 65– 75–
AA 0.00000 0.00000 0.00000 0.00000 0.00000 0.00128 0.01356 0.01356
Aa 0.04229 0.10752 0.18632 0.34897 0.67973 1.00000 1.00000 1.00000
aa 0.33210 0.52025 0.65508 0.81660 0.69059 1.00000 1.00000 1.00000
AA 0.00000 0.00034 0.00034 0.00089 0.00296 0.00557 0.01079 0.01079
Aa 0.00018 0.06780 0.06780 0.11060 0.19775 0.26221 0.34613 0.34613
aa 0.73222 0.99643 0.99643 0.99846 0.99957 0.99980 0.99992 0.99992
ly not complete, however, with half of the relatives carrying the DR4 allele in multiple-case RA families remaining unaffected (Taneja et al. 1993). A very recent study, comparing the linkage results from 21 previously published genome-wide scans and searching for loci contributing to susceptibility to autoimmune disorders, indicated that several clinically distinct autoimmune diseases may be controlled by a common set of susceptibility loci (Becker et al. 1998). Moreover, genetic mapping in animal models also suggests that genetic risk factors are shared among several autoimmune diseases (Remmers et al. 1996). These results suggest that both RA and OAID may have a common, and genetically influenced etiology, which may include the MHC as well as other inherited factors. One possible explanation for this phenomenon could be the inheritance of a common gene in the MHC predisposing to autoimmunity, together with other genetic and environmental factors. In the present study, 7.8% and 2.8% of the first-degree relatives of RA probands had ATD and IDDM, respectively. This is significantly higher than the 1% and 0.35% prevalence in the general population (Myers 1994; Tunbridge et al. 1977). There have been other papers reporting familial aggregation of autoimmune diseases in multiple relatives of RA probands. However, no study to date has measured the extent of the familial aggregation or performed segregation analysis to identify the pattern of inheritance. This is the first investigation that uses odds-ratio regression analysis of familial aggregation of autoimmune diseases in the relatives of RA probands and the first to use complex segregation analysis to test the genetic models. To compare our results with previous studies, we also analyzed RA only in our families using the same methods.
481
Results from the odds-ratio regression analysis, taking RA only into account, showed that the prevalence of RA in first-degree relatives of RA probands was 15.5 times as high as that of the control proband’s relatives, whereas for RA and/or OAID, the prevalence in relatives was increased 11.4 times. The conclusions reached were statistically significant. For RA only, under the assumption of common risk and a constant odds ratio over all pairs of first-degree relatives, there was no significant evidence of familial aggregation. However, when assuming different odds ratios depending on different familial relationship classes, parent–sibling pairs showed significant evidence of familial aggregation. For RA and/or OAID, there was significant evidence of familial aggregation under the common risk and constant odds ratio assumption. Three inter-class measures (parent–sibling, sibling–offspring, and parent–offspring pairs) showed significant familial aggregation under the assumption of different odds ratios, dependent on different familial relationship classes. Individual effects of the proband’s status was stronger in RA only (15.5) than RA and/or OAID (11.4), whereas the pairwise effect in the odds-ratio regression model was opposite. The reason for this result is that only a small increase in the number of cases of OAID in the control proband’s relatives would change the individual effect dramatically, as the relationship of relatives to the proband is not taken into account in the logistic model. However, the pairwise odds-ratio measurement, which takes the relationship of the relative to the proband into account, is more robust and a more accurate measurement of familial aggregation. Three inter-class measures, but no intra-class measure, showed significant familial aggregation for RA and/or OAID. This indicates that a strong environmental effect could be ruled out. In the segregation analysis, when taking RA only into account, the most parsimonious model was the mixed model. It was not significantly better than the major gene model. As the many previously published studies indicate, it is unlikely that a single gene could account for all cases of RA. For example, the monozygotic twin concordance rate for RA was reported as only 15.4% only (Deighton and Walker 1991). In addition, although there appears to be a strong association between RA susceptibility and HLADR4, 50% of RA patients do not carry the HLA risk allele (Pritchard 1994). In our study, when both RA and OAID were analyzed together, the most parsimonious model was a mixed model with both a major gene and polygenes having an effect on the phenotype. It may be hypothesized that this major gene effect is from the MHC and, together with small polygenic effects, the familiality of many autoimmune diseases, including RA, can be explained. An excess of females to males (3:1) affected with RA has long been noticed, as well as for many OAID, including thyroid disease and SLE. Our study showed that the occurrence of substantially more RA (and OAID) cases in females than in males may be attributed to a major gene susceptibility locus. This may result from the greater dominance effect found in women (Tables 5, 6). Further studies using a larger sample size will help to better explain these findings.
Although attempts were made to avoid inaccuracies in all aspects of the study, some limitations remained. For assigning diagnoses, we did not have a strong face-to-face history with physical and laboratory examinations; some of the reports were not from personal interviews, but made by other relatives. However, since we used the same approach in both RA and control pedigrees, the bias should be minimized. The results of comparing self and relative reports in RA/OAID disease status indicated that using relative reports was very conservative, and the chance of error was very low (0.007) when at least two relatives reported an individual to be affected. Due to the relatively small number of families in the study, several other models of interest, including the general mixed model, could not be tested in the segregation analysis. The estimates of the heritability under both polygenic and mixed models in both RA only and RA/OAID as outcome were at the upper bound (0–1), and thus were most likely to be overestimated. A sample with more families would allow more accurate parameter estimates. Even in this small series, however, significant differences between RA and control proband families could still be detected. This indicates the sample had enough power to detect differences, although parameter estimation may not be highly precise. The lack of a blood collection component to the study, and hence the unavailability of MHC typing information, prevents us from drawing any conclusion about the specific role of that locus in susceptibility to autoimmune disease. However, the strong suggestions from previous studies allow us to hypothesize that the MHC may be one of the important loci contributing to RA and/or OAID, which our data support. The results of this study, taken together, suggest that it may be more accurate to consider RA and OAID together as a single, pleiotropic outcome of a major gene with additional polygenic effects. Using this phenotype definition in future, genetic linkage studies may provide more power to detect the involvement of the major gene component of autoimmune disease aggregating in families. Acknowledgements We thank Dr. Sandra Hasstedt for providing PAP software and for advice in doing the segregation analyses, and Drs Kung-Yee Liang and Terri Beaty for providing the GEE software. This work supported in part by grant AR44422.
References Agrawal S, Aggarwal A, Dabadghao S, Naik S, Misra R (1995) Compound heterozygosity of HLA-DR4 and DR1 antigens in Asian Indians increases the risk of extra-articular features in rheumatoid arthritis. Br J Rheumatol 34:41–44 Akaike H (1974) A new look at the statistical model identification. IEEE Trend Automatic Control 19:716–723 Arnett FC, Edworthy SM, Bloch DA, McShane DJ, Fries JF, Cooper NS, Healey LA, Kaplan SR, Liang MH, Luthra HS, Medsger TA Jr, Mitchell DM, Neustadt DH, Pinals RS, Schaller JG, Sharp JT, Wilder RL, Hunder GG (1988) The American Rheumatism Association 1987 revised criteria for the classification of rheumatoid arthritis. Arthritis Rheum 31:315–324 Becker KG, Simon RM, Bailey-Wilson JE, Biddison WE, McFarland HF, Trent JM (1998) Clustering of non-MHC susceptibility candi-
482 date loci in human autoimmune disease. Proc Natl Acad Sci USA 95:9979–9984 Bias WB, Reveille JD, Beaty TH, Meyers DA, Arnett FC (1986) Evidence that autoimmunity in man is a Mendelian dominant trait. Am J Hum Genet 39:584–602 Deighton CM, Walker DJ (1991) The familial nature of rheumatoid arthritis. Ann Rheum Dis 50:62–65 Dizier MH, Eliaou JF, Babron MC, Combe B, Sany J, Clot J, Clerget-Darpoux F (1993) Investigation of HLA component involved in rheumatoid arthritis (RA) by using the marker association–segregation χ2 (MASC) method: rejection of the unifyingshared-epitope hypothesis. Am J Hum Genet 53:715–721 Gregerson PK, Silver J, Winchester RJ (1987) The shared epitope hypothesis. An approach to understanding the molecular genetics of susceptibility to rheumatoid arthritis. Arthritis Rheum 30:1205–1213 Grennan DM, Sanders PA (1988) Rheumatoid arthritis. In: Grennan DM (ed) Clinical rheumatology, vol 2(3). Genetics of rheumatic diseases. Bailliere Tindall, London, pp 585–601 Hasstedt SJ (1994) Pedigree analysis package. Department of Human Genetics, University of Utah. Email:
[email protected]. utah.edu Hasstedt SJ, Clegg DO, Ingles L, Ward RH (1994) HLA-linked rheumatoid arthritis. Am J Hum Genet 55:738–746 Hochberg MC, Spector TD (1990) Epidemiology of rheumatoid arthritis: update. Epidemiol Rev 12:247–252 Kwoh CK, Venglish C, Lynn AH, Whitley DM, Young E, Chakravarti A (1996) Age, sex, and the familial risk of rheumatoid arthritis. Am J Epidemiol 144:15–24 Lalouel JM (1979) Gemini – a computer program for optimization of a nonlinear function. Department of Medical Biophysics and Computing, University of Utah, Technical Report No. 14 Lawrence JS (1970) Rheumatoid arthritis – nature or nurture? Ann Rheum Dis 29:357–379 Liang KY, Beaty TH (1991) Measuring familial aggregation by using odds-ratio regression models. Genet Epidemiol 8:361–370 Linos A, Worthington JW, O’Fallon WM, Kurland LT (1980) The epidemiology of rheumatoid arthritis in Rochester, Minnesota: a study of incidence, prevalence, and mortality. Am J Epidemiol 111:97–98 Lippman SM, Arnett FC, Conley CL, Ness PM, Meyers DA, Bias WB (1982) Genetic factor predisposing to autoimmune diseases. Autoimmune hemolytic anemia, chronic thrombocytopenic purpura, and systemic lupus erythematosus. Am J Med 73:827–840 Lynn AH, Kwoh CK, Venglish CM, Aston CE, Chakravarti A (1995) Genetic epidemiology of rheumatoid arthritis. Am J Hum Genet 57:150–159 Mulhern LM, Masi AT, Shulman LE (1966) Hashimoto’s disease. A search for associated disorders in 170 clinically detected cases. Lancet 2:508–511 Myers AR (1994) Medicine: the national medical series for independent study, 2nd edn. Harwal Pub. Co., Philadelphia, pp 424–425 Pritchard MA (1994) Evidence for a hypothetical non-HLA susceptibility gene in rheumatoid arthritis. Br J Rheumatol 33:457–479
Qaqish BF, Liang KY (1992) Marginal models for correlated binary responses with multiple classes and multiple levels of nesting. Biometrics 48:939–950 Remmers EF, Longman RE, Du Y, O’Hare A, Cannon GW, Griffiths MM, Wilder RL (1996) A genome scan localizes five non-MHC loci controlling collagen-induced arthritis in rats. Nat Genet 14:82–85 Rigby AS (1992) HLA haplotype sharing in RA sibships: risk estimates in siblings. Scand J Rheumatol 21:68–73 Rigby AS, Voelm L, Silman AJ (1993) Epistatic modeling in rheumatoid arthritis: an application of the Risch theory. Genet Epidemiol 10:311–320 Risch N (1990) Linkage strategies for genetically complex traits. I. Multilocus models. Am J Hum Genet 46:222–228 Sakamoto Y, Ishiguro M, Kitagawa G (1986) Akaike information criterion statistics. KTK Scientific Publishers, Tokyo, pp 83–85 Sels F, Westovens R, Emonds M-P, Vandermeulen E, Dequeker J (1997) HLA typing in a large family with multiple cases of different autoimmune diseases. J Rheumatol 24:856–859 Suarez-Almazor ME, Tao S, Moustarah F, Russell AS, Maksymowych W (1995) HLA-DR1, DR4, and DRB1 disease-related subtypes in rheumatoid arthritis. Association with susceptibility but not severity in a city-wide community-based study. J Rheumatol 22:2027–2033 Taneja V, Singh RR, Malaviya AN, Anand C, Mehra NK (1993) Occurrence of autoimmune diseases and relationship of autoantibody expression with HLA phenotypes in multicase rheumatoid arthritis families. Scand J Rheumatol 22:152–157 Teitsson I, Thorsteinsson J, Arnason A, Valdimarsson H (1985) Rheumatic diseases in an Icelandic family. Scand J Rheumatol 14:109–118 Thomas DJB, Young A, Gorsuch AN, Bottazzc GF, Cudforth AG (1983) Evidence for an association between rheumatoid arthritis and autoimmune endocrine disease. Ann Rheum Dis 42:279–300 Thompson EA, Cannings C (1979) Sampling schemes and ascertainment. In: Sing CF, Skolnick M (eds) The genetic analysis of common diseases: applications to predictive factors in coronary heart disease. Alan Liss, New York, pp 363–382 Thomson G (1995) HLA disease associations: models for the study of complex human genetic disorders. Crit Rev Clin Lab Sci 32:183–219 Tunbridge WMG, Evered DC, Hall R, Appleton D, Brewis M, Clark F, Grimley Evans J, Young E, Bird T, Smith PA (1977) The spectrum of thyroid disease in a community: the Whickham survey. Clin Endocrinol 7:481–493 Walker DJ, Griffiths M, Griffiths ID (1986) Occurrence of autoimmune diseases and autoantibodies in multicase rheumatoid arthritis families. Ann Rheum Dis 45:323–326 Weyand CM, Hicok KC, Conn DL, Goronzy JJ (1992) The influence of HLA-DRB1 genes on disease severity in rheumatoid arthritis. Ann Intern Med 117:801–806 Yamashita TS, Khan MA, Kushner I (1986) Genetic analysis of families with multiple cases of rheumatoid arthritis. Dis Markers 4:113–119