Dr. Gregersen is a member of the Abbott Scholar Scientific Review Board and receives consulting fees (more than $10,000) from Abbott.
Dissecting the heterogeneity of rheumatoid arthritis through linkage analysis of quantitative traits
Article first published online: 28 DEC 2006
Copyright © 2006 by the American College of Rheumatology
Arthritis & Rheumatism
Volume 56, Issue 1, pages 58–68, January 2007
How to Cite
Criswell, L. A., Chen, W. V., Jawaheer, D., Lum, R. F., Wener, M. H., Gu, X., Gregersen, P. K. and Amos, C. I. (2007), Dissecting the heterogeneity of rheumatoid arthritis through linkage analysis of quantitative traits. Arthritis & Rheumatism, 56: 58–68. doi: 10.1002/art.22325
- Issue published online: 2 JAN 2007
- Article first published online: 28 DEC 2006
- Manuscript Accepted: 6 OCT 2006
- Manuscript Received: 13 JAN 2006
- NIH. Grant Numbers: K24-AR-02175, ES-09912
- North American Rheumatoid Arthritis Consortium was supported by the National Arthritis Foundation
- National Institute of Allergy and Infectious Diseases. Grant Number: N01-AR-72232
- National Institute of Arthritis and Musculoskeletal and Skin Diseases. Grant Number: R01-AR-44222
- General Clinical Research Center, Moffitt Hospital, University of California, San Francisco
- NIH National Center for Research Resources from the USPHS. Grant Number: 5-M01-RR-00079
- Top of page
- PATIENTS AND METHODS
- AUTHOR CONTRIBUTIONS
To dissect the heterogeneity of rheumatoid arthritis (RA) through linkage analysis of quantitative traits, specifically, IgM rheumatoid factor (IgM-RF) and anti–cyclic citrullinated peptide (anti-CCP) autoantibody titers.
Subjects, 1,002 RA patients from 491 multiplex families recruited by the North American RA Consortium, were typed for 379 microsatellite markers. Anti-CCP titers were determined based on a second-generation enzyme-linked immunosorbent assay, and IgM-RF levels were quantified by immunonephelometry. We used the Merlin statistical package to perform nonparametric quantitative trait linkage analysis.
For each of the quantitative traits, evidence of linkage, with logarithm of odds (LOD) scores of >1.0, was found in 9 regions. For both traits, the strongest evidence of linkage was for marker D6S1629 on chromosome 6p (LOD 14.02 for anti-CCP and LOD 12.09 for RF). Six other regions with LOD scores of >1.0 overlapped between the 2 traits, on chromosomes 1p21.1, 5q15, 8p23.1, 16p12.1, 16q23.1, and 18q21.31. Evidence of linkage to anti-CCP titer but not to RF titer was found in 2 regions (chromosomes 9p21.3 and 10q21.1), and evidence of linkage to RF titer but not to anti-CCP titer was found in 2 regions (chromosomes 5p15.2 and 1q42.3). Several covariates were significantly associated with 1 or both traits, and linkage analysis exploring the covariate effects revealed striking effects of sex in modulating linkage signals for several chromosomal regions. For example, sex had a striking impact on the linkage results for both quantitative traits on chromosome 6p (P = 0.0007 for anti-CCP titer and P = 0.0012 for RF titer), suggesting a sex–HLA region interaction.
Analysis of quantitative components of RA is a promising approach for dissecting the genetic heterogeneity of this complex disorder. These results highlight the potential importance of sex or other covariates that may modulate some of the genetic effects that influence the risk of specific disease manifestations.
Rheumatoid arthritis (RA) is a chronic inflammatory disease with the potential to cause substantial disability, primarily as a result of the erosive and deforming process in joints. The disease is characterized by a spectrum of clinical and laboratory manifestations, with resultant variation in the RA phenotype expressed by affected individuals. Presumably, this phenotypic heterogeneity reflects differences in underlying disease mechanisms, which might be based, at least in part, on genetic differences (1, 2).
Like most chronic diseases, RA is etiologically complex, with important contributions from multiple genetic and environmental factors (3). The most rigorous assessment of RA heritability, which was based on a quantitative analysis of Finnish and English twins, indicates that 50–60% of the occurrence of RA in twins is explained by genetic effects (4). Genome-wide screening in multicase families, which involves the analysis of a large set of informative, polymorphic markers that are approximately evenly spaced across the genome, to identify regions with evidence of linkage to disease, has been a popular approach for identifying susceptibility genes. Several full genome screens have now been performed in families with multiple RA cases (5–10). Although these studies reveal some overlap in genomic regions of linkage, as highlighted by a recent meta- analysis (11), overall the results of these individual studies are notable for their differences. In view of the phenotypic heterogeneity of RA, it is likely that some of these differences reflect underlying phenotypic differences in the study populations. Thus, linkage analysis of more homogeneous subgroups may shed light on these disparate results and elucidate underlying disease mechanisms with distinct genetic associations.
Although RA has typically been investigated as a categorical trait, i.e., one that is either present or absent, other human traits or disorders are more commonly examined as continuous, or quantitative, traits. Linkage analyses of quantitative traits are currently of great interest, in part due to the potentially increased power of quantitative, compared with qualitative, outcomes (12) and due to advances in statistical methodology (13). For example, analyses of asthma-associated phenotypes as quantitative traits (e.g., levels of serum immunoglobulins) have revealed novel genetic loci for this complex disorder (14, 15).
In the present study, we sought to apply this powerful methodology to the identification of RA disease loci by performing a genome-wide linkage analysis of 2 quantitative traits among a large collection of multiplex RA families recruited by the North American RA Consortium (NARAC). Specifically, we examined serum levels of 2 RA-related autoantibodies, IgM rheumatoid factor (IgM-RF) and anti–cyclic citrullinated peptide (anti-CCP). RF has been used diagnostically and prognostically in RA for years, as a result of well-established associations with disease risk and outcome (16). There is also growing interest in anti-CCP titers in RA due to their high specificity (17) and the fact that these autoantibodies may be detected many years before RA onset (18). Anti-CCP autoantibodies are also associated with disease severity, particularly the development and progression of the erosive process in joints (19, 20).
PATIENTS AND METHODS
- Top of page
- PATIENTS AND METHODS
- AUTHOR CONTRIBUTIONS
We studied 1,002 affected siblings in 491 multicase RA families recruited as a collaborative effort by the NARAC (21). The majority of families included a single sibpair (88.4%), and the remaining families included 3 siblings (10.4%), 4 siblings (1%), or 6 siblings (0.2%). Details of our enrollment procedures have been published previously (1), and clinical and genetic marker data are available at http://www.naracdata.org/index.asp. Briefly, eligible families had to meet the following criteria: ≥2 siblings satisfied the American College of Rheumatology (ACR; formerly, the American Rheumatism Association) 1987 criteria for RA (22), ≥1 sibling had documented erosions on hand radiographs, and ≥1 sibling had disease onset between the ages of 18 and 60 years. The presence of any other diseases associated with similar articular symptoms, such as psoriasis or inflammatory bowel disease, was an exclusionary criterion for families. Informed consent was obtained from every subject, including all participating family members, and approval of the local institutional review board was secured at every recruitment site prior to enrollment.
Basic demographic and clinical information, including ethnic background, age at onset of RA symptoms, and exposure to tobacco smoke, was collected from all RA patients by telephone. Confirmation of RA diagnosis, including which ACR criteria were fulfilled, was obtained from patients' rheumatologists. At the time of study enrollment, RA patients were examined to document the extent of joint involvement and functional disability, using validated RA instruments (23, 24). Radiographs of the hands and wrists were also obtained to document the presence and extent of joint involvement.
RF and anti-CCP testing.
Serum samples collected at study entry were tested for IgM-RF and anti-CCP autoantibodies in the Immunology Division of the Department of Laboratory Medicine at the University of Washington. RF testing was performed using a latex-enhanced nephelometric assay (Behring Diagnostics, San Jose, CA) in which human and rabbit IgG–coated latex beads are used as antigen. This assay was calibrated to the World Health Organization international standard for RF (25). RF values <12 IU were considered negative and were assigned a quantitative value of 11 for analysis. Anti-CCP titers were determined based on a second-generation enzyme-linked immunosorbent assay (Inova Diagnostics, San Diego, CA). Anti-CCP values exceeding the upper limit of linearity were rounded off to that value. Thus, all anti-CCP values >210 were assigned a quantitative value of 210 for analysis.
Microsatellite marker genotyping.
All families were genotyped for 379 microsatellite markers from the Marshfield Set 8A Combo List with additional markers in certain regions, such as the HLA complex (6). These markers provided ∼10 cM genome-wide coverage and consisted of 47 panels. Reaction conditions were standard for all markers used, as described in the Marshfield polymerase chain reaction (PCR) protocol (http://research.marshfieldclinic.org/genetics/sets/combo.html). Details of our methods for microsatellite genotyping in these families have been described previously (8).
The genetic relationships among individuals were checked using Relative (26), resulting in the identification of 29 half-sibling pairs and 4 unrelated pairs who were initially reported to be full siblings. Consistency in allele inheritance within families was then verified using the PedCheck program (27). The summary error rate, which includes the number of genotypes that were eliminated to resolve Mendelian inconsistencies, was 1.12%. Whenever a Mendelian inconsistency was found, we applied a conservative approach of removing the genotypes of all individuals whose genotypes caused the inconsistency. The data were also checked for the presence of monozygotic twin pairs, and the 8 pairs identified (based on identical genotypes) were eliminated from the analysis. Testing for Hardy-Weinberg equilibrium of the microsatellite markers was also performed, using the method developed by Guo and Thompson (28), and implemented with the SAS Allele procedure (SAS Genetics, Cary, NC). Based on these results, we repeated the linkage analysis for genomic regions containing ≥1 microsatellite marker with significant evidence of Hardy-Weinberg disequilibrium (at a genome-wide significance level of P < 0.000013). Finally, Simwalk (29) was used to verify intermarker distances against the published maps used, as previously reported (8).
Broad-level HLA–DRB1 typing for the allele groups DRB1*01 through DRB1*18, and high-resolution DRB1*04 typing were accomplished by initial PCR amplification of groups of alleles using biotinylated PCR primers, followed by hybridization to immobilized sequence-specific oligonucleotide probes in a linear array format. Positive hybridization reactions were detected using a streptavidin–horseradish peroxidase conjugate and a soluble colorless substrate, 3,3′,5,5′-tetramethylbenzidine (30). A computer algorithm based on the sequence-specific oligonucleotide probe hybridization pattern and the Anthony Nolan 1999 HLA sequence database (http://www.ebi.ac.uk/imgt/hla/) was used to assign genotypes. The following alleles detected were classified as shared epitope positive: DRB1*0101, 0102, 0104, 0105, 0401, 0404, 0405, 0408, 0409, 1001, 1402, and 1406. Substantial previous research has documented association of HLA–DRB1 alleles encoding the shared epitope sequence with RA (3, 31).
Linkage analysis for the 2 quantitative traits, RF titer and anti-CCP titer, was performed using the Multipoint Engine for Rapid Likelihood Inference (Merlin) statistical package (32) (http://www.sph.umich.edu/csg/abecasis/Merlin/). We chose to use the nonparametric quantitative trait linkage statistic, which implements a score test as proposed by Whittemore and Halpern (33). The procedure also allows for variable marker information following the approach suggested by Kong and Cox (34). Because our sample was selected to include individuals with RA, reference to the population mean was not an effective approach for analysis. We therefore used the “deviates” option, in which the population mean is fixed at zero, in order to allow for the sampling scheme used (i.e., families selected based on RA diagnosis). Prior to analysis we subtracted a “mean” value of 11 from all RF titers because all 20 normal subjects assayed had a recorded value <11. Similarly, we subtracted 4.6 from anti-CCP titers, since this was the mean for the normal subjects we assayed. To obtain asymptotic P values we converted all positive scores to a chi-square deviate with a 1:1 mixture of 0 df and 1 df, by multiplying them by 4.6 and then identifying that part of the resulting chi-square distribution exceeding the observed logarithm of odds (LOD) score. Allele frequencies were calculated by counting the number of alleles genotyped in each subject. As reported by Boehnke, this approach provides a consistent estimate of the allele frequencies that is adequate given the large number of independent families studied (35).
Because variance components procedures can yield excess false-positive findings in the presence of non-normal trait distributions, we also performed simulation studies to assess the empirical P values. For these simulation studies, we maintained the same pedigree structure and phenotypes as were observed. We then simulated marker data according to the marker allele frequencies, availability of samples from individuals, and distances among markers, using Allegro version 1.2c (36). For chromosome 6, which showed the most significant results, we performed 10,000 replicate simulations and then used the same Merlin setup described above to obtain empirical P values. We then compared the resultant empirical (i.e., experimental) P values with asymptotic P values calculated from a mixture of chi-square distributions (i.e., derived from asymptotic theory). We also studied the impact that transforming the data would have on our analyses. However, we found that standard approaches such as Box-Cox or probit transformation did not reduce the skewness or kurtosis because of the right (for anti-CCP) or left (for RF) truncation of values associated with the measurements.
Several covariates were associated with 1 or both of the quantitative traits. Therefore, we repeated the linkage analysis for each trait after adjusting for the associated covariates. The following covariates, determined at the time of anti-CCP and RF testing, were considered for these analyses: sex, ethnicity, age at study entry, disease duration (time since RA diagnosis), exposure to tobacco smoke, and HLA–DRB1 shared epitope positivity. Linkage analysis incorporating the significant covariates for each trait was performed using covariate-adjusted trait residuals in Merlin (deviates option). The mean levels of the traits were reset to the mean among affected subjects prior to analysis. We assessed our regression analyses for potential effects from outliers by evaluating the Cook's D measure.
To further evaluate the impact of specific covariates in the linkage analysis, we performed stratified analyses. The stratification variables that we studied included sex, disease duration, exposure to tobacco smoke, and presence of the shared epitope. In the first analysis, we treated the families as units without further modification except when affected individuals within a family were discordant for an exposure, in which case we created pairs of discordant relatives. To assess significance, we compared the evidence of linkage after dividing the families into strata with the evidence of linkage that we obtained using the same family structures but without dividing the families into strata. After multiplying the difference in LOD scores for the 2 analyses by 4.6, the scores so obtained approximately followed a chi-square distribution with 2 df. Thus, we obtained P values for this test by comparison with a chi-square distribution. However, because the family sizes for the stratified analyses varied for this analysis, the mean LOD scores per family are not meaningful. For descriptive purposes, we therefore divided the families into all possible pairs of affected relatives and then presented the mean LOD scores per family among the strata. Finally, we compared linkage results for the 2 quantitative traits with nonparametric linkage results for RA defined qualitatively among the same group of families, using the nonparametric linkage statistic in Merlin.
- Top of page
- PATIENTS AND METHODS
- AUTHOR CONTRIBUTIONS
Table 1 summarizes demographic and clinical characteristics of the 1,002 affected individuals. The majority of subjects were Caucasian, and 77% were female. The mean age at RA diagnosis was relatively young, and most patients had well-established disease at study entry. The majority of affected individuals had evidence of erosions, as expected based on our eligibility criteria. Eighty-four percent of subjects were HLA–DRB1 shared epitope positive. Table 1 also shows the mean ± SD values for the 2 quantitative traits, IgM-RF and anti-CCP autoantibody titers. The distribution of RF and anti-CCP titers among the study population revealed striking variation and a non-normal distribution of trait values. The median trait values were 83 for IgM-RF and 113 for anti-CCP titer. The skewness and kurtosis coefficients were 4.97 and 33.03, respectively, for RF titer and 0.07 and −1.31, respectively, for anti-CCP, reflecting the fact that some individuals had very high RF titers (positive skewing and kurtosis). The distributions of IgM-RF and anti-CCP titers are shown in Figure 1. We also assessed skewness and kurtosis of the residuals following adjustments and checked for evidence of influence. For anti-CCP titer, no Cook's D observations were near a critical value of 1.0 and the largest value was 0.01, providing evidence against excess influence from outliers in the regression analyses. For RF titer, the largest Cook's D value was 0.11, suggesting some potential excess influence.
|Age at RA diagnosis, mean ± SD years||41 ± 13|
|Disease duration, mean ± SD years||14 ± 11|
|% with erosions||92|
|% HLA–DRB1 shared epitope positive||84|
|% current smoker†||18|
|IgM-RF titer, mean ± SD||262 ± 542|
|Anti-CCP titer, mean ± SD||109 ± 78|
We used the ACT program (http://www.epigenetic.org/Linkage/act.html) (37) to estimate the heritability of these 2 quantitative traits. The results of these estimates showed 67% heritability for anti-CCP titer and 14% for RF titer. Although it is important to keep in mind that the ascertainment scheme may have influenced these results (i.e., all individuals examined had RA), the relatively higher heritability estimate for anti-CCP versus RF titer suggests that anti-CCP titer is more strongly influenced by genetic factors, whereas RF titer might be more strongly influenced by environmental or other factors. After adjusting for covariates, the heritability of anti-CCP titer converged to 1.0, while the heritability estimate for RF titer was 37%.
Table 2 summarizes the genomic regions with the strongest evidence of linkage to each of the quantitative traits (i.e., LOD >1.0), including the name and position of the best marker in each linkage region. These results do not include adjustment for covariates. For each of the quantitative traits, evidence of linkage, with LOD scores >1.0, was found in 9 regions. The marker that showed the strongest evidence of linkage was the same for both traits, i.e., D6S1629 in the major histocompatibility complex on chromosome 6p (anti-CCP LOD 14.02, P < 10−15; RF LOD 12.09, P < 10−13). Six other regions with LOD scores of >1.0 overlapped between the 2 traits, on chromosomes 1 (D1S1631), 5 (D5S1462), 8 (D8S277), 16 (D16S403, D16S516), and 18 (D18S858). Evidence of linkage to anti-CCP but not RF titer was found in 2 regions (chromosomes 9 [D9S1121] and 10 [D10S1221]), and evidence of linkage to RF but not anti-CCP titer was found in 2 regions (chromosomes 5p [D5S817] and 1q [D1S235]).
|Chromosome (cM)||Best marker||Anti-CCP||IgM-RF|
|6 (47.7)||D6S1629||14.02||9.99 × 10−16||12.09||8.82 × 10−14|
Simulation studies of the linkage results for chromosome 6p were consistent with very strong evidence of linkage and also indicated that the asymptotic P values conformed well to the empirical P values (Figure 2), except that for anti-CCP the asymptotic significance level for P values <1 × 10−5 appeared to be underestimated. We therefore report asymptotic P values.
Comparison of unadjusted linkage results for both quantitative traits with results of nonparametric linkage analysis of RA defined qualitatively (i.e., the presence of RA) revealed generally similar regions with evidence of linkage. However, the strength of evidence of linkage varied across the 3 traits, as illustrated in Figure 3, which displays the results of linkage analysis for all 3 traits, for the 8 chromosomes with LOD scores >1.0 for at least 1 of the quantitative traits. For example, the evidence of linkage on chromosome 1 was strongest for RF titer, whereas the linkage evidence on chromosome 18 was strongest for anti-CCP titer, and the linkage evidence on chromosome 16 was strongest for RA defined qualitatively.
Several covariates were associated with the quantitative traits examined. Specifically, male sex was significantly associated with both quantitative traits (P < 0.0001 for anti-CCP titer and P = 0.020 for RF titer), whereas disease duration and exposure to tobacco smoke (at the time of autoantibody testing) were significantly associated with RF titer (P = 0.021 for disease duration and P = 0.0007 for tobacco exposure) but not anti-CCP titer. The HLA–DRB1 shared epitope was significantly associated with anti-CCP titer (P < 0.0001) but not RF titer.
Based on these results, we repeated the linkage analysis for anti-CCP titer after adjusting for sex and HLA–DRB1 shared epitope positivity (see Patients and Methods for definition of shared epitope alleles). Similarly, we repeated the linkage analysis for RF titer after adjusting for sex, disease duration, and exposure to tobacco smoke. Results of linkage analyses that substantially changed the LOD scores for either RF or anti-CCP are presented in Table 3 and in Figure 4, which shows results obtained before and after adjustments for covariates. Covariate adjustments influenced the results of linkage analysis for both traits, but particularly RF titer. For example, adjustment of RF titer for sex, disease duration, and exposure to tobacco smoke revealed 1 region with a 1.90 increase in the LOD score (on chromosome 3 [D3S1304]) compared with the unadjusted analysis. Five regions demonstrated a decrease of >1.0 in the LOD score (range 1.19–9.36 decrease) in the covariate-adjusted linkage analyses. Most striking was the 9.36 decrease in LOD score for chromosome 6 (marker D6S1629). Of note, this analysis did not include HLA–DRB1 shared epitope positivity as a covariate, since it was not significantly associated with RF titer in the crude analyses.
|Chromosome (cM)||Marker||Change in LOD|
|IgM-RF titer||Anti-CCP titer|
Adjustment of anti-CCP titer for sex and DRB1 shared epitope positivity revealed 1 region with a 5.38 decrease in LOD score, on chromosome 6 (from 14.02 to 8.64 at marker D6S1629). It was of interest that the peak LOD score on chromosome 6 was high (8.64) even after adjusting for DRB1 shared epitope positivity and sex, suggesting substantial additional genetic contribution to anti-CCP titer in this region. Further, the decrease in LOD score was greater with adjustment for all shared epitope–positive alleles compared with adjustment for only the subgroup of DR4+ shared epitope alleles (decrease of 4.09, from 14.02 to 9.93), providing evidence of the importance of this group of alleles (versus individual DRB1 alleles) in relation to anti-CCP titers. Overall, the changes in linkage results after adjustment for specific covariates suggest possible interaction between a gene or genes in these regions and the specific covariates analyzed, or that covariate effects explain part of the linkage signal.
In order to evaluate further the impact of specific covariates on the linkage results for these quantitative traits, we performed stratified analyses, defined by specific covariates, on the results shown in Table 3 and Figure 4. Table 4 shows results for chromosome 6, from analyses of sex strata that were formed by organizing the data into pairs of individuals and then treating these as independent pairs. Similarly, smoking status strata and duration of disease strata (using the median value as cut point) in relation to the RF titer were analyzed for chromosomes 1, 3, 5, 6, and 16, and HLA–DRB1 shared epitope strata in relation to anti-CCP titer were analyzed for chromosome 6.
|Stratum||Anti-CCP titer||IgM-RF titer|
|No. of pairs analyzed||Peak LOD score†||Mean LOD score per sibpair||No. of pairs analyzed||Peak LOD score†||Mean LOD score per sibpair|
Results of these stratified analyses indicated that sex had a striking impact on the linkage results for both quantitative traits on chromosome 6. Table 4 shows the mean LOD score per sibpair by sex stratum, in the anti-CCP and RF titer linkage analyses. The sex of the sibling pair had a dramatic impact on the strength of the linkage signal on chromosome 6p. For example, the per-pair LOD score associated with anti-CCP was 0.0235 for female/female pairs, 0.0558 for female/male pairs, and 0.1058 for male/male pairs. There was also significant evidence that the sex of the pairs influences the evidence of linkage (P = 0.0007 for anti-CCP titer and P = 0.0012 for RF titer). Significance was assessed by summing the LOD scores across strata and then comparing the summed LOD score with that obtained from pooling the data. The finding suggests a sex–HLA region interaction and may explain why adjustment of the quantitative trait levels for the covariate sex resulted in a substantial decrease in the observed LOD scores. Analyses in which we did not divide the families into sibling pairs showed similar results (P = 0.014 for anti-CCP, P = 0.027 for RF).
On chromosome 16, we detected a significant interaction with sex prior to division of the families into pairs (P = 0.0043 for RF) and an effect that was not significant after division into pairs (P = 0.080 for RF). Results of stratifications according to the smoking status and disease duration of the sibpair members did not reveal significant heterogeneity for the analysis of RF titer in relation to these chromosomes (1, 3, 5, 6, and 16), suggesting that the covariate sex was of primary importance in the covariate-adjusted analyses.
Finally, results of stratification according to shared epitope strata for the analysis of anti-CCP titer in relation to chromosome 6 revealed striking differences (P < 10−7), as expected based on the well-documented importance of the HLA region to anti-CCP autoantibody production in RA. Further, although shared epitope positivity was not significantly associated with RF titer in univariate analyses, results of stratified linkage analysis for RF titer in relation to chromosome 6 revealed a significant effect of the HLA–DRB1 shared epitope (P < 10−5).
- Top of page
- PATIENTS AND METHODS
- AUTHOR CONTRIBUTIONS
Significant progress has been made during the past 5 years in the identification of genomic regions likely to contribute to RA etiology, primarily through genome-wide linkage analyses. However, in view of the clinical heterogeneity of the disorder, as well as evidence that genetic differences explain at least part of this variation (1), it is likely that analysis of more homogeneous subgroups will be required to fully unravel the genetics of this complex disorder, as has been the case for a number of other diseases, such as asthma and diabetes (14, 15, 38).
Our goal for this study was to begin to dissect the complexity of RA through linkage analysis of 2 quantitative RA-related traits, specifically serum levels of 2 autoantibodies. The availability of a large and well-characterized collection of RA families recruited by the NARAC, in conjunction with recent developments in methods of quantitative trait linkage analysis, has been essential to our ability to perform the present study.
Comparison of linkage results for the 2 quantitative traits examined with results for RA defined qualitatively revealed similarity in regions where evidence of linkage was found. However, there was variation in the strength of evidence of linkage to these 3 phenotypes, which presumably reflects the presence of certain genes or genomic regions that primarily influence the production of 1 or both autoantibodies, and other genes or regions that influence other RA disease processes.
The similarity of linkage results for the 2 quantitative traits (i.e., anti-CCP and RF titers) suggests the existence of genomic regions that influence titers of both autoantibodies. This phenomenon is known as pleiotropy. Although we sought to formally evaluate the evidence of pleiotropy in the genomic regions exhibiting the most striking similarities in linkage results for the 2 quantitative traits examined, we were unable to successfully perform these analyses given the characteristics of our sample (i.e., the selection of RA patients resulted in a sample of individuals with high levels of both traits) and limitations of the currently available software programs. Further, despite the similarity of linkage findings for RF and anti-CCP titers, the lower heritability and high skewness of RF titers indicate that results of analysis of RF titers are less reliable than results of analysis of anti-CCP titers. Nonetheless, given the potential value of such analyses (for example, as a means of identifying chromosomal regions that control autoantibody production in RA and possibly other autoimmune diseases), this is an important area of future methodologic research.
It is interesting to note that the recently discovered autoimmunity gene PTPN22 is located within the first peak on chromosome 1. A missense single-nucleotide polymorphism (SNP) in this gene is strongly associated with RF-positive RA and other autoimmune diseases characterized by prominent autoantibody production (39–41). Our results indicated that the linkage evidence in this region was strongest for RF titer. However, the missense SNP alone does not explain the linkage signal in this region, since the proportion of variance of anti-CCP explained by PTPN22 R620W genotypes was 0.02% for anti-CCP titer and 0.2% for RF titer; thus, there are likely to be other genetic variants in this region that also influence autoantibody production.
It is also of note that we failed to detect a strong linkage signal on chromosome 1p in the region of the PADI4 gene, which is known to influence citrullination and has been shown to be a risk factor for RA in Asian populations (42). One might have anticipated that anti-CCP levels would show linkage in this region, and our results suggest that variants of PADI4 that may influence anti-CCP levels are uncommon in our study population.
Clearly, additional work will be needed to confirm our findings and to narrow linkage regions. Future studies might involve higher-resolution genome screens using dense panels of SNPs and/or genome screens in other multiplex RA family cohorts. We have recently completed a genome screen, using SNP markers, on families in the NARAC cohort (10), which will provide a valuable resource for refining the genomic regions that most strongly influence these important RA-related phenotypes. This work has important implications regarding our ability to identify genetic loci relevant to specific etiologic processes and hence to elucidate distinct underlying mechanisms of disease.
We also sought to investigate genetic heterogeneity through incorporation of specific demographic, genetic, and environmental covariates in the linkage analyses. The association of specific covariates with these quantitative traits, and their influence on the results of linkage analysis, provide evidence of the genetic heterogeneity of the disorder, which should be pursued more completely in future studies. These findings were the most novel results of the present study and warrant careful consideration in future genetic and epidemiologic studies of RA. Of particular interest was the substantial impact of sex on the evidence of linkage to RF titer in the HLA region on chromosome 6. Other chromosomal regions also exhibited substantial changes in linkage results for RF titer after adjustment for covariates (see Table 3), suggesting that this trait may be importantly influenced by nongenetic factors, perhaps as a result of gene–environment interaction.
Prior research has also provided evidence of an important role of male sex and environmental factors in RF autoantibody production (43–45). The differences in LOD score according to the sex of the sibpair members observed for certain chromosomes, such as the HLA region on chromosome 6p, are particularly striking and suggest interaction between sex and genes within these genomic regions. Also of interest was the residual evidence of linkage in the HLA region to anti-CCP titer after adjustment for the HLA–DRB1 shared epitope, which is considered to be the primary RA genetic association in this region. Further, adjustment for DR4+ shared epitope alleles had a smaller impact on the linkage results compared with adjustment for the entire group of shared epitope alleles, supporting the notion that this entire group of alleles is important, at least with regard to anti-CCP titers.
As mentioned above, the development of statistical methods for linkage analysis of quantitative traits is an area of active investigation in the field of statistical genetics. One of the challenges faced in the current analysis relates to the sampling scheme for families. Specifically, the selection of families with multiple affected siblings introduced additional complexity into our analysis, since most methods have been developed based on the assumption that samples will be obtained from the general population. Due to this design feature we investigated a number of different statistical tests and performed simulation studies to evaluate significance values for specific methods. Although we chose to use a specific method that is robust with regard to this design feature, it may also have influenced our ability to fully investigate covariate effects and will require additional methodologic work in future studies. The selection of multicase families and other design features of the NARAC collection also influenced the characteristics of the study population, resulting in a group of patients with younger age at RA onset and more severe disease compared with RA patients in the general population. Further, the majority of enrolled families were of Caucasian ancestry. Thus, it will be important in future studies to evaluate the relevance of potential quantitative trait loci in RA patients with different ethnic or clinical characteristics, such as later-onset sporadic disease.
In summary, we have performed genome-wide linkage analysis of 2 RA-related quantitative traits in order to dissect the genetic contribution to this complex disease. Given the clinical heterogeneity of RA, careful analysis of quantitative traits may facilitate the identification of genes that contribute to this disease and shed light on distinct underlying mechanisms. To the extent that such pathways are associated with distinct outcomes and responses to treatment, this information will guide the development of more specific diagnostic and therapeutic tools. These results also highlight the potential importance of specific covariates and may be useful in identifying regions containing genes that interact with sex or with specific environmental factors to influence the risk of RA or specific disease manifestations.
- Top of page
- PATIENTS AND METHODS
- AUTHOR CONTRIBUTIONS
Dr. Criswell had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study design. Drs. Criswell, Wener, Gregersen, and Amos.
Acquisition of data. Drs. Criswell, Jawaheer, Wener, and Gregersen.
Analysis and interpretation of data. Dr. Criswell, Ms Chen, Mr. Lum, and Drs. Wener, Gregersen, and Amos.
Manuscript preparation. Dr. Criswell, Ms Chen, and Drs. Jawaheer and Amos.
Statistical analysis. Dr. Criswell, Ms Chen, Mr. Gu, and Dr. Amos.
- Top of page
- PATIENTS AND METHODS
- AUTHOR CONTRIBUTIONS
The authors wish to thank participants in the NARAC family collection and the NARAC investigative team. We are also grateful to the following individuals for their assistance with genotyping for this project: Michael F. Seldin, Russell Shigeta, Susan Dowbak, Xiangli Xiao, Joanita Monteiro, and Dong Chen. Naila Ahmad and Kirsten Pfeiffer also provided valuable assistance with this project. Gonzalo Abecasis provided considerable guidance in the use of Merlin.
- Top of page
- PATIENTS AND METHODS
- AUTHOR CONTRIBUTIONS
- 16Rheumatoid factors: what do they tell us? J Rheumatol 2002; 29: 234–40..