Prospective analysis of the impact of HLA-DR and -DQ on joint destruction in recent-onset rheumatoid arthritis

Rheumatology Advance Access published February 28, 2003 Rheumatology 2003;42:1 of 10 doi:10.1093/rheumatology/keg190, available online at www.rheumatology.oupjournals.org Prospective analysis of the impact of HLA-DR and -DQ on joint destruction in recent-onset rheumatoid arthritis U. Wagner, S. Kaltenhäuser, M. Pierer, W. Seidel, M. Tröltzsch, H. Häntzschel, J. R. Kalden 1 and R. Wassmuth 1,2 Objective. To evaluate the differential impact of HLA-DR and -DQ on the progression of erosive disease in the clinical course of early rheumatoid arthritis (RA). Methods. HLA genotyping for HLA-DR and -DQ was carried out in a prospective study of 87 patients with early RA. The progression of erosive disease was assessed by radiological scores over a period of 2 yr in all patients and over 4 yr in 77 patients. The impact of HLA markers was evaluated by univariate comparisons and by multiple logistic regression analyses. Results. Patients expressing the RA-associated shared epitope (SE) on a DRB1*01-positive or, most prominently, on a DRB1*04-positive allele had higher Larsen scores at all time-points analysed when compared with SE-negative patients. A similar impact on radiological progression was seen for the RApredisposing DQ3, but not for DQ5 heterodimers. In the presence or absence of the DRB1 SE, no additional effects could be discerned for RA-associated DQ molecules. The presence of a DERAA-positive DRB1 allele was associated with a slower pace of joint destruction. While gene dosage effects were seen for SE compound homozygosity, no effect for DQ3 homozygosity could be discerned. Conclusion. Although a significant influence of HLA-DQ3 heterodimers on the progression of erosive joint destruction was seen, the analysis of the HLA-DQ locus did not add additional information over the study of HLA-DR including the determination of the SE and the DERAA motif in order to predict the development of severe progressive joint destruction. KEY WORDS: Rheumatoid arthritis, Joint destruction, Immunogenetics, Human leucocyte antigens (HLA), Epidemiology. Rheumatoid arthritis (RA) can be regarded as an autoimmune disease on a permissive genetic background. The human leucocyte antigens (HLA) of the class II region are important for the predisposition as well as the rate of progression of joint destruction in this autoimmune disease. Most prominently, different DRB1*04 alleles (*0401, *0405, *0408) and DRB1*01 subtypes (*0101, *0102) and DRB1*1001 contribute to disease susceptibility. As a common denominator, these positively associated DRB1 alleles share a sequence motif within the third hypervariable region (shared epitope, SE) w1, 2x. This epitope can be found in up to 80 90% of Caucasoid RA patients, encompasses amino acid positions 70 74 and is characterized by the amino acid cassette QRRAA, QKRAA or RRRAA w3x. Whereas the contribution of HLA class II markers to disease susceptibility is rather limited, it has become apparent in recent years that immunogenetic markers Department of Medicine IV, University of Leipzig, Leipzig, 1 Institute for Clinical Immunology, Department of Medicine III, University of Erlangen-Nürnberg, Erlangen and 2 Institute for Transplantation Diagnostics and Cell Therapeutics, Düsseldorf University Medical Center, Düsseldorf, Germany. Submitted 16 April 2002; revised version accepted 6 November 2002. Correspondence to: R. Wassmuth, Institute for Transplantation Diagnostics and Cell Therapeutics, University Medical Center, University of Du sseldorf, Moorenstrasse 5, Building 14.80, D-40225 Du sseldorf, Germany. E-mail: ralf.wassmuth@itz.uni-duesseldorf.de Rheumatology 42 ß British Society for Rheumatology 2003; all rights reserved

2of10 U. Wagner et al. substantially modulate the progression of joint destruction. As shown in a prospective study w4 7x, there is substantial evidence that patients carrying the SE, particularly on an HLA-DR4 haplotype, are at greatest risk of severe erosive disease w8x. The distinction between the amino acid cassettes QKRAA and QRRAA may also be of relevance to the clinical course w9x. While initial investigations of the linked DQB1 locus did not indicate any confounding effects for DQB1 alleles on the susceptibility to RA w10x or yielded controversial results w8, 11 16x, the predominant role of the DRB1 locus, particularly the SE, has been challenged recently by experimental studies in transgenic mice and subsequent investigations in humans w17, 18x. Functionally, this model of DQ predisposition involves the presentation of HLA-DRB1-derived autologous peptides by HLA-DQ antigens in the predisposition to and progression of RA w17, 19x. RA-predisposing DQ molecules include DQ3, designated DQ3RA, and DQ5, designated DQ5RA, heterodimers. According to the most recent modification of the DQ model, designated DQRA herein, heterodimers composed of DQB1*0301, *0302, *0303 or *0304 in combination with DQA1*0301 or *0302 and DQB1*0501 in combination with DQA1*0101 or *0104 respectively confer susceptibility, while the DQ7 molecule encoded by DQB1*0301u DQA1*0501, seen in linkage disequilibrium with DRB1 alleles encoding DR5 and DR6 heterodimers, is excluded. The exemption of DQ7 when positive for DQA1*0501, however, is not supported by the susceptibility effects seen for DRB1*1402 (Dw16), which is in strong linkage disequilibrium with DQ7, encoded by DQB1*0301uDQA1*0501. In addition to the predisposing effects of HLA-DR and -DQ antigens, protective effects have been ascribed to the DRB1 alleles *0103, *0402, *1102 and *1103 as well as *1301, *1302 and *1304. These DRB1 alleles are characterized by the amino acid cassette DERAA in the HVR3 region (third hypervariable region of DRB1) w18x. Gene dosage effects may be important, as in this study two doses of predisposing DQ alleles were strongly associated with RA, even in the presence of a single SE-positive haplotype, while the DERAA motif conferred dominant protection in DQ5-positive individuals. Moreover, the presence of rheumatoid factors (RF) and rheumatoid nodules are less frequent in DERAA-positive patients w20x. Thus, particular combinations or haplotypes of DR and DQ alleles, hereafter referred to as the interactive DRuDQ model, rather than the solitary influence of a single locus, may be relevant to susceptibility or resistance to RA. While previous studies were concerned with disease susceptibility, the contribution of HLA-DQ to the course of RA, in particular the progression of erosive disease, remains to be determined in a prospective fashion. A recent study investigating the protective effects of the HLA class II region showed a graded influence of DQ3RA vs DQ5RA not only on disease susceptibility but also on disease activity at study entry in early arthritis patients w21x. Taking advantage of our prospective study in patients with early arthritis, the aim of our analysis was to determine for the first time the differential role of HLA-DR and -DQ markers on the radiological progression of RA over a follow-up period of up to 4 yr. Patients and methods Study design The clinical data presented in this work are derived from a prospective observational study focusing on the identification of RA patients at high risk of rapid joint destruction early in the course of disease, and has been described in detail recently w22x. Approval of the study was obtained from the University of Leipzig ethics committee. We recruited patients fulfilling the American College of Rheumatology criteria for the diagnosis of RA, who had a disease duration of less than 2 yr before study inclusion and had not received disease-modifying antirheumatic drugs (DMARDs) before enrolment. The intended protocol for treatment included monotherapy with sulphasalazine or methotrexate combined with low-dose corticosteroid therapy if necessary to control clinical disease activity. Therapeutic biologicals inhibiting tumour necrosis factor a (TNF-a) or interleukin 1 (IL-1) were not available or employed during the study and follow-up period. Treatment could be modified in case of adverse side-effects or lack of efficacy. Patients were excluded from the study if lost to follow up (n=6). In addition to the collection of clinical data, hand and feet radiographs were taken for the documentation of radiological progression of joint destruction upon study entry, after 6 months and after 1, 2 and 4 yr of observation. Radiographs were scored according to Larsen s method by two independent radiologists, as published previously w22x, with low inter-observer variability. Immunoglobulin (Ig) M and IgA RF were determined upon study entry. Clinical characterization of the study population A detailed characterization of the study population has been published previously w22x and is summarized in Table 1. Of TABLE 1. Characteristics of the patient populations at study entry Patients followed for 24 months Patients followed for 48 months Number of patients 87 77 Sex (MuF) 14u73 13u64 Age at disease onset (yr) 54 (37.9 63.2) 52.6 (36.5 62.2) Disease duration before study 6.1 (3.4 11.3) 6.3 (3.7 11.5) entry (months) Patients positive for IgM RF 65.5% 63.6% IgM RF concentration (IUuml) 70.6 (0 206) 70.5 (0 206) Patients positive for RF IgA 25.3% 27.3% IgA RF concentration (IUuml) 0 (0 31.5) 0 (0 45.7) Patients with erosions at study 32.2% 33.8% entry Larsen score at study entry 0 (0 4.0) 0 (0 4.5) Larsen score at the end of the 14.0 (0 32.7) 24 (5 42) observation period Number of DMARDs 2 (1 2) 2 (1 2) Disease duration before study entry is the time between the first appearance of symptoms and enrolment in the study. Number of DMARDs is the number of successive treatment attempts with different DMARDs. Percentages of patients using the different medications refer to DMARD use at the time of the 24- or 48-month analysis wmedian (interquartile range)x.

HLA markers in rheumatoid arthritis 3of10 the entire study cohort, 87 patients had completed a 2-yr observation period and 77 patients had been followed for 4 yr. The median time between the first appearance of symptoms and study enrolment was 6.1 months winterquartile range (IQR) 3.4 11.3 monthsx and the median age at the onset of disease was 54 yr (37.9 63.2 yr). At initial presentation, 65.5% of the patients were positive for IgM RF (median concentration 70.6 IUuml, IQR 0 206 IUuml) and 18 patients (25.3%) were IgA RF-positive at study entry. Radiographs taken at initial presentation showed the presence of erosive changes in 28 patients (32.2%). HLA genotyping HLA-DRB1 genotyping was performed by oligonucleotide hybridization of enzymatically amplified DNA as described previously w8x. The method allowed low-resolution HLA-DRB1 genotyping comprising specificities DRB1*01 to DRB1*17 and the identification of alleles pertinent to the assessment of the SE and the DERAA-positive DRB1 alleles. In cases of ambiguity, direct sequencing of enzymatically amplified DNA was carried out. DQB1 alleles were determined following the XII International Histocompatibility Workshop and Conference protocol w23x. Using this approach, most of the DQB1 alleles (DQB1*0501-*0504, *0601-*0604, *0605u06, *02, 0301-*0304) could be differentiated. Differentiation between associated and non-associated DQ7 molecules was based on their differential linkage disequilibrium with respect to HLA-DR. Biometrical analyses Odds ratios (ORs) were calculated according to Woolf s method as cross-product ratios of a 2 32 contingency table wor=(a 3d)u(c 3b)x w24, 25x. Haldane s correction for the OR was used when either all patients were positive or all controls were negative for a particular specificity or allele w26x. The level of significance was assessed by Yates-corrected x 2 analysis or Fisher s exact test as appropriate. Correction of P values for each HLA specificity or allele was carried out according to the Bonferroni inequality method by multiplying the P values by the number of antigensualleles of the locus in question. In addition to the x 2 value, the upper and lower 95% confidence intervals (CI) are indicated. Stratification analysis (Mantel Haenszel test) was used to detect relative influences of individual HLA markers w27x. McNemar test was used to test the null hypothesis of an equal probability of a positive result under the adjacent pair of the test in the different HLA-stratified groups of patients, as carried out previously w28x. Differences in medians or means between groups were analysed with the Mann Whitney or t-test as appropriate. Correlations were evaluated using the Spearman rank correlation coefficient method w29x. Logistic regression analysis was performed by stepwise removal of independent variables using the likelihood ratio test as the criterion. The estimation of ORs (OR lr )in multiple logistic regression analysis for independent variables was estimated by OR lr =eb, where b is the regression coefficient w29x. SPSS (SPSS Inc., Chicago, IL, USA) was used for all calculations. Results Immunogenetic profile of the study cohort There was no significant difference in the distribution of HLA markers between the entire study cohort (n=87) followed for 24 months and the subgroup of patients followed for 48 months (Table 2). Genotyping for HLA TABLE 2. Distribution of HLA-DRB1, -DQB1 and RA-associated amino acid motifs Patients followed for 24 months (n=87) Patients followed for 48 months (n=77) HLA marker DRB1*01 27 (31.0%) 23 (29.9%) DRB1*04 36 (41.4%) 17 (42.8%) DRB1*0401 29 26 DRB1*0404 5 5 DRB1*0405 0 0 DRB1*0408 1 1 SE + 55 (63.2%) 59 (63.6%) SE +u+ 12 (13.8%) 10 (12.9%) SE+DR4 34 (39.1%) 31 (40.2%) DERAA 9 (10.3%) 7 (9.1%) DQRA 59 (67.8%) 53 (68.8%) DQ3RA 38 (43.7%) 35 (45.5%) DQ5RA 33 (37.9%) 23 (29.8%) The table shows absolute numbers and percentages for HLA DRB1 specificities, DRB1*04 subtypes and the distribution of DR- and DQ-derived sequence motifs. DRB1 and DQB1 revealed a substantial overlap between the presence of RA-associated DRB1 alleles and the presence of either DQRA or DQB1*0301uDQA1*0501. Nevertheless, a substantial fraction of the patient population (n=16) was either DQRA-positive or DQB1*0301uDQA1*0501-positive in the absence of the DRB1 SE, while all SE-positive patients were also DQRA-positive (McNemar test, P < 0.001). When DQB1*0301uDQA1*0501-positive individuals, i.e. DQ7- non-dr4, were excluded, only four patients of our study cohort remained classified discordantly according to the SE and the interactive DQuDR model (McNemar test, P=0.125). Thus, there was no significant difference in distribution between DQRA and the DRB1 SE. HLA markers and rheumatoid factors The production of IgM RF was neither associated with the presence of RA-associated DRB1 alleles (P=0.279) nor with SE-positive DR4 alleles (P=0.527) or DQRA (P=0.374). Interestingly, three of the nine DERAApositive patients were IgM RF-positive upon initial presentation compared with 65.4% (51u78) being seropositive in the rest of the study population (P=0.075). Only one of them remained IgM RF-positive after 2 yr of observation compared with 48.7% (38u78) in the DERAA-negative patients (P=0.038). Association of HLA markers with the progression of joint destruction Univariate comparisons of patients positive and negative for RA-associated DRB1 alleles indicated a more rapid course of joint destruction for predisposing HLA-DR markers. Patients expressing the RA-associated SE on a DRB1*01- or *04-positive allele had higher Larsen scores at all time-points analysed when compared with SE-negative patients (P=0.04 at initial presentation, P=0.01 after 1 yr of observation, P=0.001 after 2 yr and P=0.022 after 4 yr). In addition, SE-positive patients showed a faster rate of annual progression, i.e.

4of10 U. Wagner et al. a greater increase in Larsen score during the first 2 yr of observation using the values at study entry as baseline and calculating the increment during the following years (median Larsen score 12 vs 0, P=0.005). Most prominently, the presence of the RA-associated SE on a DRB1*04 allele was associated with even greater differences in Larsen score seen upon univariate comparison between SE+DRB1*04-positive and -negative patients (P=0.027 at initial presentation, P=0.004 after 1 yr of observation, P < 0.001 after 2 yr, and P=0.003 after 4 yr) (Fig. 1A). Similarly, the increase in Larsen score compared with baseline was greater during the first and second years of observation (median 4 vs 0, P=0.015, and median 5 vs 0, P=0.002 respectively) in SE+DRB1*04-positive compared with negative patients. The DRB1 protective motif DERAA was associated with a significantly lower Larsen score at initial presentation compared with DERAA-negative patients (median 0, IQR 0 0 vs median 0, IQR 0 7, P=0.046). Furthermore, none of the nine patients expressing the DERAA motif had evidence of erosive joint destruction, i.e. a Larsen score above 0 at initial presentation, while 36% (28 of 78) DERAA-negative patients did present with evidence of erosive disease at study entry. A significant difference in Larsen score was seen after 4 yr when DERAA-positive and -negative patients were compared (median 2 vs 27; IQR 0 17.75 vs 10 44, P=0.03) (Fig. 2). Considering the increase in Larsen score from baseline at study entry during the 4-yr period, DERAA-positive patients showed significantly FIG. 2. Means and standard errors of the means for DERAApositive patients (circles; n=9 for the time-points 0, 12 and 24 months and n=7 for 48 months) and compared with the rest of the study population (squares; n=78 for the time-points 0, 12, 24 and n=70 for 48 months). *P=0.08; **P=0.03. smaller differences in Larsen scores compared with DERRA-negative patients (median difference 2 vs 10, P=0.04). FIG. 1. Means and standard errors of the means of the Larsen score at the time-points 0, 12 and 24 months (n=87) and 48 months (n=77) for patients characterized by the immunogenetic markers indicated. (A) Patients classified according to the presence or absence of DR-derived markers. Levels of significance are given for the comparison of SE+DRB1*04-positive patients with SE-negative patients. (B) Comparison of Larsen scores of patients positive or negative for DQRA heterodimers. Levels of significance are given for the comparison of DQRA-positive patients with DQRA-negative patients. (C) Results for DQ3RApositive, DQ5RA-positive and DQRA-negative patients. Levels of significance are given for the comparison of DQ3RA-positive patients with DQRA-negative patients.

HLA markers in rheumatoid arthritis 5of10 When patients were divided according to the absence or presence of DQRA heterodimers, significant differences between patient groups in median Larsen scores were observed (median 0 vs 0, IQR 0 0 vs 0 8.75, P=0.046 at initial presentation; median 0 vs 8, IQR 0 11.5 vs 0 24.25, P=0.018 after 1 yr of observation; median 0 vs 19, IQR 0 20 vs 2.5 37, P=0.002 after 2 yr, and median 13 vs 29, IQR 0 10 vs 30 3.49, P=0.01 after 4 yr) (Fig. 1B). When DQ3RA- and DQ5RA-positive patients were compared separately with DQRA-negative patients, a more pronounced effect of DQ3RA than of DQ5RA on joint destruction was seen. DQ3RA-positive patients had significantly higher Larsen scores throughout the study compared with DQRA-negative patients (median 1 vs 0, IQR 0 10 vs 0 10, P=0.014 at initial presentation, median 11.5 vs 0, IQR 2 26 vs 0 11.5, P=0.006 after 1 yr of observation; median 27.5 vs 0, IQR 9 40 vs 0 20, P < 0.001 after 2 yr, and median 34 vs 13, IQR 16 51.75 vs 0 30, P < 0.001 after 4 yr of observation) (Fig. 1C). In contrast, no difference was seen between DQ5RA-positive and DQRA-negative patients at these time-points (P=0.549, 0.326, 0.179 and 0.956 respectively). Similarly, the differences in Larsen score between DQ3RA and DQ5RA did not reach significance after 2 and 4 yr of observation (median 27.5 vs 10, IQR 9 40 vs 0 34.75, P=0.084 after 2 yr and median 34 vs 18, IQR 16 51.75 vs 5.25 45, P=0.095 after 4 yr). When patients were divided according to the presence or absence of erosive disease at study entry, a significant influence of the RA-associated SE on a DRB1*04 allele and of DQ3RA was seen only in the patient group without erosive disease at study entry (n=59) (Fig. 3A and B). In patients in whom erosions were present at study entry, no significant effect was discerned for DR and DQ markers after 1 and 2 yr of observation. While a trend towards higher Larsen scores was seen both in SE+DRB1*04-positive and in DQ3RA-positive patients after 4 yr of observation (n=13), this difference did not reach significance (P=0.29 and P=0.1 respectively) (Fig. 3C and D). HLA markers and DMARD treatment and response The analysis of current DMARD therapy revealed no differences between immunogenetically defined patient groups. In SE+DRB1*04-positive patients, SE-positive and SE-negative patients, the median number of DMARDs used was 2 (range 1 2) and the use of methotrexate, either alone or in combination with cyclosporin A, sulphasalazine or hydroxychloroquine, was not significantly different between groups. In order to assess a potential influence of immunogenetic markers on treatment response as the underlying reason for different radiographic outcomes, patients were also classified as ACR 20 responders (n=51) or non-responders (n=36) under the current therapy according to the American College of Rheumatology 20% improvement criteria w30x. As the most significant difference between immunogenetically stratified groups was seen after 2 yr of observation, this follow-up time was used for comparison. The percentage of responders was not significantly different between SE+DRB1*04- positive, SE-positive and SE-negative patients (48.4, 57.1 and 60.5% respectively). Moreover, comparison of patients in the DQ3RA-positive, DQRA-positive and DQRA-negative groups indicated similar percentages of responders (48.6, 56.6 and 61.8%). Influence of DQ and DR markers on the progression of joint destruction after stratification for SE+DRB1 alleles and DQRA In order to dissect SE- and DQRA-dependent effects on radiographic progression, patients were stratified for the presence of either the SE or DQRA. As all SE+ DRB1-positive patients are also positive for DQRA, no additional predispositional effects were discerned for DQRA. The inclusion of patients expressing DQB1*0301u DQA1*0501 heterodimers (DQ7) in the DQRA-positive patient group has been controversial. When SE-negative, DQRA-positive patients, including DQB1*0301u DQA1*0501-positive patients (at 2 yr n=16; at 4 yr n=14), were selected and compared with all other patients, significantly lower median Larsen scores were seen after 2 yr (0 vs 18, P=0.026), whereas no significant difference could be observed after 4 yr of observation (11 vs 25, P=0.18). DQ7-positive patients had significantly lower values of the Larsen score at all time-points when compared with SE+DRB1*04-positive patients. Furthermore, this group also showed a trend towards lower Larsen scores compared with SE-negative patients. The presence of DQRA, i.e. DQ5 and DQ3 after exclusion of DQB1*0301uDQA1*0501 in SE-negative patients (n=4) was also not associated with higher Larsen scores at any of the time-points analysed (P=0.332 at initial presentation, P=0.763 after 2 and P=0.172 after 4 yr of observation). Similarly, SE-positivity among DQRA-positive patients had no significant influence on the Larsen scores observed (P=0.805 at initial presentation, P=0.493 after 2 and P=0.977 after 4 yr of observation). In the DQRA-positive group, patients positive for SE+ DRB1*04 alleles, however, did have higher Larsen scores after 2 and 4 yr of observation (P=0.04 and P=0.027 respectively) compared with DQRA-positive SE+DRB1*04-negative patients. Gene dosage effect for SE-positive DRB1 alleles and DQRA haplotypes When patients expressing SE-positive HLA DRB1 alleles on both haplotypes were compared with the rest of the study population, they were found to have higher Larsen scores after 2 yr (P=0.013) and 4 yr (P=0.016) of observation (Table 3). Within the group of SE-positive patients, there was also a non-significant trend for SE compound homozygotes to reach higher Larsen scores after 2 yr (median 32 vs 19, IQR 13 44.5 vs

6of10 U. Wagner et al. FIG. 3. Time course of Larsen scores (mean and standard error of the mean) for patients without erosive disease at study entry (n=59, A and B) and for patients in whom erosions were already present at the initial radiograph (n=28, C and D). (A) Comparison of Larsen scores for initially non-erosive patients, who were positive (full circles; n=17 for the time-points 0, 12 and 24 months and n=8 for the time-point 48 months) or negative (squares; n=23 for the time-points 0, 12 and 24 months and n=18 for the time-point 48 months) for an RA-associated DRB1*04 allele. (B) Comparison of DQ3RA-positive (triangles; n=19 for the time-points 0, 12, 24 and n=10 for 48 months) and DQ3RA-negative (diamonds; n=40 for the time-points 0, 12, 24 and n=29 for 48 months) patients without erosive disease at study entry. (C) Patients showing erosive disease at study entry who were positive (circles; n=17 for the time-points 0, 12 and 24 months and n=7 for 48 months) and negative (squares; n=11 for the time-points 0, 12 and 24 months and n=6 for 48 months) for an RA-associated DRB1*04 allele. (D) Patients with erosive disease at study entry who were either DQ3RA-positive (triangles; n=19 for the time-points 0, 12, 24 and n=9 for 48 months) or DQ3RA-negative (diamonds; n=9 for the time-points 0, 12, 24 and n=4 for 48 months). 2 32.75, P=0.072), which reached significance after 4 yr of observation (median 51 vs 25, IQR 20 65.25 vs 7 45.5, P=0.031). Larsen scores in DQRA-homozygous patients were significantly different from the rest of the study population only after 4 yr (P=0.043) (Table 3).

HLA markers in rheumatoid arthritis 7of10 TABLE 3. Effects of the DRB1 SE and compound homozygosity for immunogenetic markers on Larsen scores wmedian (IQR)x reached during the observation period SE+DRB1-negative SE+DRB1-positive n Larsen score n Larsen score P At baseline 32 0 (0 0) 55 0 (0 7.75) 0.04 After 1 yr 32 0 (0 13) 55 8 (0 26) 0.01 After 2 yr 32 0 (0 20) 55 20 (4.25 37) 0.001 After 4 yr 28 19 (0 32) 49 30 (4.25 47.5) 0.022 SE+DRB1 allele +u2 or 2u2 SE+DRB1 allele +u+ At baseline 75 0 (0 3.5) 12 1 (0 8.5) 0.27 After 2 yr 75 10 (0 28) 12 32.5 (13 44.5) 0.013 After 4 yr 67 23 (2 39) 10 43.5 (16 69) 0.016 DQRA dimers +u2 or 2u2 Relative risks for severe joint destruction associated with HLA-DR and -DQ markers For the calculation of relative risks conferred by SE+DRB1*04 alleles, DQRA, DQ3RA and DQ5RA, patients were stratified into two groups according to the severity of erosive disease using the Larsen scores reached by one-third of the study population wls 027.7 after 2 yr of observation (n=29) and LS 035.0 after 4 yr (n=28)x as cut-off, as published previously w22, 31, 32x (Table 4). After 2 yr, DQRA (OR 6.57) was found to confer a marginally higher risk of severe joint destruction than the presence of the SE (OR 5.83) or of SE+DRB1*04 alleles (OR 4.3) (Table 4). After 4 yr of observation, however, the highest risk was associated with the presence of DQ3RA heterodimers (OR 4.35), followed by SE+DRB1*04 alleles (OR 3.86) and DQRA DQRA dimers +u+ At baseline 72 0 (0 2) 15 2 (0 14.5) 0.092 After 2 yr 72 10 (0 28.5) 15 28 (9.75 39.25) 0.081 After 4 yr 64 23 (1 39) 13 34 (11.5 57.7 0.043 The table shows Larsen scores reached after the observation period indicated in SE-positive and -negative patients and in compound homozygotes for the RA-associated SE or DQRA haplotypes (OR 3.8). Nevertheless, confidence intervals overlapped and the distribution of SE+DRB1*04 alleles and DQ3RA-positive alleles was not significantly different neither in the patient group with the most severe joint destruction nor in the group with Larsen scores below the cut-off (P=0.5 and P=0.5 for DQ3RA vs SE+DRB1*04 in both groups) in the adjacent pair test (McNemar test). Similarly, no significant difference in the distribution of DQRA and SE was seen (P=0.5 and P=0.5 for DQRA vs SE in both groups). Subsequently, patients were stratified for the presence or absence of SE or SE+DRB1*04 to assess the contribution of DQRA and vice versa. The presence of DQRA did not confer additional risk in SE- or SE+ DRB1*04-positive patients. Similarly, the presence of DQ3RA did not confer additional risk in SE+ TABLE 4. Association of immunogenetic markers with the development of severe erosive disease in recent-onset RA patients HLA marker Larsen score < 27 (n=58) Larsen score 027 (n=29) OR 95% CI P (two-sided) 24 months of observation (87 patients) SE 30 (51.7%) 25 (86.2%) 5.83 1.8, 18.88 0.004 SE+DRB1*04 16 (27.6%) 18 (62.1%) 4.3 1.67, 11.06 0.002 DQRA 33 (56.9%) 26 (89.7%) 6.57 1.78, 24.17 0.005 DQ3RA 19 (32.8%) 19 (65.5%) 3.9 1.52, 10.0 0.007 DQ5RA5 17 (29.3%) 10 (34.5%) 1.27 0.49, 3.29 0.806 Larsen score < 35 (n=49) Larsen score 035 (n=28) 48 months of observation (77 patients) SE 28 (57.1%) 22 (78.6) 2.75 0.95, 7.98 0.099 SE+DRB1*04 14 (28.6%) 17 (60.7%) 3.86 1.45, 10.29 0.012 DQRA 30 (61.2%) 24 (85.7%) 3.8 1.14, 12.67 0.046 DQ3RA 16 (32.7%) 19 (67.9%) 4.35 1.6, 11.75 0.006 DQ5RA5 16 (32.7%) 8 (28.6%) 0.83 0.3, 2.28 0.917 Patients with recent-onset RA were grouped according to the Larsen scores reached after 24 and 48 months of observation (above or below the Larsen score reached by two-thirds of the study population), and sensitivity and specificity were calculated for immunogenetic markers. Comparisons were made for the presence or absence of the DRB1 SE (SE), the DRB1 shared epitope on DR4 (SE+DRB1*04), DQRA, DQ3RA and DQ5RA.

8of10 U. Wagner et al. DRB1*04-positive patients. The presence of DQ3RA in SE-positive patients was associated with an OR of 2.25 (95% CI 0.7 7, P=0.254) for more severe joint destruction after 24 months and an OR of 3.4 (95% CI 0.9 11.8; P=0.093) after 48 months of observation. SE+DRB1*04 alleles, but not the presence of the SE per se, were associated with a trend to increased ORs for higher Larsen scores in DQRA-positive patients after 2 and 4 yr of observation (OR=2.4, P=0.182 and OR=2.8, P=0.132). With regard to gene dosage, neither SE compound homozygosity nor the presence of two DQRA-positive heterodimers was associated with a significantly increased relative risk of reaching a Larsen score 027.7 after 2 or 035.0 after 4 yr of observation compared with the remaining study population. Modelling In order to account for the simultaneous contributions of different HLA markers, modelling using multiple logistic regression analysis was carried out. Radiological evidence of joint destruction, measured with the Larsen score (cut-off 27.7 and 35.0 for the 24- and 48- month periods respectively), was used as the dependent variable. When HLA-DR specificities (DRB1*01, DRB1*17) were used as independent variables, a contribution to the model equation was seen for DR4 after 2 yr (OR lr =3.64, P=0.007) and after 4 yr (OR lr =2.5, P=0.07) for the development of severe destructive RA. Considering DQB1 alleles alone, DQB1*0301 contributed only after 2 yr (OR lr =2.45, P=0.098). After 4 yr, none of the DQB1 alleles remained variable in the model. For the joined inclusion of DRB1 specificities and DQB1 alleles, DRB1*04 (OR lr =3.63, P=0.007) contributed to the model only after 24 month and after 48 months (OR lr =2.51, P=0.068). When derived DR and DQ markers (SE+DRB1*04, QKRAA, QRRAA, RRRAA and DERAA as well as DQ3RA and DQ5RA) were included, the model equation was formed by QKRAA (OR lr =5.13, P=0.003) and QRRAA (OR lr =3.87, P=0.013) after 2 yr (overall prediction of 70.1%, P=0.014), whereas only SE+DRB1*04 (OR lr =3.86, P=0.007) was necessary to model the development of severe joint destruction after 4 yr (overall prediction of 67.5%, P=0.006; positive prediction of 60.7% and negative predictive value 71.4%). Discussion Although advances have been made in the understanding of immunological factors in the pathogenesis of RA, there are still considerable problems in the study of the epidemiological aspects of the role of immunogenetic factors of disease susceptibility, the course of disease and prognostication. Despite intriguing immunopathogenic suggestions to explain the role of HLA class II molecules in RA, the dissection of HLA class II-dependent susceptibility to RA has been hampered substantially by linkage disequilibrium within the HLA-DR and -DQ region. Moreover, current models of HLA-encoded susceptibility are not unequivocally supported by epidemiological data and it may be extremely difficult to decide with certainty whether either HLA class II loci alone or haplotypic combinations influence disease susceptibility, the course of disease and the outcome. In this immunogenetic analysis of a prospectively followed study cohort with recent-onset RA, we focused on the comparison of the contribution of HLA-DR and -DQ to the development of erosive joint destruction as a most informative outcome measure w33x. Consistently with previous findings, our results confirm a highly significant influence of HLA-dependent markers on the progression of joint destruction. This influence was already apparent at initial presentation and was sustained after both 2 and 4 yr of observation. The presence of the DRB1 SE was associated with earlier occurrence of the first erosive changes, which is illustrated by the increased Larsen scores in SE-positive patients at baseline and by the greater increase in the score in previously non-erosive patients during the first and second years of observation. The results of the multivariate analysis showed, however, that this association reflects not merely a greater capacity to develop erosions but also faster progression of joint destruction throughout the observation period. Consistent with a graded influence of RA-associated HLA-DR markers, the most marked contribution to the progression of erosive joint disease was seen for SE+DRB1*04 alleles when the DRB1 locus was considered, while the presence of the SE on a DRB1*01 allele conferred little additional risk. Similarly, separate analysis of the HLA-DQ locus showed a highly significant influence of DQ3RA and a less pronounced effect, if any, for the DQ5RA heterodimers. In line with previous observations w34x, however, results presented here clearly rule out an influence of DQ7 alleles that are not in linkage disequilibrium with SE-positive DRB1*04 alleles. The influence of HLA markers on the progression of erosive joint disease was dependent on the presence or absence of erosive disease at study entry. The significant influence of the RA-associated SE on a DRB1*04 allele and of DQ3RA was most prominent in patients without erosive disease at study entry. Thus, HLA genotyping may have its greatest predictive value in patients who do not show any erosive disease at onset or during the very early course of disease. It has been argued that one of the major shortcomings of the SE model one that can be overcome by the interactive DRuDQ model is the inability of the former to explain the differences in disease susceptibility and severity associated with the different SE-positive alleles. In this prospective analysis, however, a graded effect of DQ3RA vs DQ5RA was also seen. DQ3RA had a pronounced effect on joint destruction, but this was not observed in DQ5RA-positive patients. Thus, the influence of HLA-DQ on erosive joint destruction was also subject to hierarchical effects. Most importantly, these gradations followed the pattern observed for HLA-DRB1*04 and *01, as expected because of the linkage disequilibrium. In contrast to the SE model,

HLA markers in rheumatoid arthritis 9of10 however, this graded influence of HLA-DQ was not dose-dependent, as seen for the SE. In the presence of the SE, no additional effects could be discerned for RA-associated DQ molecules, which argues against a prominent epidemiological role for HLA-DQ in the destructive process. Moreover, in the absence of the SE, the presence of DQRA did not exert a significant influence on erosive joint disease. Although clearly protective effects of the DERAA motif were observed, the influence of DERAA-positive DRB1 alleles on severe joint destruction is limited because only a small fraction of patients will be DERAApositive. Moreover, in longstanding RA (5 30 yr), the protective effect of the DERAA motif may be apparent only in the absence of the SE w35x. The findings were supported by modelling the HLAdependent influence using different sets of HLA markers as independent variables to predict the development of a Larsen score above the tercentile of the total study cohort. Significant positive influences were detected only for DR, i.e. QKRAAuQRRAA and SE+DRB1*04, when DR- and DQ-derived markers were considered in the model. The progression of joint destruction in RA may not only be subject to the influence of immunogenetic markers and RF and the presence of early erosive disease, but may also be modulated by treatment, as recently shown for DMARD and TNF-a blockade w36, 37x. Therefore, we assessed the distribution of treatment modalities as a potential confounding variable with respect to the progression of joint destruction. Because we found no significant association of any of the immunogenetic markers analysed with the magnitude of the response to DMARD treatment, it appears highly unlikely that differences in treatment contributed to the differential outcome of erosive joint disease. Nevertheless, it is generally possible that the immunogenetic markers that appear to modulate radiological outcome are in fact markers of treatment response. Clarification of this issue, however, would require a study design different from that used here. From this study, it is concluded that the analysis of HLA-DQ does not provide information additional to that given by the study of HLA-DR, including the determination of the SE and the DERAA motif, in order to predict the development of severe progressive joint destruction. Moreover, the gradation of influence on joint destruction is similar for markers of both HLA class II loci. Acknowledgements This work was supported by grants from the Deutsche Forschungsgemeinschaft (SFB 263), the Bundesministerium für Bildung und Wissenschaft (Verbundprojekt Struktur und Funktion des Bindegewebes bei rheumatischen und Autoimmunerkrankungen ; grant 01 ZZ 9103 u 14), the Interdisziplinäres Zentrum für Klinische Forschung Leipzig, Teilprojekt A 15 and the Kompetenznetzwerk Entzündlich-rheumatische Systemerkrankungen, Teilprojekt C2.7. References 1. Gregersen PK, Silver J, Winchester RJ. The shared epitope hypothesis. An approach to understanding the molecular genetics of susceptibility to rheumatoid arthritis. Arthritis Rheum 1987;30:1205 13. 2. Cairns JS, Curtsinger JM, Dahl CA, Freeman S, Alter BJ, Bach FH. Sequence polymorphism of HLA DR beta 1 alleles relating to T-cell-recognized determinants. Nature 1985;317:166 8. 3. Weyand CM, Hicok KC, Conn DL, Goronzy JJ. The influence of HLA-DRB1 genes on disease severity in rheumatoid arthritis. Ann Intern Med 1992;117:801 6. 4. Gough A, Faint J, Salmon M et al. Genetic typing of patients with inflammatory arthritis at presentation can be used to predict outcome. Arthritis Rheum 1994;37:1166 70. 5. Toda Y, Minamikawa Y, Akagi S et al. Rheumatoidsusceptible alleles of HLA-DRB1 are genetically recessive to non-susceptible alleles in the progression of bone destruction in the wrists and fingers of patients with RA. Ann Rheum Dis 1994;53:587 92. 6. Boki KA, Panayi GS, Vaughan RW, Drosos AA, Moutsopoulos HM, Lanchbury JS. HLA class II sequence polymorphisms and susceptibility to rheumatoid arthritis in Greeks. The HLA-DR beta shared-epitope hypothesis accounts for the disease in only a minority of Greek patients. Arthritis Rheum 1992;35:749 55. 7. Salvarani C, Macchioni P, Mantovani W, Portioli I. Polymyalgia rheumatica and seronegative rheumatoid arthritis: some considerations based on a northern Italian population. J Rheumatol 1993;20:756 7. 8. Wagner U, Kaltenhauser S, Sauer H et al. HLA markers and prediction of clinical course and outcome in rheumatoid arthritis. Arthritis Rheum 1997;40:341 51. 9. Weyand CM, McCarthy TG, Goronzy JJ. Correlation between disease phenotype and genetic heterogeneity in rheumatoid arthritis. J Clin Invest 1995;95:2120 6. 10. Nepom GT. Structural and genetic features of human leukocytic antigen class II elements associated with rheumatoid arthritis. Am J Med 1988;85:12 3. 11. Sansom DM, Bidwell JL, Maddison PJ, Campion G, Klouda PT, Bradley BA. HLA DQ alpha and DQ beta restriction fragment length polymorphisms associated with Felty s syndrome and DR4-positive rheumatoid arthritis. Hum Immunol 1987;19:269 78. 12. Stephens HA, Sakkas LI, Vaughan RW, Teitsson I, Welsh KI, Panayi GS. HLA-DQw7 is a disease severity marker in patients with rheumatoid arthritis. Immunogenetics 1989;30:119 22. 13. Clarkson R, Bate AS, Grennan DM et al. DQw7 and the C4B null allele in rheumatoid arthritis and Felty s syndrome. Ann Rheum Dis 1990;49:976 9. 14. McCusker CT, Reid B, Green D, Gladman DD, Buchanan WW, Singal DP. HLA-D region antigens in patients with rheumatoid arthritis. Arthritis Rheum 1991; 34:192 7. 15. Singal DP, Green D, Reid B, Gladman DD, Buchanan WW. HLA-D region genes and rheumatoid arthritis (RA): importance of DR and DQ genes in conferring susceptibility to RA. Ann Rheum Dis 1992;51:23 8. 16. Hillarby MC, Ollier WE, Davis M, Davidson J, Sanders PA, Grennan DM. Unusual DQA-DR haplotypes in rheumatoid vasculitis. Br J Rheumatol 1993;32:93 6. 17. Zanelli E, Krco CJ, Baisch JM, Cheng S, David CS. Immune response of HLA-DQ8 transgenic mice to peptides from the third hypervariable region of

10 of 10 U. Wagner et al. HLA-DRB1 correlates with predisposition to rheumatoid arthritis. Proc Natl Acad Sci USA 1996;93:1814 9. 18. van der Horst-Bruinsma IE, Visser H, Hazes JM, Breedveld FC et al. HLA-DQ-associated predisposition to and dominant HLA-DR-associated protection against rheumatoid arthritis. Hum Immunol 1999;60:152 8. 19. Zanelli E, Krco CJ, David CS. Critical residues on HLA-DRB1*0402 HV3 peptide for HLA-DQ8-restricted immunogenicity: implications for rheumatoid arthritis predisposition. J Immunol 1997;158:3545 51. 20. Seidl C, Korbitzer J, Badenhoop K et al. Protection against severe disease is conferred by DERAA-bearing HLA-DRB1 alleles among HLA-DQ3 and HLA-DQ5 positive rheumatoid arthritis patients. Hum Immunol 2001;62:523 9. 21. Vos K, van Der Horst-Bruinsma IE, Hazes JM et al. Evidence for a protective role of the human leukocyte antigen class II region in early rheumatoid arthritis. Rheumatology 2001;40:133 9. 22. Kaltenhauser S, Wagner U, Wassmuth R et al. Immunogenetic markers and seropositivity predict radiological progression in early rheumatoid arthritis independent of disease activity. J Rheumatol 2001;28:734 44. 23. Bignon JD, Fernandez-Vina MA, Cheneau ML et al. HLA-DNA class II typing by PCR-SSOP: 12th International Workshop experience. In: Charron D, ed. Genetic diversity of HLA, functional and medical implication. Proceedings of the Twelfth International Histocompatibility Workshop and Conference. Paris: EDK, 1997:21 5. 24. Miettinen O. Estimability and estimation in case-referent studies. Am J Epidemiol 1976;103:226 35. 25. Woolf B. On estimating the relation between blood group and disease. Ann Hum Genet 1955;19:251 3. 26. Haldane S. The estimated significance of the logarithm of a ratio of frequencies. Ann Hum Genet 1956;20:309 11. 27. Mantel N, Haenzsel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer Inst 1959;22:719 48. 28. Michelsen B, Wassmuth R, Ludvigsson J, Lernmark A, Nepom GT, Fisher L. HLA heterozygosity in insulin-dependent diabetes is most frequent at the DQ locus. Scand J Immunol 1990;31:405 13. 29. Rosner BA. Fundamentals of biostatistics. Boston: Duxbury Press, 1982. 30. Felson DT, Anderson JJ, Boers M et al. American College of Rheumatology. Preliminary definition of improvement in rheumatoid arthritis. Arthritis Rheum 1995;38:727 35. 31. Mottonen T, Paimela L, Leirisalo-Repo M, Kautiainen H, Ilonen J, Hannonen P. Only high disease activity and positive rheumatoid factor indicate poor prognosis in patients with early rheumatoid arthritis treated with sawtooth strategy. Ann Rheum Dis 1998;57:533 9. 32. Fex E, Jonsson K, Johnson U, Eberhardt K. Development of radiographic damage during the first 5 6 yr of rheumatoid arthritis. A prospective follow-up study of a Swedish cohort. Br J Rheumatol 1996;35:1106 15. 33. Emery P, Symmons D. What is early rheumatoid arthritis? Definition and diagnosis? In: Woolf AD, van Riel PLCM, eds. Bailliere s Clinical Rheumatology. London: Bailliere Tindall, 1997:13 26. 34. de Vries N, van Elderen C, Tijssen H, van Riel PL, van de Putte LB. No support for HLA-DQ encoded susceptibility in rheumatoid arthritis. Arthritis Rheum 1999; 42:1621 7. 35. Mattey DL, Hassell AB, Plant MJ et al. The influence of HLA-DRB1 alleles encoding the DERAA amino acid motif on radiological outcome in rheumatoid arthritis. Rheumatology 1999;38:1221 7. 36. O Dell JR, Nepom BS, Haire C et al. HLA-DRB1 typing in rheumatoid arthritis: predicting response to specific treatments. Ann Rheum Dis 1998;57:209 13. 37. Lipsky PE, van der Heijde DM, St Clair EW et al. Infliximab and methotrexate in the treatment of rheumatoid arthritis. N Engl J Med 2000;343:1594 602.