Immunogenetic risk and protective factors for juvenile dermatomyositis in Caucasians

Authors


Abstract

Objective

To define the relative importance (RI) of class II major histocompatibility complex (MHC) alleles and peptide binding motifs as risk or protective factors for juvenile dermatomyositis (DM), and to compare these with HLA associations in adult DM.

Methods

DRB1 and DQA1 typing was performed in 142 Caucasian patients with juvenile DM, and the results were compared with HLA typing data from 193 patients with adult DM and 797 race-matched controls. Random Forests classification and multiple logistic regression were used to assess the RI of the HLA associations.

Results

The HLA–DRB1*0301 allele was a primary risk factor (odds ratio [OR] 3.9), while DQA1*0301 (OR 2.8), DQA1*0501 (OR 2.1), and homozygosity for DQA1*0501 (OR 3.2) were additional risk factors for juvenile DM. These risk factors were not present in patients with adult DM without defined autoantibodies. DQA1 alleles *0201 (OR 0.37), *0101 (OR 0.38), and *0102 (OR 0.51) were identified as novel protective factors for juvenile DM, the latter 2 also being protective factors in adult DM. The peptide binding motif DRB1 9EYSTS13 was a risk factor, and DQA1 motifs F25, S26, and 45(V/A)W(R/K)47 were protective. Random Forests classification analysis revealed that among the identified risk factors for juvenile DM, DRB1*0301 had a higher RI (100%) than DQA1*0301 (RI 57%), DQA1*0501 (RI 42%), or the peptide binding motifs. In a logistic regression model, DRB1*0301 and DQA1*0201 were the strongest risk and protective factors, respectively, for juvenile DM.

Conclusion

DRB1*0301 is ranked higher in RI than DQA1*0501 as a risk factor for juvenile DM. DQA1*0301 is a newly identified HLA risk factor for juvenile DM, while 3 of the DQA1 alleles studied are newly identified protective factors for juvenile DM.

Juvenile dermatomyositis (DM) is a rare systemic autoimmune disease characterized by chronic inflammation of skeletal muscle and skin (1). Although the pathogenesis of juvenile DM is unknown, growing evidence suggests that there is an association with environmental exposures in genetically susceptible individuals (2, 3). Juvenile-onset DM is thought to share certain features with adult DM, including characteristic skin rashes as well as cellular and complement-mediated perivascular inflammation of blood vessels in the muscles. However, juvenile DM is thought to differ from adult DM in other ways, including increased frequencies of dystrophic calcification, ulceration, and vasculopathy with angiogenesis (4, 5).

The extended major histocompatibility complex (MHC) European ancestral haplotype HLA–A*0101;B*0801;Cw*0701;DRB1*0301;DQA1*0501;DQB1*0201 has been found to be the strongest risk factor for adult-onset myositis and other autoimmune diseases (6–8). The class II MHC allele HLA–DQA1*0501 has been identified as a major risk factor for juvenile DM in Caucasian patients, in small studies in which the primary focus was the distribution of DQA1 alleles (9–11).

In the present study, we examined the class II MHC alleles HLA–DRB1 and HLA–DQA1 and peptide binding motifs as risk or protective factors in the largest population of patients with juvenile DM studied to date, in order to define their relative importance (RI) through multivariable statistical approaches, and to compare these risk and protective factors with those in adult DM.

PATIENTS AND METHODS

Patients.

One hundred twenty-seven Caucasian patients with juvenile-onset DM (as defined by disease onset before age 18 years) and 15 Caucasian patients with juvenile DM in association with another autoimmune condition were included in this study. All patients fulfilled the Bohan and Peter criteria (12) for probable or definite DM. All subjects were enrolled in investigational review board–approved National Institutes of Health (NIH) Clinical Center and US Food and Drug Administration (FDA) protocols. The diagnosis of juvenile DM overlap syndrome was defined by the presence of the Bohan and Peter criteria for DM (12) as well as the criteria for another connective tissue disease. Associated autoimmune diagnoses included systemic lupus erythematosus (5 patients), scleroderma (4 patients), diabetes mellitus (2 patients), juvenile rheumatoid arthritis (2 patients), immune thrombocytopenic purpura (1 patient), and ulcerative colitis (1 patient).

HLA typing data from patients with juvenile DM were compared with the typing data from 797 Caucasian control subjects and 193 Caucasian patients with adult DM (as defined by disease onset at age ≥18 years) who were referred to the NIH Clinical Center, the FDA, and other referral centers (University of Texas Houston Health Science Center, Mayo Clinic, Rochester, and University of Pittsburgh Medical Center) between 1983 and 2002 (6).

HLA typing.

Purified genomic DNA was utilized for high-resolution HLA typing using commercial reagents for polymerase chain reaction–mediated sequence-specific oligonucleotide probe hybridization (GenoVision, West Chester, PA) and sequence-specific primer techniques (Dynal Biotech, Lafayette Hill, PA). High-resolution (i.e., allele-specific) HLA typing was utilized to identify specific susceptibility factors within larger groups of alleles that were defined at lower resolutions (i.e., group-specific analyses). Typing for 28 DRB1 and 14 DQA1 high-resolution alleles was performed in 108 patients, and typing for low-resolution alleles was performed in all 142 patients with juvenile DM. HLA allele assignments were consistent with those reported by the World Health Organization Nomenclature Committee for Factors of the HLA System (Thirteenth International Histocompatibility Workshop, Victoria, British Columbia, Canada). HLA results were generally reported in high resolution, except when significant differences between groups were detected only with low-resolution typing and in certain peptide binding motif analyses.

HLA peptide binding motifs were defined as follows: for DRB1, 9EYSTS13 (DRB1*03, *11, *13, and *14), for DQA1, F25 (DQA1*0103, *0201, and *0601), S26 (DQA1*01, *02, *04, and *06), and 45(V/A)W(R/K)47 (DQA1*01 and *02) (13).

Statistical analysis.

Analyses were performed using GraphPad InStat software for Windows 95, version 3.00 (GraphPad, San Diego, CA) and the SAS System for Windows, version 8.02 (SAS Institute, Cary, NC). Fisher's exact test was used to calculate P values for 2 × 2 tables. The rate of allele positivity was determined by the number of allele-positive subjects divided by the total number of subjects for whom complete high- or low-resolution HLA data were available at a given locus. P values were adjusted for multiple comparisons using Holm's procedure (14). P values were determined as significant when the adjusted P values were less than or equal to 0.05. Alleles that were of higher or lower frequency in patients with juvenile DM compared with controls, prior to correction for multiple comparisons, were termed possible risk or protective factors, respectively.

The RI of each individual high-resolution HLA–DRB1 and HLA–DQA1 allele and peptide binding motif as a genetic predisposing or protecting factor for juvenile DM was estimated using a statistical learning machine, utilizing the Random Forests algorithm (see http://stat-www.berkeley.edu/users/breiman/RandomForests/) (6). All HLA–DRB1 alleles, HLA–DQA1 alleles, and peptide binding motifs identified among the patients with juvenile DM and control subjects were simultaneously analyzed by Random Forests modeling. The Random Forests classification analyses were performed using the R programming language (R software, version 2.1.0 [2005-04-18, ISBN 3-9000-07-0], from the R Foundation for Statistical Computing). All allelic variables in the test population were ranked by their RI score, which indicates their ability to discriminate between patients with juvenile DM and control subjects. The number of decision trees was set to a value whereby the RI estimates converged; we confirmed that models containing 500, 1,000, 5,000, 10,000, 15,000, and 20,000 independent classification trees demonstrated identical results. Individual decision trees were constructed from combined case and control training data sets, utilizing bootstrap sampling with replacement (approximately two-thirds of the data sets were sampled per tree) and random feature selection. Training and test data are continually and randomly reutilized in the construction of individual decision trees.

Logistic regression analyses were performed as an independent method of confirming the associations identified by the Random Forests modeling. The top 10 alleles or motifs from the Random Forests classification were used in the logistic regression models. The multiplicative models of allele interactions were analyzed using PROC LOGISTIC in SAS. To determine strengths of association, interaction of risk factors, and the role of linkage disequilibrium, an allele interaction analysis was also performed (15).

RESULTS

Associations of HLA–DRB1 and DQA1 with juvenile DM.

HLA–DRB1*0301 (odds ratio [OR] 3.9, P < 0.0001), DQA1*0301 (OR 2.8, P < 0.0001), and DQA1*0501 (OR 2.1, P < 0.0001) were identified as risk factors for juvenile DM in Caucasian patients (Table 1). DRB1*1201 was found to be a possible risk factor for juvenile DM (OR 3.3, P = 0.03), but was not significant after correction for multiple comparisons.

Table 1. Summary of immunogenetic risk factors in Caucasian patients with juvenile dermatomyositis (DM)*
HLA class II allele or motifJuvenile DM, %Controls, %POR (95% CI)
  • *

    Carriage rates were determined by the number of allele-positive subjects divided by the number of subjects for whom complete HLA data were available at a given locus. OR = odds ratio; 95% CI = 95% confidence interval.

  • DRB1 high-resolution typing data were available for 108 patients with juvenile DM and 568 controls, DRB1 low-resolution for 142 patients with juvenile DM and 797 controls, DQA1 high-resolution for 195 patients with juvenile DM and 353 controls, and DQA1 low-resolution for 195 patients with juvenile DM and 542 controls. Results of DRB1 typing for determination of peptide binding motif were available for 142 patients with juvenile DM and 797 controls.

  • P ≤ 0.05 after Holm's adjustment for multiple comparisons (using family-wise error rates of 5%).

HLA–DRB1 allele    
 DRB1*030150.921.0<0.00013.9 (2.6–6.0)
 DRB1*12015.61.80.0303.3 (1.2–9.2)
HLA–DQA1 allele    
 DQA1*030131.814.5<0.00012.8 (1.8–4.2)
 DQA1*050162.144.2<0.00012.1 (1.4–3.0)
 DQA1*0501 homozygosity9.43.10.0143.2 (1.3–7.9)
HLA–DRB1 peptide binding  motif 9EYSTS1375.460.20.00052.0 (1.3–3.0)

HLA–DQA1*0101 (OR 0.38, P < 0.0001), DQA1*0201 (OR 0.37, P < 0.0001), and DQA1*0102 (OR 0.51, P < 0.0004) were identified as protective factors for juvenile DM in Caucasian patients (Table 2). DRB1*01 (OR 0.51, P = 0.006), DRB1*07 (OR 0.59, P = 0.031), and DRB1*1501 (OR 0.49, P = 0.031) were found to be possible protective factors (P > 0.05, after correction for multiple comparisons).

Table 2. Summary of immunogenetic protective factors in Caucasian patients with juvenile DM*
HLA class II allele or motifJuvenile DM, %Controls, %POR (95% CI)
  • *

    Carriage rates were determined by the number of allele-positive subjects divided by the number of subjects for whom complete HLA data were available at a given locus. See Table 1 for definitions.

  • DQA1 high-resolution typing data were available for 195 patients with juvenile DM and 353 controls, DQA1 low-resolution (for determination of peptide binding motifs) for 195 patients with juvenile DM and 542 controls, DRB1 high-resolution for 108 patients with juvenile DM and 568 controls, and DRB1 low- resolution for 142 patients with juvenile DM and 797 controls.

  • P ≤ 0.05 after Holm's adjustment for multiple comparisons (using family-wise error rates of 5%).

HLA–DRB1 allele    
 DRB1*0114.124.40.0060.51 (0.31–0.83)
 DRB1*0716.224.70.0310.59 (0.37–0.95)
 DRB1*150111.120.30.0310.49 (0.26–0.93)
HLA–DQA1 allele    
 DQA1*010119.038.2<0.00010.38 (0.25–0.57)
 DQA1*010227.743.10.00040.51 (0.35–0.74)
 DQA1*020115.432.9<0.00010.37 (0.24–0.58)
HLA–DQA1 peptide binding motif    
 DQA1 F2525.147.7<0.00010.26 (0.18–0.38)
 DQA1 S2673.986.20.00020.45 (0.30–0.68)
 DQA1 45(V/A)W(R/K)4768.782.8<0.00010.46 (0.31–0.66)

Homozygosity testing showed that only homozygosity for DQA1*0501 was possibly increased in frequency in patients with juvenile DM compared with controls (P = 0.014) (Table 1), and this was independent of DRB1*0301, which is an allele known to be part of the haplotype containing DQA1*0501 (7). Homozygosity for other alleles was found to be neither a risk factor nor a protective factor for juvenile DM.

HLA peptide binding motifs as risk or protective factors for juvenile DM.

The peptide binding motif DRB1 9EYSTS13 was found to be a risk factor for juvenile DM in Caucasian patients (OR 2.0, P = 0.0005) (Table 1). The DQA1 motifs F25 (OR 0.26, P < 0.0001), S26 (OR 0.45, P = 0.0002), and 45(V/A)W(R/K)47 (OR 0.46, P < 0.0001) were protective factors for juvenile DM (Table 2).

Comparison of HLA alleles and motifs between juvenile DM and adult DM.

Because of differences in the distribution of autoantibodies between adult DM and juvenile DM and the known associations of specific autoantibodies with certain HLA alleles (16), we compared HLA risk and protective factors in patients with juvenile DM and patients with adult DM without a defined myositis-specific or myositis-associated autoantibody (Table 3). Several risk factors, including DRB1*0301, DQA1*0301, DQA1*0501, and homozygosity for DQA1*0501, were noted to be unique to juvenile DM in patients without defined autoantibodies. Patients with juvenile DM and those with adult DM shared a number of DQA1 protective alleles and motifs, although DQA1*0201 and DQA1 S26 were unique protective factors in juvenile DM. Patients with juvenile DM and those with adult DM did not differ in the proportion of any alleles or motifs.

Table 3. Similarities and differences in risk or protective factors in patients with juvenile DM and patients with adult DM without defined myositis-specific and myositis-associated autoantibodies*
HLA class II allele or motifPatient group with risk or protective factorJuvenile DMAdult DMControls, %
%POR (95% CI)%POR (95% CI)
  • *

    P values were determined to be significant (P ≤ 0.05) after Holm's adjustment for multiple comparisons (using family-wise error rates of 5%). Patients with juvenile DM and patients with adult DM did not differ in the proportion of any alleles or motifs. NS = not significant, after application of Holm's procedure (see Table 1 for other definitions).

  • DQA1 high-resolution typing data were available for 84 patients with juvenile DM and 61 patients with adult DM without defined autoantibodies and 353 controls, DQA1 low-resolution (for determination of peptide binding motifs) for the same number of patients and 542 controls, DRB1 high- resolution for 40 patients with juvenile DM and 58 patients with adult DM without defined autoantibodies and 568 controls, and DRB1 low- resolution for 52 patients with juvenile DM and 65 patients with adult DM without defined autoantibodies and 797 controls.

Risk factor        
 DRB1*0301Juvenile DM62.5<0.00016.3 (2.2–18.0)39.7NS2.5 (1.1–5.8)21.0
 DRB19EYSTS13Juvenile DM, adult DM76.90.0182.2 (1.1–4.3)80.00.0012.6 (1.4–4.9)60.2
 DQA1*0301Juvenile DM27.40.0092.2 (1.1–4.6)24.6NS1.9 (1.0–3.7)14.5
 DQA1*0501Juvenile DM69.1<0.00012.8 (1.4–5.7)57.4NS1.7 (1.0–2.9)44.2
 DQA1*0501 homozygosityJuvenile DM11.90.0024.2 (1.7–10.3)11.5NS4.0 (1.5–10.9)3.1
Protective factor        
 DQA1*0101Juvenile DM, adult DM19.10.00080.38 (0.17–0.85)18.00.0020.36 (0.18–0.71)38.2
 DQA1*0102Juvenile DM, adult DM25.00.0030.44 (0.22–0.89)21.30.0010.36 (0.19–0.68)43.1
 DQA1*0201Juvenile DM15.50.0010.37 (0.16–0.88)16.4NS0.40 (0.20–0.82)32.9
 DQA1 F25Adult DM39.3NS0.71 (0.44–1.2)26.30.0020.39 (0.21–0.72)47.7
 DQA1 S26Juvenile DM76.20.020.51 (0.29–0.90)83.6NS0.82 (0.40–1.7)86.2
 DQA1 45(V/A)W(R/K)47Juvenile DM, adult DM16.7<0.00010.04 (0.02–0.08)70.50.020.49 (0.27–0.90)82.8

In comparing the risk and protective factors between patients with juvenile DM and those with adult DM, including those with myositis-specific or myositis-associated autoantibodies, DRB1*0301, DQA1*0501, and the DRB1 motif 9EYSTS13 were common risk factors, and the DQA1*01 allele and the DQA1 F25 motif were common protective factors for both juvenile DM and adult DM. HLA–DQA1*0301, a risk factor for juvenile DM and a possible risk factor for adult DM in patients without defined autoantibodies, was present in higher frequency in patients with juvenile DM than in those with adult DM (31.8% versus 18.1%; P = 0.003). The high-resolution alleles DQA1*0101, DQA1*0102, and DQA1*0201, and motifs DQA1 S26 and DQA1 45(V/A)W(R/K)47 were protective factors in patients with juvenile DM only, and not in patients with adult DM without defined autoantibodies.

RI scores and rank order of risk and protective factors.

The ranking of the RI scores for HLA–DRB1 and DQA1 alleles and motifs as risk or protective factors for juvenile DM was defined by Random Forests classification analyses. DRB1*0301 was ranked as having the highest RI for the development of juvenile DM (RI score 100%, GINI impurity rank 5.1) (Table 4). Two other risk factors for juvenile DM, DQA1*0301 and DQA1*0501, were ranked second and third in order of RI as risk factors for juvenile DM. Other HLA risk or protective factors from univariate analyses were found to have high RI ranking in terms of their ability to discriminate patients with juvenile DM from control subjects (Table 4). All Random Forests models consistently classified DRB1*0301 as the allele with the highest RI in discriminating patients with juvenile DM from controls, and stability in the rank order of the other risk or protective factor alleles and motifs was also observed. The error rate in these models ranged from 31.3% to 31.8%.

Table 4. Relative importance (RI) of juvenile DM–associated risk or protective factors, as predicted by Random Forests classification models in Caucasian patients*
HLA class II allele or motifRI, % (rank)
  • *

    RI scores were normalized to the highest-ranking factor in a given analysis.

  • Rank indicates the reduction of GINI scores (absolute values) calculated using the GINI impurity criterion for individual variables over all classification trees in the Random Forest. For this model, 107 patients with juvenile dermatomyositis (DM) and 327 European American controls were used, and the error rate was consistently 31.3–31.8%. Random Forests models each containing 500, 1,000, 10,000, 15,000, and 20,000 independent trees consistently classified DRB1*0301 as the allele with the highest RI in discriminating patients with juvenile DM from controls, and stability in the rank order of the other risk and protective factor alleles and motifs was also observed.

  • Risk factor.

  • §

    Protective factor.

DRB1*0301100 (5.10)
DQA1*030157 (2.89)
DQA1*050142 (2.12)
DQA1 S26 (*01, *02, *04, *06)§40 (2.02)
DQA1 45(V/A)W(R/K)47 (*01, *02)§37 (1.90)
DQA1*0201§37 (1.87)
DRB1*1501§33 (1.69)
DRB1 9EYSTS13 (*03, *11, *13, *14)32 (1.59)
DQA1 F25 (*0103, *0201, *0601)§27 (1.38)
DRB1*0101§27 (1.36)
DQA1*0101§27 (1.36)
DQA1*0102§26 (1.32)

The RI of HLA risk or protective factors was also studied by logistic regression analysis, utilizing the top 10 alleles or motifs from Random Forests classification analyses in the logistic regression model (Table 5). DRB1*0301 (OR 3.4, P = 0.0002) had the highest RI as an HLA risk factor for juvenile DM in Caucasian patients. Only DQA1*0301 (OR 3.4, P = 0.001) was identified as a secondary risk factor. DQA1*0201 (OR 0.41, P = 0.038) was the primary protective factor, and DRB1*0101 (OR 0.49, P = 0.066) had borderline significance as a protective factor for juvenile DM in Caucasian patients.

Table 5. Logistic regression analysis of risk or protective factors in Caucasians with juvenile dermatomyositis (n = 107) and controls (n = 327)*
HLA class II allele or motifPOR95% Wald confidence limits
  • *

    The top 10 alleles or motifs from Random Forests classification analysis were used in the model. The error rate was 29.5%. OR = odds ratio.

  • Risk factor.

  • Protective factor.

DRB1*03010.00023.41.8–6.4
DQA1*03010.0013.41.6–7.2
DQA1*02010.0380.410.18–0.95
DRB1*01010.0660.490.22–1.0
DRB1*15010.1370.590.26–1.2
DQA1 S260.5980.710.21–2.5
DRB1 9EYSTS130.7731.10.60–2.0
DQA1 F250.8171.10.54–2.2
DQA1*05010.8660.940.49–1.9
DQA1 45(V/A)W(R/K)470.9811.00.30–3.2

We used McNemar's test to determine if there was a significant difference between the results of Random Forests classification and the results of logistic regression analysis. There was no significant difference in the error rates between Random Forests classification and logistic regression analyses (error rates of 31.8% and 29.5%, respectively).

Gene–gene interactions.

The results of allele interaction analysis (15) demonstrated that DRB1*0301 was an independent risk factor for juvenile DM, regardless of whether DQA1*0501 was present (OR 3.6, P < 0.0001) or absent (OR 17.6, P = 0.005). In contrast, DQA1*0501 had no independent association with juvenile DM in the absence of DRB1*0301 (OR 0.38, P > 1.0 in the presence of DRB1*0301 and OR 1.9, P > 0.6 in the absence of DRB1*0301). Therefore, it is likely that the association of DQA1*0501 with juvenile DM was primarily attributable to its linkage disequilibrium with DRB1*0301 (OR 21.9, P < 0.0001). DRB1*0301 was also a risk factor for juvenile DM in the presence (OR 10.2, P = 0.002) or absence (OR 4.5, P < 0.0001) of DQA1*0301. In addition, DQA1*0301 was associated with juvenile DM independent of DRB1*0301 (OR 11.7, P = 0.0003 in the presence of DRB1*0301 and OR 5.1, P < 0.0001 in the absence of DRB1*0301).

Multiplicative logistic regression models were used to examine potential gene interactions between 2 risk factor alleles (DRB1*0301 with either DQA1*0301 or DQA1*0501) or 2 protective factor alleles (DQA1*0201 with either DQA1*0101 or DQA1*0102). In the multiplicative logistic regression models, interactions of 2 risk factor alleles or 2 protective factor alleles were found to have no significant effect on the associations with risk or protective factors for juvenile DM (P > 0.29 for each pair of combinations), suggesting that there was no multiplicative interaction of any 2 given risk or protective alleles.

DISCUSSION

The HLA class II allele DQA1*0501 has previously been shown to be an immunogenetic risk factor for the development of juvenile DM in Caucasian patients (9–11). However, those studies were based on relatively small numbers of patients. Given recent findings in adults, we characterized the complete allele profiles of HLA–DRB1 and DQA1 genes in a larger population of Caucasian patients with juvenile DM to reassess the HLA risk and protective factors. We utilized multivariable statistical approaches to identify the strongest HLA associations with juvenile DM. Our results identified DQA1*0301 as a novel immunogenetic risk factor for juvenile DM and confirmed the previous association of DRB1*0301 (17). Of these, DRB1*0301 appears to be the strongest risk factor, based on its much stronger RI ranking in Random Forests classification analysis. We also confirmed DQA1*0501 as a risk factor for juvenile DM, although this allele was not as strongly associated in the multivariable analyses. Our results suggest that, as is the case in adult DM, the association of DQA1*0501 as a risk factor is likely the result of its linkage disequilibrium with DRB1*0301 (6, 8, 18).

In large studies of North American Caucasian patients with adult idiopathic inflammatory myopathy, DRB1*0301 and its linked allele DQA1*0501 were also risk factors for adult DM, including in patients with specific autoantibodies as part of the 8.1 ancestral haplotype (6, 19). Although class I or class III MHC typing data were not available in the present study, these risk factors are likely part of the 8.1 ancestral haplotype in juvenile DM. Interestingly, the association with DRB1*0301 and the linked allele DQA1*0501 was not found to be present in patients with adult DM without defined autoantibodies, suggesting that this haplotype is even more strongly associated with the antisynthetase autoantibodies PM-Scl and Ro (16). The 8.1 ancestral haplotype has been considered to be a risk factor for multiple autoimmune diseases, conferring risk for autoimmunity that is not disease specific. Although the reasons for this are not known, the 8.1 ancestral haplotype is generally associated with increased humoral immunity, increased tumor necrosis factor α and Th2 cytokine responses, and decreased cellular immunity, and thus may further contribute to autoimmune disease pathogenesis (20).

In contrast, HLA–DQA1*0301, which is not a risk factor for adult DM in patients without defined autoantibodies, is an important independent risk factor for juvenile DM, implying some genetic differences between the juvenile and adult forms of the disease. Of note, homozygosity for DQA1*0501 is also a risk factor for juvenile DM. Homozygosity for DQA1*05 has also been recognized as a risk factor in adult polymyositis, and homozygosity for any DQA1 allele was previously reported to be an additional risk factor for myositis within multiplex families (21). The increase in the number of possible genetic risk factors in juvenile DM as compared with adult DM, homozygosity for a risk allele as another risk factor, and the slightly increased strength of associations for the risk factor alleles in juvenile DM as compared with adult DM may be factors in explaining the earlier age at onset of myositis, as is also seen in juvenile rheumatoid arthritis and type I diabetes mellitus (22–25).

In this study we also identified novel immunogenetic factors that were found significantly less frequently in Caucasian patients with juvenile DM compared with controls. These protective factors included the HLA class II alleles DQA1*0201, *0101, and *0102, of which *0201 and *0101 were also clearly protective factors in multivariable analyses. The role of protective alleles in disease pathogenesis is unclear, but in type I diabetes mellitus, the protective allele DQA1*0602 is effective in binding insulin and other disease-specific autoantigens as well as binding mimicking antigens from viral proteins (26). We speculate that a difference in hydrophobicity of the peptide binding groove between the protective DQA1*06 alleles compared with the risk factor alleles may result in the binding of antigens of higher hydrophobicity to the protective alleles. Although in some immune-mediated diseases HLA risk factors have been shown to have higher affinity for inciting antigens, in other cases autoantigens may have higher affinity for protective factor alleles (27, 28). Therefore, under some conditions, a tight peptide–MHC binding may result in elimination of autoreactive T cell clones from the thymus.

The patients with juvenile DM in this study appeared to have a number of protective alleles in common with those in patients with adult DM without defined autoantibodies, but overall carried more DQA1 protective factors (Table 3). This difference in the number of protective factors in juvenile DM compared with adult DM may be related to the lower incidence of juvenile-onset disease compared with adult DM (29, 30). A higher incidence of protective alleles in the Lazio region of Italy has been proposed to partially explain the relatively lower incidence of type I diabetes mellitus in this population (31). An alternative explanation for the lower incidence of DM in childhood is that children may be exposed to fewer environmental risk factors that can overcome these protective genes to initiate disease in children.

Herein we also defined a number of new peptide binding motifs as risk factors—DRB1 9EYSTS13—or protective factors—DQA1 F25, DQA1 S26, and DQA1 45(V/A)W(R/K)47—for juvenile DM. Peptide binding motifs were stronger risk factors in adult patients with idiopathic inflammatory myopathies (6, 19), but in children, multivariable analyses suggested that these associations were secondary and that the primary risk factors for juvenile DM were the individual HLA alleles. The reasons for this are not clear, but may be related to the fact that most of the motifs are composed of a combination of low-resolution alleles, and therefore their reduced association may be attributable to the fact that, in general, peptide binding motifs are not combinations of specific alleles. Alternatively, the motifs may be less important as risk or protective factors because these particular peptide binding motifs may not be relevant to the causative agents in juvenile myositis, which currently remain unknown.

The use of multivariable statistics to examine the relative strength of associations of the different alleles and binding motifs is a strength of the current study. Random Forests classification enables all alleles and motifs to be examined simultaneously and independent of each other, providing maximal separation of patients from controls. Its disadvantage is that specific ORs are not provided. The logistic regression analysis, in contrast, can examine only those alleles used in the analysis in relation to each other, and based on the limitation of patient numbers, only the top discriminating alleles or motifs from Random Forests classification were used. It is reassuring, however, that both analyses roughly corroborated one another. The error rates for these analyses were in the expected range, given that the HLA alleles are polymorphic loci present in healthy individuals as well as in patients.

Although attempts were made to minimize confounding factors in this study, some limitations remain. First, since there are strong immunogenetic associations with particular autoantibodies (8, 16), our HLA findings could be confounded by the presence of myositis-specific or myositis-associated autoantibodies in our patients. We did not have sufficient numbers of patients with juvenile DM with a known autoantibody to analyze the HLA associations with each specific autoantibody separately. It is also possible that other autoantibodies not yet identified may be present, each of which could have separate genetic risk and protective factors. However, in assessing the data from the patients without an identified antibody, it was reassuring that the findings were similar to those in our overall population. We were limited in our statistical power, however, to assess some subgroups of patients, such as those with adult DM without myositis autoantibodies, given the relatively small populations that would have resulted from such subsetting.

Another limitation of this study is that it was not a population-based study, but rather was based on referral populations of patients with juvenile DM, enrolled from throughout the US and Canada, primarily from tertiary care medical centers. Although the control subjects were also geographically diverse and enrolled from several different regions of the country (6), variations in the ethnic composition of the patients and controls may confound some of the reported associations. Therefore, confirmation of these findings in familial studies through transmission disequilibrium testing would be reassuring and is ongoing.

In this study we defined HLA–DRB1*0301 as the major immunogenetic risk factor for juvenile DM in Caucasian patients, and we found that DQA1*0501 is likely to have only a secondary association, due to its linkage disequilibrium with DRB1*0301; similar observations have been made in patients with adult DM with myositis-specific or myositis-associated autoantibodies (6, 8). We also identified additional risk and protective alleles for juvenile DM, including DQA1*0301 and DQA1*0201, respectively. These results suggest that juvenile DM has a complex pathogenesis that may differ from that of adult DM.

Acknowledgements

We thank Drs. Elaine Remmers and Karyl Barron for their critical review of the manuscript. We also thank members of the Childhood Myositis Heterogeneity Collaborative Study Group and others who contributed to this study, as follows: Leslie S. Abramson, Barbara S. Adams, Elizabeth M. Adams, F. Paul Alepa (posthumous), Kathy Amoroso, Elif Arioglu, Frank C. Arnett, E. Arthur, Balu H. Athreya, Alan N. Baer, Susan Hyatt Ballinger, Karyl S. Barron, April C. Bingham, William P. Blocker, John Bohn, John F. Bohnsack, Gilles Boire, Michael S. Borzy, Gary R. Botstein, Susanne L. Bowyer, Richard W. Brackett, Elizabeth B. Brooks, Christine Brunet, Thomas Bunch, Victoria W. Cartwright, Gail D. Cawkwell, Stephen J. Chanock, Chun Peng T. Chao, Darryl Crisp, Randy Q. Cron, R. Culp, John Daigh, Luminita David, Frederick C. Delafield, Andrew H. Eichenfield, John F. Eggert, Melissa Elder, J. Ellman, Janet E. Ellsworth, C. Etheridge, S. Evans, Kathleen Fearn, Terri H. Finkel, Charles B. Foster, Robert C. Fuhlbrigge, Vernon F. Garwood, Abraham Gedalia, Natalie Gehringer, Stephen W. George, Harry L. Gewanter, Ellen A. Goldmuntz, Donald P. Goldsmith, Gary V. Gordon, Larry M. Greenbaum, Katherine R. Gross, Hillary Haftel, Melissa Hawkins-Holt, C. Hendrics, Michael Henrickson, Gloria C. Higgins, J. Roger Hollister, Russell Hopp, Bruce Hudson, E. Huh, Norman T. Ilowite, Lisa F. Imundo, Jerry C. Jacobs (posthumous), Rita S. Jerath, Courtney R. Johnson, Mary Jones, Olcay Jones, Lawrence K. Jung, Lawrence J. Kagen, Stuart J. Kahn, Thomas G. Kantor, Ildy M. Katona, Gregory F. Keenan, Edward C. Keystone, Yukiko Kimura, Daniel J. Kingsbury, Steven J. Klein, C. Michael Knee, J. Koenig, Bianca A. Lang, Andrew Lasky, Alexander Lawton, Johanan Levine, Carol B. Lindsley, Robert N. Lipnick, Seth H. Lourie, Elizabeth Love, Max S. Lundberg, Katherine L. Madson, Peter N. Malleson, Donna Maneice, A. Mariano, Harold Marks, Alan L. Martin, F. Matthew, John Miller, S. Ray Mitchell, Hamid J. Moallem, Penelope A. Morel, Chihiro Morishima, Frederick T. Murphy, Henry Nathan, Ann Neumeyer, Chester V. Oddis, Judyann C. Olson, Karen Onel, Barbara E. Ostrov, Lauren M. Pachman, Ramesh Pappu, Murray H. Passo, Maria D. Perez, Donald A. Person, Karin S. Peterson, Paul H. Plotz, Marilynn G. Punaro, Charles D. Radis, Linda I. Ray, Peter D. Reuman, Robert M. Rennebohm, John D. Reveille, Rafael F. Rivas-Chacon, Alan L. Rosenberg, Deborah Rothman, Peter A. Schlesinger, Kenneth C. Schuberth, Donald W. Scott, D. Seamon, Bracha Shaham, Robert M. Sheets, David D. Sherry, Sara H. Sinal, Frances J. Stafford, Howard Stang, Robert P. Sundel, Ilona S. Szer, Ira N. Targoff, Simeon Taylor, Elizabeth S. Taylor-Albert, Donald E. Thomas, Richard K. Vehe, Maria-Lourdes Villalba, Scott A. Vogelgesang, Larry B. Vogler, Emily Von Scheven, S. Wahl, Carol A. Wallace, Harry J. Wander, Arthur Weinstein, Jana Wells, Patience H. White, Grace C. Wright, John Yee, Christianne M. Yung, and Lawrence S. Zemel.

Ancillary