Table 1. The relationship between auto-epitope binding by HLA-DQ molecules and protection from autoimmune diseases is independent of disease types and DQ chains.

To test, whether the relationship between binding auto-epitopes and risk for autoimmune diseases is caused by disease- or DQ chain-specific differences in the epitope-binding of alleles, we constructed multivariate models: (1) log(OR) ∼ Fr_allele + Disease + Chain, (2) log(OR) ∼ Fr_aminoacid + Disease + Chain, where OR is the odds ratio for developing the given disease; Fr_allele is the level of disease-associated auto- epitope binding by the DQ allele; Fr_aminoacid is the level of disease-associated auto-epitope binding by DQ alleles, which contain the given amino acid and; Chain (i.e., DQA1 or DQB1) and Disease are categorical variables. Fr_allele and Fr_aminoacid showed negative effect on OR after controlling for diseases and DQ chains. Significant relationship between predictor and response variables is marked with red color.

	Allele associations				Amino acid associations
	Variable	Slope	Variance explained	P	Variable	Slope	Variance explained	P
	Frallele	–	0.4	0.01	Framinoacid	–	0.39	0.008
	Disease	NA	0.17	NA	Disease	NA	0.08	NA
	Chain	+ (DQB1)	0.02	0.47	Chain	− (DQB1)	0.08	0.004
R2	0.38 (P = 0.046)				0.5 (P = 6∗10−8)
N	22				61
BP test P	0.2				0.11

Notes.

Variance expained: The proportion of variance in log(OR) explained by the given predictor variable; P: the probability of observing relationship between the predictor and response variables by chance; R²: Total variance in log(OR) explained by the model (P corresponds to F-test P value); N: number of associations between autoimmune diseases and alleles or amino acids; BP test P: the P value for Breusch-Pagan test of heteroscedasticity, P values larger than 0.05 suggest lack of heteroscedasticity.