Proper definition of the set of autoantibody-targeted antigens relies on appropriate reference group selection

Autoimmune diseases are frequently associated with autoantibodies. Recently, large sets of autoantibody-targeted antigens ("autoantigen-omes") of patient and control sera have been revealed, enabling autoantigen-omic approaches. However, statistical standards for defining such autoantigen-omes are lacking. The z-score indicates how many standard deviations an antigen reactivity of a given sample is from the mean reactivity of the corresponding antigen in a reference group. Hence, it is a common measure to define significantly positive reactivity in autoantigen profiling approaches. Here, we address the risk of biased analyses resulting from unbalanced selection of the reference group. Three study groups were selected. Patients-of-interest were chronic inflammatory demyelinating polyneuropathy (CIDP); controls were other neuropathies (ONP); and healthy controls (HC). Each serum was screened for significant autoantigen reactivity using HuProt™ protein arrays. We compared three possible selections of reference groups for statistical z-score calculations: method#1, the control groups (ONP + HC); method #2, all groups together; method #3, the respective other groups (e.g., CIDP + HC for the ONP autoantigen-ome). The method selection seriously affected the size of the autoantigen-omes. Method #1 introduced a bias favoring significantly more antigens per patient in the CIDP group (for z >4: 19 ± 3 antigens) than in the control groups (ONP: 2 ± 1; HC: 0 ± 0). The more balanced methods #2 and #3 did not result in significant differences. This contribution may help to avoid interpretation biases and to develop guidelines for population studies revealing autoantigen-omes via high throughput studies such as protein microarrays, immunoprecipitation with mass spectrometry, or phage display assays.