An autoantibody profile identified by human genome-wide protein arrays in rheumatoid arthritis

Abstract

Precise diagnostic biomarkers of anticitrullination protein antibody (ACPA)-negative and early-stage RA are still to be improved. We aimed to screen autoantibodies in ACPA-negative patients and evaluated their diagnostic performance. The human genome-wide protein arrays (HuProt arrays) were used to define specific autoantibodies from the sera of 182 RA patients and 261 disease and healthy controls. Statistical analysis was performed with SPSS 17.0. In Phase I study, 51 out of 19,275 recombinant proteins covering the whole human genome were selected. In Phase II validation study, anti-ANAPC15 (anaphase promoting complex subunit 15) exhibited 41.8% sensitivity and 91.5% specificity among total RA patients. There were five autoantibodies increased in ACPA-negative RA, including anti-ANAPC15, anti-LSP1, anti-APBB1, anti-parathymosin, and anti-UBL7. Anti-parathymosin showed the highest prevalence of 46.2% (p = 0.016) in ACPA-negative early stage (<2 years) RA. To further improve the diagnostic efficacy, a prediction model was constructed with 44 autoantibodies. With increased threshold for RA calling, the specificity of the model is 90.8%, while the sensitivity is 66.1% (87.8% in ACPA-positive RA and 23.8% in ACPA-negative RA) in independent testing patients. Therefore, HuProt arrays identified RA-associated autoantibodies that might become possible diagnostic markers, especially in early stage ACPA-negative RA.

Keywords: HuProt array; anticitrullination protein antibody (ACPA); autoantibody; rheumatoid arthritis.

FIGURE 1

Flow chart of the study design and quality control of HuProt arrays. (A) Flow chart of the study composed of HuProt array in Phase I, customized focused array for validation in Phase II and diagnostic model of 44‐marker‐panel. (B and C) Quality control of HuProt arrays. A total of 10 RA and three control serum samples diluted 1:1000 were incubated with the human proteome microarray. Then, Cy5‐conjugated anti‐human IgG antibody were added. The colored box indicates positive autoantigens. (B) Consistency and diversity autoantibody profiling between healthy control and RA samples. (C) Chip showed that RA sera were recognized with NM_014042 (gray rectangle), NM_002824 (green rectangle), and BC093033 (orange rectangle) respectively.

FIGURE 2

The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis and GO enrichment analysis of 51 selected candidates of Huprot™ 3.1 Chip. (A) Gene Ontology (GO enrichment analysis). (B) KEGG, Kyoto Encyclopedia of Genes and Genomes.

FIGURE 3

Dot plot for four distinct patient clusters by t‐SNE algorithm based on 51 autoantibodies screened by focused array. (A) Each plot represents one patient, there were 51 dimensions representing 51 autoantibodies, which were reduced to two dimensions (t‐SNE1 and t‐SNE2) by t‐SNE analysis. There were four clusters, red and light blue dots indicate disease control and healthy control. Yellow dots represent anti‐CCP‐negative RA and dark blue represent anti‐CCP‐positive RA. (B) Constitution of patients and controls. DC, disease control; HC, healthy control; CN, anti‐CCP‐negative RA; CP, anti‐CCP‐positive RA; SLE, systemic lupus erythematosus; SS, Sjögren's syndrome; AS, Ankylosing spondylitis; OA, osteoarthritis.

FIGURE 4

Prevalence of autoantibodies in early stage ACPA‐negative RA. (A) Presence of autoantibodies in early RA. Prevalence based on the cut‐off value of fluorescence signal intensity for five autoantibodies in early RA (history < 2 years, n = 53) and established RA (history ≥ 2 years, n = 99) were displayed. Gray bars: percentage of autoantibodies‐positive patients; pink bars: percentage of autoantibodies‐negative patients (chi‐square or fisher's exact, *p < 0.05). (B) Presence of autoantibodies in early and ACPA‐negative RA. Prevalence of the five autoantibodies in early RA (history < 2 years, n = 13) and established RA (history ≥ 2 years, n = 35) with anti‐CCP negative. Gray bars: percentage of autoantibodies‐positive patients; pink bars: percentage of autoantibodies‐negative patients (chi‐square or fisher's exact, *p < 0.05). NM_014042 (anti‐ANAPC15), BC001785.1 (anti‐LSP1), NM_145689 (anti‐APBB1), NM_002824 (anti‐PTMS), and NM_032907 (anti‐UBL7).

FIGURE 5

Correlation of diagnostic autoantibodies and clinical significance. (A) Heat map indicating correlations between autoantibodies and immunological characteristics in RA (n = 152). Scale color of the filled squares indicates the strength of the correlation (r) and whether it is negative (blue) or positive (red). Spearman's rank correlation test or Pearson correlation test. **p < 0.05 and corrected p < 0.05; *p < 0.05 and corrected p > 0.05, The corrected p value is the p value calculated after a Bonferroni correction; five autoantibodies with specificity more than 20% in anti‐CCP‐negative RA are: anti‐ANAPC15 (NM_014042), anti‐LSP1 (BC001785.1), anti‐APBB1 (NM_145689), anti‐PTMS (NM_002824), and anti‐UBL7 (NM_032907); SHS, modified Sharp‐van der Heijde Score. (B–F) Fluorescence signal detected for the five autoantibodies mentioned above. The rectangles indicate the interquartile range, and the bar within the rectangle indicates the median value, SEM was showed as error bars. RA, rheumatoid arthritis; HC, healthy controls; OA, osteoarthritis; SLE, systemic lupus erythematosus; CTD, connected tissue disease; AS, Ankylosing spondylitis.

FIGURE 6

Diagnostic model with focused array‐derived autoantibodies. (A) Feature importance (feature weight w) of the 44 selected autoantibodies in SVM model. Feature selection with support vector machine (SVM) model. According to feature weight, 44 autoantibodies were selected. (B) ROC curve of RA diagnosis prediction in cross validation with training data. (C) ROC curve of RA diagnosis prediction with testing data. (B and C) training set (n = 240, 120/120) and testing set (n = 203, 62/141). Performance of diagnostic model on training set (fourfold cross validation) and testing set, AUROC = 0.86 and 0.84, respectively.