Study participants

There were 42 individuals in the preclinical UC group, 101 in the diagnosed UC group, 11 in the preclinical CD group, 21 in the diagnosed CD group, and 82,357 in the HC group, and the observation period was 677,174.4 person-years. In this cohort, the prevalence of UC at the baseline survey was 122/100,000 persons, and the crude incidence of UC was 6.2/100,000 person-years. In the preclinical UC group, the mean age at baseline was 39.1 ± 13.3 years, 47.6% were male, and the median time from baseline to the diagnosis of UC was 4.57 years (Table 1). Anti-αvβ6 and anti-EPCR titers were measured for 40 individuals in the preclinical UC group, 99 in the diagnosed UC group, 11 in the preclinical CD group, 19 in the diagnosed CD group, and 123 matched HCs, excluding 6 participants whose baseline sera were absent. The lifestyle and dietary habit information was analyzed for 42 individuals in the preclinical UC group and 1091 individuals from the matched HCs.

Table 1 Baseline characteristics of each groupHigh positivity rates for anti-αvβ6 and anti-EPCR were confirmed in the preclinical UC group

The positivity rates for anti-αvβ6 and anti-EPCR antibodies were high in both the preclinical UC group and the diagnosed UC group, with anti-αvβ6 positivity rates of 52.5% and 73.7%, respectively, and anti-EPCR positivity rates of 51.4% and 63.6%, respectively (Fig. 2, Supplemental Table 1). Both antibody titers were significantly higher in the preclinical UC group than the preclinical CD group, diagnosed CD group, and HCs. Their sensitivity and specificity for predicting UC were 52.5% and 97.6%, respectively, for anti-αvβ6 and 51.4% and 97.8%, respectively, for anti-EPCR (Table 2).

Fig. 2

The scatter and box plots illustrate a anti-αvβ6 titers and b anti-EPCR titers by disease group at baseline. The vertical axis shows each antibody titer on a log(1 + antibody titer) scale. The cutoff value for the antibody titer is indicated by the dotted line. Asterisks indicate levels of statistical significance: *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001 (Wilcoxon signed-rank test). anti-αvβ6 anti-integrin αvβ6 antibody, anti-EPCR anti-endothelial protein C receptor antibody, MFI mean fluorescence intensity, n.s. not significant, UC ulcerative colitis, CD Crohn’s disease, HCs healthy controls

Table 2 Diagnostic performance of anti-αvβ6 and anti-EPCR titers for ulcerative colitis

Based on the combination of positive and negative results for both antibodies, patients were divided into four groups: double-positive, anti-αvβ6 single-positive, anti-EPCR single-positive, and double-negative (Table 3). Double-positivity was more specific for individuals with UC (preclinical UC group: 31.4%; diagnosed UC group: 51.1%; preclinical and diagnosed CD groups, and HCs: 0%) than single-positivity.

Table 3 Combination of positive and negative results for anti-αvβ6 and anti-EPCRAnti-EPCR predicted the onset of UC as accurately as anti-αvβ6, and the combination model demonstrated even greater predictive performance

Anti-αvβ6 and anti-EPCR titers in the single and combination models were evaluated for the preclinical UC and diagnosed UC groups (Fig. 3a, b). The predictive performance of anti-αvβ6 for UC onset had an AUC of 0.89 (95%CI 0.82–0.96); that of anti-EPCR had an AUC of 0.89 (95%CI 0.83–0.96). The combination model with anti-αvβ6 and anti-EPCR (AUC = 0.92, 95%CI 0.87–0.97) showed higher predictive performance than the single models, although the difference did not reach statistical significance (versus anti-αvβ6, p = 0.38; versus anti-EPCR, p = 0.070). The UC diagnostic performance of anti-αvβ6 had an AUC of 0.93 (95%CI 0.89–0.97) and that of anti-EPCR had an AUC of 0.90 (95%CI 0.85–0.95), whereas the AUC for the anti-αvβ6 and anti-EPCR combination model increased to 0.94 (95%CI 0.91–0.98).

Fig. 3

Receiver operating characteristic (ROC) analysis of models based on anti-αvβ6, anti-EPCR, lifestyle factors, and their combinations. The ROC curves represent the performance of (a) autoantibody models for the preclinical UC group, (b) autoantibody models for the diagnosed UC group, (c) models that include anti-αvβ6 and lifestyle for 153 individuals (39 in the preclinical UC group and 114 of the HCs), and (d) models that include anti-EPCR and lifestyle for 119 individuals (34 in the preclinical UC group and 85 of the HCs). anti-αvβ6 anti-integrin αvβ6 antibody, anti-EPCR anti-endothelial protein C receptor antibody, AUC area under the curve, CI confidence interval

Positivity rates for anti-αvβ6 and anti-EPCR are high in active UC cases, and titers of both autoantibodies are positively correlated

We analyzed autoantibody titers according to clinical subtype (Supplemental Tables 2, 3; Supplemental Figs. 1, 2). In the preclinical UC group, anti-αvβ6 titers were significantly higher in cases with moderate activity than in those with mild activity (p = 0.050), and positivity rates for both antibodies were also higher in cases with moderate activity (anti-αvβ6: mild 12.5%, moderate 57.1%; anti-EPCR: mild 37.5%, moderate 58.3%). There were no other clinical subtype factors relevant to autoantibody titers.

We next examined the correlation between anti-αvβ6 and anti-EPCR (Supplemental Fig. 3). We found a significant correlation between anti-αvβ6 and anti-EPCR titers in preclinical and diagnosed UC groups (preclinical UC group: r = 0.47, p = 0.004; diagnosed UC group: r = 0.42, p < 0.001), but not in the non-UC group.

The positivity rates for anti-αvβ6 and anti-EPCR increased as diagnosis approached in the preclinical UC phase

We examined longitudinal antibody titers in the preclinical UC group (Supplemental Table 4, Supplemental Fig. 4). First, we compared 11 pairs of samples between two time points (points A and B) before UC diagnosis. The positivity rates for both antibodies were higher at the point closer to diagnosis (anti-αvβ6: 36.4% to 63.6%; anti-EPCR: 50.0% to 80.0%), and anti-EPCR titers were significantly higher at the point closer to diagnosis (p = 0.014). No significant changes in antibody titers were observed between the two time points before and after diagnosis.

Insomnia was associated with UC onset among the environmental factors investigated

The analysis of lifestyle habits was performed using data from 41 individuals in the preclinical UC group, as lifestyle data were lacking for 1 individual (Table 4). The percentage of patients with an AIS ≥ 6 points, corresponding to insomnia, was significantly higher in the preclinical UC group than among the HCs (41.5% and 24.7%, respectively; crude OR = 2.16, 95%CI 1.07–4.27, p = 0.025). The percentage of patients with a history of smoking was higher in the preclinical UC group, but the difference was not statistically significant (58.5% and 44.6%, respectively; crude OR = 1.68, 95%CI 0.85–3.37, p = 0.11). Multivariate logistic regression was performed for AIS ≥ 6 points, smoking history, and drinking history, revealing that AIS ≥ 6 points (adjusted OR = 2.14, 95%CI 1.11–4.04, p = 0.019) was significantly associated with preclinical UC.

Table 4 Univariate and multivariate analyses for lifestyle factors

The dietary habits analysis was conducted with data from 36 individuals in the preclinical UC group, as dietary data were lacking for 6 individuals in the group. No nutrients or food categories were found to be significantly associated with preclinical UC (Supplemental Tables 5, 6).

Prediction model for UC development based on serum autoantibodies and environmental factors

We developed a regression model based on the lifestyle factors of insomnia and smoking history to evaluate its predictive performance for UC development. We performed ROC analysis for the lifestyle model, serum autoantibody model (anti-αvβ6 and anti-EPCR), and the combined model (Fig. 3c, d). The lifestyle single model (AUC = 0.65, 95%CI 0.55–0.74) had a significantly lower predictive test performance than the anti-αvβ6 and anti-EPCR single models (p < 0.001 and p < 0.001, respectively). Combination with the lifestyle model slightly improved the predictive performance of the anti-EPCR single model (AUC 0.90–0.93), but not the anti-αvβ6 single model (AUC 0.91 to 0.87).