Identification and multicentric validation of soluble CDCP1 as a robust serological biomarker for risk stratification of NASH in obese Chinese

Abstract

The definitive diagnosis of non-alcoholic steatohepatitis (NASH) currently relies on invasive and labor-intensive liver biopsy. Here, we identified soluble CUB domain-containing protein 1 (sCDCP1) as a top-ranked non-invasive biomarker for NASH using Olink-based proteomics in 238 obese individuals with liver biopsies. Both the circulating concentration and hepatic mRNA abundance of sCDCP1 were significantly elevated in patients with NASH and correlated closely with each histological feature of NASH. In the pooled multicenter validation cohort, sCDCP1 as a standalone biomarker achieved an area under the receiver operating characteristic (AUROC) of 0.838 (95% confidence interval [CI] 0.789-0.887) for diagnosing NASH, which is better than those achieved with cytokeratin-18 and other non-invasive tests. Furthermore, the C-DAG model established by the combination of sCDCP1 with diabetes, aspartate aminotransferase (AST), and gender accurately rules in and rules out both NASH and fibrotic NASH (gray zones <20%). Thus, sCDCP1-based non-invasive tests can be potentially implemented for screening and early diagnosis of NASH and for ruling out low-risk individuals to avoid unnecessary liver biopsies.

Keywords: biomarkers; metabolic steatohepatitis; metabolic syndrome; non-alcoholic steatohepatitis; non-invasive diagnosis; personalized risk stratification; proteomics.

Graphical abstract

Figure 1

Olink proteomics-based identification of serum sCDCP1 as the top performer in identifying NASH in liver-biopsy-confirmed NAFLD cohort (A) Random forest feature selection (left) and SVM learning (right) showing sCDCP1 was the top-ranked protein in identifying NASH. (B) Violin plot showing the distribution of sCDCP1 levels in patients with biopsy-proven NL (n = 36), NAFL (n = 100), or NASH (n = 102). (C) Heatmap based on the sCDCP1 expression level and the distribution of clinicopathological features. The p values of logistic regression between sCDCP1 and features are shown. ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05. (D) The Spearman correlation plot of sCDCP1 with clinical indicators that closely correlated with NAFLD. (E) Odds ratio (OR) for NASH with models being controlled for established risk factors in a stepwise manner. T1 (reference), the first tertile of sCDCP1 (−1.124 to 0.203 NPX); T2, the second tertile of sCDCP1 (0.203–0.971 NPX); T3, the third tertile of sCDCP1 (0.971–3.562 NPX). Model 1, non-adjusted; model 2, adjusted for gender, age, and BMI; model 3, adjusted for gender, age, BMI, ALT, AST, and HOMA-IR. (F) Interaction of sCDCP1 with CK18, FGF21m and THBS2. Spearman’s R and p values are shown within each rectangle, with color intensity indicating the strength of association. (G and H) Receiver operating characteristic (ROC) curve indicating the performance of sCDCP1, CK18, THSB2, and FGF21 in the diagnosis of NASH and fibrotic NASH. NPX, normalized protein expression, the Olink’s arbitrary unit.

Figure 2

Validation of circulating sCDCP1 as a robust biomarker for NASH by quantitative ELISA in multiple cohorts Data were collected from the main cohort (n = 489) (A–D) and external validation cohort (n = 135) (E–H). (A and E) Boxplot showing the distribution of serum sCDCP1 levels in patients with biopsy-proven NL, NAFL, or NASH (75/223/191 in main cohort, 30/62/43 in external validation cohort). Adj. p for trend was calculated after adjustment of age, gender, BMI, and HOMA-IR. ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05, ns, p ≥ 0.05 (Dunn’s test). (B and F) OR for NASH with models being controlled for established risk factors in a stepwise manner. T1 (reference), the first tertile of sCDCP1; T2, the second tertile of sCDCP1; T3, the third tertile of sCDCP1; T1/T2/T3, 15.63–69.08/69.08–170.01/170.01–935.24 pg/mL in the main cohort and 15.63–57.78/57.78–125.98/125.98–690.56 pg/mL in the external validation cohort. Model 1, non-adjusted; model 2, adjusted for gender, age, and BMI; model 3, adjusted for gender, age, BMI, ALT, AST, and HOMA-IR. (C and G) Heatmap based on serum sCDCP1 level and the distribution of histopathological features. The p values of logistic regression between sCDCP1 and features were shown. ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05. (D and H) Serum sCDCP1 in patients with NASH stratified by fibrosis stage (F 0–1, F 2–4, and F 3–4; 377/112/35 in the main cohort, 108/27/13 in the external validation cohort). Data are presented as Tukey boxplots. ∗∗∗p < 0.001 (Dunn’s test).

Figure 3

The decline in serum sCDCP1 was closely associated with decreases in CK18 and liver enzymes after bariatric surgery (A) Line plot showing paired sCDCP1 serum levels at baseline and follow-up after the surgical intervention in patients with (n = 75) and without (n = 76) NASH. (B) Correlations heatmap. Delta values (▵) = post-operation values – pre-operation values. Spearman’s R and p values are shown within each rectangle, with color intensity indicating the strength of association. ∗∗∗p < 0.001, ∗∗p < 0.01, ∗p < 0.05

Figure 4

Comparison of the performance of sCDCP1 with CK18 and other non-invasive tests in diagnosing NASH and fibrotic NASH (A and B) ROC curve of NASH (A) and fibrotic NASH (B) diagnosis in training set, test set, external validation cohort, and pooled validation cohort. The training set (n = 326) and the test set (n = 163) were split from the main cohort (n = 489) recruited from the First Affiliated Hospital of Jinan University in a ratio of 2:1. The external validation cohort (n = 135) was enrolled from three independent centers. The test set and the external validation cohort were combined into the pooled validation cohort (n = 298). NCSS, NASH clinical scoring system.