Circulating microRNAs improve bacterial infection diagnosis and overall survival prediction in acute decompensation of liver cirrhosis

Abstract

Bacterial infections are the most frequent precipitating event in patients with acute decompensation of cirrhosis (AD) and are associated with high mortality. Early diagnosis is challenging due to cirrhosis-related systemic inflammation. Here we investigated the potential of circulating microRNAs to diagnose bacterial infections and predict survival in cirrhotic patients with AD. High throughput profiling of circulating microRNAs was performed using the Nanostring technology in 57 AD patients and 24 patients with compensated cirrhosis (CC). Circulating miRs profiling showed that: (a) miRs differentially detected in AD vs. CC were mostly down-regulated; (b) a composite score including absolute neutrophil count, C reactive protein and miR-362-3p could diagnose bacterial infection with an excellent performance (AUC of 0.825 [95% CI = 0.671-0.980; p < 0.001]); (c) a composite score including miR-382-5p, miR-592 and MELD-Na improved 6-month survival prediction. Circulating miRs are strongly dysregulated in patients with AD and may help to improve bacterial infection diagnosis and survival prediction.

Keywords: Bacteriology; Cell biology.

Graphical abstract

Figure 1

Flow-chart of the study population disposition Eighty-nine consecutive patients (61 AD and 28 CC) of the prospective observational EMIC cohort were included from June 2019 to May 2021. During the COVID-19 pandemic, 13 additional consecutive patients (8 AD and 5 CC patients) were recruited, with the same inclusion and exclusion criteria, in an external prospective cohort at the Beaujon University Hospital from July 2013 to May 2016, were also included. 18 out of the 102 patients were excluded from the miR profiling study because of inclusion criteria failures [COVID-19 infection (n = 2), HCC diagnosis during the hospitalization (n = 2), incorrect definition of AD or CC (n = 4), absence of plasma sample (n = 10)]. Two additional patients were excluded due to samples hemolysis. miR analysis was performed on 57 patients with and 25 patients with CC. In 12 patients a second run was required due to poor ligation, and 1 CC patient was finally excluded due to persistent poor ligation after 3 runs. miR profiling results are presented for 57 AD and 24 CC patients. In the AD group, 17 had a proven bacterial infection at the hospital admission while 31 patients had no suspected or proven sepsis. In 8 cases sepsis remained unproved and 1 patient had a viral infection.

Figure 2

Dysregulation of circulating miRs in AD of liver cirrhosis (A) PCA of the 96 differentially detected miRs (p < 0.05) between all AD patients (light and dark blue; n = 57) and CC patients (yellow; n = 24). AD patients with ACLF (dark blue, n = 18) and non-ACLF AD patients (light blue, n = 39) clustered together. (B) Heatmap of the top differentially detected miRs generated by hierarchical clustering. Color scale is blue to red with high values getting the red color. (C) Histogram of the top 31 miRs (log 2 old-change values). Only 2 miRs (miR-371a-5p and miR-493-3p) are significantly upregulated in AD patients.

Figure 3

Development of a bacterial infection diagnostic score in AD patients (A) PCA of the 34 differentially detected miRs between patients with proven bacterial infection (n = 17) and patients without any suspected or confirmed bacterial infection at the time of hospital admission (n = 31). (B) Median levels of cytokines (IL-6, IL-8, and TNFα), miRs (miR-362-3p, miR-612, and miR-3144-3p), ANC and CRP in infected and non-infected AD patients. ANC, CRP, IL-6 levels were significantly higher in infected patients (respectively p = 0.011, p = 0.006 and p = 0.020), as plasma counts of miR-362-3p (p = 0.002), miR-3144-3p (p = 0.020) and miR-612 (p = 0.027). (C) Focused PCA of significant biological parameters (ANC, CRP, and IL-6) and the top 3 differentially detected miRs (miR-362-3p, miR-612, and miR-3144-3p), allowing to visualize covariance and correlation strength of the different variables. After adjusting for ANC, IL-6 was not significant anymore. miRs carried redundant information whereas CRP and ANC were independent of each other. (D) ROC curves of the sepsis score (combining miR-362-3p, ANC, and CRP), miR-362-3p, CRP, ANC, and the combination of ANC and CRP in relation to bacterial infection diagnosis. Targeting a negative predictive value of 92.0%, optimal cut-off value was 0.310 (positive predictive value 70%, sensitivity 87.5%, and specificity 79.3%). Targeting a positive predictive value of 88.9%, threshold value was 0.668 (negative predictive value 77.8%, sensitivity 50.0%,and specificity 96.6%). Bacterial infection is unlikely in patients with score <0.310 while patients with score >0.668 are likely infected.

Figure 4

Predictive value of circulating miRs in AD (A) PCA of the 27 differentially detected miRs between patients alive 6 months after sample collection (n = 41) and patients with poor outcome (n = 16). (B) ROC analysis of 6-month survival probability relative to MELD-Na score or prognostic score combining MELD-Na, miR-382-5p and miR-592. C-index of MELD-Na was 0.766 (95% CI, 0.633–0.899), while c-index of composite score combining MELD-Na and the 2 miRs was 0.843 (95% CI, 0.770–0.916), representing and improvement in prediction error of 32.9%. (C) Comparison of Kaplan-Meier curves between patients with a miR-based score >1.0 (corresponding to the 25^th percentile) to patients with lower score (p < 0.0001 using log rank test). For the establishment of the new prognostic score, variables were mean centered and, using the regression coefficients, the score was calculated as follows: 0.1721x(MELD-Na-23.5) + 0.0398x(miR-382-25) – 0.0901x(miR-592-17).