A total of 197 patients that entered the prospective ACLF-I cohort study (observational study for the characterization of the pathogenesis of ACLF) between November 2020 and April 2023 and were included in this study (7). Patients with decompensated liver cirrhosis, age between 18 and 80 years, were eligible to enter the study. Hepatocellular carcinoma outside MILAN criteria, other malignancy, or severe congenital/acquired immune deficiency (e.g., HIV, immunosuppressive therapy in transplant recipients or rheumatologically diseases) and pregnancy were exclusion criteria. Demographic, laboratory and clinical characteristics are systematically recorded in a digital patient register (OSSE) (23, 24) from the clinical care data in cooperation with the Institute for Medical Informatics and the Data Integration Center at the Goethe University Hospital, Frankfurt (DIZ). Serum samples were collected on the day of study inclusion (baseline) and during follow-up. ALCF was classified according to the CLIF score and the European Association for the Study of the Liver (EASL)-CLIF criteria (17, 25). Alcohol-related liver cirrhosis was defined by a reported daily drinking average above 20 g/dL in their patient history. A cohort of healthy individuals (n = 234) was added as a control group. Of these, 200 samples were provided by the DRK Blood Donation Service Baden-Württemberg/Hessen, Frankfurt am Main, Germany, and 34 samples were obtained from healthy volunteers at Frankfurt University Hospital. Prior to participation, all participants gave their written informed consent.

This study was performed in accordance with the declaration of Helsinki and approved by the local ethics committee (ethics vote no. 20–653). All patients gave their written informed consent prior to entering the study.

Clinical data, laboratory data and serum samples were obtained at baseline and follow-up (obtained at each visit during a 3-month follow-up). Routine laboratory diagnostics include liver function tests differential white blood cell count (WBC), C-reactive protein (CRP), drinking behavior, smoking behavior, gastrointestinal bleeding, ascites, therapeutic paracentesis, albumin treatments, diabetes and diabetes treatment, transjugular intrahepatic portosystemic shunting (TIPS), hepatitis, viral infections, renal failure, respiratory failure, circulatory failure, data on bacterial and/or fungal infection development. Laboratory data and biological samples were obtained at each visit during a 3 month follow up.

Serum samples (9 mL) were taken from patients and healthy individuals and subjected to centrifugation. Serum samples were centrifuged at 1.400 x g for 10 min at 4 °C and the supernatant was taken, aliquoted and stored at −80 °C until further use.

Genomic DNA was isolated from 250 μL serum using the QIAmp® DNA Blood Mini Kit (Qiagen, Germany, Cat# 51104) according to the manufacturer’s manual under sterile conditions within a laminar flow cabinet. Prior to each extraction, the flow cabinet was decontaminated using a two-step disinfection procedure: initially with 2% Incidin Plus (Ecolab), which was applied for 15 min. to eliminate a broad spectrum of microorganisms, followed by 80% ethanol to remove residual disinfectant and enhance surface sterility. Subsequently, the cabinet was exposed to ultraviolet (UV) light for 30 min. to ensure additional decontamination under sterile conditions. To monitor for potential bacterial DNA contamination, two negative extraction controls (columns processed without serum) were included in the DNA extraction steps. The isolated DNA was eluted in 35 μL Elution Buffer (EB) and DNA concentration and purity were assessed using a NanoDrop® ND-2000 spectrophotometer (Thermo Fisher Scientific).

The following bacterial isolates were used in this study: E. faecium (DSM# 20477), E. coli Dh5α and E. coli JM109 (Promega Corp. Madison, US cat# L2005). The bacterial strains were pre-cultured from −80 °C glycerol stocks on either lysogeny broth (LB)- agar plates (LB-Agar Lennox, Carl Roth GmbH + Co. KG, 37 gL⁻¹) or on Trypticase soy yeast extract medium (TSYM) agar plates (30 gL⁻¹ Carl Roth Trypticase soy broth, 2.0 gL⁻¹ Carl Roth Yeast Extract, 15 gL⁻¹ Sigma-Aldrich Agar). The bacteria were incubated at 37 °C on agar plate’s prior application. All bacteria used in this study are listed in Supplementary Table S1.

For the quantification of EF 16S rDNA, a sequence fragment of the 16S rDNA of EF were cloned into the pGEM-T vector (26, 27) and the corresponding plasmids were used to establish real-time-quantitative PCR (RT-qPCR) standard curves. Briefly, a standard PCR was performed using a 50 μL PCR- GoTaq™ green master mix (Promega Corp. Madison, US cat# M712) containing 1 μL of EF DNA. For the 16S rDNA gene amplification EF specific primers were used: forward primer E. faecium_qPCR_16SF (5′-GCGGCTCTCTGGTCTGTAAC-3′), reverse primer E. faecium_qPCR_16SR (5′- TAAGGTTCTTCGCGTTGCTT-3′), amplifying ∼254 bp from the 16S rDNA gene of EF. The amplified PCR products were checked on 1.5% agarose gel and the PCR products were cut out and purified with the QIAquick® Gel Extraction Kit (Qiagen, Cat# 28704). The purified PCR products were ligated into the pGEM-T vector (Promega, Madison, WI, US cat#A137A) using Promega T4 DNA Ligase (Promega Corp. Madison, US cat# M180A) and transformed into E. coli JM109 competent cells (Promega, Madison, WI, US cat# A1380) according to the manufacturer’s manual instructions. Clones were picked and checked for the correct insert by using the M13 forward and reverse primers and sequencing of the PCR products by Sanger-Sequencing (EUROFINS, Ebersberg, Germany).

The abundance of EF in serum samples was measured by RT-qPCR using the above mentioned primers: E. faecium_qPCR_16SF, reverse primer E. faecium_qPCR_16SR, amplifying ∼254 bp from the 16S rDNA gene of EF From each sample 2 μL DNA were subjected to quantitative RT-qPCR using QuantiNova® SYBR® Green PCR master mix (Qiagen, Germany, Cat# 208054). For quantification two-step quantitative RT-qPCRs was performed on an Applied Biosystems StepOnePlus Real-Time PCR System (Applied BioSystems, Waltham MA, United States), with the following settings: initial cycle 95 °C for 2 min., followed by 40 cycles of 95 °C for 5 s. (denaturation) and 60 °C for 10 s. (combined annealing/extension). All samples were run in triplicate. Negative controls included nuclease-free water (no-template control, NTC) and eluted DNA from the negative extraction controls (columns processed without serum; NC). These controls were included in each RT-qPCR reaction to monitor for contamination (28, 29). To avoid cross-contamination, RT-qPCR reactions were prepared in a separate room and under a different laminar flow hood than those used for DNA extraction. RT-qPCR efficiency was evaluated using a standard curve generated from fivefold serial dilutions (in triplicate) of plasmid DNA (pGEM-T vector) containing the respective E. faecium 16S rDNA gene fragment, or genomic DNA from E. faecium. The calculated amplification efficiencies ranged from 87 to 107%, with correlation coefficients (R²) between 0.91 and 0.98. Relative gene expression was quantified using the 2^−ΔΔ Ct method, normalized to the 16S rDNA gene expression levels of the standards (30). All primers used in this study are listed in Supplementary Table S2.

For RT-qPCR samples that were tested positive for EF DNA, the PCR products were extracted and purified using the QIAquick® PCR Purification Kit (Qiagen, Germany, Cat# 28104), following the manufacturer’s instructions. For molecular identification through Sanger sequencing, 15 μL of the purified PCR product DNA was premixed with 2 μL of primer (final concentration 10 pmol/μL). For 16S rDNA gene sequencing the following primer was used: E.faecium_qPCR_16SF (5’-GCGGCTCTCTGGTCTGTAAC-3′) amplifying ∼254 bp from the 16S rDNA gene of EF. The premixed sequencing samples were sent to Eurofins for Sanger-sequencing using the TubeSeq service (Eurofins Genomics Europe, Ebersberg, Germany).

To determine the concentration of inflammatory cytokines, present in the serum samples we measured the cytokine concentration of 11 different cytokines IFN-gamma (BR29), IL-1 alpha/IL-1F1 (BR38), IL-1 beta/IL-1F2 (BR28), IL-1ra/IL-1F3 (BR30), IL-2 (BR43), IL-6 (BR13), IL-10 (BR22), IL-18/IL-1F4 (BR78), Lymphotoxin-alpha/TNF-beta (BR45), MIF (BR53), TNF-alpha (BR12) in all serum samples of the patients by using Luminex® Discovery Assay (Bio-Techne GmbH Cat# LXSAHM-16). The cytokine measurements were performed on a BioPlex 200 system powered by Luminex® xMAP™ Technology and xPONENT software V4.3 according to the manufacturer’s manual.

Prior to correlation analysis and random forest analysis, data normalization and scaling were performed using log transformation (base 10) and auto scaling (Z-transformation of each variable). Kendall or Spearman rank correlation tests were conducted to evaluate the features of interest. To determine the significance of each clinical variable, its contribution to the clinical phenotype, and its association with the presence of EF DNA, Random Forest analyses were performed using 500 trees for supervised classification. These analyses were carried out with the R software and the MetaboAnalyst package (31–33). The mean decrease in accuracy was calculated as a measure of each variable’s importance and subsequently plotted.

All variables were plotted as single data point or expressed as median (interquartile range) and were compared between the EF positive (EF+) and EF negative (EF-) groups, organized by the respective stratification classes (SDC, UDC and ACLF). In violin plots quartiles are denoted as dotted black lines, medians are denoted as joined lines. Prior to statistical analysis, a normality test was conducted. Normally distributed data were analyzed using a t-test for univariate analysis. Influence of EF and disease stages (SCD, UCD and ACLF) on different endpoints was investigated with the full model two-way ANOVA followed by a post-hoc Tukey test (HSD-test). For non-normally distributed data the Kruskal-Wallis test was used. Chi-square tests or Fisher’s exact test was performed to compare aetiologies and types of acute decompensation as well as to compare the frequency of EF in patients and controls. p-values ≤ 0.05 were considered to be statistically significant. Statistical calculations and plots of cytokine intensities and clinical lab values for all individual patients were performed and created using GraphPad Prism version 9.5.1 for Windows (GraphPad Software, San Diego, California, United States).¹

In total, blood samples of 197 patients who were prospectively enrolled in the above mentioned observational cohort were analyzed. Of these, 135 patients were male (69%) and 62 females (31%), average age 59 years (± 12). A control cohort, totaling 234 healthy individuals, consisted of 98 men (42%) and 136 women (58%), average age 41 years (± 16). Demographic and baseline characteristics of the patient cohort are depicted in Table 1. The classification of the patients’ clinical course (based on the PREDICT study criteria) identified 83 patients (42%) with SDC, of whom 28% were EF+ and 72% were EF-. Additionally, 49 patients (25%) were categorized with unstable decompensated cirrhosis (UDC), with 18% EF+ and 82% EF-. Furthermore, 65 patients (33%) developed acute-on-chronic liver failure (ACLF) within 3 months of hospital admission, with 29% EF+ and 71% EF- (Figure 1A). The prevalence of bacterial infections at baseline did not differ significantly between EF+ and EF- patients.

From the patients with acute decompensation of liver cirrhosis, sera of 51 patients (∼26%) were positive for EF DNA (EF+; by RT-qPCR), while sera of 146 patients (74%) were negative for EF DNA (EF-; Figure 1B). In contrast, only 3 samples (1.3%) from the control cohort were EF+ (p-value = 0.001; Supplementary Figure 1). In the three EF+ cases from the control cohort, the frequency of EF DNA reached up to 1.93 × 10⁴ copies of the 16S rDNA gene, corresponding to a theoretically estimated abundance of approximately 3.2 × 10³ EF CFUs, assuming 6 copies of the 16S rDNA gene per EF cell (34). In the patient cohort, EF DNA was detected at abundances of up to 3.5 × 10⁵ copies of the 16S rDNA gene, corresponding to approximately 5.8 × 10⁴ EF CFUs, based on an estimated six copies of the 16S rDNA gene per EF bacterial cell (34). Among EF+ patients, 23 (45%) were classified with having SDC, 9 (18%) classified with UDC, and 19 (37%) developed ACLF (Figure 1C). In the EF- group, 60 patients (41%) had SDC, 40 (27%) had UDC, and 32 (46%) were categorized with ACLF (Figure 1D). Of the 65 ACLF patients, 19 (29%) were EF+, with no significant differences in ACLF grades 1 to 3 between EF+ and EF- patients (data not shown). RT-qPCR analysis revealed significantly lower CT values in ACLF patients (mean CT=35.80) compared to those with SDC (mean CT=37.48, p< 0.0001) and UDC (mean CT=36.81, p= 0.0254; Supplementary Figure 2), indicating a higher abundance of EF 16S rDNA in their serum samples.

To assess the concordance between conventional microbiological diagnostics and molecular detection of EF DNA, we compared results from standard clinical specimens—blood cultures, urine samples, and VRE swab tests—with serum-based RT-qPCR results obtained from the same patient cohort (n=51). Classical microbiological analyses identified EF in a total of 7 patients: 1 in blood culture, 5 in urine samples, and 7 in VRE swabs (note that some patients had multiple positive specimen types). In contrast, RT-qPCR targeting the 16S rDNA of EF detected bacterial DNA in 51 out of 51 patients, indicating a substantially higher sensitivity. A Chi-square test confirmed that this difference is highly statistically significant (χ² = 64.68, p< 0.0001), supporting the conclusion that RT-qPCR is markedly more sensitive than classic microbiological-based diagnostic methods in this cohort (Supplementary Figure 3).

A chi-square analysis revealed a significant difference in the distribution of liver disease aetiologies between EF+ and EF- patients with ACLF (p= 0.0008), indicating a potential link between the underlying cause of liver disease and the detection of EF DNA in serum (Supplementary Figure 4A). Additionally, there was a significantly (p= 0.0254) higher proportion of EF+ patients with ACLF presented with portal hypertension compared to EF– patients. This finding suggests a potential association between the presence of Enterococcus faecium DNA and advanced portal hypertension in ACLF (Supplementary Figure 4B).

The analysis of inflammation-related markers between EF+ and EF- patients stratified by disease phenotype revealed significant differences in leukocyte counts, CRP, and IL-6 levels (Figure 2). Leukocyte counts were the highest in ACLF EF+ patients, followed by ACLF EF- patients. Both ACLF groups exhibited significantly higher leukocyte levels compared to UDC and SDC patients (p≤ 0.0001 and p≤ 0.005, respectively; Figure 2A). Among UDC patients, EF+ individuals tend to show elevated leukocyte counts relative to EF- patients, with significant enhanced leucocytes in ACLF EF+ compared UDC EF- patients (p≤ 0.0001; Figure 2A). In contrast, SDC patients had the lower leukocyte counts and showed no significant differences between the SDC EF+ and EF- group but a significant difference could be observed between SDC EF- and ACLF EF+ (p≤ 0.005) as well as SDC EF+ and ACLF EF- (p≤ 0.05; Figure 2A). These results underscore a possible link between EF DNA positivity and SI, particularly in patients with advanced disease stages such as ACLF and UDC. C-reactive protein (CRP) levels (Figure 2B) followed a similar trend. ACLF EF+ patients had the highest median CRP concentrations compared to EF+ and EF- SDC patients (p< 0.005; Figure 2B). Among UDC patients, there was no significant difference between EF+ and EF- individuals (p≥ 0.05). In the SDC group, CRP levels remained low overall, with no significant differences between EF+ and EF- patients (Figure 2B). As shown in Figure 2C, IL-6 levels were markedly elevated in ACLF EF+ patients, who exhibited the highest median concentrations among all liver disease phenotypes IL-6 levels in ACLF EF+ patients were significantly higher than those in SDC EF+ patients (p≤ 0.05). Additionally, IL-6 levels were significantly elevated in ACLF EF− patients compared to both SDC EF+ and SDC EF− patients (p < 0.05; Figure 2C). Among all groups, SDC EF+ patients had the lowest IL-6 levels, with no significant differences between EF+ and EF- individuals in the SDC category (Figure 2C). These observations point to a potential link between IL-6 elevation and EF DNA positivity in more advanced stages of liver disease. Overall, the data support the idea that EF DNA may contribute to an intensified SI response, particularly in UDC and ACLF patients.

Our analysis of liver inflammation markers revealed significant variations in plasma bilirubin and AST levels across disease phenotypes and EF DNA status (Figure 3). Plasma bilirubin levels (Figure 3A) were markedly elevated in ACLF EF+ patients, who showed the highest concentrations across all groups. Statistically significant differences were observed between ACLF EF+ and other subgroups, including SDC EF+, SDC EF-, UDC EF+, and UDC EF- (p≤0.05 to p ≤ 0.005; Figure 3A). These findings suggest that bilirubin elevation is primarily driven by disease severity rather than EF DNA status. Serum AST levels (Figure 3B) were less different between the groups with UDC EF+ patients showing the highest levels which were significantly elevated compared to SDC EF+ patients (p≤ 0.05; Figure 3B). However, no significant differences were detected between EF+ and EF- patients within the same phenotype. The findings related to liver function emphasize the link between liver dysfunction markers and disease severity, particularly in ACLF patients. While EF DNA presence does not independently alter bilirubin or AST levels within each disease stage, it may contribute to worsening liver function in more advanced phenotypes such as UDC and ACLF.

Analysis of plasma creatinine levels revealed a correlation EF DNA detection in patient sera and kidney function (Figure 4). Creatinine concentrations were the highest in ACLF EF+ patients compared to most other subgroups including SDC EF-, SDC EF+ and UDC EF- groups, indicating more severe kidney dysfunction in this group (p≤ 0.0001 to p ≤ 0.005). Additionally, creatinine levels in ACLF EF- patients were significantly higher than those in SDC EF+ and SDC EF- groups (p≤ 0.005), further underscoring the impact of disease severity on renal function (Figure 4). Among the SDC and UDC groups, creatinine levels were relatively lower and showed no statistically significant differences between EF+ and EF- patients. These findings indicate that the presence of EF DNA is associated with worsened kidney function in patients with ACLF, while its impact is less evident in earlier stages of liver disease. Despite marked differences in creatinine levels, sodium and potassium concentrations remained consistent across all patient groups (data not shown). These results indicate that the presence of EF DNA is associated with more severe kidney dysfunction, particularly in ACLF patients, where SI and renal impairment are most pronounced.

We additionally examined how the presence of EF DNA relates to clinical data and its possible contribution to the development of ACLF. To do this, we used Random Forest and correlation analyses on a subset of 106 patients from our clinical dataset to assess how EF DNA affects different clinical parameters. Supplementary Figure 5A presents the top 25 clinical parameters most closely associated with EF DNA positivity within this cohort. Notably, serum levels of Macrophage Inhibitory Factor (MIF) emerged as the most significant clinical predictor of ACLF, showing a statistically significant correlation (p = 0.013). Additionally, while IL-18, platelet count, IL-1β, albumin, IFN-γ, IL-10, IL-1α, IL-6, bilirubin, serum creatinine, IL-1ra, INR, and MELD score were positively correlated with ACLF development, these associations did not reach statistical significance. In contrast, markers such as GGT, TNF-β, TNF-α, AST, ALT, hemoglobin, and hematocrit showed negative correlations with ACLF development, though these correlations were also non-significant. (Supplementary Figure 5A). We also examined which laboratory parameters were linked to the presence of EF DNA using the random forest classification, identifying Macrophage Inhibitory Factor (MIF), platelet count and sodium as most important predictors (Supplementary Figure 5B).

Cumulative hospital days were stratified by disease stage (SDC, UDC, ACLF) and Enterococcus faecium DNA status. Within each disease class (SDC, UDC, ACLF), no significant differences in cumulative hospital stay were observed between EF- and EF+ patients (Supplementary Figure 6A). However, a clear trend was noted in the ACLF group, where EF+ patients showed longer cumulative hospital stays compared to EF- cases. Overall, cumulative hospitalization was primarily determined by disease severity, with ACLF patients experiencing the longest stays regardless of EF status (Supplementary Figure 6A).

The length of hospital stay was assessed across disease stages (SDC, UDC, ACLF) in relation to EF DNA status. No significant differences were found between EF- and EF+ patients within any disease class. In ACLF, EF+ cases showed a tendency toward longer hospital stays compared to EF- cases, although this trend did not reach statistical significance (Supplementary Figure 6B).