The Ethics Committee of Shengjing Hospital of China Medical University approved this study (2022PS1067K) and written informed consent was obtained from all patients involved in the study for the publication of any potentially identifiable data included in this article. We retrieved the records of patients with a diagnosis of PLA from January 2017 to December 2021 from the electronic medical database at our institution (n = 290). According to the presence or absence of DM, these patients were divided into DM group (n = 189) and Non-DM group (n = 101).

Demographic information, laboratory findings, underlying diseases, hospital length of stay (LOS), the incidence of intensive care unit (ICU) admission and extrahepatic migratory infection (EMI) were obtained from medical records. Diagnostic criteria for PLA, DM, EMI, biliary tract disease and digestive system cancer were as previously described (Chang et al., 2015; Ren et al., 2020; Wang et al., 2022).

From January 2020 to December 2021, a total of 62 pus samples were prospective and continuously collected for conventional culture and 16S rDNA sequencing. According to the presence or absence of DM, these samples were divided into DM group (n = 38) and Non-DM group (n = 24). Of these, 38 pus samples (DM group, n = 22; Non-DM group, n = 16) were collected for untargeted metabolomics analysis. All collected pus samples were analyzed and no samples were artificially removed. The workflow of this study is shown in Figure 1.

The inclusion criteria were: (a) presence of a focal lesion or lesions in the liver on contrast enhanced CT images; (b) frank pus aspirated from the abscess cavity through diagnostic, and/or surgical drainage procedures; and (c) positive microbiological culture results from liver abscess and/or blood cultures.

The methods used for abscess drainage and pus collection were as described in a previous study (Song et al., 2014). All percutaneous procedures which strictly follows the principle of asepsis were performed under ultrasound guidance. Eighteen gage P.T.C Chiba needles of varying lengths were used to puncture the abscesses. The aspirated isolates including cultures were sent for microbiological analysis, 16S rDNA gene sequencing and untargeted metabolomics analysis. Under fluoroscopic guidance, 2–4 mL of undiluted contrast media was instilled into the abscess cavity through the 18G needle and then a 0.035-inch wire was inserted into the abscess cavity. After serial dilatation, an 8.5F pigtail catheter was inserted into the abscess. After collection, pus samples were stored at −80°C in a biological sample bank until used for 16S rDNA sequencing and untargeted metabolomics analysis.

The full details regarding microbiome methodology are provided in the Supporting Information. The 16S rDNA sequencing was performed to characterize the pus microbiome composition. Length-heterogeneity PCR fingerprinting was routinely used to rapidly survey our samples and standardize the community amplification. We then characterized the microbial taxa associated with the pus microbiome using Multitag Sequencing of samples. Statistical data analysis was performed based on the feature table and feature sequence obtained as described in the Supporting Information. The full analysis script is available in the Supporting Information. Alpha and beta diversity analyses were calculated using QIIME2 and all the graphs were constructed using the R package (R Foundation for Statistical Computing, Vienna, Austria). PICRUSt was used for inferred metagenomic analyses as previously described (Puri et al., 2018). Significant difference analysis was performed to compare the differential species at the genus level between the two groups (p < 0.05). Enrichment analysis of abundant taxa or functions between the two groups was performed using the linear discriminant analysis effect size (LEfSe) with a threshold logarithmic linear discriminant analysis (LDA) score set at 3.0. BugBase was used to predict the bacterial phenotype (Meyer et al., 2008; Liao et al., 2022).

The full details regarding untargeted metabolomics analysis are provided in the Supporting Information. Here, we briefly describe the sample analysis. Partial Least-Squares Discriminant Analysis (PLS-DA) was conducted using metaX to discriminate the different variables between groups and the variable importance for the projection (VIP) value was calculated. A VIP cut-off value of 1.0 was used to select important features. A volcano plot was used to compare differential metabolites between the two groups. Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis was performed on significantly different metabolites (satisfying ratio ≥ 2 or ratio ≤ 0.5; a value of p ≤ 0.05; VIP > 1). All graphs were constructed using the R package (R Foundation for Statistical Computing).

Continuous variables were expressed as the mean ± standard deviation or median (interquartile range). Categorical variables were expressed as absolute numbers with frequencies (%). Differences between continuous data were analyzed using Student’s t-test or the Mann–Whitney U-test. Differences between categorical data were analyzed using the chi-squared test or Fisher’s exact test, as appropriate. All statistical analyses were performed using SPSS statistical software for Windows, version 25.0 (IBM, Armonk, NY, United States). Two-tailed p-values < 0.05 were considered statistically significant. Correlation heatmaps with signs were performed using the OmicStudio tools. Scatter plots were constructed using Prism 8 software (GraphPad, San Diego, CA, United States).

A total of 290 PLA patients (DM, n = 189; non-DM, n = 101) were studied (Table 1). As expected, there are a number of laboratory findings of PLA patients in both DM and non-DM groups falling outside the reference range. For example, leukocyte, neutrophils (NEUT),NEUT% and so on. Compared with the non-DM group, PLA patients with DM had a higher C-reactive protein (CRP, an inflammatory marker)level, higher EMI incidence and longer LOS, which showed that PLA patients with DM have more severe clinical manifestations.

To determine the effects of DM on the pus microbiome composition in PLA patients, pus samples of 62 consecutive PLA patients (DM, n = 38; non-DM, n = 24) were performed using 16S rDNA sequencing. The clinical data of these 62 patients are shown in Table 2. As expected, compared with non-DM patients, PLA patients with DM had a higher CRP level. Although there was a statistical difference in platelets (PLT) in Table 2, the PLT in both groups were all within the reference range and had no clinical significance. Besides, there was no statistical difference in PLT in Table 1. Therefore, we speculate that the statistical difference in PLT in Table 2 may be due to the small sample size.

Supplementary Table S1 shows that 16SrDNA sequencing was able to detect the microbiome composition in conventionally cultured negative samples and was able to detect more bacteria in samples that were conventionally cultured as a single strain.

Alpha diversity represents the species evenness and richness within the microbiota, while beta diversity represents the shared diversity within the microbiota at different ecological distances (Yang et al., 2022). The Chao 1 index describing species richness revealed no significant difference (DM vs. non-DM; p = 0.12) (Figure 2A). In addition, the Shannon diversity index did not show significant difference between the DM vs. non-DM patients; p = 0.89 (Figure 2B). We then evaluated the beta diversity of the two groups. Using Bray Curtis PCoA, we found some differences in species classifications between DM and non-DM patients (Figure 2C), but there was no statistical difference. We then conducted a correlation analysis between Chao1, Simpson, Shannon and Pielou-e indices of α diversity and clinical data, which suggested (Supplementary Table S2) that the Simpson and Pielou-e indices were negatively correlated with glycosylated hemoglobin HbA1c levels (Figures 2D,E).

To determine the effects of DM, we first compared the microbial taxa in DM vs non-DM patients. Compared with non-DM patients, at the genus level, DM patients showed significantly greater relative abundance of Klebsiella (Figure 2F). Additionally, DM patients showed lesser levels of the genera, Acidovorax, Enterococcus, Bacteroides, Citrobacter and Streptococcus, when compared with the non-DM group.

The statistical results of LEfSe included two parts, namely, a histogram of the LDA value distribution and a cladogram. The histogram of LDA values (Figure 3A) showed that there were 39 differentially abundant taxa at different taxonomic levels, of which 24 were from the DM group and 15 were from the non-DM group. The genus Klebsiella had the largest LDA score. The cladogram (Figure 3B) showed that the relative abundance of genus Klebsiella was greater in DM patients, when compared with non-DM patients.

Taxa relative abundance in the DM and non-DM groups for the discriminating genera is shown in Figure 3C. There were 17 discriminating genera between the two groups at the genus level, among which Klebsiella was the most discriminating taxa.

To determine the functional relationships of microbiome composition and laboratory findings, we performed correlation analysis between the relative abundance of the top 10 microbiota at the genus level and using laboratory findings (Figure 3D). The results showed that the relative abundance of Klebsiella was positively correlated with blood glucose and CRP levels, which suggested that the relative abundance of Klebsiella was related to a high glucose environment and severe inflammatory reaction. The relative abundance of Acidovorax and Faecalibacterium was negatively correlated with CRP levels, but there was no correlation with blood glucose levels.

As shown by the potential predictions of phenotypic functions in the bacterial community in PLA, nine potential microbial phenotypes were detected. We observed a significantly higher representation of those containing mobile elements bacteria, facultatively anaerobic bacteria, potential pathogenic bacteria, bacteria related to biofilm formation and stress-tolerant bacteria in the DM group, when compared with the non-DM group (Table 3).

PICRUSt2 was used for functional profiling of the microbial community by using the KEGG database. Finally, we identified 38 KEGG pathways at level 3, which were significantly changed in percentage means between the two groups (p < 0.01). The top 30 pathway names sorted by p-values are included in Figure 4, with the most significant differential metabolic pathway involving ABC transporters.

To further identify metabolomic changes in PLA patients with DM, we performed a non-targeted metabolomics study of 38 pus samples (DM, n = 22; non-DM, n = 16). PLS-DA, which has good discrimination ability, was used to analyze the metabolic profiles based on class information (Yu et al., 2021). The PLS-DA model showed that there was separation between the two groups (Figure 5A).

A volcano plot (Figure 5B) was constructed to show trends in differentially abundant metabolites between the two groups. Using KEGG enrichment analysis (Supplementary Figure S1), seven main pathways (Cellular Processes, Drug Development, Environmental Information Processing, Genetic Information Processing, Human Diseases Metabolism and Organismal Systems) were identified, with the most abundant being Metabolism. In this pathway, we found that the differential metabolites enriched in the DM group were mainly involved in amino acid biosynthesis and metabolism, lipid metabolism, glucose metabolism and polycyclic aromatic hydrocarbon degradation.

In the environmental information processing pathway, we found the ABC transporters pathway showed differences between the two groups, which was consistent with the 16SrDNA prediction. Based on variable importance in the projection values >1 in the loading plot, FC ≥ 2 or FC ≤ 0.5, and p < 0.05 with KEGG annotations, seven differentially accumulated metabolites were identified in the ABC transporters pathway, of which four metabolites (2′-deoxyuridine, 2′-deoxyinosine, 3-dehydroxycarnitine, and inosine) were enriched and three metabolites (phosphoric acid, taurine, and orthophosphate) were depleted (Supplementary Table S3; Figure 5C).

To identify functional relationships of pus microbiota and differentially accumulated metabolites in the ABC transporters pathway, we performed correlation analysis based on Pearson’s correlation coefficients. The top 10 relative abundance microbiota at the genus level and the above seven differentially-accumulated metabolites in the ABC transporters pathway were included for analysis, which showed that the metabolites was correlated with pus microbiota. The heatmap of the correlation is shown in Figure 5D. The results showed that phosphoric acid, taurine, and orthophosphate were positively correlated with Acidovorax, Caulobacter, Citrobacter, and Pseudomonas, while they were negatively correlated with Klebsiella. The 2′-deoxyuridine, 2′-deoxyinosine, and 3-dehydroxycarnitine were negatively correlated with Acidovorax and Caulobacter.

Figure 5E shows the correlation heatmap for the seven differential metabolites in the ABC transporters pathway and the laboratory findings. The results suggested that all these metabolites were closely related to fasting blood glucose (FBG). The 2′-deoxyuridine, 2′-deoxyinosine, 3-dehydroxycarnitine and inosine were positively related to FBG, while phosphoric acid, taurine and orthophosphate were negatively correlated with FBG. These metabolites were also related to albumin (ALB), alanine aminotransferase (ALT) and albumin-bilirubin (ALBI), which are closely correlated with liver function.