Enhanced pathogen detection and gut microbiome alterations in pyogenic liver abscess: insights from next-generation sequencing

Abstract

Objectives

Pyogenic liver abscess (PLA) is a life-threatening infection with high mortality in Asia. Although Klebsiella pneumoniae is commonly implicated, emerging data suggest a more diverse microbial spectrum. This study investigated pathogen detection using conventional culture and next-generation sequencing (NGS) and characterized gut microbiome alterations in PLA patients compared to healthy controls.

Method

This was a prospective, multicenter cohort study conducted across eight tertiary hospitals. We enrolled 100 PLA patients who underwent percutaneous aspiration. Abscess aspirates underwent both conventional culture and 16 S rRNA-based NGS. Stool samples from PLA patients and 100 healthy controls were analyzed for gut microbiome composition using NGS.

Results

Culture positivity was 82%, with abscess cultures positive in 77 cases and blood cultures in 32. K. pneumoniae was the most frequently isolated pathogen (67%), and polymicrobial infections were identified in only 3% of cases by culture. NGS of abscess aspirates was available in 92% of patients, including 15 culture-negative cases. NGS identified polymicrobial infections in 16.3% of patients—more than fivefold higher than culture. Among 77 patients who underwent both culture and NGS, 13 (16.9%) showed discordance, mostly due to polymicrobial or anaerobic organisms identified by NGS. Stool NGS analysis revealed significantly reduced alpha diversity in PLA patients compared to healthy controls (Shannon index 2.9 vs. 3.5, p < 0.001), increased abundance of Enterococcus species (27.1% vs. 8.6%, p < 0.001), and depletion of SCFA-producing genera including Faecalibacterium, Roseburia, and Lachnospira species. Despite K. pneumoniae dominance in abscesses, its stool abundance did not significantly differ between PLA patients and controls.

Conclusion

NGS improves the detection of anaerobes and mixed infections in PLA. The gut microbiota of PLA patients shows marked dysbiosis, suggesting a potential role in disease pathogenesis and future therapeutic targets.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13099-026-00799-4.

Introduction

Over the past decades, pyogenic liver abscess (PLA) has continued to pose a considerable clinical challenge globally despite significant advancements in diagnostic imaging, antimicrobial therapy, and interventional drainage techniques [1].

In East Asia, including Taiwan and South Korea, the burden of PLA is particularly noteworthy [2, 3]. Our previous investigations have revealed a concerning trend; the incidence of PLA has been steadily increasing in recent years [4]. This rise occurs despite a generally high standard of medical care in South Korea, suggesting that underlying factors, potentially related to host susceptibility or microbial virulence, require further elucidation. Compounding this issue, we have also observed that the mortality rate associated with PLA in the Korean population has remained stubbornly high, showing no significant decline over several decades [4]. In our previous studies, we demonstrated that anaerobic bacteria may play a more critical role in the pathogenesis and treatment of liver abscesses than previously appreciated [5]. Notably, empirical antibiotic regimens that included anaerobic coverage were associated with significantly improved survival outcomes [5]—a finding that contrasts with the current emphasis on Klebsiella species as the primary causative agents of PLA.

Traditional diagnostic approaches, such as blood and abscess fluid cultures, have limitations in identifying the full spectrum of pathogens, particularly anaerobic and fastidious organisms. Recent advancements in sequencing technologies—particularly next-generation sequencing (NGS)—have enabled a more comprehensive and culture-independent analysis of microbial communities [6–8]. However, the application of NGS to PLA remains limited, and few studies have directly compared its performance to conventional culture methods.

Moreover, while the gut-liver axis has been increasingly recognized in various hepatic conditions, its role in liver abscess formation has not been fully elucidated [9, 10]. The gut microbiota may serve as an important reservoir for pathogenic translocation, especially in the setting of mucosal barrier disruption [11]. Yet, few studies have examined the stool microbiome in patients with liver abscesses, and even fewer have compared these findings to healthy individuals.

Therefore, this study was designed to conduct a comprehensive microbial investigation in Korean patients with pyogenic liver abscess. We aimed (1) to compare the diagnostic utility and microbial profiles obtained from traditional culture versus abscess-based NGS, (2) to characterize the gut microbiome in PLA patients using stool NGS and compare it with that of a healthy control group, and (3) to explore potential associations between the microbial findings in the abscess and the gut, seeking to provide a more holistic understanding of PLA microbiology and potentially identify novel targets for diagnosis and therapy.

Methods

Study design and setting

This was a prospective, multicenter cohort study conducted at eight university-affiliated tertiary hospitals in the Gyeongin region of South Korea. The study enrolled adult patients hospitalized with PLA between January 1, 2021, and the end of the enrollment period (November 31, 2024). This study was approved by the Institutional Review Board of Soonchunhyang University Bucheon Hospital (approval number: SCHBC 2021-08-014; approval date: October 7, 2021) and the Ethics Committee of GC Laboratories (GCL-2024-1055-01) and was registered in the Clinical Research Information Service (CRIS) under the registration number KCT0008934. Informed consents were acquired from all study participants.

Patient enrollment and inclusion criteria

Eligible patients were those diagnosed with PLA who underwent percutaneous aspiration or drainage, allowing direct sampling of abscess material. All patients suspicious of liver abscess in CT images were screened for inclusion. Patients with amebic liver abscess confirmed by microscopy of abscess fluid were excluded. For each included patient, blood culture, abscess fluid culture, and abscess NGS were routinely performed. Stool NGS was additionally conducted in patients who agreed to provide the stool. In this prospective study, abscess fluid samples were obtained during percutaneous drainage before the initiation of antibiotic therapy. However, for stool NGS, while we aimed for collection upon admission, some specimens were obtained shortly after starting treatment in cases where immediate defecation was difficult due to patient factors such as constipation.

For comparison with PLA patients, healthy controls were recruited from individuals undergoing routine health examinations at the same participating centers. Eligibility criteria included: absence of known liver disease, no clinical suspicion of infection at the time of sampling, no antibiotic exposure within the preceding three months, and no history of immunosuppressive or probiotic therapy. Age- and sex-matching (1:1) with PLA patients was performed for stool microbiome analysis.

Classification of liver abscess patients based on microbial etiology

The patients were classified into mono-microbial and poly-microbial infections based on the culture or NGS results, respectively. With culture results, patients were classified as monomicrobial when only one kind of bacteria was identified from blood or liver abscess aspirate and polymicrobial infection when more than two species of bacteria were identified. With NGS results, the patients were grouped into a monomicrobial infection when the NGS results revealed more than 99.9% of the bacteria DNA was from a single bacteria species.

Sample collection and laboratory methods

Abscess fluid samples were obtained via percutaneous catheter drainage (PCD) or aspiration using a sterile swab kit. The samples were subjected to both conventional microbiological culture and 16 S rRNA gene sequencing. NGS was performed by GC Genome (Yongin, South Korea). All samples were transported to the laboratory within one week of collection and stored under refrigerated conditions (1–10 °C) before shipment.

DNA extraction and 16 S rDNA sequencing

Fecal samples and liver abscess specimens were collected using a stool collection kit (NBgene-GUT kit; Noble Biosciences, Republic of Korea) containing preservatives. All samples arrived within three days of sampling at GC Genome Corp. To minimize any alterations in the microbial profile, DNA extraction was performed immediately upon arrival of the samples using the Chemagic DNA Stool Kit (PerkinElmer, USA) with a modified bead-beating pretreatment step from fecal samples and the MAX™ Microbiome Ultra Nucleic Acid Isolation Kit (ThermoFisher Scientific, Waltham, MA, USA) from liver abscess samples. The V4 hypervariable region was amplified using a NEXTflex 16 S V4 Amplicon-Seq kit (BioO Scientific, Austin, TX, USA) and sequenced using an Illumina MiSeq Reagent Kit v2 (500 cycles) following the manufacturer’s protocol.

Outcome measures

The primary outcome of this study was the sensitivity of conventional culture versus 16 S rRNA NGS for microbial detection in liver abscess samples. Secondary outcomes included the concordance between NGS and culture results in patients with dual-positive findings and the characterization of stool microbiome profiles according to the causative organism of liver abscess. This prospective multicenter study followed patients during hospitalization and for 1 year after discharge.

Sample size

All eligible patients admitted during the study period were prospectively screened. Due to budgetary constraints associated with NGS testing, the study aimed to include a maximum of 100 patients for full NGS analysis.

Bioinformatics and statistical analysis

Continuous variables were compared using the Wilcoxon p-value as appropriate. Categorical variables were analyzed using the chi-square test. Logistic regression models were used to identify clinical or microbiological predictors of outcomes.

At least 20,000 reads were obtained per sample and sequence reads were analyzed using the QIIME 2 framework. Demultiplexed and primer-trimmed data were quality-filtered and denoised using DADA2 by default parameter [12, 13]. Amplicon sequence variants (ASVs) with fewer than 10 reads or present in only a single sample were removed. To ensure data quality and exclude background noise, a No Template Control (NTC) was included in each sequencing batch, and samples showing predominantly contaminant profiles were excluded. Given that polymicrobial abscess samples contained a limited number of dominant taxa, this filtering strategy was unlikely to remove clinically meaningful low-abundance pathogens. To ensure the accuracy of species-level identification for the Enterobacteriaceae family, all ASV sequences initially classified into this group were manually verified using BLAST against the NCBI RefSeq database. Only ASVs showing high-identity matches to K. pneumoniae were assigned to that species. Taxonomy was assigned to each ASV using the naive Bayes machine learning taxonomy classifiers in the q2 feature classifier against the NCBI RefSeq database with taxonomic weight assembly using q2 clawback [14, 15]. For pathogen identification using NGS, a microorganism was considered positive when its relative abundance in the abscess sample exceeded 0% (i.e., at least one classified read after DADA2 denoising, quality filtering, and taxonomic assignment). Relative abundance was calculated using the number of assigned reads divided by the total quality-filtered reads per sample.

Alpha diversity was analyzed based on Shannon’s diversity index. Principal coordinates analysis based on weighted UniFrac distances was used to construct a visualization of the data. Differences in beta diversity measures between groups were assessed by ANOSIM [16]. ANOSIM p-values were generated using the Bray–Curtis measure of dissimilarity. For differential abundance analysis, we used Analysis of Compositions of Microbiomes with Bias Correction (ANCOM-BC), which estimates the unknown sampling fractions and corrects the bias induced by their differences among samples [17]. Differences between the groups were analyzed using the Wilcoxon test.

Results

Patient demographics and clinical characteristics

A total of 100 patients with pyogenic liver abscesses were included in the study (Table 1). The mean age was 59.99 ± 11.78 years, and 53% were male. The mean height and weight were 164.7 ± 8.2 cm and 64.9 ± 11.5 kg, respectively, with a mean body mass index of 24.8 ± 3.8 kg/m². Several comorbidities were noted: 37% of patients had diabetes mellitus, 40% had hypertension, 25% had dyslipidemia, 12% reported alcohol consumption, and 26% were current smokers. Chronic kidney disease and cirrhosis were present in 7% and 8% of patients, respectively. Additionally, 16% had a prior history of malignancy, and 5% were on proton pump inhibitor therapy at the time of presentation.

Table 1.

Baseline characteristics

Characteristic	N = 100
Anthropometrics
Age	59.99 ± 11.78
Gender: Female	47 (47.0%)
Gender: Male	53 (53.0%)
Height (cm)	164.71 ± 8.16
Weight (kg)	64.91 ± 11.51
Body mass index (kg/m2)	24.78 ± 3.79
Comorbidities & History (%)
Alcohol drinking	12 (12.0%)
Active smoker	26 (26.0%)
Diabetes	37 (37.0%)
Hypertension	40 (40.0%)
Dyslipidemia	25 (25.0%)
Chronic kidney disease	7 (7.0%)
Cirrhosis	8 (8.0%)
History of malignancy	16 (16.0%)
Current use of proton pump inhibitor	5 (5.0%)
Laboratory Values
White blood cell count (x103/uL)	14.13 ± 6.20
Neutrophil (%)	81.84 ± 10.39
Lymphocyte (%)	9.67 ± 8.87
Neutrophil-lymphocyte ratio	18.55 ± 20.79
CRP (mg/dL)	16.95 ± 6.68
Hb (g/dL)	11.63 ± 1.99
Platelet (x103/uL)	249.97 ± 154.15
Glucose (mg/dL)	148.12 ± 77.94
HbA1c (%) (N = 63)	6.90 ± 1.83
AST (IU/L)	86.02 ± 135.71
ALT (IU/L)	84.07 ± 100.71
Alkaline phosphatase (IU/L)	201.64 ± 136.22
Total bilirubin (mg/dL)	1.26 ± 1.10
Serum albumin (g/dL)	3.38 ± 0.62
PT INR	1.17 ± 0.16

Continuous variables are presented as mean ± standard deviation (SD), and categorical variables are expressed as number (percentage)

Laboratory results indicated a systemic inflammatory response, with a mean white blood cell count of 14.13 ± 6.20 × 10³/µL and a mean CRP of 16.95 ± 6.68 mg/dL. Neutrophil percentage averaged 81.8%, while lymphocyte percentage was reduced (9.67%), resulting in a markedly elevated neutrophil-to-lymphocyte ratio (NLR: 18.55 ± 20.79). Liver function tests showed elevated AST (86.0 ± 135.7 IU/L), ALT (84.1 ± 100.7 IU/L), and alkaline phosphatase (201.6 ± 136.2 IU/L).

Antibiotics regimens and clinical outcomes

Initial empirical antibiotic regimens varied among the 100 patients, with the most common being a combination of third-generation cephalosporin and metronidazole, administered in 82 patients (82.0%) (Table 2). Piperacillin/tazobactam alone was used in seven patients (7.0%), and in combination with metronidazole in five patients (5.0%). Less frequently used regimens included carbapenem plus metronidazole (2.0%), third-generation cephalosporin alone (2.0%), quinolone plus metronidazole (1.0%), and carbapenem monotherapy (1.0%). A total of 98 patients were treated with antibiotics with anti-anaerobic activity.

Table 2.

Treatment regimens and clinical outcomes

	N = 100
Antibiotics regimen
3rd generation cephalosporine + metronidazole	82 (82.0%)
Piperacillin/tazobactam	7 (7.0%)
Piperacillin/tazobactam + metronidazole	5 (5.0%)
Carbapenem + metronidazole	2 (2.0%)
3rd generation cephalosporine only	2 (2.0%)
Quinolone + metronidazole	1 (1.0%)
Carbapenem only	1 (1.0%)
Clinical Course & Management
Length of hospital stay (days)	16.0 (13.0–21.0)
Duration of percutaneous drainage placement (days)	10.0 (7.0–16.0)
Metastatic infection *	8 (8.0%)
Lung	4 (4.0%)
Vertebra	1 (1.0%)
Prostate	2 (2.0%)
Eye	2 (2.0%)
Death	0 (0%)

* Can occur in multiple instances in the same patient

Variables are presented as number (percentage)

The median length of hospital stay was 16.0 days (interquartile range [IQR], 13.0–21.0), and the duration of percutaneous drainage placement was 10.0 days (IQR, 7.0–16.0). Eight patients (8.0%) developed metastatic infections, including pulmonary (4.0%), vertebral (bony) (1.0%), prostatic (2.0%), and ocular (2.0%) involvement. Half of the patients with metastatic infection were caused by K. pneumoniae. Importantly, there were no deaths reported during hospitalization in this cohort.

When comparing patients with bacteremia (blood and abscess culture-positive, n = 27) to those with localized infection (abscess culture-positive only, n = 50), patients with bacteremia (n = 27) had a significantly higher burden of comorbidities, including diabetes (55.6% vs. 22.0%, p = 0.003) and hypertension (59.3% vs. 28.0%, p = 0.007), compared to those with localized infection (n = 50) (Supplementary Table 1). The bacteremic group also exhibited a longer mean hospital stay (19.37 vs. 16.22 days; p = 0.049) and higher neutrophil percentage (85.74% vs. 80.79%; p = 0.046). However, differences in CRP levels (p = 0.304) and metastatic infection rates (p = 0.236) were not statistically significant.

Extended pathogen distribution based on culture and NGS

Among all patients, blood cultures were positive in 32 cases (32.0%), and abscess cultures were positive in 77 (77.0%), with an overall culture positivity rate of 82.0%. K. pneumoniae was the most common pathogen, isolated in 67 patients (67.0%) across all culture types—26 from blood and 64 from abscess fluid (Table 3). Other identified organisms included Escherichia coli (5.0%), Morganella morganii (2.0%), viridans streptococcus (5.0%), and polymicrobial infections in 3.0%. Polymicrobial combinations included E. coli + Bacteroides fragilis, K. pneumoniae + E. faecium, and K. pneumoniae + Burkholderia cepacia.

Table 3.

Microbial detection in liver abscess: comparison between conventional culture and next-generation sequencing

Microorganisms isolated from blood and abscess cultures
N = 100	Blood culture		Abscess culture		Culture overall
Positivity rate	32 (32.0%)		77 (77.0%)		82 (82.0%)
Microorganism	K. pneumoniae 26 (26.0%) E. coli 1 (1.0%) M. morganii 1 (1.0%) B.cepacia 1 (1.0%) Viridans streptococcus 1 (1.0%) Polymicrobial 2 (2.0%) (E.coli + B.fragilis, K.pneumoniae + E.faecium)		K. pneumoniae 64 (64.0%) E. coli 5 (5.0%) M. morganii 2 (2.0%) Viridans streptococcus 5 (5.0%) Polymicrobial 1 (1.0%) (E.coli + B.fragilis)		K. pneumoniae 67 (67.0%) E. coli 5 (5.0%) M. morganii 2 (2.0%) Viridans streptococcus 5 (5.0%) Polymicrobial 3 (3.0%) (E. coli + B. fragilis, K. pneumoniae + E. faecium, K. pneumoniae + B. cepacia)
NGS results compared to conventional culture
N = 100	Culture overall	NGS overall		NGS results in culture-positive cases		NGS results in culture-negative cases
Total number	82	92		77		15
Microorganism	K. pneumoniae 67 (67.0%) E. coli 5 (5.0%) M. morganii 2 (2.0%) Viridans streptococcus 5 (5.0%) Polymicrobial 3 (3.0%)	K. pneumoniae 71 (71.0%) Fusobacterium nucleatum 2 (2.0%) Burholderia 1 (1.0%) Viridans streptococcus 3 (3.0%) Polymicrobial 15 (15.0%)		K. pneumoniae 62 (62.0%) viridans streptococcus 3 (3.0%) Polymicrobial 12 (12.0%)		K. pneumoniae 9 (9.0%) Fusobacterium nucleatum 2 (2.0%) Burholderia 1 (1.0%) Polymicrobial 3 (3.0%)

NGS of abscess samples was performed in all 100 patients, and test results were available in 92 patients (Fig. 1). We suppose that the main cause of failure of NGS could be low bacterial burden due to the error in the sample collection and transportation process. For NGS negative samples, the PCR for Entamoeba histolytica was performed, and all the results were negative. The results of 77 patients who had both positive culture and NGS results and 15 patients with NGS results were analyzed. NGS revealed a broader microbial spectrum than culture. While K. pneumoniae remained dominant (71.0%), NGS additionally identified Fusobacterium nucleatum (2.0%) and Burkholderia cepacia (1.0%) as causes of mono-microbial infection. (Table 3). The polymicrobial infections comprised 16.3%—more than a fivefold increase compared to culture (3.0%). The polymicrobial infections were mainly caused by the Prevotella, Bacteroides, and Campylobacter species.

Fig. 1.

Study cohort flow with NGS results

In culture-positive cases, 3.2% of the PLA caused by K. pneumoniae was identified as polymicrobial infection by NGS, whereas 75% of PLA caused by non-K. pneumoniae was identified as polymicrobial.

Concordance between microbiological diagnostic methods

The concordance between NGS and conventional culture was assessed among patients who tested positive by both methods. Venn diagrams showing the overlap between positive blood cultures (N = 32) and abscess cultures (N = 77) are shown in the supplementary Fig. 1. Among 27 patients who revealed both blood and abscess culture positive, 93% (n = 25) showed concordant pathogen results, with only 2 cases showing discordance (Supplementary Fig. 1). Similarly, a comparison between culture and abscess NGS positive in 77 patients showed concordant results in 64 cases (83.1%), with 13 (16.9%) showing discordance. Most discrepancies involved the detection of polymicrobial infections by NGS, whereas conventional culture identified a single or different organism.

Stool Microbiome profiles and comparison with healthy control

Stool NGS was performed in 65 liver abscess patients and compared with 100 age- and sex-matched healthy controls. Among the 49 patients with K. pneumoniae-dominant liver abscess (KLA), 63.3% exhibited a prominent Klebsiella signature in stool samples, compared to 66% of healthy controls. In contrast, non-KLA patients (n = 16) showed higher proportions of polymicrobial signatures and more variable stool profiles.

Figure 2A illustrates the taxa bar plots comparing the three groups—KLA, non-KLA, and controls—revealing clear differences in taxonomic composition at the phylum and genus levels. Both K. pneumoniae and non-KLA patient groups exhibited significantly lower microbial diversity compared to healthy controls (Wilcoxon p < 0.001, Fig. 2B). However, no significant difference was observed between the KLA and non-KLA groups (p = 0.186). To further illustrate the variability in stool microbial composition among individuals, we additionally generated sample-level stacked bar plots for all participants. These plots are provided as Supplementary Fig. 2. Individual-level microbiome composition can also be inspected interactively through the accompanying QIIME2 visualization file.

Fig. 2.

Taxonomic composition of the stool microbiome and stool microbial diversity in liver abscess patients and controls (A) Relative abundance of stool microbiota in K. pneumoniae, non-K. pneumoniae patients, and controls (taxa bar plot), (B) Alpha-diversity (Shannon index), and Beta-diversity (PCoA using Bray-Curtis dissimilarity)

Regarding beta diversity, the control group showed significant compositional differences from both patient groups (K. pneumoniae and non-K. pneumoniae, ANOSIM p = 0.001), whereas no significant difference was observed between the two patient groups (ANOSIM p = 0.849). The overall three-group comparison showed a borderline significance (ANOSIM p = 0.055).

Microbial link between gut and abscess

Differential abundance analysis at the genus level showed that both KLA and non-KLA patient groups showed significant depletion of several short-chain fatty acid (SCFA)-producing genera compared to controls, such as Faecalibacterium, Roseburia, and Lachnospira (Supplementary Fig. 3 A, Fig. 3B). No genera were significantly different between KLA and non-KLA groups. Similarly, analysis at the family level revealed significant reductions in SCFA-producing families, including Ruminococcaceae and Lachnospiraceae, in both KLA and non-KLA groups compared to controls (Supplementary Fig. 3 C). As with genus-level analysis, no significantly different taxa were observed between KLA and non-KLA groups. However, Enterococcus showed increased relative abundance in both abscess groups compared to controls (p < 0.001), although the difference between the two patient groups was not significant (Fig. 3A).

Fig. 3.

Differential microbiome features in the liver abscess and stools (A) Comparison of relative abundance of taxa in the liver abscess, (B) Relative abundance of Klebsiella pneumoniae in stool samples

Contrary to expectations, the relative abundance of K. pneumoniae in stool samples was not significantly higher in KLA patients compared to non-KLA patients or healthy controls (Fig. 3B). Despite the predominance of K. pneumoniae in abscess cultures among KLA patients, this trend was not mirrored in the gut microbiome. The relative abundance distribution of Klebsiella spp. across all individuals is summarized in Supplementary Table 2.

Discussion

This prospective, multicenter study investigated the utility of NGS in improving pathogen detection in PLA and explored gut microbiome alterations in affected patients compared to healthy controls. We found that NGS was more sensitive than conventional culture methods, particularly in identifying anaerobic and polymicrobial infections, and also discovered that PLA patients exhibited significant gut microbial dysbiosis.

Our motivation for this study stemmed from our previous epidemiologic observations in South Korea, where the incidence of PLA has steadily increased over the past decade [4]. Surprisingly, despite the country’s advanced medical infrastructure, mortality rates associated with PLA have remained largely unchanged. This discrepancy prompted us to investigate factors related to patient outcomes. In our prior analysis, we found that the use of empirical antibiotics with anaerobic coverage was significantly associated with improved survival, whereas patients who received regimens without anaerobe coverage had worse outcomes [5]. This analysis led us to question whether anaerobes might still play a substantial role in PLA pathogenesis, even in an era where K. pneumoniae is commonly reported as the predominant causative pathogen.

NGS substantially improved the pathogen detection rate, identifying organisms in 15% of patients who had negative culture results. We suppose the negative results of the NGS originated from the suboptimal specimen storage or transportation, considering E. histolytica PCR results were uniformly negative for those samples. Importantly, NGS detected polymicrobial infections in 15% of patients—five times higher than the rate detected by culture. Many of the discordant results between culture and NGS were due to anaerobic or fastidious organisms, reflecting the difficulty in routine culture systems to identify those bacteria. These findings suggest that anti-anaerobe coverage is essential, especially when non-K. pneumoniae bacteria were identified from the blood or liver abscess culture [18, 19].

Another important aspect of our study is the exploration of the gut-liver axis. Since PLA often arises via ascending infection through the portal vein, we hypothesized that alterations in gut microbiota may contribute to abscess formation [20]. Stool NGS analysis revealed marked dysbiosis in PLA patients, including reduced alpha diversity and depletion of SCFA-producing genera. These microbes play a key role in maintaining intestinal barrier integrity and immune modulation [21–23]. Our findings are consistent with previous studies reporting gut microbial imbalance in hypervirulent K. pneumoniae infections [24–26].

Several epidemiologic studies have reported an increased incidence of colorectal cancer among patients with pyogenic liver abscess, suggesting a potential link through the gut–liver axis [27]. Additionally, a meta-analysis by Mohan et al. [28] reported that patients with cryptogenic PLA had a sevenfold increased risk of colorectal cancer compared to controls. This association suggests a potential microbial or mucosal basis for this link. Colorectal cancer is well known to be accompanied by profound alterations in gut microbial composition and barrier function, both of which may facilitate microbial translocation into the portal circulation [29]. Therefore, the observed co-occurrence of PLA and colorectal cancer may reflect a shared pathophysiologic pathway rooted in gut dysbiosis, lending further support to the gut-liver axis hypothesis in the development of PLA. However, colorectal cancer was not systematically evaluated in the present cohort, and this association should therefore be interpreted as background context based on prior literature rather than as a finding of this study.

Interestingly, despite the predominance of K. pneumoniae in abscess samples, its relative abundance in the stool was not significantly higher in KLA patients than in non-KLA patients or healthy controls. This unexpected result suggests that hepatic infection by K. pneumoniae may not be driven simply by gut overgrowth. However, because strain-level analysis was not feasible using 16 S rRNA sequencing, potential explanations such as strain-specific virulence or enhanced invasion capacity remain speculative. Supporting this notion, prior studies have demonstrated that hypermucoviscous K. pneumoniae (especially K1 and K2 serotypes) can translocate efficiently even when present in low abundance in the gut [30, 31]. Notably, our study suggests that stool carriage rates of K. pneumoniae in healthy Korean adults exceed 40%, indicating that colonization alone may not be sufficient to explain disease development. Instead, specific pathogenic traits of the organism or disruption of host defenses may be required for progression to hepatic infection.

Beyond pathogen-specific virulence, our findings also suggest that host–microbiome interactions may contribute to the pathogenesis of liver abscesses. Both KLA and non-KLA patients showed a profound loss of SCFA–producing commensals and an enrichment of potentially pathogenic taxa such as Enterococcus, indicating substantial gut microbial dysbiosis. Such dysbiosis has been shown to impair intestinal barrier integrity, promote bacterial translocation through increased mucosal permeability, and augment gut–liver axis inflammation [32]. Therefore, rather than the absolute abundance of K. pneumoniae or other causative organisms in the gut, a dysbiotic intestinal environment may increase the likelihood that virulent strains gain access to the bloodstream and ultimately seed the liver. This conceptual framework reconciles the clinical observation that hepatic infection can develop even when pathogenic organisms are not dominant in the gut microbiome. These observations are partly consistent with previous studies reporting gut dysbiosis in liver abscess patients, in which depletion of SCFA-producing bacteria and enrichment of pro-inflammatory taxa were implicated in bacterial translocation [33]. However, unlike some earlier reports suggesting that overgrowth of the causative pathogen in the gut predisposes to hepatic infection [34], our data did not demonstrate an increased stool abundance of K. pneumoniae in KLA patients. This discrepancy suggests that pathogen-specific proliferation may not be the primary driver of infection and that dysbiosis-mediated barrier disruption may provide an alternative mechanistic explanation [11].

Our study has several limitations. First, although this was a multicenter prospective study, our study is limited by its relatively small sample size, especially in the non-KLA subgroup, which may have led to Type II errors. Due to the high cost of NGS, the cohort size was restricted, and thus our results should be interpreted as exploratory until validated in larger populations. Accordingly, the lack of a significant difference in alpha diversity between KLA and non-KLA groups should be interpreted cautiously, as the non-KLA group was small and effect size estimation was not feasible. Second, because stool microbiome analysis was cross-sectional, causality between gut dysbiosis and pyogenic liver abscess cannot be established. Moreover, prior outpatient antibiotic or medication exposure before admission cannot be fully excluded and may have influenced baseline microbiome composition. Third, although healthy controls were matched for age and sex, other factors known to influence gut microbiota—including diabetes, lifestyle behaviors, diet, and body mass index—were not fully matched or adjusted for and may have contributed to the observed differences. Another limitation of our study is that molecular diagnostics such as quantitative PCR or NGS were not performed on blood samples of patients who were abscess-positive but blood-culture negative. At the time of enrollment, a validated protocol for blood-based pathogen sequencing/qPCR was not available among participating centers, and the study protocol focused on abscess-targeted NGS. Because this study was exploratory in nature, NGS results were not used to guide antibiotic modifications, and their impact on clinical outcomes could not be assessed. Future prospective studies incorporating real-time NGS–guided treatment decisions are warranted.

In conclusion, our study demonstrates that NGS enhances microbial detection in pyogenic liver abscesses, particularly for anaerobes and polymicrobial infections that are often missed by conventional culture. Gut microbiome analysis revealed marked dysbiosis in PLA patients. These results suggest that incorporating NGS into routine diagnostic workflows may enhance the identification of clinically relevant pathogens in PLA. In addition, the observed gut microbial alterations point to the gut-liver axis as a biologically plausible contributor to disease pathogenesis and a potential focus for future research and intervention.

Abbreviations

ANCOM-BC

Analysis of Compositions of Microbiomes with Bias Correction

ANOSIM

Analysis of Similarities

ASVs

Amplicon sequence variants

AST

Aspartate aminotransferase

ALT

Alanine transaminase

CRIS

Clinical Research Information Service

CRP

C-reactive protein

DNA

Deoxyribonucleic acid

HbA1c

Glycated hemoglobin

IQR

Interquartile range

KLA

K. pneumoniae-dominant liver abscess

NGS

Next-generation sequencing

NLR

Neutrophil-to-lymphocyte ratio

PCD

Percutaneous catheter drainage

PLA

Pyogenic liver abscess

rRNA

Ribosomal ribonucleic acid

SCFA

Short-chain fatty acid

Author contributions

Author ContributionsConceptualization: Min Jae Kim, Jeong-Ju YooMethodology: Ju Sun Song, Young Kul Jung, Min Jae Kim, Jeong-Ju YooInvestigation: Solbi Kweon, Seong-Hee Kang, Hyung Joon Yim, Seung Kak Shin, Gwang Hyeon Choi, Eun Sun Jang, Hae Lim Lee, Sung Won Lee, Jung Hwan Yu, Ji Eun Han, Soon Sun Kim, Sang Gyune Kim, Young Seok KimData curation: Ju Sun Song, Young Kul Jung, Seong-Hee Kang, Hyung Joon YimFormal analysis: Ju Sun Song, Young Kul Jung, Jeong-Ju YooWriting – original draft: Ju Sun SongWriting – review & editing: Jeong-Ju Yoo, Min Jae KimSupervision: Min Jae Kim, Jeong-Ju YooFunding acquisition: Min Jae Kim, Jeong-Ju Yoo.

Funding

This study was funded by Korean Association for the Study of the Liver, Gyeong-In Area, Asan Medical Center fund (2020IF0005) and was supported by the Soonchunhyang University research fund.

Data availability

The raw 16 S rRNA sequencing data generated in this study are not publicly available due to ethical restrictions and patient privacy concerns, as they contain potentially identifiable human genetic information. The study protocol was approved by the Institutional Review Board of Soonchunhyang University Bucheon Hospital (SCHBC 2021-08-014), and the informed consent obtained from participants does not permit the public deposition of their individual-level data. However, processed data supporting the conclusions of this article, such as amplicon sequence variant (ASV) tables, taxonomic assignments, and diversity metrics, are available from the corresponding author, Jeong-Ju Yoo (puby17@naver.com), upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Institutional Review Board of Soonchunhyang University Bucheon Hospital (approval number: SCHBC 2021-08-014; approval date: October 7, 2021) and the Ethics Committee of GC Laboratories (GCL-2024-1055-01) and was registered in the Clinical Research Information Service (CRIS) under the registration number KCT0008934. Informed consents were acquired from all study participants.

Competing interests

The authors declare no competing interests.

Trial registration number

Clinical Research Information Service (CRIS) KCT0008934.