Longitudinal dysbiosis of gut bacteriome and mycobiome in patients with severe acute pancreatitis and its association with mortality.

This was a prospective observational study conducted in the pancreatic intensive care unit of the First Affiliated Hospital of Nanchang University, China from May 2019 to December 2020. Adult patients aged 18–75 who were diagnosed with SAP and admitted to pancreatic ICU within 72 h after the onset of symptoms were enrolled. SAP was defined in accordance with the revised Atlanta classification [ 9 ]. Exclusion criteria included (1) treatment with antibiotics or probiotics before the first sample was collected; (2) length of hospital stay less than 7 days; (3) history of abdominal surgery; and concomitant with (4) pregnancy; (5) gastrointestinal disease; (6) malignant cancer. Acute Physiology and Chronic Health Evaluation (APACHE-II) and Sequential Organ Failure Assessment (SOFA) scores were recorded by specialized pancreatic ICU physicians at admission. Basic information was also obtained from each patient, including age, sex, body mass index (BMI), etiology, history of smoking and drinking, antibiotic usage, infectious complications, and mortality within 90 days. The healthy volunteers who met the following inclusion criteria were recruited from the physical examination center: (1) asymptomatic; (2) had not taken antibiotics or probiotics within 4 weeks; (3) not pregnant; and free of chronic (4) metabolic; (5) cardiovascular; (6) gastrointestinal diseases. Age- and sex-matching processes was performed to select comparable controls. Written informed consent was obtained from each participant following protocols approved by the Medical Ethics Committee of the an First Affiliated Hospital of Nanchang University (No. 2019055). This study was retrospectively registered with the Chinese Clinical Trial Registry (registration number PID 283891). The collection of fecal samples from the patients started within 72 h after their admission and continued weekly until they were discharged or deceased. The first samples collected were labeled as T1, which represented the acute phase of the patients’ condition. The final samples collected were labeled as T3, indicating the endpoint of the patients’ treatment. The samples collected between T1 and T3 were labeled as T2, representing the progression stage of the patients’ condition. Stool samples from healthy individuals were collected once as the control group. Immediately following defecation, fresh stool samples were divided into three parts and frozen at −20 ℃, and then stored at −80 ℃ until analysis. Microbial genomic DNA was extracted from fecal samples and amplified using primers of bacterial 16S rRNA gene and the fungi internal transcribed spacer (ITS) rRNA gene. Purified amplicons were pooled in equimolar amounts and then paired-end sequenced (2 × 300) on an illumina Miseq PE300 platform (Illumina, USA) according to the standard protocols by Majorbio Bio-Pharm Technology Co. Ltd. (Shanghai, China). Raw sequencing reads were demultiplexed, quality-filtered by Fastp version 0.19.6 and merged by FLASH version 1.2.11 [ 10 , 11 ]. Operational taxonomic units (OTUs) with 97% similarity cutoff were clustered using UPARSE version 11 and chimeric sequences were removed [ 12 ]. The taxonomy of each OTU representative sequences was analyzed by RDP Classifier version 2.13 against Silva 16S rRNA database (v138) for bacteria and UNITE (8.0) for fungi using confidence threshold of 0.7 [ 13 ]. The microbial data was analyzed on the Majorbio Cloud Platform ( http://www.majorbio.com/ ) [ 14 ]. Alpha diversity indexes, including Sobs, Shannon and Chao1, were calculated by Mothur v.1.30.2 to evaluate the richness and evenness of gut microbiome. Beta diversity was measured by partial-least squares discriminant analysis (PLS-DA) using R (v3.3.1) and permutational multivariate analysis of variance (PERMANOVA) was applied to figure out whether the microbiome composition was significantly different among different groups. The microbial taxonomic distributions of each group were visualized using R (v3.3.1). Wilcoxon signed rank-sum test was used to analyze the differential taxa for paired samples. Significant taxa were reported at FDR-adjusted p  < 0.05. Co-occurrence network with nodes representing 97% cutoff OTUs and edges representing correlations between these OTUs was constructed using R package igraph. To eliminate the effects of random sampling, we repeated the operations for 1000 times echo network. We performed principal component analysis taking topological features of number of vertices, number of edges and modularity as input. We chose the representative network, whose score on the first principal component axis was the closest to the mean value of all the networks. The two representative networks were explored and visualized with the interactive platform gephi. The nodes in networks represent OTUs and the edges that connect these nodes represent correlations between OTUs. Network modularity was recalculated using the gephi built-in algorithm with a resolution of 8. The Sankey plot was drawn using R library “ggalluvial”. We used random forest classifier to predict sample type, taking stool community composition of bacteria and/or fungi, or four SCFAs as features. All available timepoints (T1-T3) were included in the random forest analysis to maximize the sample size. The genera composition data of bacteria and fungi were normalized separately before analyses. To filter out redundant features, up to 20 variables were selected by using the minimum Redundancy Maximum Relevance (mRMR) filterranking algorithm, using the R package “mRMRe” [ 15 ]. We use nested cross validation to get feature importance scores and sample predictions for all samples using the classify-samples-ncv pipeline in qiime2 (2019.4) [ 16 ]. Receiver operating characteristic (ROC) figures were drawn using R package “pROC” [ 17 ]. Fecal samples (25 mg) were complemented with 500 μl of water containing 0.5% phosphoric acid and 50 μg/ml of 2-ethylbutyric acid, and then extracted according to the manufacturer’s protocol (Majorbio Bio-Pharm Technology Co., Ltd., Shanghai, China). The contents of SCFAs were determined using an Agilent 8890B-5977B GC–MS system (Agilent Technologies Inc., USA) equipped with a HP FFAP column (30 m × 0.25 mm × 0.25 μm, Agilent J&W Scientific, USA). The ion fragments of target SCFAs were automatically identified by MassHunter quantitative software (Agilent Technologies Inc., USA, No. V10.0.707.0), and the SCFAs concentrations were calculated by standard curve. SCFAs standards were mixtures of acetic acid, propanoic acid, butyric acid. SPSS 26.0 software was used for statistical analysis and p  < 0.05 was considered statistically significant. Measurement data conforming to normal distribution were expressed as Mean ± SD. If the variance was homogeneous, two independent samples T test were selected for comparison. For variables that were not normally distributed, the approximate T test was carried out and the measurement data of skew distribution was expressed by median (quartile method). Mann–Whitney U test was performed to compare nonparametric variables of two sample groups, while multiple group comparisons were made using Kruskal–Wallis test.

A total of 73 SAP patients admitted to pancreatic ICU within 72 h and 28 healthy volunteers were enrolled in this study. The flow chart illustrating the process of patient screening was shown in Fig.  1 . The most common cause of SAP among the enrolled patients was hypertriglyceridemia (HTG), accounting for 41% of cases. Biliary pancreatitis accounted for 33% of cases, followed by alcoholic pancreatitis at 19%. Out of the 73 patients, 17 (23%) died within 90 days after admission, while 56 (77%) were survivors. Patients’ characteristics were summarized in Table  1 . Age and sex were matched between the patients and healthy controls ( p  > 0.05). Compared to survivors, non-survivors were older and had higher APACHE-II score and infection rates (Table S1). Fig. 1 Flow diagram of the patient selection process and sample collection analysis. SAP, severe acute pancreatitis Table 1 Basic demographics of patients and healthy subjects SAP ( n  = 73) Healthy controls ( n  = 28) p value Age, median years (IQR) 47(37–59) 49(43–56) 0.56 Sex, male/female 49/24 13/15 0.253 BMI, median kg/m 2 (IQR) 25(24,28) 23(20,25) 0.652 Etiology, n (%)  HTG 30(41%)  Biliary 24(33%)  Alcohol 14(19%)  Other 5(7%) APACHE-II score, median (IQR) 11(9–15) SOFA score, median (IQR) 4(3–6) Infectious complications (%) 45(62%) Hospital stay, days, median (IQR) 19(13–39) ICU stay, days, median (IQR) 12(7–25) Mortality (%) 17(23%) HTG Hypertriglyceridemia, APACHE II The Acute Physiology and Chronic Health Evaluation II score, SOFA The Sequential Organ Failure Assessment score Flow diagram of the patient selection process and sample collection analysis. SAP, severe acute pancreatitis Basic demographics of patients and healthy subjects HTG Hypertriglyceridemia, APACHE II The Acute Physiology and Chronic Health Evaluation II score, SOFA The Sequential Organ Failure Assessment score A total of 12,315,418 reads were obtained from 16S rRNA gene sequencing and ITS1 sequencing yielded 17,311,489 reads. To elucidate the initial alterations of gut microbiome, we first analyzed the stool samples from SAP patients at baseline (T1). Analysis of alpha diversity revealed that both the richness and diversity, as calculated by Sobs, Shannon and Chao1 indices, of bacteria and fungi were remarkably lower in SAP patients than in healthy controls (HC) (Fig.  2 A-B). Partial least square discriminant analysis (PLS-DA) plot showed that samples of SAP and healthy subjects were separated into two strikingly distinct groups in both bacteria and fungi community (PERMANOVA, R 2  = 0.075, p  = 0.001 for bacteria; R 2  = 0.070, p  = 0.001 for fungi) (Fig.  2 C-D). Fig. 2 Gut microbial variations in individuals with SAP. Alpha diversity of bacteria ( A ) and fungi ( B ) represented by observed species (Sobs), Shannon, Chao1 indices in SAP patients and healthy control (HC) individuals (Wilcoxon rank-sum test, * p  < 0.05, ** p  < 0.01, *** p  < 0.001). Partial least squares discriminant analysis (PLS-DA) of bacteria ( C ) and fungi ( D ) revealed compositional variations between SAP and HC. Relative abundance of bacterial and fungi at phylum ( E , F ) and genus ( H , J ) level. Bacterial ( G ) and fungi ( I ) genera with significant differences (Wilcoxon rank-sum test with Benjamine-Hochberg correction; FDR [false discovery rate] < 0.05) between SAP and HC. SAP, severe acute pancreatitis Gut microbial variations in individuals with SAP. Alpha diversity of bacteria ( A ) and fungi ( B ) represented by observed species (Sobs), Shannon, Chao1 indices in SAP patients and healthy control (HC) individuals (Wilcoxon rank-sum test, * p  < 0.05, ** p  < 0.01, *** p  < 0.001). Partial least squares discriminant analysis (PLS-DA) of bacteria ( C ) and fungi ( D ) revealed compositional variations between SAP and HC. Relative abundance of bacterial and fungi at phylum ( E , F ) and genus ( H , J ) level. Bacterial ( G ) and fungi ( I ) genera with significant differences (Wilcoxon rank-sum test with Benjamine-Hochberg correction; FDR [false discovery rate] < 0.05) between SAP and HC. SAP, severe acute pancreatitis In keeping with previous work, we found an increase in the bacterial phylum Proteobacteria and a decrease in Firmicutes and Actinobacteriota in SAP patients compared to the control group (Fig.  2 E). Furthermore, SAP group was dominated by pathogenic genus Escherichia-Shigella , Enterococcus , and Klebsiella , the abundances of which were up to 49%, compared to only 11% in healthy subjects (Fig.  2 G-H). Conversely, the potential beneficial genera such as Blautia , Bifidobacterium , and Romboutsia, were statistically less abundant in SAP patients than in healthy controls (Fig.  2 G). In the analysis of the fungal microbiota, we found that the gut mycobiome was primarily comprised of two phyla: Ascomycota and Basidiomycota. A notable feature was the increased abundance of Ascomycota in SAP patients, which was balanced by an equivalent decrease in Basidiomycota (Fig.  2 F). At the genus level, Candida , Penicillium , Saccharomyces , and Aspergillus were the most abundant fungi (Fig.  2 J). Upon specific analysis, we noted that SAP patients had higher proportions of Candida compared to healthy controls (Fig.  2 I). On the contrary, Saccharomyces and Aspergillus were more abundant in healthy individuals than in SAP patients (Fig.  2 I). Altogether, early dysbiosis of gut bacterial and fungal communities was observed at the onset of SAP. Next, we classified the patients into two groups: survivors (S) and deceased (D), and then examined the changes of gut bacterial composition from early after admission at T1, to progression (T2 and T3). In brief, both richness and diversity values were lower at T2 and T3 compared to T1 (Fig.  3 A-B). The Sobs index, a measure of species abundance, declined more sharply in non-survivors compared to the survivors (Fig.  3 A). Similarly, the Shannon index, which measures species diversity, continuously dropped in the non-survivors, while it remained relatively stable in the survivors (Fig.  3 B). When using PLS-DA and PERMANOVA on the basis of Bray–Curtis distance, separation among samples from T1, T2 and T3 was observed in survivor patients (R 2  = 0.04, p  = 0.001) (Fig.  3 C). However, no significant differences were noted in the T1-T2-T3 distances within the D group, which indicates a comparatively lesser microbial variation in non-survivor patients. Fig. 3 Gut bacteriome alterations in SAP patients with different clinical outcomes during hospitalization. Stool samples were collected from admission (T1) through progression (T2) to discharge (T3). Patients were divided into survivors and non-survivors according to their prognosis during 90-day follow-up. A - B Comparison of alpha diversity among different groups. C Profile of bacterial taxonomy was shown by PLS-DA. D Relative abundance of bacterial genera between survivors and non-survivors at different time points. E Bacterial genera with significant differences between survivors and non-survivors at admission (Wilcoxon rank-sum test with Benjamine-Hochberg correction; FDR < 0.05). F Differential bacterial genera between admission (T1) and discharge (T3) in survivors. S, survivors; D, non-survivors; T1, admission stage; T2, progression stage; T3, discharged stage. * p  < 0.05, ** p  < 0.01, *** p  < 0.001 Gut bacteriome alterations in SAP patients with different clinical outcomes during hospitalization. Stool samples were collected from admission (T1) through progression (T2) to discharge (T3). Patients were divided into survivors and non-survivors according to their prognosis during 90-day follow-up. A - B Comparison of alpha diversity among different groups. C Profile of bacterial taxonomy was shown by PLS-DA. D Relative abundance of bacterial genera between survivors and non-survivors at different time points. E Bacterial genera with significant differences between survivors and non-survivors at admission (Wilcoxon rank-sum test with Benjamine-Hochberg correction; FDR < 0.05). F Differential bacterial genera between admission (T1) and discharge (T3) in survivors. S, survivors; D, non-survivors; T1, admission stage; T2, progression stage; T3, discharged stage. * p  < 0.05, ** p  < 0.01, *** p  < 0.001 To explore differences in microbial variations across different disease stages, we compared the gut microbiota composition between survivors and non-survivors at the time of admission and during progression. In survivors, a notable depletion in the Firmicutes/Proteobacteria ratio was observed at T1 (admission), with a subsequent recovery by T3 (recovery). In contrast, no significant change in this ratio was found in the non-survivors (Fig. S1). At the genus level, Enterococcus was statistically more abundant in the non-survivors compared to the survivors, while Bacteroides was less abundant (Fig.  3 E). Among survivors, the relative abundance of Klebsiella was higher in the T2 (progression stage) samples than in the T1 (admission stage) samples. It was then found to be lower once again in the T3 (recovery stage) samples (Fig.  3 D). In contrast, the D group patients had higher levels of Klebsiella in the end stage samples compared to the admission and progression stage samples (Fig.  3 D). Moreover, the abundance of Bifidobacterium was remarkably elevated in the T3 (recovery stage) samples from survivor patients compared to those collected at T1 and T2 stages (Fig.  3 F). Of note, the overgrowth of Enterococcus , which was observed in patients compared to healthy subjects, was much more profound in non-survivors when compared to survivors at the time of admission (Fig.  3 E). Short chain fatty acids (SCFAs) are widely recognized as beneficial metabolites produced by the gut microbiota. However, their changes in patients during hospitalization have not been fully understood. To determine the alterations of SCFAs from admission to discharge, we performed a paired comparison between survivors and non-survivors. The levels of SCFAs, especially acetic acid and propionic acid, were found to increase in recovering patients (Fig. S2). In contrast, no significant variation was observed in the non-survivors. Collectively, the restoration of SCFAs is associated with the good prognosis in SAP patients. The alpha diversity indexes, including Sobs and Shannon, remained similar throughout the hospital stay for both survivors and non-survivors (Fig.  4 A-B). However, beta diversity determination using PLS-DA and PERMANOVA revealed significant differences in fungal composition with the recovery of SAP in survivors (R 2  = 0.02, p  = 0.003) (Fig.  4 C), suggesting that the fungal community changed during that recovery phase in these patients. In contrast, no such response was noted in non-survivor patients, indicating a persistent dysbiosis. Non-survivors displayed a higher abundance of Candida , which reached up to 60%, throughout the entire course of the disease compared to survivors (Fig.  4 D). Specifically, the pathogenic genera Penicillium , Talaromyces , and Botrytis , which were overrepresented in SAP patients compared to healthy subjects, were statistically reduced at the last follow-up in survivors. There were no such alterations in non-survivors (Fig.  4 H-J). In comparison to survivors, the non-survivors exhibited persistent fungi dysbiosis from admission onwards, characterized by an overgrowth of pathogenic Candida , especially Candida albicans , and a loss of commensal Cladosporium (Fig.  4 E-G). These findings imply that the divergent alterations of gut fungi across the course of SAP are correlated with disease prognosis. Fig. 4 Temporal changes of gut mycobiome in SAP patients with different outcomes during hospitalization. Alpha diversity indexes, including Sobs ( A ) and Shannon ( B ), were calculated in survivor patients and non-survivor patients from T1 (admission) through T2 (progression) to T3 (discharge). C Beta diversity of mycobiome variation between S and D group across different stages. D Relative abundance of fungal genera. Differential proportion of sequences attributable to Candida ( E ), Candida albicans ( F ), Cladosporium ( G ) between survivors and non-survivors at admission. Discrepant proportion of sequences attributable to Penicillium ( H ), Talaromyces ( I ), Botrytis ( J ) between admission and discharge in survivors. S, survivors; D, non-survivors; T1, admission stage; T2, progression stage; T3, discharged stage. * p  < 0.05, *** p  < 0.001 Temporal changes of gut mycobiome in SAP patients with different outcomes during hospitalization. Alpha diversity indexes, including Sobs ( A ) and Shannon ( B ), were calculated in survivor patients and non-survivor patients from T1 (admission) through T2 (progression) to T3 (discharge). C Beta diversity of mycobiome variation between S and D group across different stages. D Relative abundance of fungal genera. Differential proportion of sequences attributable to Candida ( E ), Candida albicans ( F ), Cladosporium ( G ) between survivors and non-survivors at admission. Discrepant proportion of sequences attributable to Penicillium ( H ), Talaromyces ( I ), Botrytis ( J ) between admission and discharge in survivors. S, survivors; D, non-survivors; T1, admission stage; T2, progression stage; T3, discharged stage. * p  < 0.05, *** p  < 0.001 To elucidate whether the microbiome co-occurrence pattern is associated with disease progression, we analyzed the interkingdom between bacteria and fungi in patients with favorable and poor prognosis. In the S (survivor) group, module 6, 1, and 11 were primarily represented, with relatively equal proportions. In contrast, the D (non-survivor) group was dominated by module 4 (Fig.  5 A). The modularity index, which measures how well a microbiome network is organized into separate functional teams, dramatically decreased from 0.639 in survivors to 0.379 in non-survivors across all time points during hospitalization (Fig.  5 D). Various network indices, including the number of nodes, edges, average degree, closeness centrality, degree centrality, and density were higher in the non-survivors than the survivors. This suggests stronger interactions within the microbiome during disease deterioration (Fig.  5 B, D, Fig. S3). The average path distance and diameter were statistically reduced in the D group, indicating a high degree of interconnectedness and efficient communication within this network (Fig. S6). A higher betweenness centrality value in the D group also suggests frequent flow of information and resources among nodes in the network (Fig. S3). Interestingly, fungi made up 6% of the nodes in the survivors, but this proportion rose dramatically to 14% in the non-survivors (Fig.  5 C). The module 16 in the survivors was totally composed of fungi, which positively correlated with each other and relatively independent from other modules. In contrast, the fungi in the non-survivors were distributed in four modules, including module 4, 12, 1, and 6 (Fig.  5 C). Taken together, patients with favorable prognosis exhibited distinct bacteria-fungi co-occurrence patterns compared to those with poor outcomes. Fig. 5 Interkingdom co-occurrence networks. A Networks contained both bacterial and fungal taxa showing a more balanced pattern of modular distribution in the survivors compared to those in the non-survivors. Modules represented by nodes with different colors. Blue and red colors of the edges indicated positive and negative correlations, respectively. B Comparison of node-level topological features (degree and closeness centrality) demonstrating the higher degree and closeness centrality values for the hub taxa in non-survivor group than survivor group. C Sankey diagram of bacteria and fungi composition attributed to different modules of survivor and non-survivor group at the genus level. D Lower modularity and higher edge numbers were observed in non-survivors than those in survivors during hospitalization. T1, admission stage; T2, progression stage; T3, discharge stage; S, survivors; D, non-survivors. *** p  < 0.001 Interkingdom co-occurrence networks. A Networks contained both bacterial and fungal taxa showing a more balanced pattern of modular distribution in the survivors compared to those in the non-survivors. Modules represented by nodes with different colors. Blue and red colors of the edges indicated positive and negative correlations, respectively. B Comparison of node-level topological features (degree and closeness centrality) demonstrating the higher degree and closeness centrality values for the hub taxa in non-survivor group than survivor group. C Sankey diagram of bacteria and fungi composition attributed to different modules of survivor and non-survivor group at the genus level. D Lower modularity and higher edge numbers were observed in non-survivors than those in survivors during hospitalization. T1, admission stage; T2, progression stage; T3, discharge stage; S, survivors; D, non-survivors. *** p  < 0.001 To identify the characteristics of the gut microbiome associated with mortality after ICU admission, random forest models were constructed at the genus level for both bacteria and fungi, as well as for SCFAs. We assessed the efficacy of these classifiers in distinguishing between survivors and non-survivors. At the genus level, random forest classifiers based on the top 20 most important bacteria and fungi achieved area under the receiver operating characteristic curve (AUC) values of 0.812 and 0.657, respectively (Fig.  6 C-D). This suggests that the bacterial classifier performed better in differentiating prognosis than the fungi one. For the bacterial classifier, Enterococcus , which was predominant in patients, was a highly labeled microbiota followed by Bacteroides between the S (survivor) and D (non-survivor) groups (Fig.  6 A). For the fungal classifier, the major features included Candida , Penicillium , Saccharomyces and Aspergillus , although their predictive efficiency was relatively low (Fig.  6 B). However, the model based on SCFAs metabolites achieved the lowest AUC value of 0.503 for determining mortality compared to the bacterial and fungal models (Fig.  6 E). Fig. 6 Identification of gut bacteria, fungi and short chain fatty acids classifier for mortality in SAP. The top 20 main important mortality associated bacteria ( A ) and fungi ( B ) at genus level from random forest models. C - E Receiving operational curve (ROC) analysis showed bacterial markers achieved better accuracy for predicting mortality, achieving an area under curve (AUC) value of 0.812, compared to fungi (AUC: 0.657) and short chain fatty acids (SCFAs) (AUC: 0.503). S, survivors; D, non-survivors Identification of gut bacteria, fungi and short chain fatty acids classifier for mortality in SAP. The top 20 main important mortality associated bacteria ( A ) and fungi ( B ) at genus level from random forest models. C - E Receiving operational curve (ROC) analysis showed bacterial markers achieved better accuracy for predicting mortality, achieving an area under curve (AUC) value of 0.812, compared to fungi (AUC: 0.657) and short chain fatty acids (SCFAs) (AUC: 0.503). S, survivors; D, non-survivors Infectious complications are considered to be the most severe complications of SAP often leading to late mortality. The gut is thought to be the primary source of these infections [ 18 ]. The definitions of infectious complications were established according to a previous study [ 19 ]. We evaluated infection based on the occurrence of at least one infectious episode in 45 patients (62%) (Table S2). Pancreatic necrosis was the most common type of infection, with an incidence rate of 48% (35/73). This was followed by bacteremia at 33% (24/73), pneumonia at 21% (15/73), infected ascites at 14% (10/73), and urinary tract infections at 1% (1/73). Patients who developed infections had higher mortality rates than those without infections (38% vs 0%, p  = 0.000). We identified 78 strains of microorganisms from 45 SAP patients with infectious complications. Gram-negative bacteria accounted for 50% of these strains, primarily including Acinetobacter baumannii , Klebsiella pneumonia and Enterobacter spp. . The main Gram-positive bacteria were Enterococcus spp. and Staphylococcus spp. . Thirteen patients were found to have fungal infections, primarily Candida . Among the 45 patients with infectious complications, 13% (6/45) had infections with multi-drug resistant (MDR) bacteria. To determine if gut microbiome dysbiosis is associated with infection complications, we compared the admission samples (T1) between patients with and without infection. While the Sobs and Chao1 indices of bacteria reduced substantially in infected patients, the alpha diversity indexes of fungi did not show significant differences between the two groups (Fig.  7 A, Fig. S4A). Permutational analysis of variance (PERMANOVA) showed significant compositional differences in both bacterial (R 2  = 0.016, p  = 0.003) and fungal (R 2  = 0.011, p  = 0.01) communities between the infected and non-infected groups (Fig.  7 B, Fig. S4B). Notably, the relative abundances of opportunistic pathogens, such as Acinetobacter , Klebsiella , Candida tropicalis , were greatly expanded in the gut of the infected group compared to the non-infected group. In contrast, the beneficial probiotic Bifidobacterium was depleted in the infected group (Fig.  7 E-H). Fig. 7 Dysbiosis of the gut bacteria and fungi in SAP patients with infectious complications. A Bacterial alpha diversity, as revealed by observed species, Shannon, and Chao1, comparing infected and non-infected patients during hospitalization. B PLS-DA plot showing compositional differences of bacteria in beta diversity between infected and non-infected patients. Bacterial composition at the phylum ( C ) and genus ( D ) level. E Fungi species Candida tropicalis was significantly enriched in the gut of infected patients compared to those without infections. Differential bacterial genera, including Acinetobacter ( F ), Klebsiella ( G ), Bifidobacterium ( H ), between infected and non-infected patients. Inf, infected patients; Non_Inf, non-infected patients. * p  < 0.05, ** p  < 0.01, *** p  < 0.001 Dysbiosis of the gut bacteria and fungi in SAP patients with infectious complications. A Bacterial alpha diversity, as revealed by observed species, Shannon, and Chao1, comparing infected and non-infected patients during hospitalization. B PLS-DA plot showing compositional differences of bacteria in beta diversity between infected and non-infected patients. Bacterial composition at the phylum ( C ) and genus ( D ) level. E Fungi species Candida tropicalis was significantly enriched in the gut of infected patients compared to those without infections. Differential bacterial genera, including Acinetobacter ( F ), Klebsiella ( G ), Bifidobacterium ( H ), between infected and non-infected patients. Inf, infected patients; Non_Inf, non-infected patients. * p  < 0.05, ** p  < 0.01, *** p  < 0.001

The present study is among the pioneering attempts to explore the characteristics of gut bacteria and fungi in patients with SAP at various disease progression stages and with different clinical outcome (survivors and non-survivors). We found that SAP patients experienced gut dysbiosis, characterized by reduced biodiversity and disordered community structure, affecting both bacterial and fungal communities when compared to healthy individuals. Furthermore, we observed dynamic changes in bacterial and fungal populations in survivors across three stages (admission, progression, and discharge), but the non-survivors exhibited persistent microbial disturbance without response to treatment. Random forest analysis showed that the bacterial signature demonstrated superior performance in predicting mortality compared with fungi and SCFAs signatures. Additionally, we noted an overgrowth of certain pathogens, including Klebsiella , Acinetobacter , and Candida , in the gut of patients with infectious complications, suggesting that microbiome dysbiosis might be the origin of infection. Emerging evidence suggests that gut bacteria change dramatically either before or at the onset of AP, however, their alterations throughout hospitalization remain unclear [ 7 ]. Our study revealed a decrease in bacterial diversity and an increase in proportions of pathogenic Escherichia_Shigella and Enterococcus when compared to healthy individuals, consistent with previous reports [ 6 , 20 , 21 ]. Escherichia coli , an Escherichia commensal organism, has been reported to aggravate acute necrotizing pancreatitis by damaging the intestinal barrier [ 22 ]. In our study, we noted increased bacterial disturbance in patients with poor outcomes apparently within 72 h after the onset of SAP, including a flourish of Enterococcus and a loss of Parabacteroides . In another study by Yu et al., an increased abundance of Enterococcus was observed in patients with SAP compared to patients with mild AP [ 20 ]. Enterococcus faecium has been shown to be one of the best discriminators between necrotizing patients and non-necrotizing ones, and it might be a potential predictor for infection [ 23 ]. Moreover, an elevation of Parabacteroides might protect against the exacerbation of SAP by producing acetate [ 24 ]. We also noted that while bacterial diversity tended to restore at discharge in the survivor group, it progressively decreased in non-survivors. A notable loss of microbial diversity has been documented in critically ill patients and is independently associated with death during ICU stay [ 25 ]. Moreover, we observed a restoration of Bifidobacterium in survivors at discharge compared to the admission stage. A lower abundance of Bifidobacterium has been reported to be associated with AP severity, and its metabolite can inhibit pancreatic and systemic inflammation [ 26 ]. Levels of SCFAs, which have been proven to stabilize the intestinal barrier and improve pancreatic injury, were found to increase in survivors, likely due to the rejuvenation of SCFAs-producing bacteria [ 27 ]. While both our findings and existing evidence implicate gut dysbiosis in AP progression, the randomized double-blind PROPATRIA trial demonstrated that probiotic prophylaxis not only failed to reduce infection rates but unexpectedly increased mortality [ 19 ]. This unexpected outcome may be attributed to probiotic-induced gut ischemia and local inflammation in critically ill patients. Thus, future studies are warranted to investigate strain-specific effects, optimal dosing, and timing of administration, with a prioritized focus on safety-especially in patients with systemic inflammation or organ failure. Dysbiosis of the human gut mycobiome has been associated with a variety of diseases, but its role in SAP remains obscure. This study, to our knowledge, is the first to characterize the gut mycobiome in SAP and to elucidate mycobiota alterations during hospital stay in relation to prognosis. Patients with SAP displayed enteric fungal dysbiosis, characterized by a decrease in biodiversity compared to healthy individuals. The phyla Ascomycota and Basidiomycota were predominant in both SAP patients and healthy subjects, but the ratio of Ascomycota to Basidiomycota was higher in SAP patients. We identified an overgrowth of opportunistic pathogens in SAP patients, such as Candida tropicalis . C. tropicalis is increasingly recognized for its epidemiological importance and virulence. It can produce strong biofilms and adhere robustly to epithelial and endothelial cells [ 28 ]. Infection with C. tropicalis can increase intestinal permeability and exacerbate colitis, likely through the induction of robust Th1/Th17 responses [ 29 ]. Conversely, some strains of Saccharomyces , including S. cerevisiae and S. boulardii , are probiotic yeasts that are widely used to treat gastrointestinal disorders such as inflammatory bowel disease and various types of diarrhea [ 30 ]. S. boulardii has been shown to exert protective effects on acute necrotizing pancreatitis by suppressing inflammatory responses and bacterial translocation [ 31 ]. Moreover, we found a significant overgrowth of the pathogen Candida , particularly C. albicans , in non-surviving patients compared with survivors, indicating its role in mortality. Various studies have reported that fungal pathogens contribute to disturbances of gastrointestinal barrier and trigger inflammation by inducing intestinal epithelial cells to secrete chemotactic factors and pro-inflammatory cytokines [ 32 ]. In this study, patients were admitted to pancreatic ICU within 72 h of the onset of SAP, and we found a decrease in diversity along with an enrichment of opportunistic pathogens ( Enterobacteriaceae , Enterococcaceae ) over symbiotics ( Lachnospiraceae , Ruminococcaceae , Bifidobacteriaceae ). These findings are consistent with previous studies on critically ill patients, where similar bacterial imbalances, known as pathobiota, were observed regardless of the primary diagnosis [ 33 , 34 ]. Commensal bacteria such as Bifidobacterium and Lachnospiraceae have been shown to play a crucial role in resisting colonization by Enterobacteriaceae in the gut [ 35 ]. We also observed persistent disruption in the gut mycobiome of ICU patients, characterized by decreased diversity and a marked increase in Candida , which aligns with recent findings [ 36 ]. Our study identified Enterococcus and Candida as the most important bacterial and fungal features, respectively, for predicting mortality. This corroborates previous research suggesting that a predominant colonization of Enterococcus and Enterobacteriaceae in the intestine upon admission is linked with an increased risk of death or secondary infection [ 25 , 37 ]. Patients with SAP often develop infections in the later stages of the disease, which potentially contribute to poor prognosis. Numerous studies indicate that the pathogens, including both bacteria and fungi, causing infectious pancreatic necrosis primarily originate from the gut [ 4 ]. We noted that the bacteria cultured from infected patients included Klebsiella pneumoniae and Acinetobacter baumannii , while fungi mainly consisted of Candida . We also found these pathogens’ relative abundance in the gut of infected patients to be statistically higher than in non-infected ones, suggesting that gut-derived pathogens could be the primary source of infection. Dietary intake/choice of food is an important factor that determines microbial composition in the human gut. The human mycobiota diversity is influenced by the diet of an individual, including fermented food products, bread, as well as alcoholic beverages [ 38 ]. Studies on the decreased intake of bread and beer have been shown to minimize the amount of Saccharomyces cerevisiae in human stools [ 39 ]. In a controlled study, it was examined that Penicillium was correlated with a plant-based diet and Candida was enriched in animal-based diet participants [ 38 ]. Additionally, studies have shown foodborne fungi, such as Penicillium and Aspergillus , in more than 60% of stool samples from vegetarians, but these fungal taxa were infrequently found in individuals who consumed a Western diet [ 40 ]. Our findings align with a recent study demonstrating that AP patients exhibited reduced fungal diversity and Penicillium abundance, accompanied by elevated Candida abundance during acute phases [ 41 ]. The observed depletion of Penicillium and other food-associated genera may reflect disease-related dysbiosis and transient reductions due to fasting/enteral nutrition, as these taxa are commonly diet-derived. However, the consistent enrichment of Candida across studies, including ours, supports its role in AP progression. We will integrate dietary metadata and viability-based methods to refine our understanding of gut fungi in future work. The gut microbiome is a complex ecological system, and the interactions between bacteria, fungi, and their relationships are crucial for maintaining ecosystem stability [ 42 , 43 ]. Co-occurrence network analysis has become a valuable tool for understanding these ecological interactions and providing insights into community stability based on topological properties. We observed decreased modularity and increased closeness centrality and edge numbers in deceased patients compared to those who were discharged with recovery. In a healthy state, the gut microbiome exhibits a certain level of modularity, where microbial species form cohesive subgroups that interact more strongly within themselves than with other groups. However, it has been suggested that the modularity of the gut microbiome may decrease in disease states [ 44 , 45 ]. Various factors, such as systemic inflammation and altered immune responses, can lead to the loss of microbial species responsible for maintaining modularity, resulting in a more disorganized and less resilient microbial community [ 46 ]. Indeed, it is important to acknowledge the limitations of this study. First, as an observational study with a limited sample size from a single center, the generalizability of our findings may be constrained. The gut microbiota biomarkers identified in this study were not externally validated, nor were they confirmed using alternative methods such as qPCR. Further validation in larger, multicenter cohorts is warranted. Second, while our use of 16S rRNA and ITS1 gene sequencing provided valuable insights into microbial composition, this approach only assesses relative microbial abundance rather than absolute quantification and the taxonomic resolution is lower compared to shotgun metagenomics. Finally, although our prospective, longitudinal design revealed significant associations between gut dysbiosis and clinical outcomes, the underlying mechanisms remain unclear. Future studies incorporating in vivo and in vitro experiments to elucidate the causal relationships and functional consequences of these microbial changes are warranted.

In summary, our study provided insights into the early disturbances in gut bacterial and fungal composition in critically ill patients with acute pancreatitis. Notably, patients with poor prognosis exhibited significant microbial dysbiosis upon admission, particularly characterized by the proliferation of Enterococcus and Candida , which persisted despite treatment interventions. The presence of opportunistic pathogens such as Acinetobacter , Klebsiella , and Candida in the intestines of patients who subsequently developed infections suggested their origin from the gut. Additionally, gut bacteria outperformed fungi and short-chain fatty acids (SCFAs) in identifying patients at high risk of death. These findings emphasize the need for further research to develop precise strategies for modulating both bacterial and fungal imbalances to reduce the occurrence of infectious complications and improve survival rates in critically ill individuals.

Acute pancreatitis (AP) is an inflammatory disorder of the pancreas associated with substantially morbidity and mortality [ 1 ]. While most patients experience a mild disease course, approximately 20% develop severe pancreatitis (SAP), characterized by necrosis of the (peri)pancreatic tissue and/or (multiple-)organ failure [ 2 ]. In addition to respiratory, circulatory and kidney failure, gut injury is also an important concern in SAP patients. About 40% of patients experience gut failure, which is independently associated with higher mortality rates [ 3 ]. Gut failure often leads to impaired intestinal barrier function, facilitating bacterial and fungi translocation from the gut and subsequent development of infectious complications [ 4 ]. Infected pancreatic necrosis, one of the most severe complications of SAP with a mortality rate of up to 35%, is mainly attributed to bacteria originating from the gut [ 5 ]. Therefore, the gut not only suffers damage as a victim but also contributes a second hit to the progression of SAP. Accumulating evidence suggests imbalances in gut bacteria occur either before or early within the first 3 days after establishment of AP, and the causal role of gut bacteria in AP has been further demonstrated through fecal microbiota transplantation from AP donors to germfree mice [ 6 , 7 ]. However, little information is available regarding the trajectory of microbiota changes during AP progression and whether these microbial features are associated with clinical outcomes. In addition to bacteria, it is increasingly evident that fungi are essential components of the gut microbial community, and play a critical role in regulating host homeostasis and physiological processes [ 8 ]. To date, there is still a lack of comprehensive understanding regarding the composition, community structure, and interplay between gut mycobiome and bacteria in the context of AP. In this study, we conducted an observational cohort study of SAP patients admitted to pancreatic intensive care unit. Stool samples were collected from patients upon admission to discharge for 16S rRNA and ITS1 sequencing. This allowed us to decipher the temporal changes in gut bacteria and fungi throughout their hospitalization. Additionally, we also examined whether the abundances of specific taxa were associated with the mortality outcome and the occurrence of infectious complications.

Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4. Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Supplementary Material 4.