Gut mycobiome dysbiosis after sepsis and trauma

The overall experimental framework is depicted in Fig. 1A. This study was a single-center, prospective observational cohort investigation conducted from 2019 to 2023 at a Level-1 trauma center. The inclusion and exclusion criteria are detailed in Additional file 1: Table S1. Primary data collection for specific sepsis patients and healthy controls was carried out within the scope of the study titled “Persistent Inflammation, Immunosuppression, and Catabolism Syndrome (PICS): A New Horizon for Surgical Critical Care,” approved by the Institutional Review Board (IRB) under number 201400611 on October 28, 2014, and registered at ClinicalTrials.gov with the identifier NCT02276417. Initial data collection for certain trauma patients was conducted as part of the study titled “Hematopoietic Stem Cell Dysfunction in the Elderly after Severe Injury,” which received IRB approval under number 201601386 on September 6, 2016, and was registered at ClinicalTrials.gov with the identifier NCT02577731. The remaining data were gathered in the context of the study titled “Gut Microbiome Dysfunction in Sepsis and Trauma Survivors,” which received IRB approval under number 202102863 on May 3, 2022, and was registered at ClinicalTrials.gov with the identifier NCT05357170. This study’s reporting adheres to the guidelines outlined in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [22].

Clinical data, including age, gender, race, ethnicity, and medical comorbidities, were collected to profile the demographic features of the cohorts. Additionally, management and outcome parameters were documented, which included factors like the injury severity score, sepsis criteria scoring, blood product transfusions, antibiotic administration, dietary intake, as well as hospital and ICU lengths of stay. Mortality outcomes were also recorded. It is important to note that all trauma and sepsis patients included in the study underwent at least two weeks of standardized ICU management and met established criteria for Chronic Critical Illness (CCI) and PICS [23].

Stool specimens were acquired either per rectum, via a rectal tube, or from a colostomy between hospital days 14 and 21. These specimens were then divided into aliquots and preserved in vials containing Cary-Blair bacterial media (ThermoFisher Scientific, Waltham, MA) at -80ºC. Simultaneously, peripheral blood was obtained through a single venipuncture using heparinized blood collection tubes (Becton–Dickinson & Co., Franklin Lakes, NJ). The blood samples were centrifuged at 800 × g for 10 min, and the resulting plasma was preserved at − 80 °C.

Mycobiome measurement and analyses were conducted as per our previously described method [24, 25]. Briefly, we extracted high-quality genomic DNA from 200 mg of fecal samples using the QIAmp PowerFecal Pro DNA Kit (Qiagen, Hilden, Germany) and quantified it with a Nanodrop spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). The internal transcribed spacer (ITS) region of the fungal rRNA gene was amplified, following the Earth Microbiome Project benchmark protocol (www.​earthmicrobiome.​org). The resulting amplicons were purified using the AMPure^® XP magnetic beads kit (Beckman Coulter, Indianapolis, IN), and the purified products were quantified with the Qubit-4 fluorimeter (InVitrogen, Waltham, MA, USA). An amplicon library was generated following previously established methods [26]. The purified library was combined in equimolar concentrations and subjected to sequencing on an Illumina MiSeq platform (Illumina Inc., San Diego, CA, USA) using a 2 × 300 base pairs reagent kit (MiSeq reagent kit v3; Illumina Inc.) for paired-end sequencing.

Raw sequences were processed using the Quantitative Insights Into Microbial Ecology (QIIME2) bioinformatics software suite (version 2.2023.5) [27]. Quality-filtering, adapter-trimming, denoising, and removal of non-chimeric amplicons were conducted using the DADA2 pipeline through the q2-dada2-plugin with default parameters [28]. All identified amplicon sequence variants (ASVs) were aligned using MAFFT [29]. Taxonomic assignment for these ASVs was accomplished by employing the sklearn classifier, which utilized a pre-trained naïve Bayes taxonomy classifier. The classification process involved alignment against the 99% UNITE 9.0 database. Alpha-diversity was quantified using the number of observed ASVs, Chao1, and Shannon index metrics. To assess beta-diversity, the Bray–Curtis dissimilarity index was utilized, and the results were presented with a principal coordinate analysis (PcoA). Statistical analyses, including the nonparametric Kruskal–Wallis test and PERMANOVA with 999 random permutations, were applied to identify significant differences in microbial diversity and structure. The differential abundance of taxa was determined through the ANOVA-Like Differential Expression (ALDEx2) approach [30]. To predict the group based on microbial composition through supervised classification, the q2-sample-classifier plugin in QIIME2 was utilized. This process involved a nested stratified fivefold cross-validation using the Random Forest classifier, which was built with 5,000 trees. To evaluate the impact of two factors, group and sex, on taxa abundance and potential interaction effects, a two-way ANOVA was conducted. To examine the correlations between fungal taxa and bacterial taxa, gut, and plasma metabolites, Spearman’s rank correlation was used. Networks connecting fungal taxa and gut metabolites were established by calculating Spearman correlations, and significant associations (Spearman correlation coefficient > 0.30 and Benjamini–Hochberg corrected p value < 0.05) were visualized using Cytoscape v3.9.1 [31]. Data visualization was carried out using ‘R’ or ‘Python’ packages.

Patient characteristics of healthy control subjects (control), sepsis patients, and trauma patients are summarized in Additional file 1: Table S2. No significant differences in age, sex, race, or ethnicity were observed. When comparing medical comorbidities (including hypertension, hyperlipidemia, renal disease with creatinine > 1.5 mg/dL, coronary artery disease, diabetes, congestive heart failure, chronic obstructive pulmonary disease, and cancer), significant distinctions among the groups were noted only for hypertension and coronary artery disease. Both the sepsis and trauma cohorts had a higher prevalence of hypertension compared to the control group, while the sepsis cohort exhibited a greater number of patients with a history of coronary artery disease. It is noteworthy that all sepsis and trauma patients received antibiotics during their hospital stay. As expected, sepsis patients received a larger variety of antibiotics for longer durations, primarily because all sepsis patients required antibiotics for source control. All patients in the sepsis and trauma groups received enteral nutrition, which included tube feeds or an oral diet. Only one sepsis patient received total parenteral nutrition for ten days. The sepsis group experienced significantly longer stays in the ICU and the hospital compared to the trauma cohort.

To investigate alterations in the gut mycobiome following hospitalization in critically ill sepsis and trauma patients, we conducted an evaluation of the fungal community diversity and composition. Additionally, we explored differences based on host sex (Fig. 1b, c). The overall mycobiome profiles of the sepsis and trauma cohorts were found to be nearly similar, while both significantly contrasted from the composition in the control cohort. In control group, the beta-diversity distance (a measure of the similarity of microbial communities between two samples or ecosystems) between subjects, transformed into a two-dimensional plot using PCoA analysis based on dissimilarity, exhibited closer proximity among females compared to males. This resulted in more distinctly divergent arrays between healthy adults versus sepsis or trauma patients among females (Fig. 1b), indicating the existence of dissimilar mycobiome profiles between these groups. In all groups, the microbial alpha-diversity (species richness) of patients exhibited a high degree of dispersion. However, sepsis patients displayed significantly lower microbial diversity compared to healthy adults and trauma patients, as measured by the Shannon index, which measures both microbial richness and evenness. While there were no differences between groups in the total number of amplicon sequence variants (ASVs, an indirect proxy for species number), which represent the observed count of genes potentially originating from different bacteria, and microbial diversity measured by the Chao1 index-measuring solely microbial richness-indicating no variation in microbial richness among the groups, only the sepsis group exhibited distinct microbial evenness. In the healthy control group, male subjects exhibited higher diversity than their female counterparts, whereas the opposite trend was observed in sepsis and trauma patients. Only male sepsis patients demonstrated significantly lower Shannon diversity and had fewer ASVs than healthy males. Male trauma patients also exhibited modestly reduced diversity than healthy males. In contrast, female sepsis and trauma patients demonstrated almost comparable or slightly higher diversity compared to healthy females (Fig. 1c). At the phylum level, the mycobiome community was dominated by two major fungal phyla, Basidiomycota and Ascomycota. It is noteworthy that Basidiomycota, which otherwise made up a considerable proportion of the composition in healthy adults, was nearly depleted in both sepsis and trauma patients (Fig. 1d). Further deeper analyses revealed that this difference was primarily due to an explosive increase in the levels of genus Candida in both sepsis and trauma patients. In healthy adults, Rhodotorula, Saccharomyces, Penicillium, and Candida were the major genera comprising the gut mycobiome, with Candida being the sole predominant genus in sepsis and trauma patients. Particularly in sepsis patients, Candida accounted for more than 95% of the mycobiome abundance on average (Fig. 1e). In differential abundance analysis, Candida was identified as the most significantly proliferated genus in both sepsis and trauma patients, followed by Fusarium, while Penicillium and Saccharomyces were nearly depleted (Fig. 2a, b). Additionally, further analysis using ANOVA distinguished Candida, Penicillium, Fusarium, and Rhodotorula as major taxa differing significantly among the three groups (Fig. 2f). Additionally, healthy adults exhibited distinct microbial compositions based on sex. Male adults had higher levels of Candida and Penicillium; while, female adults harbored more Saccharomyces and Rhodotorula (Fig. 1e, 2e). In both sexes, the abundance of taxa other than Candida was insignificantly low in sepsis and trauma patients, but Aspergillus tended to be more prevalent in female sepsis patients than in males, and Chimonocalamus tended to be more abundant in female trauma patients compared to males. A higher level of Rhodotorula was observed in female subjects across all groups (Fig. 2c–e). Although no taxa were identified as significantly differing by gender in ANOVA, Rhodotorula and Chimonocalamus were the most affected taxa by gender, regardless of the group, among all taxa (Fig. 2f). We further trained and executed a machine-learning model to predict and capture the primary distinctions between sepsis or trauma patients and healthy subjects, aiming to understand the abnormalities of their mycobiome. The control group demonstrated the highest prediction accuracy (AUC = 0.92); while, the sepsis and trauma groups exhibited relatively lower accuracy (AUC = 0.74 and 0.75, respectively) (Fig. 2h). Within the sepsis group, samples that were incorrectly predicted were more likely to be categorized as belonging to the trauma group rather than the control group, and the same pattern was observed within the trauma group (Fig. 2g). This may be attributed to the similar microbial profiles observed in both sepsis and trauma patients. Candida was consistently identified as the most important feature for prediction, followed by Curvularia and Penicillium, and these taxa were most abundant in sepsis, trauma, and control groups, respectively (Fig. 2i).

In each analysis, Candida reproducibly proved to be the most significant and distinctive genus for distinguishing sepsis and trauma patients from healthy adults. We investigated whether the Candida profile alone could accurately characterize the groups and be exclusively utilized as a group prediction factor. Overall, not only the abundance but also its detection rate in sepsis and trauma groups was significantly higher than in the healthy control group (Fig. 2j). Although the average abundance of Candida was highest in the sepsis group, the detection rate was slightly lower than in the trauma group, where all patients had Candida carriage. All the ASVs identified as Candida were successfully assigned to the species level, resulting in six species. Candida species profiles exhibited substantial differences between the healthy control group versus sepsis and trauma patients; while, sepsis and trauma patients shared relatively similar profiles. Specifically, C. orthopsilosis and C. sake were only identified in the healthy control group and represented a significant portion of the Candida flora in this group. In contrast, C. dubliniensis and C. parapsilosis, which were major species in the sepsis and trauma groups, were either not detected or barely detected with very low abundance. C. albicans and C. tropicalis were the most abundant and highly detected species in all groups, although the sepsis and trauma groups had relatively higher abundance and detection rates compared to the control group (Fig. 2k). When examining the PCoA analysis based on Candida species profiles, the healthy control group exhibited a distinct profile and was significantly separated from the sepsis and trauma groups. Intriguingly, the overall profile within each group was nearly identical to the group structures generated with all ASVs, indicating that ASVs assigned to Candida species were the major features determining the overall group structures and driving the differences between the groups (Fig. 2l). Furthermore, the prediction accuracy of the model trained solely with Candida species showed analogous patterns to the prediction model trained with all microbiome data (Fig. 2m, n). C. tropicalis and C. albicans were consistently identified as the most important species for determining and predicting the groups (Fig. 2o).

The gut is a highly complex ecosystem with various forms of symbiosis. Changes in one domain vastly affect other domains and the microbial by-products produced after their metabolism. We have previously observed significant reduction in the gut bacterial diversity in sepsis and trauma patients, accompanied by significant shifts in the composition of the microbiota, including the emergence of specific pathobiome patterns. Additionally, both trauma and sepsis were associated with discernible changes in fecal and plasma metabolites (Additional file 2: Figure S1). Therefore, to understand the ecological niche and fundamental co-regulation of mycobiome communities within the gut, we applied correlational analyses of fungal taxa with gut bacteriome as well as with gut and plasma metabolome arrays (Fig. 3a–e). Bacterial taxa that had at least one significant co-occurring relationship with fungal taxa were divided into two branches based on their hierarchical clustering of correlational relationships. One branch primarily consisted of bacteria associated with gastrointestinal infections, such as the Enterobacteriaceae family, Escherichia–Shigella, Enterococcus, and Staphylococcus. These bacteria tended to positively correlate with fungal taxa found abundant in sepsis and trauma patients. The other branch mostly contained commensal and beneficial taxa, including Bifidobacterium, Akkermansia, and Ruminococcus, which tended to co-occur with fungal genera more abundant in healthy adults (Fig. 3a). Specifically, three major genera, Candida, Fusarium, and Penicillium, exhibited the most significant correlations with commensal bacteria, including Faecalibacterium, Anaerostipes, and the [Eubacterium] hallii group. Candida and Fusarium, which were increased in sepsis and trauma patients, negatively correlated with these bacteria, while Penicillium, more abundant in the healthy control group, showed a strong positive correlation with these bacteria. Additionally, several beneficial bacteria, such as Akkermansia and Blautia, demonstrated a positive correlation with Penicillium and a negative correlation with Candida. Meanwhile, we observed a reverse trend with Lactobacillus, which was positively correlated with fungal taxa abundant in sepsis and trauma patients (Fig. 3d).

Gut microbial metabolites were also clustered hierarchically based on their co-regulation networks with fungal taxa. One cluster contained short-chain fatty acids (SCFAs), including butyrate and propionate, which were positively correlated with fungal taxa prevalent in healthy adults. Another cluster consisted of by-products of microbial metabolism, including lactate, trimethylamine-N-oxide (TMAO), and ethanol, which were positively correlated with taxa more abundant in sepsis and trauma patients (Fig. 3b). Candida showed a strong negative correlation with SCFAs; while, Penicillium demonstrated a robust positive correlation with butyrate. These results were also reflected in the network analysis, where Candida showed a co-occurrence network with alanine and sorbitol and a mutual exclusive network with butyrate. In contrast, Penicillium showed exactly opposite networks with these metabolites (Fig. 3f). Curvularia exhibited a similar correlational profile with Candida, primarily negatively correlating with essential metabolites associated with gut integrity, such as propionate, cholate, and threonine. While Issatchenkia and Alternaria exhibited several significant correlations and networks with amino acids, their low prevalence rates (9.4% and 5.7%, respectively) suggest that the significant results may be misleading due to the skewed distribution of data. In blood (plasma) metabolome, three metabolites, including lactate, glutamate, and alanine, were found to be common with gut metabolites. Among these, only lactate displayed a similar correlational pattern with fungi as observed in the gut lactate, although it did not reach statistical significance. Overall, only Candida and Penicillium demonstrated strong and distinct correlations with plasma metabolites, and these two genera had inverse profiles. Penicillium negatively correlated with 3-hydroxybutyrate, the primary ketone bodies produced during ketogenesis, while positively correlating with acetate, glutamate, and alanine, which are involved in gluconeogenesis. Candida however showed an inverse relationship with these metabolites (Fig. 3c, e).