Introduction
The human microbiome, a collective genome of millions of bacteria, viruses, and fungi, have a sophisticated, multidirectional, and mutualistic relationship with their human host. At the core of this interaction is the gastrointestinal tract, which contains trillions of bacteria that create a complex ecosystem called the bacterial gut microbiota (BGM). The BGM affects host health and play major role in acute as well as chronic disease conditions. Autochthonous BGM impede pathogens through selective and controlled expansion, beneficial metabolite and nutrient production, and endocrine interactions while closely corresponding with the local and systemic immune system. Disruption in these mechanisms interact with other factors such as environmental (occupation and drug/toxin exposure), host (genetic predisposition, alcohol, and tobacco), and disease states (metabolic syndrome and cirrhosis). That leads to initiation of both quantitative (diversity reduction) and qualitative (perturbed functional metabolism) changes in the BGM, a process known as dysbiosis. Dysbiosis worsens existing disease (e.g., cirrhosis progression) or causes new clinical events (e.g., infections).1 The BGM plays a central role in the etiology and progression of various liver diseases (e.g., alcohol associated hepatitis) and clinical events in cirrhosis (e.g., hepatic encephalopathy), and its modulation ameliorates adverse patient outcomes.2–4 Recent evidence had demonstrated that dysbiosis of the BGM and its functions promote sepsis, “a life-threatening organ dysfunction caused by dysregulated host response to infection,” and associated organ dysfunction in affected patients. New research has shown that microbiota modulation may help improve clinical outcomes and organ dysfunction associated with sepsis.5,6 Nonetheless, the role and function of BGM in patients with advanced cirrhosis and sepsis remain unknown. The current work therefore aimed to characterize the BGM and its functions associated with specific clinical and biochemical events among decompensated cirrhosis (DC) patients who were infected (iDC) or noninfected (niDC).
Study objectives
The study objectives were to characterize significant bacterial communities and their functional metabolism between iDC and niDC patients and distinguish and identify significant bacterial groups and associated functional metabolites among iDC patients grouped into the following sepsis specific categories: (1) infection sites, (2) infection episodes (0, 1, or ≥2), (3) Systemic inflammation in sepsis evaluated by interleukin (IL)-6 (<100, 100–1,000, and >1,000 pg/dL), (4) outcome during hospital stay, and (4) outcome 180 days after admission. We also described the topology of the relationships between pertinent bacterial taxa associated with grouped parameters and events in DC with and without infections using network analysis.
Methods
Patients
From March 2019 to December 2019, consecutive patients with decompensated cirrhosis admitted to the Liver Unit through the emergency and outpatient departments were included (Fig. 1). This study protocol was approved by the Institutional (hospital) review board and was performed in accordance with the Helsinki declaration of 1975 and its revisions. All participants provided informed consent for the use of de-identified fecal samples. A comprehensive clinical and drug history was obtained at admission, including events occurring over the preceding 3 months. This included details on prior admission for infection, exhaustive evaluation during the previous admission and discharge summaries when available, and decompensation events. Decompensation was defined as the presence of either jaundice, ascites, kidney injury, hepatic encephalopathy, or a combination of those. Patients with suspected drug-induced liver injury (including complementary and alternative drugs), those listed for liver transplantation, and those with index presentation as acute on chronic liver failure were excluded. Similarly, patients with acute variceal bleeding, those with hepatic and extrahepatic malignancies, those undergoing interventional vascular and invasive hepatobiliary procedures, those with severe cardiopulmonary disease, those with advanced cirrhosis on multiple organ support, and those patients on hemodialysis were also excluded. Infection was suspected based on history and clinical and physical examination and confirmed either by radiological, biochemical, or microbiological evidence along with blood and body fluid analyses. All patients with serum procalcitonin ≥10 ng/mL were considered to harbor bacterial infection even without an identifiable source. When serum procalcitonin was <10 ng/mL infection was confirmed when either radiological, blood and body fluid, or microbiological evidence supported the same.7,8 IL6 levels (normal <7 pg/mL) were documented only to determine systemic inflammation and not to identify infection or sepsis. Decompensated cirrhosis patients with and without infections were matched for the presence of ascites, hepatic encephalopathy (HE), and acute kidney injury, Child Turcotte Pugh stage, and model for end-stage liver disease scores at baseline using the case–control matching function in MedCalc software (Ostend, Belgium). All included patients were started on intravenous third-generation cephalosporins, and antibiotics were modified per site of infection and culture sensitivity during the course of hospitalization.
Sample collection
Stool samples were collected within 24 h of hospital admission in sterile containers, immediately processed into aliquots within 1 h, and stored at −80°C in the in-house storage designated for the protocol. After collecting a minimum of 10 samples during a period, RNAlater (Ambion/Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA) was added to the aliquots, which were then transported to the main laboratory facility within 4 h while maintaining a cold chain −20°C for DNA extraction and further analysis.
DNA extraction and sequencing of bacterial communities
Approximately 200 mg of the provided stool sample was used for bacterial DNA extraction, using a defined and published protocol modification of the commercially available QIAmp DNA Stool Mini Kit1 (Qiagen, Venlo, Netherlands). Sequencing as per standardized validated methods at the V3-V4 regions was performed with an Illumina MiSeq next-generation sequencer (Illumina, San Diego, CA, USA) using the Illumina kit at 2×300 paired-end sequencing, after which taxonomic classification was performed according to the GreenGenes Database (version 13.8).
Metagenomic statistical methods
The alpha diversity, including phylogeny (i.e. summed evolutionary age of all the species in the community), was presented using phylogenetic diversity (PD or Faith’s PD) measure, which uses phylogenetic distance to calculate the diversity of a given sample. Phylogenetic distance represents the number of changes that have occurred within a particular taxon (branch) ascertained within groups or between groups. Quantitative Insights into Microbial Ecology (QIIME version 2) was used to ascertain the quantitative and qualitative microbial communities.9 The Kyoto Encyclopedia of Genes and Genomes (KEGG) Pathway was utilized to study functional metabolic interactions and relationships within sequenced bacterial communities.10 Analysis of Similarity was used to test for a statistical difference between two or more microbial communities (alpha diversity). A p-value <0.05 was considered to indicate statistically significant differences between grouped sample attributes.11 The visualization tool Circos (v.0.69-9; tool version: 0.23) was used to facilitate the exploration and analysis of similarities and differences arising from comparisons of bacterial communities. The table viewer script in Circos tools was used to format abundance data. This method shows significant positive, negative, or neutral interactions between communities specified for the variable. Output data were presented using a circular chord ideogram layout that displayed the relationships among microbial communities concerning clinical or investigation variables at the phylum/class/order/family taxonomic level. To simplify the graphical output, tabulations were made by compounding all analyzed clinical or investigation variables.12 Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt, version 1.1.1) was used for predictive metabolic functional profiling of microbial communities using 16S rRNA marker gene sequences precalculated for protein-coding genes present in KEGG gene families and 16S rRNA gene copy number.13 Linear discriminant analysis effect size (LEfSe) combined with the Kruskal-Wallis and pairwise Wilcoxon tests were utilized to identify significant differences in abundance and functionality of microbial communities between groups. We used default significance (alpha value=0.05) and linear discriminant analysis thresholds (2.0) at all taxonomic levels between time points. To simplify the graphical output, tabulations on significant taxa and the associated functional metabolic pathways were provided.14
Bioinformatics and network analysis
Networks were inferred using CoNet (v.1.1.1-beta) application within Cytoscape (v.3.7.2). The following measures, implemented by CoNet, were used to detect copresence/exclusion between bacterial communities: Pearson, Spearman, Mutual Information, Bray-Curtis dissimilarity, and Kullback Leibler dissimilarity. CoNet was also set up to determine the comparison metadata for each sample (i.e., the respective clinical or investigational variable grouping). Moreover, the following network measures were calculated using NetworkX (version 2.2): degree centrality, betweenness centrality, closeness centrality, and added to the node attributes in the graph. The centrality measure of degree assigns importance to a specific node (microbe at bacterial phylum/family/genus level) in the form of the number of direct connections to other nodes (microbes). The centrality measure of betweenness represents the number of times a microbe lies on the shortest path between other microbes, which identifies microbes that serve as the best “bridges” between other microbes in the network. This measure demonstrates the bacteria that influence flow around a system (in this case, the clinical or investigation variable). A high betweenness indicated the microbe that holds authority over different clusters in the specified network. The centrality measure of closeness represented microbial (nodal) closeness to all other nodes (microbes) in the network. These measure the shortest path between all nodes (microbes) and identifies the individual microbe best placed to influence the entire network more quickly or those who are good “broadcasters” of information (metabolic-cross talk). The topology of constructed networks and significant taxa, the top 2% of all taxa analyzed, were presented in a simplified tabular form based on these attribute indices of degree, closeness, and betweenness centrality measures.15–17 General statistical analysis was performed using MedCalc version 19.7 (Ostend, Belgium) and NCSS version 12 (Kaysville, UT, USA) statistical software. Data were reported as means and standard deviation or as medians and interquartile range between brackets as applicable. The Shapiro-Wilk test was used to test normality and Levene’s test was used for non-normal distributions. The Bartlett homogeneity test (for nominal variables) was utilized to assess the equality of variances. To decrease data variability and increase data conformity to a normal distribution, logarithmic transformation was applied. Chi-square and Fisher’s exact tests were used to compare normally distributed variables; the Mann-Whitney U test was used to evaluate continuous variables. A simplified summary of patient inclusion and methodology is shown in Supplementary Figure 1.
Discussion
This study defined and demonstrated the quantitative, qualitative, and interactive roles of the BGM in the presence and absence of infections. To the best of our knowledge, no study has comprehensively analyzed and characterized the BGM with respect to clinical events associated with infections in cirrhosis. Our study aimed to provide insights on the role of BGM in promoting sepsis and associated events among patients with cirrhosis. Accordingly, our findings showed that pathogenic genera, such as Leptotrichia, Neisseria, and Erwinia, were predominant in cirrhosis patients with infection, whereas beneficial bacteria, such as Bifidobacterium and Coriobacteria, were predominant in those without infection. Similarly, the functional metabolism of the predominant BGM included upregulation of pathways associated with oxidative stress, endotoxemia, and proinflammatory process in those with infections and endogenous antimicrobial, antioxidant, and cytokine signaling pathways in those without infections. Specific pathogenic bacterial taxa were associated with different sites of sepsis (e.g., Mycoplasmataceae in bacteremia and Aeromonas in UTIs). Repeat infections and high systemic inflammation were associated with rare, multidrug-resistant, and immune exhaustion-related intestinal bacterial groups. This change in BGM was significantly notable in those without infections (Bifidobacterium) who subsequently developed their first (Lautropia) and multiple episodes of infection (Leptotrichia). Specific bacterial genera with high propanoate functional metabolism, such as Megamonas, were associated with survival in infected patients with cirrhosis during admission and after follow-up. Network analysis showed that noninfected cirrhosis patients had richer and more diverse interactions, comprising mutual exclusions and co-occurrence between various bacterial taxa, compared with those with infections.
Host-microbiota interactions can predispose individuals to infections, promote more severe infections, and even prevent their occurrence, as demonstrated in insects, small animals, and humans. Microbe-arbitrated host protection from infections can occur through the maintenance of host immune homeostasis, interference competition with pathogenic taxa, and competitive resource utilization. A decline in host health, as seen in development and progression of cirrhosis, can cause dysbiosis, which promotes the perturbation and transition of commensal microbiota toward pathogenicity and further predisposes the host to infections.18 Reduced microbial diversity and loss of colonization resistance due to perturbation caused by dysregulated local and systemic immune homeostasis promote systemic infections originating in the gut, as demonstrated with Clostridioides difficile infection.19 Gut microbiota can be used to develop a pathogenic profile with the onset of dysbiosis driven by illness and its management or immune compromise, such as antibiotic treatments and liver disease, respectively.20,21 Liu et al.22 found that changes in gut microbial patterns promoted necrotizing enterocolitis and late-onset sepsis in babies born prematurely and that changes in gut bacteria were probably the causative factors of the development of infectious complications in predisposed patient groups. Moreover, changes in gut microbiota composition can potentially predispose patients to a state of immunosuppression, thereby increasing the risk of sepsis. Indeed, Hyoju et al.23 clearly demonstrate how a selective high-fat, sucrose-rich Western diet, antibiotic exposure, and surgical injury converge on the microbiome, causing lethal sepsis and multiorgan failure without exogenous pathogens. In immunosuppressed patients, such as those undergoing bone marrow transplantation, antibiotic-associated dysbiosis promoted a five- to nine-fold increase in the risk of bloodstream infection and sepsis. Similarly, a strong dose-response relationship between dysbiosis-causing events and subsequent severe sepsis-related hospitalization had been observed among elderly patients. Furthermore, a previous study showed that prolonged antibiotic exposure and utilization of additional antibiotic classes and broader-spectrum antibiotics during hospitalization were associated with dose-dependent increases in the risk of subsequent sepsis. This suggested that the association between antibiotic exposure and subsequent sepsis was associated with microbiome depletion and not illness severity.24–26
Critical illness, as noted in patients with decompensated cirrhosis, profoundly disturbs the gut microbiota due to low food intake and malnutrition, recurrent hospitalizations for portal hypertensive complications, worsening liver function and progressive decline in immune status, and multiple exposure to antibiotics. This phenomenon is evident in the current study, wherein patient with advanced cirrhosis requiring multiple admissions demonstrated worsening dysbiosis due to repeated infections, causing pathobiont generation, immune exhaustion, multiple organ failure, and poor survival. The effects of the gut microbiome on sepsis outcome have been clearly demonstrated in an animal model. Notably, when sepsis was introduced using the cecal ligation and puncture method in two different sets of mice purchased from different vendors, with different fecal microbiota beta diversities and immune phenotypes, higher mortality was notable in the first group. When both groups of mice were housed together, differences between beta diversities and immune phenotypes disappeared, and further sepsis introduction included similar survival outcomes. The aforementioned study highlighted the importance of the gut microbiome for survival from, and host immune response to, sepsis. Thus, the transition of a microbiome into a pathobiome could promote poor clinical outcomes and mortality from sepsis by modulating the host immune response, which was seen in our patients with repeated infections, severe unchecked inflammation, and higher mortality.27 A prior study that evaluated the effects of gut microbiome dynamics in intensive care unit (ICU) patients showed that changes in the gut microbiota were associated with patient prognosis. The proportions of Bacteroidetes and Firmicutes significantly changed during ICU stay, and extreme changes were observed in almost all patients with a poor prognosis, suggesting a correlation between qualitative and quantitative dysbiosis and sepsis outcomes. Similarly, a study conducted by our group demonstrated that progressive dysbiosis in advanced cirrhosis with repeated infections and hospitalizations promoted extreme changes in fecal bacterial communities and was associated with poor short- and long-term clinical outcomes.28 Specific classes of antibiotics have been associated with fecal microbiota changes. Accordingly, penicillins, cephalosporins, and carbapenems were associated with a reduction in beneficial Bifidobacteria and Lactobacilli and an increase in pathogenic Enterobacteriaceae/Enterococci. Published faucal microbiota data suggest the development of resistance in Enterobacteriaceae following exposure to piperacillin-tazobactam. Moreover, antibiotic-associated microbiota disruption can occur as early as 3 days and last for 12 months as shown via molecular analytical methods in those receiving short courses of fluroquinolones.29
This study found that infections upon admission in DC patients were associated with deleterious changes in the fecal microbiome, which progressively worsened in those with repeated infections. The GBM of those receiving high-end antibiotics for worsening infections transitioned toward a pathobiont profile, and associated bacterial taxa were identified as opportunistic pathogens predominant in immunocompromised states. Prior studies on patients with and without sepsis demonstrated heterogeneous patterns of intestinal microbiota, including the disappearance of bacterial genera with important functions in host metabolism. Moreover, our fecal analysis of critically ill patients with sepsis at different sites revealed significant differences in GBM profiles, similar to our findings among patients with cirrhosis.30 Agudelo-Ochoa et al.31 found that the microbiota of ICU patients with sepsis contained numerous microbes strongly associated with inflammation, such as Parabacteroides, Fusobacterium, and Bilophila species. Furthermore, a difference in the abundance of pathogenic species, such as Enterococcus spp., was observed in patients with sepsis who died. Among our patients, those with infections, those with repeated infection, and those who survived after matching for portal hypertension events, liver disease severity, and extrahepatic organ failure involvement, had a GBM that was preferentially and significantly associated with proinflammation, intestinal barrier dysfunction, and endotoxemia.
We chose to study IL6 in our patients for several reasons. Studies have shown that IL6 levels are increased in patients with cirrhosis and is linearly associated with the severity of underlying cirrhosis. IL6 was also found to play a role in the hyperdynamic circulation observed in patients with cirrhosis. In DC, plasma IL6 levels on admission were shown to provide the most sensitive and specific tool for the diagnosis of bacterial infection and were closely linked to the development of overt HE in patients with liver cirrhosis and was particularly useful in predicting mortality in patients with cirrhosis.32–35 In that context, our study revealed the important role of GM dysbiosis and specific bacterial taxa associated with varying IL6 levels in DC patients with infections. Short-chain fatty acid-producing Propionibacterium was relatively abundant in those with IL6 <100 and pathogenic taxa were associated with states of immune exhaustion and opportunistic infections predominated the gut in DC patients with IL6 >1,000. That demonstrated a plausible association of GM as a promoter and driver of systemic inflammation and eventually organ dysfunction in DC patients with sepsis.
This study is not without limitations. Apart from being a single-center, retrospective study, we did not match our patients in terms of other confounders that could have affected gut microbiome, as well as long-term outcomes, such as dietary habits, heterogenous environmental exposures, and other known or unknown chronic comorbidities. Nonetheless, the DC groups with and without infections were comparable with respect to liver disease severity, portal hypertension events, nonabsorbable antibiotic use, and extrahepatic organ dysfunction at admission. We did not assess dynamic changes in the BGM from admission to discharge or the end of follow-up, given that it could have been affected by heterogenous and possibly unmodifiable confounding factors. During follow-up, we did not account for the various causes of death in our patients. Further, the generalized clinical outcomes, such as survival or death, were based on a single point fecal microbiota analysis for ease in analysis, which could have oversimplified the role of gut dysbiosis in our patient population.
Among patients with cirrhosis, dysbiosis occurs as part of the etiology and development of liver disease and subsequently promotes infections in advanced cirrhosis due to portal hypertension and immune dysfunction. That worsens with subsequent hospitalization and therapeutic or symptom management interventions required in this special patient group, which prevents the complete recovery of the microbiota. Periods of dysbiosis predisposed DC patients to opportunistic infections, which further worsened dysbiosis, and predisposed them to sepsis, wherein the microbiota composition becomes extremely compromised and the risk of secondary infections, immunosuppression, and organ dysfunction increases. The study showed that gut microbiota changes in advanced DC patients induced a state of prolonged immunosuppression, rehospitalization because of infections, and increased mortality. Similar to the feasibility and effectiveness of fecal microbiota transplantation as a treatment option for severe C. difficile infection in the ICU, the findings suggest that gut microbiota modulation using healthy stool transplants could be an important therapeutic option in the fight against multidrug resistant bacteria in vulnerable patient populations, such as those with advanced cirrhosis, pending further large quality, well controlled observations and clinical trials.
Future directions based on the current work includes defining subtypes of patients (independent from other confounders) based on the microbial signatures or metabolites that would benefit from targeted therapies; validation of our findings in small animal models or a second patient cohort (with a different geographic or ethnic background) prospectively, and analyzing longitudinal data (e.g., after antibiotic therapy, in patients with antibiotic prophylaxis for infections such as SBP).