Although alpha diversity was globally and markedly reduced by 72hours of ICU admission, it did not predict all-cause mortality at day 28 by meta-analysis, in line with 25 of 26 studies in the systematic review (10, 15–32, 34–36, 41, 42). Confidence in this result is constrained by data availability for meta-analysis, the overall sample size, heterogeneity and moderately-high risk of bias.
The novel finding of this systematic review is that overabundance and depletion of selected pathogenic and commensal taxa respectively are more frequently predictive of mortality and other clinical outcomes. Alpha diversity metrics do not acknowledge emergence of opportunists, disproportionate loss of obligate anaerobes, infrequent or rare species or detection of syntrophic relationships(43–45). It is plausible these traditional ecological measures of richness and evenness are too ‘blunt’ to show meaningful compositional change and predict outcomes in acute critical illness.
Whilst interactions between organisms across kingdoms may be best observed in ultra-low diversity conditions they are also not captured by alpha diversity or common sequencing approaches(6). 16S rRNA amplification was the primary focus of 24 studies yet does not represent fungal species commonly observed by culture, 18S/ITS and whole genome methods. Prevalence studies report Candida spp. in up to 80% of patients 1 week into ICU stay (46). Ravi (41) noted ESKAPE group and C. albicans constituted up to 75% of some microbiomes similar to the pairings of a single bacterial OTU with C. albicans seen by Zaborin (25). Dynamic interplay with fastidious bacterial species may govern the behaviour of Candida spp. as an invader, bystander or benefactor given clinical infection is only observed in 5% of those colonised (46) and fungi can recapitulate commensal bacterial functions (47).
Robust and replicable disease-specific patterns were rare (32). It is plausible that pathogen expansion is largely patient-specific with determination of invasive behaviour by interdependent metabolic, ecological and pharmacologic factors, a hypothesis supported by animal studies (23, 48).
In this regard, antimicrobial therapy was seen to be associated with reduced alpha diversity, particularly carbapenems (38). Ojima (15) noted the proportion of phylum Actinobacteria was severely reduced by this class and was specifically associated with Pseudomonas abundance (p < 0.01 (16). Antibiotic usage beyond ICU discharge correlated with sustained loss of alpha diversity in short stay cardiothoracic patients (21).
Antibiotic duration correlated to Acinetobacter spp. and Staphylococcus aureus abundance (18) and exposure to more than 6 antibiotics correlated to development of MDR(25). Emergence of typical MDR organisms were reported in 17 of 26 studies (see supplementary Fig. 5). Garcia (17) noted a median 8 day stay in ICU before acquisition of resistance if this was to occur, the same timeframe where diversity is most depleted (49) and where recovery of commensals began, if at all (15). Circumstantially, these observations suggest day 7–8 in ICU and/or ICU discharge may be imperative junctures for antibiotic rationalisation(15, 17, 21).
There are universal features of early gut dysbiosis amongst the critically ill
Dysbiosis is complex and dynamic within critically ill patients, yet there were a series of commonalities. Taxonomic clustering shows separation of microbiomes from that of endemic volunteers, distant healthy controls and other hospitalised patients (50, 51),. Loss of biogeographic regionalisation across distant body sites was distinctive of critical illness. Inter- and intra-individual variability seen amongst ICU patients is much greater than that seen in healthy controls (10). This variability is poorly explained by diagnosis, syndrome, or univariate clinical parameters and is amplified in patients with sepsis(18). In some studies, a minor contribution is made by demographic factors and disease severity (16) and heritability may be minimal(43), rendering the microbiome an attractive target for ICU interventions.
Regarding composition, richness and evenness is progressively depleted over the first 72hours of ICU stay, with added loss in the presence of broad spectrum antimicrobials. As many as 70 of 73 taxa were decreased in this setting relative to healthy controls (41),although this does not linearly correlate with clinical outcomes. Next generation approaches have built upon culture-based studies showing disproportionate loss of commensal organisms from both major families, Bacteroidetes and Firmicutes, which are typically dominant with concentrations of 1012-1014CFU/ml. Disappearance of certain genera among critically ill patients was consistently noted, especially Faecalibacterium, Blautia, and Ruminococcus, and the functional implications are yet to be fully elucidated from multiomic work.
Pathogenic species are thought to encroach on new ecological niches created by microbiome disruption (23, 41). In all reporting studies, organisms which reached metagenomic dominance by any metric, at any body site, predict both colonisation and infectious outcomes. Longitudinal studies have shown that dominance can predate colonisation by as many as 41 days and nosocomial infection by 3–21 days(32). By contrast, in no healthy control samples did one pathogen constitute more than 70% of abundant taxa (10).
In critical illness, coincident biodiversity loss and pathogen dominance was first reported by Iapachino who found a massive presence of Enteroccocus by denatured gel electrophoresis and selective PCR (52). A decade later, Freedberg et al (23) demonstrated Enteroccocus dominance by 16S analysis (and colonisation culture) at ICU admission was an independent predictor of infection and death. In this review, up to 80% of studies featured high abundance of Enterococcus, associated with mortality in 3 studies – where alpha diversity was not – as well as other unfavourable clinical and ecological sequalae (supplementary Table 3).
Moreover, Enterococcus has been associated with emergence of drug resistance. An epidemiological study of VRE colonisation found this related to high relative abundance of the genus, and in turn, worst outcomes including mortality within 30days (53).. While posited as maladaptive ‘overgrowth’, the mechanism by which Enterococcus moderates clinical outcomes is elusive and may be antibiotic dependent. Within a cohort exhibiting Pseudomonas colonies resistant to carbapenems, piperacillin-tazobactam abolished potentially protective taxa and increased risk of Enterococcus dominance (54).
Box 2. UNIVERSAL FEATURES of the Critically Ill Gut |
There is a dynamic and dramatic reduction of richness and evenness, with disproportionate loss of beneficial commensal organisms |
Overabundance of harmful pathogen organisms accompanies loss of diversity even if antibiotic naïve patients |
Confluence of biogeography is seen between gut and remote body sites |
Dynamic ecological complexity is poorly explained by univariates or disease severity |
Organisms known to confer multidrug resistance grow abundant over time, and abundance can precede clinical culture result |
Heterogeneity and microbiome nomenclature is a barrier to consensus
A major challenge for this meta-analysis is the lack of dysbiosis consensus definition or agreement on core microbiome indices for intra- and inter-individual comparison at time of publication. The recent Strengthening the Organising and Reporting of Microbiome Studies (STORMS) guidelines represent a major step forward in the reporting domain(55).
Heterogeneity is layered in study population, design, sampling and processing methodology, reporting and visualisation. Included studies examine patients admitted to diverse or subspecialist intensive care units and several stratified their results by biomarker, colonisation or other status. Amplicon-based approaches deliver structural taxonomy – conventionally to genus level – upon which function can be inferred, whereas whole-genome sequencing delivers subspecies granularity and functional information however this may obscure hierarchical patterns, with attendant cost and time constraints (9, 56). Validated methods to adjust for these differences at each pipeline stage are nascent(57, 58).
Common alpha diversity metrics are semi-redundant, and the superior choice of these is not established. Selection of metrics utilised was non-systematic introducing high risk of reporting bias, and in cases only pairwise p values were available. Regarding taxon-based metrics, relative abundance was most frequently reported (n = 23 studies) while absolute abundance, Ratio of Firmicutes/Bacteroidetes(59) and Berger-parker index of dominance(60) were less common (3 or fewer studies). Eligible studies traversed clinical, ecological and resistome, or methodological research domains. Without standardisation, quality concerns regarding presentation of exposure and outcome variables will persist.
Beta diversity is analysed by taxonomically weighted or unweighted measures of similarity or difference, usually with PERMANOVA comparison and visualisation by Principle Component Orthogonal Analysis plot (see supplementary Fig. 7). This has been qualitatively assessed as comparative clustering is not readily leant to aggregation. As authors are increasingly publishing their data to open-access platforms, pooled or network analyses warrant evaluation. Future directions are laid out in Table 3.
Limitations
The included studies were small, the largest registering 301 participants (23). The varied clinical implications of dysbiosis are infrequently reported leading to exclusion of several high-quality ecological studies. A consensus on the optimal methods for interpretation of meta-genomic data is still maturing and heterogeneity will not be resolved. There is an attendant risk of type II error.
Several modulators were not interrogated by this systematic review due to lack of available data and inability to aggregate or transform various exposures and outcomes. The effect of vasopressors, renal replacement therapy, feeding, medications, SDD, and ventilation on gut dysbiosis is unknown.