Enteric Microbiome and Liver Transplantation

Liver transplantation (LT) has changed the life expectancy of end-stage liver disease (ELD) patients. However, important issues may hamper the early post-LT period (e.g. graft dysfunctions, infectious complications). Risk stratification in ELD patients is based on clinical scores which are often not predictive for the LT outcomes. More robust scores are therefore needed.

It is known that microbial flora may play an important role in predisposing to several pathological conditions. This is particularly true for the liver, which is constantly exposed to high load of gut microbial antigens and metabolites. The effects of these factors have not been studied on the transplanted liver yet. The investigators will study the faecal microbiome of 275 LT patients, and, in combination with a large panel of clinical, lab and functional parameters, will correlate it to different clinical outcomes.

In particular, the following possible LT outcomes will be addressed:

1. Early allograft dysfunction (30-40% estimated incidence)
2. Treated acute cellular rejection (10-15%). Evaluated through lab parameters of liver damage and, when possible, confirmed by histopathological evaluation of liver biopsies
3. Infectious complications (10-15% divided in microbiologically confirmed and clinically suspected)
4. Length of stay in the hospital after LT
5. Mortality at 30, 90 and 365 days (7-8% at 1 year)
6. Biliary complications (10-15%)

220 adult patients undergoing orthotopic LT (OLT) will be enrolled (months 1-18) and followed for 1 year after LT. Months 19-24: 55 pts will be enrolled as internal validation cohort, and monitored until the end of the study.

Stool and blood will be sampled at the following timepoints:

T0. Pre-LT (within the 3 months before LT) T1. Early Post-LT (7 days from surgery) T2. Late Post-LT (90 days from surgery)

Stool will be used for microbiome profiling and investigation of intestinal inflammation.

Permeability analysis, evaluation of circulating catecholamines and of bacterial metabolites will be performed also on blood.

Clinical and lab data will be collected. Clinical scores (MELD and Child-Pugh), clinical complications and graft/patient survival will be recorded throughout the observation period.

Receiver operating characteristic (ROC) curves of microbiome data will be calculated at different taxonomic levels for all investigated outcomes. Curves with an area under the curve (AUC) >0.6 and a p value ≤0.05 will be considered potentially relevant. The most informative and inclusive microbiome cutoffs at the lowest significant taxonomic level (usually the family level) will be chosen and used with all the other clinical variables in contingency tables to estimate their association with the different outcomes (Chi-square test). Single, even if less inclusive, microbiome cutoffs indicating extreme dysbiosis (occupation of >30% of the microbiota by a single predominating bacterial taxon), will also be chosen from non-significant ROC curves and further investigated. Generalized Linear Model (GLM) will then be used for each outcome except survival, for which Cox regression will be used. All P values will be adjusted for False Discovery Rate.

All the analyzed variables will be considered in multivariate analysis, together with the typical clinical assessments of liver transplantation procedures. These include: clinical scores (i.e. Child-Pugh and MELD), hematologic lab analyses (leukocytes, erythrocytes, hemoglobin, hematocrit, platelets), biochemical lab analyses (creatinine, urea, sodium, potassium, ALT, AST, total Bil, GGT, ALP, albumin, ammonium, CRP, circulating catecholamines), coagulation tests (PT, PTT), and drug treatments at the different time points (including antibiotics, immunosuppressive regimens and laxatives). The predictive model by the "best subset" approach optimizing the Akaike Information Criterion (AIC) will be selected. The model selection will also consider possible interactions with different underlying conditions, such as hepatocellular carcinoma, nonalcoholic fatty liver disease/nonalcoholic steatohepatitis, and comorbidities such as diabetes and renal insufficiency In this phase the investigators will also estimate the model performance (accuracy, sensitivity, specificity, positive predictive value, negative predictive value) by 10-fold cross validation to avoid too optimistic estimates. As comparison, a Machine Learning model will also be fit.

As the data of the patients enrolled in the second year will be available, the investigators will validate the predictive model in the independent sample.