Exploratory Research On The Relationship Between Gut Microbiota And Post-hepatectomy Liver Failure(PHLF) In Hepatitis B Virus-related Hepatocellular Carcinoma(HBV-HCC)

The study is organized into three parts:1.Construction of a predictive model for post-hepatectomy liver failure（PHLF）in hepatitis B virus-related hepatocellular carcinoma（HBV-HCC）based on gut microbiome.2.Gut microbiome combined with clinical factors to construct the prediction model of post-hepatectomy liver failure（PHLF）associated with hepatitis B virus-related hepatocellular carcinoma（HBV-HCC）.3.Integrated omics analysis: changes in intestinal Klebsiella before and after surgery in patients with hepatitis B virus-related hepatocellular carcinoma（HBV-HCC）and the relationship of post-hepatectomy liver failure（PHLF）.A study of 112 patients in liver ward 2 as the training set and 48 patients in liver ward 1 as the validation set collected in our center was carried out: on preoperative state,the differential or characteristic flora of the patients with PHLF and in the patients without PHLF can provide the reference evidence to the microbiome dimension for the prediction of PHLF in HBV-HCC patients.Furthermore,a new prediction model is established by fitting clinical characteristic factors on the basis of gut microbiome,which makes it possible to optimize the disease prediction.Subsequently,29 patients with HBV-HCC were continuously collected from our center,the potential mechanism of Klebsiella involvement in PHLF was revealed by integrating analysis of microbiome and metabolomic data.These provide a potential possibility for further study of the mechanism between GM and PHLF.Part Ⅰ: Construction of a Predictive Model for Post-hepatectomy Liver Failure（PHLF）in Hepatitis B Virus-related Hepatocellular Carcinoma（HBV-HCC）Based on Gut MicrobiomeObjectives To study the potential relationship between gut microbiome（GM）and PHLF in patients with HBV-HCC;based on the differential flora or characteristic OTU level,various machine learning models were used to construct the prediction model of PHLF in HBV-HCC patients respectively;Explore the capabilities of predictive models using the various of model’s evaluation metrics.Multi-level and multi-angle demonstrate the importance of intestinal flora as a potential biomarker in auxiliary predict the occurrence of PHLF in patients with HBV-HCC.Methods According to the inclusion and exclusion criteria,from September 2020 to August 2021,HBV-HCC patients in the two liver wards of Guangxi Medical University Cancer Hospital were continuously collected.And Finally,a total of 160 patients were included as the study subjects,including 112 patients in liver ward 2 as the training set and 48 patients in liver ward 1 as the validation set.The clinical data and stool samples（before operation）of the enrolled patients were collected,and in accordance with the standard of International Study Group of Liver Surgery（ISGLS）,the group with PHLF（bo.PHLF）and the group without PHLF（nbo.PHLF）were judged.Stool samples was sequenced and analyzed by 16 S r RNA using Illumina Novaseq6000 platform.LEf Se analysis was used to screen differential GM and Boruta algorithm was used to screen characteristic OTU.40 bo.PHLF patients and 72 nbo.PHLF patients in liver ward 2 were used as discovery cohort（training set）to identify potential biomarkers with predictive value,using the methods of Random forest（RF）,Support Vector Machine（SVM）,and XGBoost built the prediction models respectively.15 bo.PHLF patients and 33 nbo.PHLF patients in ward 1 were included in the independent validation set.Area Under Curve（AUC）,sensitivity,specificity,Yuden index,false negative rate,recall rate and coincidence rate were used to evaluate the evaluation ability of GM models Under different screening methods.Z-test was used to evaluate whether there was any difference in AUC between models.Results 1.There was no significant difference in α diversity between bo.PHLF and nbo.PHLF（P>0.05）.Weighted unifrac（Stress =0.093）of NMDS indicated that β diversity was well differentiated between bo.PHLF and nbo.PHLF groups.2.Analysis of predictive markers in LEf Se analysis（LDA=2）showed that:On Genus level,in bo.PHLF group,Faecalibacterium,Dorea,Plesiomonas,Ralstonia,Marvinbryantia,Erysipelotrichaceae_UCG-003, unidentified_Ruminococcaceae,unidentifie_[Eubacterium]_coprostanoligenes_group,Haloactinopolyspora,Enhydrobacter,F0332,Oribacterium were increased;in nbo.PHLF,Bacteroides,Dialister,Paraprevotella,Parabacteroides,Flavonifractor,Oscillibacter,Hungatella,Defluviitaleaceae_UCG-011,JGI₀001001-H03,GCA-900066755,CL500-29_marine_group were increased.On Species level,in bo.PHLF,Faecalibacterium_prausnitzii,Plesiomonas_shigelloides,Ralstonia_pickettii,[Clostridium]_leptum,Moraxella_osloensis,Aggregatibacter_segnis were increased;in nbo.PHLF,Bacteroides_vulgatus,Bacteroides_stercoris,Bacteroides_uniformis,Bacteroides_xylanisolvens,Bacteroides_thetaiotaomicron,Parabacteroides_sp_CT06,Parabacteroides_merdae,Dialister_pneumosintes were increased.3.A total of 7 features are found at the OTU level: OTU₄,OTU₂88,OTU₉06,OTU₁021,OTU₁756,OTU₅926,OTU₆143.4.Constructing a diagnostic model of GM and predicting performance with3 machine learnings: the worst model performance at the Species level.Both Genus and OTU level models had their own advantages and disadvantages on predicting PHLF,among which XGBoost model on Genus level,SVM model and XGBoost model on OTU level all have better predictive performance.The predicted results are as follows: genus-level XGBoost model（sensitivity 80.00%,specificity 75.80%,Yuden index 0.558,false negative rate 20.00%,recall rate80.00%,coincidence rate 77.08%,AUC 0.7960（95%CI:0.6500-0.9419））.OTU level SVM model（sensitivity 73.30%,specificity 72.70%,Yuden index 0.46,false negative rate 26.67%,recall rate 73.30%,coincidence rate 72.92%,AUC0.7273（95%CI : 0.6016-0.8530））.OTU level XGBoost model（sensitivity73.30%,specificity 75.80%,Yuden index 0.491,false negative rate 26.67%,recall rate 73.30%,coincidence rate 75.00%,AUC 0.7616（95%CI :0.6216-0.9017））.Conclusions This study suggests that there is a strong association between gut microbiome（GM）and PHLF.GM can provide the reference evidence to the microbiome dimension for the prediction of PHLF in HBV-HCC patients.Detecting the GM has the advantages of damage-less and non-invasive application.GM,as a key predictor,using machine learning methods to build a predictive model of PHLF in HBV-HCC patients,has potential clinical utility.This study confirmed the feasibility of GM in the study of PHLF in HBV-HCC patients;found a new entry point between GM taxonomy and clinical application;it provides a new idea for the accurate diagnosis and treatment of this kind of diseases.It may provide some basis for early implementation of clinical intervention.Part Ⅱ: Gut Microbiome Combined with Clinical Factors to Construct the Prediction Model of Post-hepatectomy Liver Failure（PHLF）Associated with Hepatitis B Virus-related Hepatocellular Carcinoma（HBV-HCC）Objectives Gut microbiome（GM）,combined with clinical characteristic factors,is used to establish a new prediction model of PHLF in HBV-HCC patients,and to explore whether the model’s efficiency was optimized compared with the prediction model of bacterial flora alone,and whether it was more feasible to improve the value of clinical application.Methods Boruta was used to screen the characteristics of clinical factors and colinear analysis was performed on the selected clinical factors.The selected clinical factors were analyzed by distance-based Redundancy analysis（db-RDA）.Then,random forest（RF）,support vector machine（SVM）and XGBoost new prediction models were established based on mathematical transformation of selected clinical factors combined with Genus level and OTU level of Part I.The ability of each model was comprehensively evaluated based on Area Under Curve（AUC）,sensitivity,specificity,Jorden index,false negative rate,recall rate and coincidence rate.The effectiveness of clinical application was compared between all models.Results 1.The correlation coefficient between the selected clinical factors TBIL and PA was-0.0280,and its absolute value was <0.8,indicating no collinearity between them.2.db-RDA analysis: TBIL: r2 = 0.0745,P = 0.0190);PA: r2 = 0.0981,P =0.0045 showed significant influence on the microbial community in the samples.The degree of distinction is statistically significant.3.At the genus-level and OTU level,GM combined with clinical factors TBIL and PA to construct three new machine learning models respectively.The XGBoost model of OTU combined with TBIL and PA finally achieved the best performance in comprehensive evaluation and prediction ability and improvement of clinical detection rate.On the test set,the sensitivity was93.30%,the specificity was 75.80%,the Yuden index was 0.691,the false negative rate was 6.70%,the recall rate was 93.30%,the coincidence rate was81.25%,and the AUC was 0.8424（95%CI: 0.7275-0.9574）.Conclusions The combination of GM and clinical factors TBIL and PA has better clinical application value in predicting PHLF in HBV-HCC patients than when using bacterial flora alone.Among them,the advanced machine learning new XGBoost model may be more suitable for the integrated clinical data dimension integration GM model at the OTU level,thereby improving the efficiency of clinical application and providing the possibility to optimize the disease prediction.Part Ⅲ Integrated Omics Analysis: Changes in Intestinal Klebsiella Before and After Surgery in Patients with Hepatitis B Virus-related Hepatocellular Carcinoma（HBV-HCC）and the Relationship of Post-hepatectomy Liver Failure（PHLF）Objectives To explore the correlation between the significant increase of Klebsiella in gut microbiome（GM）after surgery and the occurrence of PHLF in HBV-HCC patients.Comprehensive omics was used to analyze and explore the underlying mechanism.Methods According to the inclusion and exclusion criteria,a total of 29HBV-HCC patients from the liver ward of Guangxi Medical University Cancer Hospital were continuously collected as the study subjects.Patients with HBV-HCC were divided into Groups PHLF and non-PHLF based on the presence or absence of PHLF.Baseline data,preoperative and postoperative stool and blood samples were collected.Changes in GM of a total of 29 patients before and after operation were compared using 16 S r RNA analysis,and preoperative and postoperative GM changes in the patients with PHLF and the patients without PHLF respectively;preoperative GM changes in the patients with PHLF and in the patients without PHLF,and postoperative GM changes in the patients with PHLF and in the patients without PHLF.The preoperative and postoperative metagenomic information of patients with PHLF was predicted by PICRUST2.In addition,non-targeted metabolites were detected by（LC-MS/MS）method,and the potential mechanism of Klebsiella involvement in PHLF was revealed by combining multiple omics.Results 1.Surgical treatment lead to GM disorder in HBV-HCC patients.2.In addition,there was significant difference in the change amount of Klebsiella before and after operation between the liver failure group（P<0.05）,there was no statistical significance in the postoperative and preoperative group without liver failure（P>0.05）.In the preoperative status,the number of Klebsiella were no statistical significance between the two groups.In the postoperative status,the number of Klebsiella in the postoperative liver failure group was significantly higher than that in the non-postoperative liver failure group（P<0.05）.3.In the liver failure group,Klebsiella plays a role in 13 amino acid related pathways,and it also plays a significant role in the branched-chain amino acid metabolic pathway.4.In the liver failure group,3-methyl-2-oxybutyric acid（BCAA）,a hub metabolite shared by Klebsiella with feces and serum,had a high negative correlation in the BCAA pathway（P=0.02,r=0.51）.Conclusions Hepatectomy can lead to an imbalance of GM in HBV-HCC patients.Due to its potential connections with 3-methyl-2-oxobutanoic acid in the BCAA pathway,significantly increased Klebsiella has the potential to be an evaluation indicator of PHLF in HBV-HCC patients.Moreover,3-methyl-2- oxobutanoic acid has research value in PHLF-targeted treatments.All these provide a potential possibility for further study of the mechanism between GM and PHLF.