Prediction Of Acute GVHD And Herpesvirus Infection By Metabolic Biomarkers Using GC-MS After Allo-HSCT

PART ? Prediction of acute graft-versus-host disease(aGVHD)by metabolic biomarkers after allo-HSCT based on GC-MS approach[Objective]To explore potential metabolic biomarkers for acute graft-versus-host disease(aGVHD)and their relationship with relapse in allogeneic hematopoietic stem cell transplantation(allo-HSCT)patients based on gas chromatography-mass spectrometry(GC-MS)approach.[Methods]GC-MS was employed to analyze four hundred and fifty-six serum samples from one hundred and fourteen consecutive patients who underwent allo-HSCT from January 2012 to May 2014 collected on day-9,0,7 and 14.The stability of GC-MS approach was validated by relative standard deviation(RSD)of quality control(QC)samples.The metabolites in serum were identified by AMDIS,NIST,and retention index.We randomly selected 13 patients with 0-? aGVHD and 15 with ?-? aGVHD to set up a discovery set.The rest of the patients were included in the validation set.T-test,ROC curves and false discovery rate(FDR)were used to detect metabolite difference between aGVHD and non-aGVHD patients.Logistic regression was used to establish prediction model of aGVHD.Relapse was analyzed with Gray's test and competing risk regression.[Results]1.Validation of GC-MS approach.In QC samples,the proportion of metabolites with common peak area RSD below 15%was 69%,which indicated the GC-MS approach met the criteria of serum sample analysis.2.Serum metabolites identification.16 metabolites were identified based on AMDIS,NIST and retention time.They were pyruvic acid,lactic acid,valine,leucine,urea,glycerol,proline,phenylalanine,galactose,glucose,palmitic acid,mannose,linoleic acid,oleic acid,stearic acid,and cholesterol.3.The median follow-up time for the whole cohort was 42(range,0.2 to 60 months)months.The median time for neutrophil recovery was 12 days(range,10 to 37 days),and the median time for platelet recovery was 14 days(range,8 to 90 days).Of the 114 patients evaluated for aGVHD,72 patients had grade 0-? aGVHD and 42 patients had grade ?-?aGVHD.The median day for developing aGVHD was day 30.Fifty-one patients passed away and the 5-year probability of overall survival(OS)was 55.26%.Thirty-one patients had relapsed by the time of the last follow-up.The incidence of relapse was 27.19%.4.In aGVHD discovery set,the t-test and ROC curve analyses showed significant differences in palmitic acid(PA)and stearic acid(SA)between the two groups of patients on day 7.PA levels were lower and SA levels were higher in the 0-? aGVHD group of patients.We further analyzed the ratio of these two metabolites(i.e.,SA:PA)because PA can be converted to SA.And SA:PA was significantly different between the two groups of patients on day 7 based on t-test and FDR(p=0.02).The SA:PA ratio(AUC=0.836)was more effective in the diagnosis of ?-? aGVHD compared with either of the two metabolite biomarkers alone(PA AUC=0.754,SA AUC=0.723).These results were also observed in validation set.5.Univariate analysis showed that stem cell source and donor/recipient sex match,which were included in the subsequent models,were associated with the occurrence of aGVHD.Multivariate analysis showed that SA decreased,whereas PA increased,the likelihood of ?-? aGVHD,and that patients with high SA:PA ratios on day 7 after HSCT were less likely to develop ?-? aGVHD than patients with low SA:PA ratios(odds ratio[OR]=0.06,95%CI 0.02-0.18,p<0.001).Comparison of ROC curves for the clinical characteristics alone with those for the clinical characteristics combined with PA,SA,and the SA:PA ratio to predict aGVHD showed significant increases in the AUC in all markers when combined with clinical characteristics;the SA:PA ratio had a significantly larger AUC than PA(p=0.006)or SA(p=0.01).According to the leave-one-out cross validation,the prediction accuracy of the model with SA:PA was 0.807.Therefore,low SA:PA level is an independent risk factor of grade ?-? aGVHD and could be an excellent marker for aGVHD prediction.6.We discussed the relationship between SA:PA and relapse because patients with aGVHD are less likely to relapse.Using Gray's test,relapse was more frequent in the patients with high compared with low SA:PA ratios(p=0.04),although there was no significant difference in nonrelapse mortality between the two groups(p=0.89).After the adjustment for clinical characteristics,patients in the high SA:PA ratio group were significantly more likely to relapse than those in the low ratio group(HR=2.26,95%CI 1.04-4.91,p=0.04).Thus,high SA:PA level is an independent risk factor of relapse.[Conclusion]The SA:PA ratio on day 7 after HSCT is an excellent biomarker to predict both aGVHD and relapse.Low SA:PA level is an independent risk factor of grade ?-? aGVHD,and it is also an independent protective factor of relapse.PART ? Prediction of herpesvirus infection by metabolic biomarkers after allo-HSCT based on GC-MS approach[Objective]To explore potential metabolic biomarkers for cytomegalovirus(CMV)and Epstein-Barr virus(EBV)infection in allo-HSCT patients based on GC-MS approach.[Methods]GC-MS approach,samples and metabolite identification were the same as those in Part?.Combined with clinical characteristics,one hundred and two patients were involved in CMV analysis,whereas one hundred and fourteen patients were evaluated for EBV infection.T-test and false discovery rate(FDR)were used to detect metabolite difference between patients with herpesvirus infection and those without herpesvirus infection.Logistic regression was used to establish prediction models of CMV and EBV infection.[Results]1.The cumulative incidence of CMV reactivation was 50.00%.The median time for CMV reactivation was 2.9 months.The cumulative incidence of EBV infection was 14.91%.The median time to EBV reactivation was 8 months.2.Mannose can be used as a biomarker for CMV infection prediction.The t-test analysis showed significant differences in urea(day-9,day 0),mannose(day-9 and day 14),glucose(day 7),valine(day 14),and leucine(day 14)between patients with CMV infection and those without CMV infection.After FDR correction,only urea(0 day)was significantly different in two groups.We further analyzed the ratio of mannose on day-9 and on day 14(i.e.,mannose ratio)because mannose levels were significantly different on day-9 and day 14 in two groups.After FDR correction,mannose ratio was still marginally significantly different between two groups.After the adjustment for clinical characteristics in logistic regression,high mannose ratio was a risk factor of CMV infection(OR=4.36,95%CI,1.59-12,p=0.004).The AUC of the prediction model with clinical characteristics was 0.732,whereas the AUC of the model with the mannose ratio was 0.786(p=0.09768).Therefore,high mannose ratio was an independent risk factor of CMV infection.3.Valine,leucine,galactose and glucose can be used as biomarkers for EBV infection prediction.The t-test and FDR analyses showed significant differences in galactose(day-9,day 0)and mannose(day 0)between patients with EBV infection and those without EBV infection.We performed propensity score matching(PSM)based on clinical characteristics(1:1 match).And then,the galactose changes from day-9 to day 0 was calculated.The t-test analysis showed that there was marginally significant difference of galactose changes between Bu/Cy and TBI/Cy groups(p=0.07).Thus,different conditioning could affect galactose levels.So we further analyzed the metabolites in Bu/Cy and TBI/Cy groups independently.In Bu/Cy group,the t-test and FDR analyses showed that there was no significantly different metabolites between patients with EBV infection and those without EBV infection.In TBI/Cy group,the t-test and FDR analyses showed significant differences in valine,leucine,galactose and glucose on day 7 between patients with EBV infection and those without EBV infection.Those who had low valine,leucine,galactose and glucose levels were more likely to develop EBV infection after HSCT.Principal component analysis(PCA)was applied to analyzing valine and leucine since both of them belong to branched-chain amino acids.And the first principal component accounted for 96.070%of the variability.So the first principal component was defined as ValLeu(ValLeu=0.980*Val+0.980*Leu)and was treated as a new marker.Similarly,we used PCA to analyze galactose and glucose because both of them are carbohydrates.And the first principal component was defined as GalGlu(GalGlu=0.999*Gal+0.999*Glu)and accounted for 99.830%of the variability.In logistic regression,four prediction models were developed.They were base model(only clinical characteristics involved),model with ValLeu,model with GalGlu,and model with ValLeu and GalGlu.As a result,patients with high ValLeu and high GalGlu were less likely to develop EBV infection(ValLeu OR=0.05,95%CI,0-0.68,p=0.024;GalGlu OR=0.14,95%CI,0.01-1.72,p=0.124).Comparison of ROC curves for the four models,the model with Valleu and GalGlu(AUC=0.889)had larger AUC than that of the base model(AUC=0.732),which was significantly different(p=0.04).Therefore,Valine,leucine,galactose and glucose can be used as biomarkers for EBV infection prediction in patients receiving TBI/Cy.[Conclusion]Our findings suggest that mannose could be a biomarker to predict CMV infection after HSCT.Valine,leucine,galactose and glucose can be used as biomarkers for EBV infection prediction in patients receiving TBI/Cy.