Prediction Of Severe Acute Graft-versus-host Disease By Transcriptome Sequencing Of Mixed Lymphocyte Reaction Between Donor And Recipient Before Transplantatio

BackgroundAllogeneic hematopoietic stem cell transplantation(allo-HSCT)is an important treatment for multiple malignant and nonmalignant hematologic diseases,and severe acute graft-versus-host disease(aGVHD)is a major cause of non-relapsed death after transplantation.Effective prediction for this disease is essential to prevent and treat aGVHD and improve the prognosis of transplantation.The mixed lymphocyte reaction(MLR)is the classic experiment that mimics allogeneic rejection in transplant donors and recipients.With the widespread use of high-throughput sequencing,RNA sequencing results can provide a multidimensional perspective of cellular status at the transcriptome level.Machine learning algorithms can parse disease-related high-throughput results in a non-linear fitting manner to identify key features in biological and medical research data,and in recent years this algorithm has been rapidly applied in aGVHD risk prediction studies.Therefore,exploring the relationship between donor-recipient MLR cell status and aGVHD at the transcriptome level,combined with machine learning algorithms to predicting the risk of aGVHD occurrence may provide new insights and strategies for predicting severe aGVHD and optimising donor selection prior to transplantation.ObjectiveThe aim of this study was to explore the relationship between MLR cell status and aGVHD from the transcriptome level and to clarify whether transcriptome sequencing results of donor-recipient MLR cells combined with machine learning algorithms can effectively predict severe aGVHD.In addition,we would like to further evaluate the biological significance of MLR-related feature genes in aGVHD and its target organs,screen for therapeutic targets with intervention potential,which may provide new ideas and insights for the management of aGVHD.MethodsA total of 86 donor-recipient blood samples and their clinical data were collected in this study,while an additional 57 human leukocyte antigens(HLA)(i.e.,less than 5/10 HLA compatibility)HLA mismatched allogeneic paired(HMAP)blood samples were collected to simulate more violent rejection reactions.RNA was collected and sequenced a few days after MLR on mononuclear cells purified from peripheral blood samples.Bioinformatics analysis of genes associated with severe aGVHD from MLR samples was performed.Machine learning algorithms were used to identify feature genes in severe aGVHD.establish predictive models.To further evaluate the role of MLR-related feature genes in aGVHD and its target organs,we introduced sgRNA libraries targeting feature genes and Cas9 genes into human T cells as donor T cells in a humanized aGVHD mouse model to evaluate the role of genes in aGVHD.Results1.Compared to the transcriptomic results of 0-Ⅱ grade aGVHD MLR cells,the biological processes of MLR cells of grade Ⅲ-Ⅳ aGVHD group and HMAP group exhibited upregulation of antigen processing/presenting and cell killing processes,but downregulation of lipid metabolism and response to vitamin.2.We found that the MAPK cascade signaling pathway was highly correlated with the occurrence of aGVHD through weighted correlation network analysis(WGCNA)and competitive endogenous RNA(ceRNA)analysis.3.The patient data containing MLR sequencing results and clinical information were randomly divided into training and validation sets.The random forest(RF)algorithm and support vector machine(SVM)algorithm were applied in the training set to calculate feature parameter genes that contribute to the prediction accuracy and establish models.The area under the curve(AUC)of the RF model and SVM model for predicting severe aGVHD in the validation set were 0.9 and 0.92,respectively.4.A humanized aGVHD mouse model was constructed by infecting T cells with an sgRNA lentiviral library targeting feature genes,and the results of analyzing the human T cell genome of each target organ enriched at the onset of aGVHD in mice revealed that the distribution of feature genes in different target organs differed at aGVHD.Our analysis also identified genes such as FOXP3、ZBTB7B、TNFRSF14 和 RUNX1 as possible potential targets for intervention in preventing and treating aGVHD.ConclusionsTranscriptome results from pre-transplant donor-recipient MLR cells combined with machine learning algorithms can effectively predict severe aGVHD,and the results provide a new strategy for predicting severe aGVHD and optimizing donor selection prior to transplantation in the clinical setting.BackgroundPatients with hematologic diseases are at high risk for infection with carbapenem-resistant Enterobacteriaceae(CRE),and CRE bloodstream infections(BSI)have a high mortality rate.Early empiric treatment targeting CRE significantly decreases mortality in patients with CRE BSI.Patients with colonized CRE and patients with previous CRE infection are usually at higher risk of subsequent CRE BSI,but early prophylaxis for all patients would lead to antibiotic abuse and increase the economic burden for patients without subsequent CRE BSI.Therefore,the development of prediction tools for the development of subsequent CRE BSI in hematologic patients with CRE detected is urgently needed to identify high-risk patients eligible for empiric anti-CRE therapy and thereby improve the prognosis of this group of patients.ObjectiveTo develop a predictive risk model for subsequent BSI in hematological patients with CRE isolated from perianal swabs could be useful to target preventive strategies.MethodsThis was a single-center retrospective cohort study at a 769-bed tertiary blood diseases hospital which included all hematological patients hospitalized from 10 Oct 2017 to 31 July 2021.We developed a predictive model using multivariable logistic regression and internally validated it using enhanced bootstrap resampling.ResultsOut of 421 included patients with CRE isolated from perianal swabs CRE BSI occurred in 59.According to the multivariate logistic analysis,age(OR=1.04,95%CI:1.01-1.06,P=0.004),both meropenem and imipenem minimal inhibitory concentration(MIC)of isolate from perianal swabs>8ug/ml(OR=5.34,95%CI:2.63-11.5,P<0.001),gastrointestinal symptoms(OR=3.67,95%CI:1.82-7.58,P<0.001),valley absolute neutrophil count>0.025(109/L)(OR=0.07,95%CI:(0.02-0.19,P<0.001)and shaking chills at peak temperature(OR=6.94,95%CI:(2.60-19.2,P<0.001)were independently associated with CRE BSI within 30 days and included in the prediction model.At a cut-off of prediction probability ≥21.5%the model exhibited a sensitivity,specificity,positive predictive value and negative predictive value of 79.7%.85.6%,96.27%and 47.47%.Discrimination and calibration of the prediction model was good on derivation data(C-statistics,0.8898;Brier score,0.079)and enhanced bootstrapped validation dataset(adjusted C-statistics.0.881;adjusted Brier score,0.083).The risk prediction model is freely available as mobile application at https://liujia1992.shinyapps.io/dynnomapp/.ConclusionsThe prediction model based on age.meropenem and imipenem MIC of isolate from perianal swabs,gastrointestinal symptoms,valley absolute neutrophil count and shaking chills may be used for better targeting interventions in hematological patients with CRE isolated from perianal swabs.