Construction Of A Prognostic Model For Acute Myeloid Leukemia Based On Endoplasmic Reticulum Stress And M6A Modification-related Gene

Part Ⅰ Construction of endoplasmic reticulum stress related prognostic model in acute myeloid leukemiaBackgroundAcute myeloid leukemia(AML)is a highly invasive hematologic malignancy characterized by complex cytogenetic abnormalities and heterogenous genetic mutation,which causes strong heterogeneities and poor prognosis in patients.Prognostic assessment can optimize treatment and improve patient survival.Traditional ELN risk stratification contains characteristics of chromosomal aberrations and genetic mutations,but needs to be improved due to complex assessment system,long detection periods and limitation of fitness in all patients.In addition,risk stratification of ELN does not fully consider the problem of patient chemotherapy response,since it is mainly is derived from pathogenic abnormal karyotypes and mutated genes,which cannot reflect all the information of poor chemotherapy response.However,the basic research believes that the main reason of occurring refractory and relapsed AML after chemotherapy in patients is due to multi-drug resistance.Studies have found that endoplasmic reticulum(ER)stress can integrate the stimulation of the internal and external environment and coordinate multiple biological functions and mechanisms to regulate cell survival and death,which has an important impact on the chemotherapy response of AML.ObjectiveTo screen differentially expressed genes(DEGs)between sensitive and refractory or relapsed(R/R)AML patients,and establish an ER stress-related prognostic model for AML in public transcriptome database to refine the ELN risk stratification.Methods15 sensitive and 26 R/R species from patients were collected and conducted with RNA-seq to screened DEG,which formed set A.After filtering clinical information and follow-up data,115 TCGA patients were included as the model construction training cohort.A total of 723 genes associated with ER stress were searched out in Gene Oncology(GO)databases.The 723 genes were then conducted by univariate COX regression to obtain genes with prognostic significance,which formed set B.Set A and set B were further intersected to screen genes both related to therapeutic response and prognosis.LASSO regression and multivariate COX regression were performed on the filtered genes to construct the prognostic model in training cohort.The GES10358 and GSE37642 data sets were used as the external validation cohorts,respectively.KaplanMeier curve,multivariate COX regression and time dependent receiver operator characteristic curve(tROC)were used to evaluate the prognostic efficacy of the model in the training and validation cohorts,which was further compared with ELN risk stratification.We used the "pRRophetic" algorithm to predict chemotherapy response of cytarabine,sorafenib and midostaurin,etc.,and compared differences between low-risk and high-risk patients.Results644 DEGs were screened out between sensitive and R/R patients,which formed set A.81 out of 723 ER stress related genes were prognostically significant in TCGA cohort,which formed set B.20 genes both related to chemotherapeutic response and prognosis were identified by intersecting set A and set B.The 20 genes were further analyzed with LASSO regression and 5 important ER stress-related genes were identified.The 5 genes were then conducted with multivariate COX regression and a linear model was established with coefficients and gene expression values as follows:risk score=(0.1540)*RTN4R+(-0.6358)*PDIA6+(-0.5071)*CYP2E1+0.3349*CALCRL+0.0897*ARGE.Kaplan-Meier curve and tROC analysis showed that the HR of the model in the training set was 4.86(2.79-8.44),and the AUCs at 1,3 and 5 years were 0.74(0.65-0.84),0.83(0.73-0.93)and 0.89(0.79-0.99),respectively.Multivariate analysis indicated the prognostic independency of this model.In the validation sets GSE10358 and GSE37642,the HRs were 2.57(1.37-4.80)and 1.71(1.34-2.1),respectively.The 1-year and 3-year AUCs were 0.67(0.55-0.79)and 0.75(0.61-0.89)in GSE10358,and 0.63(0.88-0.69)and 0.66 0.60-0.73)in GSE37652.These results suggested a good prognostic efficiency of the model.In training set,compared to ELN stratification with an AUC of 0.72(0.62-0.82)for 3-year survival prediction,the AUC 0.83(0.73-0.93)of the prognostic model was higher,suggesting the model was better than ELN risk stratification.Drug response analysis showed the model could distinguish chemotherapy response to cytarabine,sorafenib and midostaurin in AML patients.ConclusionAn AML prognostic model is established with ER stress-related genes,which can effectively predict the survival of AML patients.The model could refine the predictive efficiency of ELN risk stratification and provide potential value to guide clinical treatment decision-making.Part Ⅱ Construction of m6A modification related prognostic model in FLT3 mutated acute myeloid leukemiaBackgroundIn Part Ⅰ,we found the ER stress-related prognostic model had a relatively poor prognostic capacity in acute myeloid leukemia(AML)with FMS-like tyrosine kinase(FLT3)gene mutation.FLT3 mutation is the most common molecular subtype of AML,accounting for 20%to 30%of AML patients.The protein product of FLT3 mutation can highly activate intracellular proliferation-related signaling pathways,resulting in malignant proliferation of leukemia cells,which is closely related to chemotherapy resistance in patients.Therefore,patients with FLT3 mutated AML have a relatively poor prognosis and can be heterogenous due to mutation types,mutation sites and allelic ratios,causing difficulties in prognosis assessment.It is necessary to develop personalized prognostic assessment for FLT3 mutated AML.N6-Methyladenosine(m6A)modification is the most common RNA modification in mammals.m6A modification can participate in mRNA stability,splicing,decay and translation,which dramatically modulates gene expression.Studies have reported the abnormal expression of m6A modified proteins in FLT3 mutated AML and is associated with disease progression and therapeutic resistance.We established a prognostic model based on m6A modification-related genes in FLT3 mutated AML.ObjectiveTo establish a prognostic model for FLT3 mutated acute myeloid leukemia(AML)patients based on m6A modification-related genes.MethodsGSE6981,GSE61804 and GSE10358 datasets were used to analyze the differential m6A enzymes and reading proteins between AML patients with positive and negative FLT3 mutation.Correlation analysis was used to screened m6A modified-related genes in expression profiles and venn diagram presented the genes correlated to m6A modification among the three cohorts to establish the model.83 FLT3 mutated AML samples were included by integrating the TCGA and BEAT clinical and follow-up data.The prognosis model was constructed by univariate COX regression and LASSO regression analysis.Kaplan-Meier curve was used to analyze the survival difference between patients in low and risk group.We used multivariate COX regression to evaluate the prognostic independency.The time-dependent receiver operating curve(tROC)was used to evaluate the predictive effectiveness of the model.ResultsA total of 14 m6A modified enzymes or reading proteins were abnormally expressed between FLT3 mutation positive and negative patients.Correlation analysis and venn diagram presented 2476 m6A modified related genes among the three GEO datasets to establish the model.The 2476 genes were further conducted with univariate COX regression in TCGA and BEAT integrated data,which identified 132 prognostic genes.By LASSO regression,seven candidate genes were selected to establish the prognostic model:risk score=(-0.28*AKAP9)+(-0.11*AVEN)+(0.13*DMAC1)+(0.10*DPYD)+(0.05*FAR2)+(-0.18*GPHN)+(0.73*SPECC1L).KaplanMeier curve showed the risk model evaluated overall survival(OS)with a hazard ratio(HR)of 5.08(95%CI 2.54-10.14).HR in multivariate COX regression was 3.86(95%CI 1.88-7.94).The area under the curve in tROC for 1-and 3-year OS were 0.83(95%CI 0.73-0.93)and 0.94(95%CI 0.86-1.02),respectively.ConclusionWe utilized m6A modification-related genes to establish a prognostic model effectively evaluating the prognosis and survival of FLT3 mutant AML,which is expected to potentially help improve the clinical prognosis stratification and treatment decision-making.