Cellular abundance-based prognostic model associated with deregulated gene expression of leukemic stem cells in acute...

Background: Previous studies have reported that genes highly expressed in leukemic stem cells (LSC) may dictate the survival probability of patients and expression-based cellular deconvolution may be informative in forecasting prognosis. However, whether the prognosis of acute myeloid leukemia (AML) can be predicted using gene expression and deconvoluted cellular abundances is debatable. Methods: Nine different cell-type abundances of a training set composed of the AML samples of 422 patients, were used to build a model for predicting prognosis by least absolute shrinkage and selection operator Cox regression. This model was validated in two different validation sets, TCGA-LAML and Beat AML (n = 179 and 451, respectively). Results: We introduce a new prognosis predicting model for AML called the LSC activity (LSCA) score, which incorporates the abundance of 5 cell types, granulocyte-monocyte progenitors, common myeloid progenitors, CD45RA + cells, megakaryocyte-erythrocyte progenitors, and multipotent progenitors. Overall survival probabilities between the high and low LSCA score groups were significantly different in TCGA-LAML and Beat AML cohorts (log-rank p-value = $$ and $$, respectively). Also, multivariate Cox regression analysis on these two validation sets shows that LSCA score is independent prognostic factor when considering age, sex, and cytogenetic risk (hazard ratio, HR = 2.17; 95% CI 1.40-3.34; p < 0.001 and HR = 1.20; 95% CI 1.02-1.43; p < 0.03, respectively). The performance of the LSCA score was comparable to other prognostic models, LSC17, APS, and CTC scores, as indicated by the area under the curve. Gene set variation analysis with six LSC-related functional gene sets indicated that high and low LSCA scores are associated with upregulated and downregulated genes in LSCs. Conclusion: We have developed a new prognosis prediction scoring system for AML patients, the LSCA score, which uses deconvoluted cell-type abundance only.