Abstract
Acute Myeloid Leukemia (AML) is a devastating disease associated with high morbidity and mortality. The role of supportive immune cells such as NK cells and mast cells in AML tumorigenesis and relapse has been increasingly scrutinized. However studies to evaluate impact of varying cell type fraction on the clinical outcome are lacking. Recently, a software tool CIBERSORTx (Cell-type identification by estimating relative subsets of RNA transcripts) has unveiled opportunities to use bulk gene expression data for cell type fraction estimation. CIBERSORTx uses a reference matrix of cell-specific expression and mathematical deconvolution to estimate cell-type abundances from bulk tissue transcriptomes. The leukocyte signature matrix (LM22), which consists of 22 hematopoietic lineages, is a well-characterized reference matrix based on the expression profiles of 547 genes. In this study, we leveraged the CIBERSORTx algorithm to develop a cell type score (CiberScore) using the gene-expression data obtained from diagnostic bone marrow specimens (using Affymetrix U133A microarray) from pediatric AML patients treated on the multi-site AML02 clinical trial (NCT00136084). 163 patients with both gene expression and outcome data were included and clinical endpoints were defined as: i) Minimal residual disease after induction 1 (MRD1): positive MRD1 if ≥ 1 leukemic cell per 1,000 mononuclear bone marrow cells (≥ 0.1%) determined using flow cytometry; ii) Event Free Survival (EFS) defined as the time from study enrollment to induction failure, relapse, secondary malignancy, death, or study withdrawal for any reason, with event-free patients censored on last follow-up; iii) Overall Survival (OS) defined as the time from study enrollment to death, with living patients censored on the date of last follow-up. In a step-wise approach, we first deconvoluted the bulk gene-expression data through CIBERSORTx using the LM22 as a reference panel to derive the percent cell fraction of 22 distinct hematopoietic cell lineages across all patients. We evaluated the inter-patient differences in cell fraction against clinical endpoints defined above and observed significant association between survival endpoint and cell type fractions of mast cells resting and NK cells activated. We further utilized a machine-learning approach and performed 1,000 iterations of 10-fold cross-validation of Cox proportional hazard regression modeling with LASSO penalty using OS as a clinical endpoint. Cell-type fractions passing at least in 800 of 1000 models (9 hematopoietic cell types Figure 1a) were selected to create a CiberScore equation by multiplying cell fraction with average coefficient obtained from 1000 models. Patients were further classified into two groups: Low-CiberScore (n=122, Quartile 1-3) and High-CiberScore (n=41, Quartile 4). No significant difference by CiberScore groups was observed for patient characteristics such as age, initial risk group status, FLT3-ITD or cytogenetics. With respect to clinical outcome patients within High-Ciberscore group had greater MRD1 positivity (p=0.010) and poor EFS (HR 2.96, p<0.0001) and OS (HR 4.78, <0.0001) as compared to patients within Low-Ciberscore group. Kaplan-Meier survival curves for OS and EFS by CiberScore are shown in Figure 1b. In multivariable analysis, CiberScore remained significantly associated with EFS and OS after adjusting for induction 1-MRD1 status, risk-group, FLT3-status, WBC-count at diagnosis and age, implying it to be an independent prognostic factor. In conclusion, in this study we leveraged CIBERSORTx to generate cell type fraction using bulk gene-expression data from pediatric AML patients and further utilized machine learning algorithms to develop a prognostic model based on 9 leukocyte lineages. CiberScore remained significant in independent RNA seq cohorts from NCI TARGET pediatric AML repository. Ongoing studies are focused on further validation of this signature in other cohorts as well as comparing the similarities and differences in adult vs. pediatric AML with respect to the cell fraction differences of prognostic relevance. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have