Elevated levels of TNF and its targets characterize better-risk older acute myeloid leukemia patients

Résumé fourni par la source

Acute myeloid leukemia (AML) is a heterogeneous cancer. The incidence rate of AML increases while the survival rates decrease with age (median age of diagnosis 69 years) [ 1 ]. Intensive combination chemotherapy remains an upfront treatment option [ 2 ]; however, some AML patients over the age of 60 (aged AML; aAML) are unfit for intensive treatment and/or fail to achieve remission [ 3 ]. A key step in treatment selection is risk classification. We reported an aAML-specific risk classifier from patients enrolled in ECOG-ACRIN clinical trial E3999 [ 4 ]- that identified two patient risk-groups (low-risk characterized by Nucleophosmin 1 ( NPM1) mutations (G1) or female sex (G2), and high-risk characterized by other molecular abnormalities) [ 5 ]. We hypothesized that molecular and cellular characteristics of these two risk groups define biological differences important for therapeutic choices. While transcriptional profiling has informed leukemogenesis mechanisms and treatment responses [ 6 ], there remains an unmet need to identify biological processes that associate with gene expression patterns in aAML. A recent report suggested that a high inflammation gene score (iScore) is associated with poor risk features and worse clinical outcomes in AML patients [ 7 ]. We wondered if this signature would segregate our risk groups. To address this question, we generated RNA-sequencing from disease-cell-enriched diagnostic specimens isolated from aAML patients ( n = 154: low-risk ( n = 51) and high-risk ( n = 103); Supplementary Table 1a ) enrolled in E3999. When applied, the 38 gene-based iScore algorithm could identify survival groups in our study cohort (iScores: −4.05 and −3.48 for low- and high-risk groups respectively; p < 0.05); however, they were not concordant with our risk groups (Supplementary Fig. 1a ). Furthermore, when applied to differentially expressed genes (DEGs) between aAML patient specimens and normal controls ( n = 13,703), and between our defined risk groups ( n = 2314; Supplementary Table 2 ), the iScore algorithm did not identify any genes that associated with clinical outcomes (Supplementary Fig. 1b, c ). However, functional enrichment analysis of the DEGs between the risk groups (Supplementary Table 2 ) identified significant enrichment for genes in inflammatory pathways in the low-risk group (35% of all gene sets; Fig. 1A ; Supplementary Table 3 ), which did not significantly overlap with the iScore genes (Supplementary Fig. 1d ). We confirmed the enrichment for inflammatory pathways using single sample gene set enrichment analysis (ssGSEA; Wilcox test p < 0.05; Supplementary Fig. 2A, B ; Supplementary Table 4 ). We next aimed to identify regulatory factors responsible for the observed enrichment results between the risk groups. 107 upstream regulators were predicted to be either activators or inhibitors for the observed gene expression changes (B-H corrected p -value < 0.05 and absolute Z-score ≥2; Fig. 1B , Supplementary Table 5 ). Tumor necrosis factor (TNF) was predicted to be one of the top activators (Fig. 1B ), and accordingly, its transcript levels were significantly higher in the low-risk aAMLs (Fig. 1C ). Physiological concentrations of cytokines were estimated in serum obtained from a subset of the patients (low-risk: n = 16 and high-risk: n = 26). TNFα was significantly higher in the low-risk group (Fig. 1D ; Supplementary Table 6 ). This finding could not be explained by differences in T-cell population diversity between the risk groups (Supplementary Fig. 3 ). Collectively, our results suggested a pro-inflammatory state in the low-risk aAML patient group. Fig. 1: Low-risk aAMLs are characterized by a pro-inflammatory state. Full size image A Dot plot representing the gene set enrichment analysis results for all the MSigDB hallmark genesets. X-axis represents the normalized enrichment score (NES) and Y-axis represents the negative logarithm of the FDR-corrected p -value. Dots colored in black are significantly enriched pathways (FDR corrected p -value < 0.05 and |NES|> 1). Significantly enriched inflammation-associated pathways are colored red. Gene sets that are not-significantly associated at the chosen thresholds are colored gray. B Dot plot representing results from upstream analysis of DEGs identified between low- and high-risk aAML groups. X-axis represents the activation Z-score used to infer the activation states of predicted transcriptional regulators. Y-axis represents the negative logarithm of BH-corrected p -values. Upstream regulators with an absolute activation z-score greater than 2 and corrected p -value less than 0.05 are highlighted (Blue: Activators, Orange: Inhibitors). Top 5 regulators with highest Z-score are labeled. C Transcript levels of the gene encoding tumor-necrosis factor (“TNF”). Y-axis indicates log10-converted normalized (‘vst’) gene expression values. Statistical significance was estimated using Wilcoxon-rank-sum test. D Serum levels of TNF-α in low and high-risk aAMLs estimated using a Luminex assay. The statistical significance between the levels was estimated using Wilcoxon-rank-sum test. Next, to determine what cell types and states associated with the pro-inflammatory signature observed, we first performed GSEA based on a ranked list of genes obtained from comparing low- versus high-risk patients with the high-risk group as baseline (Supplementary Table 2 ). The upregulated genes in the low-risk aAMLs were enriched for gene signatures identified in NPM1 mutated (NPM1c+) AMLs (Fig. 2A ), and more differentiated hematopoietic cells (Fig. 2B, C ). These patterns were not specific to the NPM1c+ aAMLs (G1) since pre-ranked GSEA analysis on expression differences between G2 and high-risk aAMLs resulted in similar results (Supplementary Fig. 4A–C ; Supplementary Table 7 ). The low-risk aAMLs also had enrichment of AML gene signatures associated with tumor myeloid and monocytic cells as well as a previously known committed subtype of NPM1c+ AMLs (Supplementary Fig. 5A–D ; Supplementary Table 8 ). Cell surface protein and transcript levels assessed by flow cytometry and relative levels in RNA-sequencing data were concordant with these findings (Wilcox test p -value < 0.05; Supplementary Fig. 6A–E , Supplementary Table 9 ). Similar results were obtained when we estimated the proportions of distinct cell populations based on custom gene sets using deconvolution analysis (Supplementary Fig. 6F ; Wilcox test p < 0.05). Importantly, we found a significant overlap between the predicted TNF targets and genes previously identified to be expressed in tumor-monocyte-like AML cells [ 7 ] (Supplementary Fig. 7 ). Fig. 2: Gene expression analysis identifies divergent gene signatures in low- and high-risk aAMLs. Full size image A – C Geneset enrichment analysis of the ranked gene expression (high expression in low-risk to low-expression in low-risk) against a NPM1c+ AML signature ( A ) and mature hematopoietic cell signatures ( B , C ). D UMAP plot depicting single-cell RNA-seq data of 11,072 cells from 2 low-risk samples and 3 high-risk samples. Each point represents a single cell, color-coded by the risk group. E UMAP plot depicting single-cell RNA-seq data of 11,072 cells from 2 low-risk samples and 3 high-risk samples. Each point represents a single cell, color-coded by cell type annotation. F Heatmap of predicted upregulated TNF targets’ expression level (identified from upstream regulator analysis; Fig. 1B ) across cells annotated as “CD14 monocytes” in both low and high-risk aAML samples in the single cell RNA-sequencing data. Columns represent cells annotated as “CD14 Monocytes”, and rows display relative expression (Z-score) of TNF targets. Gene labels in red are those identified in tumor-monocyte like cells from VanGalen et al. To validate these findings, we performed Cellular Indexing of transcriptome and epitopes by sequencing (CITE-Seq) on 5 aAML samples (2 low-risk and 3 high-risk). Unsup

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.