Background
Acute myeloid leukemia (AML) accounts for 80% of acute leukemia in adults. AML is characterized by unlimited clonal proliferation and accumulation of myeloid progenitors [
1]. The 5-year survival for AML patients is no more than 50%, which is less than 20% in elderly AML patients [
2]. To estimate risk and survival of AML patients, quite a few prediction models have been developed. European LeukemiaNet (ELN) 2017 risk stratification is the most commonly used risk model, which stratified AML patients based on recurrent cytogenetics and molecular mutations abnormalities [
3]. A comprehensive evaluation of genetic variables is crucial for risk stratification and will guide treatment decisions. Other traditional prognostic factors include age, WBC count, LDH level etc. [
4]. Whereas the biomarkers with prognostic value are still being explored to improve the risk model for AML.
The genetic alterations of transcription factors (TFs) occur frequently in AML, which exert important effects in pathogenetic process and associate with prognosis [
5‐
8]. HOXA (homeobox protein HOX cluster A) family genes, encoding highly conserve TFs with DNA binding homeobox binding motifs, play a crucial role in adult hematopoiesis [
9‐
11], while aberrant overexpression of HOXA promotes oncogenesis [
12]. The previous studies indicated that genetic alterations in AML resulted in HOXA overexpression, such as KMT2A rearrangement [
13], FLT-ITD [
13], MLL gene abnormality [
14‐
17]. HOXA9, HOXA7, HOXA11 were associated with adverse prognosis in AML [
18,
19]. Overexpression of HOXA10 was reported to interrupt hematopoietic process [
20] and lead to myeloid leukemogenesis in mice model [
21,
22]. While the prognostic significance of HOXA10 for AML has been rarely explored.
Nowadays, multidimensional information has been accumulated for AML other than gene mutations and karyotypes, including gene expression, non-coding RNA profile, gene methylation profile, copy number variation, etc. In the present study, we explored regulatory genetic or epigenetic variables of HOXA10, such as methylation of CpG, copy number variation (CNV), lncRNA, microRNA and TF, which affect the gene expression. The interaction of lncRNA and miRNA sponges form competing endogenous RNA (ceRNA) network regulating gene expression and pathways. Then Lasso-Cox model was used to fit AML survival to prediction model, including clinical features and HOXA10-associated genetic/epigenetic variables. Our work offered evidence for using HOXA10 as a prognostic marker for AML, and establishment of novel risk model to predict AML survival.
Discussion
HOXA family genes are well known as the crucial transcription factors in pathogenesis and development of AML. HOXA10 belongs to the HOXA gene superfamily, dysregulation of which has been observed in several solid tumors [
36‐
38]. HOXA10 plays a role in myeloid differentiation, leukemogenesis and chemoresistance in AML [
39‐
41]. In the present study, we investigated the prognostic significance of HOXA10 in AML, which has been rarely described and reported previously. The results of expression analysis demonstrated that AML patients have aberrant HOXA10 expression in comparison with controls. High HOXA10 expression level is significantly associated with worse OS and DFS of AML, based on Kaplan-Meier curve and logrank test. Therefore, HOXA10 may serve as a prognostic marker for AML patients.
To explore the underlying enriched pathways of DEG, the ORA was performed. In the GO part of our ORA, we uncovered that the biological processes of DEG were enriched in cell-cell signaling and cell communication, and the cell components were enriched in plasma cell membrane and extracellular matrix. Correspondingly, the molecular functions were enriched in signaling receptor activity and extracellular matrix structural constituent. To explore the detail cell-cell signaling pathway that the DEGs were involved, the following KEGG and Reactome analysis revealed that the DEGs were enriched in PI3K-Akt signaling, Ras signaling pathway, signaling by PDGF and signaling by ROBO receptor, etc. In addition to activating mutation of NRAS/KRAS in 15–25% AML, the mutations of RAS-regulating genes (NF1 and PTPN11) and RAS-signaling receptor (FLT3 and KIT) are also harbored in AML frequently [
42‐
46]. Hyperactive mutations of NF1 and PTPN11 gene are associated with inferior survival in pediatric and elderly AML [
47,
48]. FLT3-ITD mutation is also well-known as a biomarker for worse prognosis in non-APL AML patients [
49,
50]. Although NRAS and KRAS genes was not significantly differentially expressed, RAS signaling genes, including PTPN11, FLT3 and KIT, were upregulated significantly in HOXA10-high AML patients. The dysregulation of RAS signaling pathway may lead to unfavorable impact on clinical outcome of HOXA10-high group. PDGF signaling plays a proto-oncogenic role in diverse cancer cells. Imatinib turned out to block PDGF receptor at low dose, exerting a pharmacological effect for BCR-ABL positive CML and FIP1L1-PDGFRA mutated eosinophilic leukemia patients [
51,
52], suggesting PDGF signaling as an activated effector in hematological malignancies. Gołos A et al. reported increased ROBO1/2 in AML patients in comparison with normal control, and high ROBO3 expression is associated with cytogenetic high risk and poor prognosis [
53]. However, the aberrantly expression pattern of signaling pathway by PDGF/ROBO has not been fully elucidated in AML, which were rarely studied in the prognosis of AML. The enriched pathways obtained by ORA help us to identify expression signature in HOXA10-high group, screen useful biomarkers and provided novel insights into molecular investigation on AML.
Differential expression analysis and GSEA revealed that PI3K-Akt signaling pathway was suppressed associated with HOXA10 overexpression. PI3K-Akt signaling pathway is frequently activated in AML, constitutive activation of PI3K and Akt were found in 50% de novo AML patients [
54,
55]. The PI3K-Akt signaling controls leukemic blast cells proliferation and clonogenicity [
56,
57]. AML patients with constitutive PI3K-Akt activation have better OS and relapse-free survival [
58]. The unfavorable survival profile in HOXA10-high group may attributed to aberrant downregulation of PI3K-Akt signaling.
The results of GSEA demonstrated that ribosome, oxidative phosphorylation, and lysosome pathways were activated in HOXA10-high group. Ribosome pathway is a vital cellular process, and the rate-limiting step of which is the initiation of translation in ribosomes. One of major control factors in the ribosome activity, is EIF2 (eukaryotic initiation factor 2), which is regulated by phosphorylation of α subunit (EIF2α) under diverse stress. Four kinds of EIF2α kinases (EIF2AK1/2/3/4) can affect the activity of EIF2α by phosphorylation of Ser51 [
59,
60]. Notably, the expression EIFAK2 and EIFK3 were significantly increased in HOXA10-high group by our results. EIF2AK2 (also named as double-strand RNA-dependent kinase, PKR) responses to various types of stress, including DNA damage, mitochondrial stress, viral infection, growth factor deprivation, cytokines, Toll-like receptor activation and cytotoxic drugs [
61‐
65]. Also EIF2AK2 is the only EIF2α kinase exists in both the cytoplasm and nucleolus, while other 3 kinases present only in the cytoplasm [
66]. Cheng X et al. reported that high expression of EIF2AK2 was associated with worse prognosis in AML, and it reduced DNA damage response by inhibiting ataxia-telangiectasia mutated (ATM) activation, leading to accretion of leukemia in mice model [
67]. EIF2AK3 (also named as PKR-like endoplasmic reticulum kinase, PERK) is reported to promoted leukemia progress by stimulating the dissemination of leukemia cells in vivo [
68]. So, the increased EIF2AK2/3 expression and activated ribosome pathway contributed to the worse outcomes in HOXA10-high patients. The maintenance of leukemia stem cells depends on BCL2 mediated oxidative respiration, instead of glycolysis as in normal hematopoietic cells [
69]. The metformin, targeting oxidative phosphorylation (OXPHOS), induces apoptosis of human leukemia cells in an AMPK-independent way [
70]. Cytarabine resistant leukemia cells are characterized by activated OXPHOS, with the high level of reactive oxygen species. Additionally, the resistance can be reversed by agents inducing low OXPHOS status [
71]. The activation of OXPHOS in HOXA10-high patients may promoted leukemia cell maintenance and chemo-resistance, leading to inferior survival. The biological function of lysosome pathway in AML has not been fully elucidated. While considering lysosome pathway involves in autophagy, which plays a role in leukemic transformation of normal hematopoietic stem cells and chemotherapy response [
72], it may be still valuable to explore in this area. The Kaplan-Meier curves confirmed that HOXA10 expression is associated with AML survival, while it didn’t predict OS or DFS significantly in multivariable Cox hazards analysis including other clinical and genetic variables (data not shown), suggesting HOXA10 is not an independent prognostic factor. Since copy number variation (CNV), mutations (not reported in TCGA database), CpG methylation status and upstream TFs/microRNA/lncRNA are the most common gene expression regulators, we investigated whether these factors were associated with HOXA10 expression and established Lasso-Cox model fitting OS to reveal novel prognostic markers. And finally, Lasso-Cox analysis included clinical features, CNV of HOXA10, methylation and expression status of HOXA10, HOXA10 associated TFs/DEmiRs/DElncRs. Due to the obvious correlation and dependency between each variable, the traditional multivariable Cox analysis is of limited utility, where the Lasso-Cox methods showed its superiority. A few prediction models for AML overall survival have been reported (Table
3), including Huang R et al. [
73], Mihyang Ha et al. [
74], Clinseq-G model [
75], ELN2017 recommendation [
75], Zejuan Li et al. [
75]. The AUC equals to the probability, which a diagnostic classifier will rank a randomly chosen positive instance higher than a negative one, and the highest AUC is established as best practice [
76]. Our prediction model has superior AUC than above models, possibly attributing to integrated multidimensional information integrated.
Table 3
Diagnostic utility of prognostic models for AML by time-dependent ROC
Huang R et al.(75) | 0.666 | 0.713 | 0.707 |
Mihyang Ha et al.(76) | – | – | 0.613 |
Clinseq-G model(77) | – | 0.73 | – |
ELN2017 | – | 0.65 | – |
Zejuan Li et al.(77) | – | 0.7 | – |
ARS model | 0.808 | 0.839 | 0.786 |
In the ARS model, novel markers are included. IKZF2 is recently found to drive leukemia stem cell renewal and inhibit myeloid differentiation, by regulating HOXA9 and CEBP [
77]. The aberrant variation of IKZF2 was also reported in ovarian cancer cell lines [
78], adult T cell leukemia [
79] and gastric cancer [
80], suggesting IKZF2 as a potential oncogenes. The relation of AML and LINC00649 has not been explored. While notably, the relation of HOXA family genes and LINC00649 is very close. Based on Gepia 2.0 (
http://gepia2.cancer-pku.cn/), using AML RNAseq dataset of TCGA and Pearson method, LINC00649 is negatively correlated with HOXA1 (
p = 0.015), HOXA2 (
p = 0.0056), HOXA3 (
p = 0.0045), HOXA4 (
p = 0.011), HOXA5 (
p = 0.0019), HOXA6 (
p = 0.0078), HOXA7 (
p = 0.00089), HOXA9 (
p = 0.002) and HOXA10 (
p = 0.0022) significantly. Intriguingly, expression level of all above HOXA genes are significantly associated with AML survival [
8], which indicated that LINC00649 targeted genes may exert pathogenetic effect in AML. The investigation of LINC00839, FENDRR and has-miR-424-5p in AML has not been conducted.
The limitation of this model is also obvious, which is lacking prospective large-scale studies for validation. And further experiment is needed to validate the interaction of TFs/microRNAs/lncRNAs with HOXA10.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.