Higher HOPX expression is associated with distinct clinical and biological features and predicts poor prognosis in de novo acute myeloid leukemia.

Homeodomain-only protein homeobox (HOPX) is the smallest homeodomain protein. It was regarded as a stem cell marker in several non-hematopoietic systems. While the prototypic homeobox genes such as the HOX family have been well characterized in acute myeloid leukemia (AML), the clinical and biological implications of HOPX in the disease remain unknown. Thus we analyzed HOPX and global gene expression patterns in 347 newly diagnosed de novo AML patients in our institute. We found that higher HOPX expression was closely associated with older age, higher platelet counts, lower white blood cell counts, lower lactate dehydrogenase levels, and mutations in RUNX1, IDH2, ASXL1, and DNMT3A, but negatively associated with acute promyelocytic leukemia, favorable karyotypes, CEBPA double mutations and NPM1 mutation. Patients with higher HOPX expression had a lower complete remission rate and shorter survival. The finding was validated in two independent cohorts. Multivariate analysis revealed that higher HOPX expression was an independent unfavorable prognostic factor irrespective of other known prognostic parameters and gene signatures derived from multiple cohorts. Gene set enrichment analysis showed higher HOPX expression was associated with both hematopoietic and leukemia stem cell signatures. While HOPX and HOX family genes showed concordant expression patterns in normal hematopoietic stem/progenitor cells, their expression patterns and associated clinical and biological features were distinctive in AML settings, demonstrating HOPX to be a unique homeobox gene. Therefore, HOPX is a distinctive homeobox gene with characteristic clinical and biological implications and its expression is a powerful predictor of prognosis in AML patients. Copyright© Ferrata Storti Foundation.

Introduction

HOPX was first identified in the expression sequence tag database for transcripts encoding proteins related to the development of the heart in mice and zebrafish.[1,2] Human HOPX, located in chromosome 4q12, has five isoforms. The predominant one encodes a short protein of 73 amino acids with a molecular weight of 12 kd. This is by far the smallest homeodomain protein, with conservation of the homeodomain of 60 amino acids. But the difference between this and other homeodomain proteins is that the HOPX protein does not bind DNA directly. Rather, it exerts its transcriptional inhibition through sequestration of serum responsive factor by physical interaction and by recruitment of histone deacetylase.[3]

Recently, the HOPX gene has been regarded as a stem cell marker in intestine, hair follicles, and pulmonary alveolar cells.[4-7] Some studies suggested a role for HOPX in tumorigenesis with clinical implications. HOPX has been suggested to be a tumor suppressor gene in lung, colon, esophagus, pancreas, uterine and stomach cancers.[8-13] Most of the studies showed silencing of HOPX through hypermethylation of the promoter as a mechanism of its downregulation in cancer cells.[9-13] However, the mechanisms for the tumor suppression remain largely unknown.

The HOX family and HOPX belong to homeobox genes. However, while HOX family genes have been well studied in acute myeloid leukemia (AML),[14] the clinical and biological significance of HOPX in human hematopoiesis remains undefined. We are interested in exploring the roles of HOPX in AML patients, as well as comparing HOX and HOPX in the pathophysiology of malignant hematopoiesis. In this study, we compared clinical and biological characteristics between de novo AML patients with higher and lower HOPX expression and found that higher HOPX expression was strictly correlated with unfavorable prognosis of AML patients in ours and the other two independent cohorts. Multivariate analysis revealed higher HOPX expression as an independent unfavorable prognostic factor, and independent of several published gene signatures derived from multiple cohorts. Using bioinformatics approaches, we found that HOPX expression was closely associated with known hematopoietic stem cell (HSC) signatures. While HOPX and HOX family genes were highly expressed in normal hematopoietic stem/progenitor cells (HSPC), the expression patterns and associated clinical and biological features between these two classes of homeobox genes differed dramatically in AML settings. Taken together, our study suggests that HOPX has a significant impact on various clinical and biological aspects of AML, and that there is a distinction between HOX family genes and HOPX in the AML setting.

Methods

Patients

A total of 347 adult patients diagnosed with de novo AML according to the 2008 World Health Organization classification in the National Taiwan University Hospital (NTUH) who had cryopreserved bone marrow (BM) cells and complete clinical and laboratory data available for analysis were retrospectively enrolled. Among them, 227 patients received standard induction chemotherapy. Non-M3 (acute promyelocytic leukemia, APL) patients received idarubicin 12 mg/m2 per day for 2–3 days and cytarabine 100 mg/m2 per day for 5–7 days, as described previously.[15] APL patients received concurrent all-trans retinoic acid and idarubicin. The remaining 120 patients received palliative therapy with supportive care or low-dose chemotherapy due to underlying comorbidity or in accordance with patient decision. We also prospectively enrolled another cohort of 56 newly diagnosed adult de novo AML patients with adequate BM samples for more detailed studies of the HOPX gene, including expression pattern of HOPX isoforms in AML. The study was approved by the Research Ethics Committee of the NTUH.

Cytogenetic and mutation analysis

Chromosomal abnormalities[16] and mutation analyses were performed as previously described.[15-20]

Gene expression microarray datasets and data analysis

We profiled global gene expression of BM mononuclear cells from the 347 patients (NTUH dataset) using Illumina HumanHT-12 v.4 Expression BeadChip (Illumina, San Diego, CA, USA) (GSE68469 and GSE71014).[21-23] Two large microarray datasets of AML with overall survival (OS) data, including The Cancer Genome Atlas (TCGA) dataset (n=186)[24] and GSE12417 [all with cytogenetically normal (CN) AML; n=162],[25] were utilized to validate the prognostic significance of HOPX. We used TCGA-normalized level-2 intensity and GSE12417 GPL96 data (profiled with Affymetrix Human Genome U133A Array), normalized as described by Metzeler et al.[25] Gene expression profiles GSE12662 (n=91),[26] GSE24006 (n=54),[27] and GSE24759 (n=211)[28] were also included to investigate the gene expression patterns in normal hematopoiesis.

Analysis of gene expression in next-generation sequencing datasets

To investigate the absolute levels of gene expression in AML, we analyzed expression data of 179 AML samples profiled with Illumina Genome Analyzer RNA Sequencing in TCGA dataset.[24] Reads per kilobase per million mapped reads (RPKM) levels of gene expression were extracted from TCGA database.[24]

Gene signature analysis

The association of HOPX gene with stem cell characteristics was analyzed by the Gene Set Enrichment Analysis (GSEA; a Java application that can be down-loaded at )[29] and as detailed in the Online Supplementary Appendix. In order to examine whether genes are involved in HSC quiescence, we employed another gene set enrichment scoring method that averages z-values of all involved genes.[30]

Methylation microarray datasets and analysis

DNA methylation data from Illumina Infinium HumanMethylation450 BeadChips of AML (n=194) were downloaded from the TCGA database.[24] We transformed methylation beta-values to normally distributed M-values for further analysis.[31]

Expression of HOPX isoforms

Human HOPX has five isoforms including HOPXa (NM_032495), HOPXb (three variants including NM_139212, NM_139211 and NM_001145459; abbreviated hereafter as b1, b2, and b3, respectively), and HOPXc (NM_001145460) (UCSC genomic database; ) (Online Supplementary Figure S1). Analysis of HOPX isoform expression was performed by quantitative real time-polymerase chain reaction as detailed in the Online Supplementary Appendix, Online Supplementary Table S1 and Online Supplementary Figure S1.

Bisulfite treatment and methylation analysis of HOPX

We interrogated the methylation status of the CpG islands of HOPX-b2 isoform NM_139211 from −15 to +109 bp region around the transcription start site (TSS).[10] Methods are described in the Online Supplementary Appendix.

Statistical analysis

Statistical analysis was carried out as described previously;[21-23] a brief description is available in the Online Supplementary Appendix.

Results

Correlation of HOPX expression with clinical features

The 347 AML patients were divided into two groups based on the HOPX expression levels above (higher expression group) or below (lower expression group) the median level of HOPX expression on the arrays. Higher HOPX expression was associated with older age (P=0.023), higher platelet counts (P=0.008), lower white blood cell (WBC) counts (P=0.011), and lower lactate dehydrogenase (LDH) levels (P<0.001) at diagnosis (Table 1). Patients with M1 and M6 according to the French-American-British (FAB) classification more frequently had higher HOPX expression (P=0.021 and P=0.032, respectively), while those with M3 and M5 had significantly lower levels of HOPX expression (P<0.001 and P<0.006, respectively). The comparison of clinical features between higher and lower HOPX expression groups in those receiving standard chemotherapy (n=227) is shown in the Online Supplementary Table S2. The association of higher HOPX expression with higher platelet counts, lower LDH levels, and FAB subtypes remained the same in this group of patients as that of the total cohort.

Table 1.

Comparison of clinical manifestations between acute myeloid leukemia patients with higher and lower HOPX expression.

Correlation of HOPX expression with cytogenetics and molecular alterations

Chromosome data were available in 325 patients at diagnosis (Online Supplementary Table S3). Higher HOPX expression was negatively associated with favorable karyotypes, including t(8;21) and t(15;17) (both P<0.001). We also analyzed the mutation status of 16 genes and found that the patients with higher HOPX expression had significantly higher incidences of mutations in RUNX1 (P<0.001), IDH2 (P=0.001), ASXL1 (P=0.017), and DNMT3A (P=0.031), but less frequently had CEBPA double mutations (P=0.001) and NPM1 mutation (P=0.040) (Table 2).

Table 2.

Association of HOPX expression levels with other genetic alterations.

Higher HOPX expression predicts poor clinical outcome in de novo AML patients

Among the 227 patients who received standard chemotherapy, those with higher HOPX expression had a lower complete remission (CR) rate (58.3% vs. 85.5%; P<0.001) (Table 1), shorter OS (median 23.7 months vs. 116.8 months; log-rank P<0.001) and disease-free survival (DFS) (median 5.9 months vs. not reached; log-rank P<0.001) than those with lower HOPX expression after a median follow up of 57.0 months (Figure 1A and B). Univariate Cox proportional hazards analysis confirmed the prognostic value of HOPX expression as a continuous variable in predicting patients’ OS [Hazard Ratio (HR): 1.44; 95%CI: 1.21–1.71; P<0.001] and DFS (HR: 1.55; 95%CI: 1.31–1.82; P<0.001). The prognostic significance of HOPX expression could be validated in another two independent cohorts: TCGA[24] and GSE12417[25] (Figure 1C and D). The unfavorable prognostic effects of higher HOPX expression were also seen in the subgroup of patients with AML other than APL (median OS 23.7 vs. 108.1 months; P<0.001) and those with a normal karyotype (median OS 39.1 vs. 116.8 months; P=0.004) (Figure 1E and F). The results could also be validated by the TCGA cohort (Online Supplementary Figure S2A and B).

Figure 1.

HOPX expression levels and acute myleoid leukemia (AML) patients’ survival. (A and B) In NTUH cohort, the overall survival (OS) and disease-free survival (DFS) of AML patients with higher HOPX expression are significantly shorter than those with lower expression: median OS 23.7 versus 116.8 months, P<0.001; median DFS 5.9 months versus not reached (NR); P<0.001. (C and D) The observation is validated by TCGA (median OS 11.2 months vs. 24.4 months; P<0.001) and GSE12417 (median OS 7.8 months vs. 33.3 months; P<0.001) cohorts. (E and F) When we restrict the analysis in non-acute promyelocytic leukemia (APL) patients or patients with a normal karyotype in NTUH cohort, HOPX levels still significantly correlate with OS: median OS 23.7 versus 108.1 months (P=0.0003) and median OS 39.1 versus 116.8 months (P=0.004), respectively. Green line: higher HOPX expression group; blue line: lower HOPX expression group.

By univariate analysis, HOPX expression levels and several parameters exhibited a significant impact on OS (Online Supplementary Table S4). When we combined all these prognostic factors together in a multivariate analysis, higher expression of HOPX remained a poor prognostic factor for OS (P=0.005) (Table 3), independent of age, WBC counts, karyotypes, mutation statuses of FLT3, CEBPA, MLL, TP53, WT1, and RUNX1 and expression levels of HOXA9.

Table 3.

Multivariate analysis (Cox regression) on overall survival.*

Further analysis showed much higher HOPX expression in those patients who failed to achieve CR than those who obtained a CR (by array signal intensity; P<0.001) (Figure 2A) suggesting a tight association of higher HOPX expression and drug resistance. Furthermore, HOPX expression was lower in APL, which consisted mainly of leukemic cells that are blocked at the differentiation stage of promyelocytes, indicating a possible relationship between HOPX expression and maturation stages of AML cells (Figure 2B).

Figure 2.

HOPX expression levels, its correlation with treatment response, and its role as a stem cell marker by gene set enrichment analysis. (A) There was a significant difference in HOPX levels between patients with (n=166) and without (n=61) complete remission (CR) after induction chemotherapy. (B) HOPX expression is also much higher in non-APL patients. (C) Real-time PCR of HOPX in the cohort of 56 acute myeloid leukemia (AML) patients prospectively recruited showing predominant expression of isoform HOPXb2 (NM_139211) (left, isoforms a+b1+c; middle, b2; right, b3). (D, E, and F) GSEA plots of curated HSC and LSC signatures in the NTUH dataset.[35,36] Red-to-indigo bars denote the genome-wide gene list ranked based on their P-values (t-test) between samples with high (top quartile) and low (bottom quartile) expression of HOPX. Significant positive GSEA enrichment scores indicate that HOPX expression is positively associated with HSC and LSC signatures.

Comparisons between HOPX expression and published prognostic gene signatures in predicting prognosis

Several gene expression-based prognostic predictors have been developed from various study designs in AML. To compare the performance of prognostic prediction of HOPX expression with those published predictors, we performed pairwise multivariate Cox analysis between HOPX expression and each of the 3-gene, 7-gene, 11-gene, and 24-gene predictors in three datasets.[21,32-34] Remarkably, the HOPX expression remained independent (with Cox multivariate analysis P<0.05) in most of the comparison settings (11 of 12 comparisons) (Table 4). Our data suggest HOPX to be a simple and powerful alternative for prognostication in AML.

Table 4.

Comparisons of HOPX to published prognostic gene signatures.

The expression pattern and promoter methylation of HOPX isoforms in AML patients

The pattern of expression of the 5 isoforms of HOPX and CpG methylation status in primary AML are still unknown. Because of the low levels of expression and the impossibility of separating isoforms a, b1, and c, we merged these three together for quantification. We quantified the expression levels of these isoforms in prospectively recruited AML patients’ marrow by real time-PCR and found that HOPXb2 (NM_139211) was the predominant isoform in human AML cells; the other four variants were markedly under-represented (Figure 2C). In addition, quantification of the CpG methylation in HOPXb2 promoter regions revealed low levels of methylation in most AML patients (Online Supplementary Figure S3). This finding seemed to differ from that in some studies of solid cancers in which hypermethyation of HOPX in this region caused gene silencing and was associated with poor prognosis.[9-13] Further studies are needed to confirm this hypothesis.

Correlation of HOPX expression with stem cell signatures

We curated xenotransplantation-derived HSC and leukemia stem cell (LSC) gene signatures and a recently up-dated LSC signature from previous reports[35,36] and employed the gene set enrichment analysis (GSEA) method[29] to analyze their associations with HOPX expression. GSEA tests the enrichment of each gene signature in the list of global genes ranked by HOPX-associated differential expression. Higher HOPX expression was associated with upregulation of HSC and LSC genes in the NTUH dataset (enrichment scores, 0.79, 0.61, and 0.89; P<0.0005, P=0.006, and P<0.0005, respectively) (Figure 2D–F). Concordant significant enrichments were identified in TCGA and GSE12417 AML datasets (all P-values ≤0.015) (Online Supplementary Figure S4). Seventeen and five genes appeared as leading-edge genes (ie. a subset of core-enrichment genes) of the HSC and LSC signatures (Figure 2D and E), respectively, in all the three cohorts (Online Supplementary Table S5). Interestingly, an ATP-binding-cassette (ABC) transporter gene, ABCB1, was a common leading-edge gene of HSC signature (Online Supplementary Table S5). ABC transporter genes were reported to be associated with chemoresistance in AML, with higher ABCB1, ABCG1, ABCG2 expression levels being independently poor prognostic factors.[37] Expression of these three ABC genes was significantly higher in samples with higher HOPX expression (all P-values <0.001) (Online Supplementary Table S6), but the mechanistic link between ABC and HOPX expression still has to be explored in further studies.

Expression patterns of HOPX and HOX genes in normal hematopoietic cells

HOPX and HOX family genes all encode homeodomain proteins and HOX genes are well-known HSC markers.[38-40] To further delineate the similarities and the distinctions between HOPX and the HOX gene family, we first analyzed their expression patterns in normal hematopoietic cells using arrays derived from public data. We curated three public gene expression datasets derived from normal hematopoietic cells[26-28] and we chose 12 HOX genes with at least moderate expression levels (RPKM> 5 according to TCGA RNA seq data) for further analysis.[24] In GSE24006[27] and GSE12662,[26] HOPX and HOX genes were generally expressed in a concordant manner (mean correlation coefficient 0.37 and 0.35, respectively) (left panels, Online Supplementary Figure S5A and B). We further analyzed a dataset of 9 distinct normal hematopoietic cell populations (GSE24759) (Online Supplementary Figure S6).[28] HOPX and HOX family genes were all highly expressed in normal CD34+ hematopoietic cells (average z-values = 0.82 and 0.75; 1-sample t-test both P<0.001) (Online Supplementary Figure S6). The concordant expression patterns between HOPX and HOX family suggest their shared roles in normal hematopoiesis.

Expression patterns of HOPX and HOX family genes in AML

We investigated the absolute gene expression levels of HOPX and the HOX family in AML from the TCGA RNA sequencing dataset. Among them, HOPX was the second highest expressed gene (average RPKM = 25.6 in TCGA RNA sequencing dataset; n=179), after the most abundant gene HOXA9 (RPKM = 43.3). We then compared the expression patterns between HOPX and HOX family genes in AML cells. The concordance of expression patterns shown in normal hematopoietic cells were no longer present in AML cells (correlation in GSE24006 and GSE12662 −0.31 and 0.07, respectively) (Online Supplementary Figure S5A and B). We sought to investigate the similarities/distinctions between HOPX and HOX family genes by clustering of AML patients in our dataset (NTUH) according to their expression levels. Because of the unequal numbers of genes between HOPX and HOX family (1 vs. 12), we performed a 2-step hierarchical clustering to balance the potential bias in unsupervised clustering. Briefly, patients were first clustered only by the 12 HOX family genes. Subsequently, each cluster was subject to the second round of clustering with inclusion of HOPX. As a result, we were able to identify and focus on 4 distinct groups of patients for further analysis (HOX by HOPX) (Figure 3A) in whom the high/low expressions of HOX and HOPX were confirmed significant in each cluster (comparisons of average z-scores against zero; 1-sample t-test P<0.0001) (Figure 3B). The 4 groups also showed a significantly different prognosis: HOX patients had the longest OS, while HOX patients had the poorest outcome (P<0.0001) (Figure 3C).

Figure 3.

Comparison of expression patterns, clinical and biological features among subgroups of NTUH AML patients based on hierarchical clustering of HOPX and HOX family genes. (A) Heatmap of HOPX and HOX family genes in NTUH data. We identify 4 groups of patients based on a 2-step hierarchical clustering: HOX by HOPX. Molecular and clinical variables including gene mutations, cytogenetic abnormalities, and leukemia classifications are compared among the clusters. (B) Average z-scores of HOPX and HOX genes in each group. Whiskers denote average±standard deviation of the z-scores. Statistical significance of the z-scores against zero is assessed by the 1-sample t-test (****P<0.0001). (C) Overall survival of the 4 groups of patients. HOX patients have the best overall survival (OS), followed by HOX, HOX, and HOX (P<0.001).

To further compare the clinical and biological characteristics among AML patients with different expression levels of HOPX/HOX family genes, we analyzed patients’ gene mutations, cytogenetic abnormalities, and other clinical and lab parameters. WBC counts and LDH levels varied significantly among groups (ANOVA P=0.026 and 0.0009, respectively) (Figure 4A and B). HOX patients had the lowest WBC counts and LDH levels (median 5765/μL vs. 24,720/μL and 643 U/L vs. 1027 U/L, respectively; both P<0.001). Each subgroup also had distinct biological characteristics, including CD34 expression, gene mutations of FLT3, NPM1, CEBPA, RUNX1, DNMT3A, IDH1/2 (all P<0.0001) and ASXL1 (P=0.0002) (Figure 3A and 4C). FLT3-ITD and mutations in NPM1 and DNMT3A are more common in HOX patients regardless of HOPX expression levels; RUNX1 mutation is more frequent in HOPX regardless of HOX expression levels; CEBPA double mutation is predominantly seen in HOX patients; ASXL1 mutation is mainly present in HOX subgroup; IDH1/2 mutations are particularly rare in HOX patients; CD34+ blasts are low in HOX patients. Compared with other patients, HOX patients had higher incidences of FAB M0 (4 of 76 vs. 1 of 185; P=0.011), CD34 expression on leukemic cells (64 of 71 vs. 104 of 175; P<0.001), and mutations in ASXL1 (23 of 76 vs. 16 of 185; P<0.001), RUNX1 (21 of 76 vs. 15 of 185; P<0.001), and IDH1/2 (25 of 76 vs. 30 of 185; P=0.003), while HOX patients, when compared with others, had more FAB M5 (8 of 57 vs. 2 of 204; P<0.001) and mutations in NPM1 (47 of 57 vs. 19 of 204; P<0.001), FLT3 (FLT3-ITD) (26 of 57 vs. 38 of 204; P<0.001), MLL (MLL-PTD) (5 of 57 vs. 2 of 203; P=0.001), PTPN11 (8 of 57 vs. 8 of 203; P=0.005), and WT1 (8 of 57 vs. 11 of 204; P=0.026) (Figure 3A and data not shown). These results demonstrated marked distinctions between HOPX and HOX family genes in their association with genetic alterations and clinical features in AML.

Figure 4.

The distinct clinical and genetic features among acute myeloid leukemia (AML) patients stratified by expression of HOPX and HOX family genes. (A and B) Box plots of white blood cell (WBC) counts and serum lactate dehydrogenase (LDH) levels. There is a significant difference in both variables among the 4 clusters. The highest median of WBC count appears in HOX group, while that of HOX falls to the bottom of all clusters. Similar trends are seen in LDH levels. Statistical significance were tested by ANOVA tests. (C) The 4 clusters are associated with distinctive gene mutations. FLT3-ITD and mutations in NPM1 and DNMT3A are more common in HOX patients regardless of HOPX expression levels; the RUNX1 mutation is more frequent in HOPX regardless of HOX expression levels; the CEBPA double mutation is predominantly seen in HOX patients; the ASXL1 mutation is mainly present in HOX subgroup; IDH1/2 mutations are particularly rare in HOX patients. CD34+ blasts are low in HOX patients. (D) Association of quiescence HSC signature with expression of HOPX and HOX family genes. We employ a gene set enrichment scoring to quantify the overall activity of the gene signature in each sample; a positive/negative score represents a tendency toward quiescent/proliferative HSC state. Inter-group changes are tested by an ANOVA test. Asterisks denote significant differences from zero assessed by one-sample t-tests (***P<0.001).

Distinct associated HSC gene signatures between HOPX and HOX family genes

Although all HOPX and HOX family genes are stem cell markers, our data showed that the two were associated with distinct features in AML. Stem cell signatures could be divided into two groups with either quiescence or proliferation characteristics.[41] The low LDH levels and WBC counts in the AML patients with high HOPX and low HOX family gene expression (Figure 4A and B) raises the possibility that expression of HOPX may favor quiescence of stem cells. To test this hypothesis, we examined the expression profiles of each subgroup of our AML patients by a gene set scoring[30] based on a known quiescence signature in HSC.[41] A positive signature score denotes a tendency toward a quiescent HSC state. The significance level of a score against zero (representing no tendency) was tested by 1-sample t-test. Patients with high expression of HOX family genes did not exhibit a significant tendency toward the quiescence state regardless of the abundance of HOPX expression (P=0.22 and 0.45 for HOX and HOX, respectively) (Figure 4D). However, when HOX expression is low, HOPX and HOPX were significantly associated with quiescence and non-quiescence, respectively (P=0.0007 and 0.0003, respectively) (Figure 4D). Overall, our data suggested a fundamental difference between HOPX and HOX family genes in their stem cell properties in the AML setting.

Comparison of methylation patterns between HOPX and HOX family genes in AML

Besides the different expression patterns and associated clinical and biological features in HOPX compared with HOX family genes, we also sought to find out whether there was any difference in their methylation patterns by analyzing the TCGA epigenome-wide microarray dataset (n=194). Generally, these genes formed 3 clusters according to the methylation patterns (Online Supplementary Figure S7A). HOPX was largely unmethylated in most AML patients (mean methylation M-value −1.63; 1-sample t-test against zero P<0.001; area under curve with negative M 69.05%) (Online Supplementary Figure S7B). Methylation levels of HOXA3, HOXA4, HOXA5, and HOXB3 were generally high, while other HOX genes, including HOXA7, HOXA9, and HOXB4, were uniformly hypomethylated (Online Supplementary Figure S7A and B). Taken together, our data highlighted the different molecular and clinical features that distinguish between HOPX and HOX family genes in AML.

Discussion

To our knowledge, this is the first report regarding the prognostic significance of HOPX expression in de novo AML patients and the direct comparison between HOPX and the HOX family in normal and malignant hematopoiesis. The prognostic significance of HOPX expression is independent of common known clinical and genetic factors as well as several published gene signatures. We also showed that the promoter region was barely methylated in leukemic cells from AML patients, in contrast to heavy methylation in solid cancers,[8-10,12,13,42] indicating that CpG methylation is not one of the main mechanisms of regulation of HOPX gene expression in primary human AML cells. Finally, HOPX appeared to be a distinct homeobox gene in AML cells when compared with HOX family genes.

Studies have shown that HOPX is a stem cell marker of hair follicle, intestine, and lung alveolar cells.[5-7] Through functional annotation, our current study showed that HOPX expression was associated with HSC and LSC signatures in AML cells from our cohort and also two other validation cohorts, indicating that HOPX was an LSC marker in AML. Stemness is an established property pertaining to drug resistance and poor prognosis in cancer patients.[43] LSC signature is associated with unfavorable prognosis in AML patients. The underlying mechanisms by which stem cell signatures in AML cells predict poor treatment outcome have been postulated to be related to their association with chemotherapy resistance,[27] probably due to the quiescent nature of these cells. The tight association between HOPX expression and stem cell properties is likely a major reason for the unfavorable prognosis in AML patients with higher HOPX expression shown in this study. In addition, higher HOPX expression was significantly associated with the expression of some ABC transporters, a family of proteins that bind ATP as energy source to transport the endogenous or exogenous molecules through the cell membranes.[44,45] They are abundant in stem cells, including HSCs and LSCs, and are responsible for multidrug resistance in cancer treatment.[46] Therefore, leukemia patients with higher HOPX are less likely to obtain CR after induction chemotherapy. Further functional studies are needed to throw light on its significance in leukemia stemness and drug-resistance.

We showed that HOPX had distinct expression pattern and associated clinical and biological features when compared with other homeobox genes such as the HOX family in the AML setting. While they were both enriched in normal CD34+ HSPCs, their expression in AML was asynchronous. The findings that higher HOPX expression, accompanied with lower HOX expression, was closely associated with FAB M0 subtype, CD34 expression on leukemic cells, lower WBC counts, LDH levels, and quiescence stem cell signature indicates its relationship with more immature and quiescent stem cell characters.

Our study was mainly based on a retrospective cohort although we validated our results from other public array cohorts and 56 prospectively enrolled patients. Further studies in large prospective cohorts are warranted to confirm our observations. Moreover, in vivo studies are necessary to delineate the pathophysiological effects of HOPX in hematopoiesis and leukemogenesis.