Long non-coding RNA HOXB-AS3 promotes myeloid cell proliferation and its higher expression is an adverse prognostic marker in patients with acute myeloid leukemia and myelodysplastic syndrome

We retrospectively included the adult patients with newly diagnosed primary MDS and de novo AML at the National Taiwan University Hospital (NTUH) from 1992 to 2010. Among them, 157 MDS and 193 AML patients, who had available cryopreserved bone marrow (BM) cells for RNA array analysis and comprehensive clinical information, were recruited for this study. The Cancer Genome Atlas (TCGA) AML cohort available on the TCGA website (https://​cancergenome.​nih.​gov/​) and an independent cohort of 30 MDS patients subsequently diagnosed between January 2011 and May 2012 at the NTUH was served as the validation cohorts.

All patients with AML other than acute promyelocytic leukemia (non-APL AML, n = 174) underwent standard induction chemotherapy (Idarubicin 12 mg/m² per day for two to 3 days and Cytarabine 100 mg/m² per day for five to 7 days), and consolidation chemotherapy with two-to-four courses of high-dose Cytarabine (2000 mg/m² every twelve hours for 4 days, total eight doses), with or without an anthracycline (idarubicin or mitoxantrone), after they achieved complete remission (CR) as described in our previous studies [20]. Nineteen APL patients received concurrent all-trans retinoic acid (ATRA) and Idarubicin or Mitoxantrone as induction chemotherapy and ATRA with Idarubicin, Mitoxantrone or high dose Cytarabine as consolidation chemotherapy when they achieved CR. If the patients had relapsed or refractory AML, or adverse prognostic factors at diagnosis, such as adverse-risk cytogenetic abnormalities or somatic gene mutations, they underwent allogeneic hematopoietic stem cell transplantation (allo-HSCT) when they had feasible hematopoietic stem cell donors.

In the NTUH MDS training cohort, most MDS patients (70.7%) only received supportive care. Eight patients (5.1%) received AML-directed intensive chemotherapies, 18 (11.5%) received hypomethylating agent (azacitadine or decitabbine), and 20 patients (12.7%) underwent allo-HSCT.

The BM samples from AML and MDS patients were collected at diagnosis, and the mononuclear cells were isolated by Ficoll-Hypaque gradient centrifugation and cryopreserved as previously described [21, 22]. The BM cells from 20 healthy transplantation donors were used as the normal controls to compare the gene expressions with those of AML and MDS patients. This study was approved by the Institutional Review Board of NTUH (IRB number: 201507084RINA and 201503072RINC). All the patients have signed informed consents for the collection of samples and clinical information.

The BM cells were harvested directly or after one to 3 days of un-stimulated cultures. The metaphase chromosomes were banded by the G-banding method as previously described. [23] The determination of mutations in NPM1 [21, 24], AML1 (RUNX1) [25], ASXL1 [26], DNMT3A [20], EZH2 [27], IDH2 [28], NRAS [29], KRAS [29], TP53 [30], SETBP1 [31], SRSF2 [32], TET2 [33], MLL/PTD [34], SF3B1 [35], U2AF35 [36], and ZRSR2 [35] was performed as described previously.

The raw data of TCGA AML cohort was downloaded from TCGA website (https://​cancergenome.​nih.​gov/​). The detail methods of microarrays for NTUH AML and NTUH MDS cohorts were described in Additional file 1.

The expression levels of two transcript clusters, TC17002254.hg.1 and TC17002858.hg.1, on Affymetrix GeneChip® HTA 2.0 represent HOXB-AS3 (NCBI Reference Sequence: NC_000017.11) expression. TC17002254.hg.1 detects variants 1 to 5 of HOXB-AS3, and TC17002858.hg.1 detects all variants of HOXB-AS3. Because TC17002858.hg.1 detects all HOXB-AS3 variants and the expression pattern was similar between the two transcript clusters, expression of TC17002858.hg.1 was used to stratify patients.

OCI/AML3 and TF-1 were human myeloid leukemia cell lines. TF-1 cell line (BCRC number 60323) was purchased from Bioresource Collection and Research Center (BCRC), Hsinchu, Taiwan on Sep 29, 2014. BCRC (http://​www.​bcrc.​firdi.​org.​tw/​) is a nation-wide cell bank in Taiwan, and it provides the service of preservation, identification, and selling of cell lines. OCI/AML3 was a gift from Dr. Minden (Ontario Cancer Institute/Princess Margaret Hospital, Canada) in 2008. The detailed methods of cell cultures, constructions of lentiviral vectors with shRNA and lncRNA, lentiviral production, lentiviral infections, quantitative real time PCR, proliferation assay and nuclear-cytoplasm fractionation were described in Additional file 1.

BD Pharmingen™ APC BrdU Flow Kits (Cat. NO. 552598) was used for BrdU flow assay. Cells were incubated with 10 μM BrdU at 37 °C for three hours, and then harvested for BrdU flow assay. The detailed method was described in the user manual of the manufacture. The flow cytometry was performed on LSR II (BD Bioscience, San Jose, CA) through the service provided by the Flow Cytometric Analyzing and Sorting Core Facility at the NTUH, and on FACS Canto II (BD Bioscience, San Jose, CA) through the service provided by the Molecular and Immune Function Laboratory at Tai Cheng Stem Cell Therapy Center at the National Taiwan University.

For patients with de novo AML in the NTUH AML cohort, overall survival (OS) was measured from the date of diagnosis to the date of last follow-up or death. The patients were censored on the date of last follow-up if they were alive. Relapse free survival (RFS) was defined from the date of complete remission to the date of relapse, last follow-up, or death. Relapse and death were defined as events in the RFS analysis. If the patients were alive and in complete remission, they would be censored. For patients with primary MDS, OS was measured from the date of diagnosis to the date of last follow-up or death. The patients were censored on the date of last follow-up if they were alive.

Kaplan-Meier (KM) estimation was used to plot survival curves, and log-rank tests were used to calculate the difference of OS and RFS between different groups in AML patients and OS in MDS patients. Median follow-up duration was calculated by reverse KM estimation. Multivariate Cox proportional hazard regression analysis was used to investigate independent prognostic factors for OS. A P value less than 0.05 was considered statistically significant. All statistical analyses were performed with MedCalc® 15.6.1 software (https://​www.​medcalc.​org/​).

In order to find lncRNAs with significant prognostic implications in myeloid malignancies, we analyzed the microarrays data from two cohorts: one AML cohort from the NTUH (NTUH AML cohort), and another one from The Cancer Genome Atlas (TCGA AML cohort) [37]. The expression of 2824 probes in TCGA AML cohort and 4454 transcript clusters in NTUH AML cohort had prognostic significance (Fig. 1a). The intersect between the two cohorts was 907 probes/transcript clusters, among which eleven lncRNAs were validated as non-coding RNAs according to National Center for Biotechnology Information (NCBI) database (Fig. 1a). Among these lncRNAs, HOXB-AS3 is particularly interesting, because of the crucial roles of HOX genes in cell proliferation, hematopoiesis and leukemogenesis [36, 38, 39] and the clinical significance of an anti-sense lncRNA in the HOX clusters, i.e., HOTAIRM1, in leukemia [40]. Since the role of lncRNAs in myeloid malignancies remains unclear, the following studies were focused on the role of HOXB-AS3 in myeloid malignancies.

HOXB-AS3 is a lncRNA located at human HOXB cluster on the chromosome 17q21.32 and has eight transcriptional variants generated by alternative splicing (Fig. 1b). The entire gene is overlapped with HOXB5 and HOXB6 genes, and the transcriptional direction is antisense to the overlapping HOXB genes (Fig. 1b). The exons of HOXB-AS3, except for exon 2, are not overlapped with the exons of HOXB5 or HOXB6 (Fig. 1b). By using PhyloCSF pipeline, HOXB-AS3 is predicted to be a noncoding RNA (Fig. 1c), and the coding probability is low (0.126) by Coding Potential Assessment Tool (CPAT) [41]. Therefore, HOXB-AS3 is an anti-sense lncRNA in the HOXB cluster, similar to HOTAIR in the HOXC cluster, and HOTTIP and HOTAIRM1 in the HOXA cluster [13, 42, 43].

A similar anti-sense transcript, Hoxb5os (NCBI Reference Sequence: NR_131758.1), exists in the mouse HOXB cluster on chromosome 11 (NCBI Reference Sequence: NC_000077.6 Chromosome 11 Reference GRCm38.p4 C57BL/6 J), and is also predicted to be a lncRNA (Additional file 1: Figure S1). We aligned the sequences of human HOXB-AS3 with that of mouse Hoxb5os by plalign (http://​fasta.​bioch.​virginia.​edu/​) and observed two conserved regions (Additional file 1: Figure S2), one in exon 1 and the other in exon 6 of HOXB-AS3. These analyses indicated that HOXB-AS3 might be an evolutionarily conserved anti-sense transcript in the HOXB cluster in mammals.

To investigate the function of HOXB-AS3, we infected the myeloid cell line OCI/AML3 using lentivirus carrying control or each of the two HOXB-AS3 shRNAs together with GFP. The infected cells were sorted for proliferation assay. We found that the cells carrying HOXB-AS3 shRNA grew more slowly than those carrying control shRNA (Fig. 2a). The result suggested the anti-proliferative effect resulted from HOXB-AS3 depletion. To corroborate this effect, we stably knocked down HOXB-AS3 in two myeloid cell lines, OCI/AML3 and TF-1 (Fig. 2b and d). Downregulation of HOXB-AS3 suppressed the proliferation of OCI/AML3 and TF-1 cells by decreasing the cells entering the S phase (Fig. 2c and e; Additional file 1: Figure S3a and S3b). Compared with OCI/AML3 cells, the milder effect of HOXB-AS3 depletion in TF-1 cells was possibly resulted from the very low expression of HOXB-AS3 in TF-1 cells (Additional file 1: Figure S3d). In the reciprocal experiments, overexpression of HOXB-AS3 in TF-1 cell line enhanced cell proliferation by promoting cells entering the S phase (Fig. 2f and g; Additional file 1: Figure S3c). These findings indicated that HOXB-AS3 promoted cell proliferation in the myeloid cell lines.

To explore the mechanisms of HOXB-AS3 in proliferation, we investigated the expression of genes influenced by HOXB-AS3 depletion through the microarray analysis. We compared the gene expression profiles of OCI/AML3 cells infected by virus carrying shHOXB-AS3 (two stable cell lines infected with different shHOXB-AS3, shHOXB-AS3#1 or shHOXB-AS3#2) with those infected by control virus (two stable cell lines infected with shLacZ or pLKO vector).

We compared the gene expressions of HOX clusters between OCI/AML3 cells with and without HOXB-AS3 depletion because previous studies showed that most lncRNAs in HOX clusters are located in the nucleus and regulate the expressions of HOX genes in cis- or trans-manner [42, 43]. However, we found that the expressions of HOX clusters were not influenced by HOXB-AS3 depletion (Fig. 3a). Instead, we identified 147 genes and 790 transcript clusters differentially expressed when HOXB-AS3 was knocked down in OCI/AML3 cells (Fig. 3b; the criteria were that the fold change of average log2-transformed gene expressions was more than 1.5, and that ANOVA P value was less than 0.05. The analysis was computed by BRB array tool.). GSEA analysis revealed that these differentially expressed genes were involved in several pathways associated with cell cycle progression and DNA replication (Fig. 3c and Additional file 1: Figure S4). Wikipathway analysis computed by Affymetrix® TAC also showed that many differentially expressed genes were involved in cell cycle pathway, DNA replication, G1-S transition, and RB pathway (Additional file 1: Figure S5-S8). Moreover, the expressions of these genes were mostly downregulated when HOXB-AS3 was knocked down, consistent with a promoting function of this lncRNA in cell proliferation.

We validated a set of HOXB-AS3 regulated genes by RT-qPCR and confirmed that several genes involved in the cell cycle progression (CDK1, CCNB2, and CDC25A), DNA replication (PCNA), and assembly of pre-replicative complex (CDC6, MCM4, MCM6) were indeed downregulated when HOXB-AS3 was knocked down (Fig. 3d). Conversely, these genes were upregulated when HOXB-AS3 was overexpressed (Additional file 1: Figure S9). These findings suggested that HOXB-AS3 induced the expression of a number of genes involving in cell cycle progression and DNA replication to contribute to myeloid cell proliferation.

Furthermore, to identify the subcellular location of lncRNA HOXB-AS3, we used nuclear-cytoplasm fractionation to purify RNA from the nucleus and cytoplasm of OCI/AML3 cells, followed by reverse transcription and quantitative real time PCR. Our analysis demonstrated that the majority of HOXB-AS3 was located at the cytoplasm, different from the nuclear-residing MALAT1 and NEAT1 (Fig. 3e) [44, 45]. Therefore, our results suggest that HOXB-AS3 regulate the expressions of downstream genes through an indirect mechanism, which was different from other anti-sense lncRNAs in the HOX clusters.

To determine the clinical significance of HOXB-AS3 in de novo AML patients, we analyzed the microarrays and clinical data of the NTUH AML cohort and validated the results with TCGA AML cohort. In the NTUH AML cohort, HOXB-AS3 expression was determined based on the expression levels of TC17002858.hg.1, the transcript cluster representing all variants of HOXB-AS3 on Affymetrix GeneChip® HTA 2.0. The AML patients were stratified into higher- and lower-expression groups with the median expression of HOXB-AS3 as the cutoff level. The HOXB-AS3 levels in the higher-expression group were much higher than those of the healthy donors (Additional file 1: Figure S10; P <  0.000001), while the expression levels were similar between lower group and the healthy donors (Additional file 1: Figure S10). The clinical characteristics of the NTUH AML patients were listed in Additional file 1: Table S1 Patients with higher HOXB-AS3 expression were older and had higher frequencies of mutated NPM1/wild FLT3-ITD, MLL-PTD, and RUNX1 mutations, but lower frequency of CEBPAɑ double mutations than patients with lower expression.

Patients with higher HOXB-AS3 expression had similar complete remission rate to those with lower expression, but had higher relapse rate (Additional file 1: Table S1). With a median follow-up time of 88.1 months, the AML patients with higher HOXB-AS3 expressions had shorter OS and RFS than those with lower HOXB-AS3 expressions (Fig. 4a and b; median OS, 17.7 months versus not reached, P value < 0.0001; median RFS, 12.9 months versus not reached, P value 0.0070, respectively). When we stratified patients according to the expression levels of the other transcript cluster representing HOXB-AS3, TC17002254.hg.1, the results were similar (Additional file 1: Figure S11). Subgroup analysis showed that higher HOXB-AS3 expression was also a poor-risk factor for OS in the intermediate-risk cytogenetic group (Fig. 4c). We validated our findings in the TCGA AML cohort in whom higher HOXB-AS3 expression was also an adverse prognostic biomarker (Fig. 4d) [37]. Multivariate analysis showed that higher HOXB-AS3 expression tended to be an independent adverse prognostic factor for OS in AML patients (P = 0.0634, Table 1).

We further analyzed the patients with AML other than acute promyelocytic leukemia (non-APL), and stratified them according to the median expression of HOXB-AS3 (Additional file 1: Figure S12). Patients with higher HOXB-AS3 expression still had shorter OS than those with lower HOXB-AS3 expression (Additional file 1: Figure S13a). The subgroup analysis in the intermediate-risk patients based on the 2017 European LeukemiaNet (ELN) risk classification also illustrated adverse prognosis for the patients with higher HOXB-AS3 expression (Additional file 1: Figure S13b).

To investigate the clinical relevance of HOXB-AS3 expressions in primary MDS, we analyzed the microarrays and clinical data of the NTUH MDS training cohort and validation cohort. HOXB-AS3 expression was determined based on the expression levels of TC17002858.hg.1. According to the expression levels of HOXB-AS3, the MDS patients was stratified into four groups, lowest, intermediate low, intermediate high, and highest HOXB-AS3 expression groups (Additional file 1: Figure S14). Only the patients in the highest group had distinct higher HOXB-AS3 expressions compared with the healthy donors (Additional file 1: Figure S14; P <  0.000001), while patients in other groups had similar HOXB-AS3 expressions to normal controls.

With a median follow-up time of 39.2 months, the MDS patients with highest HOXB-AS3 expressions had the shortest OS compared to others (Fig. 5a). The OS was similar among the patients with lowest, intermediate low, and intermediate high expressions of HOXB-AS3. Similar results were obtained when we stratified the patients according to the expressions of another transcript cluster, TC17002254.hg.1, which also represented HOXB-AS3 (Additional file 1: Figure S15a). If we used the cutoff point between the highest and intermediate high groups (Fig. 5a) to stratify the NTUH MDS training cohort into two groups, higher HOXB-AS3 expression predicted shorter OS in the MDS patients (Fig. 5b). The clinical characteristics of the MDS patients were listed in Additional file 1: Table S2, and the correlations of HOXB-AS3 expressions and somatic gene mutations were listed in Additional file 1: Table S3. Patients with higher HOXB-AS3 expression had a high rate of leukemic transformation, higher bone marrow blast percentage and more frequently WHO higher-risk subtypes, but similar distribution of IPSS subtypes, compared to those with lower HOXB-AS3 expression. Higher HOXB-AS3 expression was associated with RUNX1, ASXL1, and IDH2 mutations. When we applied the same cutoff point to the MDS validation cohort, the prognostic significance of HOXB-AS3 expression for OS in MDS patients was also confirmed (Fig. 5c and Additional file 1: Figure S15b). Multivariate analysis demonstrated that higher HOXB-AS3 expression was an independent poor prognostic factor for OS in primary MDS patients, irrespective of other poor prognostic factors, including higher IPSS scores and adverse risk mutations (Table 2).

Furthermore, we analyzed prognostic implications of HOXB-AS3 expressions in subgroups of patients with different IPSS risk. Among the patients with IPSS lower risk MDS (low and intermediate-1 risks), those with higher HOXB-AS3 expression had shorter OS than those with lower expression (median, 29.2 months vs 77.3 months, P value 0.0194; Fig. 6a). On the other hand, HOXB-AS3 expression did not influence OS in the patients with IPSS higher risk MDS (intermediate-2 and high risks; Fig. 6b). Therefore, high HOXB-AS3 expression could further identify a subgroup of patients with high risk in the IPSS lower risk group.