Pseudogene BMI1P1 expression as a novel predictor for acute myeloid leukemia development and prognosis

The BMI1P1 levels of 144 de novo AML patients and 36 healthy donors were detected by real-time quantitative PCR (RQ-PCR). BMI1P1 was significantly down-regulated in AML compared with control (P < 0.001). A receiver operating characteristic (ROC) curve revealed that BMI1P1 expression could differentiate patients with AML from control subjects (AUC = 0.895, 95% CI: 0.835–0.954, P < 0.001). The percentage of blasts in bone marrow (BM) was significantly lower in BMI1P1 high-expressed group versus low-expressed group (P = 0.008). BMI1P1 high-expressed cases had significantly higher complete remission (CR) than BMI1P1 low-expressed cases (P = 0.023). Furthermore, Kaplan–Meier demonstrated that both whole AML cohort and non-M3-AML patients with low BMI1P1 expression showed shorter leukemia free survival (LFS, P = 0.002 and P = 0.01, respectively) and overall survival (OS, P < 0.001 and P = 0.011, respectively) than those with high BMI1P1 expression. Multivariate analysis also showed that BMI1P1 over-expression was an independent favorable prognostic factor for OS in both whole and non-M3 cohort of AML patients (HR = 0.462, 95% CI = 0.243–0.879, P = 0.019 and HR = 0.483, 95% CI = 0.254–0.919, P = 0.027). To further investigate the significance of BMI1P1 expression in the follow-up of AML patients, we monitored the BMI1P1 level in 26 de novo AML patients and found that the BMI1P1 level increased significantly from the initial diagnosis to post-CR (P < 0.001). These results indicated that BMI1P1 might contribute to the diagnosis of AML and the assessment of therapeutic effect.


INTRODUCTION
Acute myeloid leukemia (AML) is the most common type of myeloid leukemia characterized by uncontrollable heterogeneous clonal disorder and accumulation of malignant haemopoietic progenitor cells in bone marrow and blood [1]. Currently, cytogenetics, molecular genetics and clinical studies, which are associated with pathogenesis of AML, provide useful guides for identifying patients' prognosis information and better approaches to therapy [2,3]. Identifying molecular markers contributes to differentiating patients' risk and refining the prognosis of patients with AML [4].
Recently, significant attention has been paid to noncoding RNAs (ncRNAs), including microRNAs, long non-coding RNAs (LncRNAs), small interfering RNAs (siRNAs), pseudogenes, etc. [5]. It is being increasingly clear that ncRNAs play a functional role in diverse cellular processes, with their dysregulation already associated with origination and progression of cancers [6]. Pseudogenes were initially defined as unnecessary copies of coding genes by the fact that they lost the ability of coding functional protein due to gene mutations, a lack of transcription, or their inability to encode RNA [7]. Nowadays, accumulating evidence reveals that pseudogenes are associated with various diseases and www.impactjournals.com/oncotarget functions, one of which is cancer development [7][8][9]. Pseudogenes may be strongly linked to oncogenic development and can be used as diagnostic and prognostic biomarkers in different human cancers [10]. Patients with gastric cancer (GC) are characterized by lower serum levels of PTENP1 pseudogene, which shows a diagnostic ability (AUC > 0.8) when compared with healthy controls [11]. Over-expression of SUMO1P3 pseudogene has also shown its ability for discriminating GC patients from patients with benign gastric disease [12], and its over-expression was also positively correlated with the state of bladder cancer [13]. Analogously, pseudogene INTS6P1 expression is high and steady in normal people compared with hepatocellular carcinoma (HCC) patients. The pseudogene diagnostic value may be equal to that of alpha-fetal protein (AFP), the most common biomarker used in the diagnosis of HCC [14]. Besides being accurate diagnostic markers, pseudogenes also can be used as valuable prognostic markers to stratify cancer patients. For example, Hayashi et al. [15] showed that over-expressed OCT4-pg1 combined with genomic amplification like c-MYC can promote tumor cells' proliferation and angiogenesis while inhibiting apoptosis. OCT4-pg1 amplification was positively correlated with associated with a decreased overall survival in gastric cancer. As another example, the pseudogene PTENP1 affected the post-transcriptional regulation of its parental gene (PTEN) through competition for PTEN-targeting miRNAs, and patients who did express PTENP1 showed a more favorable outcome compared to those who did not express PTENP1 in clear cell renal cell carcinoma [16]. Previous works strongly suggested that pseudogenes did not only help us to understand the cancer pathogenesis but also could serve as a new panel of useful biomarkers for cancers. Until now, several pseudogenes have been identified in normal and malignant hematopoietic cell [17,18], but the function and the regulatory mechanisms of these pseudogenes for AML have not been defined in any studies yet.
BMI1 (Moloney murine leukemia virus integration site 1) is a polycomb ring finger oncogene involved in the regulation of p16 and p19, which are inhibitor genes for cell cycle progression [19]. Its expression plays a critical role in several signaling including wnt, akt, notch, hedgehog and receptor tyrosine kinase (RTK) pathway [20]. BMI1 is essential for efficient self-renewing and reconstituting activity of hematopoietic stem cells as well as leukemic stem cells and neural progenitors [21,22]. Overexpression of BMI1 has been reported in a number of human malignancies, such as bladder, skin, prostate, breast, ovarian, colorectal as well as hematopoietic malignances [23], and its over-expression is associated with poor prognostic in these malignancies. BMI1 pseudogene, namely BMI1P1, located on human chromosomal band Xq12, which has high homology with BMI1, has barely been studied in any cancers. This study was aimed to investigate the BMI1P1 expression in de novo AML patients and to analyze its clinical relevance, whether it might serve as a biomarker for predicting disease prognostic.

BMI1P1 expression in normal controls and AML patients
In our experiment, the BMI1P1 mRNA level in normal controls ranges from 0.000 to 660.68 with a median level of 9.825. The level of BMI1P1 expression in AML cases (0-83.090, median 0.039) appears significantly down-regulated than control subjects (P < 0.001, Figure 1). In addition, down-regulated level of BMI1P1 expression, which is compared with its level in control subjects (P < 0.05 for each subtype, Table 1), was found in different AML subtypes. The typical electrophoresis results of RQ-PCR products are shown in Figure 2.

Clinical and laboratory characteristics of AML
This cohort of 144 AML patients was divided into low-expressed group (< 0.159) and high-expressed group (≥ 0.159) according to the cut off value of 0.159. Age, white blood cells (WBC), hemoglobin (HB), platelets (PLT), FAB or WHO classifications and karyotypes did not differ significantly between BMI1P1 low-expressed group and high-expressed group. We further investigated whether the level of BMI1P1 was associated with patients' gene mutations. To test this hypothesis, we detected several gene mutations, such as C/EBPA, NPM1, FLT3 ITD, C-KIT, IDH1/2, DNMT3A and U2AF1. But we failed to find a significant correlation of gene mutations with BMI1P1 in these patients (data not shown). However, the rate of over-expression of BMI1P1 in female patients was significantly higher than that in male patients (P = 0.043). Also, the percentage of blasts in bone marrow (BM) was significantly lower in BMI1P1 high-expressed group versus low-expressed group (P = 0.008). BMI1P1 high-expressed www.impactjournals.com/oncotarget    cases had significantly higher complete remission (CR) than low-expressed cases (P = 0.023) ( Table 2).

Correlation between BMI1P1 expression and clinical outcome
115 AML patients with mean follow-up time of 7 months (range, 1-92 months) were included in survival analysis. Our research showed that the high level of BMI1P1 exhibited a positive impact on patients' survival. Kaplan-Meier demonstrated that patients with lowexpressed BMI1P1 had significantly shorter leukemia free survival (LFS, median 0 vs 6.5 months, respectively, P = 0.002) and overall survival (OS, median 5 vs 13 months, respectively, P < 0.001) than BMI1P1 highexpressed patients in the whole cohort of AML patients ( Figure 4A, 4B). This favorable prognosis associated with BMI1P1 over-expression was also observed in the non-M3 cohort of AML patients (LFS, median 0 vs 3 months, respectively, P = 0.01; OS, median 10.5 vs 4 months, respectively, P = 0.011) ( Figure 4C, 4D). However, we did not find that LFS and OS were obviously altered in the CN-AML group ( Figure 4E, 4F). Multivariate analysis, applying age (≤ 60 y vs > 60 y), sex (male vs female), WBC (≥ 30 × 10 9 /L vs < 30 × 10 9 /L), HB (< 110 g/L vs ≥110 g/L), PLT (100×10 9 /L vs 100 × 10 9 /L), karyotype classifications (favorable vs intermediate vs poor), gene mutations (mutant vs wild-type) and BMI1P1 expression status (high vs low) as covariates, also showed that BMI1P1 over-expression was an independent favorable prognostic factor for OS in both whole and non-M3 cohort of AML patients (HR = 0.462, 95% CI = 0.243-0.879, P = 0.019 and HR = 0.483, 95% CI = 0.254-0.919, P = 0.027, Table 3). However, we failed to find that BMI1P1 was an independent favorable prognostic factor for LFS in the two above groups (data not shown). To further investigated whether levels of BMI1P1 factored in patients' response to therapy, we monitored BMI1P1 levels of 26 patients with AML from the initial diagnosis to complete remission ( Figure 5A). As we expected, the levels of BMI1P1 increased significantly from initial diagnosis to the post-CR (P < 0.001) ( Figure 5B).

DISCUSSION
Standard chemotherapy and hematopoietic stem cell transplantation are common therapeutic protocols for patients with AML. Approximately 90% of both t (8; 21) and inv (16) AML patients achieve a complete remission by accepting anthracycline-and cytarabine-based induction chemotherapy [24]. However, these therapeutic protocols on the elderly population or some special subtypes of AML are less well defined. In the present, personalized medicine in cancer treatment is favored and admired progressively. Patients who harbor different variation of the human genome in the cancer can be treated accordingly. A more detailed classification of the cancer genome and epigenome, thus, needs to be achieved in AML. To this end, karyotypes are frequently referred to as an essential tool for the recognition of distinct subtypes of AML and have helped to identify prognostic group. What is more, molecular markers like FLT3, C/EBPA, and NPM1 gene mutations also show strong correlation with prognosis as well as some common molecular lesions, such as DNA methyltransferase 3 alpha (DNMT3A) and isocitrate dehydrogenase 1/2 (IDH1/IDH2) [25,26]. However, a classification solely based on karyotypes and pathological features has shown its limitations, and there are less than 30% AML patients owning gene mutations [27]. Similarly, our findings on gene mutations agree with this point, for the percentage of gene mutations including C/EBPA, NPM1, FLT3-ITD, C-KIT, IDH1/2, DNMT3A and U2AF1 was 13.4%, 11.0%, 12.6%, 2.4%, 5.6%, 7.9% and 3.9% in these patients, respectively. Therefore, more useful biomarkers are needed in clinical practices to divide this heterogeneous cohort of AML patients into multiple subtypes and offer guidance and evaluation in the treatment of each patient. Pseudogenes, which are highly homologous with their parental genes, are ideal candidates to sustain the expression of their parental genes by serving as competing endogenous RNAs (ceRNAs) which compete for the binding site of the same mRNAs [16,28]. In addition, some could were measured by RQ-PCR from the initial diagnosis to complete remission. (+) and (-) indicates up-regulation and down-regulation, respectively. (B) BMI1P1 was up-regulated in 92% (24/26) of post-CR versus ID (P < 0.001), the statistical significance was found by using Wilcoxon tests. Significance was defined as P < 0.05. regulate the expression of functional genes by producing endogenous small interference RNAs (siRNAs) [29,30] and antisense RNAs (asRNAs) [31,32], and some even could encode functional proteins [33,34]. It is speculated that pseudogenes can be the supplement to their parental genes via gene mutation in a particular position. Aberrant expression of pseudogenes can be used as diagnostic and prognostic biomarkers in human cancers [14][15][16]. In some cases, it has shown its higher diagnostic and prognostic trend than microRNAs and mRNAs [35]. Nevertheless, the expression levels and functions of pseudogenes in AML have been less studied.
BMI1(the parental gene of BMI1P1), a stem cell factor, was observed to be highly expressed in various types of human cancers [23,36], including AML [37]. It was reported that BMI1 was essential for leukemic reprogramming of myeloid progenitor cells (BM blasts) into leukemic stem cells [38] and played a crucial role in regulating the proliferative activity of leukemic stem and progenitor cells [21]. In this study, BMI1P1 was found to be significantly down-regulated in de novo AML compared with healthy controls. This down-regulated level of BMI1P1 was also observed in different AML subtypes. To our knowledge, this is the first report about BMI1P1 expression in cancers. Our results also indicated that low BMI1P1 expression might be a prospective biomarker for screening AML, especially CN-AML and non-M3-AML from healthy controls by ROC curves analysis. Furthermore, our results indicated that patients with lower BMI1P1 expression had significantly higher BM blasts when compared with those with higher BMI1P1. BMI1P1 may be involved in the negatively regulation of BMI1 and leads to a decline of BM blasts in turn. More researches are needed to confirm this conjecture.
Our study further demonstrated that BMI1P1 highexpressed patients achieved significantly better OS, LFS and CR in both the entire AML cohort and non-M3-AML patients. We also revealed that the expression of BMI1P1 was an independent prognostic factor for OS in both whole and non-M3 cohort of AML patients according to multivariate analyses. As prognosis guides therapy, BMI1P1 may be a future therapeutic target. As we know, assessment of gene mutations in AML contributes to identifying subgroups with markedly superior outcome (e.g, mutant NPM1 [39] or C/EBPA [40]) and inferior outcome (e.g, mutant C-KIT [41], DNMT3A [27], FLT3 ITD [42], MLL/KMT2A [27] or WT1 [43]). To determine whether BMI1P1 correlates with gene mutations in patients with AML, we tested 7 kinds of these gene mutations. However, the differences in the impact of mutations of FLT3, NPM1, C/EBPA, C-KIT on outcome were not found, and we also failed to find a significant correlation of gene mutations with BMI1P1 in these patients. Interestingly, dynamic monitoring BMI1P1 level in 26 cases of patients revealed that BMI1P1 levels were significantly increased from the initial diagnosis to complete remission by mentioned therapeutic protocols. From the results above, we concluded that determination of BMI1P1 levels could be used as an important indicator of disease prognosis and evaluation of curative effect. Obviously, prospective studies on larger series of AML patients are needed to confirm and expand our findings.
Unfortunately, limited information is available to describe the function of BMI1P1, which has never been reported as a tumor suppressor in any human cancer. However, we showed that AML patients with a high BMI1P1 expression have a favorable outcome, suggesting that pseudogene BMI1P1 might be a tumor suppressor. Pseudogene transcripts can serve as competing endogenous RNAs (ceRNAs) to regulate its parental coding genes' expression [44]. Because of their striking sequence homology, pseudogenes are the sequences that share multiple microRNA responsive elements (MREs) with their parental genes and that can compete with their parental coding genes for the binding site of shared microRNA molecules [10,44]. Taken all together, BMI1P1 may be functional by mediating miRNA expression in AML. Over-expression of BMI1P1 transcripts may be expected to arrest the functions of oncomiRs targeting essential genes to cellular repression, through competitive binding to the oncomiRs and somehow resulting in suppression of AML. The next step is to design more additional studies, including in vitro and in vivo functional assays, stem cell-associated assays and the relationship between BMI1P1 and its parental coding gene, to assess mechanisms for potentially effects of pseudogene BMI1P1 for AML. In the future, prospective screening for BMI1P1 expression and BMI1P1-targeted intervention may shed new light on the classification and treatment of AML.
In conclusion, our study showed that pseudogene BMI1P1 was down-expressed in AML. Pseudogene BMI1P1 may serve a biomarker for detection of AML. Interestingly, BMI1P1 may serve as an important prognostic and initial treatment marker for AML.

Patients and samples
The bone marrows collected from 180 samples, including 144 patients with de novo AML treated in the Affiliated People' Hospital of Jiangsu University and 36 healthy donors regarded as normal controls after obtaining the written informed consent. All the patients were standardly diagnosed according to the French-America-British (FAB) and the World Health Organization (WHO) criteria [45,46]. Treatment protocol was described in our previously reported work [47]. The main clinical and laboratory characteristics of the patient cohort were summarized in Table 1. www.impactjournals.com/oncotarget RNA isolation, reverse transcription and realtime quantitative PCR Mononuclear cells from bone marrow samples were separated by Ficoll-Hypaque gradient. Total RNA from bone marrow mononuclear cells (BMNCs) was isolated by using Trizol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions. Reverse transcription was performed on iCycler Thermal Cycler (Eppendorf, Hamburg, Germany) using reaction mixture containing 2 μg of total RNA, dNTPs 10 mM, random hexamers 10 μM, RNAsin 80 units, and 200 units of MMLV reverse transcriptase (MBI Fermentas, Hanover, USA) to synthesize cDNA. The system of reverse transcription was incubated for 10 min at 25°C, 60 min at 42°C, and then stored at -20°C.
BMI1P1 was amplified using the primers 5′-AGTGGTATCTGCTCACT-3′ (forward) and 5′-CCTCC ACAAAGCACACACAT-3′ (reverse) with expected products of 210 bp. Real-time quantitative PCR (RQ-PCR) reactions were performed on a 7500 Thermocycler (Applied Biosystems, CA, USA). Reactions mixture of 20 μL in each tube consisting of 0.25 μM of primers, 10 μL SYBR Premix Ex Taq II, 0.4 μL 50×ROX (TaKaRa, Japan) and 50 ng of cDNA. RQ-PCR was carried out at 95°C for 30 s, followed by 40 cycles at 95°C for 5 s, 63°C for 30 s, 72°C for 30 s, and 80°C for 30 s to collect fluorescence, finally followed by the melting program at 95°C for 15 s, 60°C for 60 s, 99°C for 15 s, and 60°C for 15 s. Negative and positive controls were involved in all assays. The abundance of BMI1P1 mRNA was estimated by housekeeping gene ABL (non-receptor tyrosine kinase). Relative levels of BMI1P1 expression were calculated according to the following equation: N BMI1P1 = (E BMI1P1 ) ΔCT BMI1P 1(control-sample) ÷ (E ABL ) ΔCT ABL (control-sample) ×1000‰. The parameter efficiency (E) derived from the formula E=10 (-1/slope) (the slope referred to CT versus cDNA concentration plot).

Gene mutation detection
IDH1/2, DNMT3A and U2AF1 mutations were detected according to the literatures reported previously [48][49][50][51]. The detection of nucleophosmin (NPM1) and C-KIT mutations was performed by using PCR and highresolution melting analysis (HRMA). All positive samples were confirmed by direct DNA sequencing. FLT3-ITD and C/EBPA were detected by direct DNA sequencing.

Statistical analysis
Statistical analyses were performed using the SPSS 18.0 software package (SPSS, Chicago, IL). Chi square test or Fisher exact test was used to compare the difference of qualitative data between patients groups. For comparison of quantitative data between groups; Kruskal-Wallis test (multiple groups) and Mann-Whitney U-test (two groups) were used. Receiver operating characteristic (ROC) curve and area under the ROC curve (AUC) were designed to assess the diagnostic value of BMI1P1 expression in discriminating AML patients from normal controls. Kaplan-Meier test and Cox regression analysis were applied to analyze the impact of BMI1P1 level on the prediction of survival in AML cases. For all analyses, a P value less than 0.05 (two-tail) was considered statistically significant.