Polymorphisms at microRNA binding sites of Ara-C and anthracyclines-metabolic pathway genes are associated with outcome of acute myeloid leukemia patients

Gene polymorphisms at microRNA-binding sites (poly-miRTS) may affect gene transcription and expression through miRNA regulation, which is associated with cancer susceptibility, sensitivity to chemotherapy and prognosis. This study investigated the association between poly-miRTS of Ara-C/anthracycline metabolic pathways genes and the outcome of acute myeloid leukemia (AML) in Chinese patients after Ara-C-based chemotherapy. A total of 17 poly-miRTS were selected from the SNPinfo Web Server and genotyped in 206 Chinese Han non-FAB-M3 AML patients using the SEQUENOM Mass-ARRAY system. Among these 17 poly-miRTS, five Ara-C metabolic gene single nucleotide polymorphisms (SNPs, NT5C2 rs10786736 and rs8139, SLC29A1 rs3734703, DCTD rs7278, and RRM1 rs1042919) were identified to significantly associate with complete AML remission and/or overall and relapse-free survival (OS and RFS, respectively), and four anthracycline-metabolic gene SNPs (ABCC1 rs3743527, rs212091, and rs212090 and CBR1 rs9024) were significantly associated with chemotherapy-related toxicities. Moreover, SLC29A1 rs3734703 was shown to associate with both chemotherapy response and survival (adjusted OR 2.561 in the overdominant model; adjusted HR 2.876 for OS and 2.357 for RFS in the dominant model). The data from the current study demonstrated that the poly-miRTS of Ara-C/anthracyclines metabolic genes predicted the sensitivity and side effects of AML to Ara-C-based chemotherapy and patient survival. Further study will confirm them as biomarkers for AML patients after Ara-C-based chemotherapy.

common gene SNPs in the coding regions; however, these may not consider the role of the gene 3′-untranslational region (3′-UTR), which may affect post-transcriptional gene regulation. For example, microRNAs (miRNAs) are a class of small endogenous non-coding RNA that regulate gene expression through binding to the 3′-UTR of their target mRNAs [7,8]. A growing number of studies have suggested that gene polymorphisms at microRNA binding sites (poly-miRTS) are closely related to cancer susceptibility, sensitivity to chemotherapy, and patient prognosis [9,10], indicating that these SNPs could be a useful novel class of genetic variations for further investigation and that they may potentially become novel markers in predicting treatment responses and patient survival. Mishra et al. [11] in 2007 first identified a poly-miRTS that was associated with pharmacogenomics. Thus, further study of the poly-miRTS in genes that are related to drug metabolism could help medical oncologists better select treatment regimens and predict treatment responses.
Based on this postulation, we therefore hypothesized that poly-miRTS in Ara-C and anthracycline-metabolic pathway genes could predict response to and prognosis of Ara-C-based chemotherapy in Chinese AML patients. In this study, we searched the public databases and identified SNPs at the 3′-UTR of those genes, which may affect miRNA binding and consequently influence their expression and genotyped 17 of these SNPs in 206 blood samples from Chinese AML patients. We then evaluated and associated them with survival, chemotherapy response and toxicity in AML patients after Ara-C-based chemotherapy.

Study population
A total of 206 patients with de novo AML (diagnosed according to the WHO criteria) other than M3 were treated in The First Affiliated Hospital of Zhengzhou University (Zhengzhou, China) between January 2012 and December 2016. The patients that were diagnosed with any other cancers or previously administered cytotoxic drugs or radiation were excluded from this study. Clinical data from these patients were collected from their medical records, and patients were followed up until April 2, 2017 via telephone review. All participants were Han Chinese. This study was approved by the Ethical Committees of The First Affiliated Hospital of Zhengzhou University (Zhengzhou, China), and informed consent was obtained from each patient or their family members.

Evaluation of chemotherapy response and toxicity
The treatment responses were evaluated after the second cycle of the induction chemotherapy regimen as complete remission (CR) or non-CR according to the International Working Group AML criteria [12]. The CR was defined as the following: blast cell counts in the bone marrow < 5%, absence of extramedullary disease; absolute neutrophil count > 1 × 10 9 /L, platelet count > 100 × 10 9 /L, and independence of red cell transfusion. Patients with other treatment responses, including partial remission, non-remission and early death were assigned to the non-CR group. Disease relapse was defined as the presence of more than 5% of blast cells in the bone marrow or the reappearance of blast cells in the blood or the development of extramedullary disease. Relapse-free survival (RFS) was evaluated by measurement from the date of remission until the date of relapse or death from any cause; patients not known to have relapsed or died at the last follow-up were censored on the date they were last followed up. The overall survival (OS) was measured from the date of diagnosis until death from any cause, with observation censored on the date the patient was last known to be alive or at the time of hematopoietic stem cell transplantation after their CR. The adverse events during chemotherapy were recorded and graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events version 4.0 [13]. For data analysis, more than 2 grade adverse events were collected and analyzed.

Genotyping
Blood samples were collected into Ethylene Diamine Tetraacetic Acid tubes from patients prior to chemotherapy, and genomic DNA was extracted using the TIANamp Blood DNA Kit (TIANGEN, Beijing, China) and stored at − 80 °C until use. These 17 poly-miRTS were genotyped using the SEQUENOM Mass-ARRAY system with specific amplification primers and extension primers (Additional file 1: Table S1).

Statistical analysis
Genotype deviations of the Hardy-Weinberg equilibrium were assessed using the Pearson Chi square test, while continuous data were converted into categorical data using their median. The significant difference in genotypes and clinical information between chemotherapy toxicity, CR and non-CR was calculated using Pearson/ Continuity Correction Chi Square test or Fisher's exact test. Odds ratios (ORs) and their 95% CIs were calculated to estimated the relative risk of responding to treatment using the logistic regression analysis while adjusting for age, risk stratifications and platelets. Kaplan-Meier curves and the log-rank test were performed to assess OS and RFS stratified by the genotypes of each SNP. Associations between the clinicopathological data and survival were also estimated. The hazard ratios (HR) and 95% CIs for OS and RFS were estimated using the Cox proportional hazards model while adjusting for risk stratifications. All statistical analyses were performed using SPSS 21.0 software (SPSS Inc., Chicago, IL, USA), and P < 0.05 in a two-sided test was considered to be statistically significant.

Patient characteristics and treatment outcomes
The baseline characteristics and treatment results of all 206 AML patients are summarized in Table 2

Association between clinical characteristics and chemotherapy response and toxicity, OS, and RFS
Clinical characteristics, such as platelets counts and risk stratification, have been significantly associated with chemotherapy response and toxicity, OS, and RFS (Table 3). Platelet counts at diagnosis were a unique clinical characteristic that was significantly associated with response to chemotherapy (P = 0.006); the patients with higher platelet counts had a higher CR ratio (92.2% vs. 78.6%, OR 3.225, 95% CI 1.362-7.635). The patients with platelet counts > 38.5 × 10 9 /L had a longer mean OS than the patients with counts < 38.5 × 10 9 /L (1525 days vs. 1290 days); however, this did not reach statistical significance (P = 0.057).  15:235 Risk stratification was the only clinical factor associated with chemotherapy toxicity in Ara-C-based treatment. Compared with the low-risk patients, the intermediate risk patients had a lower risk of liver function damage and cardiotoxicity (P = 0.008, OR 0.213, 95% CI 0.071-0.638; P = 0.005, OR 0.126, 95% CI 0.030-0.535, respectively). Furthermore, the risk stratification was also associated with the OS (P = 0.022) and RFS (P = 0.008). Because there was no mortality in the low risk group, we put the low and intermediate risk patients together to carry out the OS analysis. Our data showed that compared with the low and intermediate risk groups, the high risk patients had a relatively higher risk of death [the mean OS, 1048 days vs. 1473 days, P = 0.027, HR 2.325, 95% CI (1.103-4.900)]; the high risk patients compared with low risk also had poorer RFS (mean RFS of 740 days vs. 1532 days, HR 4.821, 95% CI (1.611-14.427).
However, other clinical characteristics, such as gender, age, WBC, hemoglobin count, and BM blast percent at diagnosis showed no significant differences in CR ratio, chemotherapy toxicity or survival rates (Additional file 2: Table S2, Additional file 3: Table S3).

Associations between poly-miRTS and response to chemotherapy
The allele frequencies and genotype distributions of these 17 poly-miRTS are summarized in Additional file 4: Table S4 with the Hardy-Weinberg equilibrium. Patient characteristics, such as chemotherapy regimens or risk groups, showed no significant differences according to the genotypes of these SNPs (Additional file 5: Table S5). Associations between poly-miRTS and chemotherapy response are shown in Additional file 4: Table S4. Specifically, 4 of 17 poly-miRTS (SLC29A1 rs3734703, DCTD rs3811810 and rs7278, and RRM1 rs1042919) were significantly associated with CR ( Table 4). The Rs3734703 CC+AA genotypes had a higher CR ratio than the CA genotype (89.2% vs. 77.6%, P = 0.027 in the overdominant model). A higher CR ratio also occurred in the rs3811810 AA+GA genotypes (97.4% vs. 82.6%, P = 0.018 in the dominant model) and in the rs7278 TT+TC genotypes (97.6% vs. 82.4%, P = 0.014 in the dominant model), while Rs1042919 AA+AT showed a lower CR ratio compared with the TT genotype (77.3% vs. 92.7%, P = 0.002 in the dominant model).
We then performed a logistic regression analysis for associations between the above four poly-miRTS and CR ratio after adjusting for age, risk stratifications, and platelets count. We found that only three poly-miRTS (rs3734703, rs7278, and rs1042919) were significantly associated with chemotherapy response (Table 4). In particular, a higher CR ratio occurred in patients with the rs3734703 CC+AA genotypes (P = 0.023 in the overdominant model) with an OR of 2.561 (95% CI 1.136-5.775) or rs7278 TT+TC genotypes (P = 0.040 in the dominant model) with an OR of 8.572 (95% CI 1.106-66.423). However, patients with rs1042919 AA+AT  showed a lower CR ratio (P = 0.002 in the dominant model) with an OR of 0.246 (95% CI 0.100-0.604).

Association between poly-miRTS and chemotherapy toxicity
There were only four poly-miRTS in anthracyclines pathways associated with chemotherapy toxicity (  Table S6).

Association between poly-miRTS and survival of AML patients
The Kaplan-Meier curves with the log rank test showed that there was a significant association between SLC29A1 rs3734703 and NT5C2 rs10786736 and rs8139poly-miRTS with survival (Figs. 1, 2). The multivariable analysis using Cox regression after adjusting for risk stratifications showed that rs3734703, rs10786736, and rs8139 poly-miRTS were independent predictors for OS (Table 6) and RFS (

Combined effects of SNPs on chemotherapy response, OS, and RFS
We found that rs10786736, rs8139, rs3734703, rs7278, and rs1042919 had significant associations with Chinese AML chemotherapy response and/or OS and RFS individually. To analyze the combined effects of multilocus SNPs on AML prognosis, we created a combined genotype score model [17] using the defined genotype score, i.e., favorable genotypes (rs7278 TT/TC genotype, rs3734703 CC/AA genotype, or rs1042919 TT genotype for CR; rs10786736 CC/CG genotype, rs3734703 CC genotype, or rs8139 TT/CT genotype for OS and RFS) defined as score 1 and unfavorable genotypes (i.e., rs7278 CC genotype, rs734693 CA genotype, or rs1042919 AA/AT genotype for CR; rs10786736 GG genotype, rs3734703 AA/CA genotype, or rs8139 CC genotype for OS and RFS) defined as score 0. A defined favorable response group (composite score 1, 2, or 3) in which patients have at least one favorable genotype had a higher CR ratio compared with a defined unfavorable response group (composite score 0) in which patients have three unfavorable genotypes (85.4% vs. 68.8%, P = 0.005). After adjusting for age, risk stratifications and platelets, binary logistic regression analysis indicated that the defined favorable response group still had a higher CR ratio (OR = 3.624; 95% CI 1.398-9.395; P = 0.008) ( Table 8).
A defined favorable prognosis group (including a composite score of 2 or 3) in which patients have at least two favorable genotypes showed longer OS and RFS than a defined unfavorable prognosis group (including a composite score of 0 or 1) in which patients have at least two unfavorable genotypes (1037 days vs. 1551 days, P = 0.001 for OS; 762 days vs. 1264 days, P = 0.000 for RFS, Fig. 3). After adjusting for risk stratifications,  15:235 multivariate analysis indicated that the favorable prognosis group appeared to be an independent predictive factor for longer OS and RFS (HR = 0.316, 95% CI 0.155-0.642, P = 0.001 for OS; HR = 0.389, 95% CI 0.224-0.675, P = 0.001 for RFS, Additional file 8: Table S7).

Discussion
To the best of our knowledge, the current study is the first to evaluate and associate the poly-miRTS of pharmacogenomics-related genes in the metabolism of Ara-C and anthracyclines with AML treatment responses and outcome. We identified five poly-miRTS in Ara-C-metabolic genes (NT5C2 rs10786736 and rs8139, SLC29A1 rs3734703, DCTD rs7278, and RRM1 rs1042919) to be significantly associated with complete remission after AraC-based chemotherapy and/or OS and RFS in these Chinese AML patients. The combined effects of these SNPs showed that patients with more unfavorable genotypes had worse prognosis. The combination of multilocus SNPs could improve the detection power of genetic effects associated with the treatment outcomes. We also found that four poly-miRTS in anthracyclines-metabolic genes (ABCC1 rs3743527, rs212091, and s212090 and CBR1 rs9024) were associated with chemotherapy toxicity. A future prospective study with a larger patient population will confirm the current data.
We found that NT5C2 rs10786736 and rs8139 were independent predictors for AML survival. The rs10786736 GG genotype compared with the CG genotype has a shorter OS and RFS, and the rs8139 TT+CT genotypes have longer OS and RFS than the CC genotype. Cytosolic 5′-nucleotidase II (NT5C2) functions in dephosphorylating nucleoside triphosphates and is able to deactivate Ara-C through the de-phosphorylation of Ara-C monophosphate to Ara-C, and an increase in NT5C2 expression has shown to associate with the development of resistance to nucleoside analog-based chemotherapies, such as Ara-C, gemcitabine, and cladribine, in different cancer cell lines [18,19]. Levels of NT5C2 expression correlated with the resistance of primary leukemic  15:235 cells to Ara-C treatment in vitro and predicted a poorer clinical outcome in AML patients [20,21]. However, to date, there are only a few of studies reporting the association between NT5C2SNPs and outcomes in AML patients. For example, Mitra AK [6] sequenced NT5C2 and identified 41 genetic variants with twenty-five novel SNPs and then associated NT5C2SNPs with NT5C2 mRNA levels and Ara-C sensitivity in HapMap cancer cell lines as well as with treatment responses of pediatric AML patients to Ara-C. To date, this is the only study of rs10786736 with an outcome showing no statistical association between rs10786736 and Ara-CTP levels (after 1 and 2 days of treatment) and the clinical response of AML patients. However, the GG genotype was associated with higher NT5C2 mRNA levels and Ara-C cytotoxicity as well as tumor cell resistance to Ara-C in HapMap samples, which was also associated with Ara-C sensitivity of primary AML leukemic blasts. In consensus with the HapMap results, the GG genotype was associated with a greater Ara-C LC50 value versus the CG genotype [6]. This finding supported our current data on NT5C2 variant association with AML CR and survival in AML patients.
In the current study, we also observed SLC29A1 rs3734703 to associate with CR and survival in AML patients who received Ara-C based treatment. The solute carrier family 29 (also known as the equilibrative nucleoside transporter), member 1 (SLC29A1, or ENT1) plays a crucial role in cell uptake of anticancer nucleoside agents or nucleosides from the surrounding medium, and SLC29A1 is known to transport approximately 80% of Ara-C into leukemic cells [22,23]. In mechanistic studies of different Ara-C resistances in leukemic cells, altered SLC29A1 expression was associated with treatment resistance [24]. Many studies also reported that functional abnormalities in SLC29A1were associated with AML resistance to Ara-C [25,26]. In previous studies, SLC29A1 genetic variants (such as rs693955, rs9394992, and rs324148) were associated with treatment outcomes in AML patients [27,28]. In terms of rs3734703, only two studies reported that AML patients with a high frequency of the major "C" allele (in other words, low MAF) of rs3734703 had a poor response to Ara-C-based therapy [29], while another study [30] did not find any rs3734703 associations with CR or RFS after induction   chemotherapy with AML. However, the combination of rs3734703 AA or AC genotype with the TYMS rs2612100 AA genotype was significantly associated with shorter RFS compared to the wild type, which is in line with our current data. A possible explanation of these results is that SLC29A1 rs3734703 may increase SLC29A1 expression and in turn induce the uptake of Ara-C by AML cells and increase tumor cell apoptosis and clinical CR. Furthermore, ribonucleotide reductase (RR) is composed of dimerized large (RRM1) and small (RRM2) subunits and regulates intracellular pools of deoxy-CTP (dCTP), the expression of which leads to tumor cell resistance to nucleoside analogs, such as Ara-C treatment [31]. Indeed, tumor cell lines and leukemic blasts with high dCTP expression levels were resistant to Ara-C [32][33][34]. The biochemical modulation of Ara-C by nucleoside analogs, such as fludarabine and cladribine, stimulated Ara-CTP accumulation in leukemic cells from adult and pediatric patients [35,36]. Our current study showed that RRM1 SNP rs1042919 was associated with chemotherapy response of Chinese AML patients to AraCbased chemotherapy and further supported data from a previous study [37]. The study [37] identified RRM1 and RRM2 genetic variations by sequencing of the genomic DNA from HapMap European and African ancestry panels and revealed that the RRM1 rs1042919 AT genotype Table 8 Combined effects of rs7278, rs3734703 and rs1042919 genotypes on chemotherapy response in AML patients a Combined genotype score model was created by compiling the genotyped data of SNPs rs7278, rs3734703 and rs1042919. Score 1 indicated favorable genotypes (i.e., rs7278 TT/TC genotype, rs3734703 CC/AA genotype, or rs1042919 TT genotype) and a score of 0 indicated unfavorable genotypes (i.e., rs7278 CC genotype, rs734693 CA genotype, or rs1042919 AA/AT genotype). After adding up these scores, four composite score groups were generated: composite score 0, 1, 2 and 3 b Adjusted for age, risk stratifications and platelets was associated with lower intracellular Ara-CTP levels in leukemic cells after 1 day of treatment with Ara-C alone and 2 day treatment with Ara-C in combination with cladribine. The same research team also observed an association between RRM1 rs1042919 and poor survival and the risk of relapse in patients with AML97 treatment [38]. However, in the current study, we did not observe any associations between RRM1 rs1042919 and survival in AML patients. One of the reasons for this finding may be because patients in the AML97 study received cladribine in combination with Ara-C, whereas our patients only received Ara-C-based therapy. Cladribine, as a purine analog, mimics the nucleoside adenosine to inhibit the activity of adenosine deaminase, which reduces the cellular pool of deoxynucleotide levels [39]. Our current study also assessed DCTD rs7278 polymorphisms in association with the treatment response of AML patients to AraC-based chemotherapy and found that TT/TC genotypes reached a higher CR rate compared to the wild genotype. DCTD enzyme can deaminate Ara-CMP to Ara-UMP; however, its role in Ara-C-resistance is poorly understood. Previous studies suggested a substantial role for DCTD in Ara-CMP metabolism in T-lymphoblastic leukemia [32,40,41], and the association between DCTD polymorphisms and clinical outcome warrants further investigation [40,41].
Since the 1960s, anthracyclines have been widely prescribed for AML treatment, and polymorphisms of anthracyclines-metabolic pathway genes have also been widely studied [43][44][45][46]. Association of anthracyclines disposition in blast cells and different tissues with treatment outcomes of AML patients could be influenced by SNPs of the anthracyclines pathway genes (such as CBR and NQO1) and the efflux transporters (ABCB1 and ABCC1) [16]. The current study failed to show any significant associations between the poly-miRTS of anthracycline-pathway genes and CR or OS/RFS of AML; however, we did show that four poly-miRTS were associated with chemotherapy toxicity, i.e., ABCC1 rs3743527, rs212091, and rs212090 were significantly associated with myelosuppression and gastrointestinal reaction, which were reported to associate with a risk in developing lung cancer [42]. Moreover, cardiotoxicity was the most common anthracycline-induced toxicity [43,44]; however, the current study only observed one poly-miRTS (CBR rs9024) to be significantly associated with cardiotoxicity. The CBR rs9024 GG genotype was associated with an increase in the clearance and reduction of doxorubicin exposure levels in Asian breast cancer patients [45]. However, another study did not show the GA/AA genotype to increase the risk of cardiomyopathy in childhood cancers (including 68 AML) [46]. Kalabus [47] reported that CBRrs9024 impacted CBR1 phenotypes in the liver, and the homozygosis of the major G allele was associated with significantly higher CBR1 protein levels and CBR doxorubicin reductase activity. The same research team also carried out a functional study [48] and showed that hsa-miR-574-5p and hsa-miR-921 significantly decreased the luciferase activity of CHO cells after transfection with the CBR1 3′-UTR construct carrying the major rs9024 G allele (35 and 46%, respectively), as well as a decreased level of CBR1 protein (48 and 40%, respectively) and CBR1 activity (54 and 18%, respectively) in lymphoblastoid cells that contain the homozygous major rs9024 G allele. On this basis, we speculate that the patients with CBR rs9024 GG may have a lower risk in developing cardiotoxicity after anthracycline treatment, and our current data supported this notion.
Last but not least, platelet counts at diagnosis were also an independent predictor for CR of AML patients after Ara-C based chemotherapy; these data are novel and supported previous studies of cytogenetic and molecular abnormalities as independent predictors for the prognosis of AML patients [49][50][51]. Moreover, our current study showed that adverse cytogenetic abnormalities only counted for 25.7% (n = 53) of patients and that the majority of patients were in the intermediate and favorable risk stratifications without a difference in survival. Hence, only cytogenetic alterations may not be sufficient to predict AML prognosis, and our current results indicate that poly-miRTS in genes involved in the metabolic pathways of Ara-C and may be useful biomarkers to predict treatment response and prognosis in AML. However, future functional studies are required to define the role of these SNPs in AML. There are some limitations of our current study; for example, our current study is retrospective and patients showed some differences in treatment regimens, and the study population is relatively small.

Conclusions
Although further functional evaluations and confirmatory studies are need to further support our current data, the current study demonstrated that the miRNA-binding site SNPs of Ara-C and anthracycline metabolic pathways genes could predict treatment responses, side effects and survival in AML patients after Ara-C-based chemotherapy. They could provide insightful information to identify patients with an increased risk of adverse reactions or decreased chemotherapy responses for future individualized chemotherapy and predicting outcomes in AML patients.