Meta-analysis of gene expression in relapsed childhood B-acute lymphoblastic leukemia

B-Acute lymphoblastic leukemia (ALL) accounts for 80% of childhood leukemias, and relapsed B-ALL remains as the leading cause of cancer related deaths among children [1, 2]. Despite the 5-year survival rate for pediatric ALL exceeding 90% after treatment with multi-agent chemotherapy tailored to established risk factors [3], nearly 20% of patients will still relapse and succumb to disease. Relapsed B-ALL has a dismal prognosis, with overall survival rates of 35–40% even when treated with intensified chemotherapy or stem cell transplantation [4–6]. To date, the biological mechanisms of relapsed ALL remains largely unknown. Therefore, there is a pressing need to gain better understanding of the molecular mechanisms governing relapsed ALL, with the hope of developing more effective treatment plans and to improve patients’ survival rate.

In the past decades, microarray has been widely used to identify candidate biomarkers and therapeutic targets by studying the gene expression changes at the genome wide level. Several studies on diagnosis-to-relapsed ALL have been performed to unlock the dysregulated genes and pathways essential in driving relapsed ALL [7–10]. However, only a very small number of genes were found significantly differentially expressed between diagnosis and relapse, and the results were not consistent across all these studies. These discordant results therefore have limited the reliability for further validation or development into clinically useful biomarkers and therapeutic targets. It has been well recognized that small sample sizes, different microarray platforms, and different statistical methods are among the limiting factors contributed to the discordant results. To resolve this limitation, meta-analysis represent a powerful approach to combine different datasets from different studies to improve the reliability and generalizability of the findings by increasing its statistical power analysis. Meta-analysis on gene expression data has yielded new biological insights, as well as identification of more robust and reliable candidate biomarkers and therapeutic targets [11–13].

To identify differentially expressed genes across multiple datasets, we employed a non-parametric ‘rank product’ method [14, 15]. RankProd is among the most popular tool which utilizes a non-parametric statistical method and outperforms other meta-analysis methods, including metaArray [16], GeneMeta [17], and MAMA [18], by ranking the differentially expressed genes based on false discovery rate. Matched diagnosis and relapse samples represent the most ideal biological samples to study the mechanisms for relapse. Hence, in this study, we sought to identify differentially expressed genes associated with relapsed ALL by performing a meta-analysis on three independent microarray datasets of paired diagnosis-relapsed B-ALL, with the hope of providing new insights into the molecular mechanisms of relapsed B-ALL, as well as to identify potential therapeutic options to improve patients’ outcome. Interestingly, our analysis found a long list of significantly differentially expressed genes which have been missed in individual studies, and highlighted cell cycle regulators as promising therapeutic targets amenable for relapsed childhood B-ALL.

In the past decades, microarray has been used widely to investigate differentially expressed genes and dysregulated pathways underlying cancer pathogenesis. Numerous microarray gene expression studies on pediatric ALL have been performed, with few focused on understanding the biological mechanisms underlying relapsed ALL using matched diagnosis-relapsed samples. Also, each published dataset was relatively small (n < 50) and the concordance of these studies is rather low based on the publication findings [7–9] or even with the re-analysis on individual dataset using the limma method (Fig. 1; Additional file 1: Table S1). The discrepancies could be attributed to the small size in each single dataset which is underpowered to identify reliable candidates of interest. Hence, meta-analysis which merges all qualified datasets into a single analysis using a more robust statistical method is preferable to yield more meaningful set of differentially expressed genes and to provide new insights into the biological mechanisms. Meta-analysis on multiple microarray datasets of various diseases has yielded reliable candidates of interest by increasing the statistical power and generalizability [11–13].

Our meta-analysis demonstrated that S100A8 was the top gene upregulated in relapsed ALL as compared to matched diagnosis. S100A8 is a member of the S100 multigene family of cytoplasmic EF-hand Ca2 + -binding proteins [23] and was found overexpressed in various cancer types, and is involved in regulating cell proliferation, metastasis and apoptosis [23–27]. In hematological cancers, S100A8 has been reported to be overexpressed in childhood AML and associated with a worse prognosis [28, 29]. It may be involved in mediating chemoresistance by upregulating autophagy in leukemia cells through promoting the formation of BECN1-PI3KC3 complex [30]. Also, S100A8 was found overexpressed in the more aggressive ALL subtype, infant B-ALL, as compared to non-infant B-ALL [31], and mediated prednisolone-resistant in MLL-rearranged infant ALL [32]. Preclinical study has demonstrated S100A8 promoted cell growth of murine B-cell leukemia (BJAB) and human T-cell leukemia (Jurkat) lines [33]. Numerous studies have shown inhibition of S100A8 as a viable treatment strategy for cancers, including leukemia [28, 34–37]. For instance, inhibition of S100A8 has shown increased drug sensitivity and apoptosis of leukemic cells [28]. Given that S100A8 acts as an upstream target of EGFR signaling [38], anti-EGFR therapies, including midostaurin, enzastaurin and gefitinib has been proposed as potential therapy for kidney cancer cells which overexpressed S100A8 [35]. Moreover, increased expression of S100A8 mediated the activation of MAPK and NF-κB pathways, and treatment with p38 MAPK inhibitor SB203580 and the NF-κB inhibitor Bay 11-7082 effectively abolished migration and invasion of cancer cells [39]. Other than conferring selective sensitivity to drugs which target mediators of S100A8, the knockdown of S100A8 expression with siRNA or shRNA also showed reduced invasinesss and migration of cancer cells [28, 34, 36, 37]. Taken together, S100A8 is an ideal target for relapsed ALL therapy, and warrants further investigation.

MPO appeared as the second top ranked upregulated genes, with a fold change > 2. MPO has been long considered as the hallmark marker for AML cells by the French–American–British and WHO classifications, and has been used clinically to distinguish between AML and ALL. However, several studies reported MPO also being expressed in B-ALL cells, and associated with poorer prognosis [40–43]. For instance, infant B-ALL, a subtype which associated with poorer prognosis was shown to have overexpressed MPO, with an incidence rate of 40–60% [42, 44]. Also, B-ALL patients who presented with MPO-positive showed higher incidence of relapse [45], and reduced long-term survival [46]. Our data therefore suggested that MPO may serve as strong indicator for relapse in B-ALL patients. Moreover, silencing of MPO has been shown to effectively induce apoptosis in ovarian cancer cell lines by increasing caspase-3 activity [47]. Inhibition of MPO-overexpressed cells is therefore of clinical interest.

To date, development of cell cycle inhibitors for cancer therapy is actively ongoing. The most attractive inhibitors are those that target cell cyclin dependent kinases (e.g. CDK1) and aurora kinases (e.g. AURKA, AURKB), which are abundantly expressed in various cancer types. Our meta-analysis and several earlier studies have demonstrated that overexpression of cell cycle proteins was prominent and was among the key genetic changes underpinning progression of relapsed childhood B-ALL [7–9]. From the top 100 upregulated genes list, 14 of them are cell cycle regulators and are found to be interactive with each other (Fig. 5). Of those candidates, CDK1 appeared as a key target. To date, numerous CDK inhibitors have entered into clinical trials (https://​clinicaltrials.​gov), and have shown promising clinical response in leukemia patients. For instance, AML patients treated with a combination of flavopiridol and two chemotherapeutic agents, cytarabine and mitoxantrone, showed a complete remission rate of 75% [48], as compared to 40–50% with regimens using only conventional chemotherapy [49, 50]. Also, Dinaciclib, a novel inhibitor of CDKs 1, 2, 5, and 9, has been shown to be effective in CLL patients and induced lesser myelosuppression [51]. Recently, the approval by FDA on the use of a CDK inhibitor, palbociclib, in combination with letrozole to treat advanced estrogen positive, HER2 negative breast cancer has strengthen the usefulness of CDK inhibitors as new class of anti-cancer therapies [52]. In pediatric ALL, incorporation of CDK inhibitors into standard treatment regimens is yet to be investigated, and it is believed that clinical trials of CDK inhibitors on relapsed childhood B-ALL may be justifiable options to improve patients’ survival rate.

Another candidate of cell cycle regulators, AURKA, was also found in the top 100 upregulated genes list in our meta-analysis. AURKA is one of the three aurora kinases (AURKA, AURKB, and AURKC) which play essential roles in cell proliferation, regulating cell cycle transit from G2, formation of the mitotic spindle, centrosome maturation and separation, and cytokinesis [53–55]. Overexpression of AURKA has been documented in solid tumors and hematological cancers [56–60]. Higher levels of AURKA expression were correlated with higher tumor grade, and poorer prognosis [61–64]. Furthermore, overexpression of AURKA mediated resistance to gefitinib, taxol and cisplatin in cancer cells [65–67]. Inhibition of AURKA has been shown to increase cisplatin-induced apoptosis [66]. It is noteworthy that more than 30 AURKA inhibitors have been tested in clinical studies [68]. For relapsed and refractory AML patients, an early phase I/II clinical trial on AURKA inhibitor, MLN8237, has shown 13% complete response rate, 11% partial response rate, and 49% stable disease [69]. Given that the levels of AURKA expression was elevated in relapsed pediatric B-ALL, it would be worthwhile to investigate the efficacy of AURKA inhibitor in this group of patients.

Earlier studies have identified survivin overexpression as a strong risk factor for relapse in childhood B-ALL [70]. Independent microarray studies using other analysis pipelines have reported survivin as a key gene upregulated in relapsed ALL [7, 8]. Our analysis has strengthened the fact that targeting survivin is a promising therapeutic strategy, and warrants further investigation. Survivin is part of the AuroraB-survivin-INCENP-Borealin/Dasra B complex, an essential component for cell-cycle progression and cytokinesis [71]. It plays an important role in regulating cell proliferation and apoptosis suppression. Survivin was also found to be overexpressed in adult AML and T-cell leukemia [72, 73] as well as childhood AML [74–76], and associated with poorer survival outcome. Upregulation of survivin is mediated by multiple signaling pathways and by the tumor microenvironment including PI3K, MAPK, STAT3, Wnt/-catenin, hypoxia, angiogenesis, and NF-kβ signaling pathways [53, 76–80], hence may serve as an important target for leukemia therapy. Survivin also mediates resistance to chemotherapeutic agents, including vincristine, cisplatin, and tamoxifen in tumor cells [81–83]. Down-regulation of survivin via antisense oligonucleotides was shown to enhance sensitivity of various cancer cell types to cytotoxic agents such as TRAIL [84], cisplatin [85], taxol [86], imatinib [87], as well as to cytotoxicity induced by radiation therapy [88]. To date, several clinical trials on survivin employing different approaches including antisense oligonucleotides, small molecule inhibitors and immunotherapy are in progress ([89–92], http://​www.​clinicaltrials.​gov), and is offered as an treatment option for terminally ill relapsed B-ALL patients within in the context of clinical trial.

Taken together, our meta-analysis on paired diagnosis-relapsed B-ALL has strengthened the evidence for the roles of cell cycle dysregulation as the key component of genetic alterations underpinning disease progression, and can be considered as the promising pathway for new therapeutic intervention. The efficacy of targeted cell cycle therapies to treat relapsed pediatric B-ALL patients shall be further evaluated in the context of clinical trials.