Expression profile and specific network features of the apoptotic machinery explain relapse of acute myeloid leukemia after chemotherapy

AML is heterogeneous both at the cytogenetic and the molecular level [1, 2]. Many of these alterations have prognostic impact on the clinical outcome, especially on resistance to chemotherapy and relapse rates [1, 2]. As with other tumours, one of the main obstacles to successful chemotherapy is drug resistance [3, 4]. Nearly 80% of AML patients apparently achieve CR following chemotherapy induction, however a high proportion of them relapses and eventually dies of the disease [5, 6]. To explain MRD or relapses in these patients after chemotherapy, it has been suggested that the bone marrow microenvironment may protect cancer hematopoietic cells, allowing them to resist treatment and survive [7]. This form of resistance (called de novo drug resistance) could be in part attributable to the interaction of the Very Late Antigen (VLA)-4 of leukemic cells with fibronectin within bone marrow stroma [8]. VLA-4 and other adhesion molecules are involved in homing and mobilization of hematopoietic stem cells (HSC) [9]. Treatment strategies, based on blocking adhesion molecules or other proteins involved in homing, might minimize MRD [10]. It has been demonstrated that patients that are good mobilizers are significantly more likely to be responsive to chemotherapy than poor mobilizers [10]. However, unlike other leukemic patients, good mobilizers CR AML patients are subject to significantly higher rates of relapse than patients with lower mobilization capacity; they also have a higher rate of MRD at remission [11‐13]. A high ability to mobilize CD34+ cells in peripheral blood (PB) was proved to be an unfavourable prognostic factor, independent of others such as class of cytogenetic risk or FAB morphotype [11‐13]. It has been hypothesized that intrinsic or acquired chemoresistance of leukemic bone marrow precursors may be a possible explanation for the association between high mobilization of CD34+ cells and MRD. Furthermore, it has been demonstrated that non-leukemic HSC from AML patients show differential chemosensitivity: this allowed the identification of three categories of patients, characterized by: (i) normal chemosensitivity, (ii) chemoresistance, (iii) increased chemosensitivity, respectively (Milone et al: Chemo-sensitivity of clonigenic marrow precursors in AML patients in complete remission: association with CD34+ mobilization and with disease-free survival, submitted). A statistically significant correlation between in vitro chemosensitivity of CFU-GM to Etoposide and Mafosfamide, CD34+ cells mobilization and survival was detected: patients with high mobilization capacity showed chemoresistance of their BFU-E and CFU-GM and lower survival rates than patients with normal or low mobilization activity. Dysregulation of the Apoptotic Machinery (AM) plays a primary role in the response to antineoplastic therapy [14, 15]. In a previous work by some of us, it was demonstrated that the highly interconnected nodes of the AM network (hubs) typically represent the genes with the highest number of genome, transcriptome and proteome alterations in several cancer models [16]. Moreover, we found that most of AM proteins targeted by drugs were characterized by high connectivity; in particular, there was a highly significant association between the betweenness of these proteins and their being targets of drugs [16]. Based on these findings and on our data on AM involvement in anticancer treatment (i.e., Fenretinide in neuroblastoma and Imatinib in CML) [16], we explored the hypothesis that specific AM expression profiles in CR AML patients may determine the differential chemosensitivity, shown in vitro by their bone marrow HSC. The results allow us to propose that there is a relationship between the AM network specific structure and its proneness to dysregulation.

The transcriptome of 84 human genes, assigned to the Apoptotic Machinery [17], was analyzed in bone marrow CD34+ cells from a cohort of AML patients in apparent CR and from normal controls: in all of them, BFU-E, CFU-E, CFU-GM hematopoietic clonogenic precursors were checked for in vitro chemoresistance. Chemosensitivity tests on freshly collected cells and molecular analysis after cell cryopreservation were performed on the same bone marrow samples. AML patients were selected from a cohort of 37 patients studied prospectively, in which the HSC in vitro chemosensitivity was determined. Based on their in vitro drug sensitivity in comparison to normal controls, patients were divided into three groups: (i) normally chemoresponsive, (ii) chemoresistant, (iii) highly chemosensitive. Three patients from the second group and three from the last one were selected for further detailed molecular analysis.

Drug resistance is a major cause of failure in cancer treatment [7, 36]. Chemotherapy exerts on tumour cells a strong selective pressure: accordingly, their survival relies on a dynamic mutation process, possibly leading to an ever fitting chemoresistance phenotype [37]. The bone marrow microenvironment is a secluded and potentially safe body niche, in which cancer cells can be protected against chemotoxic agents. Mobilization from this compartment is a complex process, that requires the orchestrated participation of several molecules as chemokines, adhesion molecules, and their downstream targets [10, 38]. The molecular events that regulate HSCs' engraftment and mobilization are still incompletely defined. The precise mechanism of mobilization by the Colony Stimulating Factors (CSF), the factor most commonly used to mobilize hematopoietic cells from bone marrow to PB, remains unsatisfactorily characterized [39, 40]. CSF binding to its receptor leads to the activation of several downstream signalling cascades affecting cell survival, proliferation, differentiation, migration [41]; moreover, it is well known that CSF suppresses apoptosis in both normal HSCs and cancer cells [42]. In our experiments, we observed that CR patients with high CD34+ mobilization activity showed in vitro chemoresistance of their BFU-E and CFU-GM. This was coupled to a poorer prognosis, due to an increased relapse rate respect to normally or highly chemosensitive patients. Intriguingly, chemoresistant patients showed an AM expression profile that strongly suggested the involvement of the AM network in their in vitro drug resistance. Thus, a relationship between mobilization from bone marrow, apoptosis induction and chemoresistance seems to exist in these patients. The chemoresistant class is characterized by up regulation of Birc genes (BIRC2 and BIRC3), that inhibit the action of caspases, and down expression of proapoptotic genes as APAF1 (an apoptosome component), FAS, FASL, TNFRSF25, TNFSF10. This molecular phenotype could lead to a failure in activating apoptosis in response to chemotherapy [30‐34]. It is well known that GM-CSF inhibits Fas-induced apoptosis and stimulates expression of BIRC family members [43, 44]. This protective effect of CSF can also explain the lower complete remission rates after chemotherapy in AML patients, whose cells have a higher responsiveness to hematopoietic viability factors in vitro [45]. Our data demonstrate a down regulation of proapoptotic genes CD40LG, CIDEB, TNFSF10 in chemoresistant AML patients, previously not reported in association with drug resistance. Furthermore, differently from previous reports that in vivo resistant AML cells express high levels of BCL2A1 and BCL2L1 [30, 34], we found that both genes are down regulated in chemoresistant AML patients in comparison to those from the chemosensitive class. We suggest that this discrepancy is due to the specific features of the different cell types analyzed: we analyzed in vitro selected chemoresistant, possibly non-leukemic CD34+ cells from CR AML patients, whereas Eisele et al. and Valdez et al. studied myeloid blasts from AML patients [46] or mononuclear cells from AML patients [31], respectively. Why are some CR AML patients good responders to CSF and others are not? The continued pressure of chemotherapy may induce or select genomic mutations, able to alter the cytokine cross-talking network between cell migration and apoptosis: this would result in an acquired drug resistance phenotype [8]. We may otherwise assume that within the HSC population of some CR patients, few leukemic cells survive chemotherapy and persist under the CR threshold to be eventually selected by the mobilizing treatment [47, 48]. The presence of a substantial number of MRD cells could influence the microenvironment and enhance their protective effect from chemotherapy damage. This would explain not only the high relapse rate, but also the high mobilization activity through a reduced myelotoxicity induced by chemotherapy. Based on our results, we suggest that the presence of MRD can lead to a modulation of BM microenvironment that could cause dysregulation of some components of the AM network in HSC. Interestingly, by comparing both the resistant and the sensitive class with control samples, we noticed that 23 AM genes showed an expression profile common to both resistant and sensitive classes and different respect to controls: many of these genes are involved in an increased response to chemotherapy (e.g., the down regulation of BAG3, BIRC4, BIRC8, BNIP family, CFLAR) [49, 50]. Furthermore the downregulation of DAPK1, TNFRSF11B, TNFSF8, TP73 could be considered as a CR marker: these genes are commonly over expressed in hematopoietic malignancies, but are down regulated in patients after chemotherapy [24‐26]. The detection of proapoptotic BCL10 in CR CD34+ cells, but not in normal donors (Tables 1 and 3; Figure 1), could be related to their proneness to death induction. Dysregulation of this gene is known to be involved in some types of haematological malignancies [51]. These observations suggest that non-leukemic CD34+ cells from CR AML patients have an expression profile apparently prone to apoptosis and to a positive response to chemotherapy. This molecular feature satisfactorily explains their positive response to in vivo chemotherapy induction, leading these patients to the first apparent CR. Plausibly, the molecular differences observed between chemoresistant and chemosensitive (acquired de novo mutations or residual presence of leukemic cells) could be responsible for the de novo acquired resistant phenotype, and accordingly for the relapse. Unsurprisingly, these differentially expressed genes are characterized by high centrality within the AM network. Furthermore, many published data link these genes to drug response, suggesting that they perform a critical role within AM signalling activated by pharmacological treatments. Taken together these data strongly suggest that the topology of AM network is strictly related to its proper biological functioning. Malfunctioning of these central nodes affects the stability of the network and profoundly modify the physiological cell behaviour. In agreement with other published data, our results suggest that tumour-related defects in AM hubs are preferentially selected [16, 52]. The functional impairment of a few nodes, which control directly or indirectly the activities of many others in the context of the co-occurrence of multiple genetic defects, could represent a selective advantage during neoplastic transformation and in response to pharmacological treatment. The differential expression of topologically important AM nodes in CR AML patients could seriously impair the physiological equilibrium of AM. CD34+ cells from both classes of CR AML patients showed AM network features that make these cells prone to apoptosis: accordingly, it is not surprising that these patients were in apparent CR after the first cycle of chemotherapy induction and consolidation (Figure 3). On the other hand, the AM network of the resistant class showed some critical features respect to that of the sensitive class: by impairing proapoptotic components and activating prosurvival nodes, these differences could reduce the ability of these cells to appropriately respond to death stimuli (Figure 3). Some of these AM genes are known to be dysregulated in leukaemia: up regulation of BIRC3 and down regulation of APAF1, CIDEB, FAS, TNFRSF25, TNFSF10 were previously identified by our group as specific alterations of AM genes in leukemic cells [16].

The ability of some cancer cell types to elude pharmacological apoptosis induction is apparently based on molecular mechanisms, similar to those involved in escaping physiological cell death. In this scenario , some AM genes represent critical nodes for the proper response to chemotherapy. Their dysregulation in HSCs from CR AML patients could be related to their high mobilizing ability, in vitro chemoresistance and high relapse rates. This molecular phenotype could result from de novo mutations, selected by the treatment, or be due to residual leukemic cells, positively selected by the cytokines used for mobilization. The AM expression profile of CD34+ cells seems to discriminate CR AML patients from normal controls, as well as in vitro chemoresistant CR AML patients from those who are sensitive. These differences affect some critical nodes of the AM network and could represent one of the causes of the differential in vitro resistance of these cells. Accordingly, AM profiling before chemotherapy and transplantation could allow the identification of patients with a genotype highly predisposing to relapse, in order to treat them by different, specifically designed, therapies. Specifically, low expression levels of APAF1, CD40LG, CIDEB, FAS, FASL, TNFRSF25, TNFSF10 and up regulation of BIRC2 and BIRC3 could pinpoint patients who are prone to relapse. Otherwise, the characterization of the genotype of the cells from PB could demonstrate the characteristic molecular signature of leukemic cells and accordingly guide the design of the therapeutic strategy.