The Stat3/5 Signaling Biosignature in Hematopoietic Stem/Progenitor Cells Predicts Response and Outcome in Myelodysplastic Syndrome Patients Treated with Azacitidine

This article is featured in Highlights of This Issue, p. 1829

The introduction of azacitidine has radically transformed the therapeutic approach and improved the outcome of patients with high-risk myelodysplastic syndrome (MDS). Nevertheless, the exact mechanism of action remains to be established, while both primary and secondary resistance confer a grave prognosis, as there are currently no effective alternative therapies (1). In addition, there is lack of a serviceable, widely accepted biomarker of response and/or outcome that can offer a timely and valid estimation of the expected benefit from azacitidine and help to tailor treatment (2, 3).

Stat3 and Stat5 regulate fundamental cellular processes and aberrant cell signaling via Stat3/5 is implicated in leukemogenesis (4–9). Constitutive upregulation of Stat3 and, less often, Stat5 molecules has been reported in acute myeloid leukemia (AML), but its prognostic impact is contentious (10). This is because measurement of basal Stat3/5 levels in heterogeneous cell populations by using conventional proteomic assays does not address the cytokine regulation of signaling cascades of malignant hematopoiesis and thus cannot portray the overall picture of signaling events at the single cell level (11). Single cell network profiling using multiparametric phospho-specific flow cytometry can identify phosho-Stat3/5 biosignatures at the hematopoietic stem/progenitor cell (HSPC) level which reflect the biologic behavior of AML and can distinguish patient subgroups with worse prognosis (11–13). In their seminal article, Irish and colleagues (12) have identified a G-CSF–inducible Stat3/5 double-positive (DP) subpopulation of CD34⁺ cells in aggressive AML, without however testing its prognostic value. An analogous DP subset has also been observed in Ph⁻ myeloproliferative neoplasms (14), raising the possibility of a shared signaling biosignature among myeloid neoplasms.

Both aberrant methylation and cell signaling deregulation contribute to the pathogenesis of Myelodysplastic syndromes (MDS), while epigenetic defects of genes involved in cell signaling are frequently encountered in MDS patients, particularly in the late stages of the disease (15–18). Moreover, in addition to the bidirectional interplay among the epigenetic machinery and cell signaling (19), hypomethylating agents may indirectly affect signal transduction (20–22). Despite the above, a comprehensive view of Stat3/5 signaling alterations in late-stage MDS is missing and the effect of hypomethylating therapy on leukemic signaling has not been addressed yet.

By using phospho-specific flow cytometry, we investigated phospho-Stat3/5 signaling profiles in the HSPCs from 74 MDS and low blast count AML patients during azacitidine therapy. We show that the pretreatment Stat3/5 biosignature in MDS HSPCs was strongly associated with response status and patient outcome. We further identified a G-CSF–inducible phospho-Stat3/5 DP subpopulation in the CD34⁺ cell compartment (hereafter referred to as DP subset) whose pretreatment levels were inversely associated with response. The cellular hierarchy and the molecular properties of the DP subset were consistent with a leukemia stem cell phenotype, whereas its kinetics followed the disease course and response to treatment.

Following Institutional Review Board approval, peripheral blood and bone marrow mononuclear cells from 74 patients and 10 donors with nonclonal myelopoiesis (i.e., lymphomas, solid tumors, and immune thrombocytopenia) were obtained before treatment initiation and at the indicated time points. Informed consent was obtained in accordance with the Declaration of Helsinki. All patients received azacitidine in a nonclinical trial setting at an initial dose of 75 mg/m² s.c. for 7 days on 28-day cycles. Dose reductions of 25% to 50% and/or treatment delays were considered for severe myelotoxicity or myelosuppression-related complications. Granulocyte colony-stimulating factors (G-CSF) were used at the discretion of the treating doctor, whereas no erythropoiesis-stimulating agents were administered to any patient. Response to therapy was evaluated using the International Working Group Response Criteria for MDS (23). Heavily transfused patients were defined as those requiring ≥ 4 RBC units/8 weeks (24). Mononuclear cells were isolated after density centrifugation, frozen in liquid nitrogen, and processed within 6 months after cryopreservation.

The following antibodies were used: CD34 (clone 8G12), pStat3 (Y705), pStat5 (Y694), CD2 (RPA-2.10), CD3 (HIT3a), CD4 (RPA-T4), CD8 (RPA-T8), CD19 (HIB19), CD20 (2H7), GPA (GA-R2), CD45(2D1), CD114 (LMM741), p53 (DO7) and Ki-67 (B56) all from BD Biosciences; CD38 (LS198.4.3) from Beckman Coulter; Bcl-2 (124) from Dako; CD123 (6H6), CD90 (5E10) and CD45RA (HI100) from Biolegend. Fluorescence minus one (FMO) was employed as negative control. Data were acquired on a 5-color FC-500 (Coulter) and a 4-color FACSCalibur (BD Biosciences) cytometers and analyses were performed with Flowjo software (Treestar).

Thawed cells were washed once in RPMI to remove the residual DMSO and allowed to rest for 1 hour in serum-free RPMI medium at 37°C. Cells were then distributed at 2–5 × 10⁵ cells/well in 4 aliquots. Two aliquots remained unstimulated and used as FMO control and untreated sample controls and the others were stimulated for 15 minutes at 37°C with either human recombinant G-CSF or granulocyte macrophage stimulating factor (GM-CSF, Miltenyi Biotec GmbH) at a final concentration of 20 ng/mL for both cytokines. Stimulation was halted by fixation with Cytofix Fixation Buffer (BD Biosciences) and cells were permeabilized with Perm Buffer III (BD Biosciences) and stained with phospho-Stat3 (clone Y705), phospho-Stat5 (clone Y694), and combinations of the aforementioned antibodies for 30 minutes at room temperature. Basal phosphorylation levels were expressed as the log2 ratio of mean fluorescence intensity (MFI) of unstimulated pStat3 and pStat5 divided by the FMO control, namely log2[MFI (unstimulated)/MFI (FMO)] and potentiated levels as log2[MFI (stimulated)/MFI (unstimulated)].

To map the cellular hierarchy of the CD34⁺ DP subset, cells expressing mature lineage markers were depleted from mononuclear cells after staining with an antibody cocktail consisting of anti-CD2, CD3, CD4, CD8, CD19, CD20, and GPA (25) and anti-PE immunomagnetic MicroBeads (Miltenyi Biotec). Purified Lin⁻ cells were then subjected to positive selection of CD34⁺ cells by means of CD34 microbeads (Miltenyi Biotec). Isolated Lin-CD34⁺ cells (purity always ≥ 98%) were then stained with pStat3, pStat5, CD38, CD90, CD123, and CD45RA according to the above phospho-flow protocol and analyzed as previously described (25). For the characterization of the molecular/LSC properties of the CD34⁺ DP subset, BMMNC were stained with pStat3, pStat5, and CD34 along with one of bcl-2, Ki-67, or p53.

For the determination of CSF3R protein expression samples were analyzed by flow cytometry, by staining with CD114, CD34, CD45, and the corresponding isotype controls. Data are expressed as the ratio of the CD114 MFI of CD34⁺ cells to the MFI of isotype control. Immunomagnetically purified CD34⁺ cells (purity ≥ 98%) from the above samples were used for mRNA extraction (RNAqueous Micro kit, Ambion) and reverse transcription (RETROscript, Ambion). Quantification of the CFS3R mRNA was performed by Real-Time PCR using SYBR Green (Invitrogen) and the following primers: human CSF3R forward, 5′-CATCACAGCCTCCTGCATCATC-3′, human CSF3R reverse, 5′-CTGAAGCTCTGCTCCCAGTCTC-3′, human GAPDH forward, 5′-ACTCCACGACGTACTCAGCG-3′, human GAPDH reverse, 5′-GGTCGGAGTCAACGGATTTG-3′. PCR conditions were as following: 50°C for 2 minutes, 95°C for 10 minutes, 40 cycles of 95°C for 15 seconds and 60°C for 1 minute, followed by melting curve analysis from 65°C to 90°C. Reactions were carried out in duplicate in a PTC 200 Peltier Thermal Cycler with Chromo4 Real-Time PCR Detector. Data acquisition and analysis were performed by Chromo4 Real-Time PCR Detector and Opticon Monitor 3. Relative CSF3R expression of CD34⁺ cells was calculated by 2^−ΔΔC_t method.

DNA was extracted from bone marrow mononuclear cells or peripheral blood samples collected before the initiation of azacitidine. Mutational analysis of the coding region of TET2 and TP53 genes in 30 and 11 samples, respectively, was performed using next-generation sequencing (detailed method provided in the Supplementary Methods). For the analysis of TET2, a 10% cutoff of allele fraction was used as suggested previously (26).

Comparisons were performed by using χ², Mann–Whitney, Kruskal–Wallis, Wilcoxon signed-rank, and Friedman tests, as appropriate, and survival analysis with Kaplan–Meier and log-rank test. Overall survival (OS) was defined as the time from azacitidine initiation to death from any cause and event-free survival (EFS) as the time from azacitidine initiation to disease progression, relapse, or death. Surviving patients were censored at last follow-up. Multivariate survival analysis was based on Cox proportional hazards model using a backward stepwise selection procedure with entry and removal criteria of P = 0.05 and P = 0.10, respectively. Multiple Experiment Viewer software (MeV, http://sourceforge.net/projects/mev-tm4/) was employed for unsupervised hierarchical cluster analysis with complete linkage algorithm and Euclidean distance as distance metric (12).

Patients' characteristics are listed analytically in Table 1. Two (2.7%) patients achieved marrow complete remission (mCR) with both erythroid and platelet hematologic improvement, 17 attained CR (23%), 11 (14.9%) patients showed hematologic improvements only in platelet count (HI), 17 (23%) remained stable (stable disease, SD) and 27 (36.4%) failed (failure, F) azacitidine. The representativeness of our cohort was validated by the successful application of the prognostic score proposed by Itzykson and colleagues (24) in 72 evaluable patients. Three groups with different OS (P = 0.027) were identified (Supplementary Fig. S1). Unsupervised clustering of pretreatment signaling profiles in CD34⁺ cells of MDS patients identified 2 signaling clusters (SC), SC-1 and SC-2. The two clusters displayed similar levels of constitutive Stat3/5 phosphorylation, whereas potentiated responses of Stat3/5 to G-CSF and GM-CSF stimulation were weak in SC-1 and powerful in SC-2 (Fig. 1A). No differences were observed among the two clusters regarding age, sex, WHO subtype, transfusion burden, WPSS, IPSS, IPSS-R, and TET2 mutation status (Table 1). In contrast, patients in SC-1 achieved better response to azacitidine (P = 0.01), had worse cytogenetics both by IPSS (P = 0.01) and IPSS-R (P = 0.02) and enjoyed longer median EFS (12.5 vs. 7.8 months, respectively, P = 0.01) than the ones in SC-2 (Fig. 1 and Table 1), while median OS was also prolonged in patients of SC-1, without, however, reaching statistical significance (13.5 vs. 10.4 months, respectively; P = 0.08). Multivariate analysis confirmed the independent prognostic power of the pretreatment Stat3/5 signaling biosignature for EFS (P = 0.017), whereas heavy transfusion requirements was the other independent prognostic factor for both OS (P = 0.004) and EFS (P = 0.029, Supplementary Tables S1 and S2). These findings are in line with prior observations in AML (12, 27), and strongly suggest involvement of aberrant signaling via Stat3/5 in MDS pathobiology. We further performed unsupervised clustering of patient and nonclonal samples (Supplementary Fig. S5). All patients with nonclonal myelopoiesis clustered with responding patients, further supporting the existence of an aberrant signaling biosignature in nonresponders to azacitidine.

The complementary cytometric analysis of pretreatment Stat3/5 signaling profiles in our patients revealed an identical to the G-CSF–inducible DP subpopulation previously described in AML (12) and Ph⁻ myeloproliferative neoplasms (14) The median pretreatment levels of the DP subset were significantly lower in patients who achieved CR or marrow CR (38.6% of total CD34⁺ cells, range 0.13%–83.5%) compared to those with stable disease (75.4%, 1%–90%, P = 0.008) and failure to azacitidine (74.7%, 12.1%–90%, P = 0.006), whereas patients with HI did not show any significant differences with the other groups (HI, 55.3%, 0.5%–88.2%, Fig. 2A and B). Also, the levels of the DP subset were significantly lower in patients with poor-risk cytogenetics by IPSS (38.6%, 0.5%–75.4%) compared to those with intermediate (68.1%, 1%–85.5%, P = 0.02) and good-risk (66.9%, 0.13%–90%, P = 0.005) karyotype. Similarly, poor-risk cytogenetics by IPSS-R were associated with lower percentage of the DP subset (38.4%, 1%–75.4%) compared with good-risk disease (70.8%, 0.13%–90%, P = 0.03). Of note, consistent with the results obtained by the clustering of signaling profiles, nonclonal patients had identical levels of the DP subpopulation with responders to azacitidine (CR and HI), but significantly lower compared with nonresponders (F and SD, Supplementary Fig. S5).

No differences in the levels of the DP subset were observed regarding age, gender, MDS subtype, TET2 mutation status, and transfusion requirements (data not shown). We also tested the rest GM- and G-CSF–inducible or not CD34⁺ subpopulations, for example, single positive Stat3 or Stat5 as well as Stat3/5 DP subsets, for correlations with clinical and biologic parameters, but we were unable to find any associations except significantly higher basal Stat3 levels in patients with poor-risk cytogenetics compared with the good-risk ones by both IPSS (P = 0.002) and IPSS-R (P = 0.008, Supplementary Fig. S2).

We next sought to investigate the effect of azacitidine and disease course upon the kinetics of the DP subset. The latter subpopulation was downregulated significantly on day15 of the first cycle only in patients with CR. The pretreatment median percentage in these patients (38.6% of total CD34⁺ cells, range 0.13%–84%) downregulated to 9.6% (0.3%–70.4%, P = 0.01), whereas patients with HI (47.6%, 0.5%–82.2% changed to 32%, 0.2%–72.2%, P = 0.12), SD (75.4%, 1%–87.6% changed to 76.1%, 1%–83.7%, P = 0.23), and failure (63%, 12.1%–87.2% changed to 58%, 23.8%-87%, P = 0.26), retained unaltered levels of the DP subset (Fig. 3A). Of note, day 15 of the first cycle was chosen because the hypomethylating effect of azacitidine usually peaks 15 days after its administration (18), while clonal hematopoiesis still dominates in bone marrow as shown by the identical percentage of bone marrow CD34⁺ cells before (15%, range 4%–29%) and 15 days after first azacitidine administration (14.5%, 4%–26%, P = 0.9) in our patients.

In 19 patients, the alterations of the DP subset were studied longitudinally during the disease course (Fig. 3B and C). Measurements were performed on days 0 and 15 of the first cycle, at response evaluation after 6 cycles and at disease progression or relapse. The levels of the DP subset remained unaffected throughout the disease course in patients with SD and failure to azacitidine. In contrast, the DP subset in patients achieving CR was significantly reduced on day 15 of the first cycle and remained at low levels until disease relapse, which was accompanied by a marked expansion of the DP subset (P = 0.007). Similar kinetics of the DP subset were observed in patients with HI, without however, reaching statistical significance (P = 0.13).

Thus, it appears that the kinetics of the DP subset are following the disease course and response to azacitidine, indicating potential involvement of the former subset in mechanisms underlying disease progression and resistance to azacitidine.

Recent findings in both mice and humans challenge the leukemia stem cell (LSC) model and suggest that in approximately 90% of CD34⁺ AML cases LSCs reside in the lymphoid-primed multipotent progenitor (LMPP)-like and granulocyte-macrophage progenitor (GMP)-like compartments (25, 28). Interestingly, the same CD34⁺ cell compartments are clonally expanded in late-stage MDS (29). By combining surface with phospho-staining in purified Lin⁻CD34⁺ cells of 5 patients, we assessed the cellular hierarchy of the DP subset (Fig. 4A and Supplementary Fig. S3). We observed that, in comparison with the other major signaling subset, namely the CD34⁺ G-CSF–unresponsive, Stat3/5 double negative (DN) cells of the same patient, the DP subset was enriched in LMPP-like (70.8%, range 46%–98.5%, vs. 43%, 33%–69%, respectively) and GMP-like cells (79.5%, 53%–98% vs. 43%, 25%–76%), whereas it contained less multipotent progenitor (MPP)-like (28%, 1.6%–44% vs. 52%, 28.7%-60%), common myeloid progenitor (CMP)-like (19%, 0.9%–41% vs. 44%, 6.1%–57%), and megakaryocyte-erythroid progenitor (MEP)-like cells (1.5%, 0.1%–6.4% vs. 8.5%, 0.6%–14.5%, P = 0.04 for all comparisons, Fig. 4B).

Bcl-2, p53, and Ki-67 are well established molecular indicators of resistance to apoptosis (30), oncogenesis (31), and cellular proliferation (32), respectively, and were therefore used to address the molecular properties of the DP subset. After we confirmed the compatibility of the phospho-flow protocol with the intracellular staining for the above markers (not shown), we found that the DP subset, compared with the DN one, exhibited decreased pretreatment levels of Ki-67 (52%, 22%–81.7% vs. 66.5%, 33%–87%, respectively, P = 0.01) and increased Bcl-2 MFI (9.5, 6.9–25.5 vs. 6.1, 3.7–24.1, P < 0.001) and p53 MFI (3.67, 2–14.3 vs. 2.7, 1.6–8.7, P = 0.03), indicating quiescence and increased antiapoptotic and oncogenic properties, respectively (Fig. 5A and Supplementary Fig. S4; refs. 33, 34). Of note, the DP subset displayed significantly increased p53 levels compared with the DN one in both p53 mutated (n = 3) and unmutated cases (n = 8, data not shown), whereas the expression of the above molecules in both the DP and DN subsets remained unaltered on d15 after azacitidine initiation (Fig. 5A), suggesting that the above signaling subsets represent distinct cellular entities with stable characteristics.

Taken together, the increased levels of LMPP and GMP-like cells along with the high expression of Bcl-2 and p53 and the lower Ki-67 levels imply that the DP subset is enriched in cells with a LSC phenotype.

Myeloid blasts display variable expression of CSF3R and a heterogeneous response to G-CSF stimulation (12, 27, 35). Surface CSF3R could not be assessed concomitantly in the DP and DN subsets of G-CSF–stimulated CD34⁺ blasts because it is downregulated after G-CSF ligation (36). Therefore, to determine whether CSF3R levels and/or their modulation by azacitidine contribute to the interpatient variability and azacitidine-induced modifications of the DP subpopulation, we employed two different approaches. In the first, we evaluated CSF3R expression on 8 samples that showed either absence (<4%, DP^neg) or very high expression (≥87%, DP^high) of the DP subset, while in the second we measured CSF3R levels on day 0 and day 15 of the first cycle of azacitidine in 4 other patients who displayed significant downregulation of the DP subset on day 15. Using the first approach, we found that patients with either high or null expression of the DP subset showed identical pretreatment mRNA and protein levels of CSF3R (Fig. 5B). Likewise, CSFR3 protein and transcript expression remained unaltered despite the significant downregulation of the DP subset from 66.1% (range, 23%–80%) on day 0 to 45.5% (1%–59%, P = 0.008) on day 15 in 4 patients (Fig. 5C), indicating that quantitative changes of CSFR3 are not implicated in the generation of the G-CSF–inducible DP subset.

Abnormal hematopoietic stem/progenitor cell (HSPC) signaling via Stat3/5 is typically observed in leukemic hematopoiesis. In both adult and pediatric AML, functional phenotyping of Stat3/5 signaling networks by phospho-protein flow cytometry provides important prognostic information and pathobiologic insights (12, 27, 37). Yet, despite the reciprocal interactions of DNA methylation with Stat3/5 signaling (19, 29, 38), the high rate of mutations (39) and aberrant methylation of genes involved in cell signaling (18) in high risk MDS, no current study addresses the Stat3/5 signaling alterations at the single HSPC level in such patients. Only a recent study explored signaling abnormalities in MDS, but it was mainly focused on erythropoietin-induced Stat5 phosphorylation in erythroid progenitors in early disease stages (6). In the current work, we demonstrate an abnormal Stat3/5 signaling biosignature of HSPCs in high-risk MDS, which is amenable to modulation by azacitidine and can predict treatment response and outcome.

Two signaling clusters were identified by hierarchical clustering of pretreatment basal and potentiated Stat3/5 and Stat5 phosphorylation patterns. Patients in SC-1 displayed better response to azacitidine and longer EFS, whereas OS was also prolonged, though not statistically significant. Intriguingly, although correlated favorably with prognosis, SC-1 was inversely associated with cytogenetic risk and showed no correlation with TET2 mutation status, emphasizing that signaling profiles are not merely a surrogate for the underlying molecular abnormalities, but instead provide additional information. Moreover, the characteristic similarity of SC-2 in our study with the SC-P2 reported by Irish and colleagues in adult AML (12), which was also associated with disease resistance, highlights the biologic relationship of AML with high-risk MDS and suggests a common signaling biosignature of aggressive leukemic HSPCs in adult AML and high-risk MDS. In contrast, increased pStat3 response to G-CSF ligation correlated with superior outcome in pediatric AML, potentially reflecting the biologic differences among adult and pediatric myeloid malignancies (27). Notably, patients with nonclonal myelopoiesis and responders to azacitidine shared an identical Stat3/5 signaling biosignature characterized by little or no potentiated G-CSF responses, further corroborating the role of Stat3/5 signaling aberrations in azacitidine resistance.

Further analysis of Stat3/5 signaling profiles revealed a G-CSF–inducible Stat3/5 DP subpopulation of CD34+ cells whose levels correlated inversely with response to azacitidine. A phenotypically identical, chemoresistant subpopulation has been previously reported in adult AML and Ph⁻ myeloproliferative neoplasms (12, 14), but no further investigation of the kinetics and the molecular properties of the DP subset was conducted. We observed a remarkable downregulation of the DP subset in responding patients on day 15 after the first azacitidine administration, whereas nonresponders exhibited no changes. More important, the kinetics of the DP subset in responders mirrored those of the tumor burden and treatment response, as the former subpopulation remained at low levels during CR, whereas a parallel development of resistance to azacitidine with an expansion of the DP subset occurred. These findings strongly suggest, on one hand, that azacitidine can restore and partially control the pathologic Stat3/5 signaling in responding patients and on the other involvement of the DP subset in mechanisms underlying disease progression and azacitidine resistance. Of note, though the distinction of clonal from normal hematopoiesis is often problematic in MDS, the alterations of the DP subset pertained to clonal HSPCs as shown by two findings. First, the blast percentage on day 15 of the first cycle was identical to the pretreatment one and second, three responding patients (2 with CR and one with HI) who downregulated significantly the DP subset after 6 cycles had still abnormal karyotype indicating persistence of clonal hematopoiesis. In addition, it has been clearly demonstrated that even patients in CR have residual MDS HSPCs in substantial numbers (25, 29).

Although previous studies linked the DP subset to aggressive disease, there is currently no detailed phenotypic and molecular characterization of the former subpopulation. First, we confirmed the compatibility of the phospho-flow techniques with the intracellular measurement of Bcl-2, p53, and Ki-67 and the accurate assessment of HSPC subsets as has been previously shown (refs. 40, 41; Supplementary Fig. S3). We then observed a significant enrichment of the DP subset in LMPP-like and GMP-like cells compared with the other dominant CD34⁺ signaling subset, the G-CSF–unresponsive Stat3/5-negative subpopulation. Also, Bcl-2 and p53 levels were significantly higher in the DP subset, whereas Ki-67 expression was lower compared with the DN subset. Considering the negative role of Bcl-2 and p53 in the pathobiology of MDS (33, 42, 43), the downregulation of Ki-67 in CD34⁺ cells when MDS progresses to overt leukemia (33) and recent data regarding the hierarchy of LSCs (25), we surmise that the DP subset potentially possesses properties of leukemia-propagating cells, which may account for its association with poor risk disease and resistance to azacitidine.

Hypersensitivity of HSPCs to G-CSF may predispose to leukemic transformation and propagation via overexuberant Stat3/5 activation, but there is considerable interpatient variability of G-CSF responsiveness in AML (12, 44, 45). We found that the differential response of Stat3/5 to G-CSF in our patients is not related to the protein and mRNA levels of CSF3R, suggesting that other mechanisms such as qualitative defects of CFS3R, abnormal receptor-associated proteins, dysfunctional downstream signaling elements, or abnormalities of positive or negative regulators of signaling via Stat3/5 are potentially responsible for the variability in G-CSF sensitivity (46). Of note, azacitidine downregulated the DP subset only in responding patients, implying that diverse molecular defects are responsible for the induction of the DP subset in MDS patients, only a part of which is susceptible to epigenetic reprogramming by azacitidine.

In contrast to our results, Redell and colleagues observed a positive correlation of CSF3R expression with the magnitude of pStat3 induction in pediatric AML samples. However, no association with simultaneous potentiation of Stat3 and Stat5 with CSF3R levels was reported, whereas several samples with high levels of CSF3R failed to induce a Stat3/5 response. Moreover, the fact that CSF3R knockout mice can still mobilize effectively hematopoietic progenitors provide further support for a CSF3R-independent mechanism of induction of the DP subset in MDS patients (44).

Collectively, we report for the first time a disturbed Stat3/5 signaling architecture in high-risk MDS, which is amenable to modulation by azacitidine therapy in responding patients and its alterations parallel disease activity. Aside from furnishing critical insights in MDS biology, our findings have obvious translational implications. There is paucity of a serviceable biomarker of outcome in MDS patients treated with azacitidine, whereas the mechanisms of resistance to azacitidine are largely unknown and there is currently no effective treatment after azacitidine failure. The prognostic relevance of the Stat3/5 biosignature of MDS HSPCs in our study may help to identify which patients benefit most from azacitidine, while its alterations during the disease course can provide a tool for early detection of disease progression. Also, small-molecule JAK inhibitors and siRNA-mediated knockdown of Stat3/5 decreased the growth of CD34⁺ cells from high-risk AML patients both in vitro and in vivo (47), whereas Stat3 inhibitors induced apoptosis and decreased colony formation in HSPCs from both MDS (29) and AML (45) patients. Importantly, in all studies, the pharmacologic blockade of Stat3/5 selectively targeted clonal but spared normal HSPCs. Further corroborating these observations, our findings may serve as a guidepost for the ongoing investigation of Stat3/5 inhibition as a therapeutic strategy to overcome azacitidine resistance (4, 48).

T.P. Vassilakopoulos is a consultant/advisory board member for Genesis Pharma. S.G. Papageorgiou reports receiving speakers bureau honoraria from Genesis. I. Kotsianidis reports receiving speakers bureau honoraria and commercial research grants from Genesis Pharma. No potential conflicts of interest were disclosed by the other authors.

Conception and design: I. Kotsianidis

Development of methodology: P. Miltiades, E. Lamprianidou, S.G. Papageorgiou, E. Nakou, I. Kotsianidis

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.P. Vassilakopoulos, S.G. Papageorgiou, A. Galanopoulos, S. Vakalopoulou, V. Garypidou, E. Hatjiharissi, H.A. Papadaki, E. Spanoudakis, C. Tsatalas, I. Kotsianidis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P. Miltiades, E. Lamprianidou, T.P. Vassilakopoulos, C.K. Kontos, P.G. Adamopoulos, M. Papaioannou, E. Spanoudakis, I. Kotsianidis

Writing, review, and/or revision of the manuscript: P. Miltiades, E. Lamprianidou, T.P. Vassilakopoulos, S.G. Papageorgiou, C.K. Kontos, V. Pappa, A. Scorilas, I. Kotsianidis

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P. Miltiades, E. Lamprianidou, I. Kotsianidis

Study supervision: V. Pappa, I. Kotsianidis

Other (carried out part of the mutational analysis): C.K. Kontos, P.G. Adamopoulos

This work was supported in part by an educational grant from Genesis Pharma Hellas (to I. Kotsianidis).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.