CD70/CD27 signaling promotes blast stemness and is a viable therapeutic target in acute myeloid leukemia

Acute myeloid leukemia (AML) is a group of genetically diverse and highly aggressive hematological malignancies characterized by the accumulation of immature blasts. AML represents the most common form of acute leukemia in adults and accounts for most leukemia-related deaths (Siegel et al., 2013; Döhner et al., 2015).

In recent years, genetic and molecular aberrations underlying AML pathogenesis have been identified. A first genetic alteration occurs in a hematopoietic stem/progenitor cell (HSPC), initiating clonal expansion. Subsequently, within this expanding clone, additional cooperating mutations are acquired, resulting in aberrant cell growth and a differentiation block (Jan et al., 2012; Corces-Zimmerman et al., 2014; Shlush et al., 2014; Vasanthakumar and Godley, 2014). The improved understanding of disease mechanisms has allowed defining biologically homogenous risk groups with regard to treatment response, disease relapse, and overall survival (Patel et al., 2012; Zeisig et al., 2012). The current standard of care for the majority of AML patients is still a combination of cytarabine with an anthracycline. However, the characterization of molecular abnormalities in AML has led to the development of novel targeted agents, including FLT3, IDH1/2, and KIT inhibitors (Döhner et al., 2015).

AML is hierarchically organized and maintained by self-renewing leukemia stem cells (LSCs) that sustain a pool of disease-inducing cells (Reya et al., 2001; Huntly and Gilliland, 2005; Horton and Huntly, 2012). LSCs may self-renew symmetrically or divide asymmetrically into an LSC and a more differentiated progenitor. Changes in this balance toward symmetric self-renewal will lead to an accumulation of undifferentiated malignant cells with stem cell characteristics (Kreso and Dick, 2014; Bajaj et al., 2015). For example, this was shown for the progression of chronic myelogenous leukemia (CML) from chronic to blast phase where the fraction of symmetrically dividing cells increased (Jamieson et al., 2004; Wu et al., 2007; Bajaj et al., 2015). Concordantly, high LSC numbers as well as stem cell gene signatures in blasts are negative predictors for survival (van Rhenen et al., 2005; Pearce et al., 2006; Gentles et al., 2010; Eppert et al., 2011). Therefore, targeting signals that induce LSC expansion, either by blocking proliferation or by forcing differentiation via asymmetric cell division may lead to resolution of the disease (Horton and Huntly, 2012; Bajaj et al., 2015).

CD27, a costimulatory receptor of the TNF superfamily, is constitutively expressed on lymphocytes and HSPCs (Nolte et al., 2009; Schürch et al., 2012). CD70, its only ligand, is expressed on activated lymphocytes and dendritic cells but is undetectable in homeostasis (Nolte et al., 2009). During immune activation, CD70/CD27 signaling promotes lymphocyte expansion and survival and modulates hematopoiesis by regulating HSPCs (Nolte et al., 2005, 2009). Interestingly, CD70 is aberrantly expressed on different solid tumors and lymphomas and was shown to induce local immunosuppression in glioblastoma and renal cell carcinoma (Grewal, 2008; Nolte et al., 2009).

In this study, we demonstrate that AML blasts and AML stem/progenitor cells coexpress CD70 and CD27. Soluble CD27 (sCD27), a marker for the extent of CD70/CD27 interactions in vivo, is considerably increased in the sera of newly diagnosed AML patients and is a strong prognostic biomarker for poor overall survival independently of age or cytogenetic/molecular risk group. CD70/CD27 signaling in AML cells induces stem cell gene signature pathways including canonical Wnt, JAK/STAT, Hedgehog, and TGF-β signaling and promotes an undifferentiated and malignant state by increasing symmetric cell divisions. Blocking CD70/CD27 signaling promoted asymmetric cell divisions and differentiation of AML blasts, decreased growth and colony formation, and induced differentiation of AML stem/progenitor cells in vitro. In contrast, HSPCs from healthy BM donors were not affected by this treatment. Blocking CD70/CD27 signaling by mAb in murine AML xenografts delayed disease progression, reduced the number of AML stem/progenitor cells and prolonged survival.

CD70/CD27 signaling is deregulated in solid tumors, lymphoma, and CML (Grewal, 2008; Nolte et al., 2009; Schürch et al., 2012; Riether et al., 2015). To investigate a potential role of CD70/CD27 signaling in AML, we established a liquid biobank of sera and freshly isolated blood and BM samples from untreated AML patients at first diagnosis at our institution from 2011 to 2015. Because CD27 ligation results in the release of sCD27 by shedding, sCD27 can serve as a biomarker for the extent of CD70/CD27 interactions in vivo (Nolte et al., 2009). Overall, sCD27 serum levels were significantly increased in 137 AML patients compared with healthy controls (Fig. 1 A). To assess serum sCD27 in relation to overall survival, receiver operating characteristic curve analysis was performed in 30% of randomly selected patients. This resulted in an optimal threshold of 577 U/ml to define low and high. Using this threshold for the entire cohort, Kaplan-Meier analysis revealed that patients with low serum sCD27 (≤577 U/ml) survived significantly longer than patients with high serum sCD27 (Fig. 1 B).

Cytogenetic/molecular risk group and patient age are important established prognostic parameters in AML (Zeisig et al., 2012) and possibly acted as confounding factors in our analysis. Age and serum sCD27 levels did not significantly differ between patients in the favorable, intermediate, or adverse risk groups (Fig. 1, C and D; and Table S1). As expected, Kaplan-Meier survival curves revealed statistically significant survival differences between the different risk groups (Fig. 1 E). Furthermore, serum sCD27 demonstrated prognostic value in subgroup analyses within the different risk groups (Fig. 1, F–H). In addition, serum sCD27 correlated significantly with the percentage of blasts in BM (Table S1 and not depicted). Importantly, however, multivariate analysis for sCD27 levels adjusted for risk group, patient age, percentage of blasts in BM and blood, and leukocyte counts confirmed serum sCD27 as a strong independent prognostic marker in the entire AML patient cohort (Fig. 1 I).

In contrast, high CD27 mRNA levels in two different publicly available AML microarray datasets (Valk et al., 2004; Metzeler et al., 2008) were associated with a favorable prognosis (unpublished data). However, these datasets were not generated using purified blasts but with total PBMCs containing immune cells such as lymphocytes that express substantially higher levels of CD27 than AML blasts (Fig. 2 A). In accordance, CD27 mRNA levels positively correlated with CD3, CD8, and CD4 mRNAs in these datasets (unpublished data). Therefore, high CD27 mRNA expression in AML PBMCs rather reflects an activated immune system that promotes survival and does not necessarily indicate the extent of CD27 signaling on AML blasts.

These results indicate that sCD27 in serum but not CD27 mRNA in PBMCs is an independent negative prognostic biomarker for overall survival in AML.

We next intended to test whether CD27 and/or CD70 protein can be detected on AML blasts by FACS. The CD45 and side scatter (SSC) gating strategy is an established method to identify blasts (CD45^dimSSC^lo) in phenotypically different AML patient samples, but the blast gate may include other cell populations (Borowitz et al., 1993; Kroft and Karandikar, 2007; Gorczyca, 2010). Accordingly, the blast gate can be identified in healthy individuals as well. We therefore only included patients in this analysis with a documented blast frequency of ≥40% as determined by morphology. Healthy donors served as negative controls. As expected, CD27 and CD70 were expressed on lymphocytes but not on granulocytes. Interestingly, CD27 and CD70 were detectable on AML blasts as well (Fig. 2 A). CD27 was detected on blasts in 36/42 (86%) blood and 24/25 (96%) BM samples, whereas CD70 was detected in 22/23 (96%) blood and 20/20 (100%) BM samples (Fig. 2, B and C). CD27 and CD70 were similarly expressed on blasts in blood and BM (Fig. 2, B and C). Both proteins were detected on AML blasts in 16/23 (70%) blood and 19/20 (95%) BM samples (Table S1). In contrast, cells in the blast gate of healthy controls that mainly represent monocytes and basophils (Aoun and Pirruccello, 2007) did not express CD27 or CD70 (Fig. 2 D and Fig. S1 A). The AML stem/progenitor cell population that contains the disease-initiating LSCs in the majority of AML samples is a subfraction of CD45^dimSSC^lo blasts characterized as lineage (lin)⁻CD90⁻CD34⁺ (Fig. S1 B; Blair et al., 1997; Bonnet and Dick, 1997; Sarry et al., 2011; Terwijn et al., 2014). Importantly, CD70 and CD27 were similarly detected on BM CD45^dimSSC^loCD34⁺ as well as blood CD45^dimSSC^lolin⁻CD90⁻CD34⁺ AML cells (Fig. 2, E–G), In contrast, BM HSPCs (CD45^dimSSC^lolin⁻CD90⁺CD34⁺ BM cells; Majeti et al., 2007) from patients who underwent BM biopsy for other reasons than leukemia (healthy BM donors) did not express CD70 or CD27 (Fig. 2 H).

Because CD27 ligation leads to shedding of the protein (Nolte et al., 2009), FACS analysis may underestimate its expression. Indeed, all FACS-sorted AML blasts expressed CD27 and CD70 mRNAs (Fig. 3 A). In addition, CD27 and CD70 mRNA expression in blasts from blood and BM was 10³- to 10⁵-fold higher than in HSPCs from healthy BM donors (Fig. 3 B). Furthermore, CD27 and CD70 were detected in the majority of AML cell lines analyzed (Fig. 3, C–E).

These results indicate that blasts and stem/progenitor cells of most AML patients as well as AML cell lines coexpress the TNF superfamily ligand–receptor pair CD70/CD27.

To investigate the possibility of cell-autonomous and/or paracrine CD70/CD27 signaling, we cultured FACS-sorted AML blasts from 20 different newly diagnosed patients in the presence of a blocking αCD70 mAb (clone 41D12-D; Silence et al., 2014; Riether et al., 2015) or a control mAb in vitro. sCD27 levels in cell supernatants were significantly lower in the presence of αCD70 as compared with control mAb, indicating that the CD70/CD27 interaction was efficiently blocked (αCD70: 4.39 ± 0.20 U/ml; control: 3.25 ± 0.13 U/ml; P < 0.0001). We next quantified the expression of 91 different stem cell signature and proliferation-promoting genes important in hematopoiesis using Fluidigm dynamic array (Table S2). 45 genes were differentially expressed in αCD70 compared with control mAb–treated blasts (Fig. 4 A). Hierarchical clustering revealed that blocking CD70/CD27 signaling down-regulated numerous target genes of the canonical Wnt pathway, which is essential for leukemogenesis (Fig. 4 B; Staal and Clevers, 2005). Moreover, the expression of important HSC and hematopoiesis genes, transcription factors, epigenetic modifiers, apoptosis, and cell cycle regulators as well as inflammatory and Hedgehog-signaling genes was reduced. Of note, blocking CD27 signaling reduced important stem cell genes such as Musashi and the TNF receptor associated factor 2 (TRAF2)– and NCK-interacting kinase (TNIK; Mahmoudi et al., 2009; Ito et al., 2010). In contrast, some genes related to Notch signaling, MAPK signaling, and matrix metalloproteinases were up-regulated upon blocking CD70/CD27 signaling (Fig. 4 B). Gene ontology (GO) enrichment analysis of the differently expressed genes revealed 20 different GO categories that were significantly regulated (over threefold) and established a hierarchy of pathways with cell proliferation, apoptosis, inflammation, and Wnt and MAPK signaling at the top of the hierarchy (Fig. 4 C). Furthermore, gene networks and canonical pathways within the 45 genes regulated by CD70/CD27 signaling were analyzed in silico using direct force analysis to study possible protein–protein and/or genetic interactions. This analysis revealed that stemness-associated genes that were down-regulated by blocking CD70/CD27signaling are directly triggered by Wnt pathway–associated genes (Fig. 4 D).

Therefore, CD70/CD27 signaling induces a stem cell signature at least partially via Wnt signaling, leading to a more undifferentiated and malignant state. Because disease evolution could not be tested in primary AML samples, we analyzed the expression of genes involved in the CD70/CD27 signaling pathway during the evolution of indolent chronic-phase CML via accelerated-phase to blast-phase CML, a fatal acute leukemia. Using a publicly available microarray dataset (Radich et al., 2006), we found a continuous increase in CD70 and TNIK expression from chronic- to accelerated- to blast-phase CML. In line with Ito et al. (2010), Musashi expression increased, whereas expression of the cell-fate determinant Numb decreased during CML progression (Fig. 4 E). Correspondingly, sCD27 levels in supernatants of control-treated AML blasts correlated with Musashi expression (Fig. 4 F). These data indicate that CD70/CD27 signaling is associated with the expression of stem cell genes such as Musashi and TNIK and disease progression in myeloid leukemia.

U-937 cells express CD27 and CD70 as analyzed by FACS (Fig. 3 C). To functionally investigate the pathways that resulted from the array analysis, we treated U-937 cells with blocking αCD27 or αCD70 mAb. This strongly inhibited cell proliferation, leading to reduced cell numbers after 3 d of culture (Fig. 5, A and B; and not depicted). To further investigate the effect of CD27 signaling on the Wnt pathway, we first determined the cellular localization of β-catenin, the key molecule of canonical Wnt signaling (Clevers and Nusse, 2012). ImageStream^X analysis showed nuclear translocation of active β-catenin in control-treated U-937 cells, whereas blocking CD27 resulted in preferential cytoplasmic β-catenin localization (Fig. 5, C and D). This indicates that CD27 signaling activates the Wnt pathway and confirms the GO analysis of primary AML blasts. TNIK, an essential activator of Wnt target genes (Mahmoudi et al., 2009), colocalized with β-catenin (Fig. 5, C and D). Confirmative, similarity analysis using IDEAS software (George et al., 2006) revealed significantly less nuclear colocalization of β-catenin/TNIK in αCD27- versus control-treated cells (Fig. 5 E). This correlated with significantly reduced relative luminescence units (RLU) in a T cell transcription factor (TCF)/lymphoid enhancer binding factor (LEF) luciferase Wnt signaling reporter assay (αCD27: 1,458 ± 32 RLU; control: 616 ± 24 RLU; P = 0.0052) and reduced transcription of the Wnt target genes BIRC5, CCND1, LEF, MYC, and VEGF (unpublished data). Interestingly, blocking the CD70/CD27 interaction resulted in a higher percentage of asymmetrically dividing U-937 cells, as analyzed by the intracellular distribution of the cell-fate determinant Numb (Fig. 5, F and G; Wu et al., 2007; Ito et al., 2010; Zimdahl et al., 2014). As a consequence, blocking CD70/CD27 signaling resulted in increased amounts of intracellular Numb and surface CD11b (Fig. 5, H and I), indicating cell differentiation.

In FACS-sorted blasts from different AML patients, αCD70 treatment resulted in significantly reduced cell numbers in liquid cultures (Fig. 5 J) and colony formation in methylcellulose (Fig. 5 K), confirming the results observed in U-937 cells. Moreover, preincubation of AML blasts with an irradiated CD70-expressing lymphoblastoid B cell line (LCL; Ochsenbein et al., 2004) did not significantly increase colony formation, suggesting that CD70/CD27 interactions in AML blasts are already maximal (Fig. 5 K). In contrast and in line with our previous work (Riether et al., 2015), αCD70 treatment did not affect colony formation of HSPCs from healthy BM donors (Fig. 5, L and M). We observed elevated Numb protein levels and a significantly increased expression of the differentiation marker CD11b in AML blasts after αCD70 treatment, whereas CD14 and CD33 expression remained unchanged (Fig. 5, N and O; and not depicted). Control-treated blasts divided significantly more often symmetrically than asymmetrically. Blocking CD70 reversed this balance in all patient samples and significantly increased asymmetric over symmetric cell divisions (Fig. 5 P).

Together, these results indicate that the CD70/CD27 interaction in AML blasts promotes symmetric cell division, resulting in more undifferentiated stem-like cells.

To investigate whether CD27 signaling regulates AML cell fate via TRAF2 and TNIK (Schürch et al., 2012), we knocked down these CD27 downstream molecules in U-937 cells using shRNA. TRAF2 or TNIK knockdown reduced U-937 cell growth to a similar extent as αCD27 treatment. Interestingly, αCD27 treatment of TRAF2 or TNIK knockdown cells did not further decrease cell growth (Fig. 6, A and B). Similarly to blocking CD27 signaling (Fig. 5, G and H), TNIK knockdown promoted asymmetric cell division and increased Numb expression in the U-937 cell population, which could not be further enhanced by adding αCD27 mAb (Fig. 6, C and D). Moreover, stable knockdown of TNIK in U-937 cells significantly prolonged survival of xenotransplanted NOD/LtSz-scid IL-2Rγ^null (NSG) mice (Fig. 6 E). This suggests that CD27 signaling increases symmetric AML cell division via TRAF2/TNIK signaling.

To further elaborate on these findings in a more physiological setting, we knocked down CD27, TRAF2, or TNIK in primary AML blasts. Knockdown efficiency and specificity were assessed on mRNA and protein level by quantitative real-time PCR (qRT-PCR) and ImageStream^X analysis, respectively (Fig. 6 F and not depicted). TNIK knockdown decreased Wnt target gene expression (Fig. 6 G), and CD27, TRAF2, or TNIK knockdowns reduced colony formation in methylcellulose (Fig. 6 H). Importantly, this reduction was similar between blasts with CD27, TRAF2, or TNIK knockdown and was maintained during serial replating. In contrast, cell numbers per well were considerably increased in knockdown cells during the first round of plating but significantly declined in further rounds, indicating that CD27, TRAF2, or TNIK knockdown induced differentiation, repressed self-renewal, and led to exhaustion of these cells (Fig. 6 I). As expected, the addition of αCD70 mAb to TNIK knockdown AML blasts did not further reduce colony formation (Fig. 6 J).

These results indicate that Wnt-activating and stemness-maintaining CD27 signals are mainly mediated via TRAF2 and TNIK in AML cells.

FACS-sorted CD45^dimSSC^lolin⁻CD90⁻CD34⁺ AML stem/progenitor cells from blood of six patients from different cytogenetic/molecular risk groups were cultured in the presence of blocking αCD70 or control mAb. After 3 d, cells were enumerated, and sCD27 levels were determined in supernatants. Blocking CD70 significantly reduced cell numbers per well and sCD27 levels in supernatants (Fig. 7, A and B). Importantly, sCD27 levels were also reduced when adjusted for different cell numbers (Fig. 7 C). To assess colony-forming capacity, AML stem/progenitor cells were cultured in methylcellulose in the presence of blocking αCD70 or control mAb. In line with our results from blasts (Fig. 6 J), αCD70 treatment significantly reduced colony formation from AML stem/progenitor cells (Fig. 7 D). Serial replating experiments revealed that this effect was maintained even in the absence of αCD70 mAb in the primary or secondary replating cultures (Fig. 7, D and E). In addition, the expression of the differentiation-inducing genes RUNX1, SPI1 (PU.1), CEBPα, CEBPβ, and ID1 (Rosenbauer and Tenen, 2007) was significantly increased in AML stem/progenitor cells cultured overnight in the presence of blocking αCD70 compared with control mAb (Fig. 7 F).

These findings indicate that blocking the CD70/CD27 interaction reduces AML stem/progenitor cell numbers and induces differentiation.

U-937 cells or FACS-sorted blasts from two different AML patients were injected i.v. into sublethally irradiated NSG mice. After 7 d of engraftment, NSG mice were randomized to αCD70 or control mAb treatment. αCD70 treatment significantly prolonged survival in xenotransplanted mice (Fig. 8, A–C). Similarly, stable TNIK knockdown in AML blasts prolonged survival of xenotransplanted mice (Fig. 8 D).

Human sCD27 was present in the sera of NSG mice xenografted with primary AML samples, and its levels increased as the disease progressed. Interestingly, in αCD70-treated mice, human sCD27 levels were significantly lower than in controls, indicating that CD70 blockade specifically reduced CD27 signaling on human AML blasts in these mice (Fig. 8 E). αCD70 treatment significantly reduced total human CD45⁺ cells and CD45⁺CD34⁺ AML stem/progenitor cells in xenografted mice 40 d after transplantation (Fig. 8, F and G). Importantly, Numb expression was increased in the population of CD45⁺CD34⁺ AML stem/progenitor cells, indicating a more differentiated state (Fig. 8 H). Furthermore, FACS-sorted CD45⁺CD34⁺ AML stem/progenitor cells from αCD70-treated xenografted mice down-regulated the expression of several Wnt signaling–, hematopoiesis-, and inflammation-associated genes (Fig. 8 I) and formed significantly fewer colonies in methylcellulose per well than controls (Fig. 8 J). Similarly, the CFUs of sorted CD45⁺CD34⁺ AML stem/progenitor cells per mouse were significantly reduced by αCD70 treatment (Fig. 8 K). In contrast, cell numbers per well and per colony were considerably increased (Fig. 8, L and M).

Together, these data suggest that αCD70 treatment of xenografted mice prolonged survival by inducing differentiation of AML stem/progenitor cells.

To confirm and extend these findings, we performed xenografts with additional patient samples from each risk group. Two weeks after transplantation, NSG mice were randomized and treated for 2 wk with αCD70 or control mAb. αCD70 treatment did not affect total BM cell numbers (Fig. 9 A). In contrast, human CD45^dimSSC^lo blasts and CD45^dimSSC^lolin⁻CD90⁻CD34⁺ AML stem/progenitor cells in BM were significantly reduced, irrespective of the cytogenetic/molecular risk group of the transplant (Fig. 9, B–D). Importantly, αCD70 treatment induced a significant up-regulation of CD27 on CD45^dimSSC^lolin⁻CD90⁻CD34⁺ AML stem/progenitor cells (Fig. 9 E). In addition, αCD70 treatment resulted in a trend of CD11b and Numb up-regulation in human CD45^dimSSC^lo AML blasts from xenografts (Fig. 9, F and G). Furthermore, FACS-sorted CD45^dimSSC^lolin⁻CD90⁻CD34⁺ AML stem/progenitor cells from αCD70-treated mice formed significantly fewer colonies in methylcellulose per well than controls (Fig. 9 H). In contrast, cell numbers per colony were considerably increased (Fig. 9 I). Importantly, Fluidigm dynamic array analysis of ex vivo FACS-sorted human CD45^dimSSC^lolin⁻CD90⁻CD34⁺ AML stem/progenitor cells from xenografts revealed that αCD70 treatment resulted in a down-regulation of important stem cell– and hematopoiesis-associated genes (Fig. 9, J and K), confirming and extending our in vitro data from AML blasts (Fig. 4, A and B).

To confirm the reduction of AML stem/progenitor cell numbers in vivo, we performed secondary transplantations using whole BM cells from primary xenotransplanted NSG mice. NSG mice receiving BM cells from αCD70-treated primary xenografts survived significantly longer than NSG mice transplanted with BM cells from control mAb-treated primary xenografts (Fig. 10, A–D). Importantly, this effect was found in samples from all different cytogenetic/molecular risk groups.

These results indicate that blocking CD70/CD27 signaling reduces numbers and function and induces differentiation of AML stem/progenitor cells by inhibiting stem cell gene expression programs.

An undifferentiated malignant state is a hallmark of AML blasts that is maintained by cell-intrinsic signals, but also possibly regulated by cell-extrinsic cues (Bajaj et al., 2015). In this study, we document an unexpected role for the CD70/CD27 ligand–receptor pair in the induction of Wnt signaling and stem cell gene signatures in AML. To our knowledge, this is the first report showing that the interaction of a TNF superfamily ligand–receptor pair induces stemness in cancer cells.

First, we focused on cells in the CD45^dimSSC^lo gate that identifies blasts in different AML types (Borowitz et al., 1993; Kroft and Karandikar, 2007; Gorczyca, 2010). Using sensitive xenograft assays, Sarry et al. (2011) have identified heterogeneous AML LSCs in bulk leukemia samples in fractions different from the originally reported lin⁻CD34⁺CD38⁻ population, even in cells expressing differentiation markers (Bonnet and Dick, 1997). Because stem cell gene signature profiles have been documented in bulk blast populations in AML (Gentles et al., 2010; Eppert et al., 2011), it has been suggested that differentiated blasts may exhibit plasticity and reenter the LSC state (Kreso and Dick, 2014). We then extended our analyses to lin⁻CD90⁻CD34⁺ cells, a subfraction of CD45^dimSSC^lo cells that contains the disease-initiating LSCs in the majority of AML samples (Blair et al., 1997; Terwijn et al., 2014). Importantly, CD70/CD27 signaling induced stemness in AML blasts and in phenotypically defined stem/progenitor cells.

Stem cells can self-renew either by symmetric renewal leading to an expansion of the stem cell pool or by asymmetric division, by which the pool remains constant (Wu et al., 2007). Stem cell gene signatures predict therapy resistance and aggressive disease and negatively correlate with survival in AML and other malignancies (Ben-Porath et al., 2008; Gentles et al., 2010; Eppert et al., 2011; Metzeler et al., 2013). Mechanistically, stem cell determinants such as stemness-associated genetic signatures (Gentles et al., 2010; Eppert et al., 2011; Metzeler et al., 2013) and epigenetic profiles (Figueroa et al., 2010; Bartholdy et al., 2014) often oppose differentiation-inducing programs in AML blasts, leading to a block in terminal differentiation and senescence (Tenen, 2003). Currently, all-trans retinoic acid is the only approved drug that induces differentiation in blasts of acute promyelocytic leukemia (Grimwade et al., 2010), but other differentiation-inducing agents for AML are under investigation (Nowak et al., 2009). We now show that blocking the CD70/CD27 interaction shifted the balance from symmetric self-renewal to asymmetric cell division. This reduced the pool of AML stem/progenitor cells, leading to more differentiated leukemia cells, as documented by increased expression of the cell fate determinant Numb and the differentiation marker CD11b.

The canonical Wnt pathway, which is central for HSC development and maintenance, is constitutively active in myeloid leukemia and of crucial importance for LSCs (Staal and Clevers, 2005; Wang et al., 2010; Heidel et al., 2012). Self-renewal and β-catenin signatures in murine and human AML LSCs are often induced by translocations involving the mixed lineage leukemia gene family (Krivtsov and Armstrong, 2007). In our study, the CD70/CD27 interaction activated Wnt signaling in AML cells. However, CD70/CD27 signaling triggers additional survival-inducing and proliferation-promoting pathways, such as the canonical and noncanonical NF-κB pathways and the JNK pathway (Nolte et al., 2009). Importantly, canonical NF-κB signaling enhances Wnt signaling, and the interplay of the NF-κB and Wnt pathways induces stem cell-like and tumor-initiating capacities in non–stem cells (Schwitalla et al., 2013). Interestingly, blocking CD70/CD27 signaling reduced the expression of Musashi, an RNA binding protein and Numb repressor. Genetic deletion of Musashi in a mouse model of blast-phase CML significantly reduced leukemia progression (Ito et al., 2010). Earlier results suggested that Musashi might impair asymmetric division and arrest differentiation, partially through suppression of Numb (Imai et al., 2001). The canonical Wnt pathway regulates the expression of Musashi in intestinal epithelial stem cells through a mechanism involving a functional TCF/LEF binding site on its promoter (Rezza et al., 2010). Therefore, it is likely that CD70/CD27 signaling indirectly regulates Musashi via the Wnt pathway.

sCD27 is an important indicator for the CD27/CD70 interaction in vivo. Human sCD27 increased in the AML xenotransplant mice during disease progression, indicating that sCD27 levels represent the strength of CD27 ligation on AML blasts. Importantly, the level of sCD27 in sera of AML patients is an independent negative prognostic factor for overall survival. Serum sCD27 could therefore be used clinically as a surrogate biomarker to address the stemness signature of a patient’s AML blasts and to predict outcome. Most likely, CD27 is engaged by CD70 cross-presented by other AML blasts or stem/progenitor cells in a paracrine manner. Alternatively, autocrine CD70/CD27 signaling occurs within the membrane of the same malignant cell. In addition, activated CD70-expressing lymphocytes may trigger CD27 on AML blasts or stem/progenitor cells. Importantly, the CD70/CD27 interaction can be blocked using mAb, leading to prolonged survival in primary and secondary xenotransplantation models. Because αCD70 treatment is specific for malignant cells and does not affect healthy HSPCs, blocking the CD70/CD27 interaction may represent a promising therapeutic strategy for AML.

NSG mice were a gift from J. Schwaller (Department of Biomedicine, University Hospital of Basel, Basel, Switzerland) and have been previously described (Shultz et al., 2005). Mice were housed under specific pathogen–free conditions in individually ventilated cages with food and water ad libitum and were regularly monitored for pathogens. Animal experiments were approved by the local experimental animal committee of the Canton of Bern and performed according to Swiss laws for animal protection.

The human leukemia cell lines Kasumi-1 (Asou et al., 1991), HL-60 (Gallagher et al., 1979), PL-21 (Kubonishi et al., 1984), NB4 (Lanotte et al., 1991), HT-93 (Kishi et al., 1998), U-937 (Sundström and Nilsson, 1976), MV4-11 (Lange et al., 1987), MOLM-13 (Matsuo et al., 1997), NOMO-1 (Kato et al., 1986), KG-1 (Koeffler and Golde, 1978), and HEL (Martin and Papayannopoulou, 1982) have been described before.

Peripheral blood samples (n = 42; patient age: 59.9 ± 2.1 yr; morphological blast count: 66.5 ± 2.8%), serum samples (n = 137; patient age: 58.8 ± 1.2 yr; blood leukocyte count: 30.6 ± 4.7 G/liter), and BM aspirates (n = 25; patient age: 57.9 ± 3.6 yr; morphological blast infiltration: 72.2 ± 3.9%) were obtained from untreated AML patients at diagnosis at the University Hospital of Bern (Bern, Switzerland) after written informed consent. Study data were collected and managed using REDCap electronic data capture tools hosted at the Department of Clinical Research (Harris et al., 2009). Serum samples were predominantly from a retrospective cohort (2011–2014). Blood, BM, and corresponding serum samples were collected prospectively (2013–2016). Detailed patient and control donor characteristics are listed in Table S1. Peripheral blood samples (n = 8; age 29.4 ± 1.8 yr) and serum samples (n = 5; age: 30.6 ± 2.1 yr) from young healthy individuals, serum samples (n = 10; age: 68.7 ± 2.2 yr) from elderly healthy donors, as well as BM samples from patients who underwent BM biopsy for reasons other than leukemia (n = 3; age: 73.3 ± 6.4 yr) were used as controls. Analysis of samples was approved by the local ethical committee of the Canton of Bern.

αCD27-FITC (clone LG.7F9), αCD27-allophycocyanin (APC)-Cy7 (clone LG.3A10), Armenian hamster IgG-FITC and -APC-Cy7 (clone HTK88), αCD11b-PE-Cy7 (clone M1/70), rat IgG 2bκ-PE-Cy7 (clone RTK4530), αCD34-APC (clone 561), αCD45–Pacific Blue (clone HI30), αCD33-PerCP-Cy5.5 (clone WM53), αCD90-PerCP-Cy5.5 (clone 5E10), and anti–mouse CD45-PE-Cy7 (clone 30-F11) were from BioLegend. Lineage-positive cells were excluded by staining using biotinylated αCD2 (clone RPA2.10), αCD3 (clone OKT39), αCD14 (clone HCD14), αCD16 (clone 3G8), αCD19 (clone HIB19), αCD56 (clone HCD56), and αCD235 (clone HIR2; BioLegend), followed by a second step using streptavidin–Horizon-V500 (BD). αBrdU-APC was from BD. Human αCD70 (clone 41D12-D) and a corresponding control mAb specific for the F protein of respiratory syncytial virus (palivizumab [Synagis]; AstraZeneca) were kindly provided by arGEN-X. CD70 stainings were performed by incubation with 50 µg/ml αCD70 mAb or control mAb followed by a second step using anti–human F_C-PE (BioLegend).

Samples were acquired on an LSRII (BD), and sorting procedures were conducted using an FACSAria (BD). Data were analyzed using FlowJo software (Tree Star).

Human αCD27 (clone 1A4) and the corresponding isotype control (clone 15H6) were from Beckman Coulter. Human αCD70 (clone 41D12-D) and palivizumab (Synagis) were from arGEN-X.

10⁵ cells of AML cell lines were seeded into 24-well tissue culture plates and cultured in the presence of 10 µg/ml blocking αCD27 or αCD70 mAb or the respective control mAb. Live cell numbers were counted daily using a Neubauer chamber and trypan blue exclusion.

Xenotransplantations were performed as previously described (Sanchez et al., 2009). In brief, NSG mice were sublethally irradiated (2.75 Gy) on the day before injection. 10⁶ FACS-purified CD45^dimSSC^lo blasts from the peripheral blood or BM of newly diagnosed AML patients (patients 142, 145, 161, 162, 163, 164, 167, and 168; Table S1) were injected i.v. into the tail vein. Starting 1–2 wk after transplantation, mice were randomized, and 10 mg/kg αCD70 mAb (41D12-D) or control mAb was administered i.p. every third day. Mice were monitored daily for signs of morbidity (significant weight loss, failure to groom, abnormal gait, and posture) and euthanized when terminally ill. Secondary transplantations were performed by injecting 5 × 10⁶ or 10⁷ whole BM cells from primary xenografted animals i.v. into sublethally irradiated (2.75 Gy, day before transplantation) NSG mice.

In vitro liquid cultures and methylcellulose colony assays of FACS-purified CD45^dimSSC^lo blasts from blood or BM and CD45^dimSSC^lolin⁻CD90⁻CD34⁺ AML stem/progenitor cells from blood of newly diagnosed AML patients (blasts: patients 38, 39, 72, 139, 142, 143, 145, and 146; CD45^dimSSC^lolin⁻CD90⁻CD34⁺ AML stem/progenitor cells: patients 161, 162, 163, 164, 167, and 168; Table S1) were performed as described (Schürch et al., 2013), with slight modifications. Starting cell numbers were 10⁵ for liquid cultures and 10³ for colonies, respectively. 10 µg/ml αCD70 or control mAb were added to the cultures in the first round of plating (1°). For each round of serial colony replating (first and second replating), total cells were collected from the methylcellulose, and 5 × 10³ cells were replated into methylcellulose without any treatment. Colony numbers were assessed by inverted light microscopy after 2 wk for each round of plating. In one experiment, AML blasts were preincubated overnight with 10⁵ irradiated (10 Gy) cells of a CD70-expressing LCL (Ochsenbein et al., 2004), followed by plating in methylcellulose.

U-937 cells were cultured in vitro for 72 h, and BrdU (10 µM) was added for the last 4 h of culture. BrdU staining was performed using the APC BrdU Flow kit (BD) according to the manufacturer’s instructions.

U-937 cells were transfected with TCF/LEF lentiviral particles expressing firefly luciferase or the respective positive and negative control lentiviral particles (Cignal Lenti TCF/LEF reporter [luc] kit; SABiosciences) at a multiplicity of infection of 10, in the presence of 8 µg/ml SureEntry transduction reagent (SABiosciences) according to the manufacturer’s instructions. Stable cell lines were generated under puromycin selection (2.5 µg/ml; Santa Cruz Biotechnology, Inc.). Luciferase activity was measured on an Infinite 200 microplate reader (Tecan) using the Steady-Glo Luciferase Assay System (Promega) according to the manufacturer’s instructions.

For qRT-PCR, total RNA was extracted using the RNeasy Mini kit (QIAGEN). Complementary DNA synthesis was performed using the High Capacity cDNA Reverse Transcription kit (Applied Biosystems). Gene expression analysis was performed using TaqMan Gene Expression Assays for CD27, CD70, TNIK, CCND1, BIRC5, TRAF2, LEF, MYC, VEGF, and GAPDH (Applied Biosystems), as well as using self-designed primers for CEBPA, CEBPB, RUNX1, ID1, SPI1 (PU.1), and GAPDH (Table S2) using SYBR green reaction (Applied Biosystems). qRT-PCR reactions were performed in triplicates including nontemplate controls using an ABI Prism 7500 Sequence Detection System (Applied Biosystems). Relative quantification of gene expression was normalized against a reference gene (GAPDH) and calculated as an exponent of 2 (2^ΔCt).

AML cells were fixed with 4% paraformaldehyde, followed by blocking and permeabilization with 5% goat serum/1% bovine serum albumin in 0.1% PBS–Tween 20 for 1 h in FACS tubes. After washing, cells were incubated for 1 h with mouse anti–active β-catenin (dilution 1:100; clone 8E7; EMD Millipore) and rabbit αTNIK antibody (dilution 1:50; D-16; Santa Cruz Biotechnology, Inc.), followed by incubation with anti–mouse IgG–Alexa Fluor 546 (dilution 1:500; Invitrogen) and goat anti–rabbit IgG–Alexa-Fluor 488 (dilution: 1:1,000; Cell Signaling Technology) for 30 min. Nuclei were stained with 10 µg/ml DAPI. Cells were acquired on an ImageStream^X Mark II imaging flow cytometer (Amnis/EMD Millipore) and analyzed using INSPIRE and IDEAS software (Amnis/EMD Millipore).

AML cells were fixed with 4% paraformaldehyde, permeabilized with 1× wash buffer (Dako), and blocked with 10% normal goat serum (Invitrogen) in wash buffer (Dako). Cells were incubated overnight at 4°C with the primary rabbit αNumb antibody (ab14140; Abcam) diluted 1:50 in diluent (Dako). Incubation with the secondary antibody, goat anti–rabbit IgG–Alexa Fluor 568 (dilution 1:1,000; Abcam), was performed for 1 h at room temperature. DAPI (Roche) was used to counterstain for DNA. Samples were acquired on an ImageStream^X Mark II imaging flow cytometer and analyzed using INSPIRE and IDEAS software. Dividing cells were analyzed by IDEAS software in a blinded fashion by two independent researchers. A difference in Numb intensity between daughter cells of at least 1.8-fold was defined as asymmetric cell division according to Zimdahl et al. (2014).

CD27, TRAF2, or TNIK was silenced in AML cells using transduction-ready viral particles for gene silencing (Santa Cruz Biotechnology, Inc.). In brief, 5 × 10⁴ cells were transduced overnight at 37°C and 5% CO₂ with 2 × 10⁵ infectious units of shCD27, shTRAF2, or shTNIK lentiviral particles or control scrambled (scr) RNA lentiviral particles (Santa Cruz Biotechnology, Inc.) in the presence of 5 µg/ml polybrene (Sigma-Aldrich) according to the manufacturer’s instructions. After 18 h, medium was removed, and cells were cultured in medium supplemented with 2.5 µg/ml puromycin (Santa Cruz Biotechnology, Inc.) to select for stable expression of shRNA or scrRNA. After selection, cells were cultured in medium without antibiotics.

Human sCD27 in serum samples from newly diagnosed, untreated AML patients, young and elderly healthy controls, and in serum from xenotransplanted NSG mice was measured using the PeliKine Compact human soluble CD27 ELISA kit according to the manufacturer’s instructions (Sanquin). For determination of sCD27 in cell culture supernatants, 10⁵ FACS-sorted CD45^dimSSC^loCD33⁺ AML blasts or CD45^dimSSC^lolin⁻CD90⁻CD34⁺ AML stem/progenitor cells were cultured in liquid culture in 96-well V-bottom plates in the presence of 10 µg/ml αCD70 or control mAb for 3 d. sCD27 in supernatants was measured using the PeliKine Compact human soluble CD27 ELISA kit.

10⁵ FACS-sorted CD45^dimSSC^loCD33⁺ blasts from the blood or BM of 20 different newly diagnosed AML patients were cultured in vitro in the presence of αCD70 or control mAb for 72 h, or CD45⁺lin⁻CD90⁻CD34⁺ AML stem/progenitor cells were FACS-sorted ex vivo from αCD70- or control mAb–treated NSG mice. Cells were subjected to simultaneous isolation of DNA, RNA, and proteins using the AllPrep DNA/RNA/Protein Mini kit (QIAGEN) according to an optimized protocol (Radpour et al., 2009, 2011). The quantity of extracted molecules was assessed by spectrophotometry using a NanoDrop ND-1000 (Thermo Fisher Scientific). For RNA samples, complementary DNA synthesis was performed using the High Capacity cDNA Reverse Transcription kit (Applied Biosystems). Gene expression profiling was performed on sample duplicates using the 96.96 Biomark Dynamic Array (Fluidigm) according to the manufacturer’s advanced protocol (Wong et al., 2010). Assays were designed based on EvaGreen chemistry (Biotium), and primers for targeting desired pathways were designed accordingly for amplicons of 100–150 bp using Primer3Plus (Table S2; Untergasser et al., 2007). In brief, cDNA was subjected to preamplification (specific target amplification [STA]) with a mix of primers specific for the target genes. STA was performed by denaturing at 95°C for 15 s and annealing/amplification at 60°C for 4 min, which was repeated for 15 cycles. STA products were then diluted fivefold in DNA suspension buffer according to the manufacturer’s instructions (Fluidigm) and supplemented with TaqMan Gene Expression Master Mix (Thermo Fisher Scientific) and EvaGreen DNA binding dye (Biotium). Samples and primers were loaded into the integrated fluidic circuits of the 96.96 Dynamic Array, and analysis was performed on a BioMark HD System (Fluidigm). Cycle threshold (Ct) values were calculated and visualized using BioMark real-time PCR analysis software (Fluidigm).

As a quality-control measure, we removed genes with Ct values of ≥32.0 or differences of ≥2.0 in-between sample duplicates. If the reference genes (ACTB and GAPDH) were not expressed or were removed because of the aforementioned criteria, the sample was not included in further analysis.

Raw values were normalized using the geometric mean of the two reference genes (ACTB and GAPDH). The fold difference for each sample was calculated using the comparative Ct method (Livak and Schmittgen, 2001). Relative gene expression quantities after log₂ transformation were used for unsupervised hierarchical clustering of differentially expressed genes using standard Pearson’s correlation as similarity measurement and Ward’s method for clustering the data (Radpour et al., 2011).

GO enrichment was assessed using Genomics Suite software, version 6.6 (Partek Inc.). The list of significantly differently expressed genes was grouped into functional hierarchies, and enrichment scores were calculated using a χ² test comparing the proportion of the gene list in a group to the proportion of the background in the group. A value of three or higher corresponds to differential expression of a pathway (P < 0.05).

Gene networks and canonical pathways representing differentially expressed genes were identified using the Ariadne Genomics Pathway Studio software, version 9 (Elsevier). The dataset containing gene identifiers and corresponding fold changes was uploaded into the Pathway Studio. The functional analysis identified the direct interactions between differentially expressed genes and selected pathways to facilitate the understanding beyond their regulatory networks.

Expression data were derived from a public repository for microarray data (GEO) and are available under accession nos. GSE4170 (Radich et al., 2006), GSE12417 (Metzeler et al., 2008), and GSE1159 (Valk et al., 2004).

Statistical analysis was performed using Prism 5.0 (GraphPad Software) and SAS 9.3 (SAS Institute). Data are represented as mean ± SEM. Data were analyzed using one-way ANOVA and Tukey’s or Dunnett’s multiple comparison test, two-way ANOVA, and Bonferroni post-test, unpaired or paired Student’s t test (two-tailed), or Mann–Whitney test. Receiver operating characteristic curve analysis was used to identify the optimal threshold for sCD27 for subsequent survival analysis. In brief, 30% of the entire patient cohort (n = 41) was randomly selected, and receiver operating characteristic curve analysis was performed. To verify the reliability of the cutoff, 200 bootstrapped replications were performed. Next, all patients were then classified as low or high according to the cutoff. Univariate survival analysis was performed. Survival time differences were plotted using Kaplan-Meier curves and analyzed using the log-rank test. After verification of the proportional hazards assumption, multiple Cox regression analysis was performed using the continuous sCD27 values. Effect size was determined using hazard ratios (HRs) and 95% confidence intervals, with a baseline hazard of 1.0 and a greater risk of death with HR >1.0. All P values were two-sided and considered significant when P < 0.05.

Fig. S1 shows the FACS gating strategy to identify the CD45^dimSSC^lo blast gate in healthy individuals and the gating strategy for AML CD45^dimSSC^lolin⁻CD90⁻CD34⁺ stem/progenitor cells. Table S1 lists the characteristics of AML patients, healthy controls, and patients who underwent BM biopsy for other reasons than leukemia. Table S2 lists the primer sequences used for qRT-PCR and the 96.96 Biomark Dynamic Array.

Data analyzed in this paper were generated in collaboration with the Genetic Diversity Centre (GDC, ETH Zürich, Zürich, Switzerland). We thank Jürg Schwaller for providing NSG mice and cell lines, Karen Silence for helpful discussions, and arGEN-X for providing the human αCD70 mAb.

This work was supported by grants from the Swiss National Science Foundation (31003A_149768 and 310030B_13313), the Swiss Cancer League (KFS2879-02-2012 and KLS 2342-02-2009), the Cancer League of the Canton of Bern, the Werner und Hedy Berger-Janser-Stiftung, the SWISS BRIDGE, the Research Support Foundation Vaduz (to A.F. Ochsenbein); the Swiss Cancer League (KLS 3346-02-2014), the Sassella Foundation, the Fondation Bios pour la Recherche, the Fondazione Dr. Carlo Gianella (to C. Riether); the Novartis Stiftung für medizinisch-biologische Forschung, the Empiris Foundation/Ursula Hecht Fund (to A.F. Ochsenbein and C. Riether); the Gertrud Hagmann-Stiftung für Malignomforschung, the SwissLife Jubiläumsstiftung, the Dr. Hans Altschüler-Stiftung, the Fondazione Dr. Carlo Gianella (to C.M. Schürch); the Fondazione per la ricerca sulla trasfusione e sui trapianti, the Mobiliar Jubiläumsstiftung, the Monika Kutzner Stiftung, the Stiftung Krebs-Hilfe, the Wolfermann-Nägeli-Stiftung (to C. Riether and C.M. Schürch); and the Mach-Gaensslen Stiftung (to A.F. Ochsenbein, C. Riether, and C.M. Schürch).

The authors declare no competing financial interests.

Author contributions: C. Riether and C.M. Schürch designed and performed experiments, analyzed and interpreted data, and wrote the manuscript. E.D. Bührer, M. Hinterbrander, A.-L. Huguenin, and S. Hoepner performed experiments and analyzed data. I. Zlobec designed and performed statistical analysis. T. Pabst collected and contributed AML patient samples. R. Radpour designed and performed experiments and analyzed and interpreted data. A.F. Ochsenbein designed experiments, interpreted data, wrote the manuscript, and supervised the project. All authors revised the manuscript and approved its final version.