ABSTRACT
Drug-resistance of tumor-initiating cells, impaired NK cell immune-response, PP2A loss-of-function and aberrant miRNA expression are cancer features resulting from microenvironmental- and tumor-specific signals. Here we report that genomic-imprinted MIR300 is a cell context-independent dual function tumor suppressor which is upregulated in quiescent leukemic stem (LSC) and NK cells by microenvironmental signals to induce quiescence and impair immune-response, respectively, but inhibited in CML and AML proliferating blasts to prevent PP2A-induced apoptosis. MIR300 anti-proliferative and PP2A-activating functions are differentially activated through dose-dependent CCND2/CDK6 and SET inhibition, respectively. LSCs escape PP2A-mediated apoptosis through TUG1 lncRNA that uncouples and limits MIR300 functions to cytostasis by regulating unbound-MIR300 levels. Halting MIR300 homeostasis restores NK cell activity and suppresses leukemic but not normal hematopoiesis by eradicating nearly all LSCs. Thus, MIR300 tumor suppressor activity is essential and therapeutically important for LSC-driven leukemias.
INTRODUCTION
Functional loss of protein phosphatase 2A (PP2A) tumor suppressor activity is a therapeutically-relevant cancer feature which is essential for cancer stem cells (CSCs) maintenance, tumor growth/progression and activation of natural killer (NK) cell proliferation and cytotoxic activity1–3. Microenvironment and tumor-specific factors (e.g. cytokines, oncoproteins) inhibit PP2A activity through the activation of the PP2A Inhibitory Pathway (PIP; Jak2-hnRNPA1-SET)2. Gain or loss of miRNA expression/function is a feature of cancer cells including CSCs4, 5 in which they may function as oncogenes or tumor suppressors influencing CSC’s function, tumor growth and innate anti-cancer immunity6, 7. Several miRNAs have been associated with CSC expansion and maintenance4 and PP2A inactivation8; however, a clear causal link between expression of specific miRNAs, persistence of the drug-resistant quiescent CSCs and PP2A loss-of-function is still missing. Furthermore, these miRNAs mostly act as regulators of CSC survival, maintenance into dormancy and cell cycle re-entry but it is unclear whether any of them directly promotes CSC cell cycle exit4, 9–11.
In myeloid leukemias, the mechanisms underlying LSC quiescence, survival and self-renewal, and reduced NK cell number and cytotoxicity result from integration of tumor cell-autonomous- and bone marrow (BM) microenvironment (BMM)-generated signals12, 13. The latter are triggered by BM niche-specific metabolic conditions (e.g. oxygen tension), cell-to-cell direct and, soluble and/or exosome-encapsulated factor (e.g. miRNAs, TGFβ1)-mediated interactions between tumor, mesenchymal stromal (MSCs), endothelial (EC) and immune cells6, 13–15.
Among the miRNAs predicted in silico to re-activate PP2A in LSCs, we focused on hsa-MIR300 (MIR300) because it was predicted to activate PP2A upon inhibiting multiple PIP components and PP2A-regulated factors essential for CSC maintenance, cancer development/progression and G1/S cell cycle transition. The human MIR300 (MIR300) is an intergenic miRNA that belongs to the 14q32.31 DLK1-DIO3 genomic-imprinted tumor suppressor miRNA cluster B16, 17; it was found involved in loss-of heterozygosity, inhibited in several tumor types with high mitotic index and during epithelial-to-mesenchymal transition (EMT), and associated with a CSC phenotype18–22.
By using Philadelphia-positive (Ph+) chronic myelogenous leukemia (CML) in chronic (CP) and blastic (BC) phase, and complex karyotype (CK) acute myeloid leukemia (AML), as paradigmatic examples of stem cell-derived neoplasms characterized by constitutive expression of oncogenic kinases (e.g. BCR-ABL1, JAK2), PP2A loss-of-function, altered miRNA expression and, impaired NK cell proliferation and cytotoxicity2, 6, 23–27, we show that MIR300 is a BMM (hypoxia, MSC)-induced PP2A activator with cell context-independent tumor suppressor anti-proliferative and pro-apoptotic activities which are essential for induction and maintenance of LSC quiescence and impaired NK cell immunity but can also be exploited to selectively and efficiently induce cell death in qLSC and leukemic progenitor while sparing normal hematopoiesis.
RESULTS
Inhibition of MIR300 tumor suppressor activities in leukemic progenitors and their dose-dependent differential induction in qLSCs
MIR300 levels were progressively and markedly reduced by BCR-ABL1 activity (imatinib treatment) in BM CD34+ CML (CP and BC) and CK-AML progenitors compared to normal BM (NBM) cells and up to 800-fold higher in qLSCs (CD34+CFSEmax) than dividing CD34+ CML (CP and BC) and CK-AML progenitors (Fig. 1a, b). Accordingly, MIR300 expression was higher in HSC-enriched CD34+CD38- than committed CD34+CD38+ CML (CP and BC) BM cells from patient samples at diagnosis, and comparable to those in NBM and umbilical cord blood (UCB) CD34+ cell fractions (Fig. 1a).
Consistent with the requirement of PP2A inhibition for leukemic but not normal stem/progenitor cell proliferation/survival2, 28, restoring MIR300 expression at physiological levels by clinically-relevant CpG-based MIR300 (CpG-miR-300) oligonucleotide (Fig. 2) or MIR300-lentiviruses (not shown) reduced by ≥75% proliferation and clonogenic potential, and enhanced apoptosis (spontaneous and TKI-induced) of CD34+ CML (CP and BC) but not UCB cells (Fig. 2a, b). Importantly, Ki-67/DAPI (G0: MIR300≅46% vs. scr≅10%; G1: MIR300≅15% vs. scr≅50%; S/G2/M: MIR300≅4% vs. scr≅27.4%; and subG1: MIR300≅35% vs. scr≅3%) and FUCCI2BL-mediated (G1/G0: MIR300≅40.2% vs scr≅23.6%; G1/S: MIR300≅15% vs. scr≅2.85%; S/G2/M: MIR300≅45% vs. scr≅76.6%) cell cycle analyses in CD34+ CML (CP and BC) and LAMA-84 cells indicated that MIR300 also selectively induced cell-cycle arrest with marked expansion of the G0 qLSC and sub-G1 apoptotic CD34+ cell fractions and reduced number of cells in G1 and S phases (Fig. 2c), indicating that MIR300 functions as a dual activity (anti-proliferative and pro-apoptotic) tumor suppressor miRNA.
The effects of MIR300 modulation on quiescent stem cell number (CFSE assay), proliferation, survival and self-renewal (LTC-IC and CFC/replating) were assessed in CML (CP and BC) and CK-AML and healthy individuals (Fig. 2d). Inhibition of MIR300 function (anti-miR-300) did not alter quiescent LSC and HSC number and activity (Fig. 2d, Supplementary Fig. 1). By contrast, graded MIR300 expression (Fig. 2d, inset) differentially and selectively suppressed CML and AML but not UCB LTC-IC-driven in vitro hematopoiesis, LSC-derived clonogenic and self-renewal activity (Fig. 2d, right and Supplementary Fig 1), and quiescent (CFSEmax) LSC numbers (Fig. 2d, left). Importantly, low-MIR300 doses strongly inhibited CML and AML LTC-ICs and CFC/replating activities without affecting qLSC numbers whereas high-MIR300 doses also reduced by ∼80% CML and AML qLSCs and dividing CD34+ progenitor cell numbers (Fig. 2d and Supplementary Fig 1). Thus, low MIR300 expression may reduce colony-formation by impairing qLSC ability to proliferate and undergo cytokine-induced differentiation but have no effect on LSC survival that, instead, is halted at MIR300 levels similar to those detected in qLSCs.
MIR300 acts as master PP2A activator and an inhibitor of G1/S transition
Gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) functional enrichment and clustering of MIR300 predicted and validated mRNA targets (DIANA-mirPath v.3) indicated that MIR300-induced massive apoptosis of qLSCs and leukemic progenitors likely depends on PP2A activation triggered by MIR300-induced inhibition of SET and other PIP factors (Fig. 3a). Likewise, MIR300 targeting of G1/S cell cycle regulators CCND2 (cyclin D2) and CDK6 and of several other PP2A-regulated survival and mitogenic factors (e.g. CTNNB1, JAK2, MYC, Twist1) (Fig. 3a, and Supplementary Fig. 2a left, 2b) may contribute to MIR300-induced cell death and account for cell cycle arrest and expansion of the qLSC G0 compartment.
Indeed, restoring MIR300 expression in primary CD34+ CML (CP and BC) progenitors and Ph+ cell lines by 500nM CpG-miR-300treatment, re-activated PP2A by inhibiting SET and the other PIP molecules (i.e. JAK2, hnRNPA1) and markedly reduced CDK6, CCND2, CTNNB1 (β-catenin), Twist1 and MYC expression (Fig. 3b). Note that CTNNB1, CCND1/2 and Twist1 are validated MIR300 targets20, 21. Furthermore, consistent with the notion that high levels of active PP2A are not detrimental in normal CD34+ quiescent stem and committed progenitors2, 28, CpG-miR-300 did not alter levels of MIR300 targets in CD34+ UCB cells (Fig. 3b). Importantly, expression of MIR300-insensitive Flag-tagged SET mRNAs deleted of the whole 3’UTR (Flag-SET) or just a region encompassing the high- and low-affinity MIR300-binding sites (Flag-Δ3’UTR-SET) but not that of full-length wild type SET mRNA (Flag-wt3’UTR-SET), rescued Ph+ progenitors from MIR300-induced cell cycle arrest and PP2A-dependent apoptosis (Annexin V+ cells) (Fig. 3b, right).
Interestingly, hierarchically clustering of MIR300 predicted and validated targets based on integration of algorithms (DIANA microT-CDS, mirDIP 4.1, ComiR and CSmiRTar) that also take into account MIR300 expression levels in normal and leukemic BM cells, positioned the G1/S cell cycle regulators CCND2 and CDK6 within the top 2% and SET within the 5% of MIR300 targets (Fig. 4a). Conversely, JAK2, hnRNPA1, CTNNB1 and Twist1 fell within the top 1/3, and Myc in the lower 2/3 of MIR300-interacting mRNA cluster (Fig. 4a, and Supplementary Fig. 3). Notably, other factors contributing to G1/S transition (e.g. CNNA2) and LSC re-entry into cycle for self-renewal or maturation (e.g. Notch signaling) clustered together with CCND2 and CDK6 (Supplementary Fig. 3). By contrast, MIR300 targets belonging to JAK-STATs, PI3K-Akt, Wnt, MAPK signaling pathways, which regulate survival and expansion of leukemic progenitors and disease progression, ranked below SET and within the bottom 75% of MIR300 target distribution (Supplementary Fig. 3). Because most of these MIR300 targets are also validated target of PP2A enzymatic activity (Supplementary Fig. 2b, 3), their PP2A-dependent inactivation likely occurs upon MIR300-induced SET inhibition. Thus, MIR-300-dependent inhibition of CCND2 and CDK6 should precede that of SET which should be followed by the PP2A-dependent JAK2, CTNNB2, TWIST and, lastly, MYC inactivation.
Indeed, immunoblots confirmed that CCND2 and CDK6 inhibition occurs in CD34+ leukemic stem/progenitor cells at a CpG-miR-300 concentration (100 nM) not triggering qLSC apoptosis whereas SET inhibition requires the same CpG-miR-300 dose that reduces qLSC numbers (Fig. 4b). This is consistent with the presence of 4 (1 very high affinity 9mer), 6 (3 high affinity 7mer) and 2 (1 high affinity 8mer) MIR300 binding sites in CCND2, CDK6 and SET mRNA 3’UTRs, respectively (Fig. 4a). Because CDK6/CCND2 downregulation is an essential feature of G1/G0-arrested qLSCs29, 30 and SET inhibition is sufficient for inducing PP2A-dependent LSC and progenitor cells apoptosis2, 26, the ability of MIR300 to sequentially trigger growth arrest and PP2A-mediated apoptosis of CD34+ CML/AML LSC and progenitors (Fig. 2c, d) indicates that MIR300 tumor suppressor activities are cell context-independent and suggests that MIR300 expression in LSC may account for their entry into quiescence whereas its downregulation in leukemic progenitors likely occurs to prevent apoptosis (Fig. 4b, top).
Specificity of MIR300 activity is further supported by the inability of intragenic 14q32.31 tumor suppressor17 hsa-miR-381-3p (miR-381-3p) to downregulate JAK2, SET and β-catenin levels and inhibit CD34+ CML cell proliferation and clonogenic potential (CFC assays) despite having an identical seed sequence, highly homologous flanking region and similar expression patterns in LSC-enriched CD34+CD38-, committed CD34+CD38+ progenitors and bulk CD34+ CML-CP and - BC cells (Supplementary Fig. 4a). Note that, the presence of a +4 A-to-G seed sequence mutation in the mouse mir300 (mmu-miR-300) gene significantly alters target repertoire (miRTar, MFE: ≤ - 10kcal/mol; Score: ≥136.5) and it is predicted to impair mmu-miR-300 tumor suppressor activities by inhibiting binding to JAK2, SET, CCND2 and CDK6 mRNAs (not shown).
BMM-induced C/EBPβ-dependent MIR300 tumor suppressor activity preserves LSCs
MIR300 is under the control of an intergenic differentially-methylated region (IG-DMR) preceding and controlling the expression of MEG3 (Supplementary Fig. 5a), a maternally-expressed genomic imprinted anti-proliferative lncRNA that is important for long-term HSC (LT-HSC) maintenance but strongly inhibited upon promoter methylation in virtually all types of cancer including CD34+ CML cells17, 31. Treatment with 5-Aza-2’-deoxycytidine (5-Aza) augmented by 104-105-fold MIR300 expression in Ph+ cells (Fig. 5a, left); however, nearly all cells underwent apoptosis after 24h (not shown), suggesting that MIR300 upregulation in qLSCs unlikely depends on IG-DMR demethylation and expression of all 74 cluster B tumor suppressor miRNAs (Supplementary Fig. 5a). Thus, MIR300 induction in CML/AML qLSCs and its ability to increase the quiescent fraction of leukemic but not normal CD34+ cells implies that BM osteogenic niche metabolic and cellular factors (e.g. MSCs, hypoxia), known to inhibit growth and induce quiescence of leukemic cells12–14, may selectively favor LSC entry into quiescence by increasing MIR300 expression and regulating its anti-proliferative activity in CD34+ LSCs (Fig. 5a right).
Indeed, expression of MIR300, but not of miR-381-3p (Supplementary Fig. 4b), increased at qLSC or higher levels (Fig. 1b) in primary CD34+ CML-BC and/or LAMA-84 cells exposed to hypoxia, BM-derived primary CD34-CD45-CD73+CD105+CD90+CD44+ hMSC and HS-5 conditioned medium (CM) or MIR300-containing CD63+Alix+ MSC exosomes (Fig. 5b,c). This correlated with a 40-60% reduced cell division and, importantly, with doubled numbers of CD34+ CML cells in the CFSEmax CD34+ qLSC fraction (Fig. 5b, c and Supplementary Fig. 5b). Similarly, Ph+ LAMA-84 (Supplementary Fig. 5c) and K562 (not shown) cells responded to MSC-CM exposure by slowly ceasing proliferation and modifying gene expression in a manner similar to that of CML qLSCs2, 12. In fact, exposure to MSC (CM and/or exosomes) neither altered SET and PP2AcY307 (inactive) levels (Supplementary Fig. 5c) nor induced cell-death (not shown) in CD34+ CML-BC and/or Ph+ cells, suggesting that MSCs increase MIR300 expression at levels not sufficient to trigger PP2A-mediated apoptosis.
The absolute requirement of MIR300 for BMM-induced LSC quiescence is revealed by the ability of anti-MIR300 molecules (pZIP-miR-300 or CpG-anti-miR-300) to prevent hypoxia-induced downregulation of MIR300 targets (e.g. JAK2, Myc, SET and β-catenin) in CD34+ leukemic cells, and to protect CD34+ CML cells from MSC (CM and exosome) growth-inhibitory activity when expressed in MSCs, in which they enhance β-catenin but not SET expression (Fig. 5b, c). Moreover, the evidence that apoptosis is not induced in MSC-exposed CD34+ CML cells despite the marked MIR300 increase, and that hypoxia but not MSC CM induces MIR300-dependent SET inhibition and increases mature MIR300 to levels similar to those in qLSCs (Fig. 1b and 5b) indicate that the contribution of MSC (CM and exosomes) to MIR300 expression in LSCs is not sufficient for inhibiting SET and triggering PP2A-dependent apoptosis, and that hypoxia may induce MIR300 transcription in LSCs. Accordingly, primary MIR300 (pri-miR-300) transcripts were substantially higher in CD34+ CML-BC cells cultured (n=4; 48h) in hypoxic than normoxic conditions (Fig. 5d).
Luciferase (luc) assays in normoxia- and hypoxia-cultured CD34+ CML-BC cells transduced with reporter constructs containing the full-length (p513) or a 5’-deleted (p245, p109) intergenic region, identified a hypoxia-sensitive MIR300 regulatory element within the 109 bp preceding the human MIR300 gene (Fig. 5e). Because ENCODE (V3) ChIP-Seq and PROMO analyses revealed that the CCAAT Enhancer Binding Protein B (C/EBPβ) may interact with two DNaseI hypersensitive regions within the intergenic 513bp preceding the MIR300 gene (Fig. 5e), we assessed whether hypoxia-induced MIR300 upregulation results from C/EBPβ binding and transactivation of MIR300 transcription.
Luciferase assays with a p109 construct carrying mutated C/EBPβ binding sites at position -64 and -46 bp (p109mut) indicate that the integrity of these hypoxia-sensitive C/EBPβ responsive elements is essential for MIR300 transactivation in CD34+ CML-BC cells (Fig. 5e). Indeed, increased pri-miR-300 expression in hypoxic CD34+ CML-BC cells correlated with markedly higher C/EBPβ LAP1 (transcriptionally active) levels (Fig. 5d). By contrast, lack of C/EBPβ-driven p109-luc activity in normoxic CD34+ CML-BC progenitors correlated with increased C/EBPβ LIP (inhibitory function) expression (Fig. 5d), indicating that hypoxia-induced MIR300 expression requires C/EBPβ transactivation. Importantly, consistent with the notion that BCR-ABL1 is inactivated by hypoxia to induce quiescence and allow survival of CML LSCs2, 14 and that BCR-ABL1 activity inhibits CEBPB translation32, MIR300 induction by hypoxia correlated with decreased BCR-ABL1 activity but not expression (Fig. 5d), and ectopic C/EBPβ-ERTAM mimicked the effect of hypoxia on C/EBPβ by rescuing primary (pri-miR-300) and mature MIR300 expression in normoxic CD34+ CML-BC progenitors (Fig. 5f).
Importantly, the notion that mouse c/ebpβ induces BCR-ABL+ LSK exhaustion33 does not argue against a role for human C/EBPβ as an inducer of LSC quiescence because a) LSKs are pushed into cycle by a fully active BCR-ABL1 that promotes C/EBPβ-dependent LSK maturation33; b) the hypoxia-sensitive human C/EBPβ-responsive MIR300 regulatory element is not conserved in mouse cells; and, c) an A-to-G mutation in the mmu-miR-300 seed sequence (Supplementary Fig. 4) prevents mmus-miR-300 targeting of ccnd2 and cdk6 mRNAs (not shown).
Hypoxia also substantially increased by 10-20-fold MIR300 and C/EBPβ protein but not mRNA levels in BM-derived primary MSCs and/or HS-5 cells (Fig. 5b and Supplementary Fig. 5d), suggesting that the hypoxic conditions of the osteogenic BM niche may simultaneously induce C/EBPβ-dependent MIR300 transcription in LSCs and increase the amount of MSC-derived exosomal MIR300 transferred into LSCs. Conversely, MIR300 levels in C/EBPβ−responsive CD34+ CML-BC cells and/or Ph+ cell lines were not influenced by ectopic C/EBPα expression or exposure to TGFβ1 blocking-Ab (Supplementary Fig. 5e, f).
Thus, increased MIR300 transcription occurs in a TGFβ-independent manner through the hypoxia-induced C/EBPβ LAP1 expression in LSCs and MSCs. The latter also contribute to increased MIR300 levels in LSCs via exosomal transfer.
BMM-induced MIR300 inhibits NK cell anti-cancer immunity
Cytokine-induced expression of miR-155 and SET, and inhibition of PP2A activity (Fig. 6a and Supplementary Fig. 6a) are essential for CD56+CD3- primary NK and clinically-relevant34 NK-92 cell proliferation and cytotoxicity against tumor-initiating cells35 including CD34+ CML qLSCs (Fig. 6b, left). Conversely, endosteal BMM factors elicit signals that inhibit NK cell proliferation and cytotoxicity7, 35, 36 and are, likely, responsible for dysfunctional NK cells in leukemia patients24. Thus, we investigated whether BMM-induced MIR300 inhibits NK cell killing of CML qLSCs and progenitors through downregulation of key regulators of NK cell proliferation (e.g. CCND2, CDK6) and activity (e.g. SET, miR-155) (Fig. 6a). First, we found that increased MIR300 expression is also a feature of peripheral blood (PB) CD56+CD3- NK cells isolated from CML patients at diagnosis but not of NK cells from healthy individuals (Fig. 6b). Furthermore, KEGG/GO analyses (Supplementary Fig. 2a) indicated that MIR300 may reduce levels of key regulators of NK cell function and account for suppression of innate anti-cancer immunity in CML.
Second, hypoxia (1% O2 tension) and/or BM MSC- and HS-5-derived CM and Alix+CD63+exosomes (100ug/ml) markedly increased C/EBPβ and MIR300 but reduced CCND2, CDK6 and SET expression (Fig. 6c). Consistent with the notion that downregulation of CCND2, CDK6 and SET3, 36, 37 impairs NK cell cytokine-dependent proliferation and induces PP2A-dependent NK cell inactivation, respectively, BMM (hypoxia and MSCs)-induced MIR300 upregulation was accompanied by 45% to 75% inhibition of IL-2-dependent CD56+CD3- primary NK and NK-92 cell proliferation (Fig. 6d and Supplementary Fig. 6b) and by severely impaired NK cell immunoregulatory (IFNγ production) and anti-tumor cytotoxic (K562 killing) activities (Fig. 6e). Importantly, lentiviral anti-miR-300 (pZIP-miR-300)-transduction into HS-5 cells (Fig. 6d, left) significantly suppressed MSC CM- and/or exosome inhibitory effects om NK cell proliferation and anti-cancer cytotoxicity (Fig. 6d, e). Similar effect was obtained by pre-treatment of NK-92 cells with CpG-anti-miR-300 but not with CpG-scramble (Fig. 6d). This suggests that BMM-induced impaired NK cell growth and anti-qLSC cytotoxicity occurs in a MIR300-dependent manner. In fact, CpG-miR-300 treatment mimicked the inhibitory effects of exposure to BMM (hypoxia and MSCs) and strongly suppressed IL-2-induced proliferation and cytotoxicity of CD56+CD3- NK and NK-92 cells (Fig. 6f). However, sequencing (RNAseq) of MSC exosomal RNA suggested that also other MSC-derived miRNAs may potentially impair NK cell activity (Supplementary Fig. 6c). Interestingly, MSC HS-5 exosomes also reduced miR-155 (BIC) transcription in IL-12/IL-18-stimulated and resting NK-92 cells (Fig. 6g). Because, miR-155 not only inhibits SHIP1 and PP2A to allow MAPK- and AKT-dependent NK cell proliferation and cytotoxic activity8, 35, 38 but also suppresses C/EBPβ expression39, MSC-induced transcriptional downregulation of miR-155 precursor levels likely contributed to C/EBPβ-mediated induction of MIR300 (Fig. 6a, g).
To determine whether constitutive miR-155 overexpression in NK cell enhances their anti-LSC cytotoxic activity, we employed highly proliferating lck-miR-155 (miR155-tg) transgenic NK1.1+CD3- NK cells, which show enhanced proliferation and cytotoxic activity in vivo35. miR155-tg NK1.1+CD3- NK cells killed self-renewing CFSE+ leukemic SCL-tTA-BCR-ABL1 but not normal LSC-enriched LSK cells more efficiently than wild type (wt) NK1.1+CD3- NK cells (∼70% vs ∼30% cytotoxic activity) (Fig. 6h), suggesting that therapies with allogeneic NK cells engineered to express high miR-155 levels alone or associated with MIR300 downregulation may overcome BMM-induced loss of NK cell proliferation/activity. Indeed, clinically-relevant31, 40–44 NK-92 cells already decreased by 50% TKI-resistant CML-BC qLSC numbers (Fig. 6b).
TUG1 selectively allows MIR300 anti-proliferative activity in CML and AML LSCs
Acute and chronic myeloid qLSCs display low to undetectable CCND2 and CDK6 expression but high SET (PP2A inhibition) levels2, 23, 30, and survive despite the high levels of functional (target downregulation) MIR300 (Fig. 1b, 5b), suggesting that there must be a MIR300-interactinge factor that uncouples and differentially regulates MIR300 activities to allow LSC cell cycle exit and prevents qLSC apoptosis through modulation of free MIR300 intracellular levels. In this regard, the taurine upregulated gene 1 (TUG1) was described as a MIR300-interacting lncRNA in gallbladder carcinoma and enhances tumor cell growth and EMT upon binding to miRNAs and preventing suppression of a set of mRNAs (e.g. Myc, CCND1/2, CTNNB1, Twist1) regulating stemness and described as MIR300 target in solid tumor cells22, 45, 46, suggesting that TUG1 may differentially regulate MIR300 tumor suppressor activities (Fig. 7a).
Accordingly, TUG1 levels are markedly increased in CD34+ normal (UCB and BM) and leukemic HSCs including CD34+ qLSCs from CML-BC and CK-AML patients and healthy individuals (Fig. 7b and Supplementary Fig. 7b, e). In CML, TUG1 is regulated in a manner similar to that of MIR300 (Fig. 1, 7b); in fact, TUG1 expression in qLSCs is imatinib-insensitive whereas it is markedly reduced by BCR-ABL1 activity in dividing CD34+ progenitors (Fig. 7b). In AML, TUG1 expression is induced by AML oncogenes (e.g. AML1-ETO, MLL-fusions) and readily detectable in different AML subtypes (Supplementary Fig. 7b, c). Interestingly, TUG1 levels were not downregulated in dividing CD34+ CK-AML progenitors but remained similar to those detected in CK-AML qLSCs (Fig. 7b).
Because TUG1 expression correlates with that of TGFβ1 and TGFBR1 in tumor cells undergoing EMT47 and TGFβ1 is known to be important for LSC quiescence and secreted by leukemic cells including CD34+ CML blasts48, we investigated the role of TGFβ1 in TUG1 upregulation and MIR300-induced growth-arrest of CD34+ leukemic cells and qLSC apoptosis (Fig. 2, 5). Exposure (48h) of CD34+ CML cells to a TGFβ1 blocking antibody (anti-TGFβ Ab; green bars) halves TUG1 levels and qLSC numbers but very modestly affects dividing leukemic progenitors (Fig. 7c left, 7e). Interestingly, exposure to anti-TGFβ Ab also reduces levels of FoxM1 (Fig. 7c), a transcription factor that mediates HSC quiescence and induces TUG1 expression in osteosarcoma cells49, 50. By contrast, FoxM1 and/or TUG1 expression is strongly induced in a TGFβ-dependent manner in hypoxia-cultured (48h, 1% O2) CML CD34+CFSEmax qLSCs and bulk CD34+ CML-BC cells (Fig. 7c), suggesting that hypoxia-induced TGFβ148 may enhance TUG1 levels in a FoxM1-dependent manner to neutralize MIR300 pro-apoptotic activity in LSCs. Indeed, dosing TUG1 downregulation by exposure to low and high CpG-TUG1-shRNAs (Fig. 7e inset) mimicked the dose-dependent differential triggering of MIR300 anti-proliferative and pro-apoptotic activities exerted by different CpG-miR-300 (Fig. 4); in fact, exposure of Ph+ cells to 100 nM CpG-TUG1-shRNA efficiently induced growth-arrest but modestly affected survival (Annexin V+ cells). Conversely, 500 nM CpG-TUG1-shRNA levels decreased CML-BC and CK-AML qLSC numbers in a MIR300-sensitive manner and further enhanced CpG-miR-300-induced apoptosis thereby causing a nearly complete loss of qLSCs and dividing progenitors (Fig. 7d, e). As expected, TUG1 overexpression (TUG1 RNA) antagonized CpG-miR-300-induced qLSC apoptosis in a MIR300 dose-sensitive manner (Fig. 7e). Thus, the ability of CpG-anti-miR-300 to fully prevent TUG1-shRNA-induced apoptosis of CML/AML qLSCs (Fig. 7e) indicates that hypoxia-induced TGFβ1-FoxM1-mediated signals increases TUG1 expression in CML/AML qLSCs to dynamically regulate quiescence by maintaining the amount of free functional MIR300 at levels sufficient for arresting cell cycle but not for triggering PP2A-dependent apoptosis (Fig. 7a and Supplementary Fig. 7a).
While TUG1 seems to act identically in CML and AML qLSCs, the low and high TUG1 levels in CD34+ CML-BC and CK-AML progenitors, respectively (Fig. 7b), the inability of CpG-anti-miR-300 to counteract TUG1-shRNAs-induced apoptosis of dividing CD34+ leukemic progenitors (Fig. 7e), and the nearly complete killing of CD34+ CML/AML progenitors by the combination of CpG-miR-300 and CpG-TUG1-shRNA oligonucleotides (Fig. 8e) suggest that TUG1 activity in dividing leukemic blasts may involves requires the interaction with other miRNAs regulating leukemic cell proliferation/survival.
Reportedly, TUG1 interacts with several miRNAs45, 46, 51 and bioinformatics analysis revealed that TUG1 may function as hub for tumor suppressor miRNAs with activities similar to that of MIR300 (Supplementary Fig. 8c). By RNAseq, we found that 56 of these experimentally-validated TUG1-interacting miRNAs are differentially expressed in LSC-enriched CD34+CD38- and committed CD34+CD38+ progenitor CML (CP and BC) compared to identical BM cell fractions from healthy (NBM) individuals (Supplementary Fig. 8a). Among these, 15 TUG1-interacting miRNAs are either aberrantly expressed or have a defined role in the regulation of proliferation and survival of AML/CML LSCs and/or progenitors (Supplementary Fig. 8a, c).
Halting MIR300-TUG1 interplay selectively eradicated quiescent LSCs in a PDX model of acute and chronic myeloid leukemias
By using as previously described52 a PDX-based approach suitable for determining changes in leukemic LT-HSC survival in vivo, we assessed the effects of TUG1-dependent modulation of MIR300 activity on BM-repopulating qLSCs and the therapeutic relevance of disrupting MIR300-TUG1 interplay. NRG-SGM3 mice (n=4/group) were transplanted with >95% Ph+ CD34+ CML-CP, -AP and -BC cells previously exposed for 48hr to 500nM CpG-scramble (control), CpG-miR-300 and CpG-TUG1shRNA used as single agents or in combination (Fig. 8a). At 4 and 8 weeks after engraftment, BM hCD45+CD34+ progenitors and LSC-enriched hCD45+CD34+CD38- cells were 75-97% reduced in all arms (Fig. 8b, c). Importantly, inhibiting and/or saturating TUG1 MIR300-sponging activity with CpG-TUG1-shRNA and CpG-miR-300 resulted in ∼100% killing of leukemia-initiating (hCD45+CD34+CD38-CD90+) quiescent HSCs (Fig. 8c), barely detectable BCR-ABL1 transcripts and 3.6-fold increased numbers of normal (Ph-) BM cells (Fig. 8d), and strongly reduced numbers of PB and BM CML (CP, AP and BC) hCD45+ cells (Fig. 8e). Notably, the lengthy CML-CP engraftment-time also indicated that decreased hCD34+ cell recovery resulted from reduced numbers of long-term BM-repopulating LSCs (LT-LSC) and not of transplanted CD34+ cells. Thus, disruption of MIR300-TUG1 balance suppresses acute and chronic myeloid leukemia development by selectively and efficiently eliminating nearly all drug-resistant leukemic but not normal quiescent HSCs and progenitors.
DISCUSSION
Regardless of tumor type and stage, altered miRNA expression and PP2A tumor suppressor activity are tightly linked to tumorigenesis and impaired anti-cancer immunity2, 5, 7. Here we showed that therapeutically-exploitable post-transcriptional signals, which are initiated by the tumor-naïve and likely maintained by the tumor-reshaped BMM, controls CML and AML LSC entry/maintenance into quiescence and impairs NK cell immunity. This occurs through the induction of MIR300, a PP2A-activator and tumor suppressor miRNA with dose-dependent anti-proliferative and pro-apoptotic activities which are uncoupled and differentially regulated by the sponge activity of TUG1 lncRNA. Importantly, a dose-dependent target selection mechanism53 allows the sequential activation of MIR300 anti-proliferative and pro-apoptotic functions through the inhibition of CDK6/CCND2 and SET, respectively, in qLSCs and leukemic progenitors.
MIR300 role in LSCs and leukemic progenitors
Expression studies revealed that MIR300 levels are downregulated in CML (CP and BC) and AML (CK) progenitors but not in qLSCs in which MIR300 expression is strongly transcriptionally induced by hypoxia in a C/EBPβ-dependent manner. Accordingly, Fanthom5 (http://fantom.gsc.riken.jp), TCGA (https://portal.gdc.cancer.gov) and tumor-specific miRNA sequence database54, 55 analyses (not shown) indicate that mature and/or precursor MIR300 levels are higher in CD133+ CSCs but lower to undetectable in virtually all primary solid tumors and leukemia subtypes.
Restoration of MIR300 expression arrested proliferation, expanded the G0/G1 cell cycle fraction, strongly impaired survival of dividing CD34+ leukemic stem/progenitor cells and was associated with downregulation of CDK6/CCND2, SET and other growth- and survival-promoting factors (e.g. Twist1 and CTNNB1). MIR300-induced downregulation of CCND1, Twist1 and CTNNB1 was also observed in several solid tumors; however, this was associated with growth arrest/apoptosis but, unclearly, also with the acquisition by tumor cells of a pluripotent stem cell phenotype16, 20, 21, 46.
Although MIR300-induced loss of CCND2/CDK6 (or CCND1/CDK4) likely represents the mechanism by which MIR300 anti-proliferative activity contributes to stemness, since CCND2/CDK6 inhibition characterizes quiescent long-term HSCs (LT-HSCs) and is sufficient to arrest CD34+ leukemic progenitors in G0/G129, 30, impaired expression of Twist1 and CTNNB1 unlikely determines either MIR300 anti-proliferative and pro-apoptotic functions. By contrast it is conceivable that inhibition of SET accounts for MIR300-induced PP2A mediated apoptosis because i) SET downregulation is sufficient for triggering PP2A-mediated apoptosis of acute and chronic leukemia quiescent stem and progenitor cells26, and ii) expression of SET mRNAs lacking the high affinity MIR300-binding site efficiently rescued Ph+ cells from MIR300-induced PP2A-mediated apoptosis. Accordingly, bioinformatics analysis that integrates several algorithms and takes into account also miRNA levels in BM of normal and myeloid leukemic cells to identify miRNA targets56–58, indicated that inhibition of Twist1, CTNNB1 and other PP2A-regulated survival factors (Fig. 4a, Supplementary Fig. 2b, 3) will require levels of free MIR300 higher than those suppressing SET. Thus, SET inhibition represents the main event leading to MIR300-induced PP2A activation.
Notably, the importance of CCND2/CDK6 and SET as key MIR300 effectors is likely not limited to their MIR300-induced post-transcriptional downregulation (Supplementary Fig. 2a right). In fact, SET, CCND2 and/or CDK6 transcription, mRNA nuclear export, translation and/or protein stabilization/activation might also result from MIR300-induced inhibition of other PIP factors (e.g. hnRNPA1, JAK2) and their associated XPO1 nuclear export and SET-stabilizing SETBP1 proteins26,28,59-61.
Because qLSCs tolerate high MIR300 levels and exhibit inactivation of PP2A and activation of JAK2, SET and CTNNB12, 23, it is possible that MIR300 pro-apoptotic activity is turned-off in qLSCs. This raises the questions of whether MIR300 is required for LSC quiescence; how MIR300 is regulated in qLSCs and leukemic progenitors; and, how qLSCs elude MIR300-induced apoptosis.
The requirement of MIR300 for induction and maintenance of LSC quiescence is clearly demonstrated by i) the ability of anti-MIR300 molecules to antagonize MSC-induced inhibition of leukemic cell proliferation, ii) impaired LTC-IC-driven colony formation in the absence of qLSC apoptosis in CD34+ cells exposed to CpG-miR-300concentrations that inhibit CCND2/CDK6 but not SET expression, and iii) inability of MSCs to induce SET downregulation in leukemic cells. Importantly, the notion that MIR300-dependent SET inhibition is induced by hypoxia in CD34+ leukemic stem/progenitor cells (Fig. 5b) and that SET, JAK2 and CTNNB1 are active in qLSCs2, 23, suggest that LSC entrance into quiescence is initiated prior to LSC niching into the BM endosteal area with the lowest O2 tension in which hypoxia-induced TUG1 inactivates MIR300 pro-apoptotic function. Because exposure of leukemic CD34+ stem/progenitor cells to MSC CM and/or exosomes increases MIR300 expression, suppresses their growth in a MIR300-dependent manner and double the number of CD34+ cells in G0 (Fig. 5c), it is likely that the transfer of MSC-derived exosomal MIR300 into LSCs allows MIR300-induced CDK6/CCND2 downregulation and transition into quiescence.
Mechanistically, we showed that hypoxia induces MIR300 transcription in MSCs and CD34+ leukemic cells through reduced LIP (inhibitory) and increased LAP1 (activatory) C/EBPβ that binds/transactivates a hypoxia-sensitive regulatory element 109bp upstream the MIR300 gene. Accordingly, C/EBPβ was found to be expressed in LSCs, induced by hypoxia and to negatively regulate G1/S transition and SET expression62, 63.
To our knowledge, MIR300 is the only cell context-independent tumor suppressor miRNA inducing LSC quiescence, inhibiting innate immunity and triggering LSC and leukemic progenitor apoptosis although other miRNAs regulating CSC self-renewal (e.g. miR-29, miR-125, miR-126, let-7) or survival of quiescent cells (e.g. mir-16 family) have been associated with CSC expansion and maintenance4. For example, miR-126, which targets CDK3, inhibits qLSC re-entry into cycle but does not induce quiescence and, importantly, it is not acting as a tumor suppressor in AML/CML LSCs9, 10. Less clear and controversial is the role of miR-221/222/223 in CSC quiescence since their expression is regulated by MSCs in a tumor- and cell type-specific manner but also increases in response to mitogenic stimuli to induce CDK4/CCND1-dependent cell cycle re-entry4, 11.
MIR300 role in NK cells
MIR300 is also the only tumor-naïve-induced tumor suppressor miRNA that inhibits NK cell-mediated innate anti-cancer immunity while promoting LSC quiescence. NK cells preferentially kill CSCs64–66, including BM-repopulating TKI-resistant BCR-ABL1+ qLSCs (Fig. 6b), and NK cell quantitative and functional impairment is a common event in detected in cancer at diagnosis1, 15. In CML, impaired NK cell immunity also associates with qLSC persistance in TKI-treated patients24, 67 whereas normal levels of activated NK cells characterize TKI-treated patients in sustained TFR67 and account for increased disease-free survival after T-cell-depleted stem cell transplant64, suggesting that NK cell-based therapies may lead to qLSC eradication.
Because MIR300 levels are increased in circulating NK cells from CML patients at diagnosis (Fig. 6b) and, BM hypoxia- and MSC (CM and exosomes)-induced inhibition of NK cell proliferation and anti-tumor activity is essential for immunosurveillance7 and MIR300 induction (Fig. 6c, d), it is plausible that NK cell inhibition in CML and, likely, in other cancer patients may arise from MIR300-mediated signals initiated by the naïve BMM 7 and overriding cytokine-driven NK cell activation 35. Mechanistically we showed that NK cell inhibition likely depends on BMM (hypoxia and MSCs)- induced C/EBPβ-MIR300 signals leading to CCND2/CDK6 and SET downregulation and that BMM-induced NK cell inhibition was recapitulated by MIR300 mimics and rescued by MIR300 RNAi (Fig. 6c-f).
Because loss of CCND2 and SET expression account for inhibition of NK cell proliferation36 and suppression of NK cell anti-tumor cytotoxicity37, it is conceivable that the tumor-reshaped BMM68 may exacerbate but not initiate MIR300 or other signals that suppress cytokine-driven NK cell proliferation and cytotoxicity in leukemia patients. For example, the hypoxia-induced TGFβ1 secretion by leukemic and other BM cells7, 37, 69 and the increase of TGFBR2 in NK cells 69, 70, may augment MIR300-induced CCND2 downregulation by uncoupling IL-2-dependent mitogenic and survival signals71. Seemingly, MSC exosome-induced pre-miR-155 (BIC) downregulation may contribute to MIR300-induced NK cell inhibition by antagonizing miR155-induced C/EBPβ and PP2A downregulation leading to enhanced MIR300 expression and further inhibition of MAPK/AKT-dependent NK cell inactivation, respectively8, 35, 38, 39. By contrast, TUG1 expression seems dispensable for human NK cell survival although, consistent with its activity as a MIR300 sponge, it decreases in NK cells exposed to hypoxia and MSC-derived CM (Supplementary Fig. 6d, e).
Altogether the evidence that i) MIR300 is critical for BMM-induced NK cell inactivation (Fig. 6), ii) clinically-relevant human NK-92 cells34 halve the number of TKI-resistant CML qLSCs, and that iii) ectopic miR-155 selectively and efficiently boost NK cell proliferation and cytotoxicity against mature tumor cells35, 72 and self-renewing LSCs (Fig. 6h), strongly supports the development of NK cell-based immunotherapies with agents that will prevent BMM inhibition and enhance NK cell-mediated selective qLSC and tumor cell killing by simultaneously boosting miR-155 and inhibiting MIR300 activities.
Biologic and therapeutic relevance of the MIR300-TUG1 interplay
TUG1 is an oncogenic lncRNA upregulated in CD34+ stem and progenitor cells from healthy individuals, CML and AML (Fig. 7 and Supplementary Fig. 7) and solid tumor patients in which it has strong diagnostic, prognostic and therapeutic relevance45, 73. In CML-BC and CK-AML qLSCs, TUG1 uncouples MIR300 tumor suppressor function and in a dose-dependent manner selectively suppresses only MIR300 pro-apoptotic activity. Experiments in which we used different CpG-miR-300 or CpG-TUG1-shRNA doses (Fig. 4, 7) indicated that this unprecedented lncRNA function depends on the ability of TUG1 to keep the amount of free MIR300 at levels sufficient for inducing growth arrest but not for triggering PP2A-mediated apoptosis. This implies that the nearly complete and selective (no effects on LT-HSC-driven normal hematopoiesis) killing of qLSCs and leukemic progenitors, which is induced in vitro and in PDXs by the combination of TUG1 shRNAs and MIR300 mimetics (Fig. 7, 8), does not depend on loss of TUG1 survival signals but on the effect of the freed tumor suppressor miRNA (e.g. MIR300) on its mRNA targets. Accordingly, TUG1 loss was associated with G0/G1 arrest and apoptosis whereas its overexpression with induction of proliferation, EMT, drug-resistance and stemness in different solid tumors46, in which its activity was correlated with its sponging miRNA activity46, 51 and with the upregulation of mitogenic and survival factors (e.g. CCND1/2, CDK6, CTNNB1, Twist1) described as MIR300 targets (Fig. 3). TUG1 activity seems to be MIR300-restricted in CML and AML qLSCs but not in leukemic progenitors in which TUG1-shRNAs induced apoptosis of anti-MIR300-treated CD34+ CML and AML cells (Fig. 7e). This is consistent with the association of TUG1 with dismal outcome in AML blasts73 and with the induction of TUG1 by AML1-ETO, MLL-AF4/9 and FLT3-ITD73 (Fig. 7e and Supplementary Fig. 7) and suggests that TUG1 may function as “tumor suppressor miRNA warder” controlling in a cell-type (LSCs and progenitors)-dependent the activity of specific subsets of tumor suppressor miRNAs (e.g. miRNAs inhibiting proliferation or leading to PP2A activation). This is likely facilitated by the recruitment of functionally-related miRNAs into specific miRNA-RBP ternary complexes (e.g. TUG1-hnRNPA1-SET/CCND2/CDK6)59, 61, 74. In this regard, we showed that 15 validated TUG1-interacting miRNAs are differentially expressed in CD34+CD38- and CD34+CD38+ CML (CP and BC) BM cells and have defined roles in solid tumors and AML/CML stem/progenitor cells75 (Supplementary Fig. 8). Indeed, 96.4% and 44.3% of these miRNAs are predicted to shut down CML and AML oncogenic signals, and to share with MIR300 the ability to regulate stemness, TGFβ signaling and activate PP2A (Supplementary Fig. 8). Thus, it is conceivable that balanced TUG1-MIR300 levels are essential for qLSC induction/maintenance whereas insufficient TUG1 expression will lead to qLSC and leukemic progenitor cells apoptosis by freeing miRNAs with similar tumor suppressor activities. Conversely, high TUG1 sponge activity will promote leukemic cell proliferation, survival and qLSC cell cycle re-entry (Supplementary Fig. 7a). However, an aberrant TUG1 increase at levels inhibiting also MIR300 anti-proliferative activity in LSCs may induce LSC exhaustion by impairing entry into quiescence and forcing qLSC re-entry into cycle.
Mechanistically, we showed that TUG1 expression in qLSCs depends on hypoxia-induced TGFβ1 secretion by CD34+ CML progenitors although we cannot exclude that hypoxia-induced TGFBR1/2 upregulation in LSCs and TGFβ1 secretion by other BM cell types (e.g. MSCs)48 contributes to increase TUG1 levels to differentially regulate MIR300 anti-proliferative and pro-apoptotic tumor suppressor functions.
TUG1 can also be induced by Notch signaling76 but the anti-TGFβ Ab-mediated suppression of hypoxia-induced TUG1 levels in qLSCs argues against a Notch significant contribution to TUG1 induction. However, Notch signaling may contribute to qLSC expansion through the RBPJ-mediated miR-155 inhibition that may induce TUG150 and MIR300 expression by preventing FoxM1 and C/EBPβ downregulation77, respectively. Indeed, we showed that TUG1 upregulation was dependent on hypoxia- and TGFβ-induced FoxM1 expression in qLSCs. Accordingly, FoxM1 was described as an oncogene overexpressed in solid tumors and leukemias in which it promotes CSC quiescence, tumor cell survival and cell cycle progression49, 78 upon activation by ERK, CDK6 and PI3K-AKT signals (Supplementary Fig. 7a, d).
In conclusion, the tumor naïve BMM-induced MIR300 tumor suppressor anti-proliferative and PP2A-activating functions support leukemogenesis through the induction of LSC quiescence and inhibition of qLSC killing by cytokine-activated NK cells, respectively. This may represent the initial step leading to formation and initial expansion of the qLSC pool. Once established, the leukemic clone will reshape the BMM to further support disease development and progression. TUG1-MIR300 interaction play a central role in this process levels since altering its ratio leads to the nearly complete and selective PP2A-dependent eradication of acute and chronic myeloid leukemias at qLSC levels in vitro and in PDXs. Thus, this work not only highlights the therapeutic importance of altering MIR300 levels in anti-LSC and NK cell-based approaches for leukemias and, likely, several solid tumors with similar MIR300 and TUG1 expression patterns but also indicates that the activity of a tumor suppressor can be exploited by cancer stem cells to preserve their ability to induce and maintain cancer.
METHODS
Cell lines and primary cells
Cell lines: Ph+ CML-BC K562 and LAMA-84, BM MSC-derived HS-579 and the clinically-relevant NK-9234 cells were cultured in RPMI-1640 medium. NK-92 cultures were supplemented with 150 IU/ml rhIL-2 (Hoffman-La Roche Inc.). The amphotropic-packaging 293T and Phoenix cells were cultured in Dulbecco modified Eagle medium (DMEM). All tissue culture media were supplemented with 10-20% heat-inactivated FBS (Gemini; Invitrogen), 2 mM L-glutamine and 100 U/ml Penicillin/StreptoMycin (Invitrogen). The 32D-BCR-ABL and the IL-3-dependent parental 32Dcl3 mouse myeloid precursors were cultured in Iscove modified Dulbecco Medium (IMDM) supplemented with 10% FBS and rmIL-3 (2ng/ml, Peprotech). Primary cells: Human hematopoietic progenitor (CD34+) and stem cell-enriched fractions (CD34+ CD38-) from healthy and leukemic individuals were isolated from bone marrow (BM), peripheral (PB) or umbilical cord (UCB) blood. Prior to their use, cells were kept (18h) in StemSpanTM CC100 cytokines-supplemented SFMII serum-free medium (StemCell Technologies). Human BM MSCs (hMSCs) from healthy individuals were isolated from BM cells by Ficoll-Hypaque density gradient centrifugation followed by culture in complete human MesenCultTM proliferation medium. MSC purity (>99%) was determined by flow cytometry. Human CD56+CD3- NK cells (purity >95%) from healthy (UCB or PB) and CML individuals were FACS and/or magnetic (Miltenyi Biotech Inc.) cell sorted, or RosetteSep Ab-purified (StemCell Technologies) as described37. Frozen leukemia (CML and AML) specimens were from the Leukemia Tissue Banks located at The University of Maryland (UMB, Baltimore, MD); The Ohio State University (Columbus, OH), Maisonneuve-Rosemont Research Centre, Montreal (Quebec, Canada); Hammersmith Hospital, Imperial College (London, UK); Policlinico Vittorio Emanuele (Catania, Italy); University of Utah, (Salt Lake City, UT); Hematology Institute Charles University (Prague, Czech Republic) and Aarhus University Hospital (Aarhus, Denmark); fresh UCB, NBM and PB (CML and healthy individuals) samples were purchased (Lonza Inc.) or obtained from UMB hospital and Maisonneuve-Rosemont Research Centre, Montreal (Canada).
Cells were treated for the indicated time and schedule with 1-2 μM imatinib mesylate (IM; Novartis), 5 μM 5-Aza-2′-deoxycytidine (5-Aza; Sigma), 250-500 nM CpG-scramble, -MIR300, - anti-MIR300 and -TUG1shRNA oligonucleotides (ODNs) (Beckman Research Institute, City of Hope), 1.25 μg/ml anti-TGFβ neutralizing antibody (1D11; R&D Systems), 10 ng/ml rhIL-12 and 100 ng/ml rhIL-18 (R&D Systems). Where indicated, cells were cultured for the indicated times in hypoxic conditions (1% O2), HS-5 and hMSC CM (100% vol/vol), or in medium supplemented with MSC-derived exosomes (50-100 μg/ml). During treatments, viable cells were enumerated by Trypan Blue exclusion test.
Mouse NK1.1+CD3- NK cells and Lin-Sca+Kit+ (LSK) cells were purified from spleen of 8-10wk-old healthy and leukemic (6-8 wk-induced) SCL-tTA-BCR-ABL tg mice80 and wt and lck-miR-155-tg mice39 by microbeads negative selection (Miltenyi Biotech Inc.) and cell sorting.
Plasmids
pCDH-MIR300 (hsa-MIR300) and pCDH-miR-381 (hsa-miR-381-3p) pCDH-MIR300 was generated by subcloning a 490bp NheI-BamHI PCR fragment encompassing the mature hsa-MIR300 into the pCDH-CMV-MCS-EF1-copGFP-puro (SBI) vector. Sequence was confirmed by sequencing. pCDH-miR-381 was generated by subcloning a double-stranded (ds) synthetic ODN containing the human pre-miR-381 sequence flanked by NheI and BamHI restriction sites at the 5’- and 3’-end, respectively, into the pCDH-CMV-MCS-EF1-copGFP-puro vector. pSIH-H1-Zip- MIR300: To knockdown MIR300, a 5’-EcoRI and 3’-BamHI-flanked MIR300 antisense dsODN was directionally cloned into the pSIH1-H1-copGFP vector (SBI). pSIH-H1-copGFP-shTUG1 (TUG1 shRNA): a shRNA cassette containing the targeted nt 4571 to 4589 of hTUG1 RNA76 was subcloned into pSIH-H1-copGFP vector (SBI). A non-functional scrambled TUG1 shRNA was used as a control. pLenti-TUG1: the hTUG1 into the pLenti-GIII-CMV-GFP-2A-Puro-based vector was from Applied Biological Materials Inc. pGFP/Luc-based MIR300 promoter constructs: p513-GFP/Luc, p245-GFP/Luc, and p109-GFP/Luc were generated by cloning the -522 bp, -245 bp and -109 bp regions upstream the hsa-MIR300 gene locus (Sequence ID: NC_000014.9) into pGreenFire1 (pGFP/Luc, pGF1; SBI) reporter vector. The regions upstream the hsa-MIR300 gene were PCR amplified from K562 DNA and cloned into the pGF1 EcoRI-blunted site. p109-C/EBPBmut-GFP/Luc: an EcoRI-containing double-stranded synthetic oligonucleotide spanning the -109bp MIR300 regulatory (region and containing the T/G and C/G mutated C/EBPβ consensus binding sites located at positions -64 and -46, respectively was subcloned into EcoRI-digested pGF1 vector. pCDH-Flag-SET, fluorescent ubiquitination-based cell cycle indicator reporter pCDH-FUCCI2BL, MigR1-ΔuORF-C/EBPβ-ERTAM and MigR1-ΔuORF-C/EBPα-HA constructs were described32, 81–83. The pCDH-Flag-SET 3’UTRwt-GFP and pCDH-Flag-SET Δ3’UTR-GFP: wild-type (627bp) or 3’-deleted (513bp) SET 3’UTRs were PCR amplified from K562 cDNA using a common EcoRI-linked 5’-primer and specific BamHI-linked 3’-primers for the wild-type and deleted SET mRNA 3’UTR. The PCR products were cloned into the pCDH-Flag-SET plasmid and sequenced.
MSC Conditioned Medium (CM) and Exosomes purification
HS-5- and hMSC-derived CM was obtained by culturing cells in complete RPMI (24h) and StemSpanTM SFMII (48h) medium, respectively. Alix+CD63+ exosomes were purified by differential centrifugation (3,000xg, 15 min; 10,000xg, 10 min; and 100,000xg, 70 min) from HS-5 CM cultured in exosome-free FBS-supplemented medium84, 85. When required, exosomes were precipitated (Exo-Quick; SBI) prior to isolation by ultracentrifugation.
Flow cytometry and cell sorting
CD34+ and CD34+CD38- fractions were magnetic-(CD34 MicroBead kit; Miltenyi Biotec) and/or FACS-(αCD34 APC/PE and αCD38 PE/Cy7 Abs, BD Biosciences) purified. Primary human NK cells were sorted using Alexa fluor 488 αhCD56, PE-Cy5 αhCD19 and PE-Cy7 αhCD3 Abs (BD Bioscience) and their purity assessed by PE αCD56 (Beckman Coulter, Life Sciences) and APC-eFluor780 αCD3 (eBioscience) Ab staining. Mouse NK cells were sorted using αNK1.1 APC and αCD3 FITC Abs (BD Bioscience) and LSK cells using αLin- (lineage cocktail) Pacific Blue (Biolegend), αc-Kit PE-Cy7 and αSca1 APC Abs (eBioscience). hMSCs purity was assessed by flow cytometry using anti-CD34 and CD45 FITC, CD73 PE-Cy7, CD105 Alexa 647, CD44 Percp-Cy5.5 and CD90 PE Abs (BD Biosciences). Apoptosis was quantified by FACS upon staining cells with PE Annexin-V and 7-ADD (BD Biosciences). Cells were sorted using the FACS Aria II (BD Biosciences). Data acquisition and analyses were performed at the UMBCCC Flow Cytometry Facility. Data from LSRII or CANTO II flow cytometers (BD Biosciences) were analyzed by using either the FlowJo v8.8.7 or Diva v6.1.2 software.
Lentiviral and retroviral transduction
Lentiviral or retroviral particles were produced by transient calcium phosphate transfection (ProFection mammalian transfection System; Promega) of 293T and Phoenix cells, respectively2. Briefly, viral supernatant was collected 24 and 48h after transfection and directly used or concentrated by PEG precipitation. Viral titer was determined by flow cytometry by calculating number of GFP+ 293T cells exposed (48h) to different viral dilutions. One-to-three spinoculation rounds were used for cell line infection whereas a single spinoculation with diluted viral supernatants (MOI=6) was used for primary cells. Viral supernatants were supplemented with polybrene (4 μg/ml). FACS sorting or puromycin selection was initiated 48h after transduction.
Cell cycle analysis
Cell cycle analysis was performed by 4’,6’-diamidine-2-phenylindole (DAPI)/Ki-67 staining of CpG-scramble- or CpG-MIR300-treated (500nM; 72h) primary CD34+ CML (CP and BC) and UBC cells as described83. Briefly, cells were fixed, permeabilized (Cytofix/Cytoperm kit, BD Biosciences) and, Ki-67 PE (Biolegend) and DAPI (Sigma) stained. Alternatively, cell cycle analysis was performed on pCDH-Fucci2BL-infected cells and subjected to flow-cytometric analysis. LAMA-FUCCI2BL+ CML-BC cells were exposed to 500nM CpG-scramble and CpG-miR-300oligonucleotides after being G1/S synchronized (4-6 µM aphidicolin, 6h). CpG-oligo-treated (48h) FUCCI2BL+ cells were subjected to live cell imaging for 48h and analyzed as described83. Briefly, CpG-scramble- or CpG-MIR300-treated-treated cells in G0/G1 were mVenus- PE-Texas-Red+, G1/S cells were mVenus+ PE-Texas-Red+, and S/G2/M cells were mVenus+ PE-Texas-Red-.
Long-term culture-initiating cell (LTC-IC) and colony-forming cell (CFC)/replating assays
CD34+ CML (CP and BC), AML and UCB cells were lentivirally-transduced/GFP-sorted and/or treated (500 nM, 3d) with CpG-ODN in order to ectopically express either MIR300 or anti-MIR300 RNAs prior to use them in LTC-IC and CFC assays2. Vector-transduced and CpG-scramble-treated cells served as controls. LTC-ICs: 2×105 CML and 2×103 UCB CD34+ cells were cultured with a 1:1 mixture of irradiated (80 Gy) IL-3/G-CSF–producing M2-10B4 and IL-3/KL-producing SI/SI murine fibroblasts in MyeloCult H5100 (StemCell Technologies) supplemented with hydrocortisone. Medium was replaced after 7 days, followed by weekly half-medium changes, and fresh 500 nM CpG-ODN where required. After 6 wk, adherent and floating cells were harvested, and 5×103 CML and 2×103 for UCB cells were plated into cytokine (KL, G-CSF, GM-CSF, IL-3, IL-6)-supplemented MethoCult H4435. LTC-IC-derived colonies were scored after 14 days. CFCs and CFC/Replating assays: CD34+ and CD34+ CD38- CML/AML (5×103) and UCB (2×103) cells were seeded in MethoCult H4435 supplemented with KL, G-CSF, GM-CSF, IL-3, IL-6. After 2 wk, colonies were scored and, if necessary, 104 cells underwent two rounds of serial replating/scoring.
CFSE (or eFluor670)-mediated tracking of cell division
Carboxyfluorescein diacetate succinimidyl diester (CFSE; CellTrace CFSE Proliferation Kit; Invitrogen) or eFluor670 (Cell proliferation Dye eFluor670; eBioscience)-stained cells were FACS-sorted to isolate the highest fluorescent peak, treated as indicated, harvested after 4-5 days and counterstained with Near-IR fluorescent Dye (Live/Dead Cell Stain Kit, Invitrogen) to determine the number of viable dividing (CFSE/NearIR-) and quiescent cells (CFSEmax/NearIR-). Where indicated, cells were stained with anti-CD34 APC and sorted into dividing and quiescent sub-populations. Quiescent (CFSEmaxCD34+) and dividing cells in each peak were reported as a fraction of the initial number of CD34+ cells2. For NK cells, 4 and 7 day-cultured CFSE-labeled NK-92 and primary NK cells, respectively, were αCD56 APC Ab- and Aqua Fluorescent Dye (Live/Dead Cell Stain Kit, Invitrogen)-stained, and proliferation quantitated as fold changes of CFSE Mean Fluorescence Intensity (MFI) in living CD56+ NK cells.
Live Cell Imaging and Confocal Microscopy
Wide field fluorescence imaging of CpG-ODN-treated pCDH-FUCCI2BL-transduced LAMA-84 cells was executed with an Olympus Vivaview digital camera, maintained at humidified atmosphere 37°C with 5% CO2, mounted into a microscope-equipped incubator with green and red fluorescence filters. Emission spectral ranges were green-narrow (490-540 nm) and red-narrow (570-620 nm). Imaging was carried out for 24 or 48 hours. Confocal fluorescence images were acquired using a LEISS LSM510 inverted confocal microscope equipped with 40X/1.2-NA water immersion objective with 0.55micron pixel size. Cell cycle phases were determined by plotting average red and green fluorescent intensity over time using Prism 6.0d software. The Vivaview raw images were reconstructed using ImageJ v5.2.5 and Adobe CS6 software.
PP2A phosphatase assay
PP2A assays were performed using the malachite green–based PP2A immunoprecipitation (IP) phosphatase assay kit (Millipore) as described2. Briefly, protein lysates (50μg) in 100 μl of 20mM HEPES pH 7.0/100 mM NaCl, 5 μg αPP2Ac Ab (Millipore) and 25 μl protein A–agarose was added to 400 μl of 50 mM Tris pH 7.0, 100 mM CaCl2. Immunoprecipitates were carried out at 4°C for 2h and used in the phosphatase reaction according to the manufacturer’s protocol.
Luciferase Assay
Transcriptional activity of the intergenic 513 bp region upstream the MIR300 gene was investigated by luciferase assay using the pGreenFire-Lenti-Reporter system (pGF1; SBI) in cell lines and primary CD34+ CML-BC cells. Briefly, cells were transduced with constructs containing wild type full-length 513 (p513-GFP/Luc) and, deleted 245 (p245-GFP/Luc) and 109 (p109-GFP/Luc) base pair-long region upstream the MIR300 gene locus, or with the p109-GFP/Luc construct mutated in the two C/EBPβ binding sites (p109-CEBPβmut-GFP/Luc). The empty vector (pGFP-Luc; pGF1) was used as a negative control. After 48h, cells were lysed and luciferase activity was determined by Pierce Firefly Luciferase Flash Assay Kit (Thermo Scientific).
Cytotoxicity assays
A flow cytometry-based killing assay was performed using K562 cells as targets86. Briefly, HS-5-derived CM (100% vol/vol)- or exosomes (50-100 ug/ml)-preconditioned (36h in 50-150 IU/ml IL-2) NK-92 cells were co-incubated (3.5h) with CFSE-labeled K562 cells at ratio 5:1. Killing was evaluated by assessing the percentage of Annexin-V+ cells on CD56- and/or CFSE+ target cells. When CpG-ODNs were used, NK-92 cells were preconditioned (36h) in medium lacking IL-2. NK-92 cell-killing of qLSCs was performed using CFSE-labeled CD34+ CML-BC cells cultured (4d) in cytokine-supplemented StemSpanTM II serum-free medium exposed (18h) to NK-92 cells. Spontaneous cytotoxicity toward qLSCs was determined by FACS-mediated evaluation of LSC numbers in Annexin VnegCD34+CFSEmax cells before and after exposure to NK-92 cells. wt and miR-155-tg murine NK cell-mediated killing of LSK cells was determined by assessing CFC/replating efficiency in FACS-sorted CFSE+ Lin-Sca1+Kit1+LSK cells exposed to NK1.1+CD3-NK cells (18h).
Real time RT-PCR
Total RNA was extracted using miRNeasy micro Kit (Qiagen Inc.) or Trizol (Invitrogen) and reverse transcribed using a standard cDNA synthesis or the TaqMan MicroRNA Reverse Transcription Kit with mouse and/or human-specific sets of primers/probe for BCR-ABL1, CEBPB, FOXM1, TUG1, IFNγ, pri-MIR300, MIR300, miR-381, pre-miR-155 (BIC), 18S, RNU44, RNU6B and snoRNA202 (Applied Biosystems). PCR Reactions were performed using a StepOnePlus Real Time PCR System (Applied Biosystems). Data were analyzed according to the comparative CT method using RNU44, RNU6B, 18S and snoRNA202 transcripts as an internal control. Results are expressed as fold change of mean±SEM.
Immunoblot analysis
Lysate from cell lines and primary cells were subjected to SDS-PAGE and Western blotting as described2. The antibodies used were: anti-Actin, anti-Myc, anti-C/EBPβ, anti-Alix, anti-Twist, and anti-CD63 (Santa Cruz Biotechnology); anti-GRB2 (Transduction Laboratories); anti-JAK2, anti-β-catenin, anti-FoxM1, anti-CDK6, anti–pJAK2Y1007/1008 and anti-CCND2 (Cell Signaling); anti-SET (Globozymes); anti-ABL (Ab-3), anti-CCND1/2, anti-PY (4G10) and anti-PP2AcY307 (EMD); anti-hnRNPA1 (Abcam); and anti-Flag (M2; Sigma).
LSC engraftment and disease development in NRG-SGM3 mice
In vivo analysis of ex vivo-treated BM-repopulating LSCs was performed as described52. 106 BM CD34+ CML (n=3; >95% Ph+; CP, AP and BC) cells, treated (500 nM, 48h) ex vivo with CpG-ODNs, were intravenously (iv) injected (4 mice/treatment group/patient sample) into sub-lethally irradiated (2.6Gy) 6-8 week-old NOD.Cg-Rag1tm1Mom Il2rgtm1Wjl Tg(CMV-IL3, CSF2,KITLG)1Eav/J (NRG-SGM3; Jackson Laboratory). Engraftment was assessed by anti-human CD45 (BD Biosciences) flow staining of intra-femur BM aspirates2 at 2 wk in CpG-scramble control groups and at 2, 4 and 12 wk post-transplant in CpG-MIR300, -TUG1shRNA and MIR300+TUG1shRNA CML-BC, AP and CP, respectively. Disease evolution and effect of the various CpG-ODNs were assessed by FACS-mediated analysis of hCD45+ cells in BM and PB, and of hCD45+CD34+ proliferating progenitors, hCD45+CD34+CD38− and hCD45+CD34+CD38−CD90+ CSC-enriched cell fractions with repopulating activity in BM aspirates at engraftment time and/or 8 wk post-engraftment (end point). BCR-ABL1 transcript levels were monitored by RT-qPCR-mediated (Ipsogen) analysis of BCR-ABL1/human Abl1 ratios in total RNA samples derived from total BM of mice euthanized at 8 wk post-engraftment. Interphase FISH was performed on hCD45+-sorted BM cells, obtained at 8 wk post-engraftment, by denaturing (75°C, 2 min) slides in 70% formamide/2x standard sodium citrate (SSC) and ET-OH dehydrated. Interphase cells were hybridized (18h, 37°C) with BCR-ABL1 dual-color double-fusion probe set (10 UL; Cytocell) and DAPI counterstained. Slides were analyzed using an Olympus fluorescence microscope, and images taken with a CCD camera using a FISH imaging software (MetaSystems). 200-interphase cells/sample were analyzed and FISH negativity (Ph-) was defined as the absence of BCR-ABL fusion signal.
Oligonucleotides and Primers
The partially-phosphothioated (*) 2′-O-Methyl ODNs were linked using 5 units of C3 carbon chain linker, (CH2)3 (X). The ODNs were synthesized by the DNA/RNA Synthesis Laboratory, Beckman Research Institute of the City of Hope.
CpG-oligonucleotides and probes
CpG-MIR300: 5’-g*g*tgcatcgatgcagg*g*g*g*gxxxxxuauacaagggcagacucucucu-3’
CpG-anti-MIR300: 5’-g*g*tgcatcgatgcagg*g*g*g*gxxxxxagagagagucugcccuuguaua-3’
CpG-TUG1 shRNA: 5’-g*g*tgcatcgatgcagg*g*g*g*gxxxxxuuacucugggcuucugcac-3’
CpG-scramble: 5’-g*g*tgcatcgatgcagg*g*g*g*gxxxxxguagaaccguacucgucacuua-3’
pCDH-CMV-MCS-EF1-copGFP-puro miRNA constructs
MIR300(F): 5’-gctagctgtgactagttgtaccttag-3’; MIR300 (R): 5’-gatcctctcttccagaaagttcttg-3’
miR-381(+):5’-ctagctacgcaaagcgaggttgccctttgtatattcggtttattgacatggaatatacaagggcaagctctctgtgagtag-3’
miR-381(-):5’-gatcctactcacagagagcttgcccttgtatattccatgtcaataaaccgaatatacaaagggcaacctcgctttaagta-3’
pSIH1-H1-copGFP constructs
Zip-MIR300 ODN: 5’-gatatgttcccgtctgagagagagaaggacagtcttctctctcagacgggaacatataaaaacttaa-3’
shTUG1(+): 5’-gatccgtgcagaagcccagagtaattcaagagattactctgggcttctgcactttttg-3’
shTUG1 (-): 5’-aattcaaaaagtgcagaagcccagagtaatctcttgaattactctgggcttctgcacg-3’
pCDH-Flag-SET constructs
SET3’UTRwt-GFP(F): 5’-gaattctagcttttttcctcctttctctgtata-3’;
SET3’UTRwt-GFP(R): 5’-cggatccgtatacaagtcaaact-3’
SETΔ3’UTR-GFP(F): 5’-ggatccagagaaaagcatccaa-3’;
SETΔ3’UTR-GFP(R): 5’-cggatccgtatacaagtcaaact-3’
pGreenFire1
p522-pGF1(F): 5’-ccggaattccggcacgttttcagtatcaaatgct-3’
p245-pGF1(F): 5’-ccggaattccggtgaacctcttttactgtgactagttg-3’
p109-pGF1(F): 5’-ccggaattccggggtgtgctgctctcaccat-3’
Common(R): 5’-tgctctagagcaaatgatggcagtgacaggaa-3’
p109-C/EBPβ mutant construct
(+) strand ODN: 5’-aattc ggtgtgctgc tctcaccatg cagatcccat ctgtgtctct aaggctggct cctggagctg gtgggaactt agtcacagag gaaatggcct tcctgtcact gccatcattg-3’
(-) strand ODN: 5’-caatgatggc agtgacagga aggccatttc ctctgtgact aagttcccac cagctccagg agccagcctt agagacaca gatgggatct gcatggtgag agcagcacac cgaatt-3’
Bioinformatics tools
Statistically significant (p<0.05 with FDR correction) predicted and validated hsa-MIR300 mRNA targets according to mRNA target function and MIR300 doses were identified using DIANAmicroT-CDS (diana.imis.athena-innovation.gr), ComiR (benoslab.pitt.edu/comir), (cosbi.ee.ncku.edu.tw/CSmiRTar/) and mirDIP 4.1 (ophid.utoronto.ca/mirDIP/). Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) analyses to define the biological pathways and the functional roles of miRNA-300 and TUG-1 targets were performed using miR-Path v.3 (snf-515788.vm.okeanos.grnet.gr). miRTar algorithm (miRTar.mbc.nctu.edu.tw/) was used to predict mRNA targets of wild type and ADAR-1-edited miR-381-3p. MIR300 and TUG-1 individual gene expression profiles were obtained from curated datasets in the Gene Expression Omnibus (GEO) repository (ncbi.nlm.nih.gov/geoprofiles/). TUG1 levels in normal myelopoiesis and different AML subtypes were analyzed from BloodSpot database (servers.binf.ku.dk/bloodspot/) of healthy and malignant hematopoiesis. Integration of StarBase v2.0 (starbase.sysu.edu.cn/starbase2/index.php) database with the RNAseq MiRbase data from CD34+CD38-, CD34+D38+ CML and NBM cells was used to identify TUG-1 sponged miRNAs.
Statistics
P values were calculated by Student’s t-test (GraphPad Prism v6.0). Results are shown as mean ± SEM. A P value less than 0.05 was considered significant (∗P< 0.05, ∗∗P < 0.01, ∗∗∗ P < 0.001, ∗∗∗∗ P < 0.0001). Mixed models’ approach, the split-plot design, was used to assess differences in percent cell across three treatment groups. The percent cell change was estimated using a model with two main effects for treatment and stage of a cell cycle, as well as their interaction. Tests of fixed effects revealed that interaction of treatment and stage of a cell cycle is highly significant, p=0.01, the two main effects, treatment and stage, should be interpreted with caution, the p-values are 1.0 and <0.0001, respectively. The differences in average percent cell were tested across groups and sliced at a particular stage of cell cycle.
Study approval
Patient samples were banked at different Institutions and not collected for this study, which was carried out with a waiver of informed consent and approval from University of Maryland IRB. UCB units were collected at University of Maryland Medical Center with IRB approved protocol and written patient consent. Animal studies were performed according to IRB- and IACUC-approved protocols.
FINANCIAL SUPPORT
NHI-NCI CA163800-01 (D.P.), ACS IRG16-123-13 Pilot Grant (R.T.), Maryland Cigarette Restitution Funds (UMB Greenebaum Comprehensive Cancer Center), MSMT CZ LH15104 (K.P.M) and CRS 16255 (M.G.) for patient sample procurement and processing. S.W. is supported by a grant from the China NSFC 81872924 and Scholarship Council).
COI DISCLOSURE
All of the other authors declare no conflicts of interest related to this work
AUTHORS CONTRIBUTIONS
Conceptualization and Funding Acquisition: D.P. Writing Original Draft: D.P. Investigation (>75% effort): G.S., R.T. (equal contribution) Supporting Investigation (<25% effort): L.S., S.W., J.J.E., J.H., P.N., E.A.K. K.S., M.M.M., Y.Z.; Provided Resources: C.G.P., G.P., C.H.M.J., F.S., P.V., G.N., P.M., A.R., R.G., D.C.R., M.G., P.H., M.D., G.F., C.H., F.D., D.M., J.A., G.M., J.Q., K.M.P., Y.Z., X.F., M.R.B. and B.C.
SUPPLEMENTARY FIGURES
ACKNOWLEDGEMENTS
We thank Goloubeva O. for statistical data analysis; Underwood K.F., Mauban J.R.H., Pomicter T., Duong V.H., Emadi A., Eiring A.M, and Glynn-Cunningham N.M. for patient sample procurement, technical assistance, reagents, and/or critical reading of the manuscript.
REFERENCES
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.
- 42.
- 43.
- 44.↵
- 45.↵
- 46.↵
- 47.↵
- 48.↵
- 49.↵
- 50.↵
- 51.↵
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.
- 58.↵
- 59.↵
- 60.
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.
- 83.↵
- 84.↵
- 85.↵
- 86.↵