miR-22 has a potent anti-tumour role with therapeutic potential in acute myeloid leukaemia

MicroRNAs are subject to precise regulation and have key roles in tumorigenesis. In contrast to the oncogenic role of miR-22 reported in myelodysplastic syndrome (MDS) and breast cancer, here we show that miR-22 is an essential anti-tumour gatekeeper in de novo acute myeloid leukaemia (AML) where it is significantly downregulated. Forced expression of miR-22 significantly suppresses leukaemic cell viability and growth in vitro, and substantially inhibits leukaemia development and maintenance in vivo. Mechanistically, miR-22 targets multiple oncogenes, including CRTC1, FLT3 and MYCBP, and thus represses the CREB and MYC pathways. The downregulation of miR-22 in AML is caused by TET1/GFI1/EZH2/SIN3A-mediated epigenetic repression and/or DNA copy-number loss. Furthermore, nanoparticles carrying miR-22 oligos significantly inhibit leukaemia progression in vivo. Together, our study uncovers a TET1/GFI1/EZH2/SIN3A/miR-22/CREB-MYC signalling circuit and thereby provides insights into epigenetic/genetic mechanisms underlying the pathogenesis of AML, and also highlights the clinical potential of miR-22-based AML therapy.

A s one of the most common and fatal forms of hematopoietic malignancies, acute myeloid leukaemia (AML) is frequently associated with diverse chromosome translocations (for example t(11q23)/MLL-rearrangements, t(15;17)/PML-RARA and t(8;21)/AML1-ETO) and molecular abnormalities (for example, internal tandem duplications of FLT3 (FLT3-ITD) and mutations in nucleophosmin (NPM1c þ )) 1 . Despite intensive chemotherapies, the majority of patients with AML fail to survive longer than 5 years 2,3 . Thus, development of effective therapeutic strategies based on a better understanding of the molecular mechanisms underlying the pathogenesis of AML is urgently needed.
MicroRNAs (miRNAs) are a class of small, non-coding RNAs that post-transcriptionally regulate gene expression 4 . Individual miRNAs may play distinct roles in cancers originating from different tissues or even from different lineages of hematopoietic cells 4 . It is unclear whether a single miRNA can play distinct roles between malignancies originating from the same hematopoietic lineage, such as de novo AML and myelodysplastic syndrome (MDS). Although around 30% of MDS cases transform to AML, the genetic and epigenetic landscapes of MDS or MDS-derived AML are largely different from those of de novo AML 5,6 . MDS and MDS-derived AML are more responsive to hypomethylating agents than de novo AML 7 . The molecular mechanisms underlying the distinct pathogenesis and drug response between MDS (or MDS-derived AML) and de novo AML remain unclear.
The ten-eleven translocation (Tet1/2/3) proteins play critical transcriptional regulatory roles in normal developmental processes as activators or repressors [8][9][10] . In contrast to the frequent loss-offunction mutations and tumour-suppressor role of TET2 observed in hematopoietic malignancies [11][12][13] , we recently reported that TET1 plays an essential oncogenic role in MLL-rearranged AML where it activates expression of homeobox genes 14 . However, it is unknown whether TET1 can also function as a transcriptional repressor in cancer. Moreover, Tet1-mediated regulation of miRNA expression has rarely been studied 10 .
In the present study, we demonstrate that miR-22, an oncogenic miRNA reported in breast cancer and MDS 15,16 , is significantly downregulated in most cases of de novo AML due to TET1/GFI1/EZH2/SIN3A-mediated epigenetic repression and/or DNA copy-number loss. miR-22 functions as an essential antitumour gatekeeper in various AML and holds great therapeutic potential to treat AML.

Results
The downregulation of miR-22 in de novo AML. Through Exiqon miRNA array profiling, we previously identified a set of miRNAs, such as miR-150, miR-148a, miR-29a, miR-29b, miR-184, miR-342, miR-423 and miR-22, which are significantly downregulated in AML compared with normal controls 17 . Here we showed that among all the above miRNAs, miR-150 and especially miR-22 exhibited the most significant and consistent inhibitory effect on MLL-AF9-induced cell immortalization in colonyforming/replating assays (CFA) (Supplementary Fig. 1a). In contrast to the reported upregulation of miR-22 in MDS 16 , our original microarray data 17 (Fig. 1a,b) and new quantitative PCRindependent validation data ( Supplementary Fig. 1b) demonstrated a significant and global downregulation of miR-22 in de novo AML relative to normal controls. Notably, miR-22 is significantly downregulated in AML samples (Po0.05) compared with all three sub-populations of normal control cells, that is, normal CD34 þ hematopoietic stem/progenitor cells (HSPCs), CD33 þ myeloid progenitor cells, or mononuclear cells (MNCs) (Fig. 1a). Expression of miR-22 is significantly downregulated in all or the majority of individual subsets of AML samples than in the normal CD33 þ or CD34 þ cell samples (Fig. 1b).
To rule out the possibility that the inhibitory effect of miR-22 shown in Supplementary Fig. 1a was due to a non-specific effect of our miR-22 construct, we included the MSCV-PIG-miR-22 construct from Song et al. 16 in a repeated CFA. Both miR-22 constructs dramatically inhibited MLL-AF9-induced colony formation (Fig. 1c). As the 'seed' sequences at the 5 0 end of individual miRNAs are essential for the miRNA-target binding 18 , we also mutated the 6-bases 'seed' sequence of miR-22 and found that the miR-22 mutant did not inhibit colony formation anymore (Fig. 1c). In human AML cells, forced expression of miR-22, but not miR-22 mutant, significantly inhibited cell viability and growth/proliferation, while promoting apoptosis ( Supplementary Fig. 1c,d).
In accordance with the potential anti-tumour function of miR-22 in AML, miR-22 was expressed at a significantly higher level (Po0.05) in human normal CD33 þ myeloid progenitor cells than in more immature CD34 þ HSPCs or MNC cells (a mixed population containing both primitive progenitors and committed cells) (Fig. 1a,b), implying that miR-22 is upregulated during normal myelopoiesis. Similarly, we showed that miR-22 was also expressed at a significantly higher level in mouse normal bone marrow (BM) myeloid (Gr-1 þ /Mac-1 þ ) cells, relative to lineage negative (Lin À ) progenitor cells, long-term hematopoietic stem cells (LT-HSCs), short-term HSCs (ST-HSCs), and committed progenitors (CPs) ( Supplementary Fig. 1e), further suggesting that miR-22 is upregulated in normal myelopoiesis.
The anti-tumour effect of miR-22 in the pathogenesis of AML. Through bone marrow transplantation (BMT) assays, we showed that forced expression of miR-22 (but not miR-22 mutant) dramatically blocked MLL-AF9 (MA9)-mediated leukemogenesis in primary BMT recipient mice, with a more potent inhibitory effect than miR-150 ( Fig. 1e; Supplementary Fig. 2a). All MA9 þ miR-22 mice exhibited normal morphologies in peripheral blood (PB), BM, spleen and liver tissues (Fig. 1f), with a substantially reduced c-Kit þ blast cell population in BM ( Supplementary Fig. 2b). Forced expression of miR-22 also almost completely inhibited leukemogenesis induced by MLL-AF10 ( Fig. 1g; Supplementary Fig. 2a). Conversely, miR-22 knockout significantly promoted AML1-ETO9a (AE9a)-induced AML (Fig. 1h). Thus, the repression of miR-22 is critical for the development of primary AML. Notably, forced expression of miR-22 in MLL-AF9 and MLL-AF10 leukaemia mouse models caused only a 2-3-fold increase in miR-22 expression level ( Supplementary Fig. 2a), in a degree comparable to the difference in miR-22 expression levels between human AML samples and normal controls (Fig. 1a), suggesting that a 2-3-fold change in miR-22 expression level appears to be able to exert significant physiological or pathological effects.
To examine whether the maintenance of AML is also dependent on the repression of miR-22, we performed secondary BMT assays. Forced expression of miR-22 remarkably inhibited progression of MLL-AF9-, AE9a-or FLT3-ITD/NPM1c þinduced AML in secondary recipient mice (Fig. 2a-d), resulting in largely normal morphologies in PB, BM, spleen and liver tissues ( Fig. 2b; Supplementary Fig. 2c). Collectively, our findings demonstrate that miR-22 is a pivotal anti-tumour gatekeeper in both development and maintenance of various AML.
Identification of critical target genes of miR-22 in AML. To identify potential targets of miR-22 in AML, we performed a series of data analysis. Analysis of In-house_81S (ref. 21) and TCGA_177S (ref. 22) data sets revealed a total of 999 genes exhibiting significant inverse correlations with miR-22 in expression. Of them, 137 genes, including 21 potential targets of miR-22 as predicted by TargetScan 18 (Supplementary Table 1), were significantly upregulated in both human and mouse AML compared with normal controls as detected in two additional inhouse data sets 14,23 . Among the 21 potential targets, CRTC1, ETV6 and FLT3 are known oncogenes [24][25][26][27][28][29] . We then focused on these three genes, along with MYCBP that encodes the MYC-binding protein and is an experimentally validated target of miR-22 (ref. 30) although due to a technical issue it was not shown  in the 21-gene list (Supplementary Table 1), for further studies. As expected, all four genes were significantly downregulated in expression by ectopic expression of miR-22 in human MONO-MAC-6/t(9;11) cells (Fig. 3a). The coincidence of downregulation of those genes and upregulation of miR-22 was also observed in mouse MLL-ENL-ERtm cells, a leukaemic cell line with an inducible MLL-ENL derivative 31 , when MLL-ENL was depleted by 4-hydroxy-tamoxifen (4-OHT) withdrawal ( Fig. 3b; Supplementary Fig. 3a). While MLL-AF9 remarkably promoted expression of those four genes in mouse BM progenitor cells, coexpressed miR-22 reversed the upregulation (Fig. 3c). In leukaemia BM blast cells of mice with MLL-AF9-induced AML, the expression of Crtc1, Flt3 and Mycbp, but not Etv6, was significantly downregulated by co-expressed miR-22 (but not by miR-22 mutant) (Fig. 3d). Because miR-22-mediated downregulation of Etv6 could be observed only in the in vitro models (Fig. 3a-c), but not in the in vivo model (Fig. 3d), which was probably due to the difference between in vitro and in vivo microenvironments, we decided to focus on the three target genes (that is, Crtc1, Flt3 and Mycbp) that showed consistent patterns between in vitro and in vivo for further studies. The repression of Crtc1, Flt3 and Mycbp was also found in leukaemia BM cells of mice with AE9a or FLT3-ITD/NPM1c þ -induced AML (Fig. 3e,f). As Mycbp is already a known target of miR-22 (ref. 30), here we further confirmed that FLT3 and CRTC1 are also direct targets of miR-22 (Fig. 3g,h). The downregulation of CRTC1, FLT3 and MYCBP by miR-22 at the protein level was confirmed in both human and mouse leukaemic cells ( Supplementary Fig. 3b,c). Overexpression of miR-22 had no significant influence on the level of leukaemia fusion genes ( Supplementary Fig. 3d).
Co-expression of the coding region (CDS) of each of the three target genes (that is, CRTC1, FLT3 and MYCBP) largely reversed the effects of miR-22 on cell viability, apoptosis and proliferation ( Fig. 4a-e). More importantly, in vivo BMT assays showed that co-expressing CRTC1, FLT3 or MYCBP largely rescued the inhibitory effect of miR-22 on leukemogenesis (Fig. 4f, Table 2; Supplementary Fig. 3f). In leukaemic BM blast cells from primary and secondary BMT recipients, overexpression of miR-22 (but not miR-22 mutant) significantly downregulated expression of Cdk6 and Hoxa7, while upregulating Rgs2, which could be reversed by co-expressing CRTC1 ( Supplementary Fig. 3g,h).
These results suggest that miR-22 represses the CREB signalling pathway in AML by targeting CRTC1.
MYCBP, a MYC-binding protein, is essential for MYC-mediated gene regulation 30 . FLT3 is an upstream regulator of MYC 17 . In leukaemic BM cells, forced expression of miR-22, but not miR-22 mutant, significantly repressed expression of MYC downstream oncogenic targets Bmi1, Fasn and Hmga1 (refs [36][37][38]; the repression could be reversed by coexpressing MYCBP or FLT3 ( Supplementary Fig. 3i,j). Those three genes all showed significant inverse correlations with miR-22 in expression in human AML (Supplementary Table 2; Supplementary Fig. 3f). The miR-22-induced repression of Bmi1, Cdk6 and Hmga1 at the protein level was also observed ( Supplementary Fig. 3c).     Fig. 4a,b). Similarly, in analysis of three publically available AML data sets, we found that 7-9% of the AML cases carried loss of one or even two alleles of the miR-22 locus ( Supplementary Fig. 4c). Therefore, DNA copy-number loss in miR-22 gene locus does exist in AML cases.
Expression of miR-22 is epigenetically repressed in AML. It was reported that TET2 is repressed by miR-22 as its direct target in   breast cancer and MDS 15,16 . Here we analysed the expression patterns of TET1/2/3 and miR-22 in three independent AML patient data sets 21,22,44 (Supplementary Table 3). To our surprise, we found that TET2 (and likely also TET3) exhibited a positive correlation, whereas only TET1 exhibited a negative correlation, with miR-22 in expression in AML (Supplementary Table 3; Fig. 5a). The primary, precursor and mature miR-22 levels were all significantly downregulated by MLL-AF9, MLL-AF10, AE9a and FLT3-ITD/NPM1c þ in colony-forming cells, while Tet1 (but not Tet2 or Tet3) was upregulated (Fig. 5b). Conversely, in MLL-ENL-ERtm cells 31 , Tet1, but not Tet2 or Tet3, was downregulated when miR-22 was upregulated after withdrawal of 4-OHT (Fig. 5c). Thus, Tet1, instead of Tet2, exhibited an inverse correlation with miR-22 in expression in both human and mouse leukaemic cells. Tet1 also exhibits an inverse correlation with miR-22 in expression during mouse normal myeloid differentiation ( Supplementary Fig. 5a). Furthermore, as miR-22 and TET1 were expressed at a significantly higher and lower level, respectively, in human normal CD33 þ myeloid progenitor cells than in CD34 þ HSPCs or MNCs (see Fig. 1a and ref. 14), the inverse expressional correlation between miR-22 and TET1 likely also existed in human normal hematopoietic cells. However, forced expression of miR-22 caused no noticeable changes in Tet1 expression in MLL-AF9, AE9a or FLT3-ITD/ NPM1c þ colony-forming cells (Fig. 5d). Similarly, neither miR-22 knockout nor overexpression resulted in any significant changes of Tet1/TET1 expression (Fig. 5e,f). In contrast, Tet1 knockout remarkably increased the levels of pri-, pre-and mature miR-22 (Fig. 5g). Thus, our data suggest that miR-22 is a downstream target of and negatively regulated by Tet1, and that there is no negative feedback of miR-22 on Tet1 expression.
Tet1 has been shown to cooperate with Polycomb repressive complex 2 (PRC2) components and cofactors, such as Ezh2 and Sin3a, to repress transcription of their co-target genes in mouse embryonic stem cells 8,9 . Our luciferase reporter assay showed that forced expression of Tet1 significantly repressed the transcriptional activity controlled by the miR-22 promoter 45 , suggesting that miR-22 is a direct repressed target of Tet1 (Fig. 6a). In an all-trans retinoic acid (ATRA)-induced THP-1/t(9;11) monocytic differentiation model 46 , we showed ARTICLE that on treatment with ATRA, TET1 (but not TET2 or TET3), EZH2 and SIN3A were significantly downregulated, accompanied by the upregulation of miR-22 (Fig. 6b). WDR81 is the gene that is located closely (within 500 bp) but oppositely to the miR-22 gene loci (Fig. 6d). We also tested the potential influence of ATRA on the expression level of WDR81 in the same model. ATRA treatment showed no significant effects on WDR81 level ( Supplementary Fig. 5b), suggesting that TET1 specifically inhibits the transcription of miR-22, but not its neighbouring gene with the opposite orientation.
While miR-22 expression level had a more than fivefold increase on ATRA treatment, the degrees of decrease in expression levels of TET1, EZH2 and SIN3A are relatively mild (though statistically significant) (Fig. 6b). To identify additional transcription factor(s) that is (are) more responsive to ATRA treatment and can facilitate TET1 binding to miR-22 promoter region, we searched for transcription factors that have evolutionarily conserved binding sites within the CpG island of miR-22 locus. Among a set of such transcription factors (including GFI1, STAT, PAX4, HMX1 and SRF), only GFI1 exhibited a significant inverse correlation with miR-22 in expression in all large-scale AML cohorts (Supplementary Table 3). Interestingly, it was reported previously that ATRA treatment could significantly diminish the binding of GFI1 to the loci of many of its target genes, for example, IL-6R, JAK3, E2F6 and so on 47 . Thus, we chose GFI1 for further studies. Notably, we found that ATRA treatment substantially reduced the transcription level of GFI1 in AML cells and its decrease degree was greater than that of TET1, EZH2 or SIN3A (Fig. 6b). We further showed that GFI1 is a binding partner of TET1 in both THP1 and HEK-293T cells ( Fig. 6c; Supplementary Fig. 5c). ATRA treatment remarkably reduced the binding of GFI1, TET1, EZH2 and SIN3A, but not that of MLL protein, to the miR-22 promoter region (Fig. 6d). H3K27me3 modifications and RNA polymerase II (RNA pol II) occupancy were significantly decreased and increased, respectively, while H3K4Me3 modifications showed no significant change (Fig. 6d). Noticeably, the enrichment of GFI1 to this region was diminished by ATRA to a greater degree than that of TET1, EZH2 or SIN3A (Fig. 6d), suggesting that GFI1 might be the primary effector of ATRA treatment in regulating miR-22 expression. Consistently, knockdown of GFI1 resulted in a dramatic increase in miR-22 expression (44 fold; Fig. 6e), associated with a significant decrease in the binding of TET1, EZH2, SIN3A and GFI1 itself to the miR-22 promoter region (Fig. 6f). Knockdown of expression of TET1, EZH2 or SIN3A resulted in a 2-3-fold increase in miR-22 expression (Fig. 6g), with no effects on GFI1 expression ( Supplementary Fig. 5d); only their combinational knockdown could cause a similar level of increase in miR-22 expression ( Fig. 6g; Supplementary Fig. 5e) to that induced by GFI1 knockdown (Fig. 6e). As expected, the expression level of WDR81 was not changed on knockdown of GFI1, TET1, EZH2 or SIN3A (Supplementary Fig. 5f,g). The above data suggest that GFI1, TET1, EZH2 and SIN3A are all involved in transcriptional repression of miR-22 expression; and when treated with ATRA, GFI1 likely functions as the primary effector that facilitates the binding of TET1/EZH2/SIN3A complex to the miR-22 promoter region.
As TET1 is a methylcytosine dioxygenase 8-10 , we conducted bisulfite sequencing analysis to investigate whether TET1 affects the methylation status of the miR-22 promoter. The analysis showed that the miR-22 promoter was hypomethylated in AML cells, no matter with ATRA treatment or not ( Supplementary  Fig. 5h). The hypomethylated status of the miR-22 promoter region in various AML was confirmed by analysing the TCGA_194S data set with DNA methylation information ( Supplementary Fig. 5i). The methylation status of the miR-22 promoter showed no significant correlation with miR-22 expression level in AML ( Supplementary Fig. 5j). These data suggest that the hypomethylation status of miR-22 promotor region does not lead to a high-level expression of miR-22 in AML, and TET1-mediated repression of miR-22 transcription is unlikely related to its methylcytosine dioxygenase activity.
The miR-22-associated regulatory circuit in AML. The above data suggest that repression of miR-22 in AML is attributed to both DNA copy-number loss and especially TET1-mediated transcriptional suppression. Interestingly, among the nine AML samples with DNA copy-number loss of miR-22 locus ( Supplementary Fig. 4a,b), the AML samples with both copynumber loss and TET1 overexpression generally exhibited a more significant repression of miR-22 expression than those with copynumber loss alone ( Supplementary Fig. 5k). Thus, those two mechanisms are not mutually exclusive and can have synergistic effect on reducing miR-22 expression. Collectively, our studies revealed a previously unappreciated genetic/epigenetic regulatory circuit in AML (Fig. 6h). In this circuit, oncogenic fusion genes or gene mutants (for example, MLL-fusions, AE9a and FLT3-ITD/NPM1c þ ) function as the 'drivers'. They promote the expression of TET1, which in turn, through recruiting polycomb cofactors such as EZH2 and SIN3A, represses the transcription of miR-22 by increasing H3K27me3 and decreasing RNA Pol II binding at the miR-22 promoter. When AML cells are treated with ATRA, ATRA substantially diminishes the enrichment of GFI1, a binding partner of TET1, at the miR-22 promoter, and thereby inhibits the recruitment of the TET1/EZH2/SIN3A complex to this region. In addition, miR-22 can also be compromised in its function by genetic mechanism(s) such as DNA copy-number loss in a portion (7-18%) of the AML cases. The inactivation of miR-22 results in the de-repression of its critical oncogenic targets such as CRCT1, MYCBP and FLT3, and thereby the activation of both CREB and MYC signalling pathways, leading to cell transformation and leukemogenesis.

Restoration of miR-22 expression and function to treat AML.
To investigate the therapeutic potential of restoration of miR-22 expression/function in treating AML, we employed amine-terminated, generation 7 (G7) poly(amidoamine) (PAMAM) dendrimers ( Supplementary Fig. 6a,b), an effective non-viral gene delivery vector with minimal side effects 48 . Nanoparticles carrying miR-22 oligos significantly delayed AML progression in both MLL-AF9 and AE9a-induced secondary leukaemic recipients (Fig. 7a,b). Notably, at least 40% of the treated mice seemed to be completely cured by the miR-22 nanoparticles as the pathological morphologies in PB, BM, spleen and liver tissues all became normal ( Fig. 7d; Supplementary Fig. 6c). In contrast, the miR-22 mutant nanoparticles exhibited no significant therapeutic effect (Fig. 7a,b,d; Supplementary Fig. 6c). As expected, miR-22 oligos, but not miR-22 mutant oligos, significantly inhibited expression of its critical targets (that is, Crct1, Flt3 and Mycbp) in BM cells of the treated mice ( Supplementary Fig. 6d). The miR-22-nanoparticles showed no noticeable effects on blood cell lineages (Supplementary Table 4).
We then tested the miR-22 nanoparticles in a xenotransplantation model 49 . Similarly, the nanoparticles carrying miR-22 oligos, but not miR-22 mutant, significantly delayed AML progression induced by human MV4;11/t(4;11) cells (Fig. 7c). The miR-22-nanoparticle administration also resulted in less aggressive leukaemic pathological phenotypes in the recipient mice ( Supplementary Fig. 6e). Thus, our studies demonstrated the therapeutic potential of using miR-22-based nanoparticles to treat AML.

Discussion
It remains poorly understood how TET proteins mediate gene regulation in cancer. Here we show that in de novo AML, it is TET1, but not TET2 (a reported direct target of miR-22 in MDS and breast cancer 15,16 ), that inversely correlates with miR-22 in expression and negatively regulates miR-22 at the transcriptional level. Likely together with GFI1, TET1 recruits polycomb cofactors (for example, EZH2/SIN3A) to the miR-22 promoter, leading to a significant increase in H3K27me3 occupancy and decrease in RNA pol II occupancy at that region, and thereby resulting in miR-22 repression in AML cells; such a repression can be abrogated by ATRA treatment. Thus, our study uncovers a novel epigenetic regulation mechanism in leukaemia involving the cooperation between TET1/GFI1 and polycomb factors.
Besides GFI1, it was reported that LSD1 is also a binding partner of TET1 (ref. 50). Interestingly, LSD1 is known as a common binding partner shared by TET1 and GFI1, and mediates the effect of GFI1 on hematopoietic differentiation 51,52 . Thus, it is possible that LSD1 might also participate in the transcriptional repression of miR-22 as a component of the GFI1/TET1 repression complex.
We previously reported that TET1 cooperates with MLL fusions in positively regulating their oncogenic co-targets in MLL-rearranged AML 14 . Here we show that TET1 can also function as a transcriptional repressor (of a miRNA) in cancer. The requirement of TET1-mediated regulation on expression of its positive (for example, HOXA/MEIS1/PBX3) 14 or negative (for example, miR-22) downstream effectors in leukemogenesis likely explains the rareness of TET1 mutations in AML 53 , and highlights its potent oncogenic role in leukaemia.
The aberrant activation of both CREB and MYC signalling pathways has been shown in AML [24][25][26]54,55 , but the underlying molecular mechanisms remain elusive. Our data suggest that the activation of these two signalling pathways in AML can be attributed, at least in part, to the repression of miR-22, which in turn, results in the de-repression of CRTC1 (CREB pathway), FLT3 and MYCBP (MYC pathway), and leads to the upregulation of oncogenic downstream targets (for example, CDK6, HOXA7, BMI1, FASN and HMGA1) and downregulation of tumoursuppressor downstream targets (for example, RGS2).
In summary, we uncover a TET1/GFI1/EZH2/SIN3ABmiR-22BCREB-MYC signalling circuit in de novo AML, in which miR-22 functions as a pivotal anti-tumour gate-keeper, distinct from its oncogenic role reported in MDS or MDSderived AML 16 . Thus, our study together with the study of Song et al. 16 highlight the complexity and functional importance of miR-22-associated gene regulation and signalling pathways in hematopoietic malignancies, and may provide novel insights into the genetic/epigenetic differences between de novo AML and MDS.
Our findings also highlight the possibility of using miR-22based therapy to treat AML patients. Our proof-of-concept studies demonstrate that the nanoparticles carrying miR-22 oligos significantly inhibit AML progression and prolong survival of leukaemic mice in both BMT and xeno-transplantation models. Notably, miRNA-based nanoparticles have already entered clinical trials 56 . It would be important, in the future, to further test the combination of miR-22-carrying nanoparticles (or smallmolecule compounds that can induce endogenous expression of miR-22) with standard chemotherapy agents (cytosine arabinoside and anthracycline), or with the emerging small molecule inhibitors against MYC and/or CREB pathway effectors, to achieve optimal anti-leukaemia effect with minimal side effects. Overall, our results suggest that restoration of miR-22 expression/ function (for example, using miR-22-carrying nanoparticles or small-molecule compounds) holds great therapeutic potential to treat AML, especially those resistant to current therapies.

Methods
AML and MDS samples and cell lines. The AML and MDS patient samples were obtained at the time of diagnosis with informed consent at the University of Chicago Hospital (UCH), and were approved by the University of Chicago Hospital Institutional Review Board (UCHIRB). All patients were treated according to the protocols of the corresponding institutes/hospitals. THP-1, KOCL48, MV4;11, MEF and HEK-293T cells were purchased from ATCC (Manassas, VA) and maintained in the lab. The MLL-ENL-ER cell line was a gift from Dr Robert Slany 31 . All the cell lines were tested for mycoplasma contamination yearly using a PCR Mycoplasma Test Kit (PromoKine).
Preparation of AML and MDS samples. The primary AML and MDS samples were stored in liquid nitrogen until used. Blasts and mononuclear cells were purified by use of NycoPrep 1.077A (Axis-Shield, Oslo, Norway) according to the manufacturer's manual.
Human normal hematopoietic control cell samples. The MNC normal control samples were isolated from normal BM cells purchased from AllCells, LLC (Emeryville, CA) by use of NycoPrep 1.077A (Axis-Shield, Oslo, Norway) according to the manufacturer's manual.
Mouse normal BM cell population sorting. As described previously 23 , wild-type C57BL6/J mice were used for the sorting. All laboratory mice were maintained in the animal facility at the University of Chicago and the University of Cincinnati. All experiments on mice in our research protocol were approved by Institutional Animal Care and Use Committee (IACUC) of the University of Chicago and the University of Cincinnati.
RNA extraction and quantitative RT-PCR. Total RNA was extracted with the miRNeasy extraction kit (Qiagen, Valencia, CA) and was used as template to synthesize complementary DNA for quantitative reverse transcription PCR (qRT-PCR) analysis in a 7900HT real-time PCR system (Applied Biosystems, Foster City, CA). TaqMan qPCR assay was performed to validate the differential expression patterns of miR-22 using commercial kits from Applied Biosystems (Cat. no. 4427975). Sequences for the controls are: sno202: 5 0 -GCTGTACTGACT TGATGAAAGTACTTTTGAACCCTTTTCCATCTGATG-3 0 ; RNU6B: 5 0 -CGCA AGGATGACACGCAAATTCGTGAAGCGTTCCATATTTTT-3 0 . qPCR with SYBR Green dye (Qiagen) was used to determine expression of mRNA genes. snoRNA202, RNU48, Gapdh or GAPDH were used as endogenous controls for qPCR of miRNA and mRNA, respectively. Each sample was run in triplicate. qPCR primers are available on request. For determining the miR-22 DNA locus copy number, TaqMan qPCR assay was used as described previously 57 .
microRNA microarray and exon array assays. As described previously 17,23 , our miRNA expression profiling assay of 85 (including 10 t(8;21), 9 inv(16), 9 t(15;17), 10 MLL-rearranged, 11 ( þ 8), 29 normal karyotype and 7 others) AML samples, and 15 human normal BM samples was performed by Exiqon (Woburn, MA) using the miRCURY LNA arrays (v10.0; covering 757 human miRNAs). The 15 normal BM controls included six CD34 þ hemtopoietic stem/progenitor, five CD33 þ myeloid progenitor and four MNC samples. In terms of patient samples, MNCs isolated from the BM or PB cells of the 85 AML patients were used. The expression values are log2 (Hy3/Hy5) ratios, which were obtained on the basis of the normalized data where replicated measurements on the same slide have been averaged. In addition, as described previously 14,17,23 , a total of 100 human AML (including 30 t(8;21), 27 inv(16), 31 t(15;17) and 12 MLL-rearranged), and 9 normal BM samples (including three each of CD34 þ hematopoietic stem/ progenitor, CD33 þ myeloid and MNC samples) were analysed by use of Affymetrix GeneChip Human Exon 1.0 ST arrays (Affymetirx, Santa Clara, CA). The QC test and Affymetrix exon array assays were done in the core facility of National Human Genome Research Institute, NIH (Bethesda, MD). Robust multiarray average (RMA) 58 was used for the data normalization with Partek Genomics Suite (Partek Inc., St Louis, MI). The complete microarray data set has been deposited in the GEO database under the accession codes GSE34184 and GSE30285.
Among the above 100 human AML samples, 81 samples (that is, the In-house_81S; including 29 t(8;21), 26 inv(16) and 26 t(15;17) AML) have been also included in the Exiqon microRNA array assay 21 . The microarray data set of those 81 AML samples has been deposited in GEO database under the accession code GSE27370.
Affymetrix gene arrays of mouse samples. As described previously 17 , a total of 15 mouse BM samples including 6 primary (including three each of negative control and MLL-AF9) and 9 secondary (including three negative control and six MLL-AF9) obtained from the in vivo mouse BM reconstitution assays were analysed by use of Affymetrix GeneChip Mouse Gene 1.0 ST Array (Affymetirx). The RNA quality control, cDNA amplification, hybridization and image scan were conducted in the Functional Genomics Facility of the University of Chicago. RMA 58  with DNA methylation data as detected by Infinium HumanMethylation450 BeadChip were referred to as TCGA_194S. The mRNA/miRNA expression data and methylation data were downloaded from https://tcga-data.nci.nih.gov/tcga/dataAccessMatrix.htm? mode=ApplyFilter& showMatrix=true&diseaseType=LAML& 12:28 PM 4/6/ 2016tumorNormal=TN&tumorNormal=T&tumorNormal=NT.
For the ATRA-treatment study, THP-1 cells were seeded at a concentration of 0.4 Â 10 6 ml À 1 and treated with ATRA (1 mmol l À 1 ) or vehicle control (DMSO, 0.001%) for 72 h before cells were collected for RNA analysis or chromatin immunoprecipitation (ChIP) assays.
All the cell lines were mycoplasma negative.
Lentivirus production and infection. All the plasmid for packaging lentivirus, including pMD2.G, pMDLg/pRRE and pRSV-Rev, were purchased from Addgene Viability and proliferation assays. These experiments were conducted as described previously 17,23 with some modifications. For apoptosis and viability assays, 48 h after transfection, cells were collected and seeded with requested concentration. Cell apoptosis and viability were assessed using ApoLive-Glo Multiplex Assay Kit (Promega, Madison, WI) following the corresponding manufacturer's manuals. For cell proliferation assays, per million cells were electroporated with 1.5 mg plasmid. Twenty-four hours after transfection, cells were seeded in 96-well plates at the concentration of 10,000 cells per well. Cell numbers were counted at the indicating days.  45 ) was PCR-amplified using primers: forward 5 0 -AATAATGAGCTCAAGG TCGGACG-3 0 and reverse 5 0 -AATAATGATATCCTTTAGCTGGGTC-3 0 , and cloned into the SacI and EcoRV sites of the pGL4.15 Luciferase Reporter Vector (Promega). The MSCV-Tet1 construct was as described previously 14 . All the above insertions were confirmed by DNA sequencing.
Luciferase reporter and mutagenesis assays. Luciferase reporter and mutagenesis assays were conducted as described previously 17,23 , with some modifications. Briefly, for transfection, HEK-293T cells were plated in 96-well plates at a concentration of 6,000 cells per well in triplicate for each condition. For the miR-22 targeting CRTC1 and FLT3 experiments, after overnight incubation, cells were transfected with 20 ng of the pMIR-REPORT bearing the CRTC1 or FLT3 3 0 UTR or the 3 0 UTRs with miR-22 binding site mutations, and 20 ng of MSCV-miR-22 or an empty MSCV vector using Effectene Transfection Reagent (Qiagen) according to the manufacturer's protocol. pMIR-REPORT Betagalactosidase Reporter Control Vector (Ambion) (1 ng) was co-transfected for transfection efficiency control in all transfections. Cells were lysed and firefly luciferase and b-galactosidase activities were detected using Dual-Light Combined Reporter Gene Assay System (Applied Biosystems, Foster City, CA) 48 h post transfection. Firefly luciferase activity was normalized to b-galactosidase activity for each transfected well. For the Tet1 targeting miR-22 study, HEK-293T cells were transfected with 20 ng MSCV-Tet1 construct and/or 20 ng pGL4.15-miR-22 promoter. The succeeding luciferase reporter assay was conducted according to the manufacture's protocol (Promega). Each experiment was performed in triplicate and repeated three times.
Packaging of recombinant retroviruses and CFA assays. Those experiments were conducted as described previously 17,23 with some modifications. Briefly, retrovirus vectors were co-transfected with pCL-Eco packaging vector (IMGENEX, San Diego, CA) into HEK-293T cells using Effectene Transfection Reagent (Qiagen) to produce the retroviruses. BM cells were collected from a cohort of 4-6-week-old B6.SJL (CD45.1) donor mice after 5 days of 5-fluorouracil (5-FU) treatment, and primitive hematopoietic progenitor cells were enriched with Mouse Lineage Cell Depletion Kit (Miltenyi Biotec Inc., Auburn, CA). An aliquot of enriched hematopoietic progenitor cells were added to retroviral supernatant together with polybrene in a conical tube, which were centrifuged at 2,000g for 2 h at 32°C (that is, 'spinoculation' 14,17,23 ) and then the media was replaced with fresh media and incubated for 20 h at 37°C. Next day, the same procedure was repeated once.
Then, on the day following the second spinoculation, an equivalent of 2.0 Â 10 4 cells were plated into a 35-mm Petri dish in 1.5 ml of Methocult M3230 methylcellulose medium (Stem Cell Technologies Inc., Vancouver, Canada) containing 10 ng ml À 1 each of murine recombinant IL-3, IL-6, and granulocytemacrophage colony-stimulating factor (GM-CSF), and 30 ng ml À 1 of murine recombinant SCF (R&D Systems, Minneapolis, MN), along with 1.0 mg ml À 1 of G418 and/or 2 mg ml À 1 of puromycin. For each transduction, there were two duplicate dishes. Cultures were incubated at 37°C in a humidified atmosphere of 5% CO 2 in air. The colonies were replated every 7 days under the same conditions. The colony-forming/replating assays were repeated three times.
Primary and secondary BMT. These experiments were conducted as described previously 14,17,23 with some modifications.
For primary BMT assays shown in Fig. 1e, normal bone marrow cells of B6.SJL (CD45.1) mice were retrovirally transduced with MSCV-neo þ MSCV-PIG (as control; Ctrl), MSCV-neo þ MSCV-PIG-miR-22 (that is, miR-22), MSCV-neo-MLL-AF9 þ MSCV-PIG (that is, MA9), MSCV-neo-MLL-AF9 þ MSCV-PIG-miR-22 (that is, MA9 þ miR-22) or MSCV-neo-MLL-AF9 þ MSCV-PIG-miR-22 mutant (that is, MA9 þ miR-22mut), through two rounds of spinoculation. Then, retrovirally transduced cells were plated into methylcellulose medium supplied with a set of cytokines to form colonies as described in the CFA assays. Seven days later, colony cells were collected and washed, and then were injected by tail vein into lethally irradiated (960 rads) 8-10-week-old C57BL/6 (CD45.2) recipient mice with 1.5 Â 10 5 donor cells plus a radioprotective dose of whole BM cells (1 Â 10 6 ; freshly collected from a C57BL/6 mouse) per recipient mouse. Notably, as the colony cells were under selection of both G418 (1.0 mg ml À 1 ) and puromycin (2 mg ml À 1 ) for a week, all donor cells (that is, the collected colony cells) must be positive for retroviral transductions of both MSCVneo-and MSCV-PIG-based constructs. Thus, MLL-AF9 and miR-22 (or miR-22mut) must be ectopically co-expressed in MA9 þ miR-22 (or MA9 þ miR-22mut) donor cells, which actually were confirmed by qPCR. Indeed, due to the potent inhibitory effect of miR-22 on MLL-AF9-induced colony forming, we had to prepare more mouse BM progenitor cells for the co-transduction of MLL-AF9 and miR-22. Thus, we plated them in a larger number of dishes than what we did for other groups of co-transductions. After BMT, all recipient mice were watched for leukemogenesis for a period of 200 days or till the end point that the mice developed full-blown AML or other severe illness.
For secondary BMT assays shown in Fig. 4f, leukaemic BM cells isolated from the primary leukaemic mice bearing MLL-AF9 fusion were retrovirally transduced with MSCV-neo þ MSCV-PIG (that is, . Again, retrovirally transduced cells were plated into methylcellulose medium supplied with G418 and puromycin (for selection) as well as a set of cytokines to form colonies. Seven days later, the colony cells were collected and washed, and then were transplanted into sub-lethally irradiated (480 rads) 8-10-week-old C57BL/6 (CD45.2) secondary recipient mice via tail vein injection, with the dosage of 1.5 Â 10 5 donor cells per recipient mouse.
The maintenance and monitoring of mice. C57BL/6 (CD45.2), B6.SJL (CD45.1) mice were purchased from the Jackson Lab (Bar Harbor, ME, USA) or Harlan Laboratories, Inc (Indianapolis, IN, USA). NSGS (NSG-SGM3) immunodeficient mice 49 and miR-22 À / À (ref. 20) mice were purchased from the Jackson Lab and were bred and maintained in house. Both male and female mice were used for the experiments. All laboratory mice were maintained in the animal facility at the University of Chicago and the University of Cincinnati. All experiments on mice in our research protocol were approved by Institutional Animal Care and Use Committee (IACUC) of the University of Chicago and the University of Cincinnati. The maintenance, monitoring and end-point treatment of mice were conducted as described previously 14,17,23 .
DNA copy-number analysis of miR-22 gene locus in human AML. The copynumber data of AML from The Cancer Genome Atlas (TCGA) project were downloaded from Broad Firehose's analyses runs. The putative copy-number calls were determined using GISTIC 2.0 (ref. 65). The latest GISTIC analyses data were obtained using the following shell command: 'firehose_get -o 'GISTIC' analyses latest LAML'.
The.cel files of Affymetrix SNP 6.0 data for GSE21107 (ref. 66) and GSE23452 (ref. 67) were downloaded from NCBI GEO. The raw data were preprocessed using PennCNV 68 . Then ASCAT 69 was used to obtain the copy-number alterations. The putative copy-number calls were determined using GISTIC 2.0 as described above.
Software and statistical analyses. The miRNA and gene/exon array data analyses, as well as qPCR data analyses were conducted by use of Partek Genomics Suite (Partek Inc.), TIGR Mutiple Array Viewer software package (TMeV version 4.6; TIGR, Rockville, MA) 70 and/or Bioconductor R packages. The miRNA-gene expression correlation was analysed by use of Partek Genomics Suite (Partek Inc.). The t-test, Kaplan-Meier method and log-rank testand so on were performed with WinSTAT (R. Fitch Software), GraphPad Prism version 5.00 (GraphPad Software, San Diego, CA) and/or Partek Genomics Suite (Partek Inc.). The P values o0.05 were considered as statistically significant. Significance analysis of microarrays, embedded in the TMeV package (TIGR, Rockville, MA), was used to identify the genes that are significantly (qo0.05; false discovery rate, FDRo0.05) dysregulated in the MLL-AF9-mediated mouse leukaemia samples or human AML samples relative to the normal controls. Pearson correlation was used in the analysis of the correlation between miR-22 and candidate genes in expression. The list of transcription factors that have evolutionarily conserved binding sites within the miR-22 promoter region (that is, the adjacent upstream CpG island) was obtained by searching UCSC Genome Browser (https://genome.ucsc.edu/cgi-bin/ hgTracks?db=hg19&position=chr17%3A1614689-1623188&hgsid=467686877_ 3vyTlry3a40ZiT7dfAaAIAsYA2R6).
Data availability. Data referenced in this study are available in The Gene Expression Omnibus. The Affymetrix exon array data and the microarray data are available under accession codes GSE34184 and GSE30285. Additional exon array data are available under accession code GSE27370. The mouse microarray data is available under accession code GSE34185. The AML samples collected by the German AMLCG study group are available under accession code GSE37642. The mRNA gene expression data of 183 adult de novo AML cases is available from the TCGA.