Functional Characterization of Cooperating MGA Mutations in RUNX1::RUNX1T1 Acute Myeloid Leukemia

MGA (Max-gene associated) is a dual-specificity transcription factor that negatively regulates MYC-target genes to inhibit proliferation and promote differentiation. Loss-of-function mutations in MGA have been commonly identified in several hematological neoplasms, including acute myeloid leukemia (AML) with RUNX1::RUNX1T1, however, very little is known about the impact of these MGA alterations on normal hematopoiesis or disease progression. We show that representative MGA mutations identified in patient samples abolish protein-protein interactions and transcriptional activity. Using a series of human and mouse model systems, including a newly developed conditional knock-out mouse strain, we demonstrate that loss of MGA results in upregulation of MYC and E2F targets, cell cycle genes, mTOR signaling, and oxidative phosphorylation in normal hematopoietic cells, leading to enhanced proliferation. The loss of MGA induces an open chromatin state at promotors of genes involved in cell cycle and proliferation. RUNX1::RUNX1T1 expression in Mga-deficient murine hematopoietic cells leads to a more aggressive AML with a significantly shortened latency. These data show that MGA regulates multiple pro-proliferative pathways in hematopoietic cells and cooperates with the RUNX1::RUNX1 T1 fusion oncoprotein to enhance leukemogenesis.


Introduction
Max-gene associated (MGA) is a transcription factor that uses an n-terminal T-box domain and a c-terminal MYClike basic helix-loop-helix (bHLH) domain to regulate MAX-network and T-box family targets 1 .MGA interacts with MAX, L3MBTL2, E2F6, and PCGF6 as a part of the non-canonical polycomb repressive complex (ncPRC1.6) 2 and is required for PRC1.6 complex formation.In non-hematopoietic cells, the recruitment of MGA to T-box and MAXnetwork target genes was shown to result in the deposition of repressive histone marks such as H2AK119ub1 and H3K27me3 2,3 .MGA is required for embryogenesis and its depletion leads to aberrant embryonic stem cell differentiation and embryonic lethality [4][5][6] .
Heterozygous somatic alterations of MGA, the most common of which lead to loss-of-function truncation/deletion of the MYC-like bHLH domain, occur in 5% of all cancers and are commonly seen in lung adenocarcinoma, endometrial carcinoma, and colorectal cancer [7][8][9][10] .The loss of MGA in lung adenocarcinoma cells, which occurs in ~ 8% of lung adenocarcinoma patients, was shown to lead to an upregulation of MYC-and E2F6-target genes, resulting in an increase in cancer cell proliferation and invasiveness 11,12 .MGA has also been identi ed as a common genetic alteration in hematological neoplasms, including acute myeloid leukemia (AML), chronic lymphocytic leukemia (CLL), natural killer/T-cell lymphoma, B-cell acute lymphoblastic leukemia (B-ALL), and T-cell acute lymphoblastic leukemia (T-ALL) [13][14][15][16][17][18] .In particular, we previously identi ed MGA mutations as recurrent alterations in AMLs with RUNX1::RUNX1T1 fusions 17 .Likewise, similar mutations have been observed in AMLs with KMT2A-PTD 16 .
Despite the recurrence of these mutations, the molecular role of MGA in normal hematopoiesis, as well as in hematopoietic malignancies, has been understudied.This is partly due to the embryonic lethality of in vivo MGA de ciency models, which we circumvented in this study by developing a conditional knockout mouse model.
Harnessing a multidisciplinary approach to characterize the hematopoietic function of MGA in both human and mouse cells, we establish that MGA loss leads to an increase in proliferation via the upregulation of several cell cycle pathway genes and results in a shortened latency in a mouse model of RUNX1::RUNX1T1-driven leukemia.

Results
The reported mutations of MGA in pediatric RUNX1::RUNX1T1 leukemias are heterozygous, somatic mutations at variable variant allele frequencies and result in the deletion of the MYC-like bHLH domain, potentially resulting in a protein with altered or abolished function (Fig. 1A) 17 .We rst assessed the protein localization of MGA, including full-length MGA (WT-MGA) and 3 MGA truncations (MGA p.E1953*, MGA p.S812*, and MGA p.C623*) identi ed in patient samples, by transfecting GFP fusion expression vectors in HEK293T cells (Fig. 1A).We also included an nterminus truncation which deletes the T-box domain; this construct expresses exons 14-23, including the bHLH domain, to serve as control along with WT-MGA.WT-MGA was localized in both the cytoplasm and nucleus, while the MGA-ex14-23 control was exclusively found in the cytoplasm, which suggests the n-terminal region is required for nuclear localization (Fig. 1B).The MGA p.S812* and MGA p.C623* mutant proteins were expressed exclusively in the nucleus, while the MGA p.E1953* truncation maintained normal localization in both the nucleus and cytoplasm.The retained nuclear localization of MGA truncations suggests these mutations could still be functional.
We then performed co-immunoprecipitation (Co-IP) assays followed by western blot to determine if MGA truncations interacts with components of the ncPRC1.6complex.We co-expressed ag-tagged WT-MGA, MGA-ex14-23, and MGA p.C623* with HA-tagged MAX in HEK293T cells and performed IP-western blots.We found that MGA p.C623* does not signi cantly bind L3MBTL2, PCGF6, or MAX, suggesting it loses its interaction with the ncPRC1.6complex (Fig. 1C).Although nuclear MGA p.C623* does not interact with components of the ncPRC1.6complex, it still could interact with the genome and hamper the genomic binding of MGA and the ncPRC1.6complex in a dominant-negative manner at MGA binding sites.To examine this, we expressed HA-tagged MGA p.C623* in MOLM-13 cells and observed a global loss in genomic binding by Cleavage Under Targets and Release using Nuclease (CUT&RUN) assays.In addition to this genome-wide decrease in binding of the MGA mutant, binding was abolished at de ned MGA target sites such as CCND2 (Supp.Figure 1A&B).
The loss of interaction with the ncPCR1.6complex and the loss of DNA binding of MGA p.C623* supports the hypothesis that the truncating mutations of MGA identi ed in patients are loss-of-function mutations, leading to haploinsu ciency.To better understand the impact of MGA loss in hematopoietic cells, we designed a knock-out (KO) model of MGA using a CRISPR-based approach in MOLM-13 cells (Fig. 2A).The loss of MGA in these cells led to a slight increase in proliferation (Fig. 2B&C).RNA-sequencing (RNA-seq) analysis showed that the loss of MGA led to the upregulation of several genes that play a critical role in proliferation, genome stability, and tumorigenicity, including Cyclin D1 (CCND1), Cyclin D2 (CCND2), and Structural Maintenance of Chromosomes 1B (SMC1B) (Fig. 2D and Supp. Figure 1C) [19][20][21] .Supporting our RNA-seq data, we found that MGA binds to the CCND2 and SMC1B promoters and that MGA deletion reduces PCGF6 binding at these promoters and at the genome-wide level (Fig. 2E and Supp. Figure 1D).In contrast, MYC binding is minimally disrupted and potentially enhanced at CCND2 and SMC1B promoters (Fig. 2E, Supp. Figure 1D).We also reveal that the loss of MGA resulted in little to no change in the total number of TSS (transcription start site) occupied by H2AK119Ub1, H3K27me3, and H3K4me3.However, its deletion led to an increase in the total number of TSS with the active histone mark of H3K27Ac (Supp.Figure 1E&F).In addition, the intensity (log2Fold change) of active marks H3K4me3 and H3K27Ac at TSS are signi cantly higher, while the intensity of repressive mark H3K27me3 is lower in MGA-KO MOLM-13 cells, suggesting that MGA depletion leads to a more active epigenetic landscape (Supp.Figure 1G).Supporting these data, ATAC-seq analysis demonstrates that the loss of MGA leads to a more open chromatin state at the promoters of CCND2 and SMC1B with minimal global chromatin accessibility changes (Fig. 2F&G).
These observations were validated in primary cord blood CD34 + cells immortalized by RUNX1::RUNX1T1, in which disruption of MGA by CRISPR leads to a reduction of PCGF6 binding to the CCND2 promoter associated with an overall increase in CCND2 expression, as well as an increase in self-renewal capacity (Supp.Figure 1H-J).
Considering that MGA loss leads to a cell growth advantage in leukemic cell lines, we wanted to assess the role of MGA in hematopoietic cell development.To this aim, we designed a conditional knock-out model in the C57BL/6 background, in which LoxP sites were introduced that ank exon 3 of Mga (Supp.Figure 2A).When crossed with Vav1-Cre, this leads to the deletion of exon 3 and the effective deletion of Mga in hematopoietic cells.This conditional hematopoietic cell KO model circumvented the known embryonic lethality of homozygous Mga loss 6 , as offspring with both heterozygous (Vav1-cre tg/+ x Mga /+ = Mga(+/-)) and homozygous (Vav1-cre tg/+ x Mga / = Mga(-/-)) deletions of Mga induced by Vav1-Cre were viable with normal Mendelian ratios.Although the mutations identi ed in patients are heterozygous, we also established a cohort of mice with homozygous deletion of Mga to better de ne the functional role of Mga in hematopoiesis.RNA-seq on lineage-(lin-),cKit + hematopoietic stem and progenitor cells (HSPCs) isolated from the bone marrow (BM) of Mga(+/+) (control), Mga(+/-), and Mga(-/-) mice demonstrated dysregulation of several critical pathways in Mga(-/-) but not in control or Mga(+/-) HSPCs, including the upregulation of Myc and E2F targets and the downregulation of extracellular signaling pathways including JAK-STAT, TGFβ, and TNFα signaling (Fig. 3A&B, Supp. Figure 2C).Similar to our ndings in the MOLM-13 MGA knock-out model, Mga(-/-) HSPCs had signi cant upregulation of cell cycle and proliferation genes, including Ccne1, Ccnd1, Ccna2, Hdac2, and Smc1b (Supp.Figure 2D).To assess the epigenetic changes resulting from the loss of Mga, we performed CUT&RUN and ATAC-seq on cKit + HSPCs from these mice.Unfortunately, an anti-murine MGA antibody compatible with CUT&RUN is not available.However, both heterozygous and homozygous deletions of Mga led to increases in H3K4me3 and H3K27Ac marks and a decrease in H3K27me3 at the promoters of genes such as Ccne1, Smc1b, and Cdk1 (Supp.Figure 3A-C).ATAC-seq analysis showed no signi cant changes in promoter accessibility of Ccne1 and Cdk1, however, the promoter of Smc1b, which is closed in control HSPCs, showed signi cantly more ATAC-seq signals in both Mga(+/-) and Mga(-/-) HSPCs (Fig. 3C).Globally, both heterozygous and homozygous deletions of Mga led to a signi cant increase in open chromatin at gene promoters compared to WT controls (5043 promoters (p < 0.05) and 3984 promoters (p < 0.05), respectively) (Fig. 3D, Supp. Figure 3D).There is a ~ 50% overlap of open promoters between Mga(+/-) and Mga(-/-) HSPCs leading to a nearly identical gene set enrichment with both conditions showing a strong enrichment of peaks of promoter regions of cell cycle-related genes (Fig. 3E&F).When comparing the upregulated genes from RNA-seq in Mga(-/-) HSPCs with the enriched ATAC-seq open promoter genes for Mga(-/-), we found an overlap of 1006(18.2%)genes (Fig. 3G).This overlapping gene list is enriched for cell cycle, pro-proliferation, and chromosome stability pathway activation, suggesting that these cells have improved tness and proliferate faster than controls (Fig. 3H).
In addition to these molecular perturbations, loss of Mga in hematopoietic cells also has functional consequences.Mga(-/-) HSPCs have a signi cant increase in cell growth and proliferation and in self-renewal capacity when compared to controls, while Mga(+/-) HSPCs have only a slight advantage (Fig. 4A-D).These ex vivo analyses of Mga de cient cells propose that the loss of Mga may alter hematopoietic homeostasis.However, Mga(+/-) and Mga(-/-) mice at 3-6 months displayed normal CBCs and demonstrated no defect in the development of B cells (B220+), T cells (CD3e+), or myeloid cells (CD11b+) (Supp.Figure 3E-F).Ki67 staining of the B cell, T cell, or myeloid cell populations showed a slight increase in cycling B and T cells but not myeloid cells (Supp.Figure 3G).HSPC immunophenotyping by ow cytometry showed no change in the lymphoid progenitor (lin-cKit-, Sca-1 low ), myeloid progenitor (lin-, cKit+) or LSK (lin-, cKit+, Sca-1+) populations from both Mga(+/-) and Mga(-/-) mice (Fig. 4E).Aged control, Mga(+/-), and Mga(-/-) mice (16 months) did not spontaneously develop a hematopoietic malignancy (data not shown).Despite the observed self-renewal phenotype, serial noncompetitive BM transplants of HSPCs isolated from control, Mga(+/-), and Mga(-/-) mice (Supp.Figure 4A) lead to no signi cant changes in WBC counts or in the ability to reestablish mature blood cell production over a 16 week period (Supp.Figure 4B-G).

Discussion
Here we generated both human and mouse models to assess the role of MGA in normal and malignant hematopoiesis.Our data support the well-documented antiproliferative role of MGA and show that loss of MGA leads to an open and active epigenetic chromatin status, and subsequently the upregulation of several critical pathways for cell growth and proliferation (Fig. 7) 2,3 .Previous reports in non-hematopoietic models show that components of the ncPRC1.6complex, such as MGA and E2F6, regulate transcriptional control of critical regulators of meiosis such as STAG3, SMC1B, MEIOC, and CNTD1 and overall result in transcriptional silencing 2,25 .While components of the ncPRC1.6complex are known to play a critical role in embryonic stem cell maintenance, as depletion of MGA, L3MBTL2, and PCGF6 leads to defects in pluripotency, proliferation, and differentiation, the impact of disrupting this complex has not been thoroughly evaluated in hematopoietic cells 3,6,26 .Our hematopoietic MGA-KO models support these reports, revealing a similar transcriptional landscape, including the increased expression of genes such as Cdk1, Smc1b, and Ccne1, when Mga is disrupted.Ccne1 and Cdk1 are known regulators of cell cycle progression, while Smc1b is critical for genomic stability and is suggested to play a role in cell proliferation 20,21,27,28 .We believe the upregulation of these genes and others is due to the direct loss of MGA-dependent ncPRC1.6 binding and regulation leading to the serial replating and cell growth advantage of Mga(-/-) HSPCs.Overexpression of these genes has also been observed in several cancer types including AML, hepatocellular carcinoma, and ovarian carcinoma [29][30][31] .The lack of changes in H2AK119Ub1 upon MGA depletion was surprising.However, the decrease in H3K27me3 and increase in H3K27Ac suggest both that there are functional redundancies with other PRC1 complexes and that the binding of MGA-dependent ncPRC1.6 to the loci is more critical for transcriptional repression 32,33 .To further support this, the open chromatin of MGA target genes in MGA de cient cells is likely due to the loss of L3MBTL2, which can regulate chromatin compaction regardless of histone modi cation status 34 .
Despite these clear transcriptional and epigenetic changes, the loss of MGA in the hematopoietic compartment alone is not su cient to promote profound hematopoietic defects or the development of a hematopoietic neoplasm, even under stress conditions such as serial BM transplantation and 5-FU treatments (data not shown).These ndings support the notion that the loss of function mutations in MGA observed in hematologic malignancies are cooperating mutations that provide a transcriptional state that may enhance the effects of tumor drivers.In support of this hypothesis, the latency of leukemia induced by a viral model of RUNX1::RUNX1T1 9A was signi cantly shortened when expressed in both Mga(+/-) and Mga(-/-) hematopoietic cells, and the resulting leukemic cells displayed a more immature phenotype.RUNX1::RUNXT1 AMLs have been shown to be dependent on the gene activation of D cyclins such as CCND2 and CCND1 35,36 .Overall, we propose that the shortened latency in our RUNX1::RUNXT1 9A model is likely due to MGAde cient cells being primed for enhanced proliferation via the upregulation of cyclin genes and MYC-targets, as shown by our transcriptional analysis of RUNX1::RUNXT1 9A tumors.Consistent with this, our previous work identi ed gain-of-function mutations in CCND2 in RUNX1::RUNX1T1 AMLs, which has been supported by other studies 37 .The ndings presented here suggest that MGA loss may phenocopy these CCND2 mutations.Further supporting this claim is the observation that MGA and CCND2 were mutually exclusive in our previous study 17 .Collectively, our data highlight MGA as an important regulator of pathways critical for leukemia development, such as MYC activation, cell cycle, and oxidative phosphorylation, and that loss of function mutations identi ed in patients can functionally cooperate with the RUNX1::RUNX1T1 oncoprotein 21,[38][39][40][41] .Importantly, we have also de ned the in vivo consequences of Mga loss by using a conditional knock-out approach that circumvents the embryonic lethality of constitutional knock-out models, thus providing the scienti c community with an important model.

Animals
We used Ingenious Targeting Laboratory (Stony Brook, NY, USA) to generate the conditional Mga KO model in C57BL/6 mice that bore a LoxP-anked exon 3 of Mga using embryonic stem cell-based gene targeting.CD45.1 and C57BL/6 mice were obtained from Jackson Laboratory.Blood, WBM, and spleens were harvested as previously described 42 .HSPCs were isolated from WBM using EasySep™ Mouse Hematopoietic Cell Isolation kit.All Animal studies were approved by St. Jude Children's Research Hospital Institutional Animal Care and Use Committee.

RNA-Seq Analysis
MOLM-13 cells or Lin-,cKit + HSPCs were harvested and RNA was extracted using a quick-RNA Microprep kit (Zymo Research, CA).RNA-seq was done using TruSeq Stranded Total RNA library kit (Illumina, CA) as previously described 43 .The RNA-Seq paired-end reads were mapped to the mouse mm10 genome or human hg38 genome using STAR and quanti ed using RSEM 44,45 .Differentially expressed gene analysis and GSEA was done as previously described 42,46 .

CUT&RUN Analysis
MOLM-13 cells or Lin-, cKit + HSPCs were harvested at 500,000 cells for each antibody probe.Genomic localization pro ling was performed using Cleavage Under Targets and Release Using Nuclease kit (CUT&RUN) according to the manufacture's protocols (Epicypher, NC) as previously described 47 .Library preparations of up to 6ng of isolated DNA fragments were done using NEBNEXT Ultra library prep Kit with AMPure XP beads (Beckman Coulter, CA) following the NEBNext® Ultra™ DNA Library Prep Protocol for Illumina® With Size Selection (E7370) V.2 (fragment size < 70bp).CUT&RUN libraries were sequenced using Novaseq 6000 by performing 100 cycles of paired-end sequencing (200 cycles total).CUT&RUN bam les were pre-processed and peaks were called using methods previously described 48 .Peaks from replicate samples were merged and fragments covering those peaks were counted using bedtools version 2.25 49 .Peaks differentially expressed between treatment groups were determined using a combination of the R tools limma and voom as previously described 46 .Peaks were annotated using HOMER version 4.10 50 .The annotation was simpli ed by combining the HOMER annotations into 3 categories.Peaks within 3kb of a transcription start site were called 'TSS +/-3kb'.Peaks annotated as intergenic were labeled as "Intergenic".Peaks annotated as non-coding were labeled "non-coding" and all other peaks were labeled 'Genebody'.

ATAC-seq Analysis
MOLM-13 cells or Lin-,cKit + HSPCs were harvested at 50,000 cells.Open chromatin status was assessed using the assay for transposase-accessible chromatin via sequencing (ATAC-seq) kit (Active Motif, CA) following the manufacturer's protocols.Library prep, included in kit, was done following the manufacturer's protocols.ATAC-seq libraries were sequenced using Novaseq 6000 by performing 100 cycles of paired-end sequencing (200 cycles total).Analysis was done by the Center for Applied Bioinformatics at SJCRH.Raw reads in fastq format were processed with Trim-Galore tool (v0.4.4,Krueger F. (2012)), in order to remove potential adapters and quality trim 3' end of reads with cutadapt program, followed by FastQC analysis 51,52 .A quality score cutoff of Q20 was used and the rst 15bp of each reads were also clipped to reduce the GC bias.Next, reads were mapped to the human reference genome (hg38; GRCh38.p12) with BWA mem (0.7.17-r1188), then converted to BAM format and deduplicated with fq2bam (v3.0.0.6) 53 .Subsequently, uniquely mapped properly paired reads were extracted with samtools (v1.2), and fragments were extracted with bedtools(v2.24.0) 49,54 .Chromatin status (open/closed) was de ned by a q-value of ≤ 0.05 and a log2FC of +/-2 compared to indicated WT control.

Figures
Figures

Figure 3 Loss
Figure 3

Figure 4 Loss
Figure 4