Genome-wide interference of ZNF423 with B-lineage transcriptional circuitries in acute lymphoblastic leukemia

One of the most characteristic hallmarks of acute lymphoblastic leukemia (ALL) is the maturation arrest occurring in B or T progenitor cells. This arrest has been linked to genetic alterations of transcription factors associated with cellular differentiation, such as NOTCH1, PAX5, Ikaros (IKZF1), E2A, and early B-cell factor 1 (EBF1), or to epigenetic mechanisms that critically interfere with their normal function.^1-7  Among those, Ebf1 has been established as one of the key regulators of early B-cell development.⁸  In addition, Ebf1 (alias Olf1) has been implicated in olfactory neuronal differentiation.⁹  Ebf1(Olf1) in neural development is functionally modulated by the zinc-finger protein 423 (Zfp423), which binds Ebf1 as a heterodimer leading to its sequestration and inhibition.^10,11  In contrast, Zfp423 or its human ortholog ZNF423 is not constitutively expressed during normal hematopoiesis, but its aberrant activity has previously been associated with B-cell malignancies.^7,12,13  Hypomethylation of a regulatory CpG island in conjunction with bone morphogenetic protein 2 (BMP2)-SMAD signaling has been identified as the underlying mechanism of ZNF423 activation in B-cell precursor ALL antagonizing EBF1 function to an unknown extent.⁷ 

Notwithstanding the indispensability of EBF1 in lymphopoiesis, we hypothesized that ZNF423 action in leukemogenesis was not confined to sequestration and inhibition of EBF1 alone, but could unfold additional intrinsic activities because of its multimodular structure. To elucidate the scope of ZNF423 action and interference with core transcriptional circuitries in B-lymphopoiesis on a genome-wide scale, we established a cellular model of B-cell leukemia in which we modulated ZNF423 activity to define its transcriptional targets in a comprehensive manner and to assess its impact on disease phenotype.

Human leukemia cell lines MV4-11, REH, Nalm6, CEM, MHH-Call3, 697, HeLa, SEM, and HEK293T cells were obtained from DMSZ (Braunschweig, Germany). Leukemia cell lines were maintained in RPMI-1640 medium supplemented with 10% fetal calf serum (FCS). HeLa and HEK293T cells were maintained in Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% FCS. Cells were grown in humidified incubators at 37°C and 5% CO₂. Authenticity of cell lines has been confirmed by short tandem repeat analysis. Cell lines were routinely tested for mycoplasma infection. Cell stimulation with BMP2 (HumanZyme) was performed at a concentration of 100 ng/mL reconstituted in 4 mM hydrochloric acid and 0.1% bovine serum albumin.

Transfection of HEK293T cells was carried out using Lipofectamine 2000 reagent (Invitrogen) according to the manufacturer’s instructions. Transfection of CRISPR-Cas9 constructs in leukemic cell lines was conducted using Amaxa Cell Line Nucleofector Kit R (Lonza) following the manufacturer’s recommendations. Per nucleofection, 5 × 10⁶ cells were transfected with 2 µg of vector by using program X01 on a Nucleofector II device (Amaxa Biosystems). At 24 hours after transfection, reporter-positive cells were single-cell sorted and subsequently raised to single-cell clones in conditioned media.

Murine stem cell virus particles were produced in HEK293T cells by using the ProFection Mammalian Transfection System–Calcium Phosphate (Promega) according to the manufacturer’s recommendations. Virus particles were concentrated 48 hours after transfection by centrifugation for 2 hours at 20 000 rpm, and target cells were resuspended in virus supernatant containing 8 µg/mL polybrene (Sigma) and spinoculated for 1 hour at 1000g at 32°C.

Bone marrow was obtained from 8- to 10-week-old C57BL/6 mice. Upon harvesting, cells were maintained in medium containing Iscove modified Dulbecco medium (IMDM)-GlutaMax (Thermo Fisher Scientific, 31980030), 20% FCS, 50 µM β-mercaptoethanol, murine interleukin-7 (mIL-7) at 10 ng/mL, and murine stem cell factor at 50 ng/mL. After 7 days of culture, B220⁺CD19⁺ cells were sorted using a FACSAria Fusion sorter.

For xenotransplantation, 1 × 10⁶ gene-modified or parental SEM cells were resuspended in 30 μL RPMI-1640 media and subsequently injected into the tibiae of sublethally irradiated (2 Gy) NOD.Cg-Prkdc^scidIl2rgt^m1Wjl/SzJ (NSG) mice. Animals were treated according to Arrive guidelines with approval by competent authorities of the City of Hamburg (BGV 43/13).¹⁴ 

Briefly, 25 μg crosslinked chromatin was sheared and immunoprecipitated using 5 μg of antibodies (supplemental Table 1). Immunoprecipitated chromatin was purified, and retrieved DNA fragments were incorporated into a library for next-generation sequencing. Chromatin immunoprecipitation sequencing (ChIP-seq) reads were mapped to the human genome (GRCh37/hg19) by using the Bowtie short-read aligner. Peaks were called using input as background with MACS14 software.^15,16  Called peaks were annotated using BEDTools (Quinlan Laboratory, University of Virginia). Motif discovery was performed using MEME (http://meme-suite.org/tools/meme) and HOMER motif discovery software.^17,18 

Total RNA from 5 × 10⁶ genetically modified and parental human pro-B ALL SEM cells was isolated using RNeasy Mini Kit (QIAGEN 74104). Library preparation was performed with NEBNext Ultra RNA Library Prep Kit for Illumina (NEB E7530) using 640 ng total RNA each. STAR aligner was applied to map sequencing reads to the human genome (GRCh38/hg38) and simultaneously counted per gene by enabling the quantMode option.¹⁹  Counts are based on the Ensembl Annotation 87. Subsequent data processing was done with DESeq2 and R.^20,21 

Cell viability was determined using a fluorescence-activated cell sorting (FACS)–based annexin-V/propidium iodide (PI) Apoptosis Kit (eBioscience, 88-8007). Proliferation was determined using the Click-IT EdU Cell Proliferation Kit for Imaging (Thermo Fisher Scientific, C10340) and FITC BrdU Flow Kit (BD, 557891).

For reporter assays, we used the Dual-Luciferase Reporter Assay System (Promega) and followed the manufacturer's recommendations. Luciferase activity was analyzed at 48 hours after transfection on an Infinite M200 (Tecan) device. Relative promoter activity was calculated as the ratio of target promoter-driven Photinus pyralis luciferase and constitutive Renilla reniformis luciferase activity. Western blotting was performed using the antibodies listed in supplemental Table 1. For a list of primers used for quantitative PCR (qPCR), see supplemental Table 2.

Publicly accessible data from the Gene Expression Omnibus (GSE21455, GSE13159, and GSE74812) were downloaded in raw txt, CEL, or fastq format. Differentially expressed genes (DEGs) for GSE21455 and GSE13159 were estimated with limma, and CEL files were imported into R by oligo.^22,23  GSE74812 was processed as described under “Sequencing.” Data from Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and WikiPathways enrichment analysis were obtained using the Enrichr suite (gene list enrichment analysis tools) (https://amp.pharm.mssm.edu/Enrichr/).²⁴  ChIP-X enrichment analysis was performed using ChEA3 software (https://amp.pharm.mssm.edu/chea3/) according to authors’ guidelines.²⁵ 

Statistical analyses were performed using Student t test or univariate or multivariate analysis of variance with repeated or independent measures, followed by either Bonferroni post hoc test or contrast analysis (Holm-Šídák method) to determine significant differences. Respective P values were calculated using GraphPad Prism 6.0. P values ≤ .05 were considered statistically significant. False discovery rate or q value ≤ 0.1 was considered statistically significant.

To comprehensively explore the prevalence of aberrant ZNF423 in acute leukemia, we analyzed accessible data sets from Microarray Innovations in Leukemia (MILE), which included 1354 samples from diverse age groups (at childhood, adolescence, and adulthood) that are composed of 564 B-cell precursor ALL (BCP-ALL), 174 T-cell ALL (T-ALL), and 542 acute myeloid leukemia (AML) samples with diverse karyotypes, as well as normal bone marrow cells.²⁶ Z-scores for ZNF423 expression in individual samples in each subgroup were calculated as depicted in Figure 1A. Intriguingly, the prevalence of ZNF423 across various ALL subgroups was substantially greater than anticipated, especially in adult BCP-ALL and T-ALL. ETV6-RUNX1–rearranged ALL carrying t(12;21) revealed the highest average expression of ZNF423, which confirms previous results (supplemental Figure 1).⁷  In contrast, ZNF423 was largely absent in normal bone marrow as well as AML samples. In a correlative analysis of differential gene regulation that was dependent on ZNF423 abundance in 4 major genetic subgroups of ALL, we noted that ZNF423 had the greatest impact on gene regulation in TCF3-PBX1–rearranged ALL denoted as t(1;19) (Figure 1B).

As a next step, we sought to establish a cellular system that allowed for the interrogation of ZNF423 function in an EBF1 and SMAD1/SMAD4-dependent context on a genome-wide scale. Among a variety of leukemia cell lines, KMT2A-AFF1–rearranged pro-B ALL SEM cells expressed the most abundant EBF1 as well as ZNF423 messenger RNA, largely recapitulated at the protein level (Figure 2A-B). Importantly, SEM cells exhibited phospho-activity of SMAD1/5 at baseline, which was markedly induced by BMP2 stimulation, indicative of an intact BMP2-SMAD signaling pathway leading to transactivation of ZNF423, as previously shown (Figure 2B-C).⁷  Hence, pro-B SEM cells were chosen as an experimental model to address the impact of ZNF423 action on EBF1 and SMAD1/SMAD4-dependent transcriptional programs on a genome-wide scale. To modulate ZNF423 function in this cellular model, we generated a ZNF423 knockout (ZNF423-KO) model using CRISPR-Cas9–mediated cleavage targeting exon 4 of ZNF423, thus ablating its α and β isoforms (supplemental Figure 2A). Conversely, parental SEM cells were transduced with murine stem cell virus to express recombinant FLAG-tagged ZNF423 (overexpressed ZNF423-OE) (Figure 3A). Loss-of-function of ZNF423 in SEM cells was recapitulated by using a recombinant ZNF423-deletion mutant analogous to the Cas9-ablated gene product, which was unable to repress transactivation of the CD79a promoter by EBF1 (Figure 3B).

In this ZNF423-modified cellular model, transcript levels of EBF1 target genes CD79A and CD79B were decreased upon ZNF423 overexpression and increased in the absence of ZNF423, whereas IGLL1 and VPREB1 transcripts were unaffected, which confirmed previous in vitro results (supplemental Figure 2B).⁷  Next, we assessed the biological phenotype of ZNF423-OE compared with ZNF423-KO SEM cells. ZNF423 ablation in low-passage SEM cells caused a loss of viability by induction of apoptosis rather than decreased proliferative activity as demonstrated by FACS and 5-ethynyl-2′-deoxyuridine (EdU) incorporation analyses (Figure 3C; supplemental Figure 2C). To assess a ZNF423-dependent growth or survival phenotype in an in vivo setting, we transplanted equal numbers (1 × 10⁶ cells per animal) of living ZNF423-OE, ZNF423-KO, and parental SEM cells into immunodeficient NSG mice (n = 8 per cohort). Parental or ZNF423-OE pro-B ALL recipient animals developed rapid-onset leukemia. In contrast, mice transplanted with ZNF423-KO pro-B ALL showed a significantly delayed onset of overt disease and prolonged progression-free survival associated with a smaller spleen size on average and a lower proportionate number of human CD45⁺ ALL cells at the experimental end point, as quantified by FACS (Figure 3D; supplemental Figure 2D-E). By contrast, exogenous expression of Zfp423 was nonpermissive in a primary murine B-progenitor context, and it caused a rapid depletion of the CD19⁺ cell population under IL-7 stimulation by an increase in cell cycle arrest and apoptotic death (Figure 3E-F; supplemental Figure 2F-G).

To determine direct target genes of ZNF423, we conducted ChIP-seq in ZNF423-OE and parental SEM cells using a ZNF423-directed antibody. Furthermore, we sought to investigate the impact of ZNF423 on the genome-wide binding of EBF1, so we performed EBF1-directed ChIP-seq in our pro-B ALL model. ChIP-qPCR confirmed enrichment of EBF1 as well as ZNF423 at CD79A and CD79B loci (Figure 4A; supplemental Figure 3A). To determine whether ZNF423 re-distributed EBF1 in its chromatin-bound state, we compared EBF1 enrichment in the presence or absence of ZNF423. The proportional regional binding of EBF1 was only marginally affected by ZNF423. EBF1 as a pleiotropic transactivating, repressing, and pioneering factor predominantly bound to intragenic sites (∼60% of peaks) of the genome.²⁷  Promoter binding by EBF1 was attributed to 4% of identified enrichment peaks (Figure 4B). At proximal promoter regions, ZNF423 was more frequently observed (14% of peaks ≤1 kb from transcriptional start sites [TSSs]) but less frequently in intergenic regions (21% of peaks) compared with EBF1 (∼38% of peaks). At proximal upstream regulatory sequences, both factors were enriched around TSSs (supplemental Figure 3B).

Although the majority of EBF1-binding peaks were shared between distinct SEM clones, a substantial number of EBF1-enrichment peaks were exclusively identified in ZNF423-OE (n = 7308), ZNF423-KO (n = 3539), and parental ZNF423-expressing cells (n = 1238), suggesting that EBF1 was partially redirected by ZNF423 in its chromatin-bound state (Figure 4C). An overrepresentation analysis of functional categories of promoter-enriched, nonoverlapping EBF1 peaks (n = 1351) in ZNF423-OE cells identified DNA damage response pathways previously linked to ZNF423 in ciliopathies (supplemental Figure 3C).²⁸  At upstream regulatory sequences (–2 kb to +1 kb relative to TSSs), many binding sites were shared between EBF1 and ZNF423, but a substantial proportion were exclusively bound by either factor alone. ZNF423 binding to EBF1 consensus sites in the absence of EBF1 was unanticipated and suggests an EBF1-independent function of ZNF423, although we cannot rule out a transient presence of EBF1 at those sites during gene priming for instance (Figure 4D).²⁹ 

Motif search analyses of ChIP-seq data generated from ZNF423-OE cells using the HOMER suite revealed a ZNF423-binding motif that resembled a previously described 5′-TCCCNNGGGA-3′ EBF1 motif (Figure 4E-F).³⁰  When comparing HOMER-generated ChIP-seq motifs with databases of known transcription factor binding sites, the most significant motif displayed the strongest similarity to the deposited EBF1 consensus binding site, including the ZNF423-binding motif. The core palindromic binding motif (5′-CCCNNGGG-3′) of either factor was identical; however, the EBF1-binding motif was flanked by additional 5′-T and 3′-A nucleotides (Figure 4E-F). A motif recognition subset analysis of ZNF423 peaks not shared with EBF1 in the proximity of transcription start sites (–2 kb to +1 kb) reconfirmed an enriched EBF1 consensus motif (supplemental Figure 3D). We concluded that ZNF423 was able to bind at EBF1 consensus sites in both the presence and the absence of EBF1. Taking into account further cooperative activities of DNA-binding factors in the vicinity of EBF1 and ZNF423 consensus sites, we performed additional motif search analyses and identified enriched binding sites of other transcription factors such as RUNX1, ZIC1, and CTCFL (Figure 4E-F). RUNX1 plays a prominent role in hematopoiesis, but ZIC1 has been implicated as a transcriptional activator in neurogenesis.³⁰  CTCFL also designated as brother of the regulator of imprinted sites (BORIS) has recently been described in chromatin looping and superenhancer formation in treatment-resistant neuroblastoma, whereas ZNF423 is required for retinoic acid–induced differentiation of neuroblastoma.^31-33 

Given the broad genomic binding of ZNF423, we asked whether ZNF423 would shape the chromatin landscape in pro-B ALL cells. To this end, we applied ChIP-seq using antibodies directed against activating and repressive histone marks in ZNF423-OE and ZNF423-KO cells targeting H3K4me3 and H3K27ac as activating marks including enhancer regions, and H3K27me3 and H3K9me3 as repressive marks. In a meta-analysis, the density of histone marks was visualized in the vicinity of EBF1 and ZNF423 ChIP-seq peaks (Figure 5A). Upon experimental deletion of ZNF423, activating histone marks H3K4me3 and H3K27ac were clearly enriched at EBF1 and ZNF423 binding sites. Conversely, overexpression of ZNF423 caused a significant reduction in activating marks rather than an increase in repressive histone marks (H3K27me3 and H3K9me3), which was reconfirmed at the CD79A and CD79B loci (supplemental Figure 4A). The genome-wide abundance of activating marks was markedly depleted by ZNF423, consistent with a recruitment of nucleosome remodeling deacetylase (NuRD) or other multiprotein repressor complexes to regions bound by ZNF423 (Figure 5B). Subanalyses confined to either EBF1- or ZNF423-bound regions exhibited a similar histone composition (supplemental Figure 4B-C).

To gain further insight into ZNF423-modulated enhancer gene regulatory elements, we analyzed publicly accessible ChIP-seq data on H3K27ac and H3K4me1 marks in SEM cells, which were aligned with ZNF423–ChIP-seq data generated herein.³⁴  We retrieved 9926 genomic locations with overlapping H3K27ac/H3K4me1 peaks outside promoter regions that reflect enhancer activity. Among those, 122 enhancers revealed an enrichment of ZNF423 (supplemental Table 3). Exemplarily, an intronic enhancer at the PIK3C2B locus was bound by both EBF1 and ZNF423, and it displayed a marked decrease in activating H3K27ac marks under ZNF423-OE (supplemental Figure 4D).

To define ZNF423-target genes, we examined the transcriptional regulation of parental or ZNF423-OE vs ZNF423-KO SEM cells by transcriptome-wide RNA sequencing (RNA-seq). As predicted, the greatest difference was noted between ZNF423-OE and ZNF423-KO, which showed 1410 DEGs (603 downregulated vs 807 upregulated genes) (log2 fold change ± 0.6; P ≤ .05; false discovery rate ≤0.1) (supplemental Figure 5A-B). In an effort to evaluate the usefulness and validity of the established ZNF423-modified pro-B ALL model, we focused on ZNF423-dependent differentially regulated EBF1-candidate genes identified in this study. In an unsupervised cluster analysis, a heatmap illustrates 3 clusters each of downregulated (n = 249) and upregulated (n = 133) EBF1-target genes (Figure 6A; supplemental Figure 5C). In pathway analysis with GO gene sets of those clusters, enriched functional categories of downregulated genes in ZNF423-OE cells were closely related to leukocyte development or B-cell activation as well as other signaling pathways involved in B-cell differentiation (Figure 6B, top). Similar functional categories were obtained by gene set enrichment analysis (supplemental Figure 5D-E). In turn, the analysis of ZNF423-dependent upregulated genes revealed functional categories involved in cell migration and others (Figure 6B, bottom). Interestingly, EBF1-dependent repression of a subset of alternative hematopoietic lineage factors such as CEBPA and SPI1 (PU.1) was further augmented by ZNF423 (supplemental Table 4).³⁵  Target genes requiring the pioneering activity of EBF1 such as IRF4 were not discernibly affected by ZNF423 in our pro-B ALL model.²⁶  In this regard, an EBF1-ZNF423 heterodimer might preserve the pioneering activity, which is mediated by the C-terminal domain of EBF1, or possibly ZNF423 does not bind EBF1 C-terminus–dependent target genes. In line with the latter notion, we observed full repressive activity of ZNF423 at the CD79B promoter in the absence of the C-terminal binding domain of EBF1 (supplemental Figure 5F).

Next, we asked which additional transcription factors might feed into EBF1-dependent ZNF423-modulated transcriptional circuits. To this end, we performed a ChIP-X enrichment analysis (ChEA) that ranked transcription factors (TFs) upstream of RNA-seq–generated DEGs. Upon ZNF423-OE, the majority of upstream TFs were involved in B-lineage differentiation, including EBF1, STAT1, IKZF1, GATA1, and PAX5 as top-ranked candidates. By contrast, top-ranked TFs in ZNF423-upregulated DEGs were predominantly associated with stem cell and nonhematopoietic cell lineages such as SOX2 (Figure 6C; supplemental Table 5).

To assess the extent to which ZNF423-mediated repression of EBF1 targets in pro-B ALL recapitulated a genetic loss of Ebf1 in primary lymphopoietic cells, we analyzed overlapping EBF1/Ebf1-dependent orthologs in expression data sets from Ebf1^−/− vs Ebf1^−/−::Ebf1 pre-pro-B cells.³⁶  Clustered differentially regulated mouse orthologs from Ebf1^−/− vs Ebf1^−/−::Ebf1 genotypes and corresponding human DEG orthologs from pro-B ALL are presented in Figure 6D. Importantly, we re-identified components of the pre-BCR/BCR and downstream Pik3/Akt signaling pathways such as Cd79b, Pik3c2b, and Hck, which were downregulated upon ZNF423-OE or Ebf1 deletion and vice versa.^36,37  In addition, we identified genes from other functional categories such as Mst1, Map4k1, Tgfb1, Cdkn1a, or Tcf7 involved in DNA damage response, Hippo-Yap signaling, and Tgfb signaling. Among upregulated genes, Wnt-signaling effectors (Mdfic, Cbfb, and Nkd1) were discernible (supplemental Figure 5G-I).

SMADs are recognized as binding partners of ZNF423 mediated through its central zinc finger (ZF) domain (ZF14-19), while ZF9-13 bind DNA at BMP response elements (BREs).^38-40  In addition, BMP2-dependent phospho-activation of SMAD1 causes transcriptional upregulation of ZNF423 in BCP-ALL.⁷  In an effort to link the BMP-SMAD-ZNF423 axis to EBF1-dependent transcriptional circuitries, we searched for SMAD1/SMAD4 consensus motifs in promoter regions of genes regulated by EBF1 and/or ZNF423 using the consensus motif for BMP-responsive SMADs (supplemental Figure 6A).⁴¹  ZNF423-modulated EBF1-dependent DEGs showed an enrichment of highly prevalent SMAD1 motifs over a random gene set (n = 58 051) (supplemental Figure 6B). To decipher potential ZNF423 targets co-regulated by SMAD1/SMAD4 in a more rigorous manner, we selected ZNF423-regulated DEGs by density of SMAD1 sites (n ≥ 5) at any given promoter (supplemental Figure 6C-D). ZNF423-repressed genes exhibiting multiple SMAD-binding sites were functionally assigned to myeloid activation, T-cell differentiation, and calcium signaling (supplemental Figure 6E).^42,43  Intriguingly, ChIP-seq analysis identified an enrichment of EBF1 and ZNF423 at the TGFB1 promoter, which contains several SMAD1 motifs in the vicinity of EBF1 consensus sites (Figure 7A). Transforming growth factor beta 1 (TGF-β1)–dependent pathways have previously been described in B-cell differentiation and leukemogenesis.⁴⁴  To assess the effect of EBF1 and ZNF423 on TGFB1, reporter analyses were carried out that exhibited an increase in transactivation by EBF1, which was clearly abrogated by ZNF423 outcompeting EBF1 in a concentration-dependent manner. Conversely, increasing concentrations of EBF1 did not abrogate ZNF423-mediated repression of TGF-β1, which is indicative of a dominant-inhibitory function of ZNF423 (Figure 7B; supplemental Figure 6F). To delineate the functionally relevant binding sites, we performed a mutagenesis of the most proximal EBF1 consensus site (–105 bp from TSS) derived from ChIP-seq data and computation-derived adjacent SMAD1-binding sites (–244 and −250 bp from TSS). Transactivation of the TGFB1 promoter was markedly decreased by disruption of the SMAD1-binding sites, which was indicative of strong co-regulation by SMAD1/SMAD4 and EBF1 (Figure 7C). ZNF423 unfolded its full repressive activity in the presence of an intact proximal EBF1 consensus site. To decipher the protein-binding domains of ZNF423 that are indispensable for its repressive effect on TGFB1 transactivation, we created several deletion mutants of ZNF423 that disrupted its DNA-binding domain (ZF2-8), BRE-binding domain (ZF9-13), SMAD-interacting domain (ZF14-19), or EBF1-binding C-terminus (ZF20-30). Wild-type and BRE-binding domain deleted ZNF423 maintained a marked repressive effect on TGFB1 transactivation. In contrast, deletion of the N-terminal DNA-binding, central SMAD-interacting, or C-terminal EBF1-interacting domain resulted in a substantial loss of repressive activity of ZNF423 toward TGFB1, which reflects an interaction of ZNF423 with DNA, SMAD1/SMAD4, and EBF1 (Figure 7D). However, C-terminal truncation of ZNF423 also interferes with its ability to homodimerize, which potentially compromises its function independent of EBF1 binding.¹⁰ 

To further evaluate ZNF423 function in a TGF-β1-proficient cellular context, we analyzed primary murine hematopoietic cells under IL-7–induced B-cell differentiation conditions. Exogenous expression of Zfp423 recapitulated its repressive effect on Tgfb1 transcripts, whereas Tgfbr2 was not affected. As anticipated, Tgfb1 treatment caused an increased expression of the cell cycle inhibitors Cdkn1a, Cdkn1b, and in particular Cdkn2b. In contrast, retrovirus-driven Zfp423 caused a downregulation of Cdkn1a, Cdkn1b, and Cdkn2b in the absence of exogenous TGF-β1, which was only partially reversed by the addition of TGF-β1, illustrating the disruptive effect of Zfp423 on important downstream effectors of Tgfb1 signaling (Figure 7E-F; supplemental Figure 6G).

Disruption of EBF1 has profound biological effects on B-cell lymphopoiesis, clearly discernible in genetic Ebf1-KO mouse models.^8,36  Moreover, its haploinsufficiency has been implicated as a causative factor in B-cell leukemogenesis.^3,45  However, the frequency of inactivating mutations of EBF1 that consist mostly of monoallelic deletions and structural variants such as gene fusions is rather low in B-precursor ALL (∼4%) compared with other oncogenic events affecting B-lymphopoiesis such as PAX5 or IKZF1.³ EBF1 aberrations are more prevalent in high-risk ALL, including BCR-ABL and Philadelphia chromosome-like ALL as well as TP53-, RB1-, and JAK-mutated subtypes.^46-50  An analysis of accessible integrated genomic and transcriptomic data on ALL revealed that the expression of ZNF423 is independent of EBF1 mutational status.⁴⁹  In contrast to the rarity of EBF1 mutations, ZNF423 is highly prevalent in various subtypes of ALL which might mimic an EBF1 loss-of-function genotype and might also distort the chromatin-bound state of EBF1 and unfold intrinsic EBF1-independent activities contributing to leukemogenesis. Thus, ZNF423 might constitute an additional B-lineage gene depletion mechanism that aggravates the differentiation defect in B-precursor cells. In line with this notion, transposon-mediated overexpression of Zfp423 has been shown to cooperate with Etv6-Runx1 in B-precursor leukemogenesis.⁵¹  Moreover, in a different context of p210 BCR-ABL–driven chronic myelogenous leukemia, aberrant ZNF423 has been implicated in B-cell blast crisis, which aggravates the clinical phenotype.¹³  Inversely, CRISPR-Cas9–mediated genetic ablation of ZNF423 prolonged survival upon pro-B ALL xenotransplantation as described above (Figure 3D).

The genome-wide binding pattern of ZNF423 and EBF1 identified in this study implies that ZNF423 predominantly targets genes transactivated by EBF1, thus disrupting an essential B-lineage program. In addition, a substantial proportion of EBF1 was found to be displaced or re-directed to different genomic loci by ZNF423 which was discernible by nonoverlapping enrichment peaks under conditions of ZNF423-OE vs ZNF423-KO (Figure 4C). As an unanticipated, previously unrecognized finding, ZNF423 seemed to bind EBF1 consensus sites in the absence of a discernible enrichment of EBF1 at those sites. Accordingly, an EBF1-PBD-domain deleted ZNF423-mutant partially maintained repressive activity (Figure 7D). Nevertheless, ZNF423 activity is clearly dependent on intact DNA and protein interaction domains as demonstrated at the TGFB1 promoter. We infer a variable mechanism of ZNF423 action based on EBF1 displacement by a ZNF423 homodimer, a ZNF423/EBF1 heterodimer, or a ZNF423/SMAD heterodimer. As previously described, ETV6-RUNX1 abrogates TGF-β1–mediated upregulation of Cdkn1b in B-precursor cells resulting in an expansion of a preleukemogenic ETV6-RUNX1 cell population.⁴⁴  Similarly, ZNF423 might interfere with TGF-β1 function by direct transcriptional repression through binding of EBF1 consensus sites and interaction with SMAD1/SMAD4 complexes or EBF1. Thus, TGF-β1–dependent Cdkn1a and Cdkn2b were downregulated under ZNF423/Zfp423-OE, as shown in Figure 7F and supplemental Table 4.

The predominant net effect of ZNF423 action led to significant attenuation of EBF1-dependent transcription, as shown by downregulation of critical EBF1 targets such as pre-BCR and associated PI3K-AKT or RAS-MAPK signaling pathways without compromising cell viability. Given its wide prevalence and impact on differential gene regulation in various subtypes of primary human ALL, ZNF423 might play a hitherto underestimated role in leukemogenesis.

The authors thank members of the flow cytometry core unit at Heinrich-Pette-Institute and members of the Horstmann laboratory for valuable input.

This work was supported by a grant from the Deutsche Forschungsgemeinschaft (HO2176/4-1) (M.A.H.), Madeleine Schickedanz KinderKrebsStiftung, J.J. Ganzer Stiftung, Hans-Brökel-Stiftung für Wissenschaft und Kultur, Burkhard-Meyer-Stiftung, and by the Fördergemeinschaft Kinderkrebszentrum Hamburg.

Contribution: P.I., A.-C.P., and S.R. performed in vitro and in vivo experiments; M. Spohn performed bioinformatics analyses of microarray, ChIP-seq, and RNA-seq data; M.K. generated CRISPR-Cas clones; J.M. performed animal experiments; L.B. and D.I. performed library preparation and deep sequencing; P.I., M. Seoane, and M.A.H. analyzed data and wrote the manuscript; and M.A.H. conceived and supervised the project.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Martin A. Horstmann, Research Institute Children’s Cancer Center Hamburg, Department of Pediatric Hematology and Oncology, Hamburg University Medical Center, Martinistr 52, 20246 Hamburg, Germany; e-mail: horstman@uke.de.