UHRF1 is highly expressed in AML and high UHRF1 level predicts poor prognosis

To determine the expression of UHRF1 in AML, we analyzed the bone marrow (BM) samples of AML patients and healthy controls by using RT-qPCR, Western blotting and immunohistochemistry assays. The results showed that the UHRF1 mRNA and protein levels were higher in AML patient samples compared with the healthy controls (Fig. 1a‒c; Supplementary information, Fig. S1a). We then analyzed the international Microarray Innovations in Leukemia (MILE) data,35 which showed that UHRF1 expression is particularly higher in the t(8;21), t(11q23), inv(16) and t(15;17) AML samples relative to healthy HSC controls (Fig. 1e). Moreover, we found that UHRF1 is highly expressed in CD34+ AML cells, leukemia stem cells (LSCs)36 from AML patients and AML cell lines (Fig. 1d; Supplementary information, Fig. S1b).

Fig. 1: High expression of UHRF1 predicts poor prognosis in AML.

a The qPCR analysis of UHRF1 expression in BM mononuclear cells from AML patients (M2, n = 9; M5, n = 11) and healthy subjects (n = 8). b The Western blotting analysis of UHRF1 in BM cells of AML patients (n = 16), healthy mononuclear cord blood cells (MNCs) (n = 4) and CD34+ HSPCs (n = 4). c Quantification of the Western blotting analysis for b. d The microarray analysis of UHRF1 in CD34+ leukemia cells and LSCs of AML patients from GSE76009. (CD34+, n = 110; CD34‒, n = 117; n-LSC+, n = 138; n-LSC‒, n = 89). e The differential expression of UHRF1 in mononuclear BM or PB cells of AML patients [t(15;17), n = 87; Inv(16), n = 77; t(11q23), n = 88; t(8;21), n = 98)] and HSCs in healthy subjects (HSC, n = 6). Data were obtained from the microarray analysis of bloodpool. The samples were normalized and batch corrected using ComBat for full completeness of the dataset. PCA analysis and gene signature values were then calculated. f The event free survival of patients with t(8;21) leukemia was stratified by UHRF1 expression into UHRF1 high (n = 27) and low (n = 51) groups. The survival days (UHRF1 high: 510 days; UHRF1 low: 1067 days) mean the date of the event occurrence in AML patients such as relapse and drug resistance, etc. g The differential expression of UHRF1 in the relapsed (n = 14) and non-relapsed (n = 35) t(8;21) leukemia patients. h MLL-AF9 localizes at the promoter region of Uhrf1 in the CUT&Tag analysis of 32D cells transduced with MLL-AF9. i AML1-ETO, HEB, E2A and LMO2 colocalize at the gene body region of UHRF1 in the ChIP-seq analysis of Kasumi-1 cells. j, k The mouse BM cells transduced with MLL-AF9 (j) or AE9a (k) were analyzed. The anti-Uhrf1, anti-MLL1, anti-HA and anti-Actin antibodies were used for the Western blotting analysis. l, m Knockdown of AML1-ETO in Kasumi-1 cells decreased the mRNA (l) and protein (m) levels of AML1-ETO and Uhrf1. The anti-Uhrf1, anti-ETO and anti-Actin antibodies were used for the Western blotting analysis. Data are all presented as means ± SD. Statistical analyses were performed using Student’s unpaired t-test for a, c–e, g, l. The expression values for e were log2 transformed. Statistical significance was evaluated by log-rank test for f. The mRNA expression for g were statisticed by FPKM from RNA-seq. *P < 0.05; **P < 0.01; ***P < 0.001.

As UHRF1 expression is positively correlated with cancer cell proliferation,37 it would be interesting to determine how UHRF1 would associate with the prognosis in a specific subtype of AML. To this end, we analyzed our data of t(8;21) AML patients,38 and the results showed that the patients with high UHRF1 expression had a significantly worse event-free survival compared with the patients with low UHRF1 expression (Fig. 1f). Consistent with this finding, the UHRF1 expression levels in the relapsed t(8;21) AML patients were significantly higher than those of the non-relapsed patients (Fig. 1g). Thus, these results imply a potential role of UHRF1 as a predictor of prognosis of AML patients.

To further investigate potential mechanism of UHRF1 upregulation in AML, we analyzed our chromatin immunoprecipitation (ChIP)-seq data for AETFC (Kasumi-1 cell) and CUT&Tag data for MLL-AF9 (32D cell). Intriguingly, the results revealed that AETFC components including AML1-ETO (AE), E2A, HEB and LMO2 strongly bind to the gene body of Uhrf1 (Fig. 1i) and MLL-AF9 also binds to the promoter regions of Uhrf1 (Fig. 1h). To determine whether Uhrf1 is upregulated in direct response to AE or MLL-AF9, we retrovirally transduced murine BM cells with MIGR1-AE, MIGR1-MLL-AF9 or the MIGR1 vector, and found the elevated Uhrf1 protein in AE or MLL-AF9-transduced cells (Fig. 1j, k). We then used the shRNA against the breakpoint region of AML1-ETO to knock down AML1-ETO in Kasumi-1 cells and observed decreased UHRF1 mRNA and protein levels in Kasumi-1 cells with around 70% knockdown (Fig. 1l, m). These data indicate that UHRF1 may be a potential target of the leukemogenic fusion proteins. Together, these results suggest that UHRF1 is highly expressed in AML, and UHRF1 may indicate potential prognosis of AML patients.

Uhrf1 is required for the maintenance and progression of AML

To investigate the role of Uhrf1 in leukemia maintenance, we utilized Uhrf1 conditional knockout mice for AML studies. We isolated the E14.5 fetal liver (FL) cells and adult BM cells from Uhrf1fl/fl and Uhrf1fl/flMx1-Cre mice, which were transduced with the retrovirus expressing AML1-ETO9a (AE9a) or MLL-AF9, respectively. The lethally irradiated recipient mice were injected with these transduced cells via tail vein (Fig. 2a), and poly(I:C) was injected to the recipients to induce Uhrf1 deletion. The Western blotting analysis showed that Uhrf1 was significantly depleted in the leukemia cells of Uhrf1fl/flMx1-Cre recipients upon the poly(I:C) administration (Fig. 2c). The recipients receiving AE9a-transduced Uhrf1fl/flMx1-Cre FL cells failed to develop AML post-poly(I:C) induction, while the control recipients develop AML with a latency of 167 days (Fig. 2b). In the mouse model of MLL-AF9-driven leukemia, the MLL-AF9Uhrf1Δ/Δ mice showed a significantly longer median survival (74 days) as compared with the MLL-AF9Uhrf1fl/fl mice (54.5 days; Fig. 2b). The AE9a/MLL-AF9Uhrf1Δ/Δ mice had significantly smaller spleen (Fig. 2d), and lower white blood cell (WBC) counts as compared with AE9a/MLL-AF9Uhrf1fl/fl mice (Fig. 2e). Meanwhile, the Wright’s staining analysis revealed lots of leukemia blasts emerged in the peripheral blood (PB), bone marrow (BM) and spleen cells of AE9a/MLL-AF9Uhrf1fl/fl mice, whereas normal components of hematopoietic cells in AE9aUhrf1Δ/Δ mice and less leukemia blasts in MLL-AF9Uhrf1Δ/Δ mice were shown (Fig. 2f). Similarly, the histology analysis indicated that the AE9a/MLL-AF9Uhrf1Δ/Δ mice had less infiltration of leukemia in the BM, spleen and liver compared with AE9a/MLL-AF9Uhrf1fl/fl mice (Supplementary information, Fig. S2a, b). The flow cytometry analysis showed that BM cells of AE9a/MLL-AF9Uhrf1Δ/Δ mice contained far fewer GFP+c-Kit+ leukemia blast cells and more GFP‒Mac-1+ (i.e., normal) cells compared with AE9a/MLL-AF9Uhrf1fl/fl mice (Fig. 2g‒i).

Fig. 2: Uhrf1 is required for the maintenance and progression of AML in mouse models of AE9a- or MLL-AF9-driven leukemia.

a The strategy of AE9a-expressing fetal liver cell transplantation (FLT) or MLL-AF9-expressing BM cell transplantation (BMT). Poly(I:C) was injected to AE9a recipients or MLL-AF9 recipients on week 4 or week 2 respectively. b Conditional deletion of Uhrf1 by poly(I:C) treatment significantly prolongs the survival time of recipient mice transplanted with AE9aUhrf1fl/fl Mx1-Cre (n = 12) or MLL-AF9Uhrf1fl/fl Mx1-Cre (n ≥ 16) cells. c The expression of Uhrf1 is minimal in the sorted GFP+ cells isolated from the spleen of the MLL-AF9Uhrf1fl/fl /MLL-AF9Uhrf1Δ/Δ mice. d The size and weight of the spleen were decreased in the AE9aUhrf1Δ/Δ and MLL-AF9Uhrf1Δ/Δ mice (n ≥ 3). e The WBC counts of AE9aUhrf1Δ/Δ or MLL-AF9Uhrf1Δ/Δ mice were significantly lower than AE9aUhrf1fl/fl or MLL-AF9Uhrf1fl/fl mice (n ≥ 6). f The PB, BM and spleen show less leukemia blast cells in the AE9a/MLL-AF9Uhrf1Δ/Δ mice group compared with the AE9a/MLL-AF9Uhrf1fl/fl group. g‒i Representative flow cytometry profiles (g) and quantification of the frequencies (h, i) of GFP+c-Kit+ leukemia blast cells and normal GFP−Mac1+ cells in the BM cells of AE9a/MLL-AF9Uhrf1Δ/Δ mice compared with AE9a/MLL-AF9Uhrf1fl/fl mice (n ≥ 3). j Loss of Uhrf1 significantly prolongs the survival time of recipient mice transplanted with MLL-AF9Uhrf1Δ/Δ cells compared with MLL-AF9Uhrf1fl/fl group (n = 15). k Knockdown of Uhrf1 significantly prolongs the survival time of recipient mice transplanted with AE9a cells, compared with the control shRNA (n ≥ 11). l, m The in vivo bioluminescence imaging (l) and quantification analysis (m) shows that knocking down Uhrf1 impairs the leukemia progression in recipient mice transplanted with AE9a or MLL-AF9 leukemia cells that are luciferase-positive (n = 5). Data are all presented as means ± SD. Statistical significance was evaluated by log-rank test for b, j and k. Statistical analyses were performed using Student’s unpaired t-test for e, h, i, m. *P < 0.05, **P < 0.01, ***P < 0.001.

To further identify the role of Uhrf1 in AML progression, we transplanted 1 × 105 MLL-AF9-expressing Uhrf1fl/fl or Uhrf1Δ/Δ leukemia cells into sublethally irradiated recipient mice. Uhrf1 deficiency strikingly prolonged the median survival of the secondarily transplanted recipients (39 days vs 18 days) (Fig. 2j). Two weeks after transplantation, the WBC counts of the MLL-AF9Uhrf1Δ/Δ recipients were significantly lower than those of the MLL-AF9Uhrf1fl/fl recipients, and the red blood cell (RBC) counts, platetlet (PLT) counts and hemoglobin (HGB) concentrations were higher, reflecting the impaired leukemogenesis (Supplementary information, Fig. S2c). The splenomegaly and hepatomegaly were prominently observed in the MLL-AF9Uhrf1fl/fl recipients, but were less in the MLL-AF9Uhrf1Δ/Δ recipients (Supplementary information, Fig. S2d, e). The flow cytometry analysis showed that nearly 90% of BM cells of Uhrf1fl/fl recipients were GFP+Gr-1+ or GFP+Mac-1+, but only 10% of BM cells in Uhrf1Δ/Δ recipients were GFP+Gr-1+ or GFP+Mac-1+ (Supplementary information, Fig. S2f). Meanwhile, the peripheral blood smear analysis, the cytospin analysis of the BM and spleen cells (Supplementary information, Fig. S2g), and the histology analysis of the BM, SP, and liver (Supplementary information, Fig. S2h) revealed less leukemic blasts in the MLL-AF9Uhrf1Δ/Δ recipients than the MLL-AF9Uhrf1fl/fl recipients. We also sought to knock down Uhrf1 in AE9a or MLL-AF9 cells and evaluate their growth potential in the sublethally irradiated recipients. The mice receiving the shUhrf1-expressing AE9a or MLL-AF9 cells had a longer life span than the mice receiving scrambled shRNA-transduced cells (Fig. 2k; Supplementary information, Fig. S3a, b). These mice also had smaller spleen size, suggesting the leukemogenicity of AE9a or MLL-AF9 leukemia cells was reduced by Uhrf1 knockdown in vivo (Supplementary information, Fig. S3c‒f). Since these cells are luciferase-positive, we detected more intense bioluminescent signals in the control mice compared with the recipients receiving the shUhrf1-expressing AE9a or MLL-AF9 cells (Fig. 2l, m). Together, these results suggest that Uhrf1 is required for the maintenance and progression of AML.

Uhrf1 deficiency impairs the self-renewal of leukemia initiating cells (LICs) and decreases the frequency of LICs

To explore the role of Uhrf1 in self-renewal of LICs, we crossed the Cre-ER transgenic mice with Uhrf1fl/fl mice to generate Uhrf1fl/fl Cre-ER mice. E14.5 FL or adult BM cells of Uhrf1fl/fl Cre-ER mice were transduced with AE9a or MLL-AF9 (MA-9) and LICs were sorted to the culture medium (Fig. 3a). We used 1 µM 4-hydroxy-tamoxifen (4-OHT) to induce Uhrf1 deletion in vitro and the Western blotting analysis showed that Uhrf1 has been almost totally deleted (Supplementary information, Fig. S3g). We performed the serial replating colony formation assay to evaluate the self-renewal of LICs. The results revealed that loss of Uhrf1 significantly decreased the number of colonies (Fig. 3b, c). To further study the long-term self-renewal ability of LICs, we conducted the cobble stone area formation colony assay (CAFC) and found that loss of Uhrf1 significantly decreased the number of cobble stone area colonies (Fig. 3d). Uhrf1 deficiency significantly decreased the number of CFU and CAFC colonies derived from the mouse MLL-AF9-expressing c-Kit+ leukemia blast cells (Fig. 3e, f). We performed flow cytometry analysis of LICs in BM cells of recipients with MLL-AF9-driven AML 4 weeks after Uhrf1 deletion by poly(I:C) administration, and found that Uhrf1 deletion significantly decreased the percentage of LICs from the recipient mice in the primary transplantation (Fig. 3g, h). Moreover, knocking down UHRF1 significantly decreased the number of colonies derived from human CD34+ BM cells of AML patients in CFU assay (Fig. 3i). These results suggest that Uhrf1 deficiency impairs the self-renewal of LICs.

Fig. 3: Loss of Uhrf1 impairs the self-renewal of LICs and decreases the frequency of LICs.

a The strategy of analysis of AE9a- and MLL-AF9-driven LICs. LSK cells represent LIC cells in AE9a-driven AML (Lin–Sca-1+c-Kit+), and L-GMP cells represent LICs in MLL-AF9-driven AML (Lin−Sca-1−c-Kit+CD34+CD16/32+ GMP-like leukemic cells). b, c The average number of colonies (b) generated from 3000 AE9a-expressing FL LSK cells or 800 MLL-AF9-expressing BM L-GMP cells with or without Uhrf1 deletion upon 4-OHT treatment in each replating (n = 3), and the morphology of the colonies (c). d Conditional deletion of Uhrf1 by 4-OHT treatment decreases the self-renewal capacity of AE9a- or MLL-AF9-driven LICs in Long-Term Culture-Initiating Cell (LTC-IC) assays. Shown is the CAFC numbers of colonies generated from 4000 AE9a-expressing FL LSK cells or 500 MLL-AF9-expressing BM L-GMP cells (n = 3). e The CAFC numbers of colonies generated from 3000 MLL-AF9Uhrf1fl/fl or MLL-AF9Uhrf1Δ/Δ leukemia blast cells from recipients in the primary transplantation (n = 3). f The morphology and numbers of colonies generated from 2000 MLL-AF9Uhrf1fl/fl or MLL-AF9Uhrf1Δ/Δ leukemia blast cells from recipients in the primary transplantation (n = 3). g, h Representative flow cytometry profiles (h) and quantification of the frequencies (g) of L-GMP cells in the BM from MLL-AF9Uhrf1fl/fl or MLL-AF9Uhrf1Δ/Δ recipients (n ≥ 5). i The average numbers of colonies generated from 10,000 primary AML CD34+ patient cells with UHRF1 knockdown by shRNA (n = 3). j The log-fraction plot shows the result of the limiting dilution assay by using different dilutions of leukemia cells from MLL-AF9Uhrf1fl/fl or MLL-AF9Uhrf1Δ/Δ mice. The “solid line” means confidence of intervals for 1/LICs of estimate, and “dotted line” means confidence of intervals for 1/LICs of lower and upper. k The MTT assays show that knocking down UHRF1 significantly inhibited the proliferation of AML cells (n = 3). Statistical analyses were performed using Student’s unpaired t-test for b, d–g, i and k. ELDA software was used for analysis in j. Data are all presented as means ± SD; *P < 0.05, **P < 0.01, ***P < 0.001.

To directly enumerate the frequency of LICs, we performed the limiting dilution assay of MLL-AF9- or AE9a-driven leukemia. The results showed that Uhrf1 deficiency decreased the frequency of LICs in MLL-AF9- or AE9a-driven AML mice (1:10,949 vs 1:1478 cells and 1:3374 vs 1:747 cells, respectively) (Fig. 3j; Supplementary information, Fig. S3h). In summary, our data indicate that Uhrf1 is essential for the self-renewal of LICs.

Knocking down UHRF1 affects the survival and cell cycle of AML cells

To define the function of UHRF1 in AML cells, we knocked down UHRF1 in AML cell lines by using shRNA against UHRF1 and found that inhibition of UHRF1 significantly inhibited the growth of these cells (Fig. 3k; Supplementary information, Fig. S4a, b). The flow cytometry and morphology analysis showed that knockdown of UHRF1 significantly induced the apoptosis of Kasumi-1 and THP-1 cells (Supplementary information, Fig. S4c‒e). After knocking down UHRF1, we found the increased BAX and decreased PARP in these cells (Supplementary information, Fig. S4f), and the expansion of AML cells were reduced in CFU assay (Supplementary information, Fig. S4g, h). The flow cytometry results indicated that AML cells were significantly arrested at G2/M phase and the Western blotting results showed that the level of cell cycle-related protein p27 was increased after UHRF1 knockdown (Supplementary information, Fig. S4f, I, j).

We further studied the role of Uhrf1 in the murine AML cells. The CFU assay analysis showed that the expansion of AE9a or MLL-AF9 cells also decreased after knocking down Uhrf1 (Supplementary information, Fig. S3i, j). Moreover, the flow cytometry analysis showed that AE9a or MLL-AF9 cells were significantly arrested at G2/M phase after knocking down Uhrf1 (Supplementary information, Fig. S3k, l). In summary, these results strongly suggest that UHRF1 is essential for the proliferation and survival of AML cells.

The genome-wide DNA binding patterns of UHRF1 in leukemia cells and transcriptome change induced by UHRF1 knockdown

To investigate how UHRF1 regulates leukemogenesis, we performed the RNA-seq analysis of AML cells with or without UHRF1 knockdown. The result showed that 500 genes and 899 genes were significantly up-regulated or down-regulated in both Kasumi-1 and THP-1 cells after UHRF1 knockdown by analysis of the overlapped differential expression genes (DEGs) (Fig. 4a; Supplementary information, Fig. S5a). Gene ontology (GO) analysis of DEGs showed that GO terms or pathways such as the myeloid cell homeostasis, hemi-methylated DNA-binding, cell growth, apoptotic signaling, cell cycle-related pathways are highly enriched in AML cells upon UHRF1 knockdown (Supplementary information, Fig. S5b). The gene set enrichment analysis (GSEA) revealed that genes sets related to the MYC targets, E2F targets, G2/M checkpoint, leukemia stem cell and p53 pathway are highly enriched (Fig. 4b, c; Supplementary information, Fig. S7o). Some genes that were dysregulated in AML cells with UHRF1 knockdown (e.g., MXD4, E2F1, E2F2, ILF2, LBR, POLD2, GINS1 and TXNIP) play vital roles in regulating these pathways (Fig. 4c). The mRNA levels of some of these genes form the enriched pathways were verified by the q-PCR analysis (Fig. 4d, e).

Fig. 4: The genome-wide DNA binding patterns of UHRF1 and low expression of UHRF1 changes the transcriptome in Kasumi-1 and THP-1 cells.

a RNA-seq was performed on Kasumi-1 and THP-1 cells with UHRF1 knockdown, and the overlap of DEGs in RNA-seq was shown. b The GSEA curves for the pathways involving MYC and E2F targets selected from the top 10 affected pathways in AML cells with UHRF1 knockdown. c The heat map analysis of the enriched pathways in RNA-seq data of AML cells with UHRF1 knockdown. d, e The expression of genes in the enriched pathways was examined by q-PCR analysis in Kasumi-1 (d) and THP-1 (e) cells with UHRF1 knockdown (n = 3). f The CUT&Tag analysis was performed with the anti-UHRF1 antibody and control IgG in AML cells, and a profile of UHRF1 binding, centering on the TSS, was shown. g The heat-maps of CUT&Tag peak signals of UHRF1 target genes in AML cells. h The distribution of UHRF1 binding sites in AML cells by the CUT&Tag analysis. i The overlap of UHRF1 target genes from the CUT&Tag analysis and DEGs from RNA-seq analysis shows 383 common genes regulated by UHRF1 in both Kasumi-1 and THP-1 cells. j The binding peak of UHRF1 was identified on MXD4 promoter based upon the CUT&Tag analysis in AML cells. k The ChIP-qPCR analysis of UHRF1 binding to the MXD4 loci in AML cells. Data are all presented as means ± SD. Statistical analyses were performed using Student’s unpaired t-test for d, e, and k. *P < 0.05, **P < 0.01, ***P < 0.001.

To define the direct targets of UHRF1, we performed the CUT&Tag analysis by using the anti-UHRF1 antibody and IgG control in AML cells. The profile and heat-map of signal peaks showed that the binding sites of UHRF1 on genes are mainly at transcription start sites (TSS) (Fig. 4f, g), distal intergenic regions, intron, promoters (Fig. 4h) and up-stream/down-stream of TSS (Supplementary information, Fig. S5c). To further identify the role of UHRF1-regulated target genes in leukemia cells, we analyzed the DEGs from RNA-seq combined with the CUT&Tag data and found that there were 383 overlapped target genes in AML cells including MXD4 that encodes the transcriptional repressor MXD4, an antagonist of MYC39,40 (Fig. 4i). Altogether, these results showed that UHRF1 deficiency led to decreased transcription of the MYC signaling pathway-related genes and increased expression of MXD4 in AML cells.

UHRF1-mediated MXD4 repression is essential for leukemogenesis

To assess functionally important target genes of UHRF1, we analyzed the CUT&Tag data and validated some binding events by ChIP-qPCR in AML cells. The CUT&Tag and ChIP-qPCR data showed that UHRF1 directly bound to MXD4 promoter (Fig. 4j, k). To determine whether the TSS enrichment of UHRF1 is DNA methylation independent, we performed ChIP assay in AML cells with DNMT1 knockdown by using an antibody against UHRF1 (Supplementary information, Fig. S5d). The results showed that the enrichment of UHRF1 on MXD4 promoter was not significantly changed in AML cells with DNMT1 knockdown compared with the control cells (Fig. 5a, b). We surveyed the overlapped dysregulated genes of RNA-seq and CUT&Tag data and found that MXD4 was significantly upregulated by UHRF1 knockdown in murine AML and LICs (Fig. 5e, f; Supplementary information, Fig. S5e, f). Given the vital role of UHRF1 in DNA methylation, we performed bisulfite sequencing of MXD4 to test the DNA methylation level of MXD4 and found that knocking down UHRF1 decreased the level of MXD4 DNA methylation at the CpG sites around the TSS in AML cells (Fig. 5c, d). We found that the expression of UHRF1 is negatively correlated with the expression of MXD4 in LSCs from AML patients by using microarray analysis (GSE76009) (Fig. 5g). These data suggest that UHRF1 inhibition-induced activation of MXD4 transcription is possibly due to the decreased DNA methylation. Thus, we identified MXD4 as a target gene of UHRF1 based on RNA-seq, CUT&Tag and bisulfite sequencing analyses.

Fig. 5: Repression of MXD4 expression by UHRF1 is essential for leukemogenesis.

a, b ChIP-qPCR analysis of the TSS enrichment of UHRF1 on MXD4 gene in Kasumi-1 (a) and THP-1 (b) cells with DNMT1 knockdown by using the anti-UHRF1 antibody. c, d The DNA methylation analysis of MXD4 by the bisulfite sequencing in Kasumi-1 (c) and THP-1 (d) cells with UHRF1 knockdown (n = 3). e The protein levels of MXD4 and E2F1 were examined by Western blotting analysis in AML cells with UHRF1 knockdown. f The expression of Mxd4 was examined by q-PCR analysis in murine AML cells with Uhrf1 knockdown (n = 3). g Correlation analysis of UHRF1/MXD4 expression in LSCs from AML patients was performed using the GSE76009 dataset. h RNA-seq analysis shows that the expression of MXD4 is lower in the relapsed AML patients (n = 77) compared with the non-relapsed AML patients (n = 146). i The event free survival of AML patients was stratified by MXD4 expression into MXD4-high (679 days, n = 222) and MXD4-low (405 days, n = 66) groups. j, k The survival of recipient mice receiving AE9a (n = 7) (j) or MLL-AF9 (n = 6) (k) cells with the simultaneous knockdown of Uhrf1 and Mxd4. l, m The flow analysis of GFP, Mac-1 and Gr-1 in BM cells of the recipient mice in AE9a (l) or MLL-AF9 (m) mouse model (n ≥ 3). n, o The number of colonies generated from AE9a (n) or MLL-AF9 (o) driven LICs with Uhrf1 deletion/Mxd4 knockdown (n = 3). Data are all presented as means ± SD. Statistical analyses were performed using Student’s unpaired t-test for a–d, f, h, l–o, and log-rank test for i–k. Pearson’s correlation analysis was used for g. *P < 0.05, **P < 0.01, ***P < 0.001.

To examine whether MXD4 could predict the clinical outcomes of AML patients, we analyzed 288 patients with AML by performing RNA-seq of the leukemic cells isolated from their BM, and the result showed that MXD4 expression is significantly lower in the relapsed patients (Fig. 5h). The Kaplan-Meier curves show that the patients with low MXD4 expression have a significantly higher chance of relapse (Fig. 5i). Therefore, these data indicate that low expression of MXD4 is associated with a high relapse rate of AML patients.

It was known that MYC binding with MAX stimulates the cell cycle progression and cell proliferation through regulation of E2F, and that MXD4 can inhibit MYC function by competing with MAX binding.39,41 We indeed found that E2F is decreased in AML cells with UHRF1 knockdown by the Western blotting analysis (Fig. 5e). To understand the role of MXD4 in leukemogenesis, we sought to knock down both Mxd4 and Uhrf1 in AE9a or MLL-AF9 cells (Supplementary information, Fig. S5g) and found that knockdown of Mxd4 significantly rescued Uhrf1 deficiency-induced apoptosis in AE9a and MLL-AF9 cells (Supplementary information, Fig. S7a, b). We also evaluated the growth potential of these cells in sublethally irradiated recipient mice. The recipients receiving the shMxd4 and shUhrf1-expressing AE9a or MLL-AF9 cells had a shorter life span than the mice who received shUhrf1-transduced cells (Fig. 5j, k). These mice also had more AE9a- or MLL-AF9-expressing cells (GFP+) in their PB two weeks after the transplantation, demonstrating that the delayed leukemogenesis of shUhrf1-expressing AE9a or MLL-AF9 cells was restored by Mxd4 knockdown in vivo (Fig. 5l, m; Supplementary information, Fig. S5h). Knocking down Mxd4 also rescued Uhrf1 deficiency-disrupted self-renewal of AE9a or MLL-AF9-expressing LICs in the CFU assay (Fig. 5n, o).

The interaction of UHRF1 and SAP30 is required for MXD4 repression and leukemogenesis

To further understand the role of UHRF1 in leukemogenesis, we examined the proteins interacting with UHRF1 in AML cells by the mass spectrometry analysis and identified that UHRF1 bound to Sin3A-associated protein (SAP30), which mediates protein‒DNA interactions and is involved in transcriptional regulation (Supplementary information, Fig. S6a, b).42 Analysis of clinical samples showed that SAP30 was highly expressed in AML patients35,43 and AML cell lines (Supplementary information, Fig. S6c, d), and the high expression of SAP30 predicts the poor event-free survival in AML (Fig. 6a). The expression of SAP30 was positively correlated with the expression of UHRF1 in LSCs from AML patients (Fig. 6b). We verified that UHRF1 interacts with SAP30 by using the Co-immunoprecipitation (Co-IP) assays in AML cells (Fig. 6c). Next, we performed the glutathione S-transferase (GST) pull-down assay and found a direct interaction between UHRF1 and SAP30 through the SRA domain (Fig. 6d; Supplementary information, Fig. S6e). To further identify the interaction site of UHRF1 and SAP30, we truncated the SRA domain into three segments, and the GST pull-down analysis showed that SRA-F2 and SRA-F3 directly interacted with SAP30 (Supplementary information, Fig. S6f, g). Then, we constructed the UHRF1 plasmid with different truncated fragments in the SRA-F2 and SRA-F3 regions (Supplementary information, Fig. S6h, i) and identified that the truncation of an 11-amino acid fragment (aa568‒aa578) in the SRA domain blocked the interaction between UHRF1 and SAP30 by the Co-IP assay (Fig. 6e; Supplementary information, Fig. S6j). Finally, we performed a screening of mutation in the aa568‒aa578 fragment and identified the point mutations G572R and F573R that specifically disrupt the UHRF1‒SAP30 interaction but not the UHRF1‒DNMT1 binding, as indicated by the Co-IP assay (Fig. 6f). The G572 and F573 residues are highly conserved in other species (Supplementary information, Fig. S6k), suggesting that this interaction may be evolutionary conserved.

Fig. 6: The interaction of UHRF1 with SAP30 is critical for MXD4 repression and leukemogenesis.

a The event free survival of AML patients was stratified by SAP30 expression into SAP30 high (survival days: 503 days, n = 211) and low (survival days: 646 days, n = 77) groups. b Correlation analysis of UHRF1/SAP30 expression in LSCs from AML patients was performed using the GSE76009 dataset. c The immunoprecipitation was performed using the anti-UHRF1 or anti-SAP30 antibody, and the anti-SAP30 and anti-UHRF1 antibodies were used for the Western blotting analysis in AML cells (n ≥ 3). d The GST pull-down assay shows that UHRF1 interacts with SAP30 in vitro and the SRA domain of UHRF1 is required for the interaction with SAP30. e The schematic representation of the truncations of SRA domain. f The immunoprecipitation assay was performed to examine the interaction of HA-tagged mutant UHRF1 with Flag-SAP30 in 293T cells (n = 3). g The number of colonies generated from UHRF1-deficient AML cells transduced with WT or mutant UHRF1 (n = 3). h The survival of B-NDG recipient mice receiving Kasumi-1 (n = 6) or THP-1 (n = 5) cells with the restoration of UHRF1 and UHRF1-Mut2 after UHRF1 knockdown. i, j MXD4 expression was examined by q-PCR (i) and Western blotting (j) analysis in AML cells with SAP30 knockdown (n = 3). k The CUT&Tag analysis shows that SAP30 binds to the promoter of MXD4 in AML cells. l The ChIP-qPCR analysis of UHRF1 on the promoter of MXD4 in AML cells with the knockdown of SAP30 (n = 3). m MXD4 expression was examined by q-PCR analysis in UHRF1-deficient AML cells transduced with WT or mutant UHRF1 (n = 3). n, o The representative DNA methylation profiles (n) and quantification (o) analysis of MXD4 by the bisulfite sequencing in G572R and F573R mutant AML cells with UHRF1 knockdown. Data are all presented as means ± SD. Statistical analyses were performed using log-rank test for a, h, and Pearson’s correlation analysis for b. Student’s unpaired t-test was used for g, i, l, m, o. *P < 0.05, **P < 0.01, ***P < 0.001.

To explore the biological relevance of UHRF1‒SAP30 interaction in leukemogenesis, we expressed UHRF1 and the UHRF1-Mut2 mutant in AML cells with UHRF1 suppression (shUHRF1 AML cells) for the CFU assay. Although UHRF1 and the Mut2 mutant exhibited comparable expression levels in the transduced AML cells, Mut2 significantly impaired the ability of UHRF1 to rescue the self-renewal activity of shUHRF1 AML cells (Fig. 6g; Supplementary information, Fig. S6l). To determine whether the UHRF1‒SAP30 interaction is required for leukemogenesis in vivo, we used the cell-derived xenograft (CDX) model based on the transplantation of shUHRF1 AML cells transduced with the wild type UHRF1 or mutated UHRF1. In this assay, the mice carrying UHRF1-Mut2 showed a significant delay in leukemogenesis compared with the control mice (Fig. 6h), indicating that the UHRF1‒SAP30 interaction is required for acute myeloid leukemogenesis.

To determine whether SAP30 affects the function of AML cells, we performed the 3-(4,5)-dimethylthiahiazo (-z-y1)-3,5-di-phenytetrazoliumromide (MTT) analysis and found that knockdown of SAP30 significantly decreased the proliferation of AML cells (Supplementary information, Fig. S6m), and the q-PCR and Western blotting analysis revealed that knocking down SAP30 significantly increased the expression of MXD4 in human and murine AML cells (Fig. 6i, j; Supplementary information, Fig. S5i). To investigate the target genes of SAP30 in AML cells, we performed the CUT&Tag analysis by using the anti-SAP30 antibody in AML cells and found that SAP30 directly bound to MXD4 promoter (Fig. 6k). The profile and heat-map of signal peaks showed that the binding sites of SAP30 on genes are mainly on the TSS (Supplementary information, Fig. S6n), distal intergenic regions, intron and promoters (Supplementary information, Fig. S6o). Meanwhile, the ChIP assay showed that UHRF1 is less enriched on MXD4 promoter after SAP30 knockdown in AML cells (Fig. 6l). The expression of UHRF1 but not UHRF1-Mut2 mutant rescued UHRF1 suppression-induced MXD4 upregulation at both mRNA and protein levels in AML cells (Fig.