An Epigenetic Mechanism Underlying Chromosome 17p Deletion–Driven Tumorigenesis

Chromosome copy-number variations (CNV), which generally consist of hundreds or thousands of genes, are a common feature of human cancers (1, 2). Accumulating evidence indicates that specific CNVs are recurrent in many types of cancer, and some can occur at the early stages of tumorigenesis, suggesting that at least some cancer-associated CNVs may function as the drivers of human cancers (3, 4). For example, chromosome 3p deletions are detected in more than 90% of lung cancers, whereas chromosome 8q is frequently amplified in hepatocellular carcinoma (5, 6). Among them, heterozygous chromosome 17p deletions (del(17p)) are the most frequent chromosome loss in both solid cancers and hematopoietic malignancies, including lymphomas and leukemias (2–4, 7–10). It has been proposed that tumor suppressor genes (TSG) are enriched in cancer-associated chromosome deletion regions, whereas oncogenes are enriched in chromosome amplicons (11, 12). However, given that a large number of genes are involved in each CNV, it has been challenging to functionally identify potential TSGs and oncogenes for better mechanistic insights on CNVs in cancer. It had long been assumed that the only function of del(17p) is for TP53 loss of heterozygosity, until our recent work (8). Using a conditional chromosome 11B3 (syntenic to human chromosome 17p13, a commonly deleted region in human cancers) deletion mouse model, we demonstrated that del(17p) could promote cancer development through both TP53-dependent and TP53-independent mechanisms. However, the additional TSGs in this region remain to be identified.

There are about 500 genes on chromosome 17p, and a large amount of effort has been made to identify new TSGs in this region (8, 13). According to the “two-hit” theory, most previous studies focused on searching for candidate TSGs with mutations/focal deletions in one allele and complete loss of the other allele through del(17p), similar to TP53 (14). Unfortunately, limited success has been achieved with this strategy (15, 16). Most 17p genes involved in del(17p) remain intact in the other allele and are expressed in 17p-deleted cancers, although generally at reduced levels. Recently, it has been reported that EIF5A and ALOX15B, both located on chromosome 17p13.1, can prevent tumorigenesis by promoting apoptosis of premalignant B cells (8, 13). Interestingly, both EIF5A and ALOX15B are haploinsufficient TSGs, and neither has a second hit on the other allele. These studies imply that potential TSGs involved in del(17p) have been largely underestimated and emphasize functional studies for further identifying TSGs within chromosome deletions.

In this study, we used our previous in vivo functional short hairpin RNA (shRNA) screening results (8) and analyzed the potential roles of plant homeodomain finger protein 23 (PHF23), located on chromosome 17p13.1, in immature B-lymphoblastic malignancy. As a putative epigenetic reader, the plant homeodomain (PHD) domain of PHF23 specifically binds to trimethylated lysine 4 of histone H3 (H3K4me3; refs. 17, 18). Interestingly, PHF23 is reported to also be involved in recurrent chromosomal translocations, forming the oncofusion gene NUP98–PHF23 in human acute myeloid leukemia (AML; refs. 17–20). NUP98–PHF23 promotes leukemogenesis through its dysregulated chromatin-modifying activity. In contrast, in del(17p) cancer, one allele of PHF23 is lost without recurrent mutations on the second allele, suggesting that its function may be lost in these cancers. It would be interesting to investigate whether and how an epigenetic reader involved in del(17p) plays a role in cancer biology.

In our unbiased in vivo tumorigenesis screening with an shRNA library targeting genes on mouse chromosome 11B3, syntenic to human chromosome 17p13, the common deletion region of chromosome 17p, multiple shRNAs targeting Phf23 were enriched in the resulting lymphomas/leukemias (ref. 8; Fig. 1A). By analyzing diffuse large B-cell lymphoma (DLBCL) cohorts from The Cancer Genome Atlas (TCGA), we found that PHF23 expression was significantly reduced in DLBCLs with del(17p) compared with their counterparts, consistent with its DNA copy-number loss (Fig. 1B). Therefore, we hypothesized that PHF23 might be a TSG on chromosome 17p.

To validate the role of PHF23 in B-cell malignancies, we cointroduced multiple independent Phf23 shRNAs (shPhf23.338, shPhf23.877, and shPhf23.971) or control Renilla shRNA (shRen) with GFP and Myc into pre–B cells isolated from C57BL/6 mouse bone marrow (BM). These shPhf23 or shRen cells were then transplanted into sublethally irradiated congenic recipient mice (ref. 21; Supplementary Fig. S1). The recipient mice transplanted with shPhf23 cells developed tumors with enlarged lymph nodes, whereas none of the recipient mice with shRen did in the observation period (Fig. 1C and D). The median tumor onset times were 40 days for shPhf23.338, 40 days for shPhf23.971, and 53 days for shPhf23.877, respectively. Flow cytometry analysis showed that all of the tumor cells were GFP positive, indicating that they originated from donor cells with shPhf23 (Fig. 1E). The expression of B-cell marker B220 and hematoxylin and eosin staining results indicated that these mouse tumors with shPhf23 displayed the characteristics of high-grade lymphoblastic lymphoma/leukemia (Fig. 1E and Supplementary Fig. S2A). The knockdown efficiencies of Phf23 shRNAs in the resulting immature B-lymphoblastic malignant cells from enlarged lymph nodes of recipients were confirmed by qPCR and Western blotting (Fig. 1F and G).

To further confirm its tumor suppressor function, we designed single-guide RNAs (sgRNA) targeting the genomic DNA sequence encoding the PHD domain of PHF23, which recognizes and binds H3K4me3 (20). Fetal liver cells (FLC) enriched with hematopoietic stem and progenitor cells from E13.5 Eμ–Myc mouse embryos (22) were transduced with Cas9 and sgRNAs for Phf23 (sgPhf23) or scramble sequence (sgScr) and then transplanted into sublethally irradiated congenic recipient mice (Supplementary Fig. S1). Whereas none of the sgScr recipients developed tumors during the observation period, six of eight sgPhf23 recipient mice developed lymphoma/leukemia with a median tumor-free survival of 102 days (Fig. 1H and Supplementary Fig. S2B). sgPhf23 generated loss-of-function mutations in the resulting lymphoma/leukemias, as shown by the T7E1 assay, Sanger sequencing of clonal lymphoma/leukemia, and Western blotting (Supplementary Fig. S2C–E).

To test whether our mouse lymphoma/leukemia model resembles human cancers with 17p deletions, we analyzed RNA-sequencing (RNA-seq) data of mouse lymphoma/leukemia cells with or without Phf23 loss. Gene set enrichment analyses (GSEA) showed that genes upregulated in shPhf23 lymphoma/leukemia cells were significantly positively enriched in 17p-deleted human DLBCLs compared with those without 17p deletions [Normalized Enrichment Score (NES) = 1.75, FDR q = 0.00], whereas the downregulated genes in mouse shPhf23 lymphoma/leukemia cells were significantly negatively enriched in human DLBCLs with 17p deletions (NES = −1.39, FDR q = 0.02; Fig. 1I and J). These correlations suggested that our mouse lymphoma/leukemia with Phf23 loss partially resembled human DLBCLs with 17p deletions, and PHF23 was a key TSG on 17p. Thus, with both shRNA and CRISPR/Cas9-mediated genome-editing experiments, we identified PHF23 as a TSG on chromosome 17p.

Chromosome 17p deletions usually occur as a heterozygous loss (3, 4, 23). TSGs on chromosome deletions presumably need loss of heterozygosity through various mechanisms, such as mutations, focal deletions, and epigenetic silencing during tumorigenesis (24). However, according to cBioPortal, mutations on PHF23 were rare in human cancers (Supplementary Fig. S2F), and the expression of PHF23 in patients with del(17p) DLBCL was significantly reduced compared with those with intact chromosome 17p (Fig. 1B). Similarly, some other genes involved in large chromosomal deletions, such as MLL3 on chromosome 7q, have been shown to be haploinsufficient TSGs (12, 25, 26). Hence, we hypothesized that PHF23 might also be a haploinsufficient TSG on chromosome 17p. Consistently, although the expression levels of Phf23 were significantly reduced, there were remarkable remaining Phf23 mRNA and protein in shPhf23 lymphoma/leukemia (Fig. 1F and G). Heterozygous mutations of Phf23 were also detected in sgPhf23 lymphoma/leukemia (Supplementary Fig. S2D).

To test this hypothesis, we generated Phf23 knockout mice with CRISPR/Cas9 editing, which resulted in a 14-bp insertion on exon 5 and a presumable premature stop codon (Fig. 2A and B). Phf23^−/− mice were embryonically lethal (E15.5–16.5; unpublished data). Western blotting showed that the expression levels of PHF23 in FLCs from Phf23^+/− embryos were precisely reduced to 50% of that in wild-type (WT), whereas they were completely lost in Phf23^−/− cells (Fig. 2C). Phf23 WT, heterozygous, and homozygous knockout FLCs were transduced with Myc cDNA, followed by transplantation into sublethally irradiated congenic recipient mice (Supplementary Fig. S1). Consistent with the aforementioned tumorigenesis experiments with shRNA and CRISPR/Cas9 (Fig. 1C and H), recipients transplanted with Phf23^−/− FLCs transduced with Myc had significantly shorter tumor onset times than those with WT FLCs with Myc overexpression (Fig. 2D). Surprisingly, Phf23^+/− FLCs also gave rise to tumors with a similar latency as Phf23^−/− cells. Interestingly, Myc-overexpressing Phf23^+/+ FLCs led to typical myeloid leukemia, indicated by expression of myeloid lineage markers CD11b and Gr1 but not B220, consistent with a previous report (27). In contrast, recipient mice with both Phf23^−/− and Phf23^+/− cells had enlarged lymph nodes, and their tumor cells were B220⁺CD11b⁻Gr1⁻ (Fig. 2E and F). Therefore, Phf23 heterozygosity, similar to Phf23 null, promoted MYC-driven lymphoma/leukemia. To rule out the possibility of loss of heterozygosity of Phf23 during Phf23^+/− lymphomagenesis, we measured the protein levels of PHF23 in the resulting lymphoma/leukemia by Western blotting. The results showed that the protein levels of PHF23 in Phf23^+/− lymphoma/leukemia remained at about half of that in Phf23^+/+ cells, indicating that the second allele of Phf23 remained intact during lymphomagenesis (Fig. 2G). Moreover, the WT Phf23 allele was confirmed to remain in the Phf23^+/− tumor cells (data not shown). Altogether, these results strongly suggested that PHF23 was a haploinsufficient TSG.

Codeleted TSGs within large chromosomal deletions have been suggested to be able to cooperate during tumorigenesis (8, 13). Here, we wondered whether Phf23 deficiency would cooperate with Trp53 loss, arguably the most important TSG on chromosome 17p, to promote lymphomagenesis. Although Trp53 knockdown promoted MYC-driven immature B-lymphoblastic malignancy, co-knockdown of both Trp53 and Phf23 by shRNA (shTrp53–shPhf23) significantly shortened the latency from 29 days to 14 days (Fig. 2H). A similar cooperative effect between Trp53 and Phf23 deficiencies was observed with Trp53^+/− pre–B cells (Fig. 2I). To explore the potential molecular mechanisms of the synergy of Phf23 and Trp53, RNA-seq analyses of shRen and shPhf23 B-lymphoma/leukemia cells were performed. GSEA showed that the genes in the p53 pathway were significantly negatively enriched in Phf23-deficient tumor cells compared with shRen cells (Fig. 2J). Hence, loss of PHF23 and TP53 through 17p deletions may cooperate to drive tumorigenesis through synergistic repressions of the p53 pathway.

PHF23 contains a PHD domain at its C-terminus, which has been shown to specifically bind H3K4me3 (20). In NUP98–PHF23 fusion–driven AML, NUP98–PHF23 binds a specific subset of H3K4me3 and regulates the expression of multiple AML-associated genes (18, 28). Here, we studied the functions of WT PHF23 in Ba/F3 cells, a mouse pro/pre–B-cell line (29), by chromatin immunoprecipitation–sequencing (ChIP-seq) assay. We found that most (>70%) PHF23 binding sites were located in gene promoters with a similar pattern as H3K4me3 peaks (Fig. 3A and Supplementary Fig. S3A). More than 90% of PHF23 binding peaks overlapped with H3K4me3 peaks (Fig. 3B). H3K4me3 is commonly enriched on active promoters (30), and consistently the average expression levels of H3K4me3 binding genes were significantly higher than those withoutbinding in Ba/F3 cells (ref. 31; Fig. 3C). Interestingly, PHF23 binding was associated with further mildly but significantly increased expression of genes [average log₂ transcripts per kilobase million (TPM) of genes with both PHF23 and H3K4me3 binding versus with only H3K4me3 binding: 3.46 versus 3.05, P_adj = 3.90e–17; those with only PHF23 binding versus without binding: 2.21 versus −3.28, P_adj = 5.70e-61; Fig. 3C]. A similar gene expression pattern was also observed in freshly isolated pre–B cells (Supplementary Fig. S3B). Thus, PHF23 colocalized with H3K4me3 and regulated gene activation.

Accordingly, there were more significantly downregulated genes than upregulated ones (153 vs. 14) in Phf23 knockdown pre–B cells compared with those in shRen cells (Fig. 3D and Supplementary Table S1). Consistent with the tumor suppression function of PHF23, upregulated genes in shPhf23 cells were enriched in pathways of cell survival and proliferation, such as negative regulation of apoptosis signaling pathway and positive regulation of ERK1 and ERK2 cascade, whereas the downregulated genes were enriched in pathways related to leukocyte differentiation and functions, suggesting potential roles of PHF23 in hematopoiesis (Supplementary Fig. S3C and D). GSEA showed that the HEMATOPOIETIC_STEM_CELL_DOWN gene set and B_CELL_VS_MYELOID_DOWN gene set were significantly negatively enriched in shPhf23 cells (NES = −1.99 and −2.34; FDR q = 0.00 and 0.00, respectively; Fig. 3E). Among the most downregulated genes, Cebpb and Hk3 have been shown to be important for B-cell development and function (refs. 32, 33; Fig. 3F). Interestingly, the genes bound by PHF23 were significantly overlapped with those by NUP98–PHF23 (ref. 18; Supplementary Fig. S3E). The overlapped genes included the well-known tumor suppressor genes Cdkn1b, Cdkn2c, Nf1, and Fbxw7. These data suggested that PHF23 was required for the expression of tumor suppressor and differentiation-related genes.

To directly test the function of Phf23 in B-cell development, we transplanted Phf23^+/+ or Phf23^−/− hematopoietic stem and progenitor cells together with recipient-type BM cells for irradiation protection into lethally irradiated recipient mice. Given that the Phf23^−/− mice were embryonically lethal, Phf23^+/+ or Phf23^−/− FLCs were used (Fig. 3G). Four months later, the donor-type BM cells were analyzed. The percentages of both hematopoietic stem and progenitor cells [Lin⁻Sca1⁺cKit⁺ (LSK)] and common lymphoid progenitor cells (CLP; Lin⁻Sca1^locKit^lo) were significantly increased in the BM of recipient mice with Phf23^−/− FLCs compared with those with Phf23^+/+ FLCs (Fig. 3H and Supplementary Fig. S4A). However, the B-progenitor cells (B220⁺CD43⁺) were significantly less in recipients with Phf23^−/− FLCs than those with Phf23^+/+ FLCs (Fig. 3I and Supplementary Fig. S4B). In more detail, the Phf23^−/− pre/pro–B cells (B220⁺CD43⁺CD24⁻) were comparable to WT, whereas the pro–B cells (B220⁺CD43⁺CD24⁺) were significantly reduced in recipients (1.2% versus 0.2%) with Phf23^−/− FLCs compared with those with WT FLCs (Fig. 3J and K and Supplementary Fig. S4C). Accordingly, both the percentages and cell numbers of Phf23^−/− B220⁺ cells in BM and spleen were significantly less than those of WT (Supplementary Fig. S4D–F). Thus, Phf23 deficiency impaired B-cell differentiation, which may contribute to the development of B malignancies.

To further understand how PHF23 regulated gene expression, we performed a co-immunoprecipitation mass spectrometry (Co-IP/MS) assay to identify potential PHF23 binding partners. Interestingly, the top hits were SIN3 transcription regulator family members A and B (SIN3A and SIN3B; Supplementary Table S2). SIN3 is a master scaffold protein in the SIN3–HDAC corepressor complex, which represses transcription and is involved in various biological and cellular processes (34, 35). The SIN3–HDAC (histone deacetylase) complex contains multiple tightly associated components, and notably all of them were co-immunoprecipitated with PHF23 in our Co-IP/MS assay, including HDAC1 and HDAC2, the core catalytic proteins of this complex, and 9 other factors: SAP130, FAM60A, RBBP7, RBBP4, ARID4B, SUDS3, SAP30L, BRMS1L, and SAP30 (Fig. 4A and Supplementary Table S2; refs. 36, 37). The interactions of PHF23 with SIN3A and HDAC1 were confirmed with Co-IP followed by Western blotting (Fig. 4B). The endogenous PHF23–SIN3–HDAC (PSH) complex was further validated by Co-IP with an antibody against PHF23 (Fig. 4C). We noticed that ING2, another H3K4me3 reader that has been reported to bind the SIN3–HDAC complex and associated with gene repression (38, 39), was not detected by PHF23 Co-IP/MS. These results suggested that PHF23 directly interacted with the SIN3–HDAC complex to form a PSH complex.

PHF23 contains multiple functional domains, including the PHD zinc finger domain at the C-terminus that can bind H3K4me3 in a sequence-specific manner and a coiled-coil domain in the middle that might mediate protein–protein interactions, whereas the N-terminus has been less characterized (ref. 20; Supplementary Fig. S5A). We tested which region of PHF23 interacted with the SIN3–HDAC complex with PHF23-truncated mutants. Interestingly, we found that neither coiled-coil nor PHD domain truncations (ΔCC and ΔPHD) affected the binding of PHF23 with SIN3A and HDAC1. However, the N-terminus truncation (ΔN) completely disrupted the interactions between PHF23 and SIN3A–HDAC1 (Fig. 4D). Thus, the N-terminus of PHF23 was essential for the PSH complex.

Histone deacetylation is one of the gene-regulation mechanisms, and here we focused on the deacetylation of H3K27ac, an active histone marker (30, 40, 41). To test the effect of the PSH complex on H3K27ac, we measured the total H3K27ac levels of Phf23-deficient cells by Western blotting. As shown in Fig. 4E, H3K27ac was significantly reduced in Phf23^+/− FLCs compared with that in Phf23^+/+ FLCs. To further understand the regulation of PHF23 on the functions of the SIN3–HDAC complex, we analyzed ChIP-seq data with HDAC1 or H3K27ac antibody (Supplementary Fig. S5B and C). Consistent with the interaction between PHF23 and the SIN3–HDAC complex, PHF23 binding sites significantly overlapped with HDAC1 binding peaks at both transcription start sites (TSS) and enhancer regions (Fig. 4F). Importantly, the binding of PHF23 significantly increased the H3K27ac levels compared with those with only HDAC1 binding at both TSSs and enhancer regions (Fig. 4G and H). Of TSSs with both HDAC1 and PHF23, 79.8% were associated with H3K27ac, whereas only 31.3% of HDAC1-bound TSSs were associated with H3K27ac. Similar correlations were observed in the enhancer regions (Supplementary Fig. S5D). These results suggested that PHF23 repressed the deacetylation activity of HDAC1 on H3K27ac.

Increased H3K27ac levels are associated with the upregulation of target genes (30, 41). To analyze the effect of the PSH complex on gene regulation, we analyzed RNA-seq data of shPhf23 tumor cells compared with those with WT Phf23. The expression levels of genes associated with both PHF23 and HDAC1 were significantly higher than those with only HDAC1 in lymphomas/leukemias (Fig. 4I and J). A similar expression pattern was also observed in pre–B cells (Supplementary Fig. S5E and F). H3K27ac binding sites were largely overlapped with those of H3K4me3, and both were associated with higher gene-expression levels of targets compared with those without these histone markers (Supplementary Fig. S5G and H). Even among the genes with both H3K4me3 and H3K27ac, those with PHF23 binding had significantly higher expression levels than those without PHF23 binding (Supplementary Fig. S5I and J). In particular, GSEA showed that the SENESE_HDAC1_TARGETS_UP gene set was significantly downregulated in shPhf23 pre–B cells (NES = −1.33, FDR q = 0.00), which was consistent with the interaction of PHF23 and the SIN3–HDAC complex (Fig. 4K). Thus, the PSH complex might be a critical regulator of gene expression in both premalignant and malignant cells.

The most enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways in PHF23 and HDAC1 cobound genes were pathways in cancer and leukemia (Fig. 4L). Because PHF23 was codeleted with TP53 in human cancers with 17p deletions (3, 8) and loss of both synergistically promoted tumorigenesis (Fig. 2H–J), we wondered if the PSH complex would directly regulate the p53 pathway. We found that 67 of the 197 p53 target genes (P = 1.72e-09) were bound by PHF23 (ref. 42; Fig. 4M). Consistent with the potential function of the PSH complex to link the two active histone markers H3K4me3 and H3K27ac, most of the PHF23 bound p53 target genes were also associated with H3K4me3 and H3K27ac (Fig. 4N and O and Supplementary Fig. S5K and L). As a result, the p53 pathway was significantly negatively enriched in shPhf23 lymphoma/leukemia cells compared with those with shRen (Fig. 2J). Among these PSH target genes were multiple well-known p53 targets, including Cdkn1a, Bax, and Mdm2. These genes were cobound by PHF23 and HDAC1 and also associated with histone markers H3K4me3 and H3K27ac in Ba/F3 cells (Fig. 4P and Supplementary Fig. S5M and N). Taken together, these biochemical and multiomics data suggested that the PSH complex might link the two active histone markers, H3K4me3 and H3K27ac, for synergistically regulating gene activation, including of p53 pathway genes.

Given the importance of the PSH complex in histone modifications and gene regulation, we wondered whether this complex plays significant roles in lymphoma/leukemia genesis and maintenance. SIN3A and its associated HDAC1 and HDAC2 have been shown to be required for normal hematopoiesis (43). We found that knockdown of Sin3a completely blocked the development of Phf23^+/−;Myc lymphoma/leukemia (Fig. 5A and B). Similarly, Sin3b deficiency also significantly extended the tumor-free survival of mice transplanted with Ph23^+/−;Myc FLCs (Fig. 5C and D). The latencies of Ph23^+/−;Myc lymphoma/leukemia were correlated with the knockdown efficiencies of shSin3b. These in vivo functional studies suggested that the SIN3–HDAC complex is essential for lymphomagenesis driven by Phf23 loss.

Furthermore, we tested the potential roles of the PSH complex in tumor maintenance. We reduced the expression of Sin3a by shRNA in shPhf23 lymphoma/leukemia cells as well as Ba/F3 cells, which worked as a control to rule out general cytotoxicity of Sin3a deficiency. The results showed that Sin3a knockdown specifically impaired lymphoma/leukemia cell growth with Phf23 loss but had a mild effect on Ba/F3 cells (Fig. 5E and Supplementary Fig. S6A). Sin3b suppression had a similar effect on shPhf23 lymphoma/leukemia cells as Sin3a deficiency but had little effect on untransformed Ba/F3 cells (Fig. 5F and Supplementary Fig. S6B). Furthermore, Hdac1 shRNAs significantly reduced the survival of Phf23-deficient lymphoma/leukemia compared with shRen (Fig. 5G). The HDAC inhibitors entinostat, mocetinostat, and chidamide specifically induced the death of shPhf23 lymphoma/leukemia cells but not control Ba/F3 cells at the given concentrations, suggesting a potential therapeutic window of these small molecules for Phf23-deficient cancers (Fig. 5H).

To explore the molecular effects of HDAC inhibitors on Phf23-deficient tumor cells, we performed RNA-seq analyses of shPhf23 lymphoma/leukemia cells treated with or without entinostat or chidamide. Interestingly, the upregulated genes in HDAC inhibitor–treated cells significantly outnumbered the downregulated genes, opposite to what was observed in shPhf23 cells (35 and 176 genes upregulated and downregulated, respectively, by shPhf23 versus 384 and 33 genes upregulated and downregulated, respectively, by HDAC inhibitors; log₂ fold change >1 or <-1, P_adj < 0.05; Supplementary Fig. S6C and D and Supplementary Tables S3–4). GSEA showed that the upregulated genes in shPhf23 cells were significantly negatively enriched in lymphoma/leukemia cells treated with either chidamide or entinostat (chidamide: NES = −1.49, FDR q = 0.00; entinostat: NES = −1.34, FDR q = 0.04), whereas the downregulated genes in shPhf23 cells were significantly positively enriched (chidamide: NES = 2.05, FDR q = 0.00; entinostat: NES = 1.89, FDR q = 0.00; Fig. 5I and J and Supplementary Fig. S6E). Thus, treatment with HDAC inhibitors reversed the effects of Phf23 loss on gene regulation.

Because the N-terminus of PHF23 was the key domain mediating the formation of the PSH complex, we introduced full-length or other PHF23 mutants into Phf23-deficient tumor cells. Here, we used Phf23^+/− lymphoma/leukemia cells instead of shPhf23 cells to avoid the interference between shRNAs and exogenous cDNAs. Full-length PHF23 specifically repressed Phf23^+/− lymphoma/leukemia but did not exhibit a significant effect on Ba/F3 cells, which further confirmed Phf23 as a bona fide TSG (Fig. 5K and Supplementary S6F). In contrast, ΔN PHF23 completely lost its tumor-suppression function, which strongly suggested that dysfunction of the PSH complex, mediated by the N-terminus of PHF23, was required for tumor maintenance. ΔPHD PHF23 had a significantly reduced repression effect on Phf23^+/− lymphoma/leukemia, indicating that the H3K4me3 binding activity was required for the tumor suppression function of PHF23, whereas ΔCC had a minimal effect (Fig. 5K). Taken together, these results indicated that deficiency of the PSH complex was critical for the maintenance of tumors with Phf23 loss.

Given the large number of genes located on chromosome 17p and the seemingly dominant effect of TP53, we wondered if the PSH complex played significant roles in del(17p) cancers. We transduced full-length, ΔN, ΔCC, and ΔPHD PHF23 into mouse immature B-lymphoblastic malignancy cells with deletion of chromosome 11B3, which highly resembled human cancers with 17p deletions (8). PHF23 restoration significantly reduced the survival of chromosome 11B3–deleted tumor cells, emphasizing that PHF23 significantly contributed to the pathology of chromosome 17p–deleted cancers (Fig. 5L). Furthermore, ΔPHD PHF23 had impaired tumor suppression function on chromosome 11B3–deleted lymphoma/leukemia, suggesting the importance of the PSH complex for the function of chromosome 17p. To further validate the functions of the PSH complex in human cancers, we transduced WT or mutant PHF23 into human lymphoma cells with or without 17p deletions. We found that WT PHF23 significantly inhibited the growth of 17p-deleted Granta519 and JeKo1 cells, whereas it had no effect on 17p-intact Ri1 and Namalwa cells (Fig. 5M–P). Furthermore, consistent with our findings in murine lymphoma/leukemia cells, ΔN PHF23 completely lost its tumor-suppression activity. These data strongly suggested that PHF23 deficiency significantly contributed to the pathology of del(17p) cancers, which gave rise to an unexpected susceptibility for treatments targeting the dysregulated PSH complex.

Taken together, our data identified PHF23 as a 17p TSG, loss of which was critical for both tumorigenesis and tumor maintenance and significantly contributed to the pathology of 17p-deleted cancers. PHF23 colocalized with the active histone marker H3K4me3 and directly bound and repressed the SIN3–HDAC complex through its N-terminus. The PSH complex provides a synergistic link between H3K4me3 and H3K27ac for gene activation, especially TSGs and differentiation genes (Fig. 6A and B).

Given the prevalence of CNVs in human cancers and their association with poor prognosis, it is important to understand the underlying mechanisms of CNVs in tumorigenesis and treatment. In this study, we identify PHF23 on chromosome 17p13.1 as a haploinsufficient TSG with multiple in vivo functional genetic strategies. Remarkably, PHF23 deficiency always happens together with chromosome 17p loss, and although there are very few independent mutations in human cancers, we wonder whether PHF23 loss contributes to tumorigenesis in the context of loss of other co-occurring 17p genes, especially TP53. We demonstrate that PHF23 can cooperate with TP53, which is generally codeleted with PHF23 in human cancers, through corepressing the expression of their common downstream target genes. We propose that, whereas haploinsufficiencies of either PHF23 or TP53 would have mild effects on the p53 pathway, the cooperation of both on the same pathway may have synergistic effects to give rise to significant repression of this pathway and thus promote tumorigenesis. This model would help to explain the significance of 17p deletions and potentially other CNVs with multiple TSGs, as a whole, during tumorigenesis. Altogether, our studies on PHF23 provide an example for identifying other potential TSGs and exploring their mechanisms involved in CNVs in human cancers.

PHF23 has been proposed to be an H3K4me3 reader (20), and our study demonstrates that PHF23 can form a PSH complex together with the SIN3–HDAC complex, an eraser of H3K27ac. PHF23 directly binds with the SIN3–HDAC complex through its N-terminus and restrains its deacetylation capacity on H3K27ac. Hence, the PSH complex provides a direct link between the two active histone markers for their synergistic regulation of gene activation. The PSH complex is distinguishable from the ING2–SIN3–HDAC complex (38, 39). ING2, also a PHD finger protein and H3K4me3 reader, directly binds and enhances the deacetylation activity of the SIN3–HDAC complex, which leads to gene repression (38). It would be interesting to dissect the potential competitions between the PSH complex and the ING2–SIN3–HDAC complex for H3K4me3 sites for gene regulation and their paradoxical potential cooperation for biological processes, including tumorigenesis.

It is intriguing that, in contrast to the haploinsufficient loss of PHF23 in human cancers with 17p deletions, PHF23 is also involved in AML through chromosome translocation to form an oncofusion protein, NUP98–PHF23 (8, 18, 19). NUP98–PHF23 binds to specific H3K4me3 sites and drives an AML program with upregulated HOX genes. Paradoxically, PHF23 loss in del(17p) cancers seems to have dysregulation of the SIN3–HDAC complex and results in decreased expression of B-cell differentiation genes and other TSGs. Therefore, distinct mechanisms underlie different PHF23 genetic alterations in AML and lymphoma. We believe that the reason might be the N-terminus of PHF23, which is critical for the formation of the PSH complex and the tumor suppression function of PHF23. The N-terminus of PHF23 is deleted in AML-associated NUP98–PHF23 (18, 19), which can also at least partially contribute to the pathology of human AML. Interestingly, the translocation of NUP98–PHF23 results in not only an oncogenic fusion protein (NUP98–PHF23) but also loss of one copy of NUP98 and PHF23. It is very interesting to test whether the disruption of one allele of PHF23 in NUP98–PHF23 AML would contribute to the disease. Our studies, together with others, illustrate the complexity of the molecular mechanisms underlying genetic abnormalities in human cancers.

Importantly, we show that, although only one of the several hundred genes involved in del(17p), PHF23 haploinsufficiency is required for the maintenance of human cancers with 17p deletions, which suggests an unexpected susceptibility for these refractory diseases. Future work testing HDAC inhibitors and other potential drugs related to the PSH complex may be beneficial for patients in clinical practice.

All mouse experiments were approved by the Institutional Animal Care and Use Committees of Sichuan University. Eμ–Myc mice were from Jackson Laboratories (22). Pre–B cells were enriched from 6- to 8-week-old C57BL/6 (cat. 219; Charles River) BM by autoMACS purification with Biotin anti-mouse/human CD45R/B220 (cat. 103204, RRID:AB_312989; BioLegend) and anti-Biotin microbeads (cat. 130–090–485, RRID:AB_244365; Miltenyi Biotec). Purified pre–B cells were cultured in RPMI 1640 medium supplemented with 20% FBS and 10 ng/mL IL7 (cat. 407-ML-200; R&D). For in vivo tumorigenesis, pre–B cells or E13.5 FLCs were transduced with retroviruses carrying shRNA or CRISPR/Cas9, respectively, and injected intravenously into sublethally irradiated (4.5 Gy) C57BL/6 recipient mice (6–8 weeks old, male). All recipient mice were randomly divided into each group before transplantation and monitored twice per week by palpation. The tumor monitoring process was done as blinded experiments. The immunophenotypes of resulting lymphomas were analyzed by flow cytometry using antibodies purchased from BioLegend. Statistical analysis of all survival data was accomplished with the log-rank test from Prism 8.

Phf23 knockout mice were generated in the Genetically Engineered Mouse Facility of State Key Laboratory of Biotherapy with CRISPR/Cas-mediated gene-editing technology as described before (44). Briefly, Phf23 sgRNA (targeting sequence: ATCATGGGCCGCCCCGCAA) and Cas9 mRNA were delivered into C57BL/6 zygotes by microinjection, followed by transplantation into the uterus of surrogate CD-1(ICR) female mice. A 14-bp insertion leading to a premature stop codon occurred at the sgRNA targeted sites. The resulting Phf23 knockout mouse was back-crossed to WT C57BL/6 mouse for more than eight generations to get rid of potential off-target mutations.

The BM cells from femurs and tibia of CD45.1 donor mice and E14.5 FLCs harvested from CD45.2 Phf23 knockout mice were mixed and transplanted into lethally irradiated CD45.1 C57BL/6 recipient mice. Four months after transplantation, recipient mice were sacrificed for analyses.

MirE-based shRNAs were cloned into retroviral constructs, including MSCV–shRNA–SV40–Myc–IRES–GFP, MSCV–shRNA–IRES–mCherry, and MSCV–shRNA–PGK–Puro–IRES–GFP, as previously reported (8, 13, 25). The sequences of shRNAs targeting Renilla or Trp53 are from published reports (8, 13, 25), and the sequences of shRNAs targeting Phf23, Sin3a, Sin3b, and Hdac1 are listed in Supplementary Table S5. sgPhf23: ATCATGGGCCGCCCCGCAA. Retrovirus packaging and infection were done as previously reported (8, 13, 25).

Ba/F3 (#HB-283), HeLa (#CCL-2), 293T (#CRL-1573), JeKo1 (#CRL-3006), and Namalwa (#CRL-1432) were from ATCC, and Granta519 (#ACC342) and Ri1 (#ACC585) were from DSMZ. The 17p status of JeKo1, Granta519, Ri1, and Namalwa has been confirmed by qPCR. All human lymphoma cell lines were cultured at 37°C with 5% CO₂ in RPMI 1640 medium supplemented with 10% v/v FBS and penicillin (100 U/mL)/streptomycin (0.1 mg/mL). Ba/F3 cells were maintained in RPMI 1640 medium supplemented with 10% v/v FBS and 2 ng/mL IL3 (cat. 403-ML-050; R&D). All cell lines were routinely tested for Mycoplasma by PCR. Experiments were performed within 4 weeks after fresh viable cells were thawed.

The 3xFLAG-tagged full-length, ΔN (2–119 aa deleted), Δcoiled-coil (225–283 aa deleted), and ΔPHD domain (331–393 aa deleted) Phf23 cDNA were cloned into retroviral construct MSCV–IRES–mCherry and transduced into Ba/F3 cells. mCherry⁺ cells were sorted out by FACS (Aria III; BD Biosciences). Then, the whole-cell lysates were extracted with cell lysis buffer (Cell Signaling Technology) supplemented with protease inhibitor cocktail (Roche) and incubated with anti-FLAG M2 mAb-conjugated agarose beads (cat. A2220, RRID:AB_10063035; Sigma-Aldrich). For endogenous PHF23 Co-IP, 10⁷ HeLa cells were lysed and incubated with IgG or PHF23 antibody. The horseradish peroxidase (HRP)–conjugated FLAG antibody was purchased from Sigma-Aldrich (cat. A8592, RRID:AB_439702); the anti-SIN3A antibody was from Thermo Fisher Scientific (cat. PA1–870, RRID:AB_2187625); the anti-HDAC1 antibody was from Cell Signaling Technology (cat. 34589, RRID:AB_2756821); the anti-PHF23 antibody, recognizing its C-terminus, was from Bethyl (cat. A302–320A, RRID:AB_1850230) for Co-IP and Western blotting; the anti-IgG antibody was from Abcam (cat. ab171870, RRID:AB_2687657); the anti-GAPDH antibody was from BioXcell (BX-008); and HRP-conjugated actin antibody was from Thermo Fisher Scientific (cat. MA5–32540, RRID:AB_2809817).

The preparation of samples for mass spectrometric analysis was conducted according to procedures reported previously (45). In brief, the immunoprecipitated protein products were prepared as described above and eluted with 0.1% trifluoroacetic acid. After being condensed and dried in the speed vacuum, the lyophilized protein mixtures were digested into peptides using sequence-grade trypsin resolved in triethylammonium bicarbonate. The tryptic peptides were further reduced by 10 mmol/L dithiothreitol at 56°C for 1 hour, alkylated with 55 mmol/L iodoacetamide in the dark at room temperature for an additional 45 minutes, and again dried in the speed vacuum. The lyophilized peptides were desalted with C18 ZipTip (Millipore). First, the ZipTips were equilibrated successively with 100% acetonitrile (ACN), 50% ACN/ddH₂O, and 0.1% fluoroacetic acid (FA). The peptides resolved in approximately 20 μL 0.1% FA were then loaded onto ZipTips and washed extensively with 0.1% FA and eventually eluted with 50% ACN/ddH₂O. The eluates were analyzed by a Q-Exactive Plus instrument available at the State Key Laboratory of Biotherapy.

All primers used for qPCR analysis are listed in Supplementary Table S6. To detect the knockdown efficiencies of shRNAs targeting Phf23/Sin3a/Sin3b/Hdac1, Ba/F3 cells were infected with MSCV–mirE–PGK–Puro–IRES–GFP retroviruses, followed by 1 μg/mL puromycin (Gibco). For detection of Phf23 expression level in eventual malignant cells, tumor cells were isolated from lymph nodes of diseased mice upon sacrifice. shTrp53;Myc lymphoma/leukemia cells were used as control. For qPCR, RNA was isolated with TRIzol, cDNA was synthesized with M-MLV (Invitrogen), SYBR Green PCR Master Mix (Applied Biosystems) was used, and qPCR was performed on a Quantstudio3 (Applied Biosystems).

Lymphoma/leukemia cell lines generated from Phf23 knockdown or 11B3 deletion tumor-bearing mice were cultured in BCM medium [45% DMEM, 45% Iscove's modified Dulbecco's medium, 10% FBS, 2 mmol/L glutamine, 50 μM β-mercaptoethanol, penicillin (100 U/mL)/streptomycin (0.1 mg/mL)] and were plated in 96-well plates with a cell density of 10,000 cells per well. The plated cells were treated with the indicated concentrations of HDAC inhibitors (Selleck) or DMSO. The number of living cells was counted by BD Accuri C6.

Tumor cells were infected with the indicated retroviruses, and the percentages of mCherry⁺ populations were measured every other day with LSRFortessa flow cytometry. The cell numbers were counted by BD Accuri C6 every other day. The relative cell viability of mCherry⁺ or GFP⁺ cells was calculated by dividing by the percentages of those with empty vectors.

Pre–B cells isolated from mouse BM were first infected with eMLM–shRen or eMLM–shPhf23, respectively, and 48 hours after infection, the GFP-positive cells were sorted out by FACS Aria III for RNA-seq. Phf23 knocked-down or control shRen lymphoma/leukemia cells were harvested from enlarged lymph nodes of recipient mice for RNA-seq. shPhf23 lymphoma/leukemia cells treated with HDAC inhibitors, chidamide (0.5 μmol/L), entinostat (0.5 μmol/L), or DMSO as a control for 12 hours were harvested for RNA-seq. RNA-seq libraries were prepared by the NEBNext Ultra RNA Library Prep Kit for Illumina, lymphoma/leukemia cells' RNA-seq and HDAC inhibitor–treated cells' RNA-seq were sequenced by the Illumina NovaSeq 6000 sequencing machine with 150-bp paired-end reads, and pre–B RNA-seq were sequenced by a BGISEQ500 sequencing machine (Beijing Genomic Institute, Shenzhen, China) with 50-bp single-end reads. The RNA-seq reads were aligned to the reference genome (GRCm38) by STAR_2.6.0a. Transcript abundance was normalized and measured by TPM. Differential gene expression was analyzed by DESeq2 (RRID:SCR_000154). Genes with an absolute fold change greater than 1 and P_adj ≤ 0.05 were counted as differentially expressed genes. The heat maps of differentially expressed genes were done by pheatmap (RRID:SCR_016418) and normalized by z score. The TPM data were used for GSEA. To identify functional categories of differentially expressed genes, Gene Oncology enrichment analysis was performed using the R package clusterProfiler (RRID:SCR_016884).

ChIP assays were performed according to a modified Micrococcal nuclease (MNase)–based ChIP-seq protocol reported previously (46). Briefly, 10 million Ba/F3 cells stably expressing FLAG-tagged Phf23 were harvested, resuspended in PBS, and cross-linked in 4% formaldehyde solution. Nuclei were washed, then digested using MNase (cat. M0247S, 2000 gel units/μL; NEB), and the samples were sonicated in the Diagenode Bioruptor (30 seconds on, 30 seconds off for 15 cycles). Incubation with the indicated antibodies was performed overnight at 4°C. FLAG M2 mAb-conjugated agarose beads and HDAC1 antibody were mentioned above, and H3K4me3 (cat. ab8580, RRID:AB_306649) and H3K27ac (cat. ab4729, RRID:AB_2118291) antibodies were from Abcam. DNA was eluted with 100 μL ChIP elution buffer, followed by purification with the Qiagen MinElute PCR Purification Kit, and sequenced by Novogene. Adaptor sequences were trimmed for all paired-end reads and mapped against the mm10 reference genome with STAR_2.6.0a. Bam files were transformed to bigwig files by Deeptools (RRID:SCR_016366) (47) and normalized by Bins Per Million (BPM) mapped reads. Integrative Genomics Viewer visualization was used for peak visualization. The heat map and Average Profile of chip peaks were performed by Deeptools. Peaks were called by macs2 2.1.1 with a q value <0.01. The DiffBind (RRID:SCR_012918) was used to annotate the peaks, and only the genes binding in both biological repeats were counted. Venn diagrams were generated with the R package of VennDiagram (RRID:SCR_002414). The overlapping genes were used as an input list to calculate enrichment score in all gene sets of KEGG (RRID:SCR_012773) pathways via clusterProfiler.

Statistical test methods, sample sizes, and P values are indicated in the corresponding figure legends. Statistical analyses were performed using GraphPad Prism (RRID:SCR_002798; GraphPad Software). The log-rank test was used to compare survival differences among groups for Kaplan–Meier tumor-free survival curve. Data were tested for statistical significance by an unpaired two-tailed t test, except the data in Figure 3H–K and Supplementary Fig. S4E and F, which were tested by an unpaired one-tailed t test. The relative protein gray scale was analyzed by ImageJ (RRID:SCR_003070; National Institutes of Health). P values for the Venn diagram overlap analysis were based on hypergeometric tests. Box plots were generated using ggpubr, with the horizontal center lines denoting the median and box edges denoting the interquartile range. P_adj were calculated by Wilcoxon signed-rank test and adjusted by holm.

The RNA-seq and ChIP-seq data sets can be accessed under GEO accession number GSE145190. The process files and code can be found in GitHub (https://github.com/CCLYLabBioinformatics/epigenetic-mechanism-of-PHF23).

No potential conflicts of interest were disclosed.

M. Chen: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. X. Chen: Data curation, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Li: Investigation. X. Pan: Investigation. Y. Gong: Investigation. J. Zheng: Investigation. J. Xu: Investigation. C. Zhao: Investigation. Q. Zhang: Investigation. S. Zhang: Investigation. L. Qi: Investigation. Z. Wang: Investigation. K. Shi: Investigation. B.-S. Ding: Resources. Z. Xue: Resources. L. Chen: Software. S. Yang: Resources. Y. Wang: Resources. T. Niu: Resources. L. Dai: Investigation. S.W. Lowe: Resources, funding acquisition. C. Chen: Conceptualization, resources, supervision, funding acquisition, writing–original draft, writing–review and editing. Y. Liu: Conceptualization, resources, supervision, funding acquisition, writing–original draft, project administration, writing–review and editing.

We thank Dr. Yuquan Wei for generous support, Dr. Junhong Han and Dr. Dan Xie for their invaluable advice and help on ChIP-seq, Dr. Yinglan Zhao, Dr. Hanshuo Yang, and Ruizhan Tong for their technical support, all the laboratory members in the Chen and Liu laboratory for their insightful advice and kind support, and the Core Facilities of West China Hospital. This work was supported by the National Key R&D Program of China (2017YFA0505600), the National Natural Science Foundation of China (81670182, 81770157, 81522003, 81570150), and the Sichuan Science and Technology Program (2020ZYD002, 2018JZ0077, 2017TJPT0005). The screening was performed at Memorial Sloan Kettering Cancer Center where it was supported by grants CA087497 and CA190261 from the National Cancer Institute and MSKCC Cancer Center support grant P30 CA008748. Scott W. Lowe is the Geoffrey Beene Chair for Cancer Biology and an investigator at the Howard Hughes Medical Institute.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.