Jmjd5, an H3K36me2 histone demethylase, modulates embryonic cell proliferation through the regulation of Cdkn1a expression Free

In eukaryotic nuclei, genomic DNA is efficiently packaged into chromatin, which is made up of numerous nucleosomes. In each nucleosome, ∼146 bp of DNA is wrapped around a histone octamer comprising two copies of four core histones: H2A, H2B, H3 and H4. Histone H1, the linker histone, helps to further compact the nucleosomal DNA into higher-order structures. In addition, the flexible N-termini that extend from the core histones (the histone tails) post-translationally receive various covalent modifications such as methylation, acetylation, phosphorylation, ubiquitylation and sumoylation. These histone modifications can alter the structure of the nucleosome and may also affect the assembly between chromatin and the nuclear proteins that recognize specific histone modifications (Mosammaparast and Shi, 2011).

Among these modifications, methylation of specific lysine residues on histone H3 (e.g. H3K4, H3K9, H3K27 and H3K36) is closely related to gene expression (Martin and Zhang, 2005). During transcriptional processes, methylation at H3K4 and H3K36 correlates with transcriptionally active genes, whereas methylation at H3K9 and H3K27 is linked to transcriptional repression. To add more complexity, each of the lysine residues has four distinct methylation states: un-, mono- (me1), di- (me2) or tri- (me3) methylated. Importantly, methylation is a reversible modification regulated by histone lysine methyltransferases (KMTs) or histone lysine demethylases (KDMs). The methylation status of histone H3 is dynamically regulated by KMTs and KDMs and is implicated in diverse biological processes including cellular proliferation, differentiation, DNA repair and recombination.

Two classes of KDMs have been identified (Klose et al., 2006). KDM1 (also called LSD1) encodes a flavin adenine dinucleotide-dependent amine oxidase and is known to remove mono- and di-methylated lysine modifications. The Jumonji C (JmjC) domain family of KDMs contains proteins that are Fe(II)- and α-ketoglutarate-dependent enzymes. Unlike KDM1, JmjC proteins can remove all three methylation states. Recent studies suggest that the dysregulation of JmjC members is associated with human diseases such as neurological disorders and cancers (Pedersen and Helin, 2010). For instance, mutations in KDM5C (JARID1C) were found in patients with X-linked mental retardation (XLMR) (Jensen et al., 2005), and PHF8 mutations were also observed in patients with XLMR and cleft lip/palate (Abidi et al., 2007; Koivisto et al., 2007; Laumonnier et al., 2005). Gene amplification of KDM4A (JMJD2C) was detected in esophageal squamous carcinomas, medulloblastoma and breast cancer (Ehrbrecht et al., 2006; Liu et al., 2009; Northcott et al., 2009; Yang et al., 2000), and elevated expression of KDM5B (PLU1) has been reported in breast, prostate and lung cancers (Hayami et al., 2010; Lu et al., 1999; Xiang et al., 2007). Furthermore, inactivating somatic mutations of KDM6A (UTX) have been reported in multiple tumor types (van Haaften et al., 2009).

Studies of mice genetically modified for several JmjC members suggest the involvement of JmjC proteins in various pathological/developmental processes. For example, deletion of Kdm3a (Jmjd1a) showed an adult-onset obesity phenotype and the downregulation of several metabolic genes (Inagaki et al., 2009; Tateishi et al., 2009). Deletion of Kdm2b (Fbxl10) showed an exencephalic phenotype and caused increased cell proliferation and cell death in neural progenitor cells, indicating an essential role for this gene during neural development (Fukuda et al., 2011).

Recently, JMJD5 (KDM8), a member of the JmjC family, was reported to be a demethylase for di-methylated lysine 36 of histone H3 (H3K36me2), regulating the expression of the CCNA1 gene in MCF7 breast cancer cells (Hsia et al., 2010). In Arabidopsis thaliana, Jmjd5 was reported to be co-regulated with evening-phased clock components and to be positively associated with clock genes expressed at dawn (Jones et al., 2010). These authors also showed that knockdown of JMJD5 in U2OS osteosarcoma cells led to fast-running circadian oscillations, suggesting its interchangeable role in both plant and human circadian systems. However, the physiological role of Jmjd5 in the context of a whole mammalian organism remains unknown.

In order to understand the function of Jmjd5 in cellular and developmental processes, we generated Jmjd5 mutant mice. Deletion of the Jmjd5 allele caused severe growth retardation, resulting in embryonic lethality at mid-gestation. Mouse embryonic fibroblasts (MEFs) derived from Jmjd5 hypomorphic embryos also showed the defect in cell proliferation. We identified increased expression of Cdkn1a as one of the mechanisms leading to cell growth retardation. ChIP analysis indicated that Jmjd5 is directly recruited to the Cdkn1a locus, resulting in the alteration of the H3K36me2 status of Cdkn1a. These results suggest that the Jmjd5 histone demethylase is involved in embryonic cell proliferation by fine-tuning the expression of Cdkn1a.

The targeting vector for Jmjd5-deficient mice was constructed using a recombineering system (Liu et al., 2003). The vector was first linearized by PmeI digestion and used for electroporation into the V6.5 clone derived from C57BL/6 and 129Sv embryonic stem (ES) cells (Eggan et al., 2001). The cells were subjected to G418 selection (150 μg/ml) and independent colonies were screened by Southern blotting. Genomic DNAs isolated from ES cell clones were digested with ClaI or EcoRV and hybridized with a 5′ probe (nucleotides 4641 to 5187 in NCBI reference sequence NC_000073.5) or a 3′ probe (nucleotides 13709 to 14028), respectively. These probes were generated by PCR using the following primers: for 5′ probe, 5′-TCCCTTACACGGCTATGGTC-3′ and 5′-AGCTGAGAGAATTCAGGGCA-3′; for 3′ probe, 5′-GGGTGAGTCAGAATAGTGTA-3′ and 5′-TCATTGGCCAAGGATTCAGA-3′.

One derivative of each ES cell line containing the floxed Jmjd5 allele was injected into the blastocoel cavity of E3.5 C57BL/6 blastocysts. Chimeric mice were generated at UNITECH (Japan) and mated with C57BL/6 females. The offspring (Jmjd5^+/neo) were screened for germline transmission by PCR. F1 mice were backcrossed with C57BL/6 mice. CAG-Flp transgenic mice (Kanki et al., 2006), Pgk2-Cre transgenic mice (Ando et al., 2000) and Cdkn1a-deficient mice (Deng et al., 1995) were used for the generation of Jmjd5^flox/flox mice, Jmjd5^Δ/Δ mice and Jmjd5^Δ/Δ; Cdkn1a^Δ/Δ mice, respectively. All animal experiments were approved by the Animal Care and Use Committee of Kanazawa University.

For genotyping adult mice, tail tissue was lysed in lysis buffer (20 mM Tris-HCl pH 8.0, 5 mM EDTA, 400 mM NaCl, 0.3% SDS, 0.2 mg/ml proteinase K) for DNA extraction. Genomic DNA was used for PCR using Ex-Taq DNA polymerase (Takara) or Southern blotting. For genotyping embryos, a part of the yolk sac was lysed and the lysate was used for direct PCR using MightyAmp DNA polymerase (Takara). The primers used for genotyping are listed in supplementary material Table S1. For Southern blotting, genomic DNA was digested with StuI or KpnI and hybridized with a Jmjd5-specific probe (nucleotides 10833 to 11233 in NCBI reference sequence NC_000073.5). Signals were visualized with the Gene Image Random-Prime Labeling and Detection System (Amersham). The probe was amplified using primers 5′-TTCTGACCTCCACATATGACACAG-3′ and 5′-CCAACCTGACTCAACCTACTCC-3′.

MEFs were established from wild-type and Jmjd5^neo/neo embryos using standard protocols (Nagy et al., 2003) and cultured in Dulbecco’s modified Eagle’s medium (DMEM) with 10% fetal bovine serum, 2 mM glutamine and penicillin/streptomycin at 37°C in 5% CO₂. For the cell growth assay, 3×10⁴ cells were plated on 12-well plates and the cell number was counted every 24 hours. For retroviral infection, 1.5×10⁵ cells were plated on 6-well plates and the cell number was counted 48 hours after infection.

For whole-mount in situ hybridization, a fragment of Jmjd5 cDNA (nucleotides 73 to 1502 in NCBI reference sequence NM_029842.5) or Cdkn1a cDNA (nucleotides 104 to 587 in NCBI reference sequence NM_007669.4) was amplified and cloned into the pSTBlue-1 vector (Novagen). For the luciferase assay, pTAL-Luc (Clontech) was used as a control vector (tk-luc). To prepare p53-responsive luciferase vector (p53BS), the complementary oligonucleotides 5′-GAGAACATGTCCCAACATGTTG-3′ and 5′-TCCAACATGTTGGGACATGTTC-3′, which correspond to the p53 binding site from the human CDKN1A locus (El-Deiry et al., 1993), were synthesized. Six annealed oligonucleotides were tandemly ligated and inserted into pTAL-Luc.

Mouse Jmjd5 cDNA was tagged with FLAG-His₆ tag and cloned into the pDON-5 Neo plasmid (Takara) to produce retroviruses. A Jmjd5 H319A mutant, which corresponds to the catalytically inactive JMJD5 mutant H321A in humans (Hsia et al., 2010), was generated using a PCR-based mutagenesis kit (Takara). The shRNA-expressing retroviruses were constructed and produced as described previously (Yoshida et al., 2011). Oligonucleotides for shRNA production are listed in supplementary material Table S2.

For the preparation of anti-Jmjd5 rabbit polyclonal antibody, histidine-tagged C-terminal Jmjd5 (D175 to Y396) proteins purified from genetically engineered Escherichia coli were used.

Total RNA was extracted with Trizol (Invitrogen) and cDNA prepared using the SuperScript VILO cDNA Synthesis Kit (Invitrogen). qRT-PCR analysis was performed with FastStart Universal SYBR Green Master (Roche) using the 7900HT Fast Real-Time PCR System (Applied Biosystems). PCR data were normalized to the expression of Gapdh. qPCR primers are listed in supplementary material Table S1.

ChIP experiments were performed as described previously (Kimura et al., 2008). The complexes that were immunoprecipitated with anti-K36me2 mouse monoclonal antibody or anti-Jmjd5 antibody were collected using Dynabeads coupled with protein A or G (Invitrogen). The precipitates were used for the detection of Cdkn1a genomic regions by qPCR. ChIP primers are listed in supplementary material Table S1.

Each Stealth RNAi for Cdkn1a (Invitrogen) was introduced into cells using Lipofectamine RNAiMAX (Invitrogen) by a reverse transfection procedure. The cells were collected for RNA extraction after 48 hours, or the cell number was counted after 72 hours. The Stealth RNAi Negative Control Kit (Invitrogen) was used for control. The siRNAs used for this study are listed in supplementary material Table S2.

MEFs were lysed in RIPA buffer [50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1% NP40, 0.1% SDS, 1 mM EDTA supplemented with 500 μM 4-(2-aminoethyl)benzenesulfonyl fluoride hydrochloride, 150 nM aprotinin, 1 μM E-64 and 1 μM leupeptin]. The lysates were separated on a SuperSep Ace 5-20% running gel (Wako). Anti-p21 (SX118, BD Biosciences), anti-p53 (FL393, Santa Cruz) and anti-Gapdh (6C5, Millipore) antibodies were used. As a positive control for the detection of endogenous p21 and p53, wild-type MEFs were treated with 0.03 μg/ml adriamycin for 24 hours.

SA-β-gal staining was performed as described (Kitajima et al., 2011). As a positive control for the detection of senescent cells, MEFs (1×10⁴ cells) were sparsely seeded on 6-well plates and cultured for 15 days. For immunofluorescence, MEFs were fixed with 4% paraformaldehyde in phosphate-buffered saline. The specimens were incubated with anti-single-stranded DNA (ssDNA) antibody (Dako) and treated with Alexa 546-conjugated anti-rabbit IgG antibody (Invitrogen). Nuclei were visualized with 4′,6-diamidino-2-phenylindole (DAPI). As a positive control for the detection of apoptotic cells, wild-type MEFs were treated with 0.06 μg/ml adriamycin for 24 hours.

WISH of mouse embryos was performed as described previously (Yamaguchi et al., 1993) with the exception that hybridization was performed at 70°C. Digoxygenin (DIG)-labeled RNA probes were synthesized using the DIG RNA Labeling Mix (Roche).

Reporter vector was introduced into wild-type MEFs or Jmjd5^neo/neo MEFs using the Fugene HD Transfection Reagent (Roche). The cells were collected after 48 hours and the luciferase activities determined using the Dual-Luciferase Reporter Assay System (Promega).

In order to investigate the in vivo function of Jmjd5, we generated conditional Jmjd5 knockout mice (Jmjd5^flox/flox) by the targeted disruption of exon IV, which encodes the start of the JmjC domain (Fig. 1A). Recombination of the Jmjd5^flox allele by Cre recombinase causes a frame-shift mutation, resulting in the lack of production for Jmjd5 protein with a JmjC domain. We mated Jmjd5^flox/flox female mice with spermatocyte-specific Cre-expressing transgenic male mice (Pgk2-Cre) (Ando et al., 2000), and mutant mice with a Jmjd5 deletion allele (Jmjd5^Δ) were established. Jmjd5^+/Δ heterozygous mice were viable and fertile, and no morphological changes were observed as compared with wild-type mice. To generate Jmjd5 null mice (Jmjd5^Δ/Δ), Jmjd5^+/Δ mice were intercrossed, and the genotype for each mouse was determined by Southern blot analysis with a Jmjd5-specific probe. As shown in Fig. 1B, this probe was able to recognize the 2.3 kb wild-type and 1.4 kb mutant fragments from StuI-digested genomic DNA as well as the 1.6 kb wild-type and 0.7 kb mutant fragments from KpnI-digested genomic DNA. However, we could not find Jmjd5^Δ/Δ homozygous mice among the offspring (supplementary material Table S3), suggesting that the Jmjd5^Δ/Δ genotype was embryonic lethal. Thus, we performed further genotyping analyses at various embryonic stages. The genotype for each embryo was determined by PCR analysis, which was able to detect a 1.5 kb band for the wild-type allele and a 0.5 kb band for the mutant allele (Fig. 1C). Jmjd5^Δ/Δ embryos could be detected from embryonic day (E) 6.5 to E11.5 (Fig. 1C and supplementary material Table S3), but were almost absorbed by E11.5 (5 of 5), suggesting that they died at ∼E11.0.

At the same time, we analyzed the temporal and spatial expression pattern of endogenous Jmjd5 during early embryogenesis. qRT-PCR analysis showed that Jmjd5 was expressed at all embryonic stages examined (Fig. 1D), and the expression of Jmjd5 was maintained at and just beyond the lethal stage (E11.0) of embryonic development. Whole-mount in situ hybridization (WISH) analysis using the antisense RNA probe for Jmjd5 indicated that endogenous Jmjd5 was ubiquitously expressed throughout the embryo at E8.5 (5 of 5; Fig. 1E, left), whereas the complementary sense RNA probe detected no signal (2 of 2; Fig. 1E, right). These results suggested an important role of Jmjd5 during embryogenesis, at least after the post-implantation stages.

Next, we mated Jmjd5^+/Δ mice and dissected the pregnant mice to observe the phenotype of Jmjd5^Δ/Δ embryos and littermates from E6.5 to E10.5 (Fig. 2). Although there was no phenotypic difference between Jmjd5^Δ/Δ embryos and control littermates at E6.5 (8 of 8; Fig. 2A,A′), the growth of Jmjd5^Δ/Δ embryos at E9.0 was obviously delayed compared with that of control embryos (8 of 8; Fig. 2B,B′). The Jmjd5^Δ/Δ embryos at E9.0 looked similar to the wild-type embryos at E8.5. At this stage, we were unable to detect any other morphological defects in the appearance of Jmjd5^Δ/Δ embryos. At E10.5, Jmjd5^Δ/Δ embryos showed severe growth retardation, including incomplete embryonic turning. We also observed typical angiogenesis defects in Jmjd5^Δ/Δ embryos, which had avascular yolk sacs (8 of 8; Fig. 2C′) and clumps of red blood cells (7 of 8; Fig. 2D′). This suggested that abnormal vascular development might lead to an insufficient circulatory system in Jmjd5^Δ/Δ embryos, resulting in the observed embryonic growth retardation. However, a slight growth retardation of Jmjd5^Δ/Δ embryos is already evident at E7.5, when endothelial precursors (angioblasts) start developing in the mouse embryo (data not shown). Thus, we concluded that Jmjd5 might be crucial for embryonic survival, especially for embryonic cell proliferation after the post-implantation stages.

To examine whether Jmjd5 could intrinsically regulate embryonic cell proliferation, we attempted to prepare Jmjd5-deficient MEFs to examine the effect of Jmjd5 on cell growth. However, it was difficult to establish Jmjd5^Δ/Δ MEFs because Jmjd5^Δ/Δ embryos did not survive to E14.5, one of the most suitable stages for the preparation of MEFs. Interestingly, we found that the Jmjd5^neo/neo homozygous mice also exhibited embryonic lethality without any excision of the Jmjd5 allele (see Fig. 1 and supplementary material Table S4). At E10.5, whereas the majority of Jmjd5^neo/neo embryos (25 of 37) showed severe growth retardation as observed for Jmjd5^Δ/Δ embryos (supplementary material Fig. S1B), the 12 remaining Jmjd5^neo/neo embryos seemed to be normal except for reduced body size (supplementary material Fig. S1C) and survived until E14.5. This result indicated that Jmjd5^neo/neo embryos displayed a hypomorphic phenotype. Furthermore, RT-PCR analysis indicated that endogenous Jmjd5 expression was significantly reduced in Jmjd5^neo/neo embryos (supplementary material Fig. S2). Since Jmjd5^flox/flox mice, in which the Pgk-Neo cassette has already been removed by mating with CAG-Flp mice, were viable and looked normal (data not shown), this suggested that the exogenous Pgk-Neo cassette from our targeting vector could be interfering with the expression of endogenous Jmjd5, resulting in the hypomorphic phenotype of Jmjd5^neo/neo mice. Thus, we used these Jmjd5^neo/neo embryos to establish Jmjd5 hypomorphic MEFs for further analyses.

Pregnant mice were dissected at 14.5 days post-coitum and MEFs were prepared from each embryo. Genotypes were determined by PCR analysis and a single band of 0.5 kb, corresponding to the Jmjd5^neo allele, was detected for Jmjd5^neo/neo MEFs (Fig. 3A). A ∼75% reduction of endogenous Jmjd5 expression was observed by qRT-PCR analysis in the prepared MEFs (Fig. 3B), and immunoblotting with anti-Jmjd5 antibody also showed that the expression of Jmjd5 was significantly lower than in wild type (Fig. 3C). Next, we compared cell growth between wild-type and Jmjd5^neo/neo MEFs (Fig. 3D). The wild-type MEFs proliferated exponentially and became confluent after 72 hours in culture (n=3). However, the growth of Jmjd5^neo/neo MEFs was significantly reduced compared with wild-type MEFs within 48 hours of passage (n=3, P<0.01), and the cell number decreased to ∼50% of that of control cells 72 hours after passage (Fig. 3D). This result suggested that Jmjd5 plays an important role in embryonic cell proliferation.

To examine whether the decreased cell number of Jmjd5^neo/neo MEFs might be due to induced premature senescence, we carried out senescence-associated β-galactosidase staining. β-galactosidase-positive cells were not detected in Jmjd5^neo/neo or wild-type MEFs (Fig. 3E, middle and top), although β-galactosidase-positive cells could clearly be detected in the control sparsely seeded MEFs (Fig. 3E, bottom). Next, we performed an immunofluorescence assay with anti-single-stranded DNA (ssDNA) antibody to examine whether the cells undergo apoptosis at a higher rate. Whereas many cells were stained with anti-ssDNA antibody in the positive controls (adriamycin-treated MEFs; Fig. 3F, bottom), we did not observe an elevated occurrence of ssDNA-positive cells in Jmjd5^neo/neo MEFs as compared with wild-type MEFs (Fig. 3F, middle and top). These results indicated that downregulation of Jmjd5 did not increase premature senescence and apoptosis, but rather decreased embryonic cell proliferation.

Next we examined whether the growth retardation of Jmjd5^neo/neo MEFs was dependent on Jmjd5 demethylase activity. The Jmjd5^neo/neo MEFs were infected with control retrovirus or with retrovirus expressing wild-type Jmjd5 or the catalytically inactive mutant H319A. The expression of Jmjd5 was increased to the same level when MEFs were infected with virus expressing wild-type or inactive Jmjd5 (supplementary material Fig. S3). MEF cell number was counted 48 hours after infection, and the cell growth defect of Jmjd5^neo/neo MEFs was significantly rescued by re-expressing wild-type Jmjd5 (Fig. 3G). No significant change was observed with the expression of the inactive H319A mutant as compared with the control. These results suggested that the demethylase activity of Jmjd5 is required for normal embryonic cell proliferation.

A recent report showed that JMJD5 activated the expression of the CCNA1 gene in MCF7 breast cancer cells by demethylating H3K36me2 on the CCNA1 locus, resulting in cell cycle progression (Hsia et al., 2010). Thus, we first examined whether loss of Jmjd5 caused the downregulation of Ccna1 in mouse embryonic cells. We extracted total RNA from Jmjd5^neo/neo MEFs or Jmjd5^Δ/Δ embryos to synthesize cDNA and performed qRT-PCR analysis using Ccna1-specific primers. Unexpectedly, there was no change in the expression of Ccna1 in Jmjd5^neo/neo MEFs or Jmjd5^Δ/Δ embryos as compared with wild type (Fig. 4A,D). This indicated that there might be a different regulatory function of Jmjd5 in mouse embryonic cells and human cancer cell lines.

Various genetic studies using knockout mice have shown that a number of cell cycle regulators are involved in normal mouse development (Ciemerych and Sicinski, 2005). Therefore, we attempted to examine whether the loss of Jmjd5 affected the expression of representative cell cycle regulators, including eleven cyclins, ten CDKs, eight CDK inhibitors and Rb family members, during mouse embryogenesis. The expression of Ccnd2, Cdk6 and Cdk7 was 1.57-fold, 1.44-fold and 1.43-fold higher, respectively, in Jmjd5^neo/neo than in wild-type MEFs (Fig. 4A,B). In addition, the expression of Cdkn1a and Cdkn2a (p16Ink4a) was 1.93-fold and 1.69-fold higher, respectively, than in wild type, (Fig. 4C). By contrast, the expression of other cyclins, CDKs, CDK inhibitors and the Rb family members was not significantly changed in Jmjd5^neo/neo MEFs (Fig. 4A-C and supplementary material Fig. S4). Next, we examined the expression of Ccnd2, Cdk6, Cdk7, Cdkn1a and Cdkn2a in wild-type embryos at the early somite stage (E8.25, n=3) and in stage-matched Jmjd5^Δ/Δ embryos (n=7) by qRT-PCR analysis. In agreement with our data from Jmjd5^neo/neo MEFs, we detected a significant increase of Cdkn1a transcripts (4.17-fold, P<0.01) in Jmjd5^Δ/Δ embryos (Fig. 4D). However, unlike the upregulation in Jmjd5^neo/neo MEFs, the expression of other candidates was not altered in Jmjd5^Δ/Δ embryos, and the expression of Ccnd2 showed a 1.52-fold reduction (P<0.01). Since Cdkn1a expression was commonly elevated in both Jmjd5^neo/neo MEFs and Jmjd5^Δ/Δ embryos, we selected Cdkn1a as a probable transcriptional target of Jmjd5.

To evaluate the spatial alteration of Cdkn1a expression in Jmjd5^Δ/Δ embryos, we also performed WISH analysis (Fig. 4E). The elevation of Cdkn1a was detected in the posterior half of the Jmjd5^Δ/Δ embryo including the primitive streak, where endogenous Cdkn1a was highly expressed in the wild-type embryo. Ectopic expression of Cdkn1a was also observed in the anterior half of the Jmjd5^Δ/Δ embryo (7 of 10). A complementary sense probe for Cdkn1a was unable to detect any signals (4 of 4). These results suggested that endogenous Jmjd5 could be responsible for the spatial expression pattern of Cdkn1a in whole embryos.

Next we examined whether knockdown of Jmjd5 expression could induce upregulation of Cdkn1a. We used retroviruses expressing two different small hairpin RNAs (shRNAs) for Jmjd5 (Jmjd5 sh#1 and sh#2). MEFs were infected with control retrovirus or retrovirus expressing each Jmjd5 shRNA. qRT-PCR indicated that Jmjd5 transcripts were significantly reduced upon infection with either Jmjd5 shRNA-expressing retrovirus, but not with the control retrovirus (Fig. 4F, left). We then examined the effect of Jmjd5 knockdown on the expression of Cdkn1a. Both Jmjd5-specific shRNAs led to a significant increase in Cdkn1a transcripts (1.59-fold and 1.65-fold; Fig. 4F, right). These observations suggested that Jmjd5 might be directly involved in the regulation of Cdkn1a expression in MEFs.

According to the results shown in Figs 3 and 4, we speculated that Cdkn1a upregulation caused by the reduction in Jmjd5 levels might underlie the observed growth retardation. To test our hypothesis, we performed a knockdown assay using Cdkn1a-specific siRNAs and investigated whether the reduction of Cdkn1a could rescue the inhibition of cell proliferation in Jmjd5^neo/neo MEFs. Two Cdkn1a-specific siRNAs (siRNA#1 and #2) were designed, and the knockdown efficiency of each siRNA was shown to be ∼80% (Fig. 5A). Cdkn1a siRNA#1, #2 or an unrelated control siRNA was transfected into Jmjd5^neo/neo MEFs and their growth compared. Knockdown of Cdkn1a in Jmjd5^neo/neo MEFs caused a significant increase in cell proliferation as compared with that of Jmjd5^neo/neo MEFs treated with the control siRNA or untreated Jmjd5^neo/neo MEFs (n=3, P<0.01; Fig. 5B). The cell number of Jmjd5^neo/neo MEFs treated with the Cdkn1a-specific siRNAs was still lower than that of wild-type MEFs, indicating a partial rescue of cell growth by Cdkn1a knockdown. These results suggested that the regulation of Cdkn1a expression by Jmjd5 plays an important, but not exclusive, role in embryonic cell proliferation, although incomplete knockdown of Cdkn1a might also be responsible for the partial rescue phenotype.

Since it is known that Cdkn1a is a major direct target for the p53 (Trp53) pathway (Wood and Shilatifard, 2006), we examined whether the elevation of Cdkn1a was dependent on the induction of endogenous p53 transcripts in Jmjd5-deficient cells. qRT-PCR showed that there was no difference in the expression of p53 in Jmjd5^neo/neo MEFs and Jmjd5^Δ/Δ embryos as compared with control cells (Fig. 5C). Immunoblotting also indicated that there was no alteration in the level of p53 protein between wild-type and Jmjd5^neo/neo MEFs, despite the enrichment of Cdkn1a (p21) protein in Jmjd5^neo/neo MEFs (Fig. 5D). We next measured the transcriptional activity of p53 in wild-type and Jmjd5^neo/neo MEFs by luciferase assay. We constructed a p53BS-tk-luciferase reporter (p53BS) containing p53 response elements derived from human CDKN1A (El-Deiry et al., 1993), and transfected the reporter or control plasmid (tk-luc) together with a Renilla luciferase reference plasmid into wild-type or Jmjd5^neo/neo MEFs. There was no significant difference in the luciferase activity (n=4; Fig. 5E), indicating that an alteration in p53 transcriptional activity did not occur in Jmjd5^neo/neo MEFs.

We then investigated whether the growth retardation caused by the loss of Jmjd5 was closely associated with the dysregulation of Cdkn1a in these embryos. Jmjd5-deficient mice were intercrossed with Cdkn1a-deficient mice (Cdkn1a^Δ/Δ), which were morphologically normal at E10.5 (5 of 5; Fig. 6B) (Deng et al., 1995), to generate Jmjd5 and Cdkn1a double-knockout mice (Jmjd5^Δ/Δ; Cdkn1a^Δ/Δ). Phenotypes were compared between the double-knockout and Jmjd5-deficient mice. As with Jmjd5^Δ/Δ mice, the Jmjd5^Δ/Δ; Cdkn1a^Δ/Δ genotype was embryonic lethal (data not shown) and the mice showed the severe growth retardation (4 of 4; Fig. 6D) as compared with wild-type embryos (Fig. 6A). However, Jmjd5^Δ/Δ; Cdkn1a^Δ/Δ embryos were slightly larger than Jmjd5^Δ/Δ littermates (Fig. 6C). Other abnormal phenotypes, such as the incomplete turning and the angiogenesis defect, did not seem to be rescued in the double-knockout embryos. These results suggested that the deficiency of Cdkn1a could partially rescue the phenotype of Jmjd5^Δ/Δ embryos, similar to our result from the knockdown experiment (Fig. 5B). We therefore demonstrated that a genetic interaction between Jmjd5 and Cdkn1a is at least important for embryonic cell proliferation, but also propose that other factors regulated by Jmjd5 contribute to normal embryonic cell growth.

Previous studies suggested that Jmjd5 is an H3K36me2 histone demethylase (Hsia et al., 2010) and that methylated K36 in histone H3 is an active transcriptional mark (Li et al., 2007). Thus, we assumed that the loss of Jmjd5 caused an elevation of H3K36me2 at the Cdkn1a locus, resulting in the activation of Cdkn1a in Jmjd5-deficient cells. To test our hypothesis, we designed primers to amplify several genomic regions of Cdkn1a for chromatin immunoprecipitation (ChIP) analysis (Fig. 7A). The crosslinked chromatin from wild-type or Jmjd5^neo/neo MEFs was immunoprecipitated with anti-K36me2 or anti-Jmjd5 antibody and the precipitates were used for the detection of Cdkn1a genomic regions by qPCR. Enrichment of H3K36me2 was observed in intron II (regions a and b) and exon III (region c containing the open reading frames of Cdkn1a) but not in intron III (region d) of Cdkn1a (n>9, P<0.01) in Jmjd5^neo/neo MEFs (Fig. 7B). The recruitment of Jmjd5 to the Cdkn1a locus was significantly reduced in regions a, b and c, whereas no significant change was observed in region d (n>9, P<0.01; Fig. 7C). These results suggested that reduced recruitment of Jmjd5 to several regions of the Cdkn1a locus induced increased H3K36me2. Based on the previous report showing a correlation between enriched H3K36me2 and the transcriptionally active phase (Li et al., 2007), one may propose that the increased H3K36me2 induces Cdkn1a expression in Jmjd5^neo/neo MEFs.

We also performed ChIP analysis with anti-H3K4me3 antibody to investigate whether another epigenetic mark, H3K4me3, was altered in regions a, b and c of Cdkn1a. Enrichment of H3K4me3 was observed only in region a, whereas significant changes were not detected in the other regions (supplementary material Fig. S5). Since H3K4me3 is an active transcriptional mark (Martin and Zhang, 2005), this alteration of H3K4me3 in a Jmjd5-independent manner might reflect the enhanced transcriptional activation of Cdkn1a in Jmjd5^neo/neo MEFs. Taken together, our data demonstrate that Cdkn1a is one of the downstream targets of Jmjd5 histone demethylase and that the maintenance of H3K36me2 at the Cdkn1a locus, as regulated by Jmjd5, might be required for normal embryonic proliferation.

In this study, we generated Jmjd5-deficient mice and observed severe growth retardation in Jmjd5^Δ/Δ embryos, leading to embryonic lethality at ∼E11.0. In Jmjd5^Δ/Δ embryos, significant upregulation of Cdkn1a was detected. Similarly, growth retardation as well as elevated expression of Cdkn1a were observed in Jmjd5^neo/neo MEFs. ChIP analysis indicated that, in Jmjd5^neo/neo MEFs, H3K36me2 was significantly increased in several regions of the Cdkn1a locus where the recruitment of endogenous Jmjd5 was reduced. Thus, we demonstrate for the first time that Jmjd5 moderates the expression of Cdkn1a during embryogenesis through histone demethylase activity.

Currently, there are more than 30 known mammalian JmjC domain-encoding genes, and some have been reported to be involved in cell proliferation by regulating the expression of cell cycle inhibitors. Our recent study demonstrated that Kdm6a directly demethylates H3K27me3 in Rb1 and Rbl2, increasing their expression levels and resulting in the growth repression of MEFs (Terashima et al., 2010). KDM6B (JMJD3), an H3K27 demethylase, activated the INK4A-ARF locus in response to oncogenic stress, and its overexpression led to the inhibition of cell proliferation (Agger et al., 2009; Barradas et al., 2009). Furthermore, Kdm2b, an H3K36 demethylase, regulated cell proliferation through the modulation of Cdkn2b (He et al., 2008). These results strongly suggest that the JmjC family plays an essential role in cell growth signaling. Our analysis with Jmjd5-deficient mice indicated a novel association of Jmjd5 and Cdkn1a expression that is crucial for cell proliferation during normal mouse development.

Genome-wide profiling of methylated histone H3 from yeast to humans showed that H3K36 methylation tended to be enriched across the transcribed regions of active genes rather than promoter regions (Li et al., 2007). In Saccharomyces cerevisiae, methylation of histone H3 by Set2, an H3K36 KMT, is associated with transcriptional elongation by RNA polymerase II (Krogan et al., 2003; Schaft et al., 2003). However, there is still little evidence for the physiological function of H3K36me2 in mammals. Recently, knockdown of Kdm2b in MEFs was shown to induce the activation of Cdkn2b by increased H3K36me2 at the Cdkn2b locus (He et al., 2008). Our study demonstrated that the disruption of Jmjd5 in mouse cells induced Cdkn1a transcription and the elevation of H3K36me2 at the transcribed regions of Cdkn1a, supporting the current model linking transcriptional activation and methylated H3K36.

We observed the induced expression of Cdkn1a in Jmjd5^Δ/Δ embryos and Jmjd5^neo/neo MEFs. Although Cdkn1a is a p53 target gene, we did not detect any significant changes in p53 expression at either the transcriptional or translational level. The reporter assay showed that the transcriptional activity of p53 was not changed in Jmjd5^neo/neo MEFs, and immunofluorescence and immunoprecipitation studies indicated no alteration in the subcellular localization of p53 and no direct interaction between Jmjd5 and p53 (data not shown). These results suggested that the aberrant induction of Cdkn1a seemed to occur without p53 induction. However, we cannot exclude the possibility that p53 signaling is altered, as the p53 protein receives various post-translational modifications, such as phosphorylation, methylation, acetylation and ubiquitylation, which affect its stability, specificity for target genes and affinity for co-factors (Carter and Vousden, 2009). Therefore, more detailed analyses would be necessary to prove whether there is a connection between Jmjd5 and p53.

In all of the Jmjd5^Δ/Δ embryos examined, severe growth retardation was observed, and the expression of Cdkn1a was more than 4-fold higher than in wild-type embryos. However, growth retardation is sometimes caused indirectly by abnormal placental development in genetically modified mice (Han and Carter, 2001). In fact, endogenous Jmjd5 expression was detected in the ectoplacental cone, a precursor of the placenta, at E8.5. The allantois needs to fuse with the chorion for normal placental development, but this was not observed in some Jmjd5^Δ/Δ embryos at E10.5 (3 of 8; data not shown). Thus, we could not exclude the possibility that abnormal placental development might be a reason for the growth defect in Jmjd5^Δ/Δ embryos. However, it was notable that high expression of endogenous Jmjd5 was observed in not only the ectoplacental cone but also the entire embryonic fraction. We also demonstrated that although embryonic cells purified from Jmjd5^neo/neo embryos are not under placental control, these cells still showed the defect in proliferation and the elevated expression of Cdkn1a. Furthermore, knockdown of endogenous Cdkn1a was able to rescue the growth delay significantly in Jmjd5^neo/neo MEFs. The growth retardation in Jmjd5^Δ/Δ embryos was also partially rescued in the Cdkn1a^Δ/Δ genetic background. These results indicated that the regulation of Cdkn1a by Jmjd5 played an important role in cell proliferation and embryonic development. Since the rescue effect was partial, this suggests that additional factors regulated by Jmjd5 contribute to embryonic development. We are now trying to identify other target genes that are regulated by Jmjd5 by microarray and ChIP-sequence, and hope to reveal the detailed mechanism of Jmjd5 function in embryonic cell proliferation in the near future.

Similar to the phenotype observed in Jmjd5^Δ/Δ embryos, the upregulation of endogenous Cdkn1a has been reported in several mutant mice, resulting in severe growth retardation (Ruland et al., 2001; Rantakari et al., 2010; Suzuki et al., 1997). Notably, the elevation of Cdkn1a in Brca1- or Mdm4-deficient mice has been observed (Hakem et al., 1996; Migliorini et al., 2002), and, importantly, intercrossing with Cdkn1a-deficient mice partially rescued the phenotype, suggesting that the aberrant expression of Cdkn1a is a major, but not exclusive, reason for the embryonic death (Hakem et al., 1997; Steinman et al., 2004). Interestingly, the phenotype of histone deacetylase 1 (Hdac1)-deficient mice was similar, although more severe, than with Jmjd5^Δ/Δ mice. Disruption of Hdac1 resulted in lethality before E10.5 due to severe growth retardation, and the expression of Cdkn1a and Cdkn1b was upregulated (Lagger et al., 2002). This result suggested that several epigenetic controls, including histone methylation and acetylation, are closely associated with the normal expression of Cdkn1a during embryogenesis. Therefore, clarification of the relationship between Jmjd5 and other histone-modifying enzymes would be helpful for understanding the detailed mechanism of Cdkn1a regulation by Jmjd5.

We thank Dr N. Nozaki (Hokkaido University, Japan) for providing anti-di-methylated H3K36 antibody; Drs J. Takeda (Osaka University, Japan), N. A. Jenkins and N. G. Copeland (IMCB, Singapore) for providing gene targeting vectors; Drs Y. Nakanishi and K. Yoshioka (Kanazawa University, Japan) for providing Pgk2-Cre transgenic mice; Dr S. Itohara (RIKEN, Japan) for CAG-Flp transgenic mice; and Dr A. Hirao (Kanazawa University, Japan) for Cdkn1a-deficient mice.