H3K36 trimethylation mediated by SETD2 regulates the fate of bone marrow mesenchymal stem cells

During the aging process, bone marrow mesenchymal stem cells (BMSCs) exhibit declined osteogenesis accompanied by excess adipogenesis, which will lead to osteoporosis. Here, we report that the H3 lysine 36 trimethylation (H3K36me3), catalyzed by histone methyltransferase SET-domain-containing 2 (SETD2), regulates lineage commitment of BMSCs. Deletion of Setd2 in mouse bone marrow mesenchymal stem cells (mBMSCs), through conditional Cre expression driven by Prx1 promoter, resulted in bone loss and marrow adiposity. Loss of Setd2 in BMSCs in vitro facilitated differentiation propensity to adipocytes rather than to osteoblasts. Through conjoint analysis of RNA sequencing (RNA-seq) and chromatin immunoprecipitation sequencing (ChIP-seq) data, we identified a SETD2 functional target gene, Lbp, on which H3K36me3 was enriched, and its expression was affected by Setd2 deficiency. Furthermore, overexpression of lipopolysaccharide-binding protein (LBP) could partially rescue the lack of osteogenesis and enhanced adipogenesis resulting from the absence of Setd2 in BMSCs. Further mechanistic studies demonstrated that the trimethylation level of H3K36 could regulate Lbp transcriptional initiation and elongation. These findings suggest that H3K36me3 mediated by SETD2 could regulate the cell fate of mesenchymal stem cells (MSCs) in vitro and in vivo, indicating that the regulation of H3K36me3 level by targeting SETD2 and/or the administration of downstream LBP may represent a potential therapeutic way for new treatment in metabolic bone diseases, such as osteoporosis.


Introduction
Bone marrow mesenchymal stem cells (BMSCs) are defined as a subset of PDGFRα + Sca1 + CD45 − TER119 − cells from adult mouse bone marrow in vivo [1]. BMSCs are multipotent progenitors that can differentiate into several distinct cell types, including osteoblasts and adipocytes. During the aging process, BMSCs exhibit an age-dependent decline in osteogenesis accompanied by an elevated propensity toward adipogenesis [2][3][4]. This switch decreases the number of osteoblasts and increases the number of adipocytes, contributing to osteoporosis [5,6]. Over the past decades, significant progress has been made in understanding the transcriptional control of BMSC lineage commitment, especially the lineage switch between adipocytes and osteoblasts. However, the epigenetic regulation of this process remains to be further investigated.
Histone modifications represent a large proportion of epigenetic regulations-including histone methylation, acetylation, phosphorylation, ubiquitination, sumoylation, etc.-on different histone residues [7]. The histone methylation state has been revealed as an important modulator of BMSC differentiation into osteogenic lineage or adipogenic lineage. It has been reported that repressive histone modification H3K27 trimethylation regulated by enhancer of zeste homolog 2 (EZH2) and KDM6A/B facilitate bone marrow adipogenesis in vitro [8][9][10]. H3K9 trimethylation has also been reported to promote adipogenesis and repress osteogenesis in human mesenchymal stem cells (hMSCs), owing to the KDM4B depletion [10]. Most of these reports were done in cultured human or mouse BMSCs (mBMSCs), and little is known about the in vivo effects of the histone modifications on the fate determination of BMSCs.
SET-domain-containing 2 (SETD2) is a methyltransferase responsible for H3 lysine 36 trimethylation (H3K36me3) [11]. H3K36me3 is positively correlated with actively transcribed genes and has been shown to be more enriched on transcriptionally activated genes [12,13]. H3K36me3 may be involved in some specific process such as DNA repair or transcriptional initiation and elongation, and it may contribute to the composition of heterochromatin in combination with other histone modifications [12][13][14][15][16]. Physiologically, SETD2 plays an important role in tumor suppression, aging delay, and antiviral immunity [17,18]. However, the function of H3K36me3 and SETD2 in BMSC cell fate determination remains unclear. Recently, H3K36M mutation has been suggested to be involved in chondroblastoma, indicating that the modification of H3K36 may play a role in cell differentiation [19]. It has been reported that knockdown of SETD2 by short hairpin RNA (shRNA) in hMSCs could impair osteoblast differentiation in vitro as well [20].
Here, we demonstrate that H3K36me3 mediated by SETD2 plays an essential role in MSC lineage commitment. We genetically deleted SETD2 in BMSCs by crossing Setd2 fl/fl mice with Prx1-cre mice [21] to explore the role of H3K36me3 in MSC fate in vivo and found that Setd2deficient mice (Prx1-Cre, Setd2 fl/fl mice) exhibited reduced bone formation and increased marrow fat accumulation. Notably, Lbp was identified as one SETD2 functional downstream genes. Overexpression of lipopolysaccharide-binding protein (LBP) could partially rescue the phenotype of SETD2 deficiency in BMSCs. Further mechanistic studies showed that H3K36me3 could regulate Lbp transcription level by modulating its transcriptional initiation and elongation. Thus, our study provides a new mechanism for MSC fate determination and therapeutic target for osteoporosis.

Results
H3K36me3 level changes accordingly with the SETD2 level in adipogenesis and osteogenesis of BMSCs SETD2 has been reported as a major methyltransferase responsible for H3K36me3 [12]. In order to examine the role of H3K36me3 on mesenchymal cell fate in vivo, we firstly examined the SETD2 expression level and H3K36me3 level during BMSC differentiation to adipocytes and/or osteoblasts. Both SETD2 mRNA and protein levels are decreased during the course of adipocyte differentiation (Fig 1A and 1B). Meanwhile, the H3K36me3 level-but not H3K36me1 and H3K36me2 levels-is also decreased during the adipocyte differentiation ( Fig  1A). On the contrary, mRNA and protein levels of SETD2 are elevated during the course of osteoblast differentiation, accompanied by the increase of H3K36me3 level and stable levels of H3K36me1 and H3K36me2 (Fig 1C and 1D). The dynamic alternations of SETD2 protein level and H3K36me3 during BMSC adipogenesis and osteogenesis indicate that H3K36me3 might serve as an MSC fate determinant.
To further examine the role of H3K36me3 in MSC lineage commitment in vivo, we generated a Setd2 conditional knockout mice line. In Setd2-targeted allele, 2 LoxP sites were located at intron 5 and intron 7, and a neomycin resistance gene flanked with Frt sites was inserted into intron 7 and located between exon 7 and LoxP site (S1A Fig). The DNA of targeted embryonic stem cells (ESCs) was fragmented by Pst I restriction enzyme. For the correctly targeted allele, Southern blot using 3 0 -probe would detect an extra 5.5 k band as Neomycin gene carries a Pst I restriction enzyme site besides 2 other Pst I sites in intron 4 and 8 (S1A and S1B Fig). The identified ES cells were injected into 4-to 8-blastocyst embryos to get chimeric mice and then F0 targeted germ line mice. A PCR strategy was set up for genotyping the Setd2 fl/fl mice, in which the Neomycin resistance gene was removed by breeding with Flp recombinase mice (S1C Fig). To examine the MSC fate, we crossed Setd2 fl/fl mice with Prx1-Cre mice, which express Cre recombinase in the MSCs that form the limbs and parts of the skull, but not the spine or other organs [21]. Quantitative PCR (qPCR) using the primers targeting exon 6 revealed a reduction of SETD2 in BMSCs but not in liver (S1D Fig). Western blot confirmed the deletion of SETD2 in BMSCs but not in liver (Fig 1E and S1E Fig). More importantly, a reduction of H3K36me3 but not H3K36me1/2 was observed in SETD2-deleted BMSCs ( Fig  1E). We also did immunohistochemistry assay and observed a complete loss of SETD2 in Prx1-Cre-positive cells in the tibia growth plate of Prx1-Cre; Setd2 fl/fl mice (Fig 1F, upper  panel), accompanied with the decrease of H3K36me3 level (Fig 1F, lower panel). And the H3K36me1/2 did not change significantly (S1G Fig). The expression pattern of Prx1-Cre in the bone was further confirmed by breeding Prx1-Cre mice to the tdTomato reporter mice (a stop cassette flanked with LoxP sites before tdTomado gene, hereafter called Ai9 mice) in the femur [22]. As shown in S1F in S1 Fig, BMSCs and their progeny-including chondrocytes, osteoblasts, and adipocytes-were positive for tdTomato fluorescence in Prx1-Cre, Ai9/+ mice, verifying the specific expression pattern of Prx1-Cre.

SETD2 loss of function in BMSCs enhances adipogenic differentiation and impairs osteogeneic differentiation
SETD2 deficiency leads to a decrease of H3K36me3 level, providing a unique tool to examine the role of H3K36me3 on BMSC fate. We examined the effects of SETD2 depletion on the adipocyte differentiation of BMSCs. BMSCs from Prx1-Cre, Setd2 fl/fl and wild-type (WT) mice were cultured in adipogenic medium for 6 days, and the lipid droplets were determined by Oil-red O staining. Then, we found that adipogenic differentiation was increased in Setd2 knockout cells, as demonstrated by increased Oil-red O staining compared with cells from WT mice (Fig 2A and 2B), as well as the enhanced expression of characteristic adipocyte (D) Alp activity quantification was measured by phosphatase substrate assay. Data represent mean ± SD, n = 4. (E) qPCR analysis of Setd2, Runx2, Osx, Col1α1, Alp, and Bsp expression after osteoblast differentiation for 7 days; cells were from WT and Prx1-Cre, Setd2 fl/fl mice. Data represent mean ± SD, n = 4. Data used in the generation of this figure can be found in S1 Data. Alp, alkaline phosphatase; BMSC, bone marrow mesenchymal stem cell; Bsp, bone sialoprotein; Cebpα, CCAAT enhancer binding protein α; Col1α1, Collagen type 1 alpha 1; Fabp4, fatty acid binding protein 4; Osx, Sp7 transcription factor; Pparγ1/2, peroxisome proliferative activated receptor γ1/2; qPCR, quantitative PCR; Runx2, runt-related transcription factor 2; SETD2, SET-domain-containing 2; WT, wild-type. markers-including peroxisome proliferative activated receptor γ1/2 (Pparγ1/2), CCAAT/ enhancer binding protein α (C/EBPα), fatty acid binding protein 4 (Fabp4), and Perilipin in adipocytes differentiated from Prx1-Cre, Setd2 fl/fl BMSCs compared to adipocytes differentiated from WT BMSCs (Fig 2C).
We next examined the role of SETD2 depletion in the osteoblast differentiation. Cells from Prx1-Cre, Setd2 fl/fl , and WT mice were cultured in osteogenic medium for 1 week and 3 weeks as described in the Materials and methods section. Osteogenic differentiation was decreased in Setd2 knockout cells as demonstrated by decreased alkaline phosphatase (ALP) activity ( Fig  2D, upper panel and 2E: after 1-week culture) and decreased bone matrix formation determined by Alizarin red S staining (Fig 2E, lower panel: after 3-week culture). The declined osteogenesis in Setd2 knockout cells was proved again by the decreased expression of a series of osteogenic marker genes (Fig 2F), including runt-related transcription factor 2 (Runx2), Sp7 transcription factor (Osx), Collagen type 1 alpha 1 (Col1α1), alkaline phosphatase (Alp), and bone sialoprotein (Bsp). Taken together, our data suggested that SETD2 regulates MSC fate by enhancing osteoblast differentiation and repressing adipocyte differentiation in vitro.

SETD2-deficient mice show increased bone marrow adipocytes and decreased bone mass
We further sought to examine the impact of altered BMSC differentiation on the intact animals. Prx1-Cre, Setd2 fl/fl mice were viable and fertile with a mild decreased body weight in both male and female mice (Fig 3A and 3B). However, the body fat store showed no significant difference between Prx1-Cre, Setd2 fl/fl mice and WT mice (Fig 3C and 3D). Consistent with increased adipogenic differentiation of Setd2-depleted BMSCs, 5-week-old Prx1-Cre, Setd2 fl/fl mice displayed a significant adipocyte accumulation in bone marrow examined by Oil-red O staining ( Fig 3E). Those increased bone marrow adipocytes were further accumulated in 12-week-old Prx1-Cre, Setd2 fl/fl mice ( Fig 3F). We next asked whether the accumulated adipocytes were from Prx1-Cre-positive cells. We bred Prx1-Cre, Setd2 fl/fl with Ai9 mice and observed that adipocytes, labeled by lipid drop dye, bodipy 493/503, were colocalized with tdTomato fluorescence in both Prx1-Cre, Ai9/+ mice and Prx1-Cre, Ai9/+, Setd2 fl/fl mice ( Fig  3G), demonstrating that those accumulated bone marrow adipocytes in Prx1-Cre, Setd2 fl/fl mice were derived from Prx1-Cre-expressing cells in vivo. Consistently, in vitro adipogenic differentiation assay using BMSCs obtained from Prx1-Cre, Ai9/+ mice and Prx1-Cre, Ai9/+, Setd2 fl/fl mice also proved that Prx1-cre-positive cells could turn into bodipy 493/503 positive adipocytes, and the lack of Setd2 in BMSCs led to more adipocytes (Fig 3H and 3I).
We next examined the bone mass of Prx1-Cre, Setd2 fl/fl mice to test the hypothesis that Prx1-Cre, Setd2 fl/fl mice have decreased osteogenesis. We did micro-quantitative computed tomography (μ-QCT) analysis for the femurs isolated from Prx1-Cre, SetD2 fl/fl mice and WT control mice. The distal femurs of 5-week-old Prx1-Cre, Setd2 fl/fl mice exhibited reduced bone mass (Fig 4A), as evidenced further by significantly decreased trabecular bone volume (BV/ TV) (Fig 4B), trabecular number (Tb. N) (Fig 4C), trabecular thickness (Tb. Th) (Fig 4D), and bone mineral density (BMD) (Fig 4E), accompanied by increased trabecular space (Tb. Sp) ( Fig 4F). In addition to the trabecular phenotype, Prx1-Cre, Setd2 fl/fl mice also exhibit mild decreased thickness of cortical bone (Fig 4G and 4H). Consistent with μ-QCT analysis, immunofluorescence staining of distal femur of 5-week-old mice displayed decreased expression of osteoblast marker osteopontin (Opn) and Col1α1 in Prx1-Cre, Setd2 fl/fl mice (Fig 4I and 4J). Von-kossa staining showed decreased bone mineral deposition in Prx1-Cre, Setd2 fl/fl mice ( Fig  4K). The mineral apposition rate (MAR) was also decreased in Prx1-Cre, Setd2 fl/fl mice compared with control WT littermates by fluorescent double labeling of the mineralizing front with Calcein and Alizarin red S (Fig 4L and 4M). Together, these results show that Prx1-Cre, Setd2 fl/fl mice have a phenotype consistent with enhanced adipogenesis and impaired osteogenesis in vitro, suggesting that SETD2 could regulate the fate of MSCs in vivo.

SETD2 deficiency affects genome-wide gene expression and H3K36me3 level
We next sought to investigate the molecular mechanisms by which SETD2 regulates BMSC fate. To harvest enough cultured cells and directly assess the contribution of SETD2 to the lineage switch, we cultured BMSCs of Setd2 fl/fl mice and then infected the cells with lentivirus expressing green fluorescent protein (GFP) or Cre recombinase. Setd2 fl/fl BMSCs infected with Cre lentivirus showed more intensive adipogenesis ( Fig 5A and 5B), consistent with elevated expression of adipocyte markers (S2A Fig). Reduced SETD2 expression accompanied with the decreased H3K36me3 level by Cre lentivirus expression was confirmed by western blot analysis ( Fig 5C). We then performed RNA sequencing (RNA-seq) and H3K36me3 chromatin immunoprecipitation sequencing (ChIP-seq) analysis to identify SETD2-regulated genes (Series GSE120361). Next-generation sequencing using RNA from GFP or Cre lentivirusinfected Setd2 fl/fl BMSCs, respectively, revealed that the global transcriptome was changed dramatically between Cre-expressed cells compared to the GFP-expressed cells, indicating the significant function of SETD2 in BMSCs (Fig 5D). Among a total of 17,063 genes expressed, 904 genes were up-regulated, and 913 genes were down-regulated (fold change > 1.5) in SETD2-depleted cells. The expression levels of 13 genes were examined by qPCR and validated the RNA-seq data (S2B Fig). To further identify genes directly regulated by SETD2 and H3K36me3 on a genome-wide scale, we performed ChIP using an H3K36me3-specific antibody followed by next-generation sequencing (ChIP-seq) assays. A total of 23,001 and 147,216 H3K36me3 peaks were identified in SETD2-depleted BMSCs (Setd2 fl/fl cells expressing Cre) and control cells (Setd2 fl/fl cells expressing GFP), respectively. The reduction of H3K36me3 peaks in SETD2-deficient cells (Fig 5E and 5F, S3 Fig) certified that SETD2 is the major histone methytransferase of H3K36me3, consistent with the decrease of H3K36me3 level in SETD2-deficient cells determined by western blot analysis ( Fig 5C). H3K36me3 was widely located in whole genomic regions, including gene bodies, promoters, 3 0 UTR, and intergenic zones, except regions surrounding transcription start sites (TSSs) where the occupancy of H3K36me3 peaks were dramatically decreased (Fig 5E and 5F). SETD2 deficiency led to the decrease of H3K36me3 level at gene bodies and transcription end sites (TESs + 5 kilo base pairs [Kb]), indicating that SETD2 is responsible for the trimethylation on these regions ( Fig  5E and 5F). Only a small percentage of H3K36me3 peaks resided in 5 0 proximal regions (0-5 Kb), reflecting that H3K36me3 and SETD2 regulates gene transcription mainly through other regulation zones but not promoter regions (Fig 5E and 5F). We next analyzed H3K36me3 peaks that were lost in SETD2-deficient cells but present in control cells and found that the peaks were from 17,647 genes. We integrated the ChIP-seq data with the RNA-seq data and found that 983 genes (S1 Table) showed both the change of expression level and the loss of H3K36me3 peaks by SETD2 deficiency (Fig 5G), indicating that the altered H3K36me3 level on the genomic regions might regulate the expression levels of these genes.

LBP downstream of SETD2 regulates the fate of MSCs
As our above data demonstrated that SETD2 deficiency could facilitate adipogenesis and suppress osteogenesis both in vitro and in vivo, we hypothesized that some of SETD2-and/or H3K36me3-regulated genes would affect BMSC fate determination. Previous studies have demonstrated that H3K36me3 could activate mRNA transcription initiation and elongation to increase transcription level [23,24]. Therefore, we focused on genes down-regulated by SETD2 deficiency in the previous sorted gene list (S1 Table). We also analyzed the expression pattern of those genes in BioGPS (http://biogps.org/) and focused on a list of genes that have been reported or predicted to be involved in adipocyte metabolism. Plasma long pentraxin 3 (PTX3) was reported to be associated with type 2 diabetes, human obesity, and metabolic syndrome [25][26][27]. LBP is a negative regulator of adipocyte differentiation and adipose tissue browning [28,29]. β-1,3-galactosyltransferase 2 (B3galt2) transfers galactose from UDP-galactose to substrates and is highly expressed in white and brown adipose tissue [30]. qPCR confirmed that the expression levels of these 3 candidates (Ptx3, Lbp, and B3galt2) were decreased in the bone marrow cells derived from Prx1-Cre, Setd2 fl/fl mice (S4A Fig). As a control, the expression level of some fundamental pathway-associated genes was stable (S4B Fig). We next sought to determine the effects of those targets on adipogenesis. The cultured BMSCs from 4to 6-week-old mice were infected with lentivirus expressing those candidate genes. And overexpression of LBP-not PTX3 and B3GALT2-could restrain the differentiation of adipocytes accompanied with decreased expression of adipogenesis markers, including Pparγ1, Cebpα, Lpl, and Fabp4, determined by qPCR (S4C-S4E Fig). Consistently, Lbp knockdown with lentivirus expressing Lbp shRNAs promoted BMSC adipogenesis (Fig 6A-6C). Immunofluorescence staining of distal femur using LBP specific antibody confirmed decreased expression of LBP in the bone marrow of Prx1-Cre, Setd2 fl/fl mice ( Fig 6D). Furthermore, the expression level of Lbp was decreased during adipogenesis and increased during osteogenesis, consistent with the changes of SETD2 expression (S4F and S4G Fig). To further determine the function of LBP, we cultured Setd2-deficient BMSCs and infected the cells with lentivirus expressing LBP. Overexpression of LBP could partially rescue the enhanced adipogenesis in SETD2-deficient cells, evidenced by the partially decreased Oil-red O staining (Fig 6E-6G). In addition, the expression level of adipocyte marker genes-including Pparγ1, Cebpα, Lpl, and Fabp4determined by qPCR was also decreased in SETD2-deficient cells with LBP lentivirus treatment (Fig 6H). We next examined whether LBP could regulate the osteoblast differentiation. Meanwhile, overexpression of Lbp could partially rescue the impaired osteoblast differentiation resulting from loss of SETD2, which was determined by Alp activity and its quantification (Fig 6I, upper panel and 6J) as an early-stage osteoblast differentiation marker, and Alizarin red S staining (Fig 6I, lower panel) as a later-stage mineral deposition marker. In addition, the osteoblast markers-including Runx2, Col1α1, Alp, and Bsp-were also partially rescued in SETD2-deficient cells by Lbp lentivirus (Fig 6K).
Next, we administrated recombinant LBP in the osteogenic differentiation medium in WT mBMSCs for 7 days and harvested the cells for Alp staining, Alp quantification, Alizarin red S staining, and qPCR. We found that recombinant LBP could promote osteogenesis, as evidenced by increased Alp activity and Alizarin red S staining (S5A and S5B Fig) Fig). In addition, we also added recombinant LBP into the adipogenic differentiation medium in WT mBMSCs for 6 days, harvested the cells for Oil Red O staining, and performed RT-qPCR. Consistently, we found that recombinant LBP could repress adipogenesis, determined by decreased Oil Red O staining and the decrease of the expression of adipocyte marker genes, including Pparγ1, Cebpα, fabp4, and Perilipin (S5D-S5F Fig). Taken together, these data supported that LBP, whose expression level is regulated by SETD2, is a novel regulator of MSC fate.

H3K36me3 mediated by SETD2 regulates transcriptional initiation and elongation of the Lbp gene
We next sought to answer how SETD2 regulates Lbp expression. ChIP-seq analyses demonstrated that the genomic region of the Lbp gene was occupied by H3K36me3, and Setd2 deficiency led to the decrease of H3K36me3 occupancy (Fig 7A). However, the H3K36me3 profiles showed much fewer changes on the fundamental pathway-associated gene regions (S6 Fig). The decreased H3K36me3 peaks were further confirmed by ChIP-qPCR analysis using H3K36me3-specific antibody (Fig 7B). To examine whether H3K36me3 occupancy of the Lbp gene could affect its mRNA transcriptional initiation and/or elongation, we analyzed the expression of the intron-containing nascent transcripts. The expression of almost every analyzed intronic region, from N-terminal to C-terminal, was decreased in BMSCs of Prx1-Cre, Setd2 fl/fl mice compared to WT control mice (Fig 7C). And then we performed RNA pol II ChIP assay followed by real-time PCR with primers covering different intronic regions of the Lbp gene ( Fig 7D). As expected, the binding affinity of RNA pol II was reduced significantly at nearly all examined intronic regions, especially at −0.5 kb, upstream from the TSS, and intron1 of Lbp in BMSCs of Prx1-Cre, Setd2 fl/fl mice ( Fig 7D). These data indicated that loss of H3K36me3 occupancy could affect RNA pol II binding affinity around TSS and gene body and thus impair both mRNA transcriptional initiation and elongation of Lbp.
To further confirm that the mBMSC lineage commitment was dependent on H3K36me3 mediated by SETD2, we generated a truncated WT Setd2 (amino acid 1,297-2,537), which includes function domains for catalyzing H3K36me3, and a SET domain mutant of truncated Setd2 (R1599H), which failed to trimethylate H3K36 [31]. Overexpression of tSetd2 could not inhibit the adipogenesis of WT mBMSCs (S7A and S7B Fig). The possible reason is that the overexpression of SETD2 could not make excessive trimethylation on H3K36 in WT BMSCs (S7C Fig). As expected, overexpression of WT tSetd2 and mutated tSetd2 in SETD2-deficient BMSCs showed that the WT tSetd2 instead of mutated tSetd2 could rescue the level of H3K36me3 ( Fig  7E). Consistently, the Lbp expression was up-regulated by overexpression of WT tSetd2 rather than mutated tSetd2 (Fig 7F). Furthermore, the adipogenesis was repressed by WT tSetd2 instead of mutated tSetd2 in Setd2-deficient BMSCs, as proved by the Oil Red O staining and the indicated adipocyte markers (Fig 7G-7I). These results implicated that the transcription initiation and elongation of the Lbp gene were regulated by H3K36me3 occupancy, mediated by SETD2.

Discussion
In this study, we found that H3K36me3, mediated by histone methyltransferase SETD2, plays a key role in BMSC fate. SETD2 conventional knockout mice are embryonic lethal at E10.5-Bsp, bone sialoprotein; Cebpα, CCAAT enhancer binding protein α; Col1α1, Collagen type 1 alpha 1; Fabp4, fatty acid binding protein 4; LBP, lipopolysaccharide-binding protein; Pparγ1/2, peroxisome proliferative activated receptor γ1/2; qPCR, quantitative PCR; shRNA, short hairpin RNA; WT, wild-type.   [32], and Setd2 knockout ESCs have defects for differentiation toward endoderm [33]. Here, we constructed a conditional knockout mouse model to determine the physiological functions of H3K36me3 and SETD2 during BMSC lineage commitment. Our observation that SETD2 deficiency in BMSCs leads to the decrease of H3K36me3 but not mono-or dimethylation is consistent with the previous report that homozygous disruption of Setd2 impaired H3K36me3 but not mono-or dimethylation [32], confirming that SETD2 is the major histone methyltransferase for H3K36me3.
Age-related osteoporosis is characterized by activated adipogenesis and suppressed osteogenesis in bone marrow stromal cells, leading to progressive fat accumulation and bone loss [34,35]. However, the regulations of this process remain to be further understood, especially in vivo epigenetic regulations. We observed that SETD2 deficiency in BMSCs led to reduced bone formation and accumulation of fat at a young age, resembling the phenotype of agerelated osteoporosis. Although Setd2 is also depleted in chondrocytes, Setd2 deficiency barely affected the chondrocyte formation and differentiation in vivo and in vitro (S8 Fig). We found that Setd2 and Lbp were decreased in the aged mBMSCs (S9 Fig), supporting our hypothesis that loss of Setd2 could shift the lineage commitment of BMSCs to adipogenesis by repressing the expression of Lbp. These results are consistent with the report that H3K36 methylation promotes longevity [36]. And the SETD2-deficient mice might serve as a novel model to investigate age-related osteoporosis. Zhou and colleagues demonstrated that adipocytes function as the niche component that promote hematopoietic regeneration, while Ambrosi and colleagues established that adipogenic cells reduce hematopoietic reconstitution [37,38]. In our study, we observed excess adipocyte accumulation in the bone marrow of Prx1-Cre; Setd2 fl/fl mice. In order to determine whether the lack of Setd2 in BMSCs could influence the hematopoiesis progress, we analyzed the occupation of LSK cells (Lin − Sca1 + C-kit + cells) and HSCs (CD150 + CD48 − Lin − Sca1 + C-kit + cells) in total bone marrow cells of 3 pairs of 6-month-old mice. However, we barely detected any significant impact on percentage of both LSK cells and HSCs by the accumulation of adipocytes in Setd2-deficient mice (S10 Fig). Thus, we speculated that the function of marrow adipocytes on HSCs might depend on the state and environment of the bone marrow, and the factors that adipocytes secreted, rather than the number of adipocytes only.
In past decades, the study of H3K36me3 and SETD2 mainly focus on their function as tumor suppressors, including clear cell renal cell carcinoma, chondroblastomas, lymphoma, acute leukemia, intestinal tumorigenesis, and so on [19,[39][40][41][42]. Also, we found that the proliferation rate was increased mildly in Setd2-deficient BMSCs (S11 Fig). In the mechanism, Set2, H3K36 histone methyltransferase in yeast, has been reported to be required for transcription elongation [16]. Also, SETD2 could directly regulate transcriptional initiation of Fgfr3 through histone H3K36me3 modification [15]. In our study, Setd2 could contribute to RNA pol II- been attributed to the methylation of nonhistone proteins, such as α-TUBULIN at lysine 40 and STAT1 at lysine 525 [43,44]. The function of nonhistone substrates mediated by SETD2 on MSC fate control need to be further determined.
Lysine methylation is one of the most important histone post-translational modifications that regulate chromatin structure. KDM4/JMJD2 proteins-containing KDM4A/JMJD2A, KDM4B/JMJD2B, KDM4C/JMJD2C and KDM4D/JMJD2D-are demethylases that target histone H3 on lysines 9 and 36 [45]. Demethylation of the repressive mark H3K9me3 by KDM4A could facilitate PPARγ promoter-specific binding and function in adipocytes [46]. However, KDM4C has been reported to reduce PPARγ transcriptional activity for PPARγ response element (PPRE) genes and reduce pre-adipocyte 3T3L-1 cell differentiation [47]. The reason for the opposite function on the PPARγ between the KDM4A and KDM4C remains unclear. Increase of H3K9 trimethylation by KDM4B depletion has been reported to promote adipogenesis and repress osteogenesis in hMSCs [10]. KDM4D is unique within the KDM4 family in that it lacks both the PHD and Tudor domains and thus is only half the size of KDM4A-C [48]. However, KDM4D has not been studied in bone and MSC lineage balance. Although several reports have been made to investigate the function of KDM4 on MSC lineage commitment and adipocyte differentiation in vitro, it is still worthwhile to pursue deep insight into their physiological significance.
In our study, we performed RNA-seq and ChIP-seq analysis and identified that SETD2 and H3K36me3 could regulate the transcription of a panel of different genes. Among these genes, LBP could partially rescue the phenotype of Setd2-deficient BMSCs. We also demonstrated that the lack of H3K36me3 on the Lbp gene body leads to less Pol II occupancy and decreased transcription initiation and elongation. LBP has been reported as an essential component of the innate immune responses to gram-negative bacterial infection and can protect murine liver from different injuries [49,50]. LBP knockdown using shRNA or anti-LBP antibody led to increased markers of adipogenesis and decreased adipocyte inflammation in human adipocytes [51]. In our study, overexpression of LBP could partially change the balance between osteoblasts and adipocytes, suggesting that the decrease of LBP in Setd2-deficient BMSCs is responsible to a certain degree for dysregulated fate determination of MSCs in Prx1-Cre, Setd2 fl/fl mice, indicating that circulating LBP is not only an immune regulator but also a regulator for cell fate and cell metabolism. It will be worthwhile to further investigate how Lbp regulates adipogenesis and osteogenesis. Lbp might function as a lipid-binding protein or adipokine to stimulate some specific signaling pathways. Furthermore, the specific Lbp receptor on the membrane of BMSCs remains to be elucidated.
Overall, our study identified mammalian SETD2 as a novel regulator of MSC fate and suggested that modulation of H3K36me3 could have a therapeutic potential in osteoporosis. Our study also identified LBP as a downstream target of SETD2. Considering that LBP is a secreted protein and that it can increase BMSC osteogenesis and decrease adipogenesis, LBPs might be an attractive reagent for rejuvenating our bone.

Ethics statement
Animals were bred and maintained under specific pathogen free (SPF) conditions in the institutional animal facility of the Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences. All animal experiments were performed with protocol approved by the Animal Care and Use Committee of Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences (approval number: SIBCB-NAF-14-001-S350-019).

Mouse lines
Setd2 fl/fl mice were crossed with the Prx1-Cre strain (a gift from Andrew McMahon, Harvard University) to generate Prx1-Cre, Setd2 fl/fl mice. All mice analyzed were maintained on the C57BL/6 background.

Cell culture
BMSCs were extracted from femurs and tibiae of 4-to 6-week-old mice. Marrow plug was flushed with a 1-mL syringe. These plugs were then dispelled into single cell and seeded in 10-cm dish containing α-MEM (Corning, Corning, NY) supplemented with 10% fetal bovine serum (FBS). For osteoblast differentiation, cells were cultured in α-MEM containing 10% FBS, 50 μg/ml L-ascorbic acid, and 1,080 mg/mL β-glycerophosphate. The osteoblast differentiation level assay was performed following a previously published method [52]. For quantitative analysis of Alp activity, osteoblasts were incubated with Alamar Blue and were then incubated with phosphatase substrate (Sigma-Aldrich, St. Louis, MO) dissolved in 6.5 mM Na 2 CO 3 , 18.5 mM NaHCO 3 , 2 mM MgCl 2 . Alp activity was then read with a luminometer (Envision). Bone nodule formation was stained with 1 mg/mL Alizarin red S solution (pH 5.5) after 14 or 21 days of induction. For adipocytes, BMSCs were plated in 96-well plate at a density of 30,000 cells/well and stimulated in adipogenic induction medium (α-MEM/10% FBS containing 1 uM dexamethasone, 0.1 mM rosiglitazone, 0.5 mM IBMX, 10 ug/mL insulin) for 1 day followed by adipogenic maintenance medium (α-MEM/10% FBS containing 10 ug/mL insulin) for 2 or 3 days. After mature adipocyte formation, cells were stained with 2 mg/mL Oil red O solution or Bodipy 493/503.
Human embryonic kidney cell line (HEK293T) was obtained from ATCC and maintained in DMEM (Corning, Corning, NY) containing 10% FBS.

Analysis of bone formation rate by Calcein-Alizarin red S labeling
Five-week-old mice were injected intraperitoneally with 20 mg/kg Calcein (1 mg/mL in 2% NaHCO 3 solution) and 40 mg/kg Alizarin red S (2 mg/mL in H 2 O) on day 0 and day 4 separately. On day 7, the mice were euthanized, and the bones were fixed, dehydrated, and embedded with Embed-812 (Electron Microscopy Sciences). The samples were sectioned as 5 μm with hard tissue cutter. The distance between Calcein and Alizarin red S positive bands were measured every 25 microns along the cortical bone surface for the length of the tibia using Image J. Consecutive sections were stained with silver nitrate to measure trabecular bone.

μ-QCT analysis
Preparation of skeletal tissue and μ-QCT analysis were performed as previously described [55]. The mouse femurs were skinned and fixed in 70% ethanol. Scanning was performed with the instrument μ-QCT system SkyScan1176 (Bruker, Kartuizersweg, Belgium). For analysis of femoral bone mass, a region of trabecular bone 2.0 mm wide was contoured, starting 600 microns from the proximal end of the distal femoral growth plate. For femoral trabecular bone, a threshold of 80-255 permille was used. The region of interest of the femoral cortical bone was 1.0 mm wide, starting 3.7 mm from the proximal end of the distal femoral growth plate. For femoral cortical bone, a threshold of 125-255 permille was used. Three-dimensional reconstructions were created by stacking the two-dimensional images from the indicated regions.

Real-time RT-PCR analysis
Total RNA was prepared using TRIzol (T9424; Sigma) and was reverse transcribed into cDNA with the PrimeScript RT Reagent Kit (PR037A; TakaRa). The real-time reverse transcriptase RT-PCR reaction was performed with the BioRad CFX96 system. The primer sets used were listed in S2 Table. ChIP-seq and RNA-seq A total of 10 7 bone marrow cells from Setd2 fl/fl mice were infected with lentivirus expressing GFP and Cre, respectively. After 4-day puromycin treatment, GFP-and Cre-BMSCs were fixed by 1% formaldehyde for 10 minutes and terminated by glycine for 5 minutes (final concentration = 0.125 M). After 2 times washing by precooling PBS (protease inhibitor containing), cells were scraped and resuspended in SDS-lysis buffer, respectively (50 mM Tris-HCl [pH 7.5], 10 mM EDTA, 1% SDS and protease inhibitor), and sonicated. Cells were centrifuged to obtain cell extracts, which then were added to precleaning protein G agarose and rotated for 1 hour at 4˚C. Extracts were centrifuged, and supernatants were harvested to a new tube. ChIP assay was performed using H3K36me3 antibodies. Normal IgG was used as a negative control. qPCR was used to amplify various regions of the target gene genome, and primers for ChIP-qPCR primers are listed in S2 Table. ChIP-seq data processing The high-throughput sequencing was performed by the Computational Biology Omics Core, CAS-MPG Partner Institute for Computer Biology (PICB), Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences. The SOAP version 2.20 alignment tool was used to align ChIP-seq reads to the mouse genome build mm9 [56]. The reads with fewer than 2 mismatches that uniquely mapped to the genome were used in subsequent analyses. We calculated the distance from the peak centers to the annotated TSSs and then defined the nearest genes as peak-related genes.

RNA-seq data processing
Raw reads were mapped to mm9 using the TopHat version 1.4.1 program [57]. We assigned fragment per kilo base per million (FPKM) as an expression value for each gene using Cufflinks version 1.3.0 software [58]. Then, Cuffdiff software was used to identify differentially expressed genes between treatment and control samples [59]. Differentially expressed gene heat maps were clustered by k-means clustering using the Euclidean distance as the distance and visualized using Java TreeView software [60] (series GSE120361).

Statistics
Each experiment was performed more than twice, and the data are presented as the mean ± SD. Student t test was used to compare the effects of all treatments. Statistically significant differences are indicated as follows: " � " for p < 0.05 and " �� " for p < 0.01.