HDAC5 controls MEF2C-driven sclerostin expression in osteocytes

Abstract

Osteocytes secrete paracrine factors that regulate the balance between bone formation and destruction. Among these molecules, sclerostin (encoded by the gene SOST) inhibits osteoblastic bone formation, and is an osteoporosis drug target. The molecular mechanisms underlying SOST expression remain largely unexplored. Here we report that histone deacetylase 5 (HDAC5) negatively regulates sclerostin levels in osteocytes in vitro and in vivo. HDAC5 shRNA increases, whereas HDAC5 overexpression decreases SOST expression in the novel murine Ocy454 osteocytic cell line. HDAC5 knockout mice show increased levels of SOST mRNA, more sclerostin-positive osteocytes, decreased Wnt activity, low trabecular bone density, and reduced bone formation by osteoblasts. In osteocytes, HDAC5 binds and inhibits the function of MEF2C, a crucial transcription factor for SOST expression. Using chromatin immunoprecipitation, we have mapped endogenous MEF2C binding in the SOST gene to a distal intergenic enhancer 45 kB downstream from the transcription start site. HDAC5 deficiency increases SOST enhancer MEF2C chromatin association and H3K27 acetylation and decreases recruitment of co-repressors NCoR and HDAC3. HDAC5 associates with and regulates the transcriptional activity of this enhancer, suggesting direct regulation of SOST gene expression by HDAC5 in osteocytes. Finally, increased sclerostin production achieved by HDAC5 shRNA is abrogated by simultaneous knockdown of MEF2C, indicating that MEF2C is a major target of HDAC5 in osteocytes.

Introduction

Osteocytes, the most abundant cell type in bone, are important regulators of bone remodeling by producing paracrine factors (1). Osteocytes express RANKL, a crucial osteoclastogenic cytokine (2, 3), and sclerostin (encoded by SOST), a potent inhibitor of canonical Wnt signaling, that decreases osteoblast number and activity (4). In humans, lack of SOST due to loss of function mutations causes high bone mass in the rare disease, sclerosteosis (5), while reduced SOST levels due to deletion of a downstream intergenic enhancer region causes a milder high bone mass phenotype in Van Buchem disease (6). Common SOST polymorphisms are associated with bone density and fracture risk variation in the general human population (7). Sclerostin antibodies are now under investigation for osteoporosis therapy (8).

While regulation of SOST expression by parathyroid hormone (PTH) (9–11) and mechanical forces (12–15) has been demonstrated, little is known about molecular mechanisms controlling its expression in osteocytes. The best-studied regulator of SOST expression in osteocytes is the transcription factor MEF2C. Presumably through binding to a downstream enhancer sequence in the intergenic region between SOST and its neighbor MEOX, MEF2C positively regulates SOST expression (11). Mice lacking MEF2C in osteocytes, or lacking the downstream enhancer region, display reduced SOST levels and high bone mass (16, 17).

Histone deacetylases (HDACs) are a family of enzymes capable of deacetylating lysine residues in a wide variety of cellular proteins, including histones (18). Class IIa HDACs (HDACs 4, 5, 7, and 9) contain N-terminal extensions with phosphoacceptor 14-3-3 binding sites that allow them to sense and transduce signaling information (19). In addition, the N-termini of class IIa HDACs contain a MEF2-binding site that mediates repression of MEF2-driven gene expression (20). While roles for class IIa HDACs in osteoblasts have been described (21–23), their functions in osteocytes remains largely unexplored. Recently, HDAC5 overexpression was reported to suppress SOST expression in luciferase assays in UMR106 cells (24). We used loss of function analysis under physiologic conditions to ask whether class IIa HDACs might regulate MEF2C-driven SOST expression in osteocytes. We report that HDAC5 directly controls SOST expression and that mice lacking HDAC5 show increased sclerostin levels in osteocytes, low bone density, and reduced bone formation.

Materials and Methods

Germline HDAC5−/− mice were provided by Dr. Eric Olson and studied at 8 weeks of age. Cell culture experiments were performed using a single cell subclone of Ocy454 cells (Spatz et al, manuscript in preparation). Lentiviral infections were performed using protocols available online (www.broadinstitute.org). Sclerostin ELISAs were performed using an antibody pair as described in (25). Please see supplemental information for full details.

Results

We sought to establish an in vitro system to dissect mechanisms of SOST expression in osteocytes. Ocy454 cells are a conditionally-immortalized cell line with many properties of bone-embedded osteocytes. The line was derived from bitransgenic mice for DMP1-GFP and a thermolabile SV40 large T antigen that is active at 33°C and inactive at 37°C (Spatz et al, manuscript under review). Lentiviral (LV)-based shRNA-mediated gene silencing (26–28) was used to determine the roles of specific genes in regulating SOST expression. Figure S1a demonstrates that Ocy454 cells can be infected with a puromycin-resistance-conferring LV expressing a control shRNA (shGFP) and retain sclerostin expression and its regulation by PTH.

We next used shRNA to reduce levels of known SOST-positive and -negative regulators, MEF2C (11, 16, 17) and Gsα (29, 30), respectively. Gsα shRNA increased SOST expression and sclerostin secretion (Spatz et al, manuscript under review). For MEF2C shRNA, the range of sclerostin expression was determined in each experiment using 8–10 control shRNA-expressing lentiviruses targeting non-expressed genes (LacZ, luciferase, GFP, RFP). Dotted lines indicate two standard deviations above and below the mean value of sclerostin expression in the presence of control shRNAs. LV-mediated shRNA for MEF2C selectively reduced sclerostin secretion (Figure S1b, each data point indicates a separate Mef2-targeting hairpin, data for MEF2B is not shown because MEF2B is not expressed in Ocy454 cells), and all the MEF2C-targeting shRNAs effectively reduced MEF2C protein levels (Figure S1c).

A similar approach was then used to test the function of class IIa HDACs. As shown in Figure 1a, infection with independent hairpins targeting HDAC5 (but not HDAC4, HDAC7, or HDAC9) increased sclerostin secretion across multiple experiments. For the HDAC5 shRNAs, individual hairpins are labeled next to the corresponding data points in Figure 1a and 1b. The individual hairpins (F9 and F12) that best reduced HDAC5 mRNA and protein levels increased SOST expression (Figure 1b and Figure S1d/e). In contrast, HDAC4 and HDAC7 shRNAs comparably reduced target mRNA abundance but did not increase SOST (Figure 1c and 1d). HDAC9 is not expressed in Ocy454 cells (data not shown). While the most effective HDAC4 and HDAC5 shRNAs reduced target mRNA by 60–70%, the “best” HDAC5 shRNA (F12) was more effective at reducing target protein levels than the corresponding “best” HDAC4 shRNA (D8) tested (Figure 1e). While these data support a role for HDAC5 in controlling SOST expression, we cannot rule out a contribution from HDAC4 or HDAC7, given differences in protein knockdown. Since comparable results were seen for two independent HDAC5 hairpins (F9 and F12; Figure 1b, Figure S1d, and data not shown), the F12 hairpin was used for further study.

Figure 1. Identification of HDAC5 as a putative regulator of SOST in vitro.

(A) Ocy454 cells were infected with 10 separate control shRNA-expressing lentiviruses, and 5 separate shRNAs targeting the indicated class IIa HDAC. Ten days later, sclerostin levels were determined by ELISA and normalized to cell count per well. In experiments 1 and 2, cells were infected after 1 day at 37°C. In experiments 3 and 4, cells were infected after 5 days at 37°C. The “hit” cutoff was determined as 2 standard deviations above the mean control sclerostin levels for each experiment. (B-D) Ocy454 cells were infected with individual hairpins targeting the indicated class IIa HDAC or control (shGFP) lentivirus. 10 days later, RNA was prepared and knockdown efficiency and SOST levels were determined for each hairpin (relative to shGFP). (E) Protein knockdown for shHDAC4 and shHDAC5 was determined by immunoblotting.

Stable HDAC5 (F12) shRNA-expressing Ocy454 subclones were then generated (Figure 2a). These cells display growth kinetics (Figure 2b) comparable to that of control shLacZ-expressing cells. HDAC5 knockdown increases sclerostin secretion, particularly at earlier time points after cells are switched to the non-permissive growth temperature (Figure 2c). Analysis of a panel of osteocytic genes over the course of differentiation revealed that HDAC5 knockdown targets a subset (SOST and DMP1, but not PHEX) of osteocyte markers (Figure 2d). Osteocytes express the key regulators of osteoclastogenesis, RANKL and OPG (2, 3, 31). HDAC5 shRNA does not affect basal (Figure S2a, b) or PTH-stimulated (Figure S2c) expression of these factors.

Figure 2. HDAC5 controls SOST expression in vitro.

(A) Immunoblot for HDAC5, HDAC4, and HDAC7 from stable shLacZ and shHDAC5-expressing Ocy454 subclones. Tubulin is used as a loading control. (B) Growth curve showing comparable expansion kinetics of shLacZ and shHDAC5 cells at 33°C (n=42 wells/cell line). (C) Kinetics of sclerostin secretion after identical numbers shLacZ and shHDAC5 cells were switched from 33°C to 37°C for the indicated time. Every 3–4 days, conditioned medium was collected and analyzed by sclerostin ELISA (n=4 wells/time point). P values comparing shLacZ to shHDAC5 are 0.19 at day 3, 0.011 at day 7, 0.00019 at day 10, 1.59*10⁻⁵ at day 14, 0.012 at day 17, and 0.00081 at day 20. (D) RNA was isolated from shLacZ and shHDAC5 cells after culture at 37°C for the various times and the levels of the indicated gene were determined by RT-qPCR. All p values for HDAC5 levels comparing control and HDAC5 shRNA are below 0.001. For SOST, at day 3 p=0.29, day 10 p=0.043, day 17 p=0.012. For DMP1, at day 3 p=0.84, day 10 p=0.048, day 17 p=0.0041. (E) Immunoblot showing overexpression of FLAG-tagged HDAC5 in Ocy454 cells. (F) Growth curve showing comparable expansion kinetics of control and HDAC5-overexpressing cells at 33°C (n=12 wells/cell line). (G) Overexpressing cells were analyzed for expression of indicated genes as in (C). P value comparing LV-GFP to LV-HDAC5 WT for SOST is 0.038, for DMP1 is 0.0091. P value comparing LV-GFP to LV-HDAC5 S/A for SOST is 9.22*10⁻⁵, for DMP1 is 0.0081. (H) Overexpressing cells were analyzed for sclerostin secretion by ELISA as in (C). P values comparing LV-GFP to LV-HDAC5 WT are 0.023 at day 7, 0.069 at day 10, 0.0056 at day 14, 0.0050 at day 17, and 0.017 at day 20. P values comparing LV-GFP to LV-HDAC5 S/A cells are 0.0069 at day 7, 0.042 at day 10, 0.0034 at day 14, 0.00018 at d17, and 2.2*10⁻⁵ at day 21. (I) Immunoblot showing HDAC5 and HDAC4 protein levels in HDAC5 shRNA and HDAC5 shRNA plus HDAC5 cDNA cells. (J) Sclerostin ELISA from the indicated cells as in (C). At all time point, P value comparing control to shHDAC5 is <0.001, and P value comparing control to shHDAC5 plus LV-HDAC5 is non-significant. For all panels, * indicates p<0.05.

LV overexpression was next used to increase HDAC5 (wild type (WT) or S259/498A, a constitutively nuclear “super-repressor” version (32)) levels in Ocy454 cells (Figure 2e). While overexpression of either WT or S259/498A HDAC5 does not affect growth kinetics in Ocy454 cells (Figure 2f), overexpression of HDAC5, and especially of the S259/498A nuclear mutant form, reduces SOST expression and sclerostin secretion (Figure 2g, 2h).

To rule out off target shRNA effects, HDAC5 shRNA cells were rescued with a human HDAC5 cDNA with several synonymous nucleotide changes in the murine shRNA targeting sequences (Figure 2i). As shown in Figure 2j, HDAC5 cDNA rescue reduced sclerostin expression by HDAC5 shRNA cells, providing further evidence that the phenotype due to HDAC5 (F12) shRNA is due to reduced HDAC5 protein levels.

To show that HDAC5 mRNA is expressed in osteocytes, HDAC5 mRNA levels were analyzed in primary DMP1-GFP^neg (osteoblasts and stromal cells) and DMP1-GFP^pos (enriched osteocytes) long bone cells isolated from DMP1-GFP mice. As previously reported, the GFP-negative fraction expresses high levels of the osteoblast-specific gene Keratocan (33, 34), while SOST mRNA is exclusively detected in the GFP-positive fraction. As shown in Supplemental Figure 2d, HDAC5 mRNA is detected in both fractions, with slightly higher levels present in DMP1-GFP^pos cells.

To explore the role of HDAC5 in controlling sclerostin levels in vivo, global HDAC5-knockout (KO) mice were analyzed. These mice are normal-appearing and fertile, with reported cardiac and behavioral phenotypes (35, 36). Analysis of calvarial RNA from 8 week-old mice (n=12/group) demonstrates an 84% increase in SOST mRNA levels in the absence of HDAC5 (Figure 3a), with no changes in osteocyte density (Figure 3b). As previously described (37), sclerostin immunoreactivity is largely confined to deeply-embedded osteocytes relatively far from the endocortical surface in WT mouse tibiae (Figure 3c, left). In contrast, many more sclerostin-positive osteocytes are close to the endocortical surface in HDAC5-KO sections (Figure 3c, right). The overall percentage of sclerostin-positive osteocytes is modestly increased in the absence of HDAC5 (Figure 3d). In contrast, the percentage of sclerostin-positive peri-endosteal osteocytes (defined as embedded osteocytes within the inner ¼ of the cortical area) is dramatically increased in HDAC5 knockouts (Figure 3e).

Figure 3. Increased SOST/sclerostin levels in HDAC5-deficient mice.

(A) SOST levels relative to β-actin were determined in calvarial RNA (n=12 samples/genotype) from 8 week-old HDAC5+/+ (WT) and HDAC5−/− mice. P value for the comparison between WT and HDAC5−/− is 0.00019. (B) Comparable cortical osteocyte density in WT and HDAC5−/− bone. (C) Representative photomicrographs of sclerostin IHC (brown) in cortical bone on the lateral surface of the tibia from WT (right) and HDAC5−/− (left) sections. Arrowheads denote sclerostin-positive osteocytes close to the endosteal surface present in HDAC5−/−- tissue. (D, E) Quantification of sclerostin immunoreactive cells, n=10 sections from 5 mice scored per genotype. For both comparisons, p<0.001. For all panels, * indicates p<0.01.

We next investigated the functional consequences of increased osteocytic SOST expression in HDAC5-KO mice. Sclerostin is a Wnt pathway antagonist (38, 39); therefore, we anticipated reduced activity of this pathway in HDAC5-KO mice. Wnt signaling promotes β-catenin stabilization (40), in part via inhibiting its N-terminal phosphorylation (41). Immunostaining for the non-phosphorylated (active) form of β-catenin revealed reduced staining in endosteal osteoblasts in the absence of HDAC5 (Figure 4a, bottom). Active β-catenin levels were comparable between genotypes in primary spongiosa cells (Figure 4a, top), suggesting that reduced active β-catenin in endosteal osteoblasts in the HDAC5-KO strain is a selective phenomenon that correlates with increased sclerostin at this skeletal site. AXIN2, a canonical Wnt pathway target gene (42), levels are reduced in HDAC5 KO bone RNA (Figure 4b).

Figure 4. Skeletal consequences of HDAC5 deficiency.

(A) Active (non-phospho Ser33/37/Thr41) beta-catenin IHC on tibia sections from WT and HDAC5−/− mice. In the top panel, comparable staining is seen in growth plate chondrocytes and cells adjacent to bone surfaces in the primary spongiosa. Reduced staining is observed in cells lining the endosteal surface and in the periosteum (bottom panel, arrowheads), while staining intensity is comparable in cells lining trabecular surfaces. (B) AXIN2 levels relative to beta-actin were determined in calvarial RNA (n=12 samples/genotype) from 8 week-old mice. P value comparing AXIN2 levels in WT and HDAC5−/− is <0.0001. (C) Levels of the indicated genes relative to beta-actin were determined in calvarial RNA (n=12 samples/genotype) from 8 week-old HDAC5+/+ (WT) and HDAC5−/− mice. P value for BGLAP comparing WT and HDAC5−/− is 0.0022, for PHEX is 0.027, for HDAC5 is less than 0.001. (D) Osteocalcin IHC on tibia sections from WT and HDAC5−/− mice. Reduced staining intensity is observed in endosteal osteoblasts in HDAC5−/− tissue. (E) Representative µCT sagittal image of the distal femur showing trabecular osteopenia in the absence of HDAC5 in 8 week old female mice. (F) Trabecular bone volume fraction (BV/TV) from the distal femur of 8 week old female (WT, n=11, HDAC5−/−, n=10) and male (WT, n=12, HDAC5−/−, n=9). P value for the comparison between WT and HDAC5−/− females is 0.030, males is 0.009. Additional µCT parameters are shown in Supplemental Figure 3. For this figure, * indicates p<0.05. (G) Representative fluorescent image showing dual tetracycline labeling component of dynamic histomorphometry. Arrows (light green) show calcein labeling 7 days before sacrifice, and arrowheads (light orange) show demeclocycline labeling 2 days prior to sacrifice. (H) Representative toluidine blue stain from static histomorphometry. Arrows (red) show TRAP-stained osteoclasts. Arrowheads (orange) show osteoblasts.

To determine the role of HDAC5 in regulating (directly or indirectly) the expression of molecular markers of differentiation of various bone cell types, control and HDAC5-KO calvarial RNA was profiled for osteoblast (TNALP, BGLAP), osteoclast (ACP5, which encodes Trap5b), osteoclastogenic (TNFSF11 and TNFRSF11B encoding RANKL and OPG), and osteocyte (DMP1, PHEX) marker genes (Figure 4c). No obvious compensation at the mRNA level from the related class IIa HDACs 4 and 7 was observed. While the expression of the majority of genes were unaffected, two significant differences were found. First, PHEX levels were increased in HDAC5-KO skull RNA. Second, osteocalcin (BGLAP) mRNA levels were decreased. This corresponds to reduced osteocalcin immunostaining in osteoblasts on the endosteal surface in HDAC5-KO mice (Figure 4d). Trabecular bone density in the distal femur of 8-week-old male and female wild type and HDAC5 KO mice was quantified by micro-CT. A significant reduction in trabecular bone density was observed in HDAC5 KO mice (Figure 4e,f and S3).

To determine the tissue-level consequences of HDAC5 knockout, we performed static and dynamic histomorphometry on tibiae of 8 week old female WT and HDAC5 KO mice. As previously reported (23), trabecular bone volume (BV/TV) was approximately 40% lower in HDAC5-KO mice (Table 1). This observation was supported by other structural trabecular bone parameters and all parameters showed highly significant differences. In addition, all dynamic parameters of bone formation were significantly reduced in HDAC5-KO mice, with a 40% reduction in MAR and a 50% reduction in BFR/BS (Figure 4g), indicating a prominent defect in bone formation. These findings were further confirmed by analysis of cellular parameters, demonstrating reduced osteoblast number and surface per bone perimeter in HDAC5-KO mice. In contrast, osteoclastic parameters (osteoclast number and surface and eroded surface per bone perimeter) were not significantly changed by HDAC5 deficiency (Figure 4h), although there was a trend towards reduced osteoclast parameters. These findings are consistent with our model in which high sclerostin levels in HDAC5-KO mice causes blunted bone formation. Potential causes for the discrepant cellular data presented here and reported by Obri et al (23) are addressed in the discussion.

Table 1.

Histomorphometry analysis from proximal tibia of 8 week old female mice

Parameter	WT	HDAC5−/−	p value
	(n=8)	(n=9)	(t test)
BV/TV (%)	6.31 (1.59)	3.68 (1.18)	0.00141**
Tb.Th (um)	29.8 (2.94)	22.6 (4.13)	0.00091**
Tb.N (/mm)	2.13 (0.45)	1.61 (0.31)	0.0142*
Tb.Sp (um)	478 (128)	621 (124)	0.0333*
MAR (um/day)	2.13 (0.24)	1.28 (0.32)	0.00008**
MS/BS (%)	28.7 (3.23)	24.8 (2.85)	0.02812*
BFR/BV (%/year)	1533 (224.3)	969 (287.8)	0.00106**
BFR/BS (um3/um2/day)	222 (25.6)	117 (37.8)	0.00003**
Ob.S/B.Pm (%)	15.1 (3.15)	10.4 (2.51)	0.00385**
N.Ob/B.Pm (/mm)	10.3 (2.18)	7.53 (1.69)	0.00940**
OS/BS (%)	7.15 (3.04)	2.8 (1.49)	0.00169**
O.Th (um)	3.68 (0.33)	2.85 (0.54)	0.00207**
Oc.S/B.Pm (%)	7.63 (1.22)	6.62 (1.77)	0.19850#
N.Oc/B.Pm (/mm)	2.70 (0.37)	2.22 (0.73)	0.11504#
ES/BS (%)	3.05 (1.05)	2.18 (0.94)	0.09356#

Data are expressed as mean (STD)

^*p<0.05,

^**p<0.01,

^#p=n.s.

Having established that HDAC5 controls sclerostin levels in vitro and in vivo, we focused on the molecular mechanism(s) whereby this occurs. Since class IIa HDACs are known to regulate MEF2 function in many settings (20), and MEF2C controls SOST expression in vivo (16, 17) and in Ocy454 cells (Figure S1b), we asked whether HDAC5 could regulate MEF2C function in osteocytes. Endogenous MEF2C and HDAC5 proteins associate in Ocy454 cells (Figure 5a; CREB and Sp1 are used as negative controls). A MEF2-driven reporter (43, 44) is more active in HDAC5 shRNA Ocy454 cells than in control LacZ shRNA cells (Figure 5b, MEF2C shRNA serves as a positive control).

Figure 5. Increased MEF2C activity in osteocytes lacking HDAC5.

(A) MEF2C was immunoprecipitated from Ocy454 cells followed by immunoblotting for HDAC5, CREB, and Sp1. (B) MEF2-driven luciferase activity relative to renilla was determined in the indicated Ocy454 cells. P values comparing control to HDAC5 shRNA and control to MEF2C shRNA are less than 0.001. (C) Inter-species conservation plot from UCSC Genome Browser of mouse chromosome 11 between SOST (right) and MEOX (left). Positions of ChIP PCR primer pairs are shown below. The region corresponding to the human Van Buchem’s disease deletion lies approximately between primer sets 3 and 12. (D, E) ChIP for MEF2C (D) and H3K27Ac (E) was performed following culture of Ocy454 cells for 14 days at 37°C. (F) ChIP-qPCR using primer set 9 (+45kB SOST enhancer) for MEF2C, H3K27Ac, and p300 over the course of Ocy454 cell differentiation. For all three antibodies, significantly (p<0.01) increased enrichment was noted at day 15 compared to day 1. For all panels, error bars represent triplicate biological repeats. For all panels, * indicates p<0.01.

Several conserved non-coding sequences downstream of the SOST gene are present (45) in the region deleted in Van Buchem Disease (Figure 5c; each number below the conservation plot corresponds to a different PCR primer set). When a region 45 kB downstream from the SOST transcription start site (termed ECR5 (11) by others and SOST’9’ here) is deleted in vivo, SOST levels are reduced and high bone mass is observed (17). Endogenous MEF2C association with the SOST gene has not been reported. We performed chromatin immunoprecipitation (ChIP) to determine MEF2C occupancy in Ocy454 cells cultured at 37°C for 14 days (a time point in which SOST expression is high). MEF2C chromatin association is seen in the +45 kB region amplified with primer set 9 (Figure 5d). Histone 3 lysine 27 acetylation (H3K27Ac, a mark of enhancer activity, (46)) is also found at this same region (Figure 5e). Since SOST expression is up-regulated in Ocy454 cells over time at 37°C (Figure 2c and 2h), we asked whether temporal changes in chromatin dynamics might occur at the +45kB enhancer. As shown in Figure 5f, SOST upregulation over time is accompanied by increased MEF2C, H3K27Ac, and p300 (another marker of active enhancers, (47)) occupancy at this genomic site.

Three lines of evidence support a direct role for HDAC5 in regulating MEF2C activity at the +45 kB SOST enhancer over the course of osteocyte differentiation. First, HDAC5 shRNA causes increased activity of this element, but not of the proximal SOST promoter, in luciferase assays (Figure S4a, b). MEF2C shRNA reduces the activity of this enhancer (Figure S4c). Second, HDAC5 overexpression dose-dependently inhibits the activity of a MEF2-driven reporter from the desmin (Figure S4d) and the SOST’9’ (+45kB) enhancer (Figure S4e). Third, endogenous HDAC5 association with this region in control (but not HDAC5 shRNA) Ocy454 cells can be detected by ChIP (Figure 6a). In addition to the +45kB enhancer site, HDAC5 associates with two other putative conserved enhancers with increased H3K27Ac levels. Moreover, HDAC5 +45kB SOST enhancer occupancy decreases over time at 37°C (Figure 6b), consistent with increased SOST expression. While HDAC5 itself is a weak deacetylase (48), it associates with the nuclear co-repressor NCoR and the potent class I deacetylase HDAC3 (49, 50). Like HDAC5, endogenous NCoR and HDAC3 enhancer occupancy decrease as SOST expression increases (Figure 6b).

Figure 6. MEF2C-dependent sclerostin expression is controlled by HDAC5 at the +45 kB SOST enhancer.

(A) HDAC5 ChIP results from shLacZ or shHDAC5 Ocy454 cells grown at 37°C for 14 days. Error bars represent standard error from biological triplicates, * denotes regions showing HDAC5 association that is reduced with HDAC5 shRNA. (B) ChIP-qPCR for HDAC5, NCoR, and HDAC3 was determined over the course of Ocy454 cell differentiation as in 5F. For all three antibodies, significantly (p<0.01) decreased enrichment was noted at day 15 compared to day 1. (C) MEF2C and H3K27Ac ChIP comparing control and HDAC5 shRNA Ocy454 cells cultured at 37°C for 14 days, qPCR on recovered DNA was performed using primer set 9. P value comparing shLuciferase and shHDAC5 for Mef2c ChIP is 0.0013, for H3 K27Ac ChIP is 0.011. (D) H3K27Ac ChIP comparing control (LV-GFP) and HDAC5 S259/498A overexpressing cells after 20 days at 37°C, analyzed as in (C) except that * indicates decreased H3 K27Ac observed with HDAC5 S259/498A overexpression. P value for primer set 9 comparing shLuc and shHDAC5 for H3 K27Ac ChIP is 0.00018. (E) ChIP was performed for NCoR and HDAC3 on control (shLuciferase) and HDAC5 shRNA Ocy454 cells after 3 days in culture at 37°C. qPCR was then performed for the SOST +45kB enhancer using primer pair 9. P value for NCoR comparing shLuc and shHDAC5 is 0.00231, p value for HDAC3 comparing shLuc and shHDAC5 is 0.00418. (F) Immunoblot showing simultaneous knockdown of HDAC5 and MEF2C in Ocy454 cells. (F) Sclerostin secretion by the indicated shRNA-expressing cells over time at 37°C. (G) SOST RT-qPCR by the indicated shRNA-expressing cells over time at 37°C. Error bars represent values obtained from experimental triplicates. In F and G, all differences in sclerostin secretion and SOST mRNA levels are statistically significant (p<0.001) comparing control and HDAC5 shRNA and HDAC5 shRNA to HDAC5 shRNA plus MEF2C shRNA. For all panels, * indicates p<0.05 comparing control (shLacZ or LV-GFP) and HDAC5 shRNA or overexpressing cells.

We next tested the functional consequences of HDAC5 shRNA on MEF2C binding and the presence of H3K27Ac at the SOST +45kB enhancer. Increased amounts of MEF2C and H3K27Ac are observed at this site, as well as two other conserved intergenic loci, in HDAC5 shRNA cells (Figure 6c and S5a/b). HDAC5 S259/498A overexpression reduces region 9 H3K27Ac levels (Figure 6d and S5c). Consistent with its proposed role as a sequence-specific scaffolding protein to recruit co-repressor complexes to genomic sites (51), HDAC5 shRNA decreases NCoR and HDAC3 association with the SOST enhancer (Figure 6e).

Finally, we asked whether MEF2C is required for increased SOST production caused by HDAC5 shRNA. Ocy454 cells were infected with shRNA-expressing lentiviruses to knock down HDAC5, MEF2C, or both (Figure 6f). Reducing MEF2C levels in the setting of HDAC5 shRNA decreased sclerostin secretion, indicating that MEF2C is required for the ability of HDAC5 shRNA to increase SOST (Figure 6g/h). This confirms our model that HDAC5 knockdown causes a MEF2C gain of function phenotype which drives high sclerostin production.

Discussion

Our data indicate that HDAC5 functions as a negative regulator of MEF2C-dependent SOST expression in osteocytes. This maps (at least in part) to the distal enhancer 45kB downstream from the SOST transcription start site identified by Leupin, Collette and colleagues (11, 17, 45). Based on our results, we propose a model (Figure 7) in which early in osteocyte differentiation, HDAC5 recruits the corepressors NCoR and HDAC3 to the SOST +45kB enhancer, thus suppressing SOST expression. As osteocyte differentiation occurs, HDAC5 occupancy is reduced, allowing increased binding of MEF2C along with the coactivator p300, which increases H3K27Ac and the ability of this enhancer to promote high levels of SOST expression. Future studies are required to elucidate the signals upstream of HDAC5 during osteocyte differentiation. Additionally, at this point we do not know if HDAC5 functions as a deacetylase or a scaffolding/adaptor protein in osteocytes.

Figure 7. Schematic representation of the role of HDAC5 in controlling SOST expression in osteocytes.

Early in Ocy454 cell differentiation, high levels of HDAC5 are bound to the +45 kB SOST enhancer, thereby blocking MEF2C occupancy and recruiting the co-repressors NCoR and HDAC3. Late in Ocy454 cell differentiation, HDAC5 occupancy decreases, thereby permitting increased MEF2C binding, p300 occupancy, and H3 K27 acetylation. See discussion for further details.

Multiple possible roles for HDAC5 in bone cells have been proposed (24, 52–55). Since HDAC5 is absent in all cells in the HDAC5 knockout mice, we cannot conclude that the skeletal phenotype we observe is solely due to direct actions of HDAC5 in osteocytes. Here, we report that mice lacking HDAC5 in all cells are osteopenic, with high SOST expression and reduced bone formation. In contrast, Obri et al (23) recently reported that HDAC5 functions in osteoblasts to control MEF2C-driven RANKL expression, with increased RANKL in HDAC5-KO mice causing osteopenia due to high osteoclastic activity. Here, we did not observe a role for HDAC5 in controlling RANKL levels in vitro (Figure S2a-c) or in vivo (Figure 4c). Furthermore, our histomorphometric analysis (in which TRAP-stained osteoclasts were scored in a blinded manner) revealed normal osteoclast numbers in the HDAC5 knockout mouse (Table 1 and Figure 4h). The same HDAC5-null strain (35) was employed for both the current study and the study of Obri et al.. Possibilities to explain these discordant results include differences in genetic background, gender (we analyzed females, while Obri et al analyzed males), or skeletal site (we analyzed tibia while Obri et al mainly focused on vertebra). We should note that transgenic mice overexpressing sclerostin in bone are not reported to have increased osteoclastic activity (45, 56). Proof that the osteopenia and reduced osteoblast activity is due in part to elevated sclerostin levels will require treatment of HDAC5-deficient mice with, for example, a neutralizing sclerostin antibody.

Sclerostin levels in peri-endosteal osteocytes are increased in the absence of HDAC5 (Figure 3c, e), and this correlates with reduced non-phospho beta-catenin staining in nearby endosteal osteoblasts (Figure 4a). However, we observe decreased trabecular bone volume fraction (Figure 4e-f) but normal cortical thickness (Figure S3). Quantification of sclerostin levels in cancellous osteocytes is more challenging than in cortical osteocytes, so at this time we cannot correlate this trabecular osteopenia with increased trabecular SOST levels. The absence of a cortical bone phenotype in 8-week-old mice is not entirely unexpected: transgenic mice overexpressing SOST in bone show trabecular osteopenia but normal cortical thickness (45, 56).

Class IIa HDAC function is controlled by nucleo-cytoplasmic shuttling through dynamic phosphorylation and 14-3-3 binding. In other cell types, N-terminal phosphorylation leads to 14-3-3 association and cytoplasmic retention (20, 44, 57). A number of extracellular signals regulate osteocytic SOST expression (39, 58). Recently, it was shown that PTH stimulates HDAC5 dephosphorylation, nuclear translocation, and SOST enhancer binding using overexpression in UMR106 cells (24). The functional significance of this pathway in osteocytes in vivo remains unknown. Inhibiting HDAC5 N-terminal kinases may be of therapeutic benefit to reduce SOST expression by osteocytes.

The accelerated expression pattern of sclerostin secretion in vitro in HDAC5 shRNA cells (Figure 2b) is consistent with our in vivo immunostaining results (Figure 3b). Sclerostin is considered a marker of “late”/”mature” osteocytes (59, 60), and it is possible that deeply embedded osteocytes are also more “mature” than their peri-endosteal counterparts. An alternative hypothesis is that differential mechanical forces or differences in the extracellular cytokine milieu on peri-endosteal versus deep osteocytes in vivo account for differences in sclerostin expression in these cells. We currently do not know if HDAC5 controls aspects of osteoblast to osteocyte conversion in vivo, as has been demonstrated for Gsα (29). While our data suggests a direct role for HDAC5 in regulating SOST expression, we cannot rule out indirect effects on osteocyte biology, illustrated by changes in DMP1 (Figure 2c) or PHEX (Figure 4a) seen in its absence.

GWAS data indicates that an intronic HDAC5 SNP (rs228769) is associated with variations in bone mineral density (BMD) (61). SNPs in MEF2C and SOST also control BMD (61, 62), suggesting the importance of the MEF2C/HDAC5/SOST pathway in humans. A cluster of BMD-regulating SNPs in SOST maps to the human orthologous region to the mouse +45kB enhancer (63). The effects of these sequence variants on MEF2C or HDAC5 binding remains unexplored. In conclusion, here we show that HDAC5, a gene associated with BMD variation in humans, is a direct negative regulator of MEF2C-driven sclerostin expression in osteocytes, both in vitro and in vivo.