Unveiling the signal valve specifically tuning the TGF-β1 suppression of osteogenesis: mediation through a SMAD1-SMAD2 complex

Abstract

Background

TGF-β1 is the most abundant cytokine in bone, in which it serves as a vital factor to interdict adipogenesis and osteogenesis of bone marrow-derived mesenchymal stem cells (BM-MSCs). However, how TGF-β1 concurrently manipulates differentiation into these two distinct lineages remains elusive.

Methods

Treatments with ligands or inhibitors followed by biochemical characterization, reporter assay, quantitative PCR and induced differentiation were applied to MSC line or primary BM-MSCs for signaling dissection. In vivo adipogenesis and ex vivo culture of bone explants were used to verify the functions of different SMAD complexes. Ingenuity Pathway Analysis, and analysis of transcriptomic datasets from human BM-MSCs in combination with hierarchical clustering and STRING assay were used to decipher the interplaying co-repressors. Mouse models of chronic and acute bone loss followed by biochemical assays and micro-computed tomography demonstrated the bone effects when functionally blocking the critical co-repressor HDAC1.

Results

Distinct from the TGF-β1 inhibition on adipogenesis through canonical SMAD2/3 signaling, we clarified that TGF-β1 suppresses osteogenesis by inducing the formation of previously unidentified mixed SMADs mainly composed of SMAD1 and SMAD2, in which SMAD2 recruits more TGF-β1-induced co-repressors including HDAC1, TGIF1 and ATF3, whereas SMAD1 allows directing the whole transcriptional suppression complex to the cis-elements of osteogenic genes. Depletion of the cross-activation to the mixed SMADs dismantled specifically the TGF-β1 suppression on osteogenesis without affecting its inhibition on adipogenesis. Such phenomena can be reproduced via knockdown of co-repressors such as Hdac1 or addition of HDAC1 inhibitors in TGF-β1-treated MSCs. In either the chronic or the acute bone loss model, we demonstrated that the TGF-β signaling was augmented in the bone niche during osteolysis, whereas administration of HDAC1 inhibitors significantly improved bone quality.

Conclusion

This study identifies a new signal valve through which TGF-β1 can inhibit osteogenesis specifically. Functional interruption of this valve can tilt the seesaw balance of BM-MSC differentiation towards osteogenesis, highlighting the interplaying co-repressors, such as HDAC1, as promising therapeutic targets to combat diverse degenerative orthopedic diseases.

Background

Bone marrow-derived mesenchymal stem cells (BM-MSCs) play the most critical role in maintaining postnatal bone homeostasis. They not only function as an osteoblast reservoir but also collaborate with derived osteogenic lineage cells to secrete multiple signaling factors that direct cell differentiation and subsequent osteogenic development [1, 2]. These factors are later immobilized in the bone matrix during mineralization. Intriguingly, as a common progenitor for osteoblasts and adipocytes, differentiation of BM-MSCs will lean towards adipogenesis in diverse degenerative orthopedic conditions. In patients suffering from post-traumatic bone atrophy, acute bone loss accompanied by expansion of marrow adiposity occurs in months [3, 4]. Accumulation of marrow fat is also influenced by age and sex. Specifically, the vertebral marrow fat content rises sharply in postmenopausal females, leading to around 10% higher than in males at comparable age [5]. In addition, such an imbalance in BM-MSC differentiation is also associated with a myriad of clinical conditions, including aging, diabetes, chronic renal failure and corticosteroid abuse, leading to severe bone loss and higher fracture rates [6,7,8]. As a consequence of abnormal osteolysis in these conditions, many osteogenic factors embedded in the bone matrix will be re-liberated. It thus is valuable to find signal valves from these niche factors that can reset the differentiation of BM-MSCs towards osteogenesis, thereby improving bone quality.

TGF-β is by far the most abundant cytokine in bone, with an estimated amount of 200 µg/kg [9, 10]. TGF-β belongs to the TGF-β superfamily, which binds to tetrameric receptor complexes formed by type-I and type-II serine/threonine kinase receptors to activate the SMAD signaling. Previous studies have demonstrated that TGF-β is a key regulator to control the maintenance of MSCs and also their lineage commitment towards osteogenesis and adipogenesis in the bone niche [11,12,13]. For example, as a mitogenic factor, TGF-β has been shown to promote the proliferation of MSCs via induction of SMAD3-dependent nuclear accumulation of β-catenin [14]. TGF-β further works in concert with bone morphogenic proteins (BMPs), another subfamily in the TGF-β superfamily, to show complex effects on osteogenesis of MSCs along with the differentiation stages. BMPs mainly activate the SMAD1/5/8 complex, which further induces and recruits RUNX2 to promote the expression of SP7 (Osterix) and other osteogenic genes for subsequent differentiation and maturation of osteoblasts [15, 16]. Thus, the BMP signaling is recognized to promote almost all osteogenic steps, including commitment, differentiation and maturation [17, 18]. Contrary to the above effects of BMPs, TGF-β only promotes proliferation, commitment and early differentiation of osteoprogenitors. It then switches to counteract the BMP signaling to inhibit later differentiation, maturation and subsequent matrix mineralization of osteoblasts [13, 17]. However, the molecular mechanisms within this signaling transition remain elusive.

As to adipogenesis, emerging evidence suggests that BMPs also play essential roles in promoting the adipogenic commitment of MSCs [19]. Although such an effect may vary along with different BMP ligands, BMP4 is currently considered as the most powerful BMP candidate that can consistently induce adipogenic commitment of diverse MSCs via activation of its canonical SMAD1/5/8 signaling [20, 21]. Contrary to BMP4, TGF-β effectively blocks adipogenesis both in vivo and in vitro. For example, loss of the TGF-β signaling in mouse MSCs via conditional deletion of Tgfbr2 during fetal development results in a marked expansion of adipocytes in bone marrows [22]. Consistently, TGF-β2 treatment effectively prevents bone loss and marrow adiposity in limb-suspended mice in vivo [23]. A following cell-based study demonstrates that inhibition of SMAD3 function, but not SMAD2 function, blocks the inhibitory effect of TGF-β1 on adipogenesis [24]. It currently has been proposed that SMAD3 can physically interact with C/EBP-β and C/EBP-δ; this then inhibits the expression of critical adipogenic factors, such as PPARγ [25].

From the above, it is concluded that BMPs tend to promote both osteogenesis and adipogenesis of BM-MSCs simultaneously, whereas this seesaw balance can be strictly guided by the TGF-β signaling. In this study, we found that TGF-β1 and activin A, both of which activate the canonical SMAD2/3 signaling, can block the BMP4-driven adipogenesis in MSCs, while only TGF-β1, but not activin A, blocks BMP4-mediated osteogenesis; this suggests that an unknown pathway other than the canonical SMAD2/3 signaling is involved in the TGF-β1 suppression of osteogenesis. Unlike activin A, we here reported that TGF-β1 triggers the formation of a previously unidentified mixed SMAD complex, which further recruits specific co-repressors for executing osteogenic suppression. Osteoporotic mouse studies further demonstrated that this new signal valve is tunable independently without affecting TGF-β1-mediated adipogenic suppression, highlighting its potential to treat degenerative orthopedic diseases.

Materials and methods

Animals and ethics

C57BL/6 mice were purchased from the animal center of National Yang Ming Chiao Tung University and BioLASCO. All animal experiments conformed to the Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of National Yang Ming Chiao Tung University (Permit Number: 1110216).

Recombinant proteins and chemicals

Recombinant BMP4 (314-BP), activin A (338-AC) and TGF-β1 (240-B) were obtained from R&D Systems. Recombinant insulin (91077 C) was obtained from Merck. 3-Isobutyl-1-methylxanthine (IBMX, BML-PD140) was obtained from Enzo. SB-431,542 (S4317), sodium valproate (VPA, P4543) and dorsomorphin (P5449) were obtained from Sigma-Aldrich. MS-275 (Entinostat; HY-12163) was obtained from MedChemExpress.

Cell culture

Human BM-MSCs were purchased from ScienCell (No.7500) and were maintained under the suggested conditions using a Mesenchymal Stem Cell Medium kit (ScienCell, 7501). Mouse BM-MSCs were isolated from 6 ~ 8-week-old male mice as described previously [26]. The following purification and culture of MSCs from the freshly isolated bone marrows were according to the guideline of MesenCult Expansion Kit (StemCell, 05513). For the cell line culture, C3H10T1/2, HEK293T and 3T3-L1 cells were originally sourced from the American Type Culture Collection and were maintained under handing information.

cDNA preparation, reverse transcription, and real-time PCR quantification

Total RNAs from harvested cells or bone tissues were extracted using TRIzol reagent (Invitrogen, 15596). RNAs were reverse-transcribed and the resulting cDNAs were analyzed by StepOnePlus PCR system using Power SYBR Green PCR Mix (Applied Biosystems, 4367659) and specific primer pairs. The relative expression level of target genes was normalized against Actb. The primers used for real-time qPCR were listed in Supplementary Table 1.

Lentivirus-mediated gene knockdown and cell line selection

For shRNA-mediated gene knockdown, C3H10T1/2 cells were infected with shRNA-containing lentivirus in growth medium with 0.1 mg/ml polybrene. Stable lines were selected using medium supplemented with puromycin (2 µg/ml). Knockdown efficiency was validated using real-time qPCR. The shRNA-expressing VSV-G pseudo-typed lentiviruses were purchased from RNA Technology Platform and Gene Manipulation Core from National Core Facility for Biopharmaceuticals, Academia Sinica. The list of shRNA clones was shown in Supplementary Table 2.

In vitro adipogenesis and osteogenesis

The adipogenesis protocol followed previous descriptions [27]. Briefly, C3H10T1/2 or primary MSCs were cultured in differentiation medium (complete medium/ 0.5 mM IBMX/ 0.25 µM dexamethasone/ 10 µg/ml insulin) for 2 days. Thereafter, the medium was replaced with complete medium supplemented with insulin (10 µg/ml) and refreshed bi-daily until day 12. For lipid staining, cells were fixed with 4% formaldehyde and then incubated with 60% Oil red O solution followed by washing with deionized water to remove unbound dye.

Osteogenesis protocol of C3H10T1/2 cells was developed as previously described [28]. Briefly, the cells were cultured in the osteogenic induction medium (DMEM/ 1% PSG/ 10% FBS with 10 nM dexamethasone) without or with the recombinant proteins or chemicals. The treatments were replaced every other day. On day 7 of post-induction, cells were fixed with 4% paraformaldehyde and then stained using BCIP/NBT solution (Merck, B6404) to reveal the alkaline phosphatase (ALP) activity. To evaluate mineralization capacity, mouse BM-MSCs with indicated treatments were cultured following the protocols of MesenCult Osteogenic Stimulatory Kit (StemCell, 05504). Calcium deposition was detected by staining with 2% Alizarin Red S solution.

In vivo adipogenesis assay

C3H10T1/2 cells were primed with recombinant proteins or chemicals for 2 days and then cultured with complete medium supplemented with 10 µg/ml insulin for 1 day. Approximately 2.5 × 10⁶ cells were collected, resuspended in 150 µl serum-free medium containing 50% of Matrigel (Corning, 356231), and then injected into the subcutaneous area of hindlimbs of 6–8-week-old C57BL/6 male mice. Mice were sacrificed after 9 days of post-injection, and the plugs were excised for cryosection.

Ex vivo bone culture assay

Culturing and measuring the long bones of fetal mice followed the method developed previously [29]. In brief, metatarsals and tibias from E16.5 embryonic mice were harvested and incubated in a 96-well plate with growth medium (MEMα/ 0.2% BSA/ 5 µg/ml ascorbic acid/ 1 mM β-glycerophosphate). The next day, the length of each metatarsal was measured as the start point. The growth medium with recombinant proteins and/or chemicals was refreshed every 3 days. Measurement of the bone length was performed on day 5 and day 9. RNA extraction and µCT analysis of bone tissues were performed on day 13. A detailed timeline was provided in the corresponding figure. The 3D images of tibia diaphysis were reconstructed using a SkyScan 1276 (Bruker, Kontich, Belgium). µCT analysis was done by Taiwan Mouse Clinic, Academia Sinica and Taiwan Animal Consortium.

Tissue processing and immunohistochemical staining

For paraffin sectioning, tibias from the ex vivo bone culture model were fixed, paraffin-embedded and sectioned at 4 μm for the following immunohistochemical staining of COL1A1 (Cell signaling, 72026). For the preparation of frozen sections, cell plugs from in vivo adipogenesis assays were fixed and then cryoprotected in a 10–30% sucrose gradient. Tissues were embedded in optimal cutting temperature compound (Tissue-Tek OCT; Sakura Finetek, Torrance, CA) and sectioned at 10 μm for the subsequent lipid staining using Oil red O.

Immunoblotting, immunoprecipitation (IP) and immunocytochemical (ICC) staining

To study the SMAD phosphorylation status, MSCs were starved for 4 h in serum-free medium before being treated with ligands for different time durations. Cell lysates were then subjected to Western blotting. The primary antibodies included antibodies recognizing phospho-SMAD1/5/8 (Cell signaling, 13820), phospho-SMAD2/3 (Cell signaling, 8828), total SMAD1/5/8 (Santa Cruz, sc-6031), total SMAD2/3 (Cell signaling, 8685), SMAD1(Santa Cruz, sc-7965), SMAD5 (GeneTex, GTX108971), SMAD2 (Cell signaling, 5339), SMAD3 (Cell signaling, 9523), HDAC1 (Santa Cruz, sc-81598), ATF3 (Santa Cruz, sc-518032), TGIF1 (Santa Cruz, sc-17800), Ac-H3 (Santa Cruz, sc-36616), and β-actin (ABclonal, AC038).

To perform IP against mixed SMAD complexes, the treated cells were lysed using Mammalian Protein Extraction Reagent (Thermo Scientific, 78501) with protease inhibitor (Roche, 11873580001) and phosphatase inhibitor (Sigma, 4906845001). Cell lysates containing 1000 µg of total proteins were incubated with the protein G-conjugated magnetic beads (cytiva, 28967070) and the appropriate primary antibody or IgG control (Cell signaling, 5415) at 4 °C overnight. The precipitated samples were subjected to Western blotting to detect the co-precipitated targets.

For ICC staining of SMADs, C3H10T1/2 cells starved for 4 h were treated with TGF-β1 and/or BMP4 for 30 min. The cells were fixed, permeabilized, and then incubated with the indicated primary antibody overnight at 4 °C followed by incubation with fluorescent secondary antibodies (Invitrogen, A21206). The cytoskeletons and nuclei were counterstained with phalloidin (Invitrogen, A12380) and DAPI, respectively. Visualization and image capture were accomplished using an Olympus BX-51 microscope.

Sandwich enzyme-linked immunosorbent assay (ELISA)

For establishing the sandwich ELISA to detect the mix-SMAD complexes, the SMAD1 antibody (Santa Cruz, sc-7965) was pre-coated on the 96-well microplates at a dilution ratio of 1:100. The plates were then blocked with 5% BSA and incubated with the lysates of treated C3H10T1/2 cells overnight at 4 °C. Following the wash, the SMAD2/3 antibody and the subsequent HRP-conjugated secondary antibody were applied. Binding signals were quantified using an ECL substrate solution.

Chromatin immunoprecipitation (ChIP) and DNA pull-down assays

ChIP assay against the SMAD1-binding cis-element was performed based on the instruction of enzymatic ChIP kit (Cell Signaling, 9003). Briefly, nuclei from the treated C3H10T1/2 cells were subjected to micrococcal nuclease digestion and lysed by sonication. The digested cross-linked chromatin was immunoprecipitated with SMAD2/3 antibody (Cell signaling, 8685) using Protein G-conjugated magnetic beads. The DNA fragments were de-crosslinked from complexes by incubating with proteinase K, and were purified to detect the presence of SMAD1-binding cis-element region in Id1 and Smad6 using real-time qPCR. Input DNA was used as a control for normalization.

DNA pull-down assay and the designed oligonucleotides corresponding to the wild-type and mutated BRE cis-element were adopted from the protocols described previously [30]. In brief, the nuclear extracts from treated C3H10T1/2 cells were incubated with designed biotinylated oligonucleotides and the streptavidin-coated magnetic beads (BioPioneer, MF-STA-3000) for 2 h at room temperature. The beads were then washed and the bound proteins were subjected to Western blotting to characterize the identities of binding complexes.

Luciferase reporter assay

C3H10T1/2 or HEK293T cells at 70–80% confluence were transfected with BRE-luciferase or CAGA-luciferase reporter along with pCMV-β-galactosidase vector at a 10 : 1 ratio for 16 h. The cells were then sub-cultured in serum-free medium and treated with ligands overnight followed by activity quantification. Luciferase readings were normalized to β-galactosidase activity.

Acute and chronic bone loss models

Seven-week-old C57BL/6 female mice were used to establish two types of bone loss models. The acute bone loss model involved the administration of soluble RANKL (sRANKL) protein, following the methodology described in a previous study [31]. Briefly, recombinant His₆-tagged soluble human RANKL (amino acid residues 140–317) expressed by the E. coli BL21 strain was purified. sRANKL at a dose of 2 mg/kg or PBS was injected intraperitoneally at 24-hour intervals for two days. The next day, MS-275 (10 mg/kg), VPA (300 mg/kg), or vehicle was injected intraperitoneally for 12 consecutive days. For the chronic osteoporosis model, the ovariectomy approach was applied as described previously [32]. Briefly, mice were ovariectomized bilaterally or underwent sham operation. Mice were allowed to recover for 3 days before chemical injection. MS-275 (10 mg/kg), VPA (300 mg/kg), or vehicle was injected intraperitoneally for 30 consecutive days. Mice were then euthanized, and the femurs and tibias were isolated for micro-computed tomography (µCT) scanning or bone marrow isolation. For µCT analysis, the measurement of trabecular BMD and 3D images of distal femoral metaphysis were reconstructed at 0.4–1.9 mm from the growth plate using CTAn software (µCT Systems Bruker, Kontich, Belgium).

Statistical analyses

In in vitro cell staining and chromogenic reaction assays, the percentages of stained areas were quantified using Image J software and are presented as means ± SD. For luciferase reporter assays, each treatment condition was performed in triplicate in a single experiment. All experiments for data collection were replicated independently at least three times and showed similar results. The statistical interpretation of luciferase reporter assay and gene quantification were shown as means ± SD. For ChIP and ELISA, the results were shown as means ± SEM. In the statistical analyses for comparison between the two groups, the unpaired Student’s t test was used. For comparison of multiple groups with a single variable, such as different treatment conditions, one-way ANOVA was used. For comparison of multiple groups with two independent variables, including gene silencing or inhibitor applications in combination with cytokine treatments, two-way ANOVA was used. The following post hoc analyses for multiple comparison were corrected by Dunnett posttest when comparing each mean with a control mean, or by Tukey posttest when comparing means with every other means. Bonferroni posttest was used where only two groups were compared (PRISM software version 8; GraphPad). For in vitro experimental data, only results with high statistical significance (P < 0.001) were marked unless otherwise emphasized.

Database and bioinformatic analyses

The transcriptomic datasets were retrieved from the Gene Expression Omnibus repository. GSE87487, comprising data from healthy human BM-MSCs, and GSE73175, containing data from BM-MSCs of 3 to 4-week-old C57BL/6 mice, were selected to represent the transcriptional abundance of receptors and SMADs in the TGF-β signaling module. Transcript quantification was normalized from reads per kilobase of transcript per million mapped reads or fragments per kilobase of transcript per million mapped fragments and then converted into transcripts per million (TPM) for standardizing the data representation. For visualizing the TGF-β-induced transcriptomic changes in MSCs, genes of interest in GSE46019 microarray dataset were extracted and normalized to percentage by defining the minimum and maximum RMA values as 0% and 100%, respectively. The expression pattern and hierarchical clustering were illustrated using MultiExperiment Viewer (MeV; https://doiorg.publicaciones.saludcastillayleon.es/10.1093/bioinformatics/btr490) [33]. Ingenuity Pathway Analysis (IPA) (2023 fall release; https://analysis.ingenuity.com/) was used to explore candidates and relationships of SMAD-related transcriptional factors via the ‘Grow’ tool within the ‘My Pathway’ section of the IPA database. The ‘Overlay’ function was subsequently employed to elucidate the biological roles of these candidates. The STRING database ( https://string-db.org/) [34] was utilized to explore the potential PPI networks among the selected candidates.

Results

TGF-β1 inhibits adipogenesis and osteogenesis via different signaling mechanisms

Among TGF-β superfamily ligands, TGF-β1 and activin A are not only abundantly expressed in bone marrows but also canonical ligands considered to activate the SMAD2/3 signaling [35, 36]. Their effects on adipogenesis and osteogenesis were therefore compared. In in vitro adipogenesis assays, both TGF-β1 and activin A significantly inhibited the BMP4-driven adipocyte formation and the expression of adipogenic-related genes in C3H10T1/2 MSC line (Fig. 1A-D). These inhibitory effects were consistently shown in 3T3-L1 preadipocytes (Supplementary Fig. 1). To our surprise, TGF-β1 and activin A behave differently in osteogenesis assays; only TGF-β1, but not activin A, can block the BMP4-induced activity of ALP, a primary indicator of osteogenic differentiation, and the expression of osteogenic-related genes in C3H10T1/2 (Fig. 1E-H). We then confirmed this phenomenon by monitoring the SMAD-mediated reporter response in different cells. In HEK293T, no signal crosstalk was observed between BMP4 and TGF-β1 or activin A, as neither TGF-β1 nor activin A affected the BMP4-driven SMAD1/5/8 signaling, and vice versa (Fig. 1I). By contrast, in C3H10T1/2, TGF-β1, but not activin A, significantly suppressed the BMP4-driven SMAD1/5/8 signaling while BMP4 showed no effect on the TGF-β1-driven or the activin A-driven SMAD2/3 signaling (Fig. 1J). Additionally, TGF-β1 not only inhibited BMP4-mediated SMAD1/5/8 activation in a dose-dependent manner but also dampened the basal SMAD1/5/8 activity as well as the transcripts of SMAD1/5/8 downstream genes in C3H10T1/2 (Supplementary Fig. 2A-C). Overexpression of SMAD4 did not eliminate the TGF-β1 effect on suppressing BMP4-mediated SMAD1/5/8 activation (Supplementary Fig. 2D and E), concluding that such inhibition is not due to competition for the common-mediator SMAD between these two canonical SMAD pathways. Taken together, as TGF-β1 and activin A behave similarly in blocking adipogenesis while only TGF-β1 suppresses osteogenesis, it is suggested that TGF-β1 mediates osteogenic suppression not solely via the SMAD2/3 signaling.

Fig. 1

TGF-β1 and activin A have distinct effects on osteogenesis of MSCs. A-D Oil red O staining (A) and quantification (B) of the lipid droplets as well as transcriptional quantification of the adipogenic markers, including Adipoq (C) and Plin1 (D), in C3H10T1/2 cells treated with TGF-β1 (1 nM), activin A (10 nM) and/or BMP4 (1 nM) during adipogenic induction. E-H Staining (E) and quantification (F) of the ALP activity as well as transcriptional quantification of the osteogenic markers, including Alpl (G) and Bglap (H), in C3H10T1/2 cells treated with TGF-β1 (1 nM), activin A (10 nM) and/or BMP4 (3 nM) during osteoblastogenic induction. I and J CAGA-driven and BRE-driven luciferase reporter assays in HEK293T cells (I) or C3H10T1/2 cells (J) treated with TGF-β1, activin A and/or BMP4 as indicated. The dose (nM) of each ligand is labeled behind. Results are shown as means ± SD. ***P < 0.001 compared with indicated control, one-way ANOVA followed by multiple comparison test. n = 3 independent experiments

TGF-β1 induces SMAD1/5/8 phosphorylation in diverse MSCs

TGF-β1 and activin A induced only SMAD2/3 phosphorylation in many common-used cell lines, such as HEK293T (Fig. 2A). To our surprise, while activin A exclusively induced SMAD2/3 phosphorylation, TGF-β1 was found to induce phosphorylation of both SMAD2/3 and SMAD1/5/8 simultaneously in diverse MSCs, including C3H10T1/2 (Fig. 2B), primary mouse BM-MSCs (Fig. 2C) and primary human BM-MSCs (Fig. 2D). Both TGF-β1-activated SMAD2/3 and SMAD1/5/8 can be translocated into the nuclei (Supplementary Fig. 3A and B), suggesting their subsequent gene targeting. Intriguingly, unlike the long-lasting characteristic of phosphorylated SMAD2/3 induced by TGF-β1, the induced SMAD1/5/8 phosphorylation was transient in MSCs (Fig. 2E-G). Such a profile is consistent with the expressional changes of TGF-β1-induced Id1, a direct target gene of the SMAD1/5/8 signaling, in which TGF-β1 initially augmented the expression, but the effect then turned to suppression over time (Fig. 2H). Furthermore, despite the fact that TGF-β1 strongly suppressed the BMP4-driven osteogenesis and SMAD1/5/8 signaling (Fig. 1), TGF-β1 seems only to decrease, but not completely eliminate, the amount of phosphorylated SMAD1/5/8 induced by BMP4 and did not affect its nuclear translocation in long-term treatments (Fig. 2I and J). Taken together, TGF-β1 triggers the phosphorylation of SMAD2/3 and SMAD1/5/8 simultaneously in MSCs, and the initiation of this process, but not the phosphorylated duration, would be required for the TGF-β1-mediated osteogenic suppression.

Fig. 2

TGF-β1 induces phosphorylation of both SMAD1/5/8 and SMAD2/3 simultaneously in diverse MSCs. A-D Western blotting of phosphorylated SMAD1/5/8 (pSMAD1/5/8) and phosphorylated SMAD2/3 (pSMAD2/3) in HEK293T cells (A), C3H10T1/2 cells (B), primary mouse bone marrow-derived MSCs (mBM-MSCs) (C), and primary human bone marrow-derived MSCs (hBM-MSCs) (D) treated with indicated doses of TGF-β1, BMP4 or activin A for 30 min. E-G Western blotting of the SMAD phosphorylation patterns over the indicated timeframe of TGF-β1 treatment in C3H10T1/2 cells (E), mBM-MSCs (F) and hBM-MSCs (G). H Real-time PCR quantification of Id1 transcript in C3H10T1/2 cells treated with TGF-β1 for different intervals. Results are shown as means ± SD. ***P < 0.001 compared with the no-treatment control, one-way ANOVA followed by multiple comparison test. n = 3 independent experiments. I and J The pSMAD1/5/8 was detected by Western blots (I) and ICC staining (J) in C3H10T1/2 cells treated with TGF-β1 (0.1 nM) and/or BMP4 (0.1 nM) for indicated time. Data are representative of three independent experiments

Molecular dissection of how TGF-β1 induces the formation of a mixed SMAD complex in MSCs

We noticed that co-treatment with activin A and BMP4, which also triggered phosphorylation of SMAD2/3 and SMAD1/5/8 in the same cells like TGF-β1 but just individually via different receptor modules, showed no osteogenic suppression effect (Figs. 1 and 2). In other words, we hypothesize that simultaneous phosphorylation of both SMAD populations by the same receptor module is a critical step to trigger the formation of a previously unidentified SMAD complex capable of exerting the osteogenic suppression function; this prompts us to characterize the underlying mechanism.

In terms of the receptor module, BMP4, activin A and TGF-β1 utilize distinct combinations of receptors [37]. In general, BMP4 uses the receptor complex mainly formed by BMPR2 of the type-II receptor and ALK3 or ALK6 of the type-I receptor; activin A preferentially binds to the receptor complex mainly formed by ACVR2A or ACVR2B of the type-II receptor and ALK4 of the type-I receptor. In contrast, TGF-β1 specifically binds to the type-II receptor TGFBRII; however, the flexibility in subsequent recruitment of different type-I receptors may alter the activated SMAD populations in diverse cells [38,39,40,41]. Interestingly, we found that treatment with SB-431,542 (an inhibitor for blocking SMAD2/3 phosphorylation by ALK4, ALK5 or ALK7), but not with dorsomorphin (an inhibitor for blocking SMAD1/5/8 phosphorylation by ALK1, ALK2, ALK3 or ALK6), abolished TGF-β1-mediated SMAD1/5/8 phosphorylation (Fig. 3A) as well as its inhibition on the BMP4 signaling (Fig. 3B), suggesting TGF-β1 triggers SMAD1/5/8 phosphorylation via one of the type-I receptors among ALK4, ALK5 and ALK7. By glancing through the relative expression levels of TGF-β1 superfamily receptors in C3H10T1/2 and primary BM-MSCs (Fig. 3C-E, Supplementary Fig. 4A and B), ALK5 was further narrowed down. Consistently, knockdown of Alk5, but not Alk3 (a type-I receptor for the BMP-induced SMAD1/5/8 phosphorylation), dampened the levels of both phosphorylated SMAD1/5/8 and phosphorylated SMAD2/3 induced by TGF-β1 (Fig. 3F and G, Supplementary Fig. 4C). Taken together, TGF-β1 can trigger phosphorylation of SMAD1/5/8 and SMAD2/3 simultaneously via a receptor complex formed by ALK5 and TGFBRII in MSCs.

Fig. 3

TGF-β1 induces the SMAD1/5/8 and SMAD2/3 signaling through the ALK5-containing receptor complex. A and B C3H10T1/2 cells were treated with ligand (TGF-β1 (0.1 nM) and/or BMP4 (0.1 nM)) in combination with inhibitors (SB-431542 (SB, 10 µM) or dorsomorphin (DM, 10 µM)) as indicated. The cells were subjected to Western blots (A) to evaluate changes in phosphorylated SMADs or BRE-driven luciferase reporter activity (B) to quantify SMAD1/5/8 response. Reporter activities are shown as means ± SD. ***P < 0.001 compared to the control within each group, two-way ANOVA followed by multiple comparison test. C-E Real-time PCR quantification of the TGF-β signaling-related receptor transcripts in C3H10T1/2 cells (C), mBM-MSCs (D) and hBM-MSCs (E). F and G The gene encoding Alk3 or Alk5 in C3H10T1/2 cells was knocked down by corresponding shRNA. Knockdown efficiency was confirmed by real-time PCR quantification (F). Data are means ± SD. ***P < 0.001 compared to the infection control, students’ t test. The cells with indicated gene knockdown were treated with TGF-β1 and then subjected to immunoblotting to evaluate changes in pSMAD1/5/8 and pSMAD2/3 levels (G). Blots are representative of three independent experiments

We next validated the precise receptor-regulated SMADs (R-SMADs) phosphorylated by this receptor module in MSCs. RNA quantification and transcriptomic analysis suggest that the transcripts of R-SMAD genes except Smad8 have relative abundance in MSCs (Fig. 4A-C, Supplementary Fig. 4D and E). In the SMAD1/5/8 cohort, Smad1 knockdown eliminated most of the TGF-β1-induced SMAD1/5/8 phosphorylation, while Smad5 knockdown only reduced part of it (Fig. 4D, Supplementary Fig. 5A), suggesting SMAD1 is preferentially recruited. In the SMAD2/3 cohort, we surprisingly found that knockdown of Smad2, but not Smad3, not only reversed the TGF-β1 suppression of basal SMAD1/5/8 activity but also significantly reduced the TGF-β1’s inhibitory effect on the BMP4-mediated SMAD1/5/8 activation (Fig. 4E, Supplementary Fig. 5B). As mentioned in the hypothesis above, these data further conclude that a mixed SMAD complex composed of mainly SMAD1 and SMAD2 is formed under TGF-β1 induction. Using SMAD1 antibody as a bait, formation of the mixed SMAD1-SMAD2 complex can be evidenced by co-IP and ELISA in MSCs treated with TGF-β1 (Fig. 4F and G, Supplementary Fig. 5C). A reverse experiment using ChIP-qPCR also confirmed the occupancy of SMAD2 on the SMAD1-targeted cis-elements of Id1 and Smad6 genes (Fig. 4H and I). Consistently, either Smad1 knockdown or Smad2 knockdown can effectively reverse the TGF-β1 effect on suppressing BMP4-augmented Id1 expression in MSCs (Fig. 4J, Supplementary Fig. 5D-F). To sum up, these data lead to a prototypic model that TGF-β1 triggers simultaneous phosphorylation of SMAD1 and SMAD2, and their subsequent interaction to form a mixed SMAD complex via the receptor complex composed of TGFBRII and ALK5 in MSCs. Such a mixed SMAD complex may be required for the TGF-β1 signaling to suppress osteogenesis.

Fig. 4

TGF-β1 induces the formation of mixed SMAD1-SMAD2 complex in MSCs. A-C Real-time PCR quantification of the R-SMAD transcripts in C3H10T1/2 cells (A), mBM-MSCs (B), and hBM-MSCs (C). D and E C3H10T1/2 cells with individual knockdown of indicated Smad gene were treated with TGF-β1 (0.1 nM) and/or BMP4 (0.1 nM). Western blotting (D) was used to evaluate changes in the phosphorylated SMAD levels when Smad1 or Smad5 was knocked down. BRE-driven luciferase reporter assay (E) was used to evaluate the reversion of TGF-β1-mediated suppression on the BMP activity when Smad2 or Smad3 was knocked down. Reporter activities are shown as means ± SD. F and G Detection of the mixed SMAD complex in MSCs treated with TGF-β1 (0.1 nM) and/or BMP4 (0.1 nM) for 30 min. (F) IP was performed using anti-SMAD1 antibody or IgG control. The inputs and resulting immunoprecipitates were blotted against SMAD2/3 or SMAD1. G Plates coated with anti-SMAD1 antibody were used to capture SMAD1-containing complexes from cell lysates. Anti-SMAD2/3 antibody was then used to detect the mixed SMAD1-SMAD2 complex. Data are means ± SEM, one-way ANOVA followed by multiple comparison test. n = 4 independent experiments. H and I ChIP assay. C3H10T1/2 cells treated with TGF-β1 (0.1 nM) and/or BMP4 (0.1 nM) for 24 h were lysed and subjected to the ChIP assay using anti-SMAD2/3 antibody or control IgG as indicated. Real-time PCR was then used to quantify the SMAD1-binding cis-element of Id1 (H) and Smad6 (I) genes. Data are means ± SD. ***P < 0.001 compared to no treatment control, one-way ANOVA followed by multiple comparison test. n = 3 independent experiments. J C3H10T1/2 cells with knockdown of Luc control, Smad1 or Smad2 were treated with ligands as indicated for 48 h. Id1 transcripts were quantified by real-time PCR. Data are normalized with Actb and expressed as means ± SD. For statistical analysis, ***P < 0.001 is compared to shLuc-infected control within each group and determined by two-way ANOVA followed by multiple comparison test; ###P < 0.001 is compared to no treatment control and determined by one-way ANOVA with multiple comparison test. n = 3 independent experiments

The canonical SMAD2/3 complex and the novel mixed SMAD complex induced by TGF-β1 play distinct roles in controlling MSC differentiation

In our model, although both SMAD1 phosphorylation and SMAD2 phosphorylation driven by TGF-β1 are conducted by the same receptor complex in MSCs, we surprisingly found that these two events showed differential sensitivities to the ALK5 inhibitor. High-dose SB-431,542 at 10 µM blocked both phosphorylation of SMAD1/5/8 and SMAD2/3 in TGF-β1-treated MSCs; however, low-dose SB-431,542 at 1 µM inhibited most of the SMAD1/5/8 phosphorylation, whilst having less effect on SMAD2/3 phosphorylation (Fig. 5A, Supplementary Fig. 6A). We thus used this strategy to discern the impacts of these two signals during adipogenic or osteogenic differentiation. In vitro MSC differentiation assays further confirmed that low-dose SB-431,542 treatment still retained the inhibitory ability of TGF-β1 on the BMP4-induced adipogenesis (Fig. 5B), but indeed released that on the BMP4-mediated ALP activity and mineralization, as well as that on the BMP4-induced osteogenic master gene expression (Fig. 5C, Supplementary Fig. 6B-F).

Fig. 5

The TGF-β1-driven mixed SMAD signaling inhibits osteogenesis, but not adipogenesis, of MSCs. A Western blots to evaluate the dose effects of SB on TGF-β1-induced pSMAD1/5/8 and pSMAD2/3 in C3H10T1/2 cells. Blots are representative of three independent experiments. B and C In vitro differentiation assays. C3H10T1/2 cells were primed with either low-dose or high-dose SB followed by treatment with TGF-β1 (1 nM) and/or BMP4 (1 nM) under adipogenic or osteogenic condition. (B) Oil red O staining (upper panel) and stained area quantification (low panel) were used to evaluate adipogenic capacity. C BCIP/NBT staining (upper panel) and intensity quantification (lower panel) were used to evaluate osteogenic capacity. n = 3 independent experiments. D In vivo adipogenesis assay. C3H10T1/2 cells primed with TGF-β1 (1 nM), BMP4 (1 nM) and/or SB (low or high dose) as indicated were used to perform ectopic adipocyte formation in mice. Cell plugs (yellow circle) were dissected, frozen-sectioned, and stained with Oil red O and hematoxylin. Scale bars, 100 μm. E-G Ex vivo osteogenesis assay. The foot bones from E16.5 mice were cultured in osteogenic medium supplemented with low-dose SB, TGF-β1 (1 nM) and/or BMP4 (1 nM). E The increased length of metatarsals was measured at indicated days. For more complete presentation, sample distributions in the TGF-β1 group (box a) and the TGF-β1 plus BMP4 group (box b) were placed independently on the right panels. Each point represented a single metatarsal. µCT images of the treated tibiae (F) were analyzed and a representative bone parameter (G) were shown. BV/TV: bone volume /total volume. n = 4 mice in each group. Data are means ± SD. For statistical analysis, ***P < 0.001 is compared to the control in the same group and determined by two-way ANOVA followed by multiple comparison test; ###P < 0.001 is determined by one-way ANOVA with multiple comparison test

We further used animal models to verify functional differences between the TGF-β1-induced SMAD2/3 complex and the mixed SMAD1-SMAD2 complex. In the in vivo ectopic adipocyte formation model using ligand-primed C3H10T1/2 cells, TGF-β1 can suppress the BMP4-driven adipocyte formation, and such an effect was reversed by high-dose SB-431,542, but not by low-dose SB-431,542 (Fig. 5D). By contrast, in the bone formation model using ex vivo cultures of mouse metatarsal explants, TGF-β1 not only inhibited longitudinal growth but also hindered BMP4 promotion on this event. Addition of low-dose SB-431,542 showed negligible effect on the control or the BMP4 treated group; it however totally reversed the inhibitory effects of TGF-β1 (Fig. 5E). Transcriptional quantification of osteogenic markers in cultured foot bones, which include tarsals, metatarsals and phalanges, further corroborated these phenomena (Supplementary Fig. 7A and B). In cultured tibiae, µCT and immunohistochemical analysis revealed that low-dose SB-431,542 mitigated the TGF-β1 suppression of BMP4-promoted mineralization as showing the rescued effects on diverse bone parameters (Fig. 5F and G, Supplementary Fig. 7C-F) and the COL1A1 protein staining (Supplementary Fig. 7G). In summary of these results, activation of different R-SMAD complexes seems to serve as a signaling watershed to distinguish the TGF-β1 inhibition of MSC differentiation towards adipogenesis and osteogenesis: activation of the canonical SMAD2/3 complex, which can be totally blocked by high-dose SB-431,542, can inhibit adipogenesis, whereas induced formation of the novel mixed SMAD1-SMAD2 complex, which is selectively blocked by low-dose SB-431,542, is in charge of suppressing osteogenesis.

Formation of a transcriptional suppression complex via the TGF-β1-induced mixed SMADs

The results shown in Fig. 4 suggest that certain co-repressors should be recruited into the TGF-β1-induced mixed SMAD complex to exert transcriptional repression. Intriguingly, previous studies suggest that SMAD-interactive co-repressors are preferentially recruited by SMAD2/3, but not SMAD1/5/8 [42, 43]. To decipher the potential co-repressors, we first used in silico IPA to sort out 179 SMAD2-interacting transcriptional regulators; these regulators were further narrowed down into 21 candidates by overlaying with the filter of transcriptional repression and the functional filters of MSC differentiation, including differentiation of adipocytes and differentiation of osteoblasts (Fig. 6A, Supplementary Fig. 8). Given that TGF-β1 seems to convert its original promotion effect on osteogenic gene expression into inhibition over time (Fig. 2H), we postulated that the critical co-repressors may be induced after TGF-β1 treatment. Thus, we performed hierarchical clustering on the selected 21 candidates in GSE46019, a transcriptomic dataset derived from human BM-MSCs stimulated with TGF-β1 for different intervals. This analysis revealed 7 candidates up-regulated by TGF-β1, including MECOM, HDAC1, ATF3, ZNF423, NCOR2, TGIF1, and SNAI1 (Fig. 6B). We further postulated that the critical co-repressors would interact to form a complex with the mixed SMADs. Among these candidates, STRING analysis in both human and mouse databases positioned HDAC1 as a central node in the network with confident scores higher than 0.5 to interact with TGIF1, NCOR2 and ATF3 (Fig. 6C and D). In C3H10T1/2 cells, knockdown of Hdac1, Tgif1 or Atf3, but not Ncor2, significantly abrogated TGF-β1’s suppressive effect on the BMP4-induced reporter activity and Id1 expression (Fig. 6E-I, Supplementary Fig. 9). Pull-down assay using a DNA probe containing the BMP-responsive element further suggests that HDAC1, TGIF1, and ATF3 can be recruited by the TGF-β1-induced mixed SMAD1-SMAD2 to form a novel transcriptional suppression complex in MSCs (Fig. 6J).

Fig. 6

The TGF-β1-induced mixed SMADs further form a transcriptional suppression complex. A Illustration of the correlations of SMAD2-interacting transcriptional regulators with osteogenesis and adipogenesis in the IPA database. Factors related to [Repression of RNA] were highlighted in purple. Factors related to [Differentiation of osteogenesis] process only were marked with a blue frame, while factors related to [Differentiation of adipogenesis] process only were marked with an orange frame. Co-repressors and co-activators involved in both processes are listed in the purple box and the gray box, respectively. B Hierarchical clustering of the temporal expression patterns of selected SMAD2-binding co-repressors in TGF-β1-treated hBM-MSCs. C and D STRING diagrams depicted the PPI networks between the seven TGF-β1-induced co-repressors in human (C) and mouse (D) models. Interaction confidence scores were shown, with 1 being the highest score. Significance of the PPI enrichment p-value suggests that these proteins are more interconnected than expected by chance. E-I Quantification of the BRE-driven luciferase reporter activity in C3H10T1/2 cells infected with lentivirus-packaged shLuc (E), shHdac1 (F), shTgif1 (G), shAtf3 (H), or shNcor2 (I). The cells were treated with TGF-β1 (0.1 nM) and/or BMP4 (0.1 nM) as indicated. Data are means ± SD. ***P < 0.001 compared to the indicated sample, Student’s t test. n = 3 independent experiments. J The nuclear extracts from C3H10T1/2 cells with indicated treatments for 24 h were subjected to DNA pull-down assays using the BRE-containing or mutant probes. The presence of corresponding co-repressors was confirmed by Western blot. Blots are representative of three independent experiments. K Schematic diagram of the TGF-β1 signaling in controlling MSC differentiation. TGF-β1 activates two distinct branches of SMAD signals in MSCs via the receptor complex formed by TGFBR2 and ALK5. The signaling of canonical SMAD2/3 branch counteracts the BMP-driven adipogenesis. Parts of the induced co-repressors, such as HDAC1, TGIF1 and ATF3, can couple with the novel branch of SMAD1-SMAD2 complex to suppress BMP-driven osteogenesis. Functional disruption of this novel branch is predicted to tune the osteogenic suppression of TGF-β1 down specifically

Taken together, a proposed model suggests that TGF-β1 inhibition of BM-MSC differentiation can be divided into two distinct valves: signaling through SMAD2/3 inhibits adipogenesis, whereas that through a newly identified SMAD1-SMAD2 complex, which subsequently recruits specific transcriptional co-repressors, suppresses osteogenesis (Fig. 6K). Since matrix-embedded TGF-β1 will be rapidly re-liberated during osteolysis, it is predicted that functional interruption of its osteogenic suppression valve would tilt the seesaw balance of BM-MSC differentiation towards osteogenesis, thus combating bone degeneration.

Functional interruption of the HDAC1 activity specifically dismantles the TGF-β1 suppression on osteogenesis without affecting its inhibition on adipogenesis

From the above analyses, we concluded the TGF-β1-induced osteogenic suppression complex contains at least HDAC1, TGIF1, and ATF3. Thus, knockdown strategy was first used to evaluate their influences on MSC differentiation. Contrary to not affecting TGF-β1-mediated inhibition of adipogenesis (Supplementary Fig. 10A and B), knockdown of Hdac1 significantly reversed the TGF-β1 effects on inhibiting osteogenesis alone and on suppressing BMP4-promoted osteogenesis (Fig. 7A). Similar results in reversing the TGF-β1-mediated osteogenic suppression can also be observed in C3H10T1/2 cells with knockdown of Tgif1 or Atf3 (Fig. 7B), suggesting all three factors are required in order to form a complete osteogenic suppression complex. HDAC1 is predicted as a central node to interact with other co-repressors and SMAD2. Given the availability of clinical-approved HDAC inhibitors, we targeted HDAC1 to evaluate its ability in manipulating TGF-β1-mediated osteogenic suppression in vivo by using VPA and MS-275 [44, 45], the inhibitors with high inhibitory preferences to HDAC1. Their given doses were optimized beforehand via detecting the acetylated histone H3 levels in MSCs (Supplementary Fig. 10C and D). Consistent with the phenomena shown in Hdac1-knockdown MSCs, in vitro differentiation assays demonstrated that both HDAC inhibitors indeed liberated the TGF-β1’s suppressive effect on osteogenesis (Fig. 7C), but did not alter its inhibitory effect on adipogenesis (Supplementary Fig. 10E and F). Such specificity underscores their potential in shifting the seesaw balance of MSC differentiation towards osteogenesis in vivo. Thus, their effects were further evaluated in osteoporosis mouse models.

Fig. 7

Inhibition of HDAC1 reverses the TGF-β1 suppression on osteogenesisin vitroand rescues the osteoporotic phenotypein vivo. A - C In vitro osteogenesis. C3H10T1/2 cells with knockdown of Hdac1 (A), Atf3 or Tgif1 (B), or with HDAC inhibitor supplements (C) were treated with TGF-β1 (1 nM) and/or BMP4 (3 nM) during osteoblastogenic induction. The ALP activity was revealed by representative staining images or by intensity quantification. n = 3 independent experiments. D and E In vivo chronic osteoporosis model. Mice undergoing ovariectomy (OVX)-induced osteoporosis were used for bone marrow isolation or subsequent purification of BM-MSCs. Bone marrows (D) or the derived BM-MSCs treated with TGF-β1 (1 nM) for 30 min (E) were subjected to Western blotting to detect the indicated factors. (F - H)In vivo acute bone loss model. Mice undergoing sRANKL-induced bone loss were further injected without or with different HDAC inhibitors as indicated. Bone marrows (F) were isolated and then subjected to Western blotting to detect the indicated factors. Distal femurs from mice were subjected to µCT analysis. Representative images (G) and derived bone parameters (H), including BV/TV, Tb.Sp, Tb.Th, Tb.N and bone mineral density (BMD), were shown and compared. ‘C’ denotes the vehicle control for drugs. Data are means ± SD. *P < 0.05, **P < 0.01, ***P < 0.001, compared to indicated sample. One-way ANOVA followed by multiple comparison test

We demonstrated that the ovariectomized mouse model, aiming to imitate the chronic bone loss condition, can successfully create a TGF-β1-enriched bone marrow niche (Fig. 7D), potentially via accelerating the release of matrix-embedded TGF-β1. Concurrently, increases in the phosphorylation levels of both SMAD1/5/8 and SMAD2/3 were observed (Fig. 7D). Primary BM-MSCs isolated from ovariectomized mice were more prone to forming TGF-β1-induced osteogenic suppression complex as reflected by that higher amounts of phosphorylated SMAD1/5/8, phosphorylated SMAD2/3 and identified co-repressors were detected upon TGF-β1 stimulation (Fig. 7E); this may explain the accelerated bone loss in ovariectomized mice. Intriguingly, similar profiles, such as increases in the TGF-β1 amounts and the phosphorylation levels of both SMADs in bone marrows, can also be seen in a rapid bone loss model by administration of sRANKL protein (left panel of Fig. 7F). Transcriptional increases in Tgfbr1 and Tgfbr2 in the derived BM-MSCs further indicated that these MSCs are more sensitive to TGF-β1 (Supplementary Fig. 10G and H). These data suggest a successful mimicry of TGF-β1-activated dual SMAD signals in bone marrows of mice with bone degeneration. In sRANKL-injected mice, we further incorporated administration of HDAC inhibitor, either VPA or MS-275. No significant histopathological abnormality was observed in the main organs such as the liver, kidney and heart, suggesting a lower systemic toxicity risk under our drug administration conditions (Supplementary Fig. 11A). Of interest, although administration of HDAC1 inhibitor seemed not to affect the intensity of TGF-β1-mediated SMAD activation in bone marrows (right panel of Fig. 7F), it greatly ameliorated bone deterioration (Fig. 7G, Supplementary Fig. 11B), as also evidenced by improvements in many bone indices, including bone volume, trabecular separation, trabecular number, and bone mineral density (Fig. 7H). Specifically, MS-275 would exert a more pronounced effect than VPA.

Discussion

Distinct from the TGF-β1 inhibition on adipogenesis through canonical SMAD2/3 signaling, we here unveiled that TGF-β1 induces the formation of a previously unidentified SMAD1-SMAD2 complex as a new signal valve to suppress osteogenesis. We pinpointed HDAC1 as a central node of co-repressors recruited to this valve. Furthermore, we here adopted two mouse models to mimic different situations of bone loss, wherein ovariectomy is a preclinical model for studying the chronic bone loss in postmenopausal women [46], and sRANKL injection is to imitate the acute bone loss likely occurring in some disordered conditions, such as cancer-induced hypercalcemia and post-traumatic bone atrophy [3, 47]. We demonstrated that these osteoporotic mouse models successfully create a TGF-β1-abundant bone niche, wherein functional interruption of HDAC1 via inhibitor administration can reset the TGF-β1 signaling towards favoring osteogenesis to improve bone quality; this highlights HDAC1 as a promising therapeutic target to combat diverse degenerative orthopedic diseases.

Interestingly, several studies have proposed that certain HDACs, especially some members in the class I and class IIa of HDACs, are involved in bone development [48]. For example, in the class I HDACs, it has been elucidated that HDAC1, but not HDAC2, HDAC3 or HDAC8, participates in osteoblast differentiation via being directly recruited to the promoters of osteoblast marker genes, such as RUNX2 and SP7 [49, 50]. Such occupancy is inversely correlated with occupancy of p300 during osteogenic induction, implying the negative regulation role of HDAC1 during bone formation. Conversely, genetic ablation models of some members in the class IIa HDACs, such as osteoblast-specific deletion of Hdac4 or systemic knockout of Hdac5 or Hdac9 in mice, show osteopenic phenotypes generally [51,52,53], suggesting that these HDACs by contrast act on the promotion of osteogenesis. Taken together, HDACs can have opposite effects on bone formation depending on their types; however, as our findings suggest, HDAC1 does play a significant role in inhibiting osteogenesis. Moreover, HDAC1 has also been reported to act as a critical co-repressor in chondrogenesis, potentially via suppressing the expression of cartilage-specific genes such as COMP and COL2A1 [54, 55], or via weakening Nkx3.2 protein stability [56]. In other words, inhibition of HDAC1 activity in our findings not only can promote osteogenesis, but may also benefit chondrogenesis.

Our data further revealed that at least TGIF1 and ATF3 are also indispensable and should work synergistically with HDAC1 in augmenting the TGF-β1 suppression of osteogenesis. To our knowledge, no direct interaction between HDACs and SMADs has been proven. It thus is likely explained by that these co-repressors potentially serve as interaction mediators for HDAC1 and SMADs. Via co-IP assays, it is indeed demonstrated that the interaction between SMAD2 and HDAC1 is dependent on the presence of TGIF1 [57]. Further studies reveal that TGIF1 binds HDAC1 through a specific domain, and this forming repression complex reduces the likelihood of SMAD-co-activator complex formation via competition with p300 for binding to SMAD2 [57, 58]. In the case of ATF3, it has been shown that the basic region (residues 80 to 116) on ATF3 is required for ATF3 to be recruited into the SMAD-mediated repression complex since ATF3 lacking this region will fail to interact with both HDAC1 and also SMAD3 simultaneously in cells as shown by co-IP assays [59, 60]. As evidence of direct binding with HDAC1 or SMAD3 is lacking, the sequential binding order of ATF3 to these proteins still remains elusive. Furthermore, since no specific inhibitor against TGIF1 or ATF3 has been developed, it also cannot exclude whether these co-repressors can contribute directly to osteogenic suppression in addition to acting as mediators for HDAC1 recruitment.

While TGF-β1 is widely recognized as an inhibitory factor in bone formation, we also noticed that its effects in different stages of animals and MSCs are contradictory. Conditional deletion of Tgfbr2 or Tgfbr1 (Alk5) in the undifferentiated MSCs of mouse embryos results in cartilage malformation and short limbs [61, 62], demonstrating that the TGF-β signaling can promote skeletal development during early development. It is contrary to the fact that neutralizing TGF-β in adult mice or patients with osteogenesis imperfecta syndromes greatly improves bone mass as well as bone biomechanical properties [63, 64]. As to osteogenic progression, the TGF-β signaling has been known to promote proliferation, homing, as well as osteogenic commitment of BM-MSCs; it then turns into inhibition of maturation and osteogenic functions of osteoblasts and osteocytes [17]. These phenotypic shifts can likely be explained through our model that the expression and/or the activity of the key repressors may vary across these periods. They may be absent or functionally inhibited during embryonic development or osteogenic commitment of MSCs, whereby the TGF-β-induced mixed SMAD complex acts as a promoting role in transcription; conversely, when they are expressed and/or activated in adulthood or in the later stages of osteogenesis, they then bind and transform the TGF-β-induced mixed SMADs into a transcriptional suppression complex. Thus, profiling these key repressors may serve as a prognosis index or provide therapeutic clues for TGF-β-related bone diseases.

Our case in MSCs indicates that whether appropriate co-repressors are functionally presented in the mixed SMAD complex or not determines the switch of TGF-β1’s functions between suppressing and promoting bone formation. By the same token, if TGF-β1 can induce the mixed SMAD complex in other types of cells, it may execute certain cell-specific functions via recruiting different regulators unique to those cells. Such a hypothesis may potentially be applied to a few cases. For example, a study demonstrates that TGF-β1 exhibits biphasic effects on osteoclastogenesis through different SMADs. It decreases the expression of osteoclast-specific genes and inhibits the osteoclast function potentially through SMAD1-mediated signaling; however, it stimulates osteoclast formation through activation of SMAD3 [65], which then binds to the TRAF6-Table 1-TAK1 complex and reciprocal cooperation with c-Fos to promote the expression of key regulators, such as NFATC1, required for osteoclastogenesis [66]. These findings highlight the importance of composition of SMADs as well as their associated co-factors in determining the functional outcomes of TGF-β1 signaling. Moreover, a study in endothelial cells found that the SMAD1/5 signaling induced by TGF-β3 can promote cell proliferation and migration [40]. Other studies using epithelial cells, fibroblasts or cancer cell lines as subjects suggest that the TGF-β-induced SMAD1/5 signaling can be linked to the promotion of anchorage-independent growth and epithelial-to-mesenchymal transition [38, 39, 67]. However, more issues in these conditions need to be clarified, such as whether mixed SMADs are involved, their identities, and the regulators they recruit. This further highlights that the TGF-β-mediated signaling is far more complex than what is currently known.

Conclusions

In summary, the novelty of this study lies in the first discovery of the signal valve through which TGF-β1 inhibits osteogenesis, as well as the elucidation of its potential components for exerting transcriptional suppression (Fig. 6K). The most exhilarating finding is that functional interruption of the critical co-repressor HDAC1 in this valve can selectively dismantle the TGF-β1 suppression on osteogenesis without affecting its inhibition on adipogenesis. In addition to HDAC1, we further identified at least TGIF1 and ATF3 as co-repressors, providing a new biomarker signature capable of tuning this valve for preventing bone loss. Given that TGF-β1 is the most abundant cytokine embedded in the bone matrix and can be quickly re-liberated to the bone niche during osteolysis, our concept that resets the differentiation of BM-MSCs towards osteogenesis via tuning this newly identified signal valve of TGF-β1 will provide a brand-new thinking when developing approaches to treat patients with diverse degenerative orthopedic conditions.