SMAD3 and SMAD4 have a more dominant role than SMAD2 in TGFβ-induced chondrogenic differentiation of bone marrow-derived mesenchymal stem cells

To improve cartilage formation by bone marrow-derived mesenchymal stem cells (BMSCs), the signaling mechanism governing chondrogenic differentiation requires better understanding. We previously showed that the transforming growth factor-β (TGFβ) receptor ALK5 is crucial for chondrogenesis induced by TGFβ. ALK5 phosphorylates SMAD2 and SMAD3 proteins, which then form complexes with SMAD4 to regulate gene transcription. By modulating the expression of SMAD2, SMAD3 and SMAD4 in human BMSCs, we investigated their role in TGFβ-induced chondrogenesis. Activation of TGFβ signaling, represented by SMAD2 phosphorylation, was decreased by SMAD2 knockdown and highly increased by SMAD2 overexpression. Moreover, TGFβ signaling via the alternative SMAD1/5/9 pathway was strongly decreased by SMAD4 knockdown. TGFβ-induced chondrogenesis of human BMSCs was strongly inhibited by SMAD4 knockdown and only mildly inhibited by SMAD2 knockdown. Remarkably, both knockdown and overexpression of SMAD3 blocked chondrogenic differentiation. Chondrogenesis appears to rely on a delicate balance in the amount of SMAD3 and SMAD4 as it was not enhanced by SMAD4 overexpression and was inhibited by SMAD3 overexpression. Furthermore, this study reveals that TGFβ-activated phosphorylation of SMAD2 and SMAD1/5/9 depends on the abundance of SMAD4. Overall, our findings suggest a more dominant role for SMAD3 and SMAD4 than SMAD2 in TGFβ-induced chondrogenesis of human BMSCs.

Scientific RepoRts | 7:43164 | DOI: 10.1038/srep43164 ALK5 can activate two R-SMADs, SMAD2 and SMAD3, which share similar structures including a Mad-Homology 1 (MH1) domain, linker region and MH2 domain 9,11 . Despite their similarities, SMAD2 and SMAD3 have a differential role in TGFβ signal transduction [12][13][14][15] . Whereas Smad2 knockout mice die during gastrulation 16,17 , Smad3 knockout mice develop cartilage and survive for several months after birth [18][19][20] . Moreover, SMAD3, but not SMAD2, is involved in enhanced transcriptional activity of SRY (Sex Determining Region Y)-Box 9 (SOX9); a master regulator of chondrogenesis, in human mesenchymal stem cells 21,22 . In contrast to SMAD3, SMAD2 lacks the MH1 domain. Therefore, SMAD2 cannot bind DNA without complex formation with SMAD4, which may explain the differential effects of SMAD2 and SMAD3 [23][24][25][26][27] . However, whether SMAD2 and SMAD3 have a different function during TGFβ -induced chondrogenic differentiation of human BMSCs remains largely unknown. Also the specific role of SMAD4 in chondrogenesis of human BMSCs has not been investigated. In mice, deletion of Smad4 causes early embryonic death 28,29 . Therefore, tissue-specific Smad4 knockout mice have been generated. Conditional deletion of Smad4 in the limb bud mesenchyme of mice leads to an absence of cartilage elements prefiguring the limb skeleton 30,31 , indicating that SMAD4 is required for cartilage formation. Furthermore, chondrocyte-specific Smad4 knockout mice exhibit dwarfism and impaired growth plate organization 32 . Similarly, Smad3 knockout mice have forelimb malformations, are smaller than wild-type mice and they develop spontaneous joint degeneration resembling the degenerative joint disease osteoarthritis in humans 18,19 . In humans, mutations in SMAD3 lead to skeletal anomalies and early-onset of osteoarthritis [33][34][35][36] . Moreover, SMAD3 is required for TGFβ -mediated suppression of hypertrophic differentiation of human articular chondrocytes 37,38 . Although SMAD3 does not appear directly required for embryonic cartilage and joint development, SMAD3 seems to be important for maintaining a stable cartilage phenotype by preventing cartilage degeneration.
Since SMAD2, SMAD3 and SMAD4 have been shown to differentially regulate TGFβ signaling and to have distinct functions during in vivo cartilage development and maintenance, we investigated their role in TGFβ -induced signaling and chondrogenesis of human BMSCs. We knocked down and overexpressed either SMAD2, SMAD3 or SMAD4 in human fetal BMSCs and determined the effects on TGFβ -activated SMAD phosphorylation and induction of chondrogenic differentiation.

Results
Efficient knockdown and overexpression of SMAD2, SMAD3 and SMAD4. To study the role of SMAD2, SMAD3 and SMAD4 in TGFβ -induced SMAD phosphorylation and chondrogenesis of human BMSCs, their expression was modulated in human fetal BMSCs using short hairpin RNA (shRNA)-mediated knockdown and adenoviral (ad)-mediated overexpression.
TGFβ-activated phosphorylation of R-SMADs is altered by knockdown and overexpression of SMAD2, SMAD3 or SMAD4. TGFβ signaling appears to be a straightforward cascade in which the ALK5 receptor phosphorylates SMAD2 and SMAD3, which upon binding to SMAD4 translocate to the nucleus where they regulate gene transcription 9 . However, this system is more complex as multiple mechanisms have been discovered that control TGFβ signal transduction 39 . Since it remained unknown whether R-SMAD phosphorylation depends on the amount of SMAD2 and SMAD3 present, we investigated whether modulating the expression of SMAD2 and SMAD3 affected TGFβ -activated phosphorylation of R-SMADs. We only show the effects on phosphorylated SMAD2 (pSMAD2) due to difficult detection of pSMAD3. First, we verified that TGFβ induced pSMAD2 in Ctrl-shRNA cells (Fig. 3a) and in ad-LacZ cells compared to no stimulation (Fig. 3b). TGFβ -activated SMAD2 phosphorylation was reduced in the SMAD2-shRNA condition (Fig. 3a) and it was enhanced in ad-SMAD2 (Fig. 3b). No effect on pSMAD2 was observed with SMAD3-shRNA (Fig. 3a) or ad-SMAD3 (Fig. 3b). Moreover, to determine if co-factor SMAD4 is involved in TGFβ -activated R-SMAD phosphorylation, the expression of SMAD4 was modulated. We found that pSMAD2 was lower in SMAD4-shRNA than in Ctrl-shRNA (Fig. 3a) and was similar between ad-SMAD4 and ad-LacZ (Fig. 3b).
Taken together, these data demonstrate that activation of the SMAD2/3 and SMAD1/5/9 signaling pathways by TGFβ was affected by modulating SMAD2, SMAD3 or SMAD4 expression, indicating that the levels of TGFβ -induced phosphorylated R-SMADs depend on the amount of R-SMAD2/3 and co-SMAD4.
Scientific RepoRts | 7:43164 | DOI: 10.1038/srep43164 Knocking down either SMAD3 or SMAD4 strongly inhibits chondrogenesis. We determined the contribution of SMAD2, SMAD3 and SMAD4 during TGFβ -induced chondrogenic differentiation by knocking down the expression of these SMADs and culturing the BMSCs in pellets in chondrogenic medium containing TGFβ . After 1, 7 and 14 days we evaluated expression of chondrogenesis-specific genes ACAN, COL2A1 and SOX9, formation of cartilage matrix and the macroscopic appearance of the BMSC pellets.
Next, we analyzed cartilage matrix deposition by determining the presence of glycosaminoglycans (GAGs) and collagen type II protein. In addition, we measured the pellet size. Compared to control, TGFβ -induced deposition of GAGs was not inhibited by SMAD2-shRNA at day 14 ( Fig. 4d), whereas it was significantly inhibited by SMAD3-shRNA (p = 0.010 at day 7; p < 0.001 at day 14). The same was observed after correcting the GAG content for the amount of DNA per pellet ( Fig. 4e; Supplementary Fig. S1). Despite stimulation with TGFβ , no GAGs were formed in pellets of BMSCs transduced with SMAD4-shRNA (Fig. 4d,e). Consistent with this, analyzing the presence of GAGs and collagen type II protein in 14 days-cultured pellets by histology revealed that both cartilage components were similarly present in SMAD2-shRNA and Ctrl-shRNA, whereas they were reduced in SMAD3-shRNA and even absent in SMAD4-shRNA (Fig. 4f). These observations were reflected by the size of the pellets after 14 days of culturing. Compared to Ctrl-shRNA pellets, SMAD2-shRNA pellets had the same size, SMAD3-shRNA pellets were smaller and SMAD4-shRNA pellets were smallest (Fig. 4g).
Once BMSCs have differentiated into chondrocytes, they undergo hypertrophic maturation, which is characterized by enhanced expression of collagen type X α1 (COL10A1) and runt-related transcription factor (e,f). Protein levels were normalized to GAPDH and expressed as relative to Ctrl-shRNA. Gene expression was normalized to the mean C T value of RPS27a and TBP. Data are expressed as % relative to normalized mRNA levels in Ctrl-shRNA. Bars represent mean + S.D. of triplicate pellets from 2 experiments. **p < 0.01; ***p < 0.001 compared to Ctrl-shRNA.
Altogether, these results indicate that shRNA-mediated knockdown of SMAD2 had a minor inhibitory effect on TGFβ -induced chondrogenic differentiation of human BMSCs. However, SMAD3 knockdown strongly reduced cartilage deposition, and SMAD4 knockdown completely blocked chondrogenesis.

SMAD3 overexpression results in a strong inhibition of chondrogenesis.
Since knockdown of SMAD2, SMAD3 and SMAD4 inhibited chondrogenesis, we investigated whether overexpression of these SMADs might enhance TGFβ -induced chondrogenic differentiation of human BMSCs.
Deposition of GAGs (Fig. 5d) was significantly decreased in pellets of BMSCs transduced with ad-SMAD2 compared to control after 7 days (p = 0.009) and 14 days (p = 0.003). However, this significant reduction was not observed when the GAG content was corrected for the DNA content ( Fig. 5e; Supplementary Fig. S3). Pellets of ad-SMAD3-transduced BMSCs had a significantly lower amount of GAGs per pellet (Fig. 5d: p < 0.001 at day 7; p = 0.006 at day 14) and per μ g DNA ( Fig. 5e; Supplementary Fig. S3; p = 0.007 at day 7; p < 0.001 at day 14) than the control condition. At all time points, GAG deposition (Fig. 5d,e) was similar between ad-SMAD4 and ad-LacZ. Histological analysis of 14 days-cultured pellets revealed that the abundance of GAGs and collagen type II protein (Fig. 5f) was similar between ad-SMAD2, ad-SMAD4 and ad-LacZ, whereas their abundance was strongly reduced in the ad-SMAD3 condition. In line with these observations, pellets of ad-SMAD2-transduced BMSCs were slightly smaller than control pellets, whereas pellets of ad-SMAD3-transduced BMSCs were much smaller (Fig. 5g). The pellet size was comparable between the ad-SMAD4 and ad-LacZ condition (Fig. 5g).
In addition, we found that chondrocyte hypertrophy markers, COL10A1 and RUNX2, were similarly expressed between ad-LacZ and ad-SMAD4 after 7 and 14 days ( Supplementary Fig. S4a,b). Although COL10A1 expression was lower in ad-SMAD2 than in ad-LacZ at day 7, the two conditions were comparable at day 14 ( Supplementary Fig. S4a). At both time points, RUNX2 mRNA levels were similar between ad-SMAD2 and ad-LacZ ( Supplementary Fig. S4b). Strikingly, in the ad-SMAD3 condition COL10A1 mRNA was lower ( Supplementary Fig. S4a), whereas RUNX2 mRNA was higher than in the ad-LacZ condition ( Supplementary Fig. S4b).
Our data demonstrate that TGFβ -induced chondrogenesis was slightly inhibited by SMAD2 overexpression and not affected by SMAD4 overexpression, whereas it was strongly inhibited by SMAD3 overexpression.

Discussion
The pro-chondrogenic effect of TGFβ on human BMSCs is well-known, however, the contribution of the downstream signaling molecules SMAD2, SMAD3 and SMAD4 during chondrogenesis has been less well investigated. By using human fetal BMSCs as a model for TGFβ -induced chondrogenic differentiation, we demonstrate here that modulating SMAD2 expression had a minor effect on chondrogenesis. In contrast, knockdown as well as overexpression of SMAD3 strongly inhibited cartilage formation and SMAD4 knockdown completely blocked chondrogenesis.
To the best of our knowledge, the specific role of SMAD2 and SMAD3 during chondrogenic differentiation of human BMSCs has been studied in one study so far, which demonstrated that modulating SMAD2 expression does not affect chondrogenesis 22 . In contrast, we observed a slight reduction in cartilage formation when SMAD2 was knocked down or overexpressed. We also found that SMAD2 knockdown led to reduced protein levels of SMAD3 and SMAD4, whereas Furumatsu et al. did not report an effect of SMAD2 knockdown on SMAD3 and SMAD4 expression 22 . Since knockdown of either SMAD3 or SMAD4 inhibited chondrogenesis, reduced SMAD3 and SMAD4 abundance might explain why chondrogenesis was slightly inhibited by SMAD2 knockdown. Our   finding that TGFβ -induced chondrogenic differentiation of human BMSCs is more strongly inhibited by SMAD3 knockdown than by SMAD2 knockdown is in line with the aforementioned study 22 . While we employed continuous knockdown by viral transduction with shRNA, Furumatsu and co-workers employed transient knockdown by transfection with siRNA 22 . We, thus, verified that continuous knockdown of SMAD2 and SMAD3 had effects on chondrogenesis similar to transient knockdown.
Whereas the previous study has shown that SMAD3 overexpression accelerated chondrogenesis 22 , in our study SMAD3 overexpression blocked chondrogenic differentiation of human BMSCs. This discrepancy might be caused by a difference in chondrogenic media composition (details of medium components were not reported), cell source (fetal or adult origin, bone marrow or other tissue derived) or amount of overexpression (western blots of overexpressed SMAD3 were not shown). In addition, we found that the DNA content per pellet decreased over time in cells overexpressing SMAD3 (Supplementary Fig. S3). In multiple cell types, it has been shown that overexpression of SMAD3 leads to an inhibition of proliferation [41][42][43] . Importantly, proliferation of BMSCs early during chondrogenic differentiation is required for chondrogenesis 44 . Therefore, SMAD3 overexpression might have inhibited chondrogenesis, because it blocked proliferation. Overall, this indicates that constantly high levels of SMAD3 in BMSCs do not accelerate cartilage matrix formation. Since both knockdown and overexpression of SMAD3 inhibited chondrogenic differentiation, chondrogenesis seems to rely on well-balanced levels of SMAD3.
The present study, for the first time, investigated the role of SMAD4 during chondrogenic differentiation of human BMSCs. Previous studies in mice have demonstrated that mesenchyme-specific deletion of Smad4 leads to an absence of the limb skeleton as a result of impaired mesenchymal condensation 30,31 ; a process required for initiating chondrogenesis 45 . In our study, SMAD4 knockdown did not interfere with pellet formation, implying that human BMSCs can form pellets even with a low amount of SMAD4. Although BMSCs transduced with SMAD4-shRNA formed pellets, they did not undergo chondrogenic differentiation. This observation is in line with a previous study demonstrating that BMSCs isolated from Smad4 knockout mice show reduced expression of Sox9 30 . Additionally, we found that SMAD4 knockdown inhibited TGFβ -induced SMAD1/5/9 phosphorylation and expression of COL2A1; a direct transcriptional target of SOX9 46 . In murine chondroprogenitor cells, complex formation between Smad4 and phosphorylated Smad1/5 proteins is required for transactivation of the Col2a1 promoter 47 . This might explain why chondrogenesis did not occur when we knocked down SMAD4. Furthermore, we previously demonstrated that the SMAD2/3 as well as the SMAD1/5/9 pathways are required for TGFβ -induced chondrogenesis of human BMSCs 6,10 . In the present study activation of both pathways was reduced by SMAD4 knockdown, thereby possibly explaining the absence of cartilage formation. Altogether, these findings underline the importance of SMAD4 during TGFβ -induced chondrogenic differentiation of human BMSCs.
Following chondrogenesis, BMSC-derived chondrocytes display signs of hypertrophic differentiation. This is undesired for the formation of articular cartilage, as hypertrophic chondrocytes produce cartilage that will mineralize and ossify when implanted in vivo 7 . We found that modulating SMAD2 expression did not have an effect on expression of hypertrophic differentiation markers, neither did SMAD4 overexpression. SMAD3 overexpression and SMAD4 knockdown did result in lower COL10A1 expression. In these conditions, however, next to hypertrophy markers, the markers of chondrogenesis were also much lower expressed than in the control condition, following the principle that reduced chondrogenesis leads to reduced hypertrophic differentiation. Whereas COL10A1 expression was lower by SMAD3 overexpression, RUNX2 expression was slightly, but significantly, higher. This was surprising, because SMAD3 is required for repression of RUNX2 expression and chondrocyte hypertrophy 38,[48][49][50][51] . Possibly, as a result of decreased chondrogenesis RUNX2 expression failed to go down during the early phase of chondrogenesis or RUNX2 expression levels might depend on the abundance of SMAD3, but this requires further investigation. Overall, our results failed to show effects specifically on hypertrophic differentiation of BMSC-derived chondrocytes without effects on the induction of chondrogenesis.
Besides the effects on chondrogenic and hypertrophic differentiation, this study investigated the effect of modulating SMAD2, SMAD3 and SMAD4 expression on TGFβ signal transduction. After activation by TGFβ, the ALK5 receptor phosphorylates SMAD2/3 proteins 9 . We show that SMAD2 phosphorylation was reduced by knockdown of SMAD2, while it highly increased by SMAD2 overexpression. These findings suggests that TGFβ -activated phosphorylation depends on the number of R-SMADs present in the cytoplasm rather than, for instance, on the kinase activity or number of ALK5 receptors.
Although SMAD4 is an important co-factor for translocation of activated R-SMADs to the nucleus 57 , it has not been implicated in controlling the phosphorylation of R-SMADs. Patients with Myhre syndrome have mutations in SMAD4 that lead to decreased ubiquitination of SMAD4 protein, resulting in accumulation of SMAD4 [58][59][60][61] . In skin fibroblasts from these patients, pSMAD2/3 and pSMAD1/5/9 were enhanced compared to healthy controls 58,61 . Contrary to this, in our study SMAD4 overexpression had no effect on R-SMAD phosphorylation. On the other hand, SMAD4 knockdown reduced TGFβ -induced phosphorylation of SMAD2 and completely prevented SMAD1/5/9 phosphorylation. Based on these observations, we speculate that SMAD4 is required for R-SMAD phosphorylation, or prevents R-SMADs from de-phosphorylation 62,63 or ubiquitination 64,65 , which adds a regulatory mechanism that controls TGFβ signal transduction. To conclude, this study reveals that TGFβ -activated phosphorylation of R-SMADs in BMSCs does not only depend on the levels of SMAD2 and SMAD3, but also on the presence of SMAD4. Moreover, our findings suggest that SMAD3 and SMAD4 are more important than SMAD2 for TGFβ -induced chondrogenic differentiation of human BMSCs. However, as cartilage formation was not enhanced by overexpression of SMAD4 and even inhibited by SMAD3 overexpression, induction of chondrogenic differentiation seems to rely on a delicate balance in the amount of SMAD3 and SMAD4. This also implies that continuously enhanced SMAD3 expression levels may not be a suitable strategy to improve chondrogenesis. Despite its exploratory nature, this study offers novel insights into the signaling mechanism governing the induction of chondrogenic differentiation of human BMSCs. Further research is required to delineate the function of SMAD4 in TGFβ -induced phosphorylation of R-SMADs and the mechanisms behind the detrimental effects of SMAD3 overexpression on chondrogenesis. After reaching 80% confluence, cells were passaged using 0.05% trypsin-EDTA (Gibco) and re-seeded (~6,000 cells/cm 2 ) in MSCGM. After 4 passages, cells were stored in liquid nitrogen. Per experiment, fetal BMSCs were defrosted and expanded in MSCGM for another 2 or 3 passages.
Lentivirus concentration was determined with the INNOTEST ® HIV p24 Antigen assay (Fujirebio Europe, Gent, Belgium) and expressed as pg of p24/μ L.
At 20% confluence, BMSCs were infected with 1 pg p24 per cell in MSCGM (Lonza) supplemented with 100 μ g/mL protamine sulfate (Sigma-Aldrich) for 1 day. After culture-expanding infected cells for 2 days, pellets were prepared to induce chondrogenesis. 1 day after pellet preparation, the efficiency of shRNA-mediated knockdown of SMAD2, SMAD3 and SMAD4 was analyzed by RT-qPCR and Western blot. Adenoviral-mediated overexpression of SMAD2, SMAD3 or SMAD4. Adenoviruses for SMAD2, SMAD3 and SMAD4 were kindly provided by Dr. P. ten Dijke (Leiden University Medical Center, the Netherlands). Cells at 80% confluence were incubated for 3 hours with adenovirus (multiplicity of infection of 1 pfu/cell) overexpressing either SMAD2, SMAD3, SMAD4 or LacZ (control). Following transduction, cells were washed and centrifuged to obtain pellets for chondrogenic induction. Overexpression of SMAD2, SMAD3 and SMAD4 was verified at gene (RT-qPCR) and protein (Western blot) level in BMSCs cultured for 1 day in chondrogenic medium.
Western blot analysis of (phosphorylated) SMAD proteins. To evaluate the effects of SMAD knockdown and overexpression on activation of TGFβ signaling, cells transduced either with lentivirus (SMAD-shRNA) or adenovirus (SMAD overexpression) were seeded in chondrogenic medium without TGFβ . After 18 hours, cells were stimulated with 10 ng/mL TGFβ 1 (Biolegend) for 1 hour and cell lysates were prepared as described below to determine pSMAD2 and pSMAD1/5/9 expression.
To determine protein abundance, cells were lysed (duplicate per condition) using lysis buffer (Cell-Signaling-Technology, Danvers, MA, USA) containing 1% protease inhibitor (Roche, Mannheim, Germany) and lysates were sonicated on ice. Protein concentration was determined using bicinchoninic acid assay. Duplicate samples per condition were pooled and 10 μ g protein lysate was subjected to 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis. Proteins were transferred to nitrocellulose membranes and overnight at 4 °C incubated with an antibody (1:1,000) recognizing SMAD2 (#40855, Abcam, Cambridge, UK), SMAD3 (#28379, Abcam), SMAD4 (#AF2097, R&D systems, Minneapolis, MN, USA), pSMAD2 (#3101 L, Cell Signaling Technology) or pSMAD1/5/9 (#9511 L, Cell Signaling Technology). Subsequently, membranes were incubated with HRP-linked antibody (1:1,500) against rabbit-IgG (#P0448, Dako, Glostrup, Denmark) or against goat-IgG (#P0449, Dako) for 1 hour at room temperature. To evaluate equal protein loading between conditions, GAPDH expression was determined. Membranes were overnight incubated with an antibody (1:20,000) recognizing Data analysis. Data represent mean + standard deviation of 6 pellets (from two experiments with triplicate pellets per experiment) per condition per time point. Statistical analyses were performed using SPSS version 22 (IBM, Armonk, NY, USA). Normality and variance were verified with the Shapiro-Wilk test and Levene's Test, respectively. Statistical differences between control and experimental conditions per time point were determined with two-tailed independent T-tests. Level of significance was set at P < 0.05.