Histone Deacetylase 3 Unconventional Splicing Mediates Endothelial-to-mesenchymal Transition through Transforming Growth Factor β2*

Background: Endothelial-to-mesenchymal transition (EndMT) is involved in embryonic cardiovascular development. Results: HDAC3 undergoes unconventional splicing during stem cell differentiation, which contributes to EndMT. Conclusion: HDAC3 unconventional splicing may modulate endothelial cell plasticity. Significance: Targeting HDAC3 splicing may provide new therapeutic strategies to tackle cardiovascular disease caused by endothelial plasticity. Histone deacetylase 3 (HDAC3) plays a critical role in the maintenance of endothelial integrity and other physiological processes. In this study, we demonstrated that HDAC3 undergoes unconventional splicing during stem cell differentiation. Four different splicing variants have been identified, designated as HD3α, -β, -γ, and -δ, respectively. HD3α was confirmed in stem cell differentiation by specific antibody against the sequences from intron 12. Immunofluorescence staining indicated that the HD3α isoform co-localized with CD31-positive or α-smooth muscle actin-positive cells at different developmental stages of mouse embryos. Overexpression of HD3α reprogrammed human aortic endothelial cells into mesenchymal cells featuring an endothelial-to-mesenchymal transition (EndMT) phenotype. HD3α directly interacts with HDAC3 and Akt1 and selectively activates transforming growth factor β2 (TGFβ2) secretion and cleavage. TGFβ2 functioned as an autocrine and/or paracrine EndMT factor. The HD3α-induced EndMT was both PI3K/Akt- and TGFβ2-dependent. This study provides the first evidence of the role of HDAC3 splicing in the maintenance of endothelial integrity.


Histone deacetylase 3 (HDAC3) plays a critical role in the maintenance of endothelial integrity and other physiological processes.
In this study, we demonstrated that HDAC3 undergoes unconventional splicing during stem cell differentiation. Four different splicing variants have been identified, designated as HD3␣, -␤, -␥, and -␦, respectively. HD3␣ was confirmed in stem cell differentiation by specific antibody against the sequences from intron 12. Immunofluorescence staining indicated that the HD3␣ isoform co-localized with CD31-positive or ␣-smooth muscle actin-positive cells at different developmental stages of mouse embryos. Overexpression of HD3␣ reprogrammed human aortic endothelial cells into mesenchymal cells featuring an endothelial-to-mesenchymal transition (EndMT) phenotype. HD3␣ directly interacts with HDAC3 and Akt1 and selectively activates transforming growth factor ␤2 (TGF␤2) secretion and cleavage. TGF␤2 functioned as an autocrine and/or paracrine EndMT factor. The HD3␣-induced EndMT was both PI3K/Akt-and TGF␤2-dependent. This study provides the first evidence of the role of HDAC3 splicing in the maintenance of endothelial integrity.
As a key cellular component of the circulatory system, endothelial cells (ECs) 3 play important roles in cardiovascular homeostasis and disease development. In response to local flow patterns and physiological stimuli, ECs can exhibit a wide range of phenotypic variability throughout the cardiovascular system (1). The most remarkable feature is the plasticity of endothelialto-mesenchymal transition (EndMT), which contributes to embryonic cardiogenesis (2)(3)(4)(5)(6); postnatal angiogenesis (7); and pathological processes, such as cardiac (8) and renal fibrosis (9), pulmonary hypertension (10), and tumor angiogenesis and metastasis (11,12).
EndMT defines a specific type of cell plasticity. During EndMT, resident ECs delaminate from an organized cell layer and invade the underlying tissue. These cells lose cell-cell junctions due to decreased VE-cadherin, acquire invasive and migratory properties, and lose other EC markers, such as CD31. On the other hand, these cells gain mesenchymal markers (e.g. FSP1 (fibroblast-specific protein 1), N-cadherin, and ␣-smooth muscle actin (␣SMA)) (5,8,10,13,14). EndMT was originally identified in cardiogenesis. A subset of ECs in the endocardium of mouse embryos at 9.5 days postcoitus (dpc) transform into mesenchymal cells that migrate, proliferate, and eventually remodel the cardiac cushions into heart valve leaflets and septa for a partitioned heart (2,3). Recent reports revealed that EndMT accounts for up to 40% of cancer-associated fibroblasts, which is the main source of host-derived vascular endothelial cell growth factor (VEGF) (11). Several signaling pathways are reported to be involved in EndMT, such as transforming growth factor ␤ (TGF␤) binding and Notch and Akt1 activation * This work was supported by grants from the British Heart Foundation and (5,8,15). A very recent study showed that EndMT is also involved in neointima formation (16). However, the mechanism of EndMT is still poorly understood.
Histone deacetylases (HDACs) modulate chromatin structure through regulating the acetylation status of histone tails, functioning as transcriptional co-repressors (17,18). Recent studies showed that HDACs can also modulate transcription factor activity, increase gene transcription (19), and interact with cytoskeleton and signal transducers (20 -22). There are 18 types of HDACs, classified into four categories. HDAC3 is a member of the Class I HDACs (17,23). It is an indispensable gene, removal of which in the germ cell line causes embryonic lethality at an early stage (24). Our previous study indicated that HDAC3 is essential for EC differentiation and integrity maintenance (25)(26)(27). In this study, we found that HDAC3 undergoes unconventional splicing during embryonic stem (ES) cell differentiation and development. In addition, overexpression of the splicing isoform of HDAC3, HD3␣, reprogrammed ECs into mesenchymal cells.
Cell Culture-Mouse embryonic stem cells (ES-D3 cell line, CRL-1934 (ATCC, Manassas, VA)), human aortic endothelial cells (HAECs), and 293 cells were maintained as described previously (26,28). Sca1 ϩ cells were isolated from collagen IVdifferentiated ES cells and maintained in differentiation medium (DM; ␣-minimum Eagle's medium supplemented with 10% FBS from Sigma, 5 ϫ 10 Ϫ5 mol/liter ␤-mercaptoethanol, 100 units/ml penicillin/streptomycin) for more than five passages. Because the majority of these Sca1 ϩ cells are ECs, these are therefore designated as esECs and maintained in DM (27). For ES differentiation, the cells were seeded into a collagen IVor I-coated flask in DM for 72 h, followed by assessments.
Adenoviral Gene Transfer-Ad-HD3␣ viral vector was created from pShuttle2-FLAG-HD3␣ plasmid, and the resulting adenoviral particles were amplified using an adenoviral expression system (Clontech, Takara Biosciences, St Germaine-enlaye, France) and the manufacturer's protocol. For adenoviral gene transfer, HAECs were incubated with Ad-null or Ad-HD3␣ virus at the multiplicity of infection (MOI) indicated for 6 h. After the removal of virus solution, the cells were cultured in human endothelial serum-free medium (Invitrogen) for 24 -48 h, in M199 medium supplemented with 1% FBS for 5 days, or in M199 medium supplemented with 5 ng/ml insulin for 24 h. For inhibitor or neutralization antibody assays, inhibitors (at the concentrations indicated in the figure legends) and neutralization antibody (0.2 g/ml) were included in the whole infection and incubation process. DMSO (same volume) and IgG (0.2 g/ml) were included as vehicle and negative control, respectively.
HD3␣/␤-GFP Transgenic Vector Construction-An HDAC3 DNA fragment covering exon 4 to exon 15 was amplified by PCR from genomic DNA and inserted into pLoxPneo vector. GFP coding sequences were inserted into the open reading frame of HD3␣ and -␤ within intron 12 upstream of the stop codon, whereas a LoxP-Neo-LoxP cassette was inserted into intron 12 downstream of the stop codon, creating the pLox-Pneo-HD3␣/␤-GFPkin plasmid. ES-D3 cells were transfected with this plasmid. The positive transfection clones were selected with G418, whereas the recombinant clones were further selected with ganciclovir. The positive recombinant clones were then transfected with pCMV-Cre210 plasmid (Addgene, Cambridge, MA) to remove the loxP-Neo-LoxP cassette. The positive HD3␣/␤-GFP stable cell clones were verified by PCR with primer sets flanking the GFP insertion and LoxP site, respectively. For GFP observation, HD3␣/␤-GFP-transfected ES cells were cultured in collagen I-coated dishes in DM for 3 days and observed under inverted phase and fluorescence microscopes. Images were taken and processed by Photoshop software.
Indirect Immunofluorescence Staining-Cryosections (5 m) from mouse embryos at 10.5-and 12.5-dpc stages or chamber slides of cell culture were fixed with methanol at room temperature for 15 min, permeabilized with 0.1% Triton X-100/PBS for 15 min, blocked with normal donkey serum (1:20 dilution in PBS) for 1 h, and incubated with primary antibodies (diluted in blocking serum) and Alexa Fluor546/488-conjugated secondary antibodies from donkeys (Dako) at 37°C for 1 h and 30 min, respectively. Nuclei were counterstained with DAPI. To assess the specificity of HD3␣ antibody, 10 g/ml HD3␣ antibody was incubated with a 10 g/ml concentration of the HD3␣ peptide (used to raise the antibody) at 37°C for 1 h prior to primary antibody incubation. Images were obtained with a Leica SP5 confocal microscope and LASAF software (Leica) at 37°C and processed by Adobe Photoshop software.
Mouse Embryonic and Adult Organ Tissues-Pregnant C57BL/6J female mice were sacrificed, and the whole embryos at 8.5, 10.5, 13.5, 15.5, 17.5, and 19.5 dpc and in new born mice at day 1 were harvested for RNA extraction. Meanwhile, different organ tissues were harvested from the female mice for RNA extraction. All animal experiments in this study were performed according to protocols approved by the Institutional Committee for the Use and Care of Laboratory Animals.
Immunoprecipitation and Immunoblotting-Cells were lysed by incubation with IP-A buffer (0.02 mol/liter Tris-HCl, pH 7.5, 0.12 mol/liter NaCl, 1 ϫ 10 Ϫ3 mol/liter EDTA, 1% Triton X-100 plus protease inhibitors (Roche Applied Science)) on ice for 45 min, followed by protein concentration assessment with Bradford reagent (Bio-Rad). One mg of cell lysate was mixed with 2 volumes of IP-B buffer (IP-A without Triton X-100), precleared with 2 g of normal IgG and 10 l of Easyview Protein G-agarose beads (Sigma), and then incubated with 2 g of anti-HA or anti-FLAG antibody and 10 l Easyview Protein G-agarose beads. The immunoprecipitates were separated by SDS-PAGE and detected by Western blot analysis. Fifty g of cell lysate was included as an input control. Immunoblotting was performed as a standard procedure described elsewhere.
Cellular Fractionation-HAECs were collected by scraping in a 400 l/75-ml flask of hypotension buffer (0.01 mol/liter Tris-Cl, pH 7.5, 0.01 mol/liter KCl plus protease inhibitors) and incubated on ice with vortexing every 5 min for 15 min. Twenty-five l of 10% Nonidet P-40 was added and vortexed at 200 rpm for 10 s. Nuclei were spun down at 16,100 ϫ g at 4°C for 10 s. The supernatant was recovered as a cytosol fraction. The nuclei were washed once with PBS, resuspended in 70 l of hypotension buffer containing 0.625% Nonidet P-40, and sonicated for 6 s. Nuclear extract was recovered from the supernatant by spinning at 16,100 ϫ g at 4°C for 5 min. Protein con-centration was assessed with Bio-Rad reagent. Twenty-five g of proteins was applied to Western blot analysis.
Analysis of Secreted Proteins-HAECs were infected with Ad-null or Ad-HD3␣ at an MOI of 10 in the absence or presence of inhibitors for 6 h in complete growth medium, followed by washing with serum-free M199 medium three times and incubation with M199 medium supplemented with 5 ng/ml insulin for 24 h. Conditioned medium was collected, and the cell debris was removed by centrifugation at 4000 rpm at 4°C for 5 min. The supernatant was recovered and subjected to exosome isolation and whole medium concentration, respectively. For isolation of exosomes, 1 ml of the supernatant was mixed with 500 l of total exosome isolation reagents (Invitrogen) and incubated on a rotator at 4°C for 24 h, followed by spinning at 16,100 ϫ g at 4°C for 1 h. The pellet was resuspended in 25 l of 1ϫ SDS loading buffer (0.02 mol/liter Tris-HCl, pH 8.9, 2% SDS, 10% glycerol, 0.5% 2-mercaptoethanol, 0.025% bromphenol blue). For total medium concentration, 500 l of the supernatant was applied to an Amicon Ultra-0.5 Ultracel-3 membrane unit (Millipore, Watford, UK) and centrifuged at 16,100 ϫ g at 4°C for 15 min. This procedure was repeated until 5 ml of the supernatant was concentrated to 100 l, which was mixed with 25 l of 5ϫ SDS loading buffer. Twenty-five l of exosomes or concentration sample was applied to SDS-PAGE and Western blot analysis.
siRNA Transfection Assay-HAECs were seeded in gelatincoated 75-ml flasks at 50% confluence 24 h before transfection and washed three times with PBS and incubated with both serum-and antibiotic-free M199 medium 1 h prior to transfection. Upon transfection, 60 l of 1 ϫ 10 Ϫ5 mol/liter PI3K p85 siRNA (sc-39126, Santa Cruz Biotechnology, Inc.), Akt1 siRNA (sc-29196), or control siRNA (sc-37007) was diluted into 800 l of M199 and incubated with prediluted Lipofectamine RNAiMAX (50 l in 800 l of M199 for 5 min; Invitrogen) at room temperature for 15 min. The mixture was added to the pretreated HAECs (6 ml/75-ml flask of M199 for 1 h) and incubated for 5 h, after which the medium was refreshed with complete growth medium. Forty-eight h post-transfection, the cells were infected with Ad-null or Ad-HD3␣ and cultured for 24 h, followed by incubation in serum-free M199 containing 5 ng/ml insulin for 24 h. Cell lysates and culture medium were collected for Western blot analysis of ␣SMA expression and TGF␤2 secretion, respectively.
Statistical Analysis-Data expressed as the mean Ϯ S.E. were analyzed with a two-tailed Student's t test for two groups or pair-wise comparisons or analysis of variance for more than one comparison. A value of p Ͻ 0.05 was considered to be significant.

HDAC3 Undergoes Unconventional Splicing during Stem Cell
Differentiation-Our previous study demonstrated that HDAC3 was essential for EC differentiation (26,27). When cloning mouse HDAC3 cDNA sequence using a primer set 5Ј-atgaccggtaccgtggcgtatttctacgac-3Ј and 5Ј-cacagcaagcttgctgctctaaatctccac-3Ј, we discovered a few smaller electrophoretic bands in differentiated embryonic stem (DES) cell samples (Fig.  1A). Similar bands were also found in samples from mouse  (Fig. 1C). To verify the nature of these bands, all of the bands were isolated and sequenced (GenBank TM accession numbers JN651901-JN651904). Interestingly, the bands appeared to be isoforms spliced in an unconventional way. Four different isoforms have been identified and designated as HD3␣, -␤, -␥, and -␦ (Fig. 1D). In HD3␣, -␤, and -␦, intron 12 was kept as an additional exon, in which part of the intron sequence was incorporated into the open reading frame (ORF) and ended within the intron (GenBank TM accession numbers JN651901, JN651902, and JN651904). HD3␣ and HD3␤ have the same ORF, and the only difference is that HD3␤ loses sequence from exons 9 and 10. The ORF of HD3␦ downstream exon 3 is totally different from that of HD3␣ and HD3␤. Although all three isoforms contain sequences from intron 12, the C terminus of HD3␦ is different from that of HD3␣ and HD3␤. In addition, recombination appears to be present between exons 3 and 9 for HD3␣, between exons 3 and 10 for HD3␤ and HD3␥, and between exon 3 and intron 12 for HD3␦. All four isoforms share the same far N-terminal domain as full-length HDAC3 and possess a specific C-terminal domain derived from intron 12, except for HD3␥. FLAG tag expression of the isoforms in 293 cells revealed the expected protein bands (Fig. 1E). To verify that these isoforms were not artifacts derived from RT-PCR, an HDAC3 gene construct was created (pLoxPneo-HD3␣/␤-GFPkin), in which a GFP coding sequence was inserted into intron 12 within the ORF of HD3␣/␤ upstream of the stop codon. The construct was transfected into mouse ES cells and selected sequentially with Geneticin G-418 and ganciclovir. The HD3␣/␤-GFP knock-in ES cell line was eventually established via removal of the LoxP-Neo-LoxP cassette using pCMV-Cre transfection. Normally, intron 12 is removed through conventional splicing, so GFP is absent. Although HD3␣/␤-GFP knock-in ES cells did not yield GFP-positive cells in ES culture medium, when these cells were cultured in differentiation medium for 3 days, a portion of the cell population became GFP-positive cells (Fig. 1F), suggesting that the intron 12 was kept as an additional exon, and the intron sequence was incorporated into the ORF. These results suggest that HDAC3 naturally undergoes unconventional splicing during stem cell differentiation, as far as the HD3␣/␤ isoforms are concerned. Because the ORF is different, HD3␦ cannot give rise to GFP. To distinguish HD3␣/␤ isoforms, we raised antibody in rabbit against peptide PQGDTILTSPQNDL from intron 12 (GenBank TM accession numbers JN651901 and JN651902) and performed Western blot on differentiated ES cell samples. A band corresponding to HD3␣ was detected in DES cells but not in undifferentiated ES cells (Fig. 1G, left), which disappeared when the antibody was preincubated with the blocking peptide (Fig. 1G, right), suggesting that the antibody is specific. The intensity of this HD3␣ band was slightly changed in PDGF-, TGF␤1-, or VEGF-treated DES cells. No band corresponding to HD3␤ was detected (Fig. 1G, left). Because HD3␣ is the main isoform, the following study will only focus on HD3␣.
Because HDAC3 is essential for EC differentiation (26), we tested whether the splicing event will affect such process by introducing HDAC3 and HD3␣ into DES cells and esECs (27) via plasmid transfection, followed by quantitative RT-PCR analysis of VE-Cad and SM22 gene expression. As expected, overexpression of HDAC3 increased VE-Cad expression in both ES cells (Fig. 1H, left) and esECs (Fig. 1I, left), with VEGF having a synergistic effect. HD3␣ had no effect on VE-Cad or SM22 gene expression in ES cells (Fig. 1H). In esECs, however, HD3␣ decreased VE-Cad (Fig. 1I, left) whereas it increased SM22 (Fig. 1I, right) expression, especially under PDGF treatment. These results suggest that HD3␣ may promote ECs trans-differentiation into mesenchymal cells.
HDAC3 Splicing May Play a Role in Embryonic Development-As described in the legend to Fig. 1B, spliced HDAC3 bands could be detected in mouse embryos by routine RT-PCR; it is worth investigating whether such a splicing event was involved in mouse embryonic development. Mouse embryos were harvested at 10.5 and 12.5 dpc, respectively. Cryosections were made to include the two developmental stages of the vasculogenesis, which starts from endothelial tube formation ( Fig. 2A) and ends with smooth muscle cell localization (Fig. 2B). Immunofluorescence staining with anti-HD3␣ revealed that HD3␣ emerged in the CD31-positive endothelial cells during the 10.5dpc stage when vasculogenesis begins ( Fig. 2A, b and c). At 12.5 dpc, there is no colocalization of CD31 and HD3a (Fig. 2B, a). However, HD3␣ was found to co-localize with ␣SMA-positive smooth muscle cells (Fig. 2B, b and c). Preincubation with the blocking peptide abolished HD3␣ staining ( Fig. 2A, a), confirming the specificity of anti-HD3␣ antibody. These results indicate that HD3␣ is involved in the embryonic vasculogenesis and may be a key regulator in the determination of vascular progenitor cells.
Overexpression of HD3␣ Induces EndMT via PI3K-Akt and TGF␤ Signal Pathways-To confirm whether HD3␣ is involved in EndMT, the effect of overexpression of HD3␣ in mature endothelial cell was investigated. HAECs were infected with Ad-HD3␣ virus followed by incubation with serum-free culture medium or M199 medium supplemented with 1% FBS. 24 h postinfection, the cell growth rate decreased, and some cells underwent morphology change (Fig. 3A). Western blot analysis revealed that overexpression of HD3␣ decreased the EC marker CD31, whereas it increased mesenchymal markers, such as N-cadherin, ␣SMA, VEGF, and EndMT transcription factor Snail2, in a dose-dependent manner (Fig. 3B). Immunofluorescence staining confirmed the EndMT phenotype as revealed by the decrease or redistribution of VE-Cad and the occurrence of ␣SMA in the Ad-HD3␣-infected and adjacent cells (Fig. 3C). These results suggest that overexpression of HD3␣ can induce EndMT.
HD3␣ Induces TGF␤2 Activation-The TGF␤ signaling pathway plays a critical role in EndMT (34). The above study has demonstrated that this pathway is also required by HD3␣induced EndMT. To investigate how HD3␣ drives EndMT through TGF␤ signal pathway, the activation of this pathway was first assessed in Ad-HD3␣-infected cells. As shown in Fig.  4A, overexpression of HD3␣ induced Smad2/3 phosphorylation and nuclear translocation and increased Snail2 expression and nuclear translocation. The majority of HD3␣ itself is located in the cytosol, with only a small portion located in the nuclear fraction. Further experiments indicated that exogenous TGF␤1 and -2 could induce EndMT, but only TGF␤1 had an additive effect on HD3␣-induced EndMT (Fig. 4B). In the presence of TGF␤1, overexpression of HD3␣ induced a huge accumulation of phospho-Smad2 in the nuclei, whereas the majority of HD3␣ is located on the nuclear envelope (Fig. 4C). These experiments suggest that HD3␣ drives EndMT through activating Smad2/3 phosphorylation and nuclear translocation.
In the TGF␤ signaling axis, TGF␤s bind to TGF␤ receptor-II, which recruits and phosphorylates TGF␤ receptor-I (ALK5). The latter activates Smad2/3 phosphorylation and nuclear translocation (35). As described above (Fig. 3, D and E), ALK5 is essential for HD3␣-induced EndMT. Furthermore, adjacent cells next to Ad-HD3␣-infected cells also express ␣SMA (Fig.  3C). It is reasonable to assume that HD3␣ induces the secretion and/or activation of TGF␤s, which in turn activate ALK5 in an autocrine and paracrine manner. To test this, we first detected TGF␤1, -2, and -3 mRNA levels by quantitative RT-PCR. Overexpression of HD3␣ did not seem to affect these gene expressions at mRNA levels (Fig. 4D). Interestingly, there was also only a slight increase at the ␣SMA mRNA level (Fig. 4D). Increase of ␣SMA protein may be mediated by translational regulation. Western blot analysis showed that overexpression of HD3␣ had no significant effect on TGF␤1 and TGF␤2 pro-tein levels in cell lysate as well, although Smad2 phosphorylation was significantly increased (Fig. 4E). However, the analysis of the secreted proteins in conditioned medium indicated that overexpression of HD3␣ significantly increased TGF␤2 activation via cleavage, as revealed by the occurrence of the small electrophoretic band (Fig. 4, F and G). PI3K-Akt inhibitor, LY294002, increased TGF␤1 and TGF␤2 protein levels in Ad-HD3␣-infected cells but abolished Ad-HD3␣-induced Smad2 phosphorylation (Fig. 4E) and attenuated TGF␤2 activation (Fig. 4, F and G). As expected, ALK5 inhibitor SB431542 abolished HD3␣-induced Smad2 phosphorylation (Fig. 4E), and the active TGF␤2 was accumulated in Ad-HD3␣-infected cell medium (Fig. 4, F and G). TGF␤1 is reported to be secreted by exosomes (36). To test whether the secretion of TGF␤2 is also mediated by exosomes, TGF␤2 was detected in cell lysate, exosomes, and whole medium concentrates by Western blot. Exosomes were confirmed by the presence of CD9 protein (37). As expected, a higher level of TGF␤1 was detected in exosomes as compared with the medium (Fig. 4H). However, the majority of the secreted TGF␤2, either the precursor (50 kDa band) or active band (13 kDa; arrow in Fig. 4H) locates in the whole medium concentrates but not in exosomes, suggesting that TGF␤2 is secreted by mechanisms other than exosomes. Further experiments revealed that the presence of TGF␤2 neutralization antibody attenuated HD3␣-induced Smad2 phosphorylation and ␣SMA expression (Fig. 4I). These results suggest that HD3␣ induces TGF␤2 protein secretion and activation in a PI3K-Akt pathway-dependent manner and that the active TGF␤2 in turn functions as an autocrine and/or paracrine factor to activate downstream signal pathways, leading to EndMT.
HD3␣ Physically Interacts with Akt1-PI3K-Akt pathway plays a central role in EMT (38). As described above, our study has demonstrated that an active PI3K-Akt pathway is essential to HD3␣-induced TGF␤2 activation and EndMT. The question is how PI3K-Akt cross-talks with HD3␣. To test this, Akt phosphorylation was assessed by Western blot following Ad-HD3␣  infection. As shown in Fig. 5A, overexpression of HD3␣ did not cause a significant up-regulation of Akt phosphorylation, suggesting that HD3␣ itself does not activate the PI3K-Akt pathway. Our previous study has shown that HDAC3 could associ-ate with Akt1 (25). We tested whether HD3␣ could also physically interact with Akt by co-transfecting FLAG-tagged HDAC3 and isoform plasmids together with HA-tagged Akt1 or Akt2 into 293 cells, followed by co-immunoprecipitation FIGURE 4. Overexpression of HD3␣ induces EndMT through TGF␤2 secretion and activation. A, overexpression of HD3␣ activated the TGF␤ signaling pathway. Tubulin and histone H4 were used to indicate cytosol and nuclear extract fractions, respectively. B, TGF␤1 enhanced HD3␣-induced EndMT. Ad-nullor Ad-HD3␣-infected HAECs were treated with 5 ng/ml TGF␤1 or TGF␤2 in serum-free medium for 24 h, followed by Western blot analysis. Numbers are relative amounts of phospho-Smad2 averaged from three independent experiments. C, overexpression of HD3␣ enhanced TGF␤1-induced Smad2 phosphorylation (green) and nuclear translocation. FLAG antibody was used to identify exogenous HD3␣ (red). D, overexpression of HD3␣ did not affect TGF␤s mRNA levels. E-G, LY294002 abolished Ad-HD3␣-induced TGF␤2 activation and Smad2 phosphorylation. 5 ϫ 10 Ϫ6 mol/liter LY294002 or 1 ϫ 10 Ϫ5 mol/liter SB431542 was included in the virus infection (10 MOI for 6 h) and incubation process (M199 supplemented with 5 ng/ml insulin for 24 h), followed by Western blot analysis (without 2-mercaptoethanol) of cell lysates (E) and conditioned media (F and G). DMSO was included as a vehicle control. *, p Ͻ 0.05; **, p Ͻ 0.01. H, TGF␤2 did not exist in exosomes. Ad-null or Ad-HD3␣-infected HAECs were incubated with serum-free medium containing 5 ng/ml insulin for 24 h, followed by Western blot analysis (with 2-mercaptoethanol) of cell lysate (Lys; 25 g of protein), exosomes (EM; from 1 ml of medium/lane), and whole medium concentrates (MC; from 1 ml/lane). CD9 was used to indicate exosomes. Arrow, the 13 kDa active band. I, TGF␤2 neutralization antibody attenuated HD3␣-induced Smad2 phosphorylation and ␣SMA expression. For all Western blot analyses, FLAG antibody was used to detect exogenous HD3␣, whereas GAPDH was included as a loading control. Data are representative of or the average of three independent experiments. assays. HDAC3 and all four isoforms associated with Akt1 but not with Akt2 (Fig. 5B). Further experiments revealed that HD3␣ associated with endogenous Akt1 in HAECs and that this association was not affected by TGF␤1 (Fig. 5C). Immunofluorescence staining confirmed the association of Akt1 and HD3␣ in HAECs, which did not seem to be affected by the presence of LY294002 (Fig. 5D). Because LY294002 is also reported to inhibit other signal pathways like Adamts1 (39), the direct involvement of the PI3K/Akt pathway in HD3␣-driven EndMT was confirmed by siRNA-mediated knockdown assays. As shown in Fig. 5E, knockdown of either PI3K p85␣ or Akt1 ablated Ad-HD3␣-induced ␣SMA expression. However, HD3␣-induced TGF␤2 secretion and cleavage seemed not to be affected. The deficiency of p85 and especially Akt1 significantly reduced the HD3␣ protein level. These results suggest that PI3K and Akt1 may not be directly involved in TGF␤2 secretion and cleavage but are essential for HD3␣-induced ␣SMA expression.
HD3␣ Modulates HDAC3 Functions-Being one of the isoforms, HD3␣ shares the same N terminus and part of the internal sequence of fully spliced HDAC3. Therefore, a natural ques-tion is whether HD3␣ can associate with and modulate HDAC3. Immunoprecipitation assay was first performed in 293 cells co-transfected with FLAG-tagged HDAC3 and all four isoform plasmids together with HA-tagged HDAC3 plasmid. As expected, HDAC3 could form a homodimer. HD3␣, -␤, and -␥ isoforms, but not HD3␦, could form a heterodimer with HDAC3 (Fig. 6A). Immunofluorescence staining with antibodies against the different C-terminal domains (anti-HDAC3 (Sigma, H3034) against peptide from amino acids 411-428 of the HDAC3 protein and anti-HD3␣ raised against peptide derived from intron 12) confirmed the physical association of HDAC3 and HD3␣, mainly in the cytoplasm (Fig. 6B). Interestingly, HDAC3 protein levels in HD3␣-positive cells are much higher than that in HD3␣-negative cells, suggesting that HD3␣ may stabilize HDAC3. To test whether there is a cross-talk or interaction between HDAC3 and HD3␣ on EndMT, overexpression of HDAC3 via adenoviral transfer and suppression of HDAC3 activity via selective inhibitor apicidin (40) were included in Ad-HD3␣-induced EndMT assays. Quantitative RT-PCR assays revealed that neither overexpression of HDAC3 nor inhibition of HDAC3 activity had significant effect on FIGURE 5. HD3␣ physically interacts with Akt1. A, HD3␣ did not activate Akt phosphorylation. FLAG antibody was used to detect exogenous HD3␣, and GAPDH was included as a loading control. B, Western blot detected the association of HDAC3 and its isoforms with Akt1. 293 cells were transfected with FLAG-HDAC3 and FLAG-HD3␣,␤,␥,␦ together with HA-Akt1 or HA-Akt2, followed by immunoprecipitation (IP) with anti-HA antibody and Western blot (IB) with anti-FLAG and HA antibodies. C, TGF␤1 enhanced HD3␣ association with Smad3 but did not affect HD3␣ association with Akt1. D, LY294002 did not affect the association of HD3␣ with Akt1. HAECs were infected with Ad-HD3␣ at an MOI of 10 in the presence of DMSO or 5 ϫ 10 Ϫ6 mol/liter LY294002 for 6 h and then cultured in serum-free medium in the presence of DMSO or LY294002 for 24 h, followed by immunofluorescence staining with anti-Akt1 (green) and FLAG (red) antibodies. E, PI3K and Akt1 were essential for HD3␣-induced ␣SMA expression. HAECs were transfected with p85 siRNA or Akt1 siRNA for 48 h, followed by Ad-HD3␣ infection for 48 h. Prior to harvesting cell lysate and culture medium, the cells were treated with serum-free M199 for 24 h. Data presented are representative of three independent experiments.
TGF␤2 and ␣SMA mRNA levels (Fig. 6C). Overexpression of HDAC3 itself decreased the basal level of the EndMT phenotype as revealed by the decrease of ␣SMA, pSmad2, and Snail2 protein levels in cell lysate (Fig. 6D) and decreased TGF␤2 secretion/activation in the medium (Fig. 6E). In contrast, suppression of HDAC3 by the selective inhibitor apicidin induced TGF␤2 secretion/activation and the EndMT phenotype at basal levels and enhanced HD3␣-induced EndMT (Fig. 6, D and  E). A striking finding is that overexpression of HDAC3 enhances HD3␣-induced TGF␤2 secretion, although TGF␤2 cleavage (Fig. 6E) and Smad2 phosphorylation and ␣SMA expression (Fig. 6D) were suppressed when HDAC3 and HD3␣ were expressed together. Overexpression of HD3␣ increased endogenous and exogenous HDAC3 protein level, and overexpression of HDAC3 could also increase exogenous HD3␣ protein level (Fig. 6D), confirming the immunofluorescence staining. These results suggest that HD3␣ may form a complex with HDAC3 and modulate HDAC3 function.

DISCUSSION
EndMT is one type of EMT that refers only to vascular/cardiac endothelial cells undergoing mesenchymal transition, which contributes to cardiogenesis and a range of pathological fibrosis in human diseases (41)(42)(43). In the present study, we showed a novel finding that HDAC3 undergoes unconventional splicing to produce four different isoforms, of which HD3␣ FIGURE 6. HD3␣ modulates HDAC3 function. A, Western blot analysis detected the association of HDAC3 isoforms with HDAC3 in HEK293 cells. All isoforms except for HD3␦ associated with HDAC3. B, HD3␣ physically associated with and stabilized HDAC3. HAECs were infected with Ad-HD3␣ at an MOI of 10 for 6 h in complete growth medium and cultured in serum-free medium for 24 h, followed by immunofluorescence staining with anti-HDAC3 (green) and anti-FLAG (for HD3␣; red). C-E, effect of overexpression or suppression of HDAC3 on HD3␣-induced EndMT. HAECs were co-infected with Ad-HD3␣ (10 MOI) and Ad-HDAC3 (10 MOI) or in the presence of 1 ϫ 10 Ϫ5 mol/liter apicidin in complete growth medium for 6 h and then cultured in M199 medium supplemented with 5 ng/ml insulin for 24 h, followed by quantitative RT-PCR analysis of TGF␤2 and ␣SMA mRNA levels (C), Western blot analysis of cell lysates (D), and Western blot analysis of conditioned media with the ratio of cleaved to precursor TGF␤2 indicated in the bottom panel (E). Ad-null is included as a control and to normalize variations in MOI. DMSO was included as a vehicle control. *, p Ͻ 0.05; **, p Ͻ 0.01. Data presented are representative of or the average of three independent experiments. IB, immunoblot; IP, immunoprecipitation.
contributes to stem cell differentiation and embryonic cardiovascular development. Overexpression of HD3␣ induces the EndMT phenotype in mature ECs through TGF␤2 activation. These findings provide new insights into the mechanism of EndMT.
Unconventional splicing is a common event occurring in mammalian cells in different cellular processes, which makes it possible for a single gene to produce different proteins with different or even opposite functions. In this study, the occurrence of GFP in HD3␣/␤-GFP-transfected cells, the specific band detected by Western blot, and specific staining in mouse embryos with HD3␣/␤ antibody have demonstrated that intron 12 is kept as an additional exon. In addition to intron 12 retention, there seems to be recombination between exons because the spliced different exons cannot be produced from the routine splicing mechanisms. Due to the sequence differences in intron 12 among difference species, the HD3␣ isoform seems unique to Mus musculus. We have detected several isoforms from human HDAC3, 4 which are different from those HDAC3 isoforms in M. musculus. Detailed investigations are still under way to decipher how intron 12 is kept as an additional exon, how the recombination events occur, and which human isoform possesses a function similar to that of HD3␣.
HDAC3 forms a complex with co-repressors like N-CoR/ SMRT and Ski, in which the co-repressors can activate HDAC3 deacetylase activity (44 -46). These complexes may also contain other HDACs and suppress gene transcription via histone deacetylation. Structural analyses of HDAC3 protein indicate that the C-terminal domain is the deacetylase catalytic domain and is responsible for gene transcriptional repression, whereas the N-terminal domain is responsible for oligomerization of HDAC3 itself and interaction with other HDACs and proteins, such as Akt1 (25,47). In addition to its deacetylase activity, HDAC3 may function as a scaffold in complexes, in which the N-terminal domain plays an essential role (48,49). The HD3␣ isoform retains the N-terminal domain. Therefore, it can form a complex with HDAC3 protein. Indeed, we have detected the direct association of HD3␣ and HDAC3. The formation of the complex may modulate HDAC3 functions in several ways. First, HD3␣ may participate in an HDAC3 complex. Because the intron 12-derived C-terminal domain of HD3␣ is different from that of HDAC3, the involvement of HD3␣ in the complex may interfere with HDAC3 deacetylase activity, therefore attenuating the transcriptional repression effects. Second, HD3␣ may compete with HDAC3 for binding to associated proteins, excluding HDAC3 from some complexes and ablating both the catalytic and scaffolding roles of HDAC3. It has been reported that HDAC3 can associate with Akt and Smads (25,45). In our study, overexpression of HDAC3 decreases Smad2 phosphorylation, suggesting a suppressive role of HDAC3 in TGF␤ signaling. As expected, we found the association of HD3␣ with Akt1 and Smad3. In contrast to the suppressive effect of HDAC3 on Smad2 phosphorylation, HD3␣ increases Smad2 phosphorylation, suggesting that HD3␣ may interfere with HDAC3 interaction with Smad2 or modify the HDAC3-Smad2 complex. Further evidence for the competition hypothesis derives from the observation that overexpression of HD3␣ stabilizes HDAC3 protein. Immunofluorescence staining revealed that almost all HD3␣ proteins co-localize with HDAC3 and that the protein level of HDAC3 in HD3␣-positive cells is much higher than that in HD3␣-negative cells. When these two proteins were overexpressed together, the level of each protein was much higher than that in a single overexpression system. The formation of HD3␣-HDAC3 complex may prevent the association with degrading enzymes, therefore increasing the stabilization. The dimerization of HD3␣ and HDAC3 may also introduce HDAC3 into new signal pathways via the interaction of the HD3␣ C-terminal domain with other proteins. A detailed investigation of the relationship between HD3␣ and HDAC3 is required.
Multiple factors have been reported to induce EndMT via several interactive signal pathways. The common factor is the TGF␤ superfamily, in which TGF␤2 plays an indispensable role in EndMT in vivo (34). TGF␤s bind to the TGF␤ type II receptor, which recruits and phosphorylates the type I (ALK5) receptor. The latter activates Smad signaling pathways, leading to EndMT. In this study, we demonstrate that HD3␣ induces EndMT through TGF␤2. HD3␣ activates TGF␤2 in the extracellular matrix. The active TGF␤2 functions as an autocrine and paracrine factor to activate ALK5 receptor and Smad2/3 phosphorylation, leading to Snail2 up-regulation and EndMT. The evidence comes from the following observations. HD3␣ up-regulates Smad2/3 phosphorylation and Snail2 and mesenchymal marker expression, which can be abolished by suppression of the ALK5 receptor via SB435142. Higher amounts of cleaved TGF␤2 have been detected in conditioned medium from Ad-HD3␣-infected HAECs as compared with Ad-null infected cells. Utilization of TGF␤2 neutralization antibody attenuates the HD3␣-induced Smad2 phosphorylation and ␣SMA expression. Highly expressed ␣SMA was detected not only in HD3␣-positive cells but also in adjacent cells. High levels of active TGF␤2 were detected when SB435142 was present. SB431542 binds to ALK5, blocking downstream events. The binding of SB435142 to ALK5 may inhibit the binding of TGF␤2 to the type II receptor in a negative feedback mechanism, therefore accumulating TGF␤2 in the medium.
HD3␣-induced activation of TGF␤2 occurs in the extracellular matrix via protease-mediated cleavage. TGF␤s are secreted in a latent form, associating with latent TGF␤-binding protein (50). Upon proteolytic cleavage, the latent TGF␤-binding protein and the latency-associated peptide are removed from TGF␤ molecules, giving rise to 13-kDa singlet or 25-kDa dimer mature forms. Upon the overexpression of HD3␣, no elevated TGF␤s at mRNA and protein levels were discovered. However, the higher level of mature TGF␤2 was detected in Ad-HD3␣-infected cell culture medium, suggesting that HD3␣ may activate certain proteases in the extracellular environment, whose expression and secretion were increased. Although TGF␤1 and TGF␤2 belong to the same family, their secretion mechanisms are different. TGF␤1 is secreted via exosomes, whereas TGF␤2 seems not to be. Therefore, the extracellular location or binding partners will be different for TGF␤1 from those of TGF␤2, and so is the activation mechanism involved (i.e. different proteases are responsible for the cleavage of TGF␤1 and TGF␤2). In this study, HD3␣ activates a few proteases, which locate in the cell surface or extracellular matrix but not in exosomes; therefore, they specifically initiate TGF␤2 but not TGF␤1 cleavage.
It has been reported that Smad-independent pathways are also involved in EndMT, such as MEK, PI3K/Akt, p38 and JNK MAPKs, and G-proteins (Rho). In this study, we also discovered that the PI3K/Akt pathway is essential for HD3␣-induced EndMT. However, it may not activate TGF␤2, because genetic knockdown of PI3K or Akt1 ablated HD3␣-induced ␣SMA expression but did not affect TGF␤2 activation. In p85 siRNAand Akt1 siRNA-transfected cells, HD3␣ protein levels were significantly decreased, especially in the latter. Akt1 binds to HD3␣ in the absence or presence of LY294002, indicating that Akt1 phosphorylation is not involved in the binding. The complex formation between Akt1 and HD3␣ may contribute to the stabilization of HD3␣ protein. Considering that LY294002 abolishes TGF␤2 activation, Smad phosphorylation, and EndMT, whereas p85 or Akt1 knockdown does not affect TGF␤2 activation, it appears that other signal pathways may be involved, which are susceptible to LY294002.
In summary, HDAC3 mRNA undergoes unconventional splicing, in which intron 12 is kept as an additional exon and recombination occurs between exons 3 and 9, giving rise to the HD3␣ isoform. The HD3␣ forms a complex with HDAC3 and Akt1 to promote TGF␤2 secretion and activation through a mechanism yet to be elucidated. The active TGF␤2 functions in an autocrine and paracrine manner to activate the ALK5-Smad2/3-Snail2 pathways, leading to EndMT (Fig. 7). This study provides the first evidence that HDAC3 mRNA can undergo unconventional splicing to modulate HDAC3 and induce EndMT. Further detailed investigation on the mechanisms involved in HDAC3 splicing and how the isoforms exert their actions will undoubtedly contribute to further insights into the role of HDAC3 in the maintenance of endothelium integrity.