Sumoylation of the Transcriptional Intermediary Factor 1β (TIF1β), the Co-repressor of the KRAB Multifinger Proteins, Is Required for Its Transcriptional Activity and Is Modulated by the KRAB Domain*

Small ubiquitin-related modifier (SUMO) has emerged as a key post-translational modulator of protein functions. Here we show that TIF1β, a developmental regulator proposed to act as a universal co-repressor for the large family of KRAB domain-containing zinc finger proteins, is a heavily SUMO-modified substrate. A combined analysis of deletion and punctual mutants identified TIF1β as a multilysine acceptor for SUMO which specifically targets six lysine residues (Lys554, Lys575, Lys676, Lys750, Lys779, and Lys804) within the TIF1β C-terminal repressive region. Reporter gene assays indicate that TIF1β requires SUMO-modification for its repressive activity. Indeed, sumoylation-less mutants failed to recapitulate TIF1β-dependent repression. TIF1β homodimerization properties and interaction with the KRAB domain are preserved in the mutants with lysine to arginine substitutions as confirmed by in vivo bioluminescence resonance energy transfer (BRET). Using histone deacetylase (HDAC) inhibitors, we also demonstrate that TIF1β sumoylation is a prerequisite for the recruitment of HDAC and that TIF1β SUMO-dependent repressive activity involves both HDAC-dependent and HDAC-independent components. Finally, we report that, in addition to relying on the integrity of its PHD finger and on its self-oligomerization, TIF1β sumoylation is positively regulated by its interaction with KRAB domain-containing proteins. Altogether, our results provide new mechanistic insights into TIF1β transcriptional repression and suggest that KRAB multifinger proteins not only recruit TIF1β co-repressor to target genes but also increase its repressive activity through enhancement of its sumoylation.

In attempts to unveil other determinants involved in the strong repressive activity mediated by TIF1␤, we found that this co-repressor is a SUMO target protein. Recent studies have demonstrated that post-translational modification of transcription factors and their co-regulators by the small ubiquitinrelated modifier protein, SUMO, can modulate their transcriptional activity (24 -28). The four mammalian SUMO, SUMO1, -2, -3, and -4, all resemble ubiquitin in terms of tertiary structure and their mechanism of conjugation to protein substrate. However, unlike ubiquitination, the main known functional consequences of SUMO modification appear not to be degradation (29 -33). Several reports illustrate the direct involvement of SUMO modification in gene regulation and in particular in gene repression (34 -44). To explain the transcriptional repression induced by SUMO-modified proteins, some studies have suggested that SUMO proteins display intrinsic repressive activity, whereas others have proposed that SUMO-modified transcription factors can recruit HDAC proteins or HDACcontaining complexes (35,40,42,(45)(46)(47)(48)(49)(50).
SUMO proteins form an isopeptide bond between their C-terminal glycine and a ⑀-NH 2 -amino group of a lysine residue on their target proteins (51). The target lysine to which SUMO is conjugated is usually contained within a short consensus sequence, KX(E/D) (where represents a large hydrophobic residue) (52). The SUMO conjugation process, so called sumoylation, involves SUMO-activating E1 enzyme (SAE1/ SAE2 heterodimer) and a SUMO-conjugating E2 enzyme (UBC9) (53). Furthermore, SUMO-specific E3 ligases that enhance SUMO conjugation have also been described. The known E3 ligases belong to three structurally independent protein classes represented by RanBP2, Pc2, and PIAS family members (protein inhibitor of activated STAT) (54 -57). SUMO E3 ligases display limited substrate specificity and are reported to stabilize the interaction between the E2-conjugating enzyme, UBC9, and the target substrates to be sumoylated (53, 58 -60). Like ubiquitination, sumoylation is a dynamic and reversible process because of the existence of several SUMO-specific proteases, referred to as members of the SENP (sentrin proteases) family. These proteases are involved both in the maturation of SUMO as C-terminal hydrolases and in the removal of SUMO from its target protein as isopeptidases (58).
In this study, we report that TIF1␤ is a strong substrate for SUMO modification exhibiting several sumoylation sites clustered in its C-terminal repressive region. Prevention of TIF1␤ sumoylation by using sumoylation-deficient mutants abrogates its repressive activity both as a protein artificially targeted to DNA or indirectly recruited to DNA via a KRAB domain-containing protein. Furthermore, we demonstrate that TIF1␤ SUMO modification is required for recruitment of HDAC-containing complexes. Finally, this study indicates that TIF1␤ sumoylation can be positively regulated by its homodimerization and its interaction with KRAB domain-containing multifinger proteins which are recruiting this co-repressor to target genes.

Cell Culture and Transfection
Human embryonic kidney 293T cells (HEK 293T) maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum (Wisent), 100 g/ml penicillin and streptomycin, 1 mM L-glutamine were seeded at a density of 3 ϫ 10 5 cells per 60-mm dish for cell extracts and immunoprecipitation experiments, 1 ϫ 10 6 cells per 100-mm dish for BRET experiments, and 2 ϫ 10 5 in 6-well plates for transcriptional assays. Transient transfections of plasmids were performed on the following day using the calcium phosphate precipitation method except for transcription assays where FuGENE transfection reagent (3 l/1 g DNA) (Roche Applied Science) was used. The total amount of transfected DNA was kept constant (10 g for 100-mm dishes, 6 g for 60-mm dishes, and 1-1.3 g for 6-well plates).

Cells Extracts and Immunoprecipitations
For preparing whole cell extracts to be separated by SDSpolyacrylamide gel electrophoresis, cells were lysed in denaturating Laemmli buffer (400 -600 l/dish) containing 20 mM NEM (N-ethylmaleimide), a cysteine protease inhibitor usually used to preserve the sumoylation of cellular proteins. DNA was immediately sheared using a 1-ml syringe with a 26-gauge 1 ⁄ 2 needle or sonicated to reduce sample viscosity. For immunoprecipitation, cells were lysed in SDS-solubilizing buffer (50 mM Tris, pH 7.5, 100 mM NaCl, 0.8% SDS, 20 mM NEM) (1.5 ml/dish) completed with a mixed proteases inhibitors. The following steps were done at 4°C. Extracts were passed through a 26-gauge 1 ⁄ 2 needle to reduce viscosity and centrifuged at 15,000 ϫ g for 2 min. Extracts were then precleared with protein A-Sepharose (1 ml/10 -20 l stacked beads) (Sigma) for 1 h. Immunoprecipitations of the precleared extracts (1 ml) were then carried out for 16 h using 10 l of the appropriate antibody and 40 -50 l of protein A-Sepharose. The protein A-Sepharose beads were washed three times with 1 ml of washing buffer (50 mM Tris, pH 7.5, 100 mM NaCl, 0.4% SDS, 2% Triton, 20 mM NEM). Washed immunoprecipitates were resuspended in Laemmli buffer. Whole cell extracts and immunoprecipitates were separated on SDS-polyacrylamide gel electrophoresis and electrotransferred for Western blotting. Following probing of the nitrocellulose membranes with the relevant antibodies, a chemiluminescence reagent was used (Renaissance kit, PerkinElmer Life Sciences).

BRET 2 Experiments
The BRET assays were conducted as we previously described in Ref. 14. In brief, transiently transfected cells were resuspended and distributed in 96-well microplates (white Optiplate from Packard). Upon addition of the cell permeant luciferase substrate (coelenterazine deep blue, PerkinElmer Life Sciences), the bioluminescence signal resulting from its degradation was detected using a 370 -450-nm band pass filter (donor emission peak 400 nm). The energy transferred resulted in a fluorescence signal emitted by the GFP10 acceptor (excitation peak 400 nm and emission peak 510 nm) that was detected using a 500 -530-nm band pass filter. The BRET signal (BRET ratio) was quantified by calculating the acceptor fluorescence/ donor bioluminescence ratio as previously reported. Expression level of each construct was determined by direct measurements of total fluorescence and luminescence on aliquots of transfected cell samples. The GFP10 total fluorescence was measured using a FluoroCount (PerkinElmer Life Sciences) with an excitation filter at 400 nm, an emission filter at 510 nm. The total luminescence was measured using the same cells incubated with coelenterazine H for 10 min (Molecular Probes) (emission peak 485 nm). The BRET ratios were plotted as a function of the GFP/LUC fusion protein expression ratio, both fusion proteins expression being assessed with the same cells as described above, to take into account the potential variations in the expression of individual constructs from transfection to transfection. As a negative control, the Rluc-TIF1␤/GFP-NLS BRET pair evaluating random collision was used as in Ref. 14.

Transcriptional Assays
Transient transfections were done as described above. The firefly luciferase reporter plasmid pGL3-5xGal4 corresponds to pGL3 vector under the control of the SV40 promoter (Promega) and a multimerized Gal4 DNA binding sequence (AGGGTATATAATG) ϫ5. The Renilla luciferase vector phRluc-C1 (PerkinElmer Life Sciences) (20 ng) was co-transfected to normalize for transfection efficiency. The effector plasmid corresponds to pcDNA3.1 (Invitrogen) in which Gal4DBD-(1-147) was cloned and fused to TIF1␤, the KRAB domain or their mutants (see description above). The total amount of transfected DNA was kept constant by addition of the pcDNA3.1 empty vector. Histone deacetylase inhibitor treatments were done 24-h post-transfection either with 5 mM sodium butyrate (NaBut) (from a 3 M stock solution resuspended in PBS pH 7.5) or with 200 nM trichostatin A (from a 300 mM stock solution resuspended in ethanol). Cell lysates were prepared 48 h after transfection and split into two samples for determination of the luciferase activity and assessment of the level of protein expression by Western blotting. The equivalent of 35 g of cell lysates was processed for the luciferase activity using the Dual GloTM luciferase assay kit (Promega).

SUMO Modification of TIF1␤ in Vivo-
Considering that sumoylation is often associated with a change in the activity of transcriptional regulators, we wondered if TIF1␤ co-repressor could be a target of SUMO protein as previously shown for TIF1␣, another member of the TIF1 family (69). To determine whether TIF1␤ could be covalently modified by SUMO in mammalian cells, human 293T cells were transfected with HAtagged TIF1␤ in the presence or absence of GFP-SUMO1, considering that free SUMO1 is limited in the cells (70). Western blot using anti-TIF1␤ revealed that the addition of GFP-SUMO1 results in the appearance of several bands migrating above the unmodified TIF1␤ (Fig. 1A, lanes 3 and 4). These bands of reduced mobility are predicted to correspond to TIF1␤ species covalently linked to GFP-SUMO1. Accordingly, these SUMO-modified forms were lost in cells co-transfected with a SUMO-specific isopeptidase, SENP1 that cleaves SUMO from its protein substrate, but not in cells expressing its cata-lytically inactive form, SENP1C603S (Fig. 1B, lanes 3 and 4) (66). Further confirmation was obtained in immunoprecipitation experiments where five slow migrating Flag-HA-tagged TIF1␤ species revealed by an anti-FLAG antibody were also detected by an anti-GFP antibody recognizing the GFP-SUMO1-modified TIF1␤ (Fig. 1D, lane 6 in left and right panels).
We then obtained evidence that endogenous TIF1␤ was a substrate for SUMO. Endogenous TIF1␤ sumoylation was detected with a specific anti-TIF1␤ antibody only when we overexpressed either Myc-SUMO1 or GFP-SUMO1, or when endogenous SUMO-modification of TIF1␤ was enhanced by using the E3 ligase, PIASy (Fig. 1E, lanes 2, 4, and 6). This suggests that only a small percentage of TIF1␤ is normally covalently linked by SUMO at the steady state as also reported for other sumoylated proteins (71). Altogether, these results clearly demonstrate that TIF1␤ is strongly sumoylated in mammalian cells overexpressing SUMO1 and that co-expression of The input lanes correspond to 1.25% of the whole cell extracts used for the immunoprecipitation (IP). With the anti-GFP antibody, the signal obtained is proportional to TIF1␤ concentration and the number of GFP-SUMO molecules linked to TIF1␤-sumoylated species. E, sumoylation of endogenous TIF1␤ in the presence or absence of transfected Myc-SUMO1 (6 g), GFP-SUMO (2 g), or Flag-PIASy (1.5 g) revealed by immunoblot of 293T whole cell extracts. Arrowheads point to the unmodified TIF1␤. In D, arrows point to the SUMO-modified TIF1␤ species. In A-C, the star identifies TIF1␤ modified by endogenous SUMO. NS, signifies a nonspecific cross-reacting band. The calibrated molecular mass of prestained marker proteins is indicated in kDa.
components of the SUMO pathway can regulate TIF1␤ SUMO modification.
Identification of TIF1␤ SUMO Modification Sites-Depending mostly on the exposure times, four to five sumoylated species were easily detected in cells co-transfected with TIF1␤ and either GFP-SUMO1 or GFP-SUMO2 (Fig. 1C). Thus, at least four main sumoylated sites are predicted to be present on TIF1␤. SUMO1 is less likely to give rise to polysumoylation chain formation in vivo than SUMO2 because its sequence does not include a bona fide SUMO acceptor lysine involved in SUMO chain formation (30). Accordingly, even if the sumoylation patterns obtained with SUMO1 and SUMO2 were similar (Fig. 1C), our studies to identify TIF1␤ sumoylation sites were done with SUMO1. TIF1␤ primary sequence encodes 42 lysine residues with only two being part of a classical sumoylation consensus (Lys 554 and Lys 676 ). In an attempt to delimit the TIF1␤ region(s) subjected to sumoylation, C-terminal deletion mutants were generated and subjected to sumoylation in vivo either in cells co-transfected with GFP-SUMO1 or not (Fig. 2, A  and B). To ensure that all deletion mutants are efficiently targeted to the nucleus, deletion constructs were all fused to Gal4 DNA binding domain (Gal4DBD), which includes a nuclear localization signal. The sumoylation pattern of deletion mutants was compared with that of wild-type TIF1␤, also fused to Gal4DBD. Analysis of the number and intensity of sumoylated species observed for each deletion mutant suggest that sumoylation sites are progressively lost from ⌬1 to ⌬4 deletion mutants (Fig. 2, A and B). Only the unmodified TIF1␤ was detected with the ⌬4 mutant suggesting that all sumoylation sites are located in the C-terminal part of TIF1␤ between amino acids 482 and 834. This region of around 350 amino acids contains 13 lysine residues, including Lys 554 and Lys 676 (Fig. 2, B and C).
To determine which of the 13 C-terminal lysine residues of TIF1␤ serve as a substrate for sumoylation, each of them were mutated to arginine residue (R) individually or in combination. A detailed analysis of the sumoylation profile of these mutants suggested that the first two bands migrating above the unmodified TIF1␤ correspond to monosumoylated forms exhibiting different apparent molecular weight as also reported for other sumoylated proteins (38,72). Accordingly, as seen in Fig. 2D  (compare lanes 1 and 2), the second band above the unmodified TIF1␤ was strongly reduced by mutating K554 and the faint remaining signal was completely abolished in the double K554R/K575R mutant (Fig. 2D, lane 3). This suggests that two distinct populations of monosumoylated species comigrate, one being monosumoylated on Lys 554 and one on Lys 575 . Considering the relative abundance of these monosumoylated species, Lys 554 appears as a major sumoylation site compared with Lys 575 . Two additional major sumoylation sites were evidenced with the use of the K779R and double K779R/K804R mutants (compare lane 4 with lanes 5 and 6). These mutants led to significant reduction in the first sumoylation band migrating just above the unmodified TIF1␤. Only a faint doublet of bands closely migrating with the species monosumoylated on Lys 779 or Lys 804 were seen with the K779R/K804R mutant (lane 6). These remaining bands (less than 5% of the sumoylation detected with wild-type TIF1␤) are likely to correspond, respectively, to the monosumoylation of Lys 750 (lower band) and Lys 676 (higher band) as demonstrated with mutants containing five (5KR mutant) and six (6KR mutant) mutated lysines (Fig. 2D, lanes 7-9). The above results demonstrate that TIF1␤ includes six SUMO acceptor sites. Our deletion and punctual mutant results also suggest that the bands migrating above the monosumoylated species correspond to multisumoylated forms of TIF1␤ with potentially two to six SUMO moieties linked to any of the different acceptor lysines (scheme in Fig.  2D). The TIF1␤ species expected to contain five SUMO moieties was seen on longer exposure of some gels. However, we have never observed the one predicted to contain six SUMO moieties presumably because this modified species represents a very minor product not detectable at the level of sensitivity of our assay or because the sumoylation of some lysine residues is mutually exclusive. Consistent with the multisumoylation of TIF1␤, the removal of any sumoylation site by mutation of one target lysine leads to a decrease in the intensity or in the disappearance of the highest migrating band. Altogether, our results show that among the six potential sumoylation sites, three of them are major sumoylation sites (Lys 554 , Lys 779 , Lys 804 ) and three are minor sumoylation sites (Lys 575 , Lys 676 , and Lys 750 ).
Regulation of TIF1␤ Transcriptional Repression by Sumoylation-Knowing that TIF1␤ is a strong transcriptional co-repressor involved in gene silencing, we assessed the potential impact of TIF1␤ sumoylation on its transcriptional repressive properties. We used a gene reporter assay to compare the transcriptional repressive activity of wild-type TIF1␤ with that of various sumoylation-deficient mutants containing target lysines mutated to arginines. Because TIF1␤ needs to be recruited to DNA to repress transcription (2-4), both wildtype and SUMO-deficient mutant TIF1␤ effectors were fused to Gal4DBD. This fusion protein recognizes the Gal4 response element (5xGal4UAS) placed in front of the SV40 strong promoter that drives the expression of the luciferase reporter gene (Fig. 3A). As seen in Fig. 3C, the repressive activity of wild-type TIF1␤ is dose-dependent, reaching a maximum of ϳ20-fold. At comparable maximal doses, TIF1␤ 4KR (K554R/K575R/ K779R/K804R), that includes mutations of the three major sumoylation sites, represses trancription 4 -5-fold less than wild-type TIF1␤. At doses of transfected TIF1␤ varying in the 10-fold range, a change limited to ϳ1.5-fold was observed in the repressive activity elicited by TIF1␤ 4KR in contrast to ϳ4-fold increase obtained with the wild type. This suggests that the TIF1␤ sumoylation-deficient mutant is unable to fully repress the luciferase reporter in our assay. A comparison of the transcriptional repression activity of various mutants that impair sumoylation was also carried out (Fig. 3B). Interestingly, TIF1␤ repressive activity was progressively lost by increasing the number of mutated sumoylation sites as observed in conditions where all sumoylation deficient mutants are expressed at comparable levels. These above results strongly suggest a correlation between TIF1␤ repressive activity and its sumoylation capacity.
While controversial, some studies suggested that the repressive activity of TIF1␤ was partially sensitive to histone deacetylase (HDAC) inhibitors (15,17,73). Our results using wild-type TIF1␤ or a deletion mutant (⌬N-TIF1␤), containing the previ- Mutated residues used in this study are indicated. C, illustration of the amino acid context surrounding TIF1␤ sumoylation target lysines. Target lysine residues are in bold and those that are part of a classical KXE consensus sequence are underlined. D, sumoylation profiles of lysine to arginine mutants of Gal4DBD-TIF1␤ (2-3 g) in the presence of GFP-SUMO1 (2 g) revealed by immunoblot of 293T whole cell extracts. Arrowhead points to the unmodified TIF1␤. Arrows point to the bands containing monosumoylated species, and the modified lysine is indicated. A schematization of the sumoylation profile is provided on the right. Brackets identify monoand multisumoylated species, and bands in parentheses correspond to TIF1␤ species containing potentially five (observed on some blots) or six SUMOs (never detected).
ously characterized repressive region of TIF1␤ including the six SUMO-targeted lysines (scheme in Fig. 2B), confirmed that treatment with sodium butyrate (Fig. 3D) and another inhibitor for class I and II HDAC, Trichostatin A (data not shown), reduces the repressive activity of TIF1␤ by ϳ30 -50% (HDACdependent repression) (Fig. 3D). To investigate if there is a potential functional link between HDAC recruitment and TIF1␤ sumoylation, we compared the sodium butyrate sensitivity of ⌬N-TIF1␤ and its 6KR derivative that is refractory to sumoylation. Clearly, the residual 1.5-fold repressive activity of the ⌬N-TIF1␤ 6KR mutant was not modulated by the HDAC inhibitor suggesting that the recruitment of HDAC is SUMOdependent (Fig. 3D, left panel). Furthermore, comparative analysis of the repression of ⌬N-TIF1␤ with that of the ⌬N-TIF1␤ 6KR sumoylation-less mutant in the presence of HDAC inhibitor revealed that the SUMO-related repression (see arrow on Fig. 3D) involves both HDAC-dependent and HDAC-independent components (see dotted arrows on Fig. 3D).
Because TIF1␤ acts as a co-repressor indirectly recruited to DNA via DNA-binding KRAB multifinger proteins (2, 23), we also compared the co-repressive activity of wild-type TIF1␤ and its sumoylation-deficient mutant following its indirect recruitment to DNA using a Gal4-KRAB fusion protein (Fig.  4A). Interestingly, in the presence of Gal4-KRAB, the co-repressive activity elicited by the endogenous TIF1␤ was enhanced by up to 30% by adding increasing amounts of wildtype TIF1␤ (Fig. 4A). In contrast, the co-repressive activity was decreased by 10% in the presence of comparable amounts of 4KR mutant suggesting that this sumoylation-deficient mutant exerts a dominant negative effect over the repression induced by endogenous TIF1␤ proteins (Fig. 4A). Thus, TIF1␤ co-repressor activity is clearly impaired by mutation of its major sumoylation sites. This loss of co-repressive activity of the 4KR mutant is not caused by a change in its subcellular localization because TIF1␤ sumoylation-deficient mutants display the same nuclear localization patterns as the wild-type TIF1␤ (observed Effectors proteins correspond to wild-type TIF1␤ or its lysine to arginine mutants. B and C, transcriptional activity of TIF1␤ wild type and sumoylationless mutants directly recruited to the luciferase reporter. All transient transfections were done in 293T cells using 200 ng of reporter plasmid and a fixed amount of Gal4DBD-TIF1␤ effectors as indicated in B or increasing amounts of either Gal4DBD-TIF1␤ WT or K554R/K575R/K779R/K804R (4KR) mutant (from 25 to 300 ng) (C). Fold repression represents the ratio of luciferase activity measured for the reporter alone to the activity measured in the presence of the indicated effector proteins after normalization for transfection efficiency. Error bars represent the S.D. for at least three independent experiments performed in duplicate or triplicate. D, effect of sodium butyrate, a class I and II HDAC inhibitor, on the transcriptional activity of TIF1␤ wild-type and its sumoylation-less mutant. Transfections were done using 200 ng of reporter with 500 ng of the Gal4DBD-⌬NTIF1␤ or 100 ng of the Gal4DBD-TIF1␤ constructs for untreated cells. Considering the observed ϳ5-fold increase in the expression of TIF1␤ and the normalizing Renilla luciferase in sodium butyratetreated cells, the amount of Gal4DBD-effectors (Gal4DBD-⌬NTIF1␤ 100 ng or Gal4DBD-TIF1␤ 20 ng) was adjusted to obtain similar protein expression in treated and untreated cells. Cells were treated with sodium butyrate 24 h post-transfection and recovered 24 h later for transcriptional assays. Error bars represent the S.D. for at least three independent experiments performed in duplicate. SUMO-related repression (arrow) is composed of HDAC-dependent and HDAC-independent components (dotted arrows). For all assays (B-D), the expression level of TIF1␤ was assessed by immunoblots of 293T whole cell extracts from pooled duplicate or triplicate samples derived from one independent experiment.
by immunolocalization with Flag-HA-or GFP-tagged constructs, data not shown) nor to an impaired interaction with the KRAB domain as shown by a BRET interaction assay in living cells (Fig. 4B) (14, 74). This is evidenced by the fact that similar BRET saturation curves were obtained when a fixed amount of Rluc-TIF1␤ or -TIF1␤ 4KR was co-expressed with increasing amount of GFP-KRAB-NLS (Fig. 4B). As typically seen when specific interactions occur, these curves showed an increase in the energy transfer (BRET 2 ratio) between the RLuc fusion donor and GFP fusion acceptor until a plateau is reached when the amount of acceptor is no longer limiting compared with the donor. Noticeably, the BRET 2 ratio of Rluc-TIF1␤ or -TIF1␤ 4KR at the maximal dose of GFP-KRAB-NLS is 4-fold higher than the one obtained with the negative control BRET pair Rluc-TIF1␤L307P/GFP-KRAB-NLS. The TIF1␤ L307P mutant was previously shown to have an abrogated interaction with a KRAB domain in an in vitro experiment using the RBCC fragment of TIF1␤ (12). We also ruled out that the differential co-repressive activity elicited by TIF1␤ and its 4KR mutant was caused by an impaired homodimerization because they exhibit similar BRET saturation curves (Fig. 6A).
Noticeably, the increased repression observed when overexpressing wild-type TIF1␤ in the presence of Gal4-KRAB correlated with the appearance of a higher molecular weight band characteristic of TIF1␤ sumoylated by endogenous SUMO (Fig.  4A). This result suggests that the Gal4-KRAB fusion favors TIF1␤ sumoylation as explored in detail below (Fig. 5). In contrast, the endogenous sumoylation of TIF1␤ transfected alone remains undetectable in our transcriptional assays (Fig. 3).
Regulation of TIF1␤ Sumoylation by the KRAB Domain of Multifinger Proteins-Based on observations mentioned above, we investigated the possibility that the KRAB domain could modulate the sumoylation of TIF1␤. Strikingly, co-transfection of wild-type TIF1␤ together with a prototypical KRAB domain (derived from ZNF74 zinc finger protein) significantly boosts the sumoylation level of TIF1␤ in condition in which GFP-SUMO was limiting (Fig. 5A, left panel compare lanes 1 and 2). Such enhancement was neither observed using a non-repressive truncated version of this KRAB domain (KRAB tr), which does not interact in vivo with TIF1␤ (14,63) nor with the TIF1␤ L307P mutant, which failed to bind the KRAB domain as shown in vitro (12) and confirmed here in living cells by BRET (Figs. 4B and 5A, left and middle panels). Importantly also, in the presence of Gal4-KRAB, a higher molecular weight band characteristic of TIF1␤ sumoylated by endogenous SUMO appeared above the endogenous TIF1␤ (Fig. 5B). These results suggest that the sumoylation of TIF1␤ co-repressor is favored by its interaction with the KRAB domain. To confirm that such a domain can regulate sumoylation of TIF1␤ in the context of a full-length KRAB multifinger protein, we used two isoforms of a prototypical KRAB multifinger protein, ZNF74, that include the full-length repressive KRAB domain (ZNF74-II isoform) and the non-repressive-truncated KRAB domain (ZNF74-I isoform) used above (63). Only ZNF74-II interacts with TIF1␤ as previously shown in living cells by BRET (14). Clearly also, only ZNF74-II repressive isoform increases TIF1␤ sumoylation (Fig.  5A, right panel). Taken together, our results strongly suggest that the interaction of TIF1␤ with KRAB domain-containing zinc finger proteins positively regulates its SUMO modification and consequently its repressive activity.

Requirement RBCC Oligomerization Motif and PHD Finger
Integrity for Efficient TIF1␤ Sumoylation-TIF1␤ and other members of the TRIM family are suggested to function as oligomers homodimerizing via their RBCC motif (13,14,(75)(76)(77). Interestingly, we noticed that the sumoylation of TIF1␤ RBCC punctual mutants, L271P and L307P, was significantly reduced compared with that of the wild-type TIF1␤ (Fig. 6A). These mutants were previously found to be deficient for homodimerization in vitro and to exhibit a reduced homodimerization in cell extracts while only L307P was found to be deficient for interaction with the KRAB domain as shown by Peng et al. (12). Because our BRET studies also indicate altered homodimerization properties of the RBCC mutants compared with the wildtype TIF1␤ (or the 4KR mutant) (Fig. 6A), we suspected that efficient sumoylation of TIF1␤ may require its oligomerization. In agreement with this, a HA-tagged deletion mutant lacking the RBCC dimerization motif, ⌬N-TIF1␤, did not show any significant sumoylation despite the presence of the six possible SUMO-acceptor sites (Fig. 6B, left panel). In contrast, a version of ⌬N-TIF1␤ fused to the heterologous Gal4 dimerization domain exhibits a ladder of sumoylated bands (Fig. 6B, lane 1 in middle panel) as typically observed for the wild-type protein. Considering these results and the fact that HA-⌬N-TIF1␤ retains a nuclear localization (data not shown), this suggests that homodimerization of TIF1␤ is required for its sumoylation.
As demonstrated here, the RBCC motif of TIF1␤ appears dispensable for its sumoylation when replaced by a heterologous dimerization domain. Recently however, sumoylation of another RBCC motif-containing protein, PML, was found to depend on the integrity of part of its RBCC motif, the RING finger (78,79). Interestingly, Capili et al. (80) suggested structure similarity between PML RING finger and TIF1␤ PHD fin-ger which are both cysteine-rich, zinc-binding domains. To assess the potential involvement of the PHD finger in the sumoylation of TIF1␤, we generated mutants where the PHD finger structure was disrupted by mutating two amino acids responsible for zinc-binding (H648A, C651A) (80). Interestingly, the intensity of sumoylated species was strongly reduced when the PHD finger mutations were introduced either in the Gal4-⌬N-TIF1␤ or in the full-length TIF1␤ (Fig. 6B, middle  and right panels, lanes 1 and 2). The above results indicate that the oligomerization of TIF1␤ via its RBCC motif and the structural integrity of its PHD finger domain are essential for TIF1␤efficient sumoylation.

DISCUSSION
This study identified TIF1␤, the universal co-repressor of KRAB multifinger proteins as a sumoylation substrate, which can be either mono-sumoylated on one of its six targeted lysines or multisumoylated on a combination of these lysine residues.
Our results indicate that the sumoylation of TIF1␤ is an essential determinant for the repressive activity of this co-regulator of transcription and is a prerequisite for recruitment of histone deacetylase complexes. Interestingly, TIF1␤ sumoylation is shown to be dependent on its ability to homodimerize. Furthermore, it can be positively modulated by its interaction with KRAB domain-containing proteins. These results lead us to suggest that KRAB multifinger proteins not only recruit TIF1␤ co-repressor to target genes but also increase its repressive activity through enhancement of its sumoylation.
TIF1␤, a Substrate Rich in Lysine Residues Targeted by SUMO-Using TIF1␤ lysine-acceptor and C-terminal deletion mutants, we provided evidence that TIF1␤ possesses six SUMO-acceptor sites involved in multisumoylation events. Noticeably, in the conditions of our assays using SUMO1, we ruled out that TIF1␤ is polysumoylated through the formation of SUMO chains on individual lysine residues. While the vast majority of SUMO target proteins were described to contain one or two SUMO modification sites, TIF1␤ appears as a multiacceptor protein for SUMO such as PML, Daxx, and N-CoR (44,81,82). Interestingly, all TIF1␤ sumoylation sites are positioned in its repressive region which contains a PHD finger and a bromo domain. As it has been observed for several transcription factors, SUMO modification frequently occurs within inhibitory or repression domains (83)(84)(85).
Only two of the six TIF1␤ sumoylation sites are part of the classic sumoylation consensus sequence (KX(E/D)). Among these two sites, one is characterized here as a major (Lys 554 ) and one as a minor sumoylation site (Lys 676 ). No specific consensus motif could be identified surrounding the four other SUMO acceptor sites, characterized as two major (Lys 779 , Lys 804 ) and two minor (Lys 575 , Lys 750 ) sumoylation targets. During the completion of this manuscript, three of the six sumoylation sites characterized here were also identified by others (86).
Analysis of sumoylation profiles revealed that the position of the targeted lysine within TIF1␤ primary sequence strongly influences the migration of each TIF1␤ monosumoylated species on denaturating polyacrylamide gel. Noticeably, the closer the target lysine is to the terminal end of the protein, the smaller is the apparent molecular weight of the monosumoylated species (Fig.  2D, scheme).
TIF1␤ appears as a substrate heavily sumoylated under conditions where it is overexpressed in the presence of SUMO. As for most sumoylated proteins, it was difficult to obtain evidence of the sumoylation of endogenous TIF1␤ by endogenous SUMO. However, we revealed such endogenous modification in cells transfected with either a SUMO E3 ligase, PIASy, or a KRAB domain reported here to enhance TIF1␤ sumoylation.
TIF1␤ Sumoylation, a Prerequisite for Transcriptional Repression-Our functional assays strongly suggest that the transcriptional repression mediated by TIF1␤ requires its post-translational modification by SUMO, an observation also made for other transcriptional regulators such as Sp3, Elk-1, and BKLF transcription factors (38,46,49) or co-regulators such as p300, CtBP, and N-CoR (34,44,87). Reducing the number of TIF1␤ SUMO acceptor sites by mutating single, double, and multiple sumoylation target lysine residues progressively decreased TIF1␤-dependent transcriptional repression to near a basal repressive level. Remarkably, a TIF1␤ sumoylation-deficient mutant in which all the target lysine residues were mutated to arginine residues prevents almost completely the TIF1␤-dependent repression activity. This is suggesting that the TIF1␤ repressive ability is mostly dependent on its capacity to be covalently modified by SUMO proteins.
Because no obvious difference has been observed in the subcellular localization of wild-type TIF1␤ and its various sumoylation-deficient mutants (TIF1␤ 4KR and -6KR) (data not shown), it is unlikely that a variation in their subcellular distribution contributes to their differential transcriptional repres-FIGURE 6. TIF1␤ sumoylation status depends on the integrity of its homodimerization domain and of its PHD finger domain. A, TIF1␤ optimal sumoylation requires an intact RBCC motif. Sumoylation profiles of wild-type HA-tagged-TIF1␤ (1.5 g) and coiled-coil mutants (L271P or L307P, 2.5 g) in the presence of GFP-SUMO1 (1 g) revealed by immunoblot of 293T whole cell extracts. BRET 2 titration analysis showing homodimeric interaction of TIF1␤ wild type, its coiled-coil domain mutants, or its sumoylation-less mutant in living 293T cells. The BRET 2 pairs tested were Rluc-TIF1␤/GFP-TIF1␤ (F), Rluc-TIF1␤ L271P/GFP-TIF1␤ L271P (छ), Rluc-TIF1␤ L307P/GFP-TIF1␤ L307P (‚), Rluc-TIF1␤ 4KR/GFP-TIF1␤ 4KR (E), and the negative control pair Rluc-TIF1␤/GFP-NLS (*). Briefly, 293T cells were transfected with fixed amount of Rluc fusion protein (in the range of 1-2 g) and increasing amounts of GFP fusion protein (from 0.05 to 6 g). For each pair, data of at least three independent experiments performed in duplicate were pooled. The data were represented as described under "Experimental Procedures." B, TIF1␤ sumoylation depends on its homodimerization and on the integrity of the PHD finger domain. Comparison of the sumoylation profiles obtained in 293T cell extracts between wild-type HA-tagged-TIF1␤ (2 g) and an HA-tagged-⌬N-TIF1␤ mutant devoid of the RBCC homodimerization motif (2 g) (left panel), Gal4-⌬N-TIF1␤ (10 g) and its derivative mutated in the PHD finger (middle panel) as well as Gal4-TIF1␤ and its derivative mutated in the PHD finger as shown in the right panel. GFP-SUMO1 (1 g in right panel or 2 g) was co-transfected with TIF1␤ constructs. For each experiment, the total GFP fluorescence was measured to confirm equivalent GFP-SUMO1 expression level. Arrowheads point to the unmodified TIF1␤.
sive activity. Previous studies trying to decipher the mechanisms involved in the repressive activity of sumoylated proteins have suggested that SUMO displays an intrinsic repressive activity when directly targeted to DNA as a Gal4-SUMO fusion (49,50). Furthermore, it was shown that the covalent attachment of SUMO promotes recruitment of HDAC complexes and thus inhibition of transcription both in conditions where SUMO is fused with transcriptional regulators to generate a chimera (35,49) or is conjugated to the ⑀-NH 2 group of a lysine residue from target proteins (34,40,46). Importantly, we found that the SUMO-dependent repressive activity of TIF1␤ is accounted both by HDAC-dependent and HDAC-independent mechanisms. This is consistent with the recent results of Lee et al. (86) showing that TIF1␤ sumoylation correlates with both an attenuation of histone H3 acetylation (H3-K9, -K14) and with an enhancement of histone H3 di-methylation (H3-K9) at a specific TIF1␤-targeted promoter. Because TIF1␤ is known to interact with Mi2-␣, a zinc finger helicase, which is part of NuRD histone deacetylase complexes, as well as with the SETDB1 methyltransferase (15,16), it is tempting to speculate that the recruitment of these repressive enzymatic activities is modulated by TIF1␤ sumoylation status. This may occur as a result of a differential interaction of the sumoylated and unsumoylated forms of TIF1␤ with these enzymatic complexes. Interestingly, a recent article identified SETDB1 as a protein, which directly interacts with SUMO2 (47). Because the interaction of TIF1␤ and SETDB1 has been previously characterized by yeast-two hybrid and co-immunoprecipitation experiments (16), it remains to be determined whether this interaction is direct or indirect and if it requires TIF1␤ sumoylation.
TIF1␤ Sumoylation Is Regulated by Its Interaction with KRAB-containing Proteins and Requires TIF1␤ Oligomerization-TIF1␤, a ubiquitous and abundant protein, is believed to function as the obligate co-repressor for KRAB multifinger proteins, which are mostly expressed in stage-specific and tissue-specific manner during vertebrate development (3). In the present study, we demonstrate that KRAB domain-containing proteins positively modulate TIF1␤ sumoylation, because of their direct interaction with TIF1␤ RBCC motif. Thus, the recruitment of TIF1␤ to gene promoters regulated by KRAB multifinger proteins and its subsequent KRAB-induced sumoylation may represent a specific and unique way to enhance the repressive activity of TIF1␤ at selected promoters. This could explain why only a small proportion of endogenous TIF1␤ is SUMO modified at a steady state. It must be acknowledged that most of our work was done with transient transfections and Gal4-based systems. Thus, further studies should confirm the importance of TIF1␤ sumoylation in gene repression and its regulation by KRAB-containing proteins at promotors in chromatinized and natural chromatin contexts.
Whereas it is generally recognized that TIF1␤ can repress gene transcription when targeted to DNA through KRAB domain interaction, a few studies have found that TIF1␤ could also behave as a transcriptional co-activator (88,89). More specifically, TIF1␤ has been characterized as a co-activator when recruited by transcriptional regulators such as C/EBP␤, the glucocorticoid nuclear hormone receptor and the TRIP-Br family members (88,89). Thus, one could propose that specific protein interactors such as KRAB multifinger proteins may regulate TIF1␤ repressive activity by stimulating its sumoylation, a prerequisite for efficient recruitment of repressive enzymatic complexes. In contrast, other protein complexes, such those described above, may prevent TIF1␤ sumoylation and switch this co-regulator into a co-activator with the capacity to recruit activating enzymatic complexes. In such case, sumoylation may represent a critical determinant by which TIF1␤ is regulated.
At least two hypotheses, not mutually exclusive, can be proposed to explain how the KRAB domain increases TIF1␤ sumoylation. First, it could induce a TIF1␤ conformational change favoring its interaction with the sumoylation machinery and/or a better accessibility of TIF1␤ sumoylation sites. We ruled out however the possibility that the KRAB domain itself recruits UBC9, the E2 conjugating enzyme for SUMO (data not shown). Second, because the present study demonstrates that efficient sumoylation of TIF1␤ requires its oligomerization, an attractive possibility, is that KRAB domains may enhance TIF1␤ sumoylation by reinforcing its oligomerization. Consistent with this hypothesis, another group suggested that KRAB domain can stabilize TIF1␤ homo-oligomers formation in vitro (12). Our previous results also demonstrated that a KRAB domain-containing protein, ZNF74, can bind to TIF1␤ oligomers in living cells suggesting that oligomers represent the transcriptionally active species (14). Interestingly, a dimerization-dependent sumoylation has also been reported for at least two other proteins, namely HDAC4 and PML (90,91). In the case of PML, it was recently suggested that its dimerization favors its autosumoylation, a process requiring the presumed intrinsic SUMO E3 ligase activity of the PML RING finger (78,79). Since we found here that TIF1␤ efficient sumoylation also requires the integrity of its PHD finger, a domain reported to be structurally related to the RING finger of PML (80), further studies will determine if the TIF1␤ PHD finger may exhibit a SUMO E3 ligase activity allowing the auto-sumoylation of this strong co-repressor of transcription.
Whereas the number of SUMO target proteins is still growing, the physiological determinants regulating their specific sumoylation remain unknown. We report here that members of a family of transcription factors recruiting specifically the corepressor TIF1␤ can modulate its sumoylation and consequently its repressive activity. Further studies will determine if sumoylation may antagonize other post-translational modifications of TIF1␤ co-regulator.