Repression and Coactivation of CCAAT/Enhancer-binding Protein ϵ by Sumoylation and Protein Inhibitor of Activated STATx Proteins*

CCAAT/enhancer-binding protein ϵ (C/EBPϵ) is a neutrophil-specific transcription factor whose activity is controlled by juxtaposed activating and regulatory domains. We previously determined that the function of the major regulatory domain (RD1) in C/EBPϵ was dependent on the integrity of a five-amino acid motif that was identical to the recognition site for members of the small ubiquitin-like modifier (SUMO) family of ubiquitin-related proteins. We show here that the SUMO attachment site (the regulatory domain motif) is necessary and sufficient both for the intrinsic inhibitory function of RD1 and for coactivation by PIASxα and PIASxβ, two members of the protein inhibitor of activated STAT (PIAS) family of SUMO E3 ligases. PIASxβ was a more potent coactivator than PIASxα of both full-length C/EBPϵ and fusion proteins containing the N-terminal portion of C/EBPϵ, whereas PIASxα was more active on fusion proteins containing a heterologous activation domain. Two modes of coactivation were observed, one that was dependent on the integrity of the RING finger (RF) domain and was shared by both PIASxα and PIASxβ and a second mode that was independent of the RF and was only observed with PIASxβ. Sumoylation of C/EBPϵ was enhanced by coexpression of PIASxα, suggesting that this modification is associated with the enhanced activity of the target protein. These results suggest that a complex interplay of accessory factors, including SUMO and PIAS proteins, modulates the activity of C/EBPϵ.

The CCAAT/enhancer-binding proteins form a subgroup within the basic region/leucine zipper superfamily of transcriptional regulatory proteins (1,2). There are six members of the C/EBP 1 family, four of which (C/EBP␣, C/EBP␤, C/EBP␦, and C/EBP⑀) are transcriptional activators that control the differentiation and function of cells in a large number of tissues. C/EBP⑀ was first identified based on sequence similarity to C/EBP␣ (3) and is primarily expressed in myeloid cells within the hematopoietic system (4 -6). C/EBP⑀ expression increases during neutrophil differentiation (7), and it is also expressed in macrophages (6,8). C/EBP⑀ is implicated in the regulation of a number of neutrophil-and macrophage-specific genes (6, 8 -13). Targeted disruption of the C/EBP⑀ gene in mice caused a block in neutrophil differentiation that resulted in the production of morphologically and functionally abnormal neutrophils (14). In addition, mutations in the human C/EBP⑀ gene have been detected in patients with the rare congenital disease, neutrophil-specific granule deficiency (15)(16)(17). The similar consequences of C/EBP⑀ mutations in mice and humans suggest that disruption of C/EBP⑀ activity is a major causative event in neutrophil-specific granule deficiency.
The C/EBPs are characterized by a highly conserved carboxyl-terminal DNA binding domain that directs sequence-specific binding to the palindromic sequence 5Ј-ATTGCGCAAT-3Ј and variants thereof (1,2). Each C/EBP can form homo-and heterodimers in all combinations (3) and can heterodimerize with other members of the basic region/leucine zipper superfamily. The amino-terminal portions of C/EBP␣, C/EBP␤, C/EBP␦, and C/EBP⑀ contain juxtaposed activation and inhibitory domains that control their transcriptional regulatory activities (18 -22). We recently identified two inhibitory domains in the C/EBP⑀ polypeptide that we named regulatory domains 1 and 2 (RD1 and -2). The RD1 domain functions as an autonomous transcriptional inhibitory domain when attached to heterologous transcriptional activation domains and is functionally similar to the attenuator domain of C/EBP␣ (18) and the RD1 domain of C/EBP␤ (20). Sequence comparisons of RD1 domains from each C/EBP revealed a conserved five-amino acid motif with the consensus sequence (I/V/L)KXEP. We named this sequence the regulatory domain motif (RDM) and noted its similarity both to the synergy control element initially found in members of the nuclear hormone receptor superfamily (23,24) and to the consensus sequence for attachment of SUMO proteins (25).
The SUMO family consists of four members, SUMO-1 to -4 (25)(26)(27). SUMO proteins can be covalently attached to target proteins via an enzymatic mechanism analogous to, but using distinct enzymes from, ubiquitylation. SUMO proteins are synthesized as precursors that must first undergo proteolytic processing to produce the mature polypeptide. The mature protein can then be attached to lysine residues in the target protein, usually within the consensus sequence KXE (where and X represent a hydrophobic and any amino acid, respectively), via a two-step process involving an E1 activating enzyme complex and an E2 conjugating enzyme. Although sumoylation in vitro * This work was supported by an Established Investigator Award from the American Heart Association and grants from the CH Foundation, South Plains Foundation, and SouthWest Cancer Center at UMC (to S. C. W.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
** To whom correspondence should be addressed: requires only E1 and E2 enzymes, three classes of proteins with E3 ligase activity have recently been identified (28). These proteins, which include members of the protein inhibitor of activated STAT (PIAS) family, the nuclear pore-associated RanBP2 protein, and Pc2, a member of the polycomb family, may enhance the efficiency or selectivity of the sumoylation process in vivo. We showed previously that the RDM of C/EBP⑀ was an attachment site for SUMO-1 and that the three other C/EBPs that contain RDM-like sequences (C/EBP␣, C/EBP␤, and C/EBP␦) were also sumoylation targets (22). This modification has subsequently been confirmed for both C/EBP␣ (29) and C/EBP␤ (30). A growing number of proteins have been identified as SUMO targets and include numerous proteins that function in transcriptional regulation and other nuclear processes. The consequences of sumoylation include modification of subcellular or subnuclear location, protein stability, or activity and appear to be dependent on the identity of the target protein. Since sumoylation of C/EBPs occurs within a transcriptional inhibitory domain, it has been proposed that sumoylation inhibits the activity of C/EBPs. However, we previously demonstrated that enhancing sumoylation of C/EBP⑀, at least in the context of a Gal4 fusion protein, actually increased its activity, and thus the ultimate consequence of sumoylation of C/EBP⑀ (and presumably other C/EBPs) remains to be clearly determined.
Two models have been proposed to explain the mechanism by which the RD1 domain of C/EBP proteins inhibits their activity. An intramolecular interaction model was proposed based on studies on C/EBP␤ in which interactions between the RD1 domain and the N-terminal activation domain prevented access of the AD to the transcriptional coactivators and/or components of the general transcriptional machinery (19,20). More recently, we proposed an intermolecular interaction model based on our studies on C/EBP⑀ and suggested that SUMO attachment may play a critical role in controlling interactions between the RD1 domain of C/EBP⑀ and transcriptional coregulators (22). In this report, we have extended our earlier studies on the regulation of C/EBP⑀ by sumoylation to attempt to define the mechanism by which C/EBP⑀ activity is controlled.
Expression vectors encoding C/EBP⑀-Myc-His 6 fusion proteins were constructed in pcDNA3.1 MHB(Ϫ) (Invitrogen) by PCR amplification of the full C/EBP⑀ coding sequence from pMEX C/EBP⑀ p32, pMEX C/EBP⑀ p29, pMEX C/EBP⑀ p32 (K121A), pMEX C/EBP⑀ ⌬RD1, and pMEX C/EBP⑀ ⌬RD2 (21). These PCRs used the forward primer 5Јcaacccctcactcggcgcgcc-3Ј, which anneals upstream of a BamHI restriction site prior to the C/EBP⑀ ATG codons, and the reverse primer 5Ј-gacggcaagcttgggctgcagcccccgacacc-3Ј, which inserts a HindIII site in place of the C/EBP⑀ stop codon. These PCR products were cut with BamHI and HindIII and ligated into pcDNA3.1 MHB (Ϫ) cut with the same enzymes. pCMV-FLAG-SUMO-1 was a generous gift from Dr. Giuseppina Nucifora (University of Illinois, Chicago, IL). Expression vectors encoding FLAG-tagged wild type and RING finger deletion mutant PIASx proteins were described previously (33). A FLAG-JunD expression vector was a kind gift of Dr. Curt Pfarr (TTUHSC). The C/EBP-responsive luciferase reporter (DE1) 4 -(Ϫ35Alb) Luc contains four copies of a C/EBP-response element from the albumin promoter upstream of the minimal promoter region from the same gene (20).
Cell Culture and Transfections-COS-1 cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% Cosmic Calf Serum (Hyclone, Logan, UT), 50 units/ml penicillin, and 50 g/ml streptomycin at 37°C and 5% CO 2 in a humidified incubator. Transient transfections were performed with the Effectene transfection reagent (Qiagen, Valencia, CA) according to the manufacturer's recommendations. Transfections with Gal4-based plasmids were performed with 100 ng of the Gal4-responsive G5E1bLUC reporter plasmid, 100 ng of the Gal4-based expression plasmid, and 50 or 100 ng of the PIASx expression plasmid. Transfections with C/EBP⑀-based plasmids were performed using 80 ng of the (DE1) 4 -(Ϫ35Alb) Luc reporter plasmid, 80 ng of the C/EBP⑀ expression plasmid, and 80 or 160 ng of the PIASx expression plasmid. The total DNA concentrations in each transfection were normalized using empty pCMV expression plasmid. Transfections for Western extract preparation contained 3 g of the appropriate expression plasmid. All transfections were performed at least three times in triplicate. Luciferase values were measured using the Luciferase Assay Kit (Promega, Madison, WI). Expression levels of all proteins were assessed by immunoblotting to ensure that differences in activity were not due to differences in steady-state levels of each protein (see also Ref. 22).
In Vivo Sumoylation Assays-COS-1 cells were transfected with various combinations of pcDNA3.1 C/EBP⑀-Myc (2.0 g), pCMV-FLAG-SUMO-1 (0.5 g), and pCMV-FLAG-PIASx (0.5 g), and whole cell extracts were prepared after 48 h as described previously (22). SUMOmodified and unmodified forms of C/EBP⑀ were detected by immunoblotting using a Myc-specific monoclonal antibody (9E10; Santa Cruz Biotechnology, Inc., Santa Cruz, CA). The proportion of total C/EBP⑀ protein that was sumoylated was calculated by densitometric analysis of the intensities of the appropriate bands on autoradiographic film using the Fluorchem program (Alpha Innotech Co., San Leandro, CA).
Pull-down Assays-Nuclear extracts were prepared from COS-1 cells transfected with different combinations of pDNA3.1 MHB C/EBP⑀, pCMV-FLAG-PIASx␣, pCMV-FLAG-JunD or G4-V (1.5 g of each plasmid). The expression levels of each of the proteins were analyzed by Western blotting. Immunoprecipitations were carried out by incubating equal amounts of nuclear extracts with 4 g of rabbit polyclonal anti-FLAG antiserum (Sigma) coupled to protein G-magnetic beads (Dynal Biotech, Brown Deer, WI) in immunoprecipitation buffer (150 mM NaCl, 0.5% Nonidet P-40, 50 mM Tris-HCl), pH 7.5, for 4 h at 4°C. The immunoprecipitation buffer was supplemented with 10 mM N-ethylmaleimide, 0.5 mM phenylmethylsulfonyl fluoride, and a 1:100 dilution of protease inhibitor mixture (Calbiochem). Pull-down assays were performed by mixing nuclear extracts with 50 l of Ni 2ϩ -NTA magnetic agarose beads (Qiagen) in the presence of 10 mM imidazole for 4 h at 4°C. Immune or Ni 2ϩ -NTA-bound complexes were isolated by placing tubes in a magnetic stand, and complexes were washed 3-5 times with immunoprecipitation buffer (containing 20 mM imidazole for Ni 2ϩ -NTAbound complexes). Proteins were eluted in Laemmli sample buffer (final concentration: 60 mM Tris-HCl (6.8), 5% glycerol, 1.67% SDS, 0.1 M dithiothreitol, 0.002% bromphenol blue) by boiling for 5 min. The samples were electrophoresed on 10 -12% SDS-polyacrylamide gels, which were transferred to nitrocellulose membranes and probed with 9E10 anti-Myc, M2 monoclonal anti-FLAG (Sigma), or Gal4 DNA-binding domain (Santa Cruz Biotechnology, Inc., Santa Cruz, CA) antisera.

The RDM Is Necessary and Sufficient for Transcriptional
Repression in Gal4 Fusion Proteins-We previously mapped the location of transcriptional activation and repression domains in the mouse C/EBP⑀ protein using Gal4 fusion proteins (21,22). As shown in Fig. 1A, Gal4 fusion proteins bearing the first 64 amino acids of C/EBP⑀ (referred to here as AD1) were potent activators in COS-1 cells of a Gal4-responsive luciferase reporter plasmid, whereas the inclusion of amino acids 64 -128 of C/EBP⑀ resulted in a significant decrease in transcriptional activity. Based on these data, we previously mapped the RD1 region between amino acids 64 and 128 of C/EBP⑀ and further demonstrated that mutations that disrupt the integrity of a five-amino acid motif named the RDM completely abrogated the inhibitory effect of the RD1 region (see Refs. 21 and 22 and Fig. 1A). Only one such mutation, a lysine to alanine substitution at position 121, is shown here (compare G4-⑀-(1-128) with G4-⑀-(1-128)(K121A)), but mutations in other residues of the RDM that are conserved in the consensus SUMO acceptor site (valine 120 and glutamate 122 of C/EBP⑀) also disrupted RD1 function (22). Although these loss-of-function data highlighted the importance of the RDM for RD1 function, they did not necessarily mean that the RDM was the only sequence that participated in repression, since other sequences between amino acids 64 and 128 could be important, but only in the presence of an intact RDM. Therefore, to test whether the RDM was sufficient for transcription repression in the context of Gal4 fusion proteins, we constructed expression vectors encoding Gal4 fusion proteins consisting of the C/EBP⑀ AD1 region joined to a short peptide containing either the wild type RDM sequence (VKEEP) or a variant sequence in which the lysine that serves as the attachment site for SUMO proteins was changed to arginine (VREEP). We previously showed that alanine and arginine substitutions at this position have equivalent effects, and arginine was used in this instance to conserve the positive charge at this position (22). The activity of the fusion protein containing the wild type RDM sequence was similar to that of the fusion protein containing the complete RD1 domain (compare G4-⑀-(1-128) with G4-⑀-(1-64)-RDM in Fig. 1). The fusion protein containing the VREEP sequence (G4-⑀-(1-64)-RDMm) was significantly more active than the protein containing the wild type RDM sequence, confirming the importance of the wild type sequence for the repression activity. The difference in activities between the two proteins bearing mutations in the RDM is due to the presence of additional sequences with transcriptional stimulatory activity between amino acids 65 and 97 (see Fig. 3A in Ref. 21).
To test the AD specificity of this inhibitory function, we constructed a similar set of expression plasmids containing the AD for the herpes simplex virus VP16 protein, since we previously showed that the RD1 region of C/EBP⑀ could repress the activity of this heterologous AD (21). As in Gal4-C/EBP⑀ fusion proteins, inclusion of either the wild type RD1 or RDM efficiently repressed the activity of the VP16 AD (compare G4-V-⑀RD1 with G4-V-RDM in Fig. 1B), and alteration of the lysine residues in either context relieved this repression (compare G4-V-⑀RD1(K121A) with G4-V-RDMm in Fig. 1B). We had previously proposed that the ability of the RD1 domain to repress a heterologous AD was inconsistent with an intramolecular inhibition model due to a lack of sequence conservation between the C/EBP⑀ and VP16 ADs. However, we noted that the C/EBP⑀ and VP16 ADs are both classified as acidic activation domains. Based on this observation, we decided to test whether that C/EBP⑀ RD1 region could also inhibit the activity of ADs from other classes. Therefore, we constructed expression vectors for Gal4 fusion proteins containing AD sequences from the Sp3 (amino acids 6 -426, classified as a glutaminerich AD) and CTF1 (amino acids 354 -499, classified as a proline-rich AD) transcription factors, joined to the wild type and RDM mutant RD1 domains from C/EBP⑀. Although the ADs from Sp3 and CTF-1 differed in their intrinsic activity compared with each other and the C/EBP⑀ and VP16 ADs, the activities of both ADs were efficiently inhibited in proteins containing the wild type RD1 (compare G4-Sp3 with G4-Sp3-⑀RD1 and G4-CTF1 with G4-CTF1-⑀RD1 in Fig. 1C). Similarly, the repressive effect of RD1 was relieved in both cases when the lysine within the RDM was changed to alanine (G4-Sp3-⑀RD1(K121A) and G4-CTF1-⑀RD1(K121A) in Fig. 1C). Collectively, these data suggest that the repressive activity of the RD1 domain is likely to involve intermolecular interactions between the RDM and a protein with co-repressor activity.
C/EBP⑀ Is Coactivated by the SUMO E3 Ligases, PIASx␣ and PIASx␤-To date, the only proteins known to interact with the RDM are members of the SUMO family (22). However, as mentioned above, it is not clear whether sumoylation of C/EBP⑀ is associated with repression or derepression of its activity. To address the specific consequences of sumoylation on C/EBP⑀ activity, we tested whether coexpression of two members of the PIAS family of SUMO E3 ligases, PIASx␣ and PIASx␤ ( Fig.  2A), affected the transcriptional activity of C/EBP⑀. These two PIAS family members were chosen because they are coexpressed with C/EBP⑀ during neutrophil differentiation 2 and were previously shown to function as coregulators of proteins, 2 H. Youn and S. C. Williams, unpublished results.
FIG. 1. Identification of sequences within the C/EBP⑀ RD1 domain that mediate transcriptional repression. A, the activity of a series of Gal4-C/EBP⑀ fusion proteins was tested in COS-1 cells using cotransfection assays and a Gal4-responsive luciferase reporter plasmid. The domain structures of the fusion proteins are shown on the left, and the relative locations of the AD and RD1 domains are shown. The wild type and mutated forms of the RDM are represented by the small shaded boxes within RD1. In each of the panels, the activity of proteins bearing the full functionally defined RD1 (amino acids 64 -128) is set to 1.0, and the activity of all other proteins is expressed relative to it. Each transfection was performed at least three times in triplicate, and averaged data from 2-3 experiments are presented in each figure. The data are expressed as mean Ϯ S.D. B, a similar set of proteins that contained the VP16 AD in place of the C/EBP⑀ AD were tested, and the data are presented as described above. C, proteins containing either the glutamine-rich AD from Sp3 or the proline-rich AD from CTF1 were tested as described above.
such as the androgen receptor, that contain regulatory domains similar to the RD1 domain of C/EBP⑀ (33). Expression vectors encoding either PIASx isoform were transfected into COS-1 cells along with a C/EBP⑀ expression vector and a C/EBP-dependent luciferase reporter plasmid. The full-length C/EBP⑀ protein was quite inactive in this assay and did not transactivate the reporter plasmid above the background level (data not shown). Coexpression of wild type PIASx␣ and PIASx␤ proteins with C/EBP⑀ led to significant enhancement of activity, with maximal coactivation by PIASx␤ being greater than PIASx␣ (Fig. 2B). The RING finger motif of PIAS proteins is required for their SUMO E3 ligation functions, and thus we examined the consequences of deletion of the RING finger motifs of both PIASx␣ and PIASx␤ on C/EBP⑀ coactivation. Mutant forms of each PIASx isoform lacking their RING finger motifs (PIASx␣⌬RF and PIASx␤⌬RF) were less potent coactivators, although again PIASx␤ retained a greater activity than PIASx␣ (Fig. 2B). PIASx␣ and PIASx␤ only differ at the carboxyl terminus, where PIASx␤ possesses a serine/threoninerich region and a five-amino acid motif (IISLD) that are not present in PIASx␣ (Fig. 2A). Thus, these sequences may be responsible for the enhanced activity of PIASx␤ compared with PIASx␣ in these assays.
We next determined whether specific regions of C/EBP⑀ were required for coactivation by PIASx␣ and PIASx␤. Differential usage of two in-frame translation initiation codons in the mouse C/EBP⑀ mRNA directed the synthesis of two isoforms (named C/EBP⑀ p32 and C/EBP⑀ p29) of 281 and 252 amino acids, respectively (Fig. 3A). The shorter isoform lacks a portion of the tripartite N-terminal activation domain and is a slightly weaker activator than the p32 isoform. The two C/EBP⑀ isoforms responded similarly to coexpression of each PIASx isoform, with PIASx␤ functioning as a more potent coactivator in each case (Fig. 3B).
Cotransfection experiments were performed in COS-1 cells to determine whether either or both regulatory domains within C/EBP⑀ (Fig. 3A) were required for PIASx-dependent coactivation. As reported earlier (21), deletion of RD1 or RD2 resulted in ϳ7and 1.5-fold increases in the activity of C/EBP⑀ on the C/EBP-dependent reporter plasmid (Fig. 3C). Deletion of RD2 (⌬RD2) did not dramatically affect coactivation by either PIASx␣ or PIASx␤. However, the C/EBP⑀ protein lacking the RD1 domain (⌬RD1) was essentially unresponsive to PIASx␣  (44); RF (RING finger), associated with SUMO E3 ligase activity and protein-protein interactions; NLS, putative nuclear localization signals; D/E-rich and S/T-rich, domains rich either in Asp/Glu or Ser/Thr; IISLD, a fiveamino acid motif conserved in PIASx␤, PIAS1, and PIAS3. The RING finger deletion mutant forms of the two proteins (referred to throughout as PIASx␣⌬RF and PIASx␤⌬RF) are also schematically depicted. B, C/EBP⑀ was coexpressed with the indicated PIASx proteins in COS-1 cells, and activity was measured on a C/EBP⑀-dependent reporter plasmid. Two different amounts of each PIASx expression plasmid (80 and 160 ng) were used, and data are depicted as luciferase units. This experiment was performed in triplicate and repeated at least six times, and data are represented as mean Ϯ S.D.

FIG. 3. PIASx-dependent coactivation requires RD1 of C/EBP⑀.
A, domain structures of two isoforms of C/EBP⑀ produced by alternative translation initiation codon usage. The relative location of the ADs, RDs, and DNA-binding domain are shown. B, the p32 and p29 isoforms of C/EBP⑀ were coexpressed with either PIASx␣ or PIASx␤, and luciferase activities were measured on a C/EBP⑀-dependent reporter plasmid. The data are presented as raw luciferase values, and the -fold activation observed upon coexpression of either PIASx isoform is shown above each bar. C, coactivation of the p32 isoform of C/EBP⑀ by PIASx␣ and PIASx␤ was compared with two deletion mutants of C/EBP⑀ that lacked either the RD1 or RD2 domains. and only weakly responsive to PIASx␤ (Fig. 3C). Thus, maximal PIASx-dependent coactivation was dependent on the integrity of the RD1 within the C/EBP⑀ polypeptide.
PIASx Proteins Interact with C/EBP⑀ and Enhance Its Sumoylation-PIASx␣ and PIASx␤ physically interact with several transcription factors, including members of the steroid hormone receptor superfamily (33). We therefore next tested whether C/EBP⑀ and PIASx proteins physically interact by performing co-immunoprecipitation and Ni 2ϩ -NTA pull-down assays. Nuclear extracts were prepared from COS-1 cells transfected with combinations of expression vectors encoding C/EBP⑀-Myc-His 6 , FLAG-PIASx␣, FLAG-JunD, or G4-V. JunD is a member of the AP-1 family, and Jun and C/EBP proteins are known to physically interact (34). G4-V was used as a negative control for these experiments. PIASx␣ and JunD efficiently coprecipitated with C/EBP⑀ (Fig. 4A, compare lanes 2  and 3 with lanes 4 and 5). Likewise, C/EBP⑀ efficiently coprecipitated with both PIASx␣ and JunD (Fig. 4A, lanes 7 and 8), and interactions were dependent on the presence of both protein partners in all cases (see lanes 1 and 6 and lower panels showing input levels of each protein). These interactions were specific, since G4-V did not coprecipitate with either C/EBP⑀ or PIASx␣ (Fig. 4B). C/EBP⑀ also coprecipitated with PIASx␤ (data not shown).
The RING finger domain of PIAS proteins is required for their SUMO E3 ligase function and can also function as a proteinprotein interaction interface (35). Since PIASx␣ coactivation of C/EBP⑀ was entirely dependent on the integrity of the RING finger domain, and coactivation by PIASx␤ was partially dependent on this domain, we next tested whether PIASx proteins could enhance C/EBP⑀ sumoylation in COS-1 cells. PIASx␣ was consistently expressed at significantly higher levels than PIASx␤ in the large scale transfections necessary for performing these assays; therefore, we have only tested the activity of PIASx␣ in these experiments. Myc-tagged C/EBP⑀ was expressed in COS-1 cells in the presence or absence of SUMO-1 and PIASx␣, and unmodified and modified forms of C/EBP⑀ were detected by immunoblotting with either Myc-or C/EBP⑀-specific antisera ( Fig.  4B; data not shown). Unmodified C/EBP⑀ migrated at its expected molecular weight (ϳ34,000), and the SUMO-C/EBP⑀ complex was only detected in the presence of excess SUMO-1 (Fig.  4B, lanes 1 and 2). The identity of this band as SUMO-C/EBP⑀ was confirmed by expressing a mutant form of C/EBP⑀ in which the lysine acceptor site was changed to alanine. In this case, the SUMO-C/EBP⑀ band was absent (Fig. 4B, lane 5). The intensity of the SUMO-C/EBP⑀ band increased (from 1.7 to 9.6% of total C/EBP⑀ protein) in the presence of PIASx␣; however, the increase in the intensity of the SUMO-C/EBP⑀ band was dependent on the integrity of the RING finger domain of PIASx␣. Therefore, the coactivation of C/EBP⑀ by PIASx␣ (and presumably PIASx␤) correlates with enhanced sumoylation within the RDM.
PIASx␣ and PIASx␤-dependent Coactivation of Gal4 Fusion Proteins-As shown above, the RD1 domain of C/EBP⑀ functions as an autonomous inhibitory domain in Gal4 fusion proteins. As deletion of RD1 abolished coactivation by PIASx proteins in the context of the full-length C/EBP⑀ polypeptide, we next examined whether Gal4 fusion proteins containing RD1 were also targets for PIASx coactivation. Two different RD1containing Gal4 fusion proteins introduced earlier (G4-⑀-(1-128) and G4-V-⑀RD1) were utilized in these experiments (Fig.  5A). These proteins differ in the origin of AD sequences, and we  6 -8) and subsequently analyzed by SDS-polyacrylamide gel electrophoresis to identify co-precipitated proteins. These proteins were detected with antisera directed against either the FLAG (lanes [1][2][3][4][5] or Myc (lanes 6 -8) epitopes on the co-expressed proteins. Input levels of either C/EBP⑀-Myc or FLAG-tagged PIASx␣ or JunD were analyzed by immunoblotting in the lower panels. WB, Western blot. B, similar assays were performed using Gal4-VP16 (G4-V) as a negative control to demonstrate that interactions observed in A represented specific proteinprotein interactions. The input (I) level of G4-V is shown in lanes 1 and 2 and the lack of interactions between C/EBP⑀ and G4-V and PIASx␣ and G4-V is shown in lanes 3 and 5, respectively. C, COS-1 cells were cotransfected with the indicated combinations of expression plasmids for C/EBP⑀ (wild type or K121A mutant), SUMO-1, and PIASx␣ (wild type or ⌬RF mutant). Nuclear extracts were prepared and analyzed by Western blotting with an anti-Myc antiserum to detect Myc-tagged C/EBP⑀ proteins. The positions of the unmodified C/EBP⑀ protein and the SUMO⅐C/EBP⑀ complex are shown with arrowheads. The relative intensities of each band were measured by densitometry, and the proportion of C/EBP⑀ represented as SUMO⅐C/EBP⑀ complexes was calculated.
reasoned that these two proteins would also allow us to test whether C/EBP⑀ sequences outside the RD1 domain influenced PIASx coactivation.
As shown in Fig. 5, B and C, Gal4 fusion proteins containing either the C/EBP⑀ or VP16 ADs alone (G4-⑀-(1-64) and G4-V) were 25-35-fold more active than their counterparts that included the RD1 region, consistent with data shown in Fig. 1. Both PIASx␣ and PIASx␤ were able to coactivate G4-⑀- (1-128), and, similar to the full-length C/EBP⑀ protein, PIASx␤ was a more potent coactivator (Fig. 5B). PIASx␣⌬RF did not significantly increase G4-⑀-(1-128) activity; however, PIASx␤⌬RF retained ϳ40% of the coactivator function of the wild type protein (Fig. 5B). In contrast, PIASx␣ was a more potent coactivator of the Gal4-V-⑀RD1 protein than PIASx␤, and deletion of the RING finger motif in both proteins completely abolished their coactivator functions (Fig. 5C). Thus, coactivation of C/EBP⑀ by PIASx␣ and PIASx␤ appears to involve both common and unique properties. The common coactivation properties required the RING finger domain of both isoforms and will be referred to as RF-dependent coactivation. The unique coactivation function of PIASx␤ did not require the RF domain but required the AD of C/EBP⑀ and will be referred to as RFindependent coactivation.
The RDM Is Necessary and Sufficient for PIASx-dependent Coactivation of Gal4-C/EBP⑀ Proteins-We next tested whether the RDM was required for PIASx␣ and PIASx␤ coactivation of Gal4 fusion proteins containing the RD1 of C/EBP⑀. The effect of coexpression of either PIASx␣ or PIASx␤ was examined on G4-⑀-(1-128) or G4-V-⑀RD1 proteins in which the RDM sequence had been mutated (Fig. 6, A and B). Neither PIASx␣ or PIASx␤ enhanced the activity of the proteins carrying the mutated RDM sequence, in marked contrast to their activity on proteins bearing the wild type RDM (see Fig. 5). These results therefore indicated that the RDM sequence was necessary for PIASx-dependent coactivation. To test further whether the RDM was sufficient for coactivation, similar coexpression studies were performed comparing Gal4 fusion proteins containing either the full RD1 domain or the wild type or mutant RDM sequences described earlier. As shown in Fig. 6, C and D, PIASx-dependent coactivation properties were similar on both the G4-⑀-(1-128) and G4-⑀-(1-64)-RDM and on the G4-V-⑀RD1 and G4-V-RDM proteins. Most notably, the relative potencies of each PIASx isoform were replicated on the constructs that contained just the RDM, with PIASx␤ being a more potent coactivator of proteins containing the C/EBP⑀ AD (G4-⑀-(1-128) and G4-⑀-(1-64)-RDM) and PIASx␣ being a more potent coactivator of proteins containing the VP16 AD (G4-V-⑀RD1 and G4-V-RDM). Consistent with the data shown in Fig. 6, A and B, both proteins carrying just the mutated RDM sequence were less sensitive to coactivation by either PIASx isoform (G4-⑀-(1-64)-RDMm and G4-V-RDMm in Fig. 6, C and D). The slight effects observed on both RDMm proteins were similar to those seen on proteins containing just the ADs (data not shown). Thus, the intact RDM was sufficient for PIASx-dependent coactivation.
PIASx Proteins Are Also Differential Coactivators of C/EBP␣-We previously demonstrated that each of the four members of the C/EBP family contain inhibitory domains located adjacent to their major ADs that were dependent for function on the integrity of RDM-like sequences (22). To extend this intrafamily comparison, we tested whether PIASx proteins could function as coactivators of a Gal4 fusion protein carrying the N-terminal segment of C/EBP␣. As shown in Fig. 7, a Gal4 fusion protein containing just the N-terminal AD of C/EBP␣ (G4-␣-(1-108)) was ϳ30-fold more active than G4-␣-(1-170), which contains an RD1-like domain and a RDM-like element (22). However, G4-␣-(1-170) was a target for differential coactivation by both PIASx␣ and PIASx␤, and the pattern of coactivation was similar to that observed for Gal4 fusion proteins containing the equivalent region of C/EBP⑀ (amino acids 1-128). Specifically, G4-␣-(1-170) was weakly coactivated by PIASx␣, and this activity was dependent on the RF of PIASx␣, whereas it was more strongly coactivated by PIASx␤, and some activity (about 25% of maximal) was retained by the RF-deletion mutant of PIASx␤ (Fig. 7). Thus differential coactivation by PIASx isoforms appears to be shared within the C/EBP family.

FIG. 5. PIASx-dependent coactivation of Gal4 fusion proteins.
A, the domain structure of Gal4 fusion proteins used in these experiments. The constructs differ in the identity of the AD, being derived either from C/EBP⑀ or VP16. B, COS-1 cells were cotransfected with the indicated Gal4 expression plasmid, a Gal4-responsive luciferase reporter plasmid, and two amounts of the indicated PIASx expression vectors (80 and 160 ng). Luciferase values were measured and are expressed as mean Ϯ S.D., with the activity of G4-⑀-(1-128) being set to 1.0. C, similar to B except that the activity of the G4-V-⑀RD1 was set to 1.0.

DISCUSSION
In this study, we report that two members of the PIAS family of RING finger domain proteins function as differential coactivators of the neutrophil-specific transcription factor C/EBP⑀. This study was prompted by our recent demonstration that C/EBP⑀ is one of a growing number of nuclear proteins that are targets for modification with the ubiquitin-related protein SUMO-1 and its relatives (22). SUMO modification of C/EBP⑀ occurs on a consensus SUMO attachment site (the RDM) within a domain (RD1) that inhibits the activity of C/EBP⑀, and we show here that the RDM can functionally replace RD1 in Gal4 fusion proteins. Remarkably, PIASx-dependent coactivation also required RD1, and the RDM can also functionally substitute for RD1 as a target for PIASx coactivation. The ability of a short motif to participate in both inhibitory and stimulatory regulation of transcription suggests that this motif is a site for complex regulatory interactions.
RDM-dependent Inhibition of C/EBP⑀-We initially addressed two potential models that explain the inhibitory function of the RD1 domain, the intra-and intermolecular inhibition models. The intramolecular inhibition model, originally proposed based on studies of the domain structure and regulation of C/EBP␤, suggests that physical interactions between RD1 and the N-terminal AD prevent access to the transcrip-tional machinery. However, we show here that the RD1 is capable of inhibiting the activity of examples of three different classes of AD, the acidic ADs of C/EBP⑀ and VP16, the prolinerich AD of CTF-1, and the glutamine-rich AD of Sp3. Since these ADs are unrelated at the sequence level, we propose that it is unlikely that direct physical interactions would occur between the RD1 domain and each AD. Instead, we propose that an intermolecular inhibition model, involving proteins that interact with specific sequences within RD1, is more likely. The best candidate for a relevant RD1-interacting protein is a member of the SUMO family, and this candidacy was strengthened by the observation that attachment of an intact RDM, but not a nonsumoylatable form of the RDM, was sufficient to replicate the inhibitory function of the entire RD1 domain. These data are consistent with observations in several other transcriptional regulatory proteins, where SUMO attachment sites have been shown to be critical functional components of transcriptional inhibitory domains. A model has recently been proposed that explains how transient SUMO attachment may mediate transcriptional repression and that takes into account the generally low steady-state level of the sumoylated form of most SUMO targets (28). In this model, initial transient SUMO attachment promotes association of the SUMO target with corepressor proteins or complexes. Although FIG. 6. The RDM is necessary and sufficient for PIASx-dependent coactivation. A and B, the effect of mutating the RDM sequence on PIASx-dependent coactivation was tested in cotransfection assays in COS-1 cells. Gal4 fusion proteins containing either the C/EBP⑀ or VP16 ADs fused to wild type or RDM mutant forms of RD1 were coexpressed with wild type PIASx␣ or PIASx␤, and luciferase values were measured. C and D, similar experiments were performed in COS-1 cells to compare coactivation of Gal4 fusion proteins containing either the full RD1 or just the RDM. A mutant form of the RDM in which the lysine that is the target for SUMO attachment was changed to arginine was also examined.
we do not yet know the identity of these putative corepressors, this model is consistent with the inhibitory function of the RDM in C/EBP⑀.
PIASx-dependent Coactivation: RF-dependent and -independent Functions-There are four PIAS genes in mammals that encode several different proteins due to the use of alternative splicing patterns (35). The PIAS proteins can function as transcriptional coregulators in at least three ways. First, they can inhibit DNA binding by the target protein, a function ascribed to PIAS1 in inhibiting STAT1 activity (36). Second, they can function as co-regulators by modulating SUMO attachment to a target protein, and most examples of PIAS-dependent coactivation have detected enhancement of SUMO attachment (see, for example, Refs. [37][38][39]. Third, PIAS proteins can function as SUMO-independent transcriptional coregulators (40 -42), and this latter function may involve recruitment of accessory proteins or localization of target proteins to subnuclear domains. Previous studies indicated that PIASy functions as a corepressor of C/EBP␣ (29), but the present report provides the first evidence that PIAS proteins can function as coactivators of C/EBP family members.
PIASx␣ and PIASx␤ are the products of alternatively spliced mRNAs from the same gene (35). These two mRNAs only differ at the 3Ј-end, where a single exon utilized in PIASx␣ is replaced by two alternative exons in PIASx␤. Consequently, the two proteins only differ at their C termini and are identical over the first 550 amino acids (33,43). However, despite their overall similarity, these two proteins exhibited both shared and unique C/EBP⑀ coactivation properties. The shared function was dependent on the integrity of the RING finger domain of both PIASx␣ and PIASx␤, a domain that is required for the E3 ligase function of PIAS proteins but also participates in protein-protein interactions (35). Since PIASx␣ can enhance sumoylation of C/EBP⑀, we postulate that this modification may play a role in increasing C/EBP⑀ activity. However, we will refer to this mode of coactivation as RF-dependent coactivation (Fig. 8) until a direct connection between SUMO attachment and C/EBP⑀ coactivation can be established. The second mode of coactivation was RF-independent and was only observed with PIASx␤ (Fig. 8). Maximal coactivation by PIASx␤ required RD1 and the N-terminal AD of C/EBP⑀ (or C/EBP␣), and deletion of RD1 or replacement of the AD with the VP16 AD significantly decreased PIASx␤-dependent coactivation. Since coactivation by PIASx␣ was relatively insensitive to the identity of the AD in the target protein, we conclude that specific functional interactions between PIASx␤ and the C/EBP AD are required for maximal activity. We hypothesize that these interactions involve the C-terminal domain of PIASx␤ that is distinct from PIASx␣. Interestingly, when tested as Gal4 fusions, PIASx␤ possessed an inherent transcriptional activating function that was lacking in PIASx␣ (33), and this function may contribute to the enhanced activity of PIASx␤ as a C/EBP⑀ coactivator. The C-terminal domain of PIASx␤ contains two notable features, a Ser/Thr-rich domain and a five-amino acid motif (IISLD) that is shared with PIAS1 and PIAS3. The contribution of these features to PIASx␤ coactivation is currently under investigation. Based on the fact that PIASx␤ was also a more potent coactivator of other C/EBP family members, including C/EBP␣, we propose that PIASx␤ is a specific coactivator for C/EBPs, whereas PIASx␣ is a more general coactivator of proteins with RD-1-like inhibitory domains. These differential activities may be important in cell types that express both isoforms of PIASx.
Regulation of C/EBP⑀ Activity: The Role of Sumoylation-Our data indicate that two opposing regulatory mechanisms affecting the activity of C/EBP⑀, repression through the RD1 domain and coactivation by PIASx proteins, appear to act through the same sequence in C/EBP⑀, the RDM. As mentioned above, the only proteins known to directly interact with the RDM are members of the SUMO family, and both repression and coactivation require sequences within the RDM that are part of the consensus SUMO attachment site. These observations raise the intriguing question of how the same modification could be associated with two functionally opposite effects. There are several possible explanations that will need to be addressed by further experimentation. First, repression and coactivation could be associated with attachment of different members of the SUMO family. There are now four members of this family identified in mammals, and the different proteins are likely to have different functions (25)(26)(27). We have deter-  8. A model for repression and coactivation of C/EBP⑀. The intrinsic repression function of RD1 is dependent on the integrity of the RDM and is likely to involve a transient sumoylation step accompanied by recruitment of as yet unidentified corepressor proteins. RING fingerdependent coactivation by PIASx␣ and PIASx␤ is also dependent on the integrity of the RDM and may involve the formation of multiprotein complex that includes SUMO, a PIAS protein, and possibly other unidentified proteins. RING finger independent coactivation was only observed with PIASx␤ and requires the AD from C/EBP⑀ (or C/EBP␣; see Fig. 7). The domains of C/EBP⑀ were defined in Fig. 3. mined that C/EBP⑀ can be modified by three members of the family, SUMO-1 to -3; however, we have not detected any differences in the effects of attaching different family members. 3 Second, the difference in SUMO-dependent effects could be due to the recruitment of accessory proteins with different coregulatory properties that lead to repression or coactivation. In light of our current data, we would propose that in the absence of high levels of PIASx proteins, SUMO attachment leads to recruitment of corepressors (Fig. 8), consistent with the model proposed by Girdwood and Hay (28). Alternatively, in the presence of high levels of PIASx proteins, enhanced sumoylation would lead to the recruitment of coactivators, leading to elevated C/EBP⑀-dependent activity (Fig. 8). An extension of this model would suggest that enhanced sumoylation by other PIAS family members would be dependent on the specific protein-protein interactions directed by the PIAS protein. This concept is supported by the observation that coexpression of PIASy actually decreases C/EBP⑀ activity. 4 Clearly, the identification of coregulators recruited to C/EBP⑀ in a SUMO/PIAS-dependent manner should elucidate the mechanisms underlying the complex regulation of this activity of this protein.