Versatile protein tag, SUMO: Its enzymology and biological function

Authors


Abstract

Small ubiquitin-related modifier (SUMO) is a member of a ubiquitin-like protein family that regulates cellular function of a variety of target proteins. SUMO and ubiquitin are synthesized as precursors that need to be processed prior to conjugation to target proteins, and their mature forms have a similar tertiary structure. The mechanism for SUMO conjugation is also analogous to that of the ubiquitin system, such as the utilization of E1, E2, and E3 cascade enzymes. However, the biological consequence of SUMO modification is quite different from that of the ubiquitin system. Whereas ubiquitination of most proteins is for the degradative pathway, SUMO modification of target proteins is involved in nuclear protein targeting, formation of subnuclear structures, regulation of transcriptional activities or DNA binding abilities of transcription factors, and control of protein stability. This review will summarize the recent progress made in the enzymology of SUMO and its biological significance. © 2002 Wiley-Liss, Inc.

Posttranslational modification of proteins by ubiquitin is a key regulatory cellular mechanism that is involved in a variety of cellular processes, including regulation of intracellular protein breakdown, cell cycle regulation, signal transduction, transcription, and antigen presentation (Hochstrasser, 1996; Ciechanover, 1998; Hershko and Ciechanover, 1998). Ubiquitin is a 76-amino acid polypeptide that is highly conserved from yeast to humans. It is covalently ligated to a wide variety of target proteins through the action of ubiquitin-activating enzyme (E1), ubiquitin-conjugating enzyme (E2), and ubiquitin protein ligase (E3). Free ubiquitin is activated in an ATP-dependent manner with the formation of thioester linkage between E1 and the carboxyl terminus of ubiquitin. Ubiquitin is then transferred to one of various E2s. For E3s containing the HECT domain, ubiquitin is transferred from E2 to the active-site cysteine in the HECT domain, followed by transfer to protein substrates or to multi-ubiquitin chains that are ligated to the substrates (Kornitzer and Ciechanover, 2000; Weissman, 2001). For E3s containing the RING-finger domain, which form a complex with E2 and protein substrate, E2-bound ubiquitin might be directly transferred to the substrate. Notably, poly-ubiquitin chain formation sometimes requires another factor termed E4 (Koegl et al., 1999). The proteins ligated to multiple units of ubiquitin are then degraded by the ATP-dependent, 26S proteasome (Coux et al., 1996). Although the role as a tag for protein degradation by the 26S proteasome has been known as a major function of ubiquitin, other functions of ubiquitination have been discovered (Galan and Haguenauer-Tsapis, 1997; Terrell et al., 1998; Hofmann and Pickart, 1999; Deng et al., 2000; Spence et al., 2000; Hicke, 2001). For example, mono-ubiquitination, unlike poly-ubiquitination, is not involved in the protein degradation pathway, but plays a role in at least three distinct cellular processes, such as histone regulation, endocytosis, and budding of retroviruses from the plasma membrane.

Recently, a number of other small proteins, so-called ubiquitin-like molecules (Ubls), have been identified (Saitoh et al., 1997). These proteins are structurally related to ubiquitin, and can be conjugated to various target proteins in a similar manner with ubiquitin (Tanaka et al., 1998; Hochstrasser, 2000; Melchior, 2000; Yeh et al., 2000; Müller et al., 2001). However, covalent attachment of Ubls does not result in degradation of the modified proteins, but functions in a similar way to mono-ubiquitination. To date, several Ubls, such as small ubiquitin-related modifier (SUMO)-1/Smt3, NEDD8/Rub1, UCRP (also called ISG15), and Fub have been identified. Of these, the best characterized Ubl is the mammalian SUMO-1 (also called UBL1, Sentrin, PIC1, GMP1, or SMT3c) (Boddy et al., 1996; Mannen et al., 1996; Matunis et al., 1996; Okura et al., 1996; Shen et al., 1996; Mahajan et al., 1997), which can be conjugated to a variety of cellular proteins, such as promyelocytic leukemia protein (PML), Ran-GTPase-activating protein (RanGAP1), and inhibitor of NF-κB (IκBα) (Mahajan et al., 1998; Müller et al., 1998; Desterro et al., 1999).

Whereas Saccharomyces cerevisiae contains only one SUMO homologue, Smt3p, which was originally discovered as a suppressor of mutants in the centromeric protein MIF2 (Meluh and Koshland, 1995), the mammalian SUMO family member consists of SUMO-1, -2, and -3 (Kamitani et al., 1998; Saitoh and Hinchey, 2000). In humans, SUMO-1 is a 101 amino acid polypeptide that shares 18% identity with ubiquitin. SUMO-1 shares 48% identity with SUMO-2 and 46% identity with SUMO-3. SUMO-2 and -3 share 95% identity, and can be grouped into a subfamily distinct from SUMO-1. All three members have distinct N-terminal amino acid sequences and C-terminal extensions. Although SUMO-1 shares only 18% amino acid sequence identity with ubiquitin, the three-dimensional structure of SUMO-1 determined by nuclear magnetic resonance resembles the overall structure of ubiquitin, characterized by a tightly packed ββαββαβ fold (Bayer et al., 1998). Similar to ubiquitin, SUMO-1 is synthesized as a precursor with the C-terminal extension of several amino acids, which needs to be processed to expose the C-terminal Gly residue that is essential for conjugation to target proteins. One distinct feature of SUMO-1 is that the protein has a long and highly flexible N-terminal extension that is absent in ubiquitin, although the functional importance of this sequence is unknown. Interestingly, SUMO-1 does not have the lysine residue corresponding to Lys-48 in the ubiquitin molecule that is required for the formation of poly-ubiquitin chains, explaining why SUMO-1 does not make multi-chain forms (Bayer et al., 1998).

Like SUMO-1, other members of this family can also be conjugated to target proteins (Kamitani et al., 1998). Recent studies of Saitoh and Hinchey (2000) have shown the functional heterogeneity of SUMO family members. Using an antibody that interacts with SUMO-2 and -3, but not with SUMO-1, they demonstrated that SUMO-2 and -3 are conjugated poorly to RanGAP1, a major SUMO-1 target protein, and the conjugation of SUMO-2 and -3 can be induced by protein-damaging stimuli, such as acute temperature shift. Another difference between SUMO-1 and -2/-3 is the ability of SUMO-2 and -3 in poly-SUMO chain formation (Tatham et al., 2001). Both SUMO-2 and -3, but not SUMO-1, contain a consensus motif, ψKXE (ψ for a hydrophobic amino acid and X for any amino acid), in their N-terminal region for sequential SUMO conjugation. A mutation of this lysine residue abolishes the poly-SUMO chain formation. A yeast SUMO homolog, Smt3p, also contains a similar sequence motif, and can form poly-SUMO chains (Johnson and Gupta, 2001). Although a protein that is modified by poly-SUMO chains in vivo has been identified (Tatham et al., 2001), the functional significance of this poly-SUMO chain formation is unknown.

ENZYMES INVOLVED IN SUMO CONJUGATION AND DECONJUGATION

Since SUMO-1 has a similar structure with ubiquitin, the mechanism for conjugation and deconjugation of SUMO-1 has been expected to be similar to that of ubiquitin. In the ubiquitin system, the ubiquitin molecule is covalently attached to target proteins by the serial reaction of E1, E2, and E3 enzymes. Unlike the E1 enzyme (Uba1) for ubiquitination, which is comprised of a single polypeptide, the SUMO-activating enzyme is a heterodimer consisting of Aos1 (also called SAE2) and Uba2 (SAE1) subunits (Johnson et al., 1997; Desterro et al., 1998). Like the ubiquitin system, SUMO can be covalently attached to target proteins by Ubc9, a functional homologue of the ubiquitin E2 enzymes (Johnson and Blobel, 1997; Lee et al., 1998; Saitoh et al., 1998; Schwarz et al., 1998). In most cases, the E3 ubiquitin ligases, which are required for the formation of an isopeptide bond between ubiquitin and target proteins, are considered to be responsible for the substrate specificity. In the SUMO conjugation pathway, however, the presence of E3-like enzymes for sumoylation has been obscure until recently, because the Aos1/Uba2 complex in combination with Ubc9 is sufficient for SUMO conjugation in vitro, although this activity is considerably lower than that of HeLa cells or Xenopus egg extract (Azuma et al., 2001). Recently, three groups have demonstrated that PIAS1 (also called GBP), Siz1 (Ull1), and Siz2 (Nfi1) show the E3-like activities for sumoylation of the mammalian p53 and the yeast septins, which form a 10-nm filamentous ring that encircles the yeast bud neck (Hochstrasser, 2001; Johnson and Gupta, 2001; Kahyo et al., 2001; Takahashi et al., 2001a,b).

Like deubiquitination, SUMO-1 can be proteolytically processed from its conjugated target proteins. Several desumoylating enzymes have been identified, and these are SUSP1, SENP1, SMT3IP1, and SMT3IP2 from mammals and Ulp1 and Ulp2 from S. cerevisiae (Li and Hochstrasser, 1999; Gong et al., 2000; Kim et al., 2000; Li and Hochstrasser, 2000; Nishida et al., 2000, 2001; Schwienhorst et al., 2000). In S. cerevisiae, most of the known genes required for SUMO conjugation and deconjugation, namely SMT3, UBA2, AOS1, UBC9, and ULP1, are essential for cell viability (Dohmen et al., 1995; Seufert et al., 1995; Johnson et al., 1997; Li and Hochstrasser, 1999). Mutations in these genes lead to defects in cell cycle progression, with cells accumulating around G2/M. In Drosophila melanogaster, the loss-of-function mutation of semushi, a Drosophila ortholog of Ubc9, prevents nuclear import of transcription factor Bicoid (Bcd), and results in defects in embryogenesis (Epps and Tanda, 1998). Thus, sumoylation and desumoylation cycles must be functionally operated for the regulation of cellular processes (Fig. 1).

Figure 1.

Sumoylation and desumoylation cycle. E1-, E2-, and E3-like enzymes are involved in sumoylation of target proteins, and SUSPs catalyze desumoylation and precursor processing.

SUMO-activating enzyme, Aos1/Uba2

SUMO-activating enzyme is a heterodimeric complex consisting of Aos1 and Uba2 (Johnson et al., 1997; Okuma et al., 1999). Both subunits are well conserved from yeast to humans. In humans, Aos1 is 28% identical and 56% similar to the N-terminal half of the E1 enzyme for ubiquitin. Uba2, which has similarity to the C-terminal half of the E1 enzyme for ubiquitination, contains the active site cysteine residue required for the formation of thioester bonds. However, Uba2 alone is not sufficient for the thioester bond formation. Aos1 assembles with Uba2 to form an active E1 enzyme for sumoylation in vitro.

Although Aos1 and Uba2 form a single active enzyme, its expression pattern in human cell lines is quite different (Azuma et al., 2001). The concentration of the Uba2 protein does not show any substantial change during the cell cycle in HeLa cells. In contrast, the level of the Aos1 protein increases as cells progress through S phase, followed by a substantial decrease in G2 phase. Since the abundance of some sumoylated protein species similarly peaks in S phase, the changes in the level of Aos1 may influence the conjugation of these substrates. Both Aos1 and Uba2 localize to the nucleus, but are excluded from nucleoli. This type of subcellular distribution is consistent with the idea that most protein sumoylation occurs within the nucleus.

SUMO-conjugating enzyme, Ubc9

SUMO appears to have a single E2-type conjugating enzyme, Ubc9, which is specific for SUMO. Ubc9 is highly conserved from yeast to humans. The yeast Ubc9 shares 33 and 36% amino acid identity with the yeast ubiquitin-conjugating enzymes, Ubc4 and Ubc7, respectively. The function of Ubc9 in the SUMO-conjugating pathway was first identified by Johnson and Blobel (1997). They purified this enzyme by affinity chromatography using Smt3p, a yeast homolog of SUMO-1, as a ligand. Schwarz et al. (1998) also identified Ubc9 as a specific player in the SUMO-conjugating pathway by screening the yeast E2 mutant strains, which could regain their ability in conjugate formation with exogenously introduced, HA-tagged Smt3p. The Ubc9 proteins in human and Xenopus were also characterized as a SUMO-conjugating enzyme in the pathway of RanGAP1 modification (Lee et al., 1998; Saitoh et al., 1998). Ubc9 forms thioester bonds with SUMO-1, but not with ubiquitin, indicating that Ubc9 is the specific enzyme for SUMO conjugation (Desterro et al., 1997; Johnson and Blobel, 1997; Schwarz et al., 1998).

Ubc9 was originally identified as a ubiquitin-conjugating enzyme involved in cell cycle regulation (Seufert et al., 1995). In S. cerevisiae, repression of Ubc9 synthesis prevents cell cycle progression at G2 or early M phase, causing the accumulation of large budded cells with a single nucleus, short spindle, and replicated DNA (Seufert et al., 1995). In the absence of Ubc9, yeast cells are unable to degrade B-type cyclins, the S-phase cyclin Clb5, and the M-phase cyclin Clb2 (Seufert et al., 1995). This phenotype is strikingly similar to that of the mutants in the genes for Srp1, the yeast α-importin (Loeb et al., 1995), and Cse1, which was identified in a screening for cyclin-stabilizing mutants (Irniger et al., 1995) and bears a putative Ran-binding motif (Görlich et al., 1997). α-Importin acts as a nuclear localization signal receptor, and targets karyophilic proteins to the docking sites on the nuclear pore before their translocation into the nucleus (Görlich and Mattaj, 1996). Thus, the impaired function of nuclear transport machinery appears to be responsible for the defect in cell cycle progression and cyclin degradation in ubc9 null mutants.

In spite of the overall similarity between Ubc9 and the E2 enzymes for ubiquitination, they have very different surface charge distribution (Giraud et al., 1998; Liu et al., 1999). The surface of Ubc9, which is involved in SUMO binding, is mainly positively charged, whereas the corresponding regions in the ubiquitin E2 enzymes, like Ubc4 and Ubc7, have negative or neutral charge. Likewise, the surface electrostatic potentials of SUMO-1 and ubiquitin molecules are very different (Bayer et al., 1998), despite the fact that their tertiary structures are highly conserved. The surface of SUMO-1 that binds to the positively charged Ubc9 has an overall negative charge. On the other hand, ubiquitin cannot bind to Ubc9, because it also has positive charges in the corresponding region. Thus, the difference in the surface charge of E2s might play a role in the modifier selectivity.

SUMO protein ligases

In the ubiquitin system, E3 ligases are considered to be responsible for the substrate specificity. Despite the fact that the mechanism for conjugation of SUMO is very similar to that of ubiquitin and that a variety of sumoylated proteins have been identified, the E3 enzymes for sumoylation have not been identified until recently. Unlike the ubiquitination pathway, SUMO-1 can be transferred to protein substrates in the presence of Aos1/Uba2 (E1) and Ubc9 (E2). For example, IκBα and RanGAP1 can be sumoylated at a conserved acceptor site, ψKXE, without any E3 activity under in vitro conditions, implying that Ubc9 itself might play a role in determining the substrate specificity to a certain extent (Okuma et al., 1999; Rodriguez et al., 1999, 2001; Sampson et al., 2001).

Recently, three E3-like proteins for sumoylation have been identified from yeast and mammals (Johnson and Gupta, 2001; Kahyo et al., 2001; Takahashi et al., 2001a,b). A yeast protein, named Siz1, associates with both Ubc9 and Cdc3 (a member of the septin family), and strongly stimulates the sumoylation of the septin protein. Another yeast E3-like protein, Siz2, also promotes sumoylation of protein substrates that are different from the substrates modified with Siz1. However, Siz1 and Siz2 are not essential for cell survival, unlike Aos1/Uba2 and Ubc9, in yeast. PIAS1 is a mammalian member of E3-like protein for sumoylation. This protein was isolated as a SUMO-1-binding protein by yeast two-hybrid screening, and shown to interact with both p53 and Ubc9. PIAS1 markedly increases the sumoylation of p53 in U2OS cells, when p53, SUMO-1, and PIAS1 are cotransfected. In the presence of excess amounts of Aos1/Uba2 and Ubc9, both septins and p53 can be sumoylated without Siz1 or PIAS1 in vitro, respectively. However, the amount of the sumoylated proteins greatly increases when the E3-like proteins are also present. Thus, the E3-like proteins may serve to increase the affinity between Ubc9 and the substrates by bringing them in close proximity with catalytically favorable orientation, allowing the sumoylation to occur at a maximal rate (Fig. 2).

Figure 2.

Protein modification by SUMO and ubiquitin. A: Ubc9 (E2) might interact with target proteins, but with a low affinity, hence transferring SUMO to the proteins at a low rate. The E3-like proteins for sumoylation have the binding sites for both Ubc9 and target proteins, which might bring them close together (i.e., increase their affinity), thus allowing the sumoylation of the protein at a maximal rate. The RING-like domain in the E3-like protein might provide the binding site for Ubc9. B: The E2 enzymes for ubiquitin do not directly interact with target proteins. Thus, without E3 enzymes that tether both E2 and target proteins, ubiquitination could not occur. For simplicity, only a RING-finger type E3 enzyme consisting of a single polypeptide is depicted here.

One striking feature of these newly-identified E3-like proteins is the presence of a RING-like domain that is known to mediate interaction of E3 with E2 in the ubiquitin system (Freemont, 2000). For the HECT domain-containing E3 enzymes of the ubiquitin system, E3 forms a thioester bond with ubiquitin, which is subsequently transferred to target proteins. However, no RING-type E3 enzyme that can directly form a thioester bond with ubiquitin has been identified. Instead, the RING-type E3 enzymes interact with both E2 and target proteins substrates, allowing the transfer of ubiquitin from E2 to the substrates. A mutation in the RING-like domain of PIAS1 results in the loss of binding ability of the E3-like enzyme with Ubc9, but not with p53 (Kahyo et al., 2001). A similar mutation in the RING-like domain of Siz1 abolishes the binding of the E3-like enzyme with Ubc9 (Takahashi et al., 2001a). Thus, the RING-like domain of the E3-like enzymes for sumoylation might function in a similar manner with the RING domain of E3s in the ubiquitination pathway.

SUMO-specific proteases (SUSPs)

Like ubiquitin, all Ubls are synthesized as precursor proteins with one or more amino acids following the C-terminal Gly-Gly residues of the mature Ubl proteins (Saitoh et al., 1997). Thus, the tail sequences of Ubl precursors need to be removed by Ubl-specific proteases (Ulps), prior to their conjugation to target proteins. In yeast, two SUSPs, named Ulp1 and Ulp2, have been identified (Li and Hochstrasser, 1999, 2000; Schwienhorst et al., 2000). These two enzymes have sequence similarity in the regions of about 200 amino acids called the ULP domain, which harbors the catalytically active site residues. Ulp1 is a 621-amino acid protein containing two domains, a conserved C-terminal protease fold (432–621) and a weakly conserved N-terminal domain (1–432). The C-terminal protease domain can act as a functional desumoylating enzyme (Mossessova and Lima, 2000). In vivo expression of this domain in haploid yeast revealed a dominant-negative phenotype, suggesting that unregulated cleavage of sumoylated proteins blocks cell cycle progression in an analogous manner to the Ubc9-deficient strain. Ulp1 is a dual-functional enzyme that can generate the mature form of Smt3 from its precursor and cleave the isopeptide bond between Smt3 and its target proteins (Li and Hochstrasser, 1999). Even though both desumoylating enzymes and deubiquitinating enzymes belong to the cysteine protease family, Ulp1 has no sequence similarity to deubiquitinating enzymes, but shows distinct similarity to certain viral proteases in its catalytic region (Li and Hochstrasser, 1999; Andres et al., 2001). Ulp2, the second member of the yeast SUSP and originally called Smt4, was first isolated as a high-copy-number suppressor of a defective centromere-binding protein (Meluh and Koshland, 1995). Ulp2, but not Ulp1, is dispensable for cell viability. Inactivation of Ulp2 results in the accumulation of specific Smt3-protein conjugates, which are distinct from those observed in an Ulp1-deficient strain (Li and Hochstrasser, 2000; Schwienhorst et al., 2000). This mutation also causes slow and temperature-sensitive growth, defect in cell cycle progression, hypersensitivity to DNA damage, and chromosome mis-segregation.

In mammals, a number of SUSPs exist with sizes ranging from about 200 to 1,200 amino acids (Yeh et al., 2000). Suzuki et al. (1999) have identified a 30-kDa SUSP activity from extracts of bovine brain. This enzyme can generate free SUMO-1 molecule not only from SUMO-1-peptide fusions, but also from RanGAP1/SUMO-1-protein conjugates, suggesting that the removal of SUMO-1 from its protein conjugates, like deubiquitination, may play an important role in regulation of SUMO-1-mediated cellular processes. Recently, four additional SUSPs have been isolated from humans and mice (Gong et al., 2000; Kim et al., 2000; Nishida et al., 2000, 2001). These enzymes have a conserved C-terminal domain containing the His/Asp/Cys catalytic triad, but do not have any significant similarity in their N-terminal region (Fig. 3). SUSP1 expressed in Escherichia coli efficiently releases SUMO-1 from SUMO-1-β-galactosidase, but not from other ubiquitin-like protein fusions, including Smt3-β-galactosidase, suggesting its role in generation of mature SUMO-1 specifically from its precursor (Kim et al., 2000). SENP1, SMT3IP1, and SMT3IP2/Axam2 cleave the isopeptide bond between SUMO and its target proteins (Gong et al., 2000; Nishida et al., 2000, 2001). SMT3IP2/Axam2 is a mouse ortholog of rat Axam, which binds to Axin and promotes the degradation of β-catenin (Kadoya et al., 2000). Rat Axam contains the His/Asp/Cys catalytic triad, but its desumoylating activity has not been demonstrated yet. SMT3IP2/Axam2 induces the β-catenin degradation in human SW480 cells. However, an inactive mutant form of SMT3IP2/Axam2 also shows the same effect on the degradation of β-catenin. Thus, the desumoylating activity of SMT3IP2/Axam2 appears not to be involved in the Wnt signaling pathway. Whereas SUSP1 and SMT3IP2/Axam2 are detected in cytosol, SENP1 is evenly distributed throughout the nucleus (Gong et al., 2000; Kim et al., 2000; Nishida et al., 2001). SMT3IP1 is localized almost exclusively in the nucleolus during interphase (Nishida et al., 2000). Interestingly, SUSP1 mRNA is expressed much higher in the reproductive organs, such as testis, ovary, and prostate, than in other tissues, including colon, heart, and spleen (Kim et al., 2000). These different patterns of subcellular or tissue distribution may reflect the specific function of each enzyme at its location with desumoylating activity against different species of sumoylated proteins.

Figure 3.

Primary structures of desumoylating enzymes. A: The lines show the variable sequences and the boxes indicate the conserved active site regions. B: The identical residues in the active site domains are shaded, and the catalytic residues are indicated by the bold letters. The arrowheads show the catalytic triad.

A new interesting member of the SUSP family from the pathogenic bacterial species Yersinia has recently been reported (Orth et al., 2000). Yersinia pestis, which was responsible for the Black Death in the Middle Ages, has an impressive ability to overcome host defense mechanisms (Cornelis et al., 1998). Several pathogenic bacteria share a special mechanism (type-III secretion system) that allows the extracellular bacteria to inject specialized effector proteins into the cytosol of host cells (Cornelis and Van Gijsegem, 2000). In the case of Yersinia, the injected effectors, called ‘Yops’, target the cells of the immune system (especially macrophages), block phagocytosis, and downregulate the inflammatory response. One of the effector proteins, YopJ, inhibits the host immune response by blocking the activation of the MAPK pathway and the NF-κB pathway, hence preventing the production of cytokines and activation of the host immune response and apoptotic factors (Orth et al., 1999; Boland and Cornelis, 1998; Ruckdeschel et al., 1998). The YopJ protein has the His/Glu/Cys catalytic triad that is similar to that of adenoviral protease. Moreover, a mutation of the active site cysteine of YopJ blocks its ability to inhibit the MAPK pathway and NF-κB pathway, indicating that the proteolytic activity of YopJ is necessary for its function in the host cell. From the limited sequence similarity to the C-terminal catalytic domain of Ulp1 in yeast, YopJ is regarded as the first SUSP identified in bacterial species.

BIOLOGICAL ROLES OF SUMO MODIFICATION

A growing number of target proteins that are sumoylated have been reported. Two approaches have largely been used for identification of target proteins for SUMO-1 modification. One approach is the detection of slow-migrating proteins in sodium dodecyl sulfate–polyacrylamide gels, since sumoylation should increase the size of target proteins (Matunis et al., 1996; Mahajan et al., 1997, Buschmann et al., 2000). The other approach is the use of the yeast two-hybrid screening method for interaction between SUMO-1/Ubc9 and target proteins (Boddy et al., 1996; Shen et al., 1996; Desterro et al., 1999; Müller et al., 2000). A variety of sumoylated proteins have been identified with these approaches, and their biological functions have been studied. In the following section, we describe several functional models for SUMO modification.

Roles in protein translocation

The mammalian GTPase-activating protein RanGAP1 is identified as the first substrate for SUMO-1 modification (Matunis et al., 1996; Mahajan et al., 1997). RanGAP1 is a GTPase-activating protein for the small nuclear Ras-related GTPase Ran, whose function is essential for transport of proteins into the nucleus across the nuclear pore complex (Melchior et al., 1993; Moore and Blobel, 1993). RanGAP1 is highly concentrated at the nuclear pore complex and forms a stable complex with RanBP2, a component of the cytoplasmic filaments emanating from the nuclear pore complex (Wu et al., 1995; Yokoyama et al., 1995; Matunis et al., 1996; Mahajan et al., 1997). The interaction of RanGAP1 with RanBP2 requires the sumoylation of RanGAP1. This modification was identified during protein sequencing of the SUMO-1-conjugated RanGAP1. The RanGAP1 protein was present in two different forms with apparent molecular masses of 70 and 90 kDa, and the 90-kDa form was highly enriched in the nuclear envelope. From the sequence analysis of the immuno-purified 90-kDa protein, SUMO-1 was found to be covalently attached to the 70-kDa RanGAP1. This sumoylation takes place between the carboxyl group of Gly-97 in SUMO-1 and the ε-amino group of Lys-526 in the tail domain of RanGAP1 (Mahajan et al., 1998; Matunis et al., 1998). The sumoylated RanGAP1 binds stably to RanBP2 at the cytoplasmic fibrils of the nuclear pore complex through an interaction mediated by the sumoylated tail domain of RanGAP1. Replacement of Lys-526 by arginine in RanGAP1 prevents the localization of the protein to the nuclear rim, and instead results in its accumulation in the cytoplasm, because the mutant form of RanGAP1 can no longer bind to RanBP2. Since SUMO-1 itself cannot bind to RanBP2 (Mahajan et al., 1997), sumoylation may induce conformational change to expose or create a binding surface for RanBP2 in the C-terminal domain of RanGAP1.

Desumoylating activity has been detected in vitro both in cell extracts and solubilized nuclear envelopes (Matunis et al., 1996; Mahajan et al., 1997). A similar desumoylating activity that can release SUMO-1 from the sumoylated RanGAP1 has also been detected from bovine brain extract (Suzuki et al., 1999). However, it remains unclear whether these activities are indeed involved in the control of the nuclear protein import pathway. RanBP2 has also been identified as a target protein for SUMO-1 modification in Xenopus egg extract (Saitoh et al., 1997, 1998). Xenopus homologues of RanBP2, RanGAP1, and Ubc9 form a tight complex in egg extract, and Ubc9 is required for RanGAP1 sumoylation. However, the functional consequence of RanBP2 sumoylation is unknown.

Role in subnuclear structure formation

PML is a RING-finger protein with tumor suppressor activity. In the majority of the patients with acute promyelocytic leukemia, the PML gene has undergone a fusion with the retinoic acid receptor α gene (RARα) by a reciprocal chromosomal translocation, yielding a chimeric PML/RARα protein (de Thé et al., 1991; Kakizuka et al., 1991). In addition to a RING domain, PML contains two additional Cys- and His-rich regions (B1 and B2 boxes) and a coiled-coil domain. Whereas the precise roles of the RING domain and B boxes are unknown, the coiled-coil domain is known to function as a dimerization interface for the formation of PML-PML homodimers as well as of PML-PML/RARα heterodimers (Kastner et al., 1992). PML knockout mice are viable and develop normally, but are more susceptible to viral infections and tumor development (Wang et al., 1998a,b).

PML is enriched in discrete subnuclear matrix-associated structures, called the PML nuclear bodies and also termed as ND10, Kr body, and PML oncogenic domains (PODs) (Dyck et al., 1994; Koken et al., 1994; Weis et al., 1994). PML is one of the target proteins for SUMO-1 modification. The sumoylated PML is preferentially targeted to the nuclear bodies, whereas the unmodified form remains in the nucleoplasmic fraction (Boddy et al., 1996; Sternsdorf et al., 1997; Müller et al., 1998; Zhong et al., 2000). Sumoylation is also required for the nuclear body localization of several other proteins, including Sp100, Daxx, CBP, and ISG20. In PML−/− cells, these nuclear body proteins and SUMO-1 fail to accumulate in the nuclear bodies, which lead to disaggregation of the nuclear bodies (Ishov et al., 1999; Zhong et al., 2000). Introduction of exogenous PML into the PML−/− cells causes the relocalization of the nuclear body proteins, but that of a PML mutant that can no longer be sumoylated fails to reform the nuclear bodies, indicating the role of SUMO modification of PML in recruiting other nuclear body proteins and in formation of the PML nuclear bodies.

Recently, another role of SUMO-1 modification of PML has been suggested by Lallemand-Breitenbach et al. (2001). They showed that As2O3 triggers the proteasome-dependent degradation of PML and PML/RARα, and this process requires sumoylation at a specific site, Lys-160, of PML. Sumoylation of PML is dispensable for its As2O3-induced matrix targeting and formation of primary nuclear aggregates, but is required for the formation of the secondary shell-like nuclear bodies. Interestingly, only these mature nuclear bodies harbor 11S proteasome components, which are further recruited upon As2O3 exposure. This sumoylation directly or indirectly promotes the PML degradation, suggesting that the mature nuclear bodies might be the site for the intranuclear proteolysis.

Sp100, another major nuclear body protein, is also a target of SUMO-1 modification (Sternsdorf et al., 1997). Sp100 is an interferon-inducible protein, which was initially characterized as an antigen reactive with antibodies from patients with autoimmune disorders (Szostecki et al., 1990). The SUMO-1 acceptor site in Sp100 is Lys-297, and replacement of this residue by arginine completely abolishes the SUMO-1 modification of Sp100, but without any effect on the localization of Sp100 in the nuclear bodies (Sternsdorf et al., 1999). Sumoylation of Sp100 promotes its interaction with members of the HP1 family of nonhistone chromosomal proteins, suggesting its regulatory role in chromatin organization and in functional interplay between the nuclear bodies and chromatin (Lehming et al., 1998; Seeler et al., 1998, 2001).

SUMO-1 modification of homeodomain-interacting protein kinase 2 (HIPK2) and TEL induces formation of subnuclear structures, which are distinct from the PML nuclear bodies (Kim et al., 1999; Chakrabarti et al., 2000). HIPK2 is one of the nuclear protein kinases that act as corepressors for homeodomain transcription factors and localize to nuclear speckles (Kim et al., 1998). TEL, also known as ETV6, a transcription factor specifically required for hematopoiesis within the bone marrow (Wang et al., 1998). TEL also functions as a transcription repressor that represses the target gene through the histone-deacetylase pathway (Fenrick et al., 1999; Lopez et al., 1999). The sumoylated TEL localizes to nuclear speckles in a cell cycle-specific manner (Chakrabarti et al., 2000). In both cases, mutations of the lysine residues in the SUMO acceptor sites impair specific nuclear speckle formation. Taken together, although the biological function of these nuclear bodies (or speckles) is unclear, SUMO-1 modification appears to play a key role in the formation of subnuclear structure formation.

The immediate-early proteins IE1 and IE2 from human cytomegalovirus and the BZLF1 protein from Epstein-Barr virus are modified by SUMO-1 (Müller and Dejean, 1999; Hofmann et al., 2000; Adamson and Kenney, 2001; Ahn et al., 2001). Interestingly, both IE1 and IE2 localize transiently in the PML nuclear bodies, and sumoylation of IE1 and BZLF1 occurs in parallel with desumoylation of PML and with disassembly of the PML nuclear bodies. Since sumoylation of PML recruits p53 to the PML nuclear bodies and stimulates its transcriptional and pro-apoptotic activities, it has been suggested that the virus-induced disruption of the PML nuclear bodies could be a strategy for the virus to overcome the negative effect of p53 on cell proliferation and apoptosis. However, it has recently been demonstrated that a point mutation of Lys-450, the major sumoylation site of IE1, did not interfere with its targeting to and disruption of the PML nuclear bodies (Xu et al., 2001). Thus, the function of SUMO-1 modification of the viral proteins remains unknown.

Modulation of transcriptional activity

Although the function of the PML nuclear bodies is still unclear, SUMO-1 modification of PML followed by recruitment of certain proteins to the nuclear bodies can modulate transcriptional activity. A transcriptional repressor, Daxx, can be recruited to the nuclear bodies upon sumoylation of PML, thereby, relieving the Daxx-mediated transcriptional repression of its target genes (Ishov et al., 1999; Li et al., 2000; Lehembre et al., 2001). Sumoylation of PML also recruits p53 to the nuclear bodies, but leads to a stimulation of the transcriptional and pro-apoptotic activities of p53 (Fogal et al., 2000). In addition, recruitment of p53 to the nuclear bodies has been suggested to trigger post-translational modifications in the N- and C-terminal regions, such as acetylation, which also promote the transcriptional activity of p53 (Giaccia and Kastan, 1998; Pearson et al., 2000).

The p53 protein itself also is a target for SUMO-1 modification (Gostissa et al., 1999; Rodriguez et al., 1999; Müller et al., 2000). Upon exposure of cells to UV irradiation, SUMO-1 is covalently attached to p53 at a single site, Lys-386, in the C-terminal region, which is known to regulate the DNA-binding activity of the protein (Kubbutat et al., 1998; Honda and Yasuda, 1999). Since SUMO-1 and ubiquitin do not compete for the same lysine residue in p53 for their modification, sumoylation at Lys-386 does not affect the stability of p53. Instead, the SUMO-1 modification activates the transcriptional activity of p53. In reporter assays, overexpression of SUMO-1 leads to an increase in the p53-dependent transcriptional activity. Furthermore, the apoptotic potential of a mutant form of p53, in which Lys-386 is replaced by arginine (K386R), is moderately impaired, implying that sumoylation of p53 is necessary for exerting its full apoptotic activity (Müller et al., 2000). However, contradictory results have also been reported on the function of SUMO-1 modification in the transcriptional activation of p53 (Minty et al., 2000; Kwek et al., 2001). They performed similar reporter assays, but could not detect any transcriptional activation of p53 upon co-transfection with SUMO-1. In addition, the K386R mutant suppresses the colony formation of Saos-2 and H1299 cells to a similar extent seen with the wild-type p53. Further studies are required for the role of sumoylation of p53 in its transcriptional regulation.

In addition to p53, several transcription factors, such as c-Jun, androgen receptor (AR), and heat shock transcription factor 1 and 2 (HSF1 and HSF2), have been identified as targets for SUMO modification (Müller et al., 2000; Poukka et al., 2000; Goodson et al., 2001; Hong et al., 2001). The substitution of the lysine residues, the SUMO acceptor sites, with arginine in both c-Jun and AR enhances their transcriptional activity, suggesting that sumoylation negatively regulates the activity. The consensus sequence for sumoylation on AR is also present in the N-terminal domains of other steroid receptors, such as glucocorticoid, mineralocorticoid, and progesterone receptors. Thus, SUMO modification might be a common mechanism for regulation of transcriptional activity of the steroid receptor superfamily. HSF1 and HSF2 are transcription factors that mediate the induction of heat shock protein gene expression under environmental stress conditions (Cotto and Morimoto, 1999). SUMO modification enhances the DNA-binding ability of both proteins as assessed by gel-shift assays. In addition, a mutation of the SUMO acceptor lysine on HSF1 results in a significant decrease in stress-induced transcriptional activity of HSF1 in vivo (Goodson et al., 2001; Hong et al., 2001). Thus, SUMO modification might also involve in regulation of transcriptional activity of the heat shock transcription factors.

Antagonistic role against ubiquitin

NF-κB transcription factors activate diverse genes, including those involved in immune function, inflammatory response, cell adhesion, and growth control, in an inducible manner (Baeuerle and Henkel, 1994; Siebenlist et al., 1994). In nonstimulated cells, NF-κB is kept as inactive complexes in the cytoplasm through interactions with inhibitory proteins, IκBs (Beg and Baldwin, 1993; Verma et al., 1995). Upon stimulation by effectors, such as proinflammatory cytokines, phorbol esters, oxidants, or viral infection results, IκBα, a major inhibitor of NF-κB, is rapidly phosphorylated by a signal inducible IκB kinase (IKK) complex (Mercurio et al., 1997; Woronicz et al., 1997; Zandi et al., 1997). Once IκBα is phosphorylated, it is recognized by the ubiquitin-conjugating machinery, which utilizes Lys-21 and Lys-22 as ubiquitination sites in the inhibitor protein. Poly-ubiquitination of IκBα targets it for rapid degradation by the 26S proteasome (Alkalay et al., 1995; DiDonato et al., 1996). This results in liberation of NF-κB, which then translocates from the cytosol to the nucleus to activate the transcription of the genes containing the NF-κB-binding sites in their promoter/enhancer regions. Thus, the key regulatory steps in the NF-κB signal transduction pathway is the phosphorylation and ubiquitin-dependent degradation of IκBs.

Interestingly, SUMO-1 can also be conjugated at one of the ubiquitination sites on IκBα Lys-21, and this modification results in stabilization of IκBα against degradation by the 26S proteasome (Desterro et al., 1998). The sumoylated forms of IκBα have been detected in a number of cell types, and are resistant to TNF-α-induced degradation. The sumoylated IκBα remains associated with NF-κB, and overexpression of SUMO-1 prevents the signal-induced activation of NF-κB-dependent transcription. These results indicate that SUMO-1 can function as an antagonist of ubiquitin.

Another example of the antagonistic role of SUMO-1 against ubiquitin is the stabilization of Mdm2 (Buschmann et al., 2000). Mdm2 is an E3 ligase containing a RING domain that catalyzes ubiquitination of p53 as well as of itself (Honda et al., 1997). In normal cells, most of the Mdm2 proteins are sumoylated at Lys-446, which is the same site for self-ubiquitination. Ubc9 binds to N-terminal region (amino acids 40–59) of Mdm2 (Buschmann et al., 2001). Ubc9-mediated sumoylation protects Mdm2 from destabilization, but increases p53 ubiquitination and degradation. Upon DNA damage, such as by exposure of cells to UV and/or X-rays, the stability of p53 is dramatically increased, leading to cell cycle arrest or apoptosis (Oren, 1999). The stabilization of p53 comes from degradation and consequent loss of the E3 ligase function of Mdm2 on p53. In this process, SUMO-1 is presumed to be deconjugated from the sumoylated Mdm2 prior to ubiquitination. However, this desumoylating activity that might be induced by DNA damage has not been identified yet. These results show a role of sumoylation in counteracting the ubiquitination of target proteins, thus leading to the model that SUMO acts as a protein stabilizer.

Other proteins

An increasing number of other proteins have been identified to be sumoylated, but the biological significance of their sumoylation is not yet known. These proteins include topoisomerases, Top1 and Top2 (Mao et al., 2000a,b), glucose transporter proteins GLUT1 and GLUT4 (Giorgino et al., 2000), and chromatin factor TIF1α (Seeler et al., 2001) in mammals, transcriptional repressor protein Tramtrack69 (Lehembre et al., 2000), transcription factor Dorsal (Bhaskar et al., 2000), and calcium/calmodulin-dependent kinase CaMKII (Long and Griffith, 2000) in Drosophila, and bovine papilloma virus E1 protein (Rangasamy and Wilson, 2000; Rangasamy et al., 2000) and adenovirus E1B protein (Endter et al., 2001) in viruses.

CONCLUSIONS AND PERSPECTIVES

The versatility of SUMO modification of proteins as a cellular regulatory mechanism is now well established, and appears to be comparable to that of protein ubiquitination. Indeed, sumoylation and ubiquitination share a common mechanism in their enzymology as well as in the regulation of many cellular pathways. SUMO and ubiquitin have a common tertiary structure characterized by the ubiquitin β-fold. Both utilize E1, E2, and E3 cascade enzymes for their conjugation to target proteins. Since only a few E3-type proteins containing a RING-like domain for sumoylation have been identified so far, it will be interesting to see whether other target proteins require the E3-like activities for sumoylation and whether there exists another type of E3-like enzymes that instead have a HECT-like domain as the E3 enzymes for the ubiquitination pathway. Although all of the sumoylated proteins so far identified employ a single E2-like enzyme (Ubc9), a possibility for the presence of a new type(s) of E2-like enzyme for SUMO conjugation may not necessarily be excluded. Unlike ubiquitin, the SUMO family consists of three members, SUMO-1, -2, and -3. Moreover, the latter two family members, unlike SUMO-1, have a consensus ψKXE motif for poly-SUMO chain formation. Therefore, many studies are required also regarding the enzymology of SUMO-2 and -3 conjugation, for identification of their specific target proteins and ultimately for their role in the regulation of cellular processes.

SUMO modification is a dynamic and reversible process that regulates cellular pathways, analogous to the post-translational modification by phosphorylation and dephosphorylation. Shuttling of proteins between the nucleus and the cytosol, assembly and disassembly of nuclear bodies, stabilization and destabilization of proteins, direct or indirect activation and inactivation of transcriptional factors, as well as many other SUMO-mediated processes should all be regulated by dynamic SUMO conjugation and deconjugation. In addition, all of the SUMO family members, like ubiquitin or other Ubls, are synthesized as precursors with several amino acids following the C-terminal Gly-Gly residues of the mature SUMO proteins. Ulp1 and Ulp2 are the desumoylating enzymes in yeast that are most well characterized. A number of SUSPs in humans and mice have also been identified via EST databank search, based on the sequence of the ULP domain. The presence of a large number of mammalian SUSPs leads to a speculation that distinct enzymes might have specialized functions. However, no evidence has been obtained so far for the direct involvement of these enzymes in the SUMO-mediated cellular processes in mammals. What determines the substrate specificity of SUSPs in vivo is also unknown. Genetic studies including knock-out strategies and proteomic analysis of sumoylated proteins may provide the information about the substrate specificity and physiological function of these enzymes.

Acknowledgements

We thank Young Min Rho, Soo Joon Choi, and Kyuhee Oh for their helpful discussion. We also apologize in the event that any relevant publications were inadvertently omitted.

Ancillary