Evidence for Small Ubiquitin-like Modifier-dependent Nuclear Import of the Thymidylate Biosynthesis Pathway*

Perturbations in folate-mediated one-carbon metabolism increase rates of uracil misincorporation into DNA during replication, impair cellular methylation reactions, and increase risk for neural tube defects and cancer. One-carbon metabolism is compromised by folate deficiency and common genetic polymorphisms. In this study, the mechanism for the preferential partitioning of cytoplasmic serine hydroxymethyltransferase (cSHMT)-derived methylenetetrahydrofolate to de novo thymidylate biosynthesis was investigated. The cSHMT enzyme was shown to interact with UBC9 and was a substrate for UBC9-catalyzed small ubiquitin-like modifier (SUMO) modification in vitro. SUMOylated cSHMT was detected in extracts from S phase MCF-7 cells, and cSHMT was shown to localize to the nucleus and nuclear periphery during the S and G2/M phases of the cell cycle. A common single nucleotide polymorphism (L474F-cSHMT) impaired the UBC9-cSHMT interaction and inhibited cSHMT SUMOylation in vitro. The three folate-dependent enzymes that constitute the de novo thymidylate biosynthesis pathway, cSHMT, thymidylate synthase, and dihydrofolate reductase, all contain SUMO modification consensus sequences. Compartmentation of the folate-dependent de novo thymidylate biosynthesis pathway in the nucleus accounts for the preferential partitioning of cSHMT-derived folate-activated one-carbon units into thymidylate biosynthesis; the efficiency of nuclear folate metabolism is likely to be modified by the cSHMT L474F polymorphism.

The term folate refers to a family of enzyme cofactors that carry and chemically activate single carbons at three oxidation states (1,2). Folate-mediated one-carbon metabolism is a metabolic network that is required for the biosynthesis of purines and thymidylate and the remethylation of homocysteine to methionine (Fig. 1). Folate cofactors incorporate activated formate into the 2-and 8-positions of the purine ring, provide activated formaldehyde and two electrons for the methylation of dUMP to dTMP, and provide activated methanol for the remethylation of homocysteine to methionine. Methionine can be adenosylated to S-adenosylmethionine (AdoMet), 2 which serves as a methyl donor for numerous substrates, including lipids, neurotransmitters, DNA, and histones (3). Formate generated in the mitochondria is the primary source of single carbons for cytoplasmic one-carbon metabolism ( Fig. 1) (4). Alternatively, folate-activated one-carbon units can be derived in the cytoplasm through the activity of cytoplasmic serine hydroxymethyltransferase (cSHMT), which catalyzes the reversible conversion of THF and serine to glycine and methylene-THF, but this enzyme is not expressed in all cells (5).
Perturbation of one-carbon metabolism can result from vitamin deficiencies and/or penetrant single nucleotide polymorphisms in genes that encode folate-dependent enzymes (2). One-carbon metabolism is essential for genome stability and methylation; biomarkers for impaired folate metabolism include elevated uracil content in DNA due to uracil misincorporation during replication and repair as well as depressed cellular methylation capacity (6,7). Uracil accumulation in DNA leads to genome instability, and chromatin hypomethylation affects chromatin structure, genome stability, and gene expression (8). It is not known if the associations between disruptions in folate metabolism and pathologies (including colon cancer and cardiovascular disease) and developmental anomalies (including neural tube closure defects) result from altered AdoMet synthesis and/or folate-dependent thymidylate biosynthesis (2).
Folate-dependent biosynthetic pathways compete for a limited pool of folate cofactors (9,10). The concentration of folatebinding proteins in cells exceeds that of folate derivatives; therefore, the concentration of free folate in the cell is negligible (10 -12). This competition is most pronounced for the two reactions that utilize methylene-THF: thymidylate synthesis catalyzed by thymidylate synthase (TS) and 5-methyl-THF synthesis catalyzed by methylenetetrahydrofolate reductase (MTHFR) (Fig. 1). The MTHFR-catalyzed reaction commits one-carbon units to the methionine cycle ( Fig. 1) (13). Both reactions are essentially irreversible in vivo (10). Metabolic labeling studies, in vitro binding experiments, and mathematical modeling have indicated that methylene-THF is preferentially directed toward methionine synthesis relative to thymi-* This work was supported by Public Health Service Grant DK58144. The costs dylate biosynthesis (13,14). There is increasing evidence that the partitioning of methylene-THF between methionine and thymidylate synthesis is central to the origin of folate-related pathologies. For example, a prevalent MTHFR variant (A222V) is thermolabile; carriers exhibit 40 -70% less enzymatic activity, impaired methylation, and hypomethylated DNA but more robust de novo thymidylate biosynthesis compared with carriers of the more common allele (15). The A222V polymorphism is associated with increased risk for developmental anomalies, including neural tube closure defects (16), but affords protection from colon cancer (15). However, little is known about the regulation of methylene-THF partitioning between the thymidylate and AdoMet synthetic pathways.
Recently, we have provided evidence that the cSHMT enzyme contributes to the regulation of methylene-THF partitioning between the thymidylate and AdoMet biosynthetic pathways and is rate-limiting for de novo thymidylate synthesis in MCF-7 cells (17). Previous isotope tracer studies demonstrated that cSHMT-derived methylene-THF is preferentially directed to thymidylate biosynthesis relative to the methionine synthesis pathway (17). Furthermore, cSHMT is an abundant, high affinity 5-methyl-THF binding protein (10) (Fig. 1). A 2-fold increase in cSHMT protein levels was sufficient to sequester cytoplasmic folate as 5-methyl-THF in MCF-7 cells and deplete AdoMet levels ( Fig. 1) (17). Therefore, cSHMT is poised within the one-carbon network to contribute to the etiology of folate-related pathologies. In this study, the mechanism underlying the preferential partitioning of cSHMT-derived methylene-THF into thymidylate synthesis was investigated.

EXPERIMENTAL PROCEDURES
Cell Lines and Media-Human MCF-7 mammary adenocarcinoma cells (HTB22) and HeLa cells (CCL2) were obtained from ATCC. Cells were cultured in ␣-minimal essential medium (Hyclone Laboratories) supplemented with 11% fetal calf serum (Hyclone Laboratories) and maintained at 37°C in a 5% CO 2 atmosphere.
Isolation of Mouse Embryonic Fibroblasts-Wild-type C57/BL6 and isogenic cshmt Ϫ/Ϫ embryos were harvested at embryonic day 14 and genotyped by PCR. 3 Fibroblasts were isolated and cultured on collagen-coated plates in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 11% fetal bovine serum, penicillin, and streptomycin as described elsewhere (18).
Immunoblotting-Cells were harvested by gentle scraping and washed two times with 10 ml of phosphate-buffered saline (PBS). Cell pellets were resuspended and lysed in mammalian protein extraction reagent (M-PER) (Pierce) containing protease inhibitor mixture (Sigma) for 10 -20 min on ice. Isolation of cell nuclei was performed using the NE-PER kit from Pierce. For the identification of SUMOylated proteins, a SUMO protease inhibitor, N-ethylmaleimide (10 mM), was added to the lysis buffer. Protein concentrations were determined by the Lowry method as modified by Bensadoun (19). Cellular proteins were separated by SDS-PAGE using 12% polyacrylamide gels. Proteins were transferred at 4°C to a polyvinylidene fluoride microporous membrane (Millipore Corp.) using a Transblot apparatus (Bio-Rad). Following transfer, membranes were incubated with primary antibody for 15 h at 4°C, washed with PBS containing 0.1% Tween 20, and then incubated for 4 -15 h with the appropriate horseradish peroxidase-conjugated secondary antibody. Protein bands were visualized using the horseradish peroxidase SuperSignal chemiluminescent substrate system (Pierce). For cSHMT detection, affinity-purified sheep antihuman cSHMT antibody was diluted 1:20,000, and rabbit anti-sheep IgG-horseradish peroxidase (Pierce) was diluted 1:10,000. For TS detection, affinity-purified sheep anti-human TS antibody (Abcam) was diluted 1:5000. For detection of SUMOylated proteins, rabbit polyclonal anti-human SUMO (Abcam) was diluted 1:5000, and goat anti-rabbit IgG-horseradish peroxidase (Pierce) was diluted 1:10,000. Rabbit anti-human glyceraldehyde-3-phosphate dehydrogenase antibody (Novus) was used at a 1:100,000 dilution, FIGURE 1. Folate-mediated one-carbon metabolism. THF-mediated one-carbon metabolism is required for the synthesis of purines and thymidylate and the remethylation of homocysteine to methionine. The hydroxymethyl group of serine is the major source of one-carbon units that are generated in the mitochondria in the form of formate or in the cytoplasm through the activity of cSHMT. Mitochondrially derived formate can enter the cytoplasm and function as a one-carbon unit for folate metabolism. The cSHMT enzyme also inhibits homocysteine remethylation by sequestering 5-methyl-THF in the cytoplasm. The one-carbon unit is labeled in boldface type. The inset shows the thymidylate synthesis pathway, which involves the three enzymes cSHMT, TS, and DHFR. AdoHcy, S-adenosylhomocysteine. mouse anti-rat lamin B1 (Abcam) was used at a 1:1000 dilution, and mouse anti-bovine calpain M (Affinity Bioreagents) was used at a 1:1000 dilution.
Yeast Two-hybrid Assay-A HeLa cell-derived cDNA expression library encoding Gal4 activation domain fusion proteins was screened with a cSHMT fusion protein containing the Gal4 DNA-binding domain. To generate the cDNA encoding the cSHMT fusion protein, the human cSHMT cDNA was amplified by PCR using the primers 5Ј-TAGAATTCATGACGA-TGCCAGTC-3Ј and 5Ј-TAGTCGACTTAGAAGTCAGG-CAG-3Ј, which contain the restriction sites (EcoRI and SalI, respectively) shown in boldface type. The product was cloned into the pGBK-DNA binding vector and then transformed into the yeast strain AH109. A pretransformed HeLa cDNA library in Y187 yeast cells (Clontech) was mated to AH109 cells transformed with the pGBK-cSHMT vector following the Clontech Matchmaker protocol. After mating for 24 h, cells were plated on His Ϫ , Ade Ϫ , Leu Ϫ , Trp Ϫ dropout medium containing X-␣galactosidase. Positive colonies were picked after a 4-day incubation at 30°C. Clones were tested against the negative controls lamin and empty cassette plasmids according to the Matchmaker protocol.
Coimmunoprecipitation-Cells were lysed as described above in M-PER buffer containing protease inhibitor mixture. The extract (150 g) was incubated for 30 min at 4°C with 30 l of protein A/G-conjugated agarose beads to remove proteins that bound nonspecifically to the matrix. The precleared extracts were then incubated with 5 g of either anti-cSHMT, UBC9, UBC13, SUMO, or HA (Santa Cruz Biotechnology, Inc., Santa Cruz, CA) antibodies overnight at 4°C with 30 l of protein A/G-agarose beads. The beads were collected and washed three times with PBS. SDS-PAGE sample buffer was added to the beads to release protein complexes. Immunoblots were prepared as described above.
GST Pull-down Assay-The UBC9 and UBC13 cDNAs were amplified by PCR and cloned into the pGEX4T-2 (Amersham Biosciences) vector using the following primers: UBC9 forward, 5Ј-TAGAATTCCCATGTCGGGGATCGCC-3Ј; UBC9 reverse, 5Ј-TAGTCGACTTATGAGGGCGCAAAC-3Ј; UBC13 forward, 5Ј-TAGAATTCCCATGGCCGGGCTGCCC-3Ј; UBC13 reverse, 5Ј-TACTCGAGTTAAATATTATTCATGG-3Ј). The EcoRI and SalI (UBC9) and EcoRI and XhoI (UBC13) sites are shown in boldface type. Equal amounts of GST fusion protein (5 g) and glutathione-Sepharose beads were incubated with 500 g of MCF-7 extract with end over end shaking for 3 h at 4°C. After incubation, the beads were washed five times with PBS. Proteins bound to the column were eluted with 20 mM glutathione, separated by SDS-PAGE, and identified by immunoblotting.
Quantitative ␤-Galactosidase Assay-Plasmids containing either the cSHMT, cSHMT-K38R, cSHMT-K38R/K39R, or cSHMT-L474F cDNA fused to the DNA binding domain of Gal4 were cotransformed into Y187 yeast with a plasmid encoding UBC9 fused to the Gal4 activating domain. The cells were plated on selection medium that lacked leucine and tryptophan to maintain both plasmids. Colonies were picked from the plates, grown in selection medium overnight, and then diluted to an A 600 of 0.2 in 5 ml of selection medium. After a 6-h incubation at 30°C, the A 600 was recorded, and cells were harvested and lysed in 300 l of Y-PER buffer (Pierce). After a 30-min lysis, the extracts were centrifuged to remove the insoluble material. ␤-Galactosidase assay buffer (300 l; Pierce) was added to the extracts. After a 4 -12-h incubation, the reactions were terminated by the addition of 1 M Na 2 CO 3 (400 l). The A 420 of each reaction was recorded, and the ␤-galactosidase activity was calculated using the following formula: ␤-galactosidase units ϭ (1000 ϫ A 420 )/(t ϫ A 600 ) (t ϭ time of incubation with substrate). All values were normalized to the ␤-galactosidase activity obtained from the cSHMT-UBC9 interaction.
Purification of Recombinant cSHMT-Plasmids containing cSHMT cDNAs fused to a His 6 tag (pet28a) at the N terminus were transformed into BL21 cells (Novagen) and purified as described previously (20).
In Vitro SUMOylation Assays-SUMOylation reactions were performed using a kit from ActiveMotif following the manufacturer's instructions. Briefly, a 3 M concentration of either recombinant wild-type, K38R, K39R, K38R/K39R, or L474F human cSHMT was added to the purified SUMO components: Aos1/Uba2 (E1), UBC9 (E2), and SUMO. The reactants were mixed briefly and incubated at 30°C for 3 h. Reactions with a mutant SUMO that cannot be activated by the E1/E2 machinery were performed as a negative control. When RanBP2 was included in the reaction, 5 ng of purified recombinant GST-RanBP2⌬FG (plasmid was a kind gift of Frauke Melchior (22)) was added in place of 2 l of water. SUMOylation of p53 served as a positive control (data not shown). Reactions were terminated by the addition of 6ϫ SDS-PAGE loading buffer and heated to 95°C for 10 min before immunoblotting using antibodies against cSHMT or SUMO.
Cell Cycle Synchronization and Analysis-MCF-7 or HeLa cells at 30% confluence were arrested at various stages of the cell cycle with 30 M lovastatin (Sigma), 1 mM hydroxyurea (Sigma), or 80 ng/ml nocodazole (Calbiochem). After a 20 -24-h exposure, half of the cells were collected for fluorescence-activated cell sorting analysis, and the rest were fixed with methanol and prepared for immunofluorescence confocal microscopy (described below). For cell synchronization experiments, cells were grown to 40% confluence and treated with 80 ng/ml nocodazole for 20 h. After the nocodazole was removed (t ϭ 0), the cells were washed, and ␣-minimal essential medium was applied to the cells. Cells were isolated for fluorescence-activated cell sorting and for Western analysis at time 0, 12, and 20 h. Fluorescence-activated cell sorting analysis was performed using 3 mM sodium citrate containing 1% Triton X-100 with 50 ng/ml propidium iodide as the lysis/DNA binding reagent at the Biomedical Sciences Flow Cytometry Core Laboratory at Cornell University.
Immunofluorescence Confocal Microscopy-MCF-7 cells or mouse embryonic fibroblasts were grown on sterile coverslips in ␣-minimal essential medium then fixed with 100% methanol for 5 min and allowed to air-dry. The coverslips were rinsed briefly in PBS before incubation with the blocking solution (2% bovine serum albumin, 0.1% Triton X-100 in PBS) for 2 h. The blocking solution was removed, and a solution of primary antibodies in 2% bovine serum albumin-PBS (antibody dilution was 1:500) (either cSHMT, nonspecific sheep IgG (control), or a Q-dot 605 nM conjugated cSHMT primary antibody) was applied to the coverslips for 1-2 h at room temperature in the dark. To visualize the nucleus, 10 nM TOPRO-3 (Molecular Probes, Inc., Eugene, OR) DNA binding dye was added, and the cells were then washed five times with PBS. Samples that required a secondary antibody for detection were incubated with an Alexa-fluor 488 nM conjugated donkey antisheep antibody (Invitrogen) for 1 h at room temperature in the dark. The coverslips were washed three times with PBS. Coverslips were mounted to slides using mounting medium (Sigma), and confocal images were obtained at the Microscope Imaging Facility at Cornell University. The cSHMT protein levels in the cytoplasm and nucleus were quantified using the MetaMorph program, version 6.1.
Molecular Modeling of the cSHMT-UBC9 Interaction-UBC9 from the crystal structure of a RanGAP1-UBC9 complex (23) (Protein Data Bank entry 1KPS) was manually docked with human cSHMT (Protein Data Bank entry 1BJ4) using the program O (24), positioning the molecules so that Lys 38 and Asn 42 of cSHMT chain A occupied approximately the same positions as Lys 526 and Glu 528 of RanGAP1, relative to UBC9 residues 86 -93. After overall molecular placement, side chains at the cSHMT-UBC9 interface were manually adjusted, and energy minimization of the complex was carried out using CNS (25). Buried surface area was calculated for the RanGAP1-UBC9 and cSHMT-UBC9 complexes using the EBI PISA server (available on the World Wide Web at www.ebi.ac.uk/msd-srv/prot_int/ pistart.html).

RESULTS
Identification of cSHMT as a Nuclear Protein-A yeast twohybrid screen was performed to identify proteins that physically interact with cSHMT and thereby facilitate the preferential partitioning of cSHMT-derived methylene-THF to the de novo thymidylate biosynthesis pathway. Several proteins were identified as cSHMT binding partners (Table 1), including UBC9, UBC13, and proliferating cell nuclear antigen (PCNA).
These three proteins localize to the nucleus and nuclear periphery and are involved in DNA replication and repair (26). Their interaction with cSHMT was confirmed through replicate screening and co-immunoprecipitation using rabbit anti-human UBC9 and UBC13 antibodies ( Fig. 2A). Endogenous cSHMT protein in MCF-7 cell extract was also shown to interact with GST-UBC9 and GST-UBC13 fusion proteins (Fig. 2B).
UBC13 is an E2 ubiquitin-conjugase that functions in the error-free DNA postreplication repair pathways (27). UBC9 is an E2 SUMO conjugase (28). SUMO modification protects proteins from ubiquitin-mediated degradation, targets proteins for nuclear or subnuclear localization, enables protein-protein interactions, and/or modifies protein activity (29,30). PCNA is a highly conserved polymerase-associated protein that is synthesized in G 1 and S phases of the cell cycle and localizes to sites of DNA synthesis. PCNA is SUMOylated by UBC9 in S phase to promote DNA synthesis; ubiquitination of PCNA occurs in the nucleus and is a signal for error-free DNA repair (31,32).
The preferential partitioning of cSHMT-derived methylene-THF to TS could occur through a physical TS-cSHMT interaction that permits substrate channeling. Alternatively, compartmentation of the thymidylate biosynthesis pathway apart from the methionine synthesis pathway would also enable the preferential partitioning of cSHMT-derived methylene-THF to thymidylate synthesis. There was no evidence for a direct interaction between TS and cSHMT when tested by the yeast twohybrid assay. Therefore, the nuclear compartmentation of the de novo thymidylate biosynthesis pathway was explored as a mechanism to account for the preferential partitioning of cSHMT-derived one-carbon units to the de novo thymidylate biosynthesis pathway.
Localization of cSHMT and TS to the Nucleus-Western blot analyses demonstrated that the cSHMT and TS enzymes were present in purified HeLa cell nuclear extracts (Fig. 3A). SUMOylated cSHMT enzyme was visualized only in extracts prepared with the SUMO protease inhibitor, N-ethylmaleimide, and SUMOylated cSHMT was enriched in the nuclear fraction relative to the cytosolic fraction (Fig. 3B). Immunoprecipitates generated from MCF-7 cell extracts using a rabbit anti-human SUMO antibody contained cSHMT (55 kDa) and SUMOylated cSHMT (75 kDa) (Fig. 3C). SUMO isopeptidases are highly active and located in the nucleus and nuclear periphery; they rapidly hydrolyze SUMO polypeptides from proteins in cell extracts (33). Therefore, we could not assess if all nuclear cSHMT was SUMOylated. The active cSHMT enzyme is a homotetramer in mammalian cells. It is possible that only one subunit requires SUMOylation to transport the tetramer, which would result in the presence of both SUMO-modified and -unmodified cSHMT in the nucleus. Many proteins that undergo SUMO modification are present both in the cytoplasm and nucleus, and typically only a small fraction localizes to the nucleus (34). Nuclear localization of cSHMT was verified by confocal microscopy in mouse embryonic fibroblasts isolated from C57/ BL6 and C57/BL6 cshmt Ϫ/Ϫ embryos (Fig. 4A). The cSHMT protein was present in both the nuclear and cytoplasmic com-partment. SUMO modification and SUMO-dependent nuclear localization have been shown to be cell cycle-regulated (35,36). For example, PCNA is SUMOylated during S phase, the septin proteins are SUMO-modified during mitosis (35), and topoisomerase SUMOylation occurs during G 2 /M. The cSHMT enzyme was localized almost exclusively to the cytoplasm of MCF-7 cells blocked at G 1 /stationary phase of the cell cycle, whereas cSHMT localized to the cytoplasm, nuclear periphery, and nucleus in the S phase and G 2 /M phase of the cell cycle (Fig.  4B). The levels of cSHMT in the nuclear compartment were quantified from confocal images, and nuclear cSHMT accounted for 20 -40% of total cellular cSHMT in S phase cells. Immunoblots of nocodazole synchronized cell extracts indicated that SUMOylated cSHMT was enriched in the S phase of the cell cycle (Fig. 5). These results from Fig. 4 and 5 indicate that cSHMT is predominantly SUMOylated and localized to the nucleus in S phase cells and remains in the nucleus during the G 2 /M phase of

. Nuclear localization and SUMOylation of cSHMT.
A, cSHMT is present in nuclear extracts. Nuclei were isolated and purified from HeLa cells, and immunoblots of whole cell, nuclear, and cytoplasmic extracts were probed with either anti-TS or anti-cSHMT antibodies. TS and cSHMT were present in the nuclear and cytosolic fractions. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH) is shown as a control to demonstrate that the nuclear fractions are free of cytoplasmic contamination. B, the experiment was repeated using a SUMO protease inhibitor, N-ethylmaleimide (NEM), and a band consistent with the size of SUMO-modified cSHMT was identified in the nuclear fraction. C, the cSHMT protein immunoprecipitates with an anti-SUMO antibody. MCF-7 cell extracts were treated with anti-HA, anti-cSHMT, or anti-SUMO antibodies, which were subsequently precipitated with protein A/G-conjugated agarose beads. The precipitate was washed and then incubated with 1ϫ SDS-PAGE sample buffer to elute bound proteins. Immunoblots were prepared and probed for cSHMT protein. Lane 1, MCF-7 cell extract; lane 2, anti-HAprecipitated extract; lane 3, anti-cSHMT-precipitated extract. Note an immunoreactive band ϳ12-14 kDa larger than cSHMT that may be ubiquitinated cSHMT. Lane 4, anti-SUMO-precipitated extract. Note an immunoreactive band ϳ20 kDa larger than cSHMT that is SUMOylated cSHMT. the cell cycle. These observations are consistent with the known cell cycle dependence of thymidylate biosynthesis (37)(38)(39).
cSHMT Nuclear Localization Is Ran-dependent-The role of Ran in cSHMT localization was investigated by transfecting Ran (wild-type) or RanT24N (dominant negative Ran, Ran-GDP) expression vectors into MCF-7 cells (40) (Fig. 6). Nuclear import of many proteins is dependent upon the Ran cycle that occurs in association with the importins, which traverse the nuclear pore with target proteins bound (41). Ran-GDP is bound by NTF2 (nuclear transport factor 2) and shuttles into the nucleus. In the nucleus, RanGEF (RCC1) exchanges GDP for GTP, and Ran-GTP then binds importin proteins. The Ran-GTP-importin complex is transported to the cytoplasm, where RanGAP hydrolyzes GTP to GDP, resulting in importin release from Ran-GDP and recycling the importin for another round of transport.
Importin 13 has been demonstrated to be involved in Ranmediated nuclear transport of UBC9 and presumably with Ran-mediated transport of SUMO-modified proteins (42). The nuclear pore contains several proteins that are either SUMOylated or that are components of the SUMOylation machinery, including RanBP2 (Ran-binding protein 2). RanBP2 is localized to the nuclear periphery/nuclear pore and possesses E3 SUMO ligase activity (22,43). Ran expression increased the concentration of cSHMT in the nucleus compared with mock-transfected cells, indicating a role for Ran in cSHMT nuclear localization (Fig. 6). Less cSHMT was found in the nuclear fraction of RanT24N transfected cells as compared with mock-or Ran wild type-transfected cells, presumably because RanT24N is unable to cycle importins from the nucleus (44).
In Vitro SUMOylation of cSHMT-The ability of cSHMT to serve as a substrate for UBC9-mediated SUMOylation was investigated using a commercially available kit from ActiveMotif. SUMOylation requires the presence of SUMO, a SUMOactivating enzyme (referred to as E1; Uba2/Aos1), and a SUMO conjugase (referred to as E2; UBC9). The human cSHMT primary amino acid sequence contains a conserved SUMOylation consensus sequence (⌿KXE, where ⌿ is any hydrophobic amino acid (30)) at the N terminus of the protein (IK 38 KE); lysine 38 is the consensus residue targeted for SUMO modification in cSHMT (Fig. 7D).
The cSHMT protein was a substrate for UBC9-mediated SUMOylation (Fig. 7), and in vitro SUMOylation was enhanced by RanBP2, the nuclear pore-associated E3 SUMO ligase (Fig. 7,  A and B). This suggests that SUMOylation of cSHMT occurs at the nuclear pore and is linked to cSHMT nuclear import. RanBP2 (BP2⌬FG, the E3 ligase domain of RanBP2 without FG repeats) has been shown to increase SUMOylation of RanGAP1 and Sp100 in vitro (22).
The site of SUMOylation was confirmed by site directed mutagenesis. Mutation of either cSHMT Lys 38 or Lys 39 to arginine residues impaired SUMOylation in vitro (Fig. 7C). The inability of UBC9 to SUMOylate K38R cSHMT was expected, because this is the residue targeted for modification within the consensus sequence. The inability of UBC9 to SUMOylate K39R cSHMT was unexpected, but this residue is clearly important for cSHMT to serve as a substrate for SUMOylation. A common cSHMT variant, L474F (1420 C 3 T), increases risk for cardiovascular disease but is protective against adult acute lymphocytic leukemia (21,45). Surprisingly, the L474F variant was not a substrate for UBC9-mediated SUMOylation (Fig. 7C).
The cSHMT-UBC9 Interaction and the Role of Leu 474 -The possible interaction of cSHMT with UBC9 was examined by molecular modeling (Fig. 7E). UBC9, from the crystal structure of a RanGAP1-UBC9 complex (23), was manually docked with human cSHMT, such that A chain residues Lys 38 and Asn 42 of cSHMT occupied approximately the same positions as Lys 526 and Glu 528 of RanGAP1, relative to UBC9 residues 86 -93. Ran-GAP1, the Ran GTPase-activating protein, is targeted to the nuclear pore via SUMOylation (46). In the RanGAP1-UBC9 complex, RanGAP1 Lys 526 is the residue that becomes SUMOylated; it makes a hydrogen bond to the UBC9 main chain at Asp 127 and is close to Cys 93 , which is an essential catalytic residue required for the SUMOylation reaction. In the modeled cSHMT-UBC9 complex, cSHMT Lys 38 is close to UBC9 Cys 93 , and both Lys 38 and Lys 39 can form hydrogen   Reactions supplemented with RanBP2⌬FG (kindly provided by Frauke Melchior), an E3 SUMO ligase, exhibited increased SUMOylation activity. All reactions were analyzed by immunoblotting with a rabbit anti-human SUMO antibody. B, same as in A, except the blot was probed with a sheep anti-human cSHMT antibody. C, Lys 38 , Lys 39 , and Leu 474 are essential for in vitro SUMOylation. In vitro SUMOylation reactions with wild-type cSHMT and mutant cSHMTs were performed as in A. Mutation of either Lys 38 or Lys 39 to Arg, or Leu 474 to Phe impaired SUMOylation. D, cSHMT contains a consensus SUMOylaton sequence. The sequences of several cSHMT proteins were analyzed for the presence of ⌿KXE, where ⌿ is a hydrophobic amino acid and X is any amino acid. All mammalian cSHMT sequences share a conserved IKXE motif. Note that Xenopus laevis does not have the SUMO consensus motif. E, molecular modeling of the cSHMT-UBC9 interaction. UBC9 (structure shown in red) from the crystal structure of a RanGAP1-UBC9 complex (23) (Protein Data Bank entry 1KPS) was manually docked with human cSHMT (Protein Data Bank entry 1BJ4; A chain shown in blue, B chain in green). cSHMT L474 contacts the UBC9 main chain around residues 65-67. This figure was produced using Molscript (63) and Raster3D (64). F, quantitative ␤-galactosidase assays demonstrate that the SUMO consensus and the L474F mutation are involved in the cSHMT-UBC9 interaction. Quantitative yeast two-hybrid assays were performed with the cSHMT-, cSHMT-K38R-, cSHMTK38R/K39R-, L474F-, and UBC9-encoding plasmids as described under "Experimental Procedures." All values are normalized to the ␤-galactosidase activity generated by the cSHMT-UBC9 interaction. The Lys 38 3 Arg mutation decreased ␤-galactosidase activity by 20%. The Lys 38 3 Arg and Lys 39 3 Arg mutations combined decreased ␤-galactosidase activity by 45-50%. The Leu 474 3 Phe polymorphism decreased ␤-galactosidase activity by 20%. Error bars, S.E. bonds with the UBC9 main chain. Because of the close contact between cSHMT and UBC9, mutation of lysine 38 or 39 to arginine would require some adjustment at the interface, with a consequent effect on the probability of SUMOylation of cSHMT. In addition to the two lysines, several other cSHMT residues in the 27-46 range are potentially capable of making hydrophobic or hydrogen-bonding contacts with UBC9. The buried surface area in the RanGAP1-UBC9 complex is about 600 Å 2 ; in the modeled cSHMT-UBC9 complex, it is considerably larger, about 900 Å 2 . This is due to the positioning of cSHMT surface regions comprising 1) residues 462-480 from chain A and 2) residues 87-91 from chain B (no analogous residues exist in RanGAP1) close to UBC9. Hydrophobic and polar contacts involving these additional cSHMT regions are expected to influence cSHMT-UBC9 interactions. cSHMT Leu 474 , in particular, makes a close contact with the UBC9 main chain around residues 65-67 (45,47).
A quantitative ␤-galactosidase assay was performed to determine the effect of the L474F polymorphism, Lys 38 , and Lys 39 on the UBC9-cSHMT interaction and UBC9 catalysis. Plasmids containing either cSHMT, cSHMT-K38R, cSHMT-K38R/ K39R, or cSHMT-L474F cDNA were fused to the DNA binding domain of Gal4 and cotransformed with a plasmid containing the UBC9 cDNA fused to the Gal4-activating domain into Y187 yeast. ␤-Galactosidase activity reflects the strength of the interaction of human cSHMT protein with human UBC9 protein (Fig. 7F). The results indicate that Leu 474 , Lys 38 , and Lys 39 contribute to the cSHMT-UBC9 interaction and that the L474F polymorphism weakened the interaction by 20% (Fig. 7F). The data indicate that the UBC9-L474F cSHMT interaction is probably distorted, since the UBC9-L474F cSHMT interaction is not catalytically competent for SUMO modification in vitro.

DISCUSSION
In 1976, Shin et al. (48) determined that 10% of total liver folate was present in the nuclear compartment, suggesting that folate metabolism may occur in the nucleus. The concept of nuclear folate metabolism was first described by Prem veer Reddy and Pardee (49), who proposed the existence of a nuclear "replitase," which is a putative multienzyme complex that synthesizes nucleotides de novo at the replication fork during S phase. DNA polymerase ␣, ribonucleotide reductase, thymidine kinase, and NDP kinase were initially identified as replitase-associated proteins (49). Later, Pardee and co-workers (50,51) identified dihydrofolate reductase (DHFR) and TS activity in purified replitase complexes. Others have found that the enzyme TS localizes to the nuclear periphery but not the nucleus in Saccharomyces cerevisiae, whereas it localizes to the nucleus of colorectal cancer cell lines (52,53). In support of these observations, S. cerevisiae TS does not contain a conserved SUMO consensus site (supplemental Fig. 1). In this study, we demonstrated the SUMO-dependent nuclear localization of cSHMT during S phase and confirmed the nuclear localization of TS in human cell lines. Furthermore, the three mammalian proteins that comprise the de novo thymidylate synthesis pathway (TS, cSHMT, and DHFR) were all found to contain a consensus SUMOylation motif (supplemental data). Therefore, the entire folate-dependent thymidylate biosynthesis pathway may undergo SUMO-dependent nuclear localization during DNA replication (Fig. 1, inset). Interestingly, unlike DHFR and TS, cSHMT exhibits a narrow range of tissue-specific expression (54), indicating that nuclear folate metabolism is limited to cells expressing cSHMT.
Compartmentation of TS and cSHMT in the nucleus provides a mechanism to account for previous isotope tracer experiments that demonstrate the preferential enrichment of cSHMT-derived folate-activated single carbons into thymidylate relative to methionine (17). Furthermore, increased expression of cSHMT in MCF-7 cells was shown to enhance the efficiency of de novo thymidylate biosynthesis relative to the salvage pathway, indicating that cSHMT expression is limiting for de novo thymidylate biosynthesis. Because methionine biosynthesis occurs in the cytoplasm, SUMO-dependent nuclear localization of cSHMT may enable preferential enrichment of cSHMT-derived methylenetetrahydrofolate into thymidylate and also increase the efficiency of that pathway, as has been demonstrated in MCF-7 cells (17). Increased uracil content in DNA resulting from depressed de novo thymidylate biosynthesis during DNA replication and repair is associated with impairments in folate metabolism and dietary folate deficiency; this biomarker is likely to be affected by the efficiency of cSHMT SUMOylation and nuclear import.
The S phase dependence of cSHMT SUMOylation and nuclear localization, which persists into the G 2 /M phase of the cell cycle, indicates that nuclear de novo thymidylate biosynthesis can function in both DNA synthesis and repair (55). SUMO modification has been detected in many proteins involved in DNA replication and repair, including PCNA, topoisomerase (56), thymine DNA glycosylase (57), XPC (58), and XRRC4 (59). PCNA SUMOylation occurs in S phase of the cell cycle and serves as a signal for normal DNA replication (26). Ubiquitination of PCNA by UBC13 occurs in G 2 /M and coincides with DNA repair (60). The interaction of cSHMT with UBC13 indicates that cSHMT is a target for ubiquitination as well, although the biological significance of this modification remains to be determined.
The common cSHMT single nucleotide polymorphism, L474F, increases the risk for cardiovascular disease, but only in individuals who carry the MTHFR A222V variant (45). The polymorphism weakens the interaction between UBC9 and cSHMT, and recombinant L474F cSHMT cannot be SUMOylated in vitro. These results suggest that the L474F cSHMT-UBC9 interaction does not permit catalytic transfer of the SUMO polypeptide. UBC9 (the SUMO conjugase) and UBC13 (a ubiquitin conjugase) compete for the modification of target proteins (with either SUMO or ubiquitin); both enzymes modify the same conserved lysine residue in PCNA (61), and SUMOylation has been suggested to impair ubiquitin-mediated protein degradation (62). It is possible that the human L474F polymorphism in cSHMT also impairs UBC13-mediated ubiquitination, as occurs with UBC9-mediated SUMOylation. Decreased ubiquitin modification of cSHMT may decrease rates of cSHMT turnover and increase cSHMT accumulation in the cytoplasm. Because cSHMT accumulation in the cytoplasm impairs the homocysteine remethylation pathway (17), the L474F variant would be expected to exacerbate the metabolic disruption associated with the MTHFR A222V variant, which also impairs the homocysteine remethylation pathway (45). This would provide a reasonable mechanism to explain the synergistic gene-gene interaction between the cSHMT L474F and the MTHFR A222V variants as observed in a recent epidemiological study of cardiovascular disease risk (45).