The Fanconi Anemia DNA Repair Pathway Is Regulated by an Interaction between Ubiquitin and the E2-like Fold Domain of FANCL*

Background: The E3 ligase FANCL monoubiquitinates FANCD2 in a critical step in the repair of DNA interstrand crosslinks. Results: FANCL binds ubiquitin non-covalently via its N-terminal E2-like fold. Conclusion: Monoubiquitination of FANCD2 is regulated via a non-covalent interaction between FANCL and ubiquitin. Significance: This interaction represents an additional layer of regulation of the Fanconi Anemia pathway, and a targetable interface.

makes no contacts with the other domains of FANCL (19). The ELF domain is found in all FANCL homologues, and is conserved across species. Therefore, we sought to establish the function of the ELF domain. We report here a previously undetected non-covalent interaction between the ELF domain of FANCL and ubiquitin. We use detailed biochemical and structural studies to characterize the interaction, and find that it is mediated via a functionally conserved patch on ELF and the hydrophobic patch of ubiquitin. Furthermore we show that the interaction is neither catalytic, nor required for complex formation, but is required for efficient FANCD2 monoubiquitination in cells.

Experimental Procedures
Protein Expression and Purification-All mutants were generated with the QuikChange site-directed mutagenesis kit (Stratagene). Drosophila melanogaster FANCL constructs and DRWD-RING (105-431) were expressed and purified as described previously (19).
Human ubiquitin was cloned into pRSF Duet-1 vector with no purification tag. Luria Broth (LB) growth medium cultures were grown at 37°C until A 600 0.5, at which point expression was induced with 500 M IPTG for 4 h. Cells were harvested and resuspended in PBS buffer before repelleting and flash freezing in liquid nitrogen. The pellets were resuspended in 1 mM EDTA, 25 mM Tris, pH 8.0 before sonication and clarification. The clarified lysate was filtered and loaded onto Q Sepharose HP (GE Healthcare) anion exchange column equilibrated 25 mM Tris, pH 8.0, 1 mM EDTA. The ubiquitin came off in the unbound fraction, before further purification with size exclusion chromatography 250 mM NaCl, 100 mM Tris, pH 8.0 and subsequent storage at Ϫ80°C.
For His-ubiquitin, human ubiquitin was cloned into the pRSF Duet-1 vector with a N-terminal Hexa-His tag. Expression was induced with 0.5 mM IPTG at A 600 0.5 followed by batch affinity purification on Ni-NTA (Qiagen) as above. Hisubiquitin was eluted with 300 mM imidazole and further purified by size exclusion chromatography. All interaction experiments are done using monoubiquitin.
Xenopus laevis STREP-FANCD2 cDNA was a kind gift from P. Knipscheer and J. Walter. We expressed and purified the protein as previously described (24). Briefly, Hi5 cells were infected with baculovirus, and the protein was purified on FLAG resin before elution with 3ϫ FLAG peptide and stored with 5% glycerol at Ϫ80°C.
Pull-down Assays-In a total reaction volume of 1 ml containing assay buffer, 500 mM NaCl, 100 mM Tris, pH 8.0, 250 M TCEP and 10 M ZnCl 2 , Drosophila FANCL, FANCL L81R, ELF, DRWD-RING, Xenopus FANCL, or FANCL N72R (all 250 nM) was added with an excess of His-ubiquitin (1 M), His-Ube2T (500 nM) or His-Ube2T-Ub (500 nM). Reactions were left to bind for 1 h on ice. 100 l of Ni-NTA-agarose (Qiagen) was equilibrated in assay buffer and added to the 1 ml reaction and left on a roller at 4°C for 1 h. Samples were washed with 10 ml of assay buffer and the agarose resuspended in 100 l of assay buffer. 50 l of 2ϫ SDS buffer was added before the samples were boiled. 10 l of the samples was loaded onto SDS page gel and subjected to Western blotting and probed with appropriate antibodies. Anti-ubiquitin antibody was purchased from DAKO. Anti-Drosophila FANCL and anti-human DRWD antibodies were raised from recombinant proteins by Pettingill Technology Ltd. The anti-His antibody was purchased from GE Healthcare.
NMR Spectroscopy and ELF Assignment-The Drosophila ELF domain and ubiquitin were isotopically labeled by overexpression in minimal medium enriched with [ 15 N]ammonium sulfate and [ 13 C]glucose. For NMR samples the ELF domain and ubiquitin were buffer exchanged into 50 mM NaCl, 20 mM sodium phosphate, pH 6.5. Susceptibility matched NMR tubes (Shigemi, US) were used at all stages. Sample volumes of 330 l were placed in Shigemi BMS-005V tubes and included 10% D 2 O v/v. The majority of NMR experiments were carried out at 600 MHz using a VarianUnity INOVA equipped with a 5 mm HCN z-pulse field gradient probe. Chemical shift referencing was based on the position of the water resonance with the exact value being related to the known relationship of the H 2 O resonance with temperature. Unless otherwise stated, all NMR experiments were solvent suppressed to reduce the water signal using WATERGATE that was typically obtained using gradient field strengths of 40 -50 G cm Ϫ1 . All NMR datasets were acquired at 25°C.
All NMR data processing utilized NMRPipe (25) and analyzed using the CCPN analysis package (26). Most resonances were successfully assigned manually without ambiguity. Backbone chemical shift assignments were achieved for the ELF domain using 1 H- 15  NMR Titrations-All titration experiments were recorded in 50 mM NaCl, 20 mM sodium phosphate, pH 6.5. The 15 N-labeled domain (wild type ELF or ubiquitin) was at 0.6 mM in the sample, while the binding partner was titrated in at different molar ratios (10:1, 7:5:1, 5:1, 2.5:1, 1:1). All 15 N-1 H HSQC were recorded for 30 min.
Charging Ube2T-Charged Ube2T was prepared by adapting the method described in Ref. 28. Human His-Ube2T C24S and tag-free ubiquitin G76C were purified to homogeneity as before, albeit retaining the His-tag on Ube2T. These were then dialyzed against 50 mM sodium borate, pH 8.0, 1 mM TCEP overnight. The proteins were mixed for 15 min on ice with His-Ube2T at 330 M and ubiquitin at 1 mM. A stock solution of 1,3 dichloroacetone (Sigma) was prepared in dimethylformamide at 20 mM. This was added to the mixed proteins at 0.8 mM and the samples were left rolling at 4°C for 1 h. The reaction was quenched with 10 mM ␤-ME for 1 h. The sample was then loaded onto an SD75 16/60 column (GE Healthcare) and purified, and the protein stored at Ϫ80°C with 10% glycerol.
Complete ubiquitination profile was analyzed using fluorescently labeled ubiquitin (Ub 800 ). Ubiquitin (residues 2-76) was expressed and purified bearing a GPLCGS overhang at the N terminus. The cysteine residue in the overhang was targeted for site-specific incorporation of a DyLight™ 800 Maleimide (Life Technologies) dye following the manufacturer's protocol. Labeled species was further subject to cation exchange chromatography and stored at Ϫ20°C as single-use aliquots. All ubiquitination reactions contained 2 M of Ub 800 and were terminated by boiling with LDS loading buffer. The samples were resolved by SDS-PAGE and analyzed by direct fluorescence monitoring using Li-COR Odyssey Infrared Imaging System. Integrated intensities of FANCD2 ubiquitination from five inde-pendent experiments were obtained using Image Studio™ (Odyssey) imaging software and plotted using GraphPad Prism.

FANCL Binds
Ubiquitin Non-covalently-The ELF domain of FANCL shares significant structural homology with E2-conjugating enzymes (19). E2s form a catalytic intermediate with ubiquitin via a thioester between the catalytic cysteine and the C terminus of ubiquitin (29). The ELF domain does not possess a catalytic cysteine. However, an additional feature of E2s is that they can also interact non-covalently with ubiquitin (30,31). Therefore we hypothesized that the ELF domain of FANCL might interact with ubiquitin in a similar manner. To test this hypothesis, we performed a pull-down binding assay using 6ϫ His-tagged ubiquitin as bait (Fig. 1A). His-ubiquitin pulls down both full-length Drosophila FANCL and the isolated ELF domain. In contrast, FANCL lacking the ELF domain (⌬ELF) is not pulled down. To further characterize the interaction between the ELF domain and ubiquitin, we measured the affinity of binding using isothermal titration calorimetry (ITC), and determined a dissociation constant of 42 Ϯ 12.0 M (Fig. 1B).
A Surface-exposed Patch on the ELF Domain Interacts with a Hydrophobic Patch on Ubiquitin-We next wanted to understand the molecular determinants of the interaction between FANCL and ubiquitin. E2s bind ubiquitin non-covalently via a backside interaction that involves residues from the loop connecting strands ␤2 and ␤3 (30) (Fig. 2A). The dissociation constant between ubiquitin and the ELF domain suggests that crystallization of the complex would prove challenging. Indeed, despite extensive efforts, we were unable to obtain high-resolu- tion diffracting crystals. Therefore, to understand the mode of ubiquitin binding by the ELF domain and whether it is similar to that seen in E2s, we set out to map the interacting surfaces using Nuclear Magnetic Resonance (NMR) spectroscopy. For both structural studies and ITC, milligram quantities of highquality protein are required. The mammalian and vertebrate homologues of FANCL are not amenable to large scale soluble expression (20,23); therefore, we used the more soluble invertebrate ELF domain from Drosophila, which shares ϳ65% sequence similarity (19% identity) with the human ELF domain (19). First, we determined the solution structure of the ELF domain. Two-dimensional 15 N-1 H HSQC NMR of the 15 N-labeled ELF domain yielded clear and resolved spectra, with excellent chemical shift dispersion, characteristic of a folded globular domain. We unambiguously assigned 76 out of 104 ELF residues using triple-resonance backbone datasets (Fig.  2B). Once we had determined the positions of each residue of the ELF domain in the spectra, we then titrated in increasing amounts of ubiquitin and recorded changes in the two-dimensional 15 N-1 H HSQC. Upon addition of ubiquitin, resonances were broadened to the extent that they were no longer visible, indicating a specific but transient interaction between the proteins (Fig. 3A). We then performed the reciprocal experiments by titrating increasing wild type ELF domain into 15 N-labeled ubiquitin, and identified the binding site on ubiquitin (Fig. 3B). The interaction surface on the ELF domain involves a surface comprising residues Leu-53, His-54, Leu-74, Leu-76, and Leu-81 (Fig. 4, A and B). The interaction surface on ubiquitin is the Leu8-Ile44-Val70 central hydrophobic patch commonly recognized by ubiquitin-binding proteins (Fig. 4, A and B) (32). These results reveal a novel interaction surface on the ELF domain. This surface is not a relic of the E2-like fold, as it does not coincide with the predicted surface upon overlaying the structures ( Figs. 2A and 4, C and D). To assess the requirement  for residues in the interaction surfaces, we sought to validate our structural insights. We mutated residues involved in the binding, and assayed the resulting proteins for interaction using ITC. The ELF domain point mutant L81R completely abolishes binding (Fig. 5A), as does the ubiquitin mutant I44A (Fig. 5B). To test whether the leucine to arginine mutation on the exposed solvent-accessible surface of the ELF domain disrupts the folding of the domain, we performed two-dimensional 15 N-1 H HSQC NMR of the 15 N-labeled L81R-ELF domain. These experiments yielded clear and resolved spectra, with excellent chemical shift dispersion, characteristic of a folded globular domain, with small changes compared with the wildtype spectra, consistent with a well-folded, stable, mutant protein (Fig. 5C).
Ub Binding Is a Conserved Function of Vertebrate and Invertebrate ELF Domain-The residues in the ELF domain that are important for ubiquitin binding are not well conserved (Fig.  6A). However, ubiquitin-interacting proteins and motifs such as CUE domains, UBAs, UIMs, and MIUs often share very little sequence homology yet retain functional homology (33). We therefore wanted to determine whether the function of ubiquitin binding is conserved in vertebrate FANCL homologs. We and others had difficulties to make wild type full-length human FANCL, therefore we turned to the Xenopus system. Indeed. Xenopus tropicalis FANCL recapitulates ubiquitin binding (Fig.  6B). Mutation of Asn-72, corresponding to Leu-81 in the Drosophila protein, results in almost a complete loss of interaction with ubiquitin, indicating that the same region in FANCL is required for interaction with ubiquitin (Fig. 6B), and that the interaction is conserved between species.
FANCL-Ubiquitin Binding Does Not Enhance the Interaction with Charged Ube2T or Aid Ube2T Discharge-Several E2-RING E3 ligase interactions are enhanced by the presence of the ubiquitin thioester bound on the E2 (34,35). Therefore another explanation is that FANCL interaction with ubiquitin enhances the recognition of ubiquitin-charged Ube2T (Ube2TϳUb). To test this hypothesis, we generated a stable Ube2TϳUb ester and assayed FANCL binding via pull-down. We observed no difference in the levels of FANCL binding in a comparison between ubiquitin-charged Ube2T and uncharged Ube2T (Fig.  7A). Furthermore, there was no difference in Ube2TϳUb binding when the ubiquitin binding surface of ELF was mutated. These data suggest that ubiquitin binding does not enhance FANCL's interaction with E2.
Although ubiquitin binding does not enhance interaction with Ube2TϳUb, it may be required for efficient discharge of the ubiquitin thioester. Since Ube2T auto-monoubiquitinates (22), and FANCL has been shown to stimulate this (22), we performed a thioester discharge assay to test whether the ELF-Ubiquitin interaction is important for this. We incubated fulllength Drosophila FANCL, FANCL L81R, or ⌬ELF with human Ube2T, and monitored levels of Ube2T autoubiquitination. Indeed, addition of FANCL, FANCL L81R, or ELF-deleted FANCL (⌬ELF), stimulated discharge of ubiquitin to the same extent (Fig. 7B). This suggests that the ELF domain is neither enhancing discharge nor monoubiquitination and is thus not catalytic.
Ubiquitin Binding Is Not Required for Efficient FANCD2 Ubiquitination in Vitro-We next asked whether FANCL's non-covalent interaction with ubiquitin is important for FANCD2 monoubiquitination in vitro. Wild type FANCL supports FANCD2 monoubiquitination in vitro (Fig. 7C). Mutation of the ubiquitin-binding patch in has no effect on the levels of FANCD2 monoubiquitination, further suggesting that the ubiquitin-binding by the ELF domain is not catalytic.
FANCL was originally predicted to adopt a WD40-propeller fold in place of the ELF-DRWD domains (12). In a previous study based on this prediction, mutation of the predicted WD40 repeats, including a large part of the ELF domain, was found to disrupt assembly of the FA core complex (16). Since the core complex is required for efficient FANCL-catalyzed FANCD2 monoubiquitination in cells, we next assessed whether the defective FANCI/FANCD2 monoubiquitination in TAP-FANCL (L7A, D78R, L79A) expressing cells was due to   D78R, L79A)). Cells were either 150 nM MMC-treated (ϩ) or mock treated (Ϫ), and lysates were subfractionated into high salt nuclear extract (NEX) and soluble chromatin extract (CHEX). Equal total protein amount of extracts were separated on SDS-PAGE gels and analyzed by immunoblotting using anti-FANCD2 and anti-TAP antibodies. Mutations in the ELF ubiquitin-binding site perturbed MMC-induced FANCD2 monoubiquitination. D2-Ub, monoubiquitinated FANCD2; D2, unmodified FANCD2. B, quantitation of the various ratios of monoubiquitinated FANCD2 and unmodified FANCD2 shown in A using ImageJ analysis software. Standard error of the mean is given from three independent experiments. C, cell lines described in A were exposed to 150 ng/ml MMC, whole cell extract prepared after indicated times and subjected to FANCD2 immunoblot analysis. D, quantitation of the various ratios of monoubiquitinated FANCD2 and unmodified FANCD2 shown in C using ImageJ analysis software. D, indicated cell lines were treated with 600 nM MMC (ϩ) or mock treated (Ϫ), fractionated as described in A, and analyzed by immunoplotting using anti-FANCI and anti-TAP. FANCI monoubiquitination was significantly reduced in ELF-mutated cells. I-Ub, monoubiquitinated FANCI; I, unmodified FANCI. F, TAP-tagged wild-type FANCL (TAP-FANCL) and ELF-mutated FANCL (TAP-FANCL [L7A, D78R, L79A]) were affinity-purified from corresponding DT40 cells with IgG-Sepharose, and incubated with either wild type HA-ubiquitin (WT) or I44A mutated HA-ubiquitin (I44A). Co-precipitation of the ubiquitin forms were analyzed by immunoplotting using anti-HA. Mutating the ELF domain or the ubiquitin I44 hydrophobic patch disrupted the TAP-FANCL ubiquitin interaction. G, TAP-FANCL and TAP-FANCL (L7A D78R L79A) high salt nuclear extracts were fractionated by Superose 6 size exclusion chromatography, and fractions were analyzed by immunoplotted using anti-TAP. Elution profiles of a 1-1.5 MDa complex were comparable between wild type FANCL and the ELF domain mutated FANCL.
disrupted ELF ubiquitin-binding rather than a destabilization of the FA core complex. We first tested TAP-FANCL ubiquitinbinding employing a cell free system. Affinity purified TAP-FANCL interacted with wild type ubiquitin, that was dependent on the ubiquitin Ile-44 patch. Mutated TAP-FANCL (L7A, D78R, L79A) however showed a significantly reduced ubiquitin binding, confirming that amino acids Leu-7, Asp-78, and Leu-79 encompass the ubiquitin interaction surface on chicken FANCL (Fig. 8F).
Next we assayed the formation of the FA core complexes in DT40 FANCL Ϫ/Ϫ cells expressing TAP-tagged FANCL or TAP-tagged ELF patch mutant TAP-FANCL (L7A, D78R, L79A). We observed no difference in the pattern of high molecular weight complex formation (Fig. 7A), indicating that ubiquitin binding is not required for the stable incorporation of FANCL into the core complex. Notably, TAP-FANCL (L7A, D78R, L79A) accumulates on damaged chromatin as efficient as wild type TAP-FANCL, further supporting our data that TAP-FANCL complex formation and integrity is independent of ELF ubiquitin binding (Fig. 8G). Taken together, these data suggest that in vertebrate cells, FANCI/FANCD2 monoubiquitination is dependent on the FANCL interaction with ubiquitin, mediated by the ELF domain.

Discussion
This study describes a hitherto unknown non-covalent interaction between FANCL and ubiquitin, with an affinity commonly observed in ubiquitin-protein interactions (36), that is required for efficient FANCD2 monoubiquitination. The interaction between the ELF domain of FANCL and ubiquitin is distinct from the surfaces in E2 proteins commonly used for non-covalent ubiquitin binding. This suggests that rather than being a relic of the E2 fold, ubiquitin binding is a function specific to FANCL. We define a function for the ELF domain, which is conserved among FANCL species (19), although dispensable in vitro for both substrate binding and catalysis of FANCD2 monoubiquitination (19 -21). Our finding that the ubiquitin-binding patch of the ELF domain is required for efficient FANCD2 monoubiquitination in cells suggest that ELF domain mutations would be harmful. However, there are as yet no FA patients identified with mutations in this domain. Intriguingly, a recent study aimed at identifying breast cancer susceptibility genes reports a significant occurrence of a splice isoform of FANCL in a cohort of non-BRCA breast cancer patients (37). The isoform results in removal of residues 72-91 of the ELF domain, which encompasses the ubiquitin-interacting surface (Figs. 4C and 6A).
Our findings suggest that FANCL may be binding an unidentified ubiquitinated protein as a requisite step in the monoubiquitination of FANCD2. We propose that a potential candidate is the clamp loader, PCNA. ICL repair is a complex and multistep process, that also requires components of the translesion synthesis (TLS) pathway (9,38,39). In common with the FA pathway, the TLS pathway is also regulated by a site-specific monoubiquitination event. PCNA is modified at Lys164 by the ring E3 ligase, Rad18 (40 -42). FANCL and Rad18 are epistatic for ICL sensitivity and repair (43), and co-depletion of FANCD2 and Rad18 does not increase cellular sensitivity to cisplatin, suggesting the proteins function in the same pathway (44). Rad18 does not monoubiquitinate FANCD2 (45). However, the E3 ligase activity of Rad18 is required for efficient loading of FANCD2 onto chromatin (44). In addition to the requirement for Rad18 activity, PCNA and FANCD2 interact in cells (46). FANCL and PCNA are also reported to interact in cells, via the central (DRWD) domain of FANCL (47). Although the monoubiquitination of PCNA appears to be critical for monoubiquitination of FANCD2, the molecular and mechanistic details of this interplay are poorly understood. We propose that the ELF domain of FANCL interacts with monoubiquitinated PCNA, and that may act as a trigger for FANCD2 monoubiquitination, as observed by Geng et al. (47). Potentially, monoubiquitinated PCNA reinforces the interaction between FANCD2 and FANCI. Alternatively, given that monoubiquitination of PCNA is required for accumulation of FANCA on chromatin (45), monoubiquitinated PCNA could aid in recruitment of the core complex to sites of DNA damage and/or activation of FANCL. In summary, our data provide insights into the regulation of the FA pathway, and provides the molecular details of a required interaction, potentially representing a druggable interface in the FA pathway.  Fig. 7. H. W. and A. F. A. conceived the study, analyzed results and wrote the paper.