Structural and Functional Analysis of a Novel Interaction Motif within UFM1-activating Enzyme 5 (UBA5) Required for Binding to Ubiquitin-like Proteins and Ufmylation*

The covalent conjugation of ubiquitin-fold modifier 1 (UFM1) to proteins generates a signal that regulates transcription, response to cell stress, and differentiation. Ufmylation is initiated by ubiquitin-like modifier activating enzyme 5 (UBA5), which activates and transfers UFM1 to ubiquitin-fold modifier-conjugating enzyme 1 (UFC1). The details of the interaction between UFM1 and UBA5 required for UFM1 activation and its downstream transfer are however unclear. In this study, we described and characterized a combined linear LC3-interacting region/UFM1-interacting motif (LIR/UFIM) within the C terminus of UBA5. This single motif ensures that UBA5 binds both UFM1 and light chain 3/γ-aminobutyric acid receptor-associated proteins (LC3/GABARAP), two ubiquitin (Ub)-like proteins. We demonstrated that LIR/UFIM is required for the full biological activity of UBA5 and for the effective transfer of UFM1 onto UFC1 and a downstream protein substrate both in vitro and in cells. Taken together, our study provides important structural and functional insights into the interaction between UBA5 and Ub-like modifiers, improving the understanding of the biology of the ufmylation pathway.

The covalent conjugation of ubiquitin-fold modifier 1 (UFM1) to proteins generates a signal that regulates transcription, response to cell stress, and differentiation. Ufmylation is initiated by ubiquitin-like modifier activating enzyme 5 (UBA5), which activates and transfers UFM1 to ubiquitin-fold modifierconjugating enzyme 1 (UFC1). The details of the interaction between UFM1 and UBA5 required for UFM1 activation and its downstream transfer are however unclear. In this study, we described and characterized a combined linear LC3-interacting region/UFM1-interacting motif (LIR/UFIM) within the C terminus of UBA5. This single motif ensures that UBA5 binds both UFM1 and light chain 3/␥-aminobutyric acid receptor-associated proteins (LC3/GABARAP), two ubiquitin (Ub)-like proteins. We demonstrated that LIR/UFIM is required for the full biological activity of UBA5 and for the effective transfer of UFM1 onto UFC1 and a downstream protein substrate both in vitro and in cells. Taken together, our study provides important structural and functional insights into the interaction between UBA5 and Ub-like modifiers, improving the understanding of the biology of the ufmylation pathway. UFM1 6 is a ubiquitin (Ub)-like protein (UBL) that shares structural, but little sequence, similarity with Ub and, like Ub, can be conjugated to lysine residues of itself or target proteins (1)(2)(3). UFM1 is activated by the E1 enzyme UBA5, which forms a high-energy thioester bond between its catalytic cysteine and the exposed C-terminal glycine of UFM1. Activated UFM1 is transferred to the catalytic cysteine of the E2 enzyme UFC1. Finally, a specific E3 ligase, UFL1 (UFM1 ligase 1, also known as RCAD), transfers UFM1 onto protein substrates (4 -7). The two UFM1-specific proteases, UFSP1 and UFSP2, release UFM1 from its conjugated substrates and are additionally responsible for the post-translational processing of UFM1 to expose its C-terminal glycine (2).
Study of ufmylation (covalent conjugation of proteins with mono-or polyUFM1) is still in its infancy, but new details are beginning to emerge. For example, by covalently modifying the transcriptional co-activator ASC1, polyUFM1 chains enhance the association of ASC1 with p300 and SRC1 (steroid receptor coactivator 1), leading to the transcriptional activation of estrogen receptor ␣ target genes and improved anchorage-independent growth of tumor cells (3). The role of UFBP1 (UFM1binding protein 1 containing a PCI domain, also known as C20orf116 and DDRGK1) modification by UFM1 is less clear. However, it localizes to the endoplasmic reticulum (ER) and is involved in ER stress response (4,7,8). Interestingly, conditions that induce ER stress lead to increased expression of components of the UFM1 conjugation system, whereas their downregulation exacerbates ER stress and sensitizes cells to apoptosis (5)(6)(7)(8)(9). Of note, the ufmylation pathway is indispensable for erythro-and megakaryopoiesis, with Uba5 Ϫ/Ϫ , Ufl1 Ϫ/Ϫ , and Ufbp1 Ϫ/Ϫ mice dying in utero because of severe anemia connected to the reduced numbers of functionally differentiated erythrocytes (5,8,10).
Ub and UBLs interact with specialized domains or short linear motifs present in the versatile group of proteins that effectively act as Ub and UBL receptors (11,12). Through noncovalent interactions, receptor molecules are recruited to the Ub/ UBL-conjugated proteins, regulating their localization and/or stability or mediating the assembly of molecular platforms responsible for downstream signal propagation. For example, the ubiquitin-associated (UBA) domain of the autophagic receptor p62/SQSTM1 recognizes ubiquitylated misfolded proteins and is required for their packing into inclusion bodies and subsequent lysosomal degradation (13,14). In addition to its interaction with polyUb chains, p62/SQSTM1 interacts directly with the UBL proteins LC3/GABARAPs, which are conjugated to phosphatidylethanolamine enriched in autophagic membranes. Membrane-conjugated LC3/GABARAPs mediate lipid bilayer tethering and hemifusion (15,16), drive expansion of autophagosomes, and via autophagic receptors, target autophagy cargo to the endolysosomal compartment (17). p62/SQSTM1 binds LC3/GABARAPs via a short linear sequence, designated the LC3-interacting region (LIR) and broadly defined by the core sequence (W/F/Y)XX(L/I/V), where X is any amino acid (aa) (13,18). LIRs are increasingly found in proteins involved in signal transduction, such as ULK1-ATG13-FIP200 and KBTBD6/KBTBD7, as well as proteins in which turnover is regulated by autophagy (18 -20).
Although UBA5 was originally identified as GABARAPL2/ GATE-16-interacting protein (1), the molecular mechanisms behind this interaction with UBLs are not known. In the current study, we have identified a short linear motif in the C terminus of UBA5 that drives the interaction with either UFM1 or LC3/ GABARAP proteins, defining this as a LIR/UFIM (for LC3interacting region/UFM1-interacting motif). We present structural details for its interaction with UFM1 and model its interaction with GABARAPL2 and LC3B. By combining biophysical, biochemical and cellular techniques, we have provided a detailed characterization of the new UBL-binding motif and generated evidence for its role in the ability of UBA5 to mediate UFM1 conjugation in vitro and in cells. Our data suggest that a single UBL-binding element within UBA5 is responsible for both its function as an E1 enzyme in the ufmylation pathway and the interaction with LC3/GABARAP proteins.

Experimental Procedures
DNA Construction and Site-directed Mutagenesis-Plasmid generation was performed using either standard cloning procedures or in a Gateway cloning system (Thermo Fisher Scientific). Site-directed mutagenesis was performed using PfuUltra II fusion HS DNA polymerase (Agilent Technologies) according to the manufacturer's instructions. A comprehensive list of DNA constructs used in this study is found in Table 1.
Cell Culture and Transfection-HEK293 and U2OS cells were maintained in Dulbecco's modified Eagle's medium (Gibco) supplemented with 10% fetal bovine serum (Biochrome). Transient transfection of HEK293 and U2OS cells was performed for 24 h using X-tremeGENE 9 DNA transfec-tion reagent (Roche) according to the manufacturer's instructions. UBA5 knock-out HEK293 cells were generated by a CRISPR/Cas9 technology. UBA5 guide RNA was designed using CRISPR Design website, and it was subcloned into pX330-U6-chimeric_BB-CBh-hSpCas9 (Addgene 42230), a human codon-optimized SpCas9 and chimeric guide RNA expression plasmid. HEK293 cells were transfected with the vector together with pEGFP vector (pEGFP-C1; Clontech 6084-1) and cultured for 2 days. Thereafter, the EGFP-positive cells were sorted and expanded. Loss of UBA5 was confirmed by heteroduplex mobility assay followed by immunoblot analysis with anti-UBA5 antibody.
Antibodies, Chemicals, and Peptides-The antibodies for Western blot and immunofluorescence staining used in this study are listed in Table 2. A stock solution of bafilomycin A1 (Calbiochem) was prepared in dimethyl sulfoxide (Sigma-Aldrich). The other reagents used are indicated in the text where necessary. Peptides used in this study were obtained from commercial sources and were stored on PVDF membranes or reconstituted in PBS or dimethyl sulfoxide. A list of peptides used in this study is found in Table 3.
Preparation of Peptides and Proteins for Isothermal Titration Calorimetry (ITC), NMR, and X-ray Studies-Human LC3/ GABARAPs cloned under a modified Ub tag (21) were expressed and purified as described previously (20,22,23). UFM1 was cloned under a cleavable Ub19 leader (21) and expressed and purified in a similar manner. Short artificial tetrapeptides (GAMG-and GAME-) remained N-terminally attached to each purified protein after tobacco etch virus (TEV) cleavage. All proteins were equilibrated against a buffer (50 mM sodium phosphate, 100 mM NaCl, pH 7.0) prior to NMR and ITC experiments. For ITC and NMR titration experiments, hUBA5 LIR/UFIM peptide (EIIHEDNEWGIELVSEVSE; obtained from GenScript) was dissolved in the corresponding buffer.
Nuclear Magnetic Resonance-All NMR experiments were performed at 25°C on Bruker Avance spectrometers operating at proton Larmor frequencies of 500, 700, 800, and 900 MHz and were analyzed using Sparky 3.114 software (University of California, San Francisco X-ray Crystallography-Purified UFM1 was dialyzed into a crystallization buffer (50 mM Tris, pH 8.0, 100 mM NaCl). The synthetic hUBA5 LIR/UFIM peptide was dissolved directly in the same buffer. Crystals of UFM1 in complex with the hUBA5 LIR/UFIM peptide were obtained using 1.6 M sodium phosphate and 0.1 M HEPES, pH 7.5, as a reservoir solution by the sitting-drop vapor diffusion method at 293 K. Diffraction data were collected at the Swiss Light Source, beamline PXIII, and processed with XDS (27). The crystal structure of the complex was determined by molecular replacement using the UFM1 NMR structure (PDB ID: 1J0G) as a search model. Manual model building and refinement were done with Coot, the CCP4 software suite, and Phenix (28 -30). The final statistics of refined models are shown in Table 4.
Preparation of Proteins for Interaction Assays-GST fusion proteins were expressed from pGEX-4T1 vectors in BL21-competent Escherichia coli (Thermo Fisher Scientific) in LB medium (VWR) at 18°C overnight with 1 mM isopropyl 1-thio-␤-D-galactopyranoside (EMD Millipore). Bacteria were harvested, and pellets were frozen at Ϫ80°C for 24 h. Bacterial pellets were lysed after resuspension in a lysis buffer (50 mM Tris, 100 mM NaCl, 5% glycerol, 0.2 mg/ml chicken egg lysozyme, 1 mM PMSF, and 1 g/ml DNase) by sonication (TT13 Sonotrode, 25% amplitude, for 12 min with a 0.5/0.5-s pulse). Lysates were centrifuged at 15,000 ϫ g at 4°C for 30 min. The resulting supernatants were incubated with glutathione beads (GE Healthcare, 1 ml of 50% slurry/1 liter of E. coli culture) with rotation for 1 h at 4°C. The beads were washed three times with lysis buffer. Bound GST fusion proteins were eluted twice with 10 mM glutathione, each time for 20 min, at 4°C. Elution buffer was then exchanged to 50 mM Tris, 100 mM NaCl by dialysis, and proteins were concentrated using Centricon concentrators (GE Healthcare). For the competitive GST pulldown assay, UFM1 and GABARAPL2 were expressed from pET39-Ub19 vectors in T7 Express-competent E. coli (New England Biolabs) in LB medium at 18°C overnight after induction with 1 mM isopropyl 1-thio-␤-D-galactopyranoside. The bacterial culture was processed as described above. Purified proteins were dialyzed overnight, and Ub tags were removed by tobacco etch virus cleavage. The cleaved tags were removed by binding to Talon beads (GE Healthcare) at 4°C with rotation for 1 h. The supernatant was concentrated using Centricon concentrators.
GST Overlay Assay Using Peptide Arrays-The peptide array scan of full-length UBA5 was performed exactly as described previously (19). PVDF membranes containing arrays of spotted synthetic peptides (5 nM, made by JPT) were activated for 10 s in 100% methanol and then rinsed under running tap water followed by blocking in 5% skim milk in TBS-T (50 mM Tris, 150 mM NaCl, 0.2% Tween, pH 7.6) for 1 h at room temperature with shaking. Purified GST fusion proteins were diluted in TBS-T to a final concentration of 1 g/ml. Membranes were incubated for 2 h at room temperature with shaking. After four washes with TBS-T, membranes were incubated overnight at 4°C with anti-GST antibody diluted (1:5000) in the blocking buffer. Membranes were washed three times for 10 min each in TBS-T followed by two washes with PBS. Antibodies bound to GST fusion protein-peptide complexes were detected using ECL detection reagent (GE-Healthcare) and a Bio-Rad Imager.
GST Pulldown Assay-Purified GST fusion proteins were incubated with glutathione-Sepharose 4B (GE Healthcare) for 1 h at 4°C with rotation. The beads were washed three times in wash buffer (50 mM Tris, 100 mM NaCl, 5% glycerol). HEK293 cells were transiently transfected for 24 h with X-tremeGENE 9 transfection reagent (Roche) according to the manufacturer's instructions. Cells were lysed in a lysis buffer (150 mM NaCl, 5 mM EDTA, 1% Nonidet P40, 0.5% Na-deoxycholate, 50 mM Tris, pH 7.8, 0.1% SDS) and centrifuged at 20,000 ϫ g, and the supernatant was transferred to fresh Eppendorf tubes. The protein concentration was determined using a Bradford assay (Bio-Rad). 200 g of total cell lysate was incubated with beads for 1 h at 4°C with rotation. The beads were washed three times in wash buffer, and proteins were eluted with the 1ϫ LDS loading dye (Thermo Fisher Scientific) containing 5 mM ␤-mercaptoethanol (Bio-Rad). For the in vitro competition assay, 2 M GST-UBA5 immobilized on glutathione beads was incubated with 100 M UFM1 for 1 h at 4°C with rotation to achieve saturation. Samples were washed three times in wash buffer and incubated with varying amounts of GABARAPL2 (0.01 nM-100 M) for 1 h at 4°C with rotation. After three washings, immunoblot analysis was performed.
Co-immunoprecipitation Assay-HA/GFP fusion proteins were transiently expressed in HEK239 cells. Cell lysates were prepared after 24 h of expression using a modified radioimmune precipitation assay buffer (50 mM HEPES, pH 7.8, 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 10% glycerol, 1% Triton X-100, 25 mM NaF, 10 M ZnCl 2 ). The protein concentration was determined by the Bradford assay, and 200 g of whole lysate was incubated with 15 l of anti-HA affinity matrix (Roche 11815016001)/15 l of GFP-Trap-agarose beads (Chromotech) for 1 h with rotation at 4°C. Beads were washed three times and then loaded on SDS-PAGE for analysis.
Immunoblot Analysis-Protein samples were resolved on 4 -12% BisTris gels (Thermo Fisher Scientific) with MES running buffer (50 mM MES, 50 mM Tris, 0.1% SDS, 1 mM EDTA, pH 7.3) for 40 min at 200 V. Proteins were transferred onto PVDF membranes (Bio-Rad) by wet electroblotting at 55 mA for 90 min. Membranes were blocked for 1 h at room temperature with 5% BSA in TBS-T buffer. Primary antibodies were diluted in 5% BSA in TBS-T buffer, and membranes were incu-  bated overnight at 4°C with rotation. After three washes with TBS-T (50 mM Tris, 150 mM NaCl, 0.2% Tween, pH 7.6), the membranes were incubated with secondary antibodies for 1 h at room temperature with shaking. Image acquisition was performed using an ECL detection reagent (GE Healthcare) and a Bio-Rad Chemidoc imaging system. In Vitro Thioester Formation Assay-GST-UFM1⌬C2 (Cterminal glycine exposed form), GST-UBA5, GST-UBA5(W341A), GST-UBA5(G342A), GST-UBA5 (W341A/L345A), GST-UBA5 (⌬LIR), GST-UBA5(1-330 aa), GST-UFC1, and GST-GABARAPL2 were expressed in BL21(DE3) E. coli and purified using glutathione-Sepharose 4B (GE Healthcare). Recombinant proteins were eluted by cleaving GST with PreScission protease (GE Healthcare) on the column and dialyzed against 50 mM BisTris, pH 6.5, 100 mM NaCl, 10 mM MgCl 2 , and 0.1 mM DTT (reaction buffer). The UFM1-UBA5 thioester formation assay was carried out in the reaction buffer containing 1.0 -2.5 g of UBA5 or UBA5 mutants, 0.5-1.0 g of UFM1⌬C2, and 0.1 mM ATP. The UFM1-UFC1 thioester formation assay was assayed in the above reaction mixture supplemented with 1.0 -3.0 g of UFC1. The mixtures were incubated at 25°C for 5 min (or other times as described in the figure legends), and then nonreducing NuPAGE LDS sample buffer (Thermo Fisher Scientific) was added to stop the reaction. The samples were subjected to NuPAGE followed by Coomassie Brilliant Blue staining. The signal intensities of the bands were quantified using NIH Image.
Statistical Analysis-Statistical analyses were performed using the unpaired t test (Welch test), and p values less than 0.05 were considered significant.

UBA5 Interacts with UFM1 and LC3/GABARAP Proteins via an Evolutionary Conserved Combined LIR/UFIM Motif-We
and others have reported previously that the UFM1-activating enzyme UBA5 interacts with GABARAPL2, also known as GATE-16 (1,31), and with UFM1 (1, 32). However, it remained unclear which regions of UBA5 interacted with both UBLs. First, we performed multiple sequence alignment to determine the regions of UBA5 that are most highly conserved. Notably, three regions of high conservation were observed: the catalytic adenylation domain (found in the N-terminal part of the protein) and two poorly characterized regions (found at the C terminus of UBA5) (Fig. 1A). Although the C-terminal part of UBA5 was implicated in binding to UFM1 and UFC1 (32,33), we noticed that the 14-aa-long sequence IHEDNEWGIELVSE (aa 335-348 in human UBA5) contains a stretch of amino acids remotely resembling the LIR consensus sequence (Fig. 1B). To elucidate this further in an unbiased approach, we turned to a peptide array analysis (19). We performed a peptide walk covering the complete sequence of human UBA5 using 20-mer peptides with three-amino acid intervals. The arrays were probed with recombinant GST or GST-GABARAP, and binding was detected with anti-GST antibodies. This way, a single LIR was identified in UBA5 with the 8-aa-long sequence EWGIELVS (Fig. 1C). This stretch contains the highly conserved Trp-341, Ile-343, Leu-345, and Val-346 residues, with the important feature that the spacing between the aromatic and aliphatic residues does not conform to the canonical LIR (one intermediate residue instead of the usual two (Fig. 1B)).
Given the potential biological significance of the observed UBA5-UBL interactions, we tested whether HA-tagged UBA5 would also interact with endogenous UBLs. Although UFM1 was readily co-precipitated from HEK293 cell lysates, no signal for endogenous GABARAPL2 was identified in samples containing proteins co-immunoprecipitated with HA-UBA5 (Fig.  1G). Treatment with bafilomycin A1, which induces the accumulation of autophagosomes, did not result in an improved interaction in cells (data not shown). This is in line with the co-localization experiments in which we observed no co-localization between UBA5 and GABARAP or GABARAPL2 in U2OS cells (data not shown). On the other hand, UBA5 LIR/ UFIM mutants failed to precipitate endogenous UFM1 from HEK293 cell lysates (Fig. 1G), indicating the significance of LIR/ UFIM for UBA5-UBL interactions within the cytoplasm. Thus, UBA5 interacts with both UFM1 and LC3/GABARAPs via a unique combined LIR/UFIM; however, under basal conditions in cells, UBA5 seems to preferentially interact with UFM1 but not with LC3/GABARAPs.
Characterization of the UBA5 LIR/UFIM-UBL Interaction-As truncation and mutation of LIR/UFIM completely abolished the interaction of UBA5 with both UFM1 and LC3/ GABARAPs, we decided to characterize this sequence in detail. We again used an array of synthetic 20-mer peptides in a GST overlay assay (19) and probed for interaction with UFM1 and GABARAPL2. Surprisingly, both GST-UFM1 and GST-GABARAPL2 interacted efficiently only with the peptides containing the sequence EWGIELVS ( Fig. 2A). Notably, binding to UFM1 shows a sequence dependence very similar to that of GABARAPL2, as demonstrated by the mutational (alanine substitution) analysis of LIR/UFIM peptides. Thus, alanine substitutions of the conserved Trp-341, Ile-343, and Leu-345 fully abolished peptide interaction with UFM1 (Fig. 2B). In the case of GABARAPL2, the inhibitory effects of the same mutations are obvious, albeit less pronounced because of the higher affinity of the interaction (see ITC data below). Mutations that affect both the UFM1 and GABARAPL2 interactions include Trp-341 and Val-346, with W341A having the great-est effect on both UBLs. Additionally, Gly-342 may have a bigger role in the UBA5-UFM1 interaction but less so with the UBA5-GABARAPL2 interaction (Fig. 2B).
We then characterized the interaction between UBLs and a synthetic peptide spanning the human UBA5 LIR/UFIM sequence by biophysical methods. The ITC experiment using purified UFM1 protein and the LIR/UFIM peptide showed that the affinity of the UFM1-LIR/UFIM interaction (Fig. 2C, left plot, and Table 5) lies in the middle range (K D of 8 M). On the other hand, the canonical LIR peptide from p62/SQSTM1 (13) showed no binding to UFM1, providing a negative control and highlighting the specificity of the interaction (Fig. 2C, right  plot). NMR titration experiments confirmed the ITC data for the UFM1-LIR/UFIM interaction. We observed a typical pattern for interaction of two polypeptides with a K D in the range of 10 M (Fig. 2D). Chemical shift perturbations (CSPs) induced by titration of the nonlabeled LIR/UFIM peptide into 15   A, multiple sequence alignment of UBA5 C termini from different species (ClustalW2 algorithm). The degrees of homology (identity plus similarity) for domains are given as percentages (%) for each region. Asterisk (*), fully conserved residue; colon (:), strongly similar conservation; period (.), weakly similar conservation. The phylogenetic tree (upper right plot) was built for the indicated phylogenetic groups according to the "greatest likelihood" algorithm. UBA5 secondary structure elements were predicted by JPRED algorithms. B, alignment of LIR/UFIM sequences from human UBA5 with published LIRs from indicated human proteins. Multiple sequence alignment was performed using the ClustalW2 algorithm. C, identification of a GABARAP-interacting LIR motif in UBA5. Arrays of 20-mer peptides covering full-length UBA5 were synthesized and prepared on cellulose membranes. Each peptide was shifted three amino acids relative to the previous peptide. The arrays were probed with GST-GABARAP, and binding was detected with anti-GST antibodies. Sequences of the GABARAP-interacting peptides are shown in black, and non-interacting peptides are in gray. D, a schematic map of UBA5. Schematic domain organization and deletion/mutant constructs used in this study are shown. E, analysis of UBA5 interaction with UBLs by a GST pulldown assay. Lysates of HEK293 cells expressing GFP or the indicated GFP-UBA5 constructs were precipitated by immobilized GST or GST-UBLs (LC3/GABARAPs and UFM1). Co-precipitated proteins were detected using an anti-GFP antibody. Ponceau S staining of immunoblot membranes shows loading of GST fusion proteins. Representative results from three independent experiments are shown. WB, Western blotting. F, GST pulldown assay demonstrates interaction of wild-type UBA5 with GABARAPL2, UFM1, and UFC1. LIR/UFIM is required for binding of UBA5 to UFM1 and GABARAPL2, and the C terminus contains a UFC1-binding domain (UFC1 BD). G, co-immunoprecipitation of GFP-UBA5 and endogenous UFM1 (endogenous GABARAPL2 fails to be detected in the co-immunoprecipitated samples). The indicated forms of human HA-tagged UBA5 were transiently expressed in HEK293 cell, and immunoprecipitated (IP) by an anti-HA antibody. Co-precipitated UBLs were detected using the indicated antibodies. Representative results from three independent experiments are shown. FIGURE 2. LIR/UFIM of UBA5 specifically interacts with UFM1 and LC3/GABARAP proteins. A, demonstration of dual specificity of LIR/UFIM for UFM1 and GABARAPL2. Arrays of 20-mer peptides covering the C terminus of UBA5 were synthesized and prepared on PVDF membranes. Each peptide was shifted three amino acids relative to the previous peptide. The arrays were probed with the indicated GST fusion protein, and binding was detected with anti-GST antibodies. Sequences of UBL-interacting peptides are shown in black, and non-interacting peptides are in gray. B, confirmation of the core LIR/UFIM sequence by alanine scanning. PVDF membranes with spotted 20-mer peptides (with indicated alanine substitutions marked in red in the peptide sequences) were incubated with the indicated GST-UBLs and probed with an anti-GST antibody to reveal bound proteins. C, ITC analysis of the interaction between UBA5 LIR/UFIM and UFM1. A peptide spanning the UBA5 LIR/UFIM sequence (left panel) or p62/SQSTM1 LIR as a negative control (right panel, without correction of the dilution heat) was titrated to UFM1. The upper graphs represent the raw data; the integrated heat of each injection is displayed in the lower graphs for each titration. Thermodynamic parameters are summarized in Table 5. D, NMR data for interaction between UBA5 LIR/UFIM and UFM1. An overlay of representative areas of the 1 H-15 N HSQC spectra of 15 N-UFM1 to which the nonlabeled UBA5 LIR/UFIM peptide was added stepwise. The rainbow color code indicates increasing molar ratios upon titration from free UFM1 (red) to saturation (molar ratio 1:2 (blue)). UFM1 residues show significant CSP in the fast exchange mode (e.g. Ser-2, Phe-6, Lys-7, Lys-19, Thr-30, Lys-34, Phe-35, Ala-37, and Phe-40) (Fig. 2D and data not shown). They form a cluster at the very N terminus and at the end of ␣-helix 2 of UFM1, representing the rim of the intermolecular contacts or additional weaker site(s) of interaction. Mapping of CSPs onto the UFM1 sequence and three-dimensional structure shows a specific, well defined interaction site on the UFM1 surface spanning ␤-strand 2 and ␣-helix 1 (Fig. 2E). The biggest CSPs are localized within the area composed of ␤-strand 2 and parts of ␣-helix 1 (residues 20 -23 and 31-36), comprising a hydrophobic spot on the UFM1 surface. The UFM1 residues within long loop 1, connecting ␤-strands 1 and 2, remained mostly unassigned even after formation of the complex with the LIR/UFIM, in agreement with previous observations (34). However, a few assigned residues within loop 1 showed almost no CSPs, indicating a weak contribution of this element to the interaction.
We also analyzed interactions between the LC3/GABARAP proteins and the LIR/UFIM peptide. According to the ITC data, all GABARAP-type proteins had a significant preference for LIR/UFIM binding ( Fig. 3A and Table 5), with GABARAPL2 showing the highest affinity for LIR/UFIM (K D of 1.6 M) followed by GABARAPL1 and GABARAP (with K D of 6 and 16 M, respectively). In contrast, LC3 proteins displayed an invariably low affinity (LC3A and LC3C, K D of ϳ20 M each; LC3B, calculated K D of ϳ100 M). NMR titration of the nonlabeled LIR/UFIM peptide to 15 N-labeled GABARAPL2 (the strongest interactor) revealed significant CSPs, predominantly in a slow exchange mode, where two positions for each GABARAPL2 resonances, free and bound, are visible in the spectrum at any stage of the titration (Fig. 3B, left panel). This confirms the relatively high affinity interaction (of a low micromolar range) observed for LIR/UFIM and GABARAPL2 in ITC. Titration of the nonlabeled LIR/UFIM peptide into the 15 N-labeled LC3B protein (which binds LIR/UFIM with the lowest affinity) also induced corresponding CSPs, located especially in the LC3B ␤-strand 2 and adjacent regions (22). However, the exchange regime was fast to intermediate, indicating a much weaker interaction of LC3B with the LIR/UFIM of UBA5 (Fig. 3B, right panel).
The deviation of the primary sequence of the combined LIR/ UFIM in UBA5 from canonical LIRs (Fig. 1B) raised a question regarding the structural details of the interaction between this short linear motif with the LC3/GABARAP proteins. To answer this question, we mapped CSPs (induced by titration of the LIR/UFIM peptide in NMR titration experiments) on threedimensional structures of GABARAPL2 (Fig. 3C, upper panel) and LC3B (Fig. 3C, lower panel) and compared those with the well characterized mapping of the LIR found in the selective autophagy receptor optineurin positioned on the surface of LC3B (23) (Fig. 3C, insert). This comparison revealed that the interaction of the GABARAPL2 and LC3B proteins with UBA5 LIR/UFIM is similar to the canonical LIR-dependent interaction and involves the two hydrophobic pockets HP1 and HP2 and ␤-strand 2 of the UBL as elements of the intermolecular ␤-sheet described previously (18,35). The most perturbed resonances are localized within the corresponding area of HP1 (␣-helix 2 and ␤-strand 2) and HP2 (␣-helix 3 and ␤-strand 2). However, UBA5 LIR/UFIM induced strong CSPs in the ␣-helix 1 (residues Trp-3, Glu-7, Asp-8, and His-9) of GABARAPL2, which are weak for the corresponding residues in LC3B, signifying differences between the LC3 and GABARAP proteins (Fig. 3C).
As UFM1 and LC3/GABARAPs bind the same LIR/UFIM peptide, we assessed whether this binding was competitive. We performed an NMR titration analysis of UFM1 in complex with LIR/UFIM in the absence and presence of GABARAPL2, which has the highest affinity to LIR/UFIM. First, we saturated 15 Nlabeled UFM1 (the backbone HN resonance of Phe-35, directly involved in the intermolecular contacts, is shown in a free form on Fig. 3D, left panel, top plot; two other HN resonances, Phe-40 and Lys-34, are shown on the right panel of Fig. 3D) with the LIR/UFIM peptide (second plot from top; molar ratio 1:2) to induce corresponding CSPs in UFM1. Then, we added nonlabeled GABARAPL2 to the same sample at an equimolar ratio to the LIR/UFIM peptide (Fig. 3D, third plot from top; molar ratio 1:2:2). The resonances of UFM1 turned almost quantitatively back to the initial positions, reflecting that almost no free LIR/ UFIM peptide remained available in the sample for interactions with UFM1. A further increase in the concentration of nonlabeled GABARAPL2 led to complete stripping of the LIR/UFIM peptide of UFM1 and release of free UFM1 in the sample (Fig.  3D, fourth plot from top; molar ratio 1:2:4). None of the UFM1 HN resonances in the final sample showed positions that would be significantly altered from the positions in the reference spectra, indicating that no ternary complex was formed and that the GABARAPL2⅐LIR/UFIM complex along with the unbound UFM1 were present. We also performed a competition pulldown experiment in which immobilized GST-UBA5 was prebound with purified tag-free UFM1 and subsequently incubated with increasing quantities of purified GABARAPL2. We observed that GABARAPL2 could efficiently outcompete the UFM1-UBA5 interaction, confirming that GABARAPL2 and UFM1 compete for the same binding surface on UBA5 (Fig. 3E).
Structure of UBA5 LIR/UFIM in Complex with UFM1-Given the identification of the first UFM1-interacting motif (UFIM), we undertook a detailed structural characterization of the interaction between this short motif and UFM1. We succeeded in co-crystallizing human UFM1 with the LIR/UFIM peptide from human UBA5 and solved the structure of the resulting complex by x-ray crystallography at 2.55 Å resolution (Fig. 4, A and B, and Table 4). The crystal structure of the UFM1⅐LIR/UFIM peptide complex was determined by molecular replacement using the NMR structure of the free UFM1 (PDB ID: 1J0G) as the search model. Two UFM1 molecules and one LIR/UFIM peptide (residues 338 DNEWGILV 346 visible in the complex) are present in the asymmetric unit (Fig. 4A), with an uneven contribution of the two UFM1 molecules in the interaction with the LIR/UFIM peptide. Although the first UFM1 molecule contributes the most contacts to the peptide and provides the structural basis for specificity in recognition of LIR/UFIM (Fig. 4, B and C), the second UFM1 molecule may stabilize the resulting complex via contacts with both the LIR/ UFIM peptide and the first UFM1 molecule.
The structure of UFM1 in the complex is similar to that determined previously (34) and represents that of Ub-fold pro- APRIL 22, 2016 • VOLUME 291 • NUMBER 17

JOURNAL OF BIOLOGICAL CHEMISTRY 9033
teins. It superimposes well with Ub, SUMO, and LC3 structures with root-mean-square deviations up to 1.8 Å (Fig. 4E). Although UFM1 has a fold similar to LC3, it lacks the N-terminal two helices and consequently does not have a hydrophobic pocket corresponding to HP1 in LC3, used for interaction with the aromatic residue in a typical LIR (Fig. 4E, right plot). A key residue, Ile-343, of the LIR/UFIM occupies the hydrophobic pocket (somewhat similar to HP2 in LC3) formed by the residues of Leu-21, Val-23, Val-32, and Phe-35 of UFM1 (Fig. 4, B  and C). Additionally, Trp-341 and Val-346 of UBA5 LIR/UFIM interact with Pro-28 and Val-20 of UFM1, respectively (Fig. 4B). Trp-341 does not intercalate into the hydrophobic core of UFM1 but covers a significant part of the nonpolar surface, playing an important role in the complex stabilization. Leu-345 completes the formation of the hydrophobic cluster within the complex by contacts to the hydrophobic side chains of UFM1 (Val-20, Leu-21, and Val-23) (Fig. 4B). Hydrogen bonds between the O and HN atoms of Ser-22 in UFM1 and, correspondingly, the HN and O atoms of Glu-344 in LIR/UFIM, as well as between the O and HN atoms of Val-20 in UFM1 and Val-346 in LIR/UFIM, form an intermolecular antiparallel ␤-sheet similar to complexes of the SUMO⅐RANBP2 SIM (SUMO-interacting motif) (PDB ID: 3UIN) and GABARAP⅐K1 peptides (PDB ID: 3D32) (Fig. 4D). A salt bridge between the side chains of Glu-344 in LIR/UFIM and Lys-3 in UFM1 further stabilizes the complex (Fig. 4B, right plot). In summary, our structural studies reveal that UFM1 interacts with the short linear LIR/UFIM of UBA5 in a way similar to previously described SUMO-SIM and LC3/GABARAP-LIR interactions, with the interesting feature that the LIR/UFIM peptide binds UFM1 in an antiparallel orientation.
Formation of UFM1 Conjugates Depends on Intact LIR/UFIM within UBA5-Finally, we asked what functional significance the LIR/UFIM might have for UBA5 biology. Therefore, we tested whether the formation of UBA5-UFM1 thioester intermediates required for productive ufmylation is affected by mutating the LIR/UFIM sequence in UBA5. Indeed, when expressing wild-type UBA5 in cells, we readily observe a protein band corresponding to the size of the UBA5-UFM1 intermediate. This band runs above UBA5, is immunoreactive with anti-UFM1 antibody, and can be enriched by immunoprecipitation (Fig. 1G). Clearly, disruption of the LIR/UFIM in UBA5 led to strongly reduced formation of the UBA5-UFM1 intermediate (Fig. 1G, compare the wild type with W431A/I345A, ⌬LIR, and 1-330 aa). This result suggests that the noncovalent interaction between the LIR/UFIM of UBA5 and UFM1 may be critical for either efficient activation of UFM1 or its transfer to the catalytic cysteine of UBA5 (Fig. 1D).
To exclude any role for additional cellular proteins in this process, we made use of a previously published in vitro thioester formation assay (4) in which the production of thioester intermediates (UBA5-UFM1 and UFC1-UFM1, respectively) reflects the activity of UBA5 as a UFM1-activating (E1) enzyme. In the first reaction, involving purified UBA5 and UFM1, we observed that removal of either the LIR/UFIM or the whole C terminus resulted in reduced amounts of the UBA5-UFM1 intermediate. More subtle disruption of LIR/UFIM, by W341A, G342A, or W341A/L345A mutation, had almost no effect on UBA5-UFM1 conjugate formation in vitro (Fig. 5A, left half of the gel). Interestingly, in the second reaction, also in which purified UFC1 was included together with UBA5 and UFM1, the effect of the LIR/UFIM mutation on the formation of UBA5-UFM1 intermediates was much more pronounced, so that the double mutant W341A/L345A also had a measurable inhibitory effect on the abundance of the UBA5-UFM1 thioester (Fig. 5A, right half of the gel). W341A and G342A single mutants did not produce any measurable effect under these settings either. The formation of UFC1-UFM1 intermediates was also affected by mutation (W341A/L345A) and removal of LIR/UFIM in UBA5 (Fig. 5A, right half of the gel). The strongest effect on the UFC1-UFM1 intermediate formation by the UBA5(1-330 aa) mutant, in which the whole C terminus was deleted, could in part be explained by the presence of the UFC1binding domain (UFC1-BD) within the very C terminus of UBA5 (Ref. 33 and this study (Fig. 1, A and D)). Importantly, the observed inhibitory effect of LIR/UFIM disruption on the thioester intermediates in vitro was rather transient (reaction time of 5 min). Longer reaction times (i.e. 30 min) allowed for thioester formation resulted in an almost intact formation of UBA5-UFM1 and UFC1-UFM1 thioester intermediates even in the complete absence of LIR/UFIM (data not shown).  Fig. 2 legend). A, ITC analysis of the interactions between LIR/UFIM and LC3/GABARAPs. A peptide spanning the LIR/UFIM sequence was titrated to indicated LC3/GABARAP proteins. The upper graphs represent the raw data; the integrated heat of each injection is displayed in the lower graphs for each titration. Thermodynamic parameters are summarized in Table 5. B, NMR data for interaction between UBA5 LIR/UFIM and GABARAPL2 versus UBA5 LIR/UFIM and LC3B. An overlay of representative areas of the 15 N-1 H HSQC spectra of 15 N-GABARAPL2 (left plot) and 15 N-LC3B (right plot) to which the nonlabeled UBA5 LIR/UFIM was added stepwise. The rainbow color code indicates increasing peptide molar ratios upon titration from free GABARAPL2 and LC3B (red) to saturation (magenta, 1:2 in the case of GABARAPL2 and 1:4 for LC3B). C, three-dimensional modeling of the interaction between UBA5 LIR/UFIM and GABARAPL2 versus UBA5 LIR/UFIM and LC3B. CSPs upon interaction with the UBA5 LIR/UIFM were calculated for the GABARAPL2 (upper plot) and LC3B (lower plot) residues and compared with the well characterized LC3B-optineurin (OPTN)-LIR interaction (shown in the boxed insert). The dashed lines represent S.D. over all resonances, except the three most perturbed ones. CSP mapping of the respective structures (GABARAPL2 (PDB ID: 1EO6) and LC3B (PDB ID: 3VTU)) is shown. Residues that are not or only slightly affected (CSP Ͻ 1ϫ S.D.) or intermediately (1ϫ S.D. Ͻ CSP Ͻ 2ϫ S.D.) or strongly (CSP Ͼ 2ϫ S.D.) affected by the UBA5 LIR/UIFM peptide binding are colored in gray, yellow, and red, respectively. D, demonstration of UFM1 and GABARAPL2 competition for LIR/UFIM by an NMR competition assay. 15 N-1 H HSQC spectra of labeled 15 N-UFM1 were recorded for the unbound form (red) to which the nonlabeled LIR/UFIM peptide from human UBA5 was added to saturation (yellow, 1:2 molar ratio). To this complex, nonlabeled GABARAPL2 was added to a molar ratio of 1:2:2 (green) and 1:2:4 (blue). Finally, we tested whether conjugation of UFM1 to target proteins in the cell depended on the intact LIR/UFIM in UBA5. We made use of the cellular ufmylation assay in which FLAGtagged wild-type and LIR/UFIM mutant forms of UBA5 are co-expressed with the C-terminal glycine-exposed form of FLAG-UFM1 as well as Myc-UFL1 and Myc-UFBP1 (the latter serves as a substrate for ufmylation) in HEK293 cells in which UBA5 was deleted by CRISPR/Cas9 technology (4). UFBP1 is an E3 adaptor protein, dependent on its Lys-267 ufmylation (3); therefore, we used wild-type UFBP1 and its K267R mutant The asymmetric unit comprises two UFM1 molecules (green and cyan; schematic representation) and one UBA5 LIR/UFIM peptide (orange; backbone is shown schematically and side chains of Trp-341 and Ile-343 as sticks). Although the first UFM1 molecule (green) represents the most relevant interactions to LIR/UFIM, the second one stabilizes the complex and could be considered a crystallographic artifact (however, we cannot completely exclude such a dual UFM1 coordination near UBA5 LIR/UFIM). The LIR/UFIM of UBA5 binds to a hydrophobic pocket on the first UFM1 molecule (similar to HP2 for LC3/GABARAP proteins) mainly over the Ile-343 side chain. B, specific structural characteristics of the UBA5 LIR/UFIM⅐UFM1 complex (PDB ID: 5HKH). The left plot shows a schematic representation of the UFM1 first molecule (green) with the key residues shown as sticks (labeled in green). The UBA5 LIR/UFIM (ball-stick presentation; carbon, oxygen, and nitrogen atoms shown in orange, red, and blue, respectively) binds to a hydrophobic pocket similar to HP2 for LC3/GABARAP proteins. The right plot shows a two-dimensional LigPlot diagram with the intermolecular interface between the first UFM1 molecule and the UBA5 LIR/UFIM in a complex. Hydrophobic interactions are represented by red semicircles and hydrogen bonds by green dashed lines. C, comparison of the UFM1⅐UBA5-LIR/UFIM complex with those of Ub⅐UIM, SUMO⅐SBM, and LC3⅐LIR. Surface and schematic diagrams of UBLs are in the same orientation, with ␤2 and ␣1/␣3 at the front. Interacting peptides (orange) are shown for the UBA5 LIR/UFIM⅐UFM1 complex (PDB ID: 5HKH; first UFM1 molecule is colored green), the human Ub in complex with the VPS27 UIM (PDB ID: 2KDI), the human SUMO-1 in complex with the SUMO-binding motif (SBM) from PIASX (PDB ID: 2ASQ), and human LC3B in complex with the p62 LIR peptide (PDB ID: 2K6Q; all light blue). D, schematic representation of the UBA5 LIR/UFIM orientation in complex with UFM1. Typical for the UBL, ␤-strand ␤2 and ␣-helix ␣1 (␣3 in LC3/GABARAP proteins) are shown in the presence of UBA5 LIR/UFIM peptide (backbones are shown as sticks; reverse rainbow color code: blue for the N terminus and red for the C terminus). The arrows indicate the direction of the peptide (from N to C terminus). Examples of parallel and antiparallel orientations to ␤2 peptide in known SUMO and LC3/GABARAP proteins are listed. E, superimposition of the solved crystal structure of UFM1 (green) over those of Ub (left plot, blue), SUMO1 (middle plot), and LC3B (right plot). The structures are represented schematically, root-mean-square deviation values for positions of the C␣ atoms in each pair are given in Å, and the PDB ID code for each structure is indicated.
form as a negative control in this experiment. Upon overexpression of wild-type FLAG-UBA5, or the W341A and G342A single mutants, we observed increased formation of ufmylated UFBP1 species (UFBP1-UFM1, Fig. 5B) and the uncharacterized UFM1-protein conjugates (X-UFM1), for which ufmylation in turn depends on ufmylated UFBP1 (Fig. 5B). Indeed, UFBP1-UFM1 conjugate formation was abolished in samples expressing UFBP1(K267R) mutants (Fig. 5B). In contrast, similar expression levels of LIR/UFIM mutants (W431A/L345A and ⌬LIR/UFIM) as well as C-terminal truncation (1-330 aa) of UBA5 resulted in strongly reduced amounts of UFBP1-UFM1 and X-UFM1 conjugates. Equally, the levels of UFC1-UFM1 intermediates were strongly reduced in the samples with W431A/L345A, ⌬LIR/UFIM, and 1-330-aa UBA5 mutant expression (UFC1-UFM1, Fig. 5B). Intriguingly, despite the aforementioned clear effect on UFM1 conjugates and intermediates, there was no visible effect of LIR/UFIM mutation/deletion on the formation of UBA5-UFM1 intermediates in this assay, likely due to the strong overexpression of both UBA5 and UFM1 (UBA5-UFM1, Fig. 5B). We thus concluded that the intact LIR/UFIM is the prerequisite for the full biological activity of UBA5 toward activation and transfer of UFM1 to downstream substrates both in vitro and in cells.

Discussion
In this study, we have provided evidence for the presence of a short linear motif in the C terminus of UBA5 that we term LIR/UFIM. It is responsible for the dual binding of UBA5 to distant types of UBLs: UFM1, the cognate UBL for this E1 enzyme; and LC3/GABARAPs, UBLs associated with vesicular transport and autophagy (and so far not shown to be involved in ufmylation). We initially identified LIR/UFIM by its remote resemblance to the published LIR consensus sequence (W/F/ Y)XX(L/I/V) (36). However, unlike the canonical LIRs, the combined LIR/UFIM of human UBA5 has up to four core amino acids (EDNEWGIELVSE), shown in bold, which comprise one aromatic (Trp-341) and three aliphatic (Ile-343, Leu-345, and Val-346) residues and show an unusual spacing (one intermediate residue between the aromatic and the aliphatic amino acids instead of the two in the canonical LIR). Alanine substitutions in the peptide arrays and LIR/UFIM mutations/deletions in the binding assays unequivocally demonstrate the significance of these residues for interactions with both UFM1 and LC3/GABARAPs.
To our knowledge, this is the first report that describes a UFIM in a protein. By solving the crystal structure of the UBA5 UFM1⅐LIR/UFIM complex, we have provided an insight into the nature of the noncovalent interaction between UBA5 and UFM1 (Fig. 4B). Thus, unlike in known LIR-LC3 interactions (e.g. (37)), the tryptophan (Trp-341) of LIR/UFIM does not occupy the hydrophobic core of UFM1. Rather, it covers a part of the UFM1 surface and engages in nonpolar interaction with a proline residue, thereby stabilizing the interaction. Ile-343 of LIR/UFIM anchors the interaction by occupying the hydrophobic pocket of UFM1 (which is topologically similar to HP2 of LC3). Leu-345 and Val-346 are both engaged in hydrophobic interactions with the side chains of UFM1. Therefore, the interaction of UBA5 LIR/UFIM with UFM1 shares some features with that of LIR-LC3 but also has unique properties. One of them is also the orientation of the LIR/UFIM in the complex. Short linear motifs reported to date, such as LIR (36), SIM (38,39), and UFIM (this study), bind UBLs in a roughly similar way by establishing nonpolar interactions between the hydrophobic core of the linear sequence and hydrophobic pockets of the cognate UBL. However, these hydrophobic interactions are context-dependent and can be modulated further via the presence of polar groups, which engage in electrostatic interactions. Consequently, there are two types of LIRs and SIMs, depending on their orientation on LC3 and SUMO. One type binds UBLs in parallel to their ␤2 (37,40), whereas the other binds UBLs in an anti-parallel way (40,41). Here, we have reported the structure of an anti-parallel type of UFM1⅐LIR/UFIM complex. As this is the first such example, further studies will be required to conclude whether there are also two types of UFIM or if this is the only possible orientation of a UFIM interacting with Ufm1.
The alignment of UBA5 LIR/UFIM with other LIRs may suggest that it is more similar to those found in NDP52 and TAX1BP1, which completely lack an aromatic residue within their LIRs (Fig. 1B). Therefore, the role of the aromatic residue (trp-341) in this non-canonical LIR sequence may be to support the dual interaction with UFM1 and LC3/GABARAPs. Interestingly, LIR/UFIM preferentially binds GABARAPs, similar to the LIRs found in NBR1 (14), NIX/BNIP3L (42), KBTBD6/7 (20), ATG13, and ULK1 (19), and ALFY (43). The molecular determinants for this preference are currently not clear. Yet, the increased affinity to the GABARAP subfamily may have a biological significance. Thus, GABARAPs have a comparable affinity to LIR/UFIM as UFM1 (K D of 8.1 M for UFM1 versus 1.7-14.4 M for GABARAPs (Table 5)) and may outcompete UFM1 from its complex with UBA5 (Fig. 3, D and E). These findings suggest a potential role for LC3/GABARAP proteins in  function by in vitro UFM1-UBA5 and UFM1-UFC1 thioester formation assay. Purified UFM1⌬C2 (with exposed C-terminal glycine) and UBA5, or the indicated UBA5 mutants, were incubated with UFC1 (lanes 1-7) or without UFC1 (lanes 8 -14) at 25°C for 5 min. As a negative control, the reactions were performed in the absence of ATP (lanes 1 and 8). The samples were resolved on a nonreducing NuPAGE followed by Coomassie Brilliant Blue staining. Relative ratios of UFM1-UBA5/UBA5 and UFM1-UFC1/UFC1 were determined by the intensity of the bands. Data shown are representative of three independent experiments. Error bars represent Ϯ S.E.; p values were determined by unpaired t test (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). B, demonstration of the functional significance of LIR/UFIM for UBA5 function by the cellular ufmylation assay. HEK293 cells, in which UBA5 was deleted by CRISPR/Cas9 technology, were transfected with the indicated plasmids, and immunoprecipitation experiments (IP) with anti-FLAG M2-agarose were performed. FLAG-IP (right half of the gel) showed a remarkable reduction of the UFC1-UFM1 intermediate and the UFM1-UFBP1 and X-UFM1 conjugates (where "X" represents an uncharacterized substrate protein) in the samples transfected with UBA5 LIR/UFIM mutants or C-terminal deletion construct (1-330 aa). Relative ratios of UBA5-UFM1/UFM1, X-UFM1/UFM1, UFC1-UFM1/UFM1, and UFBP1-UFM1/UFBP1 (derived by FLAG-IP) were determined by the intensity of the bands. Data shown are representative of three independent experiments. Error bars represent Ϯ S.E.; p values were determined by unpaired t test (*, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001). WB, Western blotting.
shown). In addition, the physical segregation of UBA5 and LC3/ GABARAPs and/or masking of the binding surfaces by posttranslational modifications is likely to take place in vivo, which further complicates the analysis of the possible cross-talk between the two UBL systems. Our co-immunoprecipitation and co-localization studies failed to detect a strong association between UBA5 and GABARAPL2 ( Fig. 1G and data not shown). This striking lack of association in cells is in stark contrast to the strong and specific interaction between UBA5 and GABARAPL2 in interaction assays in vitro (Fig. 1, E and F, and Fig. 3E). It will be important to study under which conditions UBA5 and LC3/GABARAPs interact in cells and what functional consequences this binding might have for the ufmylation pathway or autophagy. The LIR/UFIM of UBA5 is required for its role as E1, i.e. a UFM1-activating enzyme (Fig. 5, A and B). Deletion of the C terminus (1-330-aa mutant) in the UBA5 inhibits the formation of UFM1 thioester intermediates and protein conjugates, likely due to the combined lack of UFC1-BD and LIR/UFIM (Fig. 1D). UFC1 interaction via the very C-terminal binding motif has been suggested previously by Xie (33) and confirmed by us in this study (Fig. 1F). However, deletion of LIR/UFIM alone (which preserves UBA5-UFC1 binding (Fig. 1F)) is sufficient to inhibit the formation of UBA5-UFM1 and UFC1-UFM1 (Fig. 5A) and UFBP1-UFM1 and X-UFM1 conjugates (Fig. 5B). The fact that the longer reaction times for the thioester formation in vitro override the dependence of UBA5 on LIR/UFIM (data not shown), and that UFM1 conjugate formation in the cell is not blocked completely when using UBA5 LIR/UFIM mutants (Fig. 5B), suggests that the catalytic (adenylation) domain of UBA5 alone (Fig. 1A) allows for sufficient, albeit transient, interaction with UFM1, providing compensation for the lack of LIR/UFIM-mediated UFM1 transfer under optimal conditions. Further studies will address the interplay between LIR/UFIM and the catalytic domain of UBA5 and elucidate the relative contribution of the LIR/UFIM to UFM1 activation and transfer along the E1-E2-E3-substrate axis (our current view on that process is summarized in Fig. 6).
In summary, our discovery and characterization of LIR/ UFIM and its role in UBA5-UFM1 and UBA5-LC3/GABARAP interactions provide an important insight into the biology of UBA5 and the ufmylation system. Furthermore, the discovery and characterization of the first UFIM will allow identification of further proteins that bind UFM1 noncovalently. By analogy to Ub signaling, in which receptor proteins harboring UBDs (ubiquitin-binding domains) drive cellular events (12,44), we predict that knowledge of the identity of the UFM1-binding proteins will expedite elucidation of the biological role of this important UBL conjugation pathway.