The General Definition of the p97/Valosin-containing Protein (VCP)-interacting Motif (VIM) Delineates a New Family of p97 Cofactors^*

Bioinformatics

Multiple sequence alignments were calculated with the L-INS-I algorithm of the MAFFT software (20). Construction of generalized sequence profiles and profile-to-sequence comparison was performed by the pftools software (21). Generation of hidden Markov models (HMMs) from sequence alignments and HMM-to-HMM comparison were performed by the HHPRED software package (22). Disordered regions in proteins were predicted by GlobPlot (23).

Plasmids

Plasmids for the bacterial expression of N-terminal hexahistidine fusion proteins of human p97, p97ΔN, and p97ΔC (24); for the bacterial expression of N-terminal GST fusion proteins of human UBXD1, UBXD1^5×PUB, and UBXD1(1–150) (24) and human p47 (25); for the mammalian expression of N-terminally FLAG epitope-tagged UBXD1 and UBXD1^5×PUB (24); for the expression of p97 as a Gal4 activation domain fusion in yeast (24); and for the expression of Saccharomyces cerevisiae CDC48 in yeast (26) were described previously. Molecular cloning of the coding regions of human UBXD1 into pET28-His₆-SUMO1 (27), of human ZNF744 (ANKZF1) into pGBDU (28) and pCMV-Tag2B (Stratagene), of yeast VMS1 (YDR049W) with and without C-terminal 3HA epitope tag into YCplac22 (29), and of the N domain (residues 1–213) of p97 into pHLT (30) was performed according to standard procedures. Details are available from the authors upon request. Site-directed mutagenesis of ZNF744 (R655A, A661L/A662L, and R665A/L666A), UBXD1 (Q53A/M54A, A58L/A59L, and R62A/L63A), p97 (D35A/N36A/S37A/V38A, S37A/V38A, D55A/T56A, and D150A), Vms1 (R617A, A623L/A624L, and R627A), and Cdc48 (D45A/N46A/S47A/V48A, S47A/V48A, D65A/T66A, and D160A) was performed using the QuikChange II XL kit (Stratagene) according to the manufacturer's instructions and verified by sequencing of the entire coding regions.

Yeast Strains and Media

Reporter strain PJ69-4a (28) was used for yeast two-hybrid interaction assays. All other yeast experiments were performed in the DF5a strain background (31). The Δvms1::hphNT1 mutant was generated by disruption of the coding sequence using standard procedures (32). CDC48 shuffle strains were generated exactly as described (26). Yeast cells were grown in standard YPD and in synthetic complete media lacking the appropriate nutrients.

Protein Expression and Purification

Bacterial expression and affinity purification of His₆-p97 (24, 25) and of GST fusions of UBXD1 (24) and p47 (25) were performed exactly as described. His₆-SUMO1-UBXD1 was expressed in Escherichia coli BL21(DE3) pRIL (Novagen) and purified by Ni²⁺-NTA affinity chromatography using standard protocols. The His₆-SUMO1 moiety was removed by incubation with recombinant, purified His₆-SenP2(364–489) protease (27) during overnight dialysis at 4 °C against 50 mm Tris-HCl, pH 8.0, 150 mm NaCl, 1 mm DTT. After dialysis, the reaction mixture was reapplied to a Ni²⁺-NTA-agarose column, and untagged UBXD1 was recovered as flow-through. Isotope-labeled p97 N domain was prepared by bacterial expression in K-MOPS minimal medium containing ¹⁵NH₄Cl. Cells were lysed by sonication, and the hexahistidine fusion protein was purified by Ni²⁺-NTA affinity chromatography using standard protocols. The lipoyl domain fusion tag was removed by TEV protease digestion followed by a second Ni²⁺-NTA affinity chromatography. The final purification step was size exclusion chromatography using a HiLoad 26/60 Superdex 75 column (GE Healthcare). The purified, isotope-labeled p97 N domain was concentrated to 50–75 μm in 25 mm HEPES-NaOH, pH 7.5, 125 mm NaCl, 5 mm DTT, 0.01% NaN₃, 5% ²H₂O.

Binding Assays

In vitro pull-down assays using immobilized biotinylated peptide (Biotin-GGSDREKRALAAERRLAAQ-COOH; PANATecs GmbH, Tübingen, Germany) or GST fusion proteins were performed as described (26), using 10 μl of beads, 8 nmol of peptide, 0.76 nmol of GST or GST fusion proteins, and 0.2 nmol of His₆-p97. For the in vitro co-immunoprecipitation of p97 and UBXD1, 0.2 nmol of p97 were incubated overnight at 4 °C in immunoprecipitation (IP) buffer (50 mm Tris-HCl, pH 7.5, 150 mm NaCl, 2 mm MgCl₂, 0.1% Nonidet P-40, 10% glycerol) with an affinity-purified, polyclonal anti-Cdc48 antibody (26). p97-antibody complexes were immobilized by incubation with 10 μl of Protein A-Sepharose (GE Healthcare), followed by two wash steps with IP buffer and one wash step with 1× TBST (25 mm Tris-HCl pH 7.5, 137 mm NaCl, 2.6 mm KCl, 0.1% Triton X-100). UBXD1 (0.2 nmol) was incubated with immobilized p97 for 1 h at 4 °C in 1× TBST, followed by four wash steps in the same buffer and Western blot analysis of the p97-bound fraction using a mouse monoclonal antibody against UBXD1 (33). Yeast two-hybrid assays (24) and immunoprecipitation experiments using yeast (26) or HEK293T (24) cell lysates were performed exactly as described.

NMR Spectroscopy

¹H-¹⁵N HSQC spectra of the isotope-labeled p97 N domain were recorded on a Bruker DRX-600 spectrometer at 298 K (25 °C). Sample and acquisition conditions were identical to those in a previous report (34), allowing for the use of the published cross-peak assignments. Peptide binding to the p97 N domain was detected by monitoring chemical shift changes upon titration of peptide to final molar ratios (N domain/peptide) of 1:0.26, 1:0.8, 1:1.8, 1:3.6, 1:7, and 1:12. Chemical shift changes were quantified as root mean squared weighted chemical shift perturbation, ((δ¹H)² + (δ¹⁵N/5)²)½.

A General Consensus Sequence for the VIM

In order to assess the role of short sequence motifs in the specific recognition of p97, we performed a bioinformatics screen of candidate motifs. To that end, multiple-sequence alignments of all known or suspected p97 interactors and their orthologs were constructed. These alignments were visually scanned for short patches of local residue conservation, embedded within unstructured or poorly conserved regions. Next, all candidate motifs were used for the construction of short HMMs, which were then used for a comparison against each other and against a collection of precalculated HMMs derived from genome-wide sets of multiple-ortholog alignments.⁴ Several motifs were found to be present in multiple p97-binding proteins, where they often overlap with known BS1, VBM, and VIM motifs. Of note, the residues contributing most to the bioinformatically defined similarity score were not always those highlighted in the previously published consensus sequences (data not shown). The most prevalent of the bioinformatically defined motifs is characterized by two flanking arginine and two central alanine residues in an invariant spacing (RX₅AAX₂R; Fig. 1 and supplemental Fig. S1). Besides other proteins, this motif is found in the p97 cofactors gp78 and SVIP (Fig. 1), where it overlaps with the previously defined VIM (12, 17). We therefore propose the minimal consensus sequence RX₅AAX₂R as the new, general definition of the VIM.

Importantly, the revised RX₅AAX₂R consensus allowed the identification of several additional proteins as true VIM-containing cofactors (Fig. 1). These include the ERAD regulator VIMP (also known as SELS) (35); the zinc finger-containing protein of unknown function ZNF744 (also known as ANKZF1), whose yeast homolog Vms1 (Ydr049w) was recently found to be involved in mitochondrial stress response and ERAD (36, 37); the fungal SUMO-dependent isopeptidase Wss1 implicated in proteasomal degradation of SUMOylated substrates (38, 39); the uncharacterized fungal Ygl108c protein; the known p97 cofactor UBXD1 (also known as UBXN6) (24, 33, 40); and the AN1-type zinc finger protein ZFAND2b (also known as AIRAPL), a proteasomal regulator (41) that had not been linked to p97 before. Interestingly, the latter two proteins lack the otherwise highly conserved first arginine residue of the RX₅AAX₂R consensus, suggesting that it may be dispensable for p97 binding at least in certain contexts. Taken together, our bioinformatic analysis provides a general definition of the previously ill defined VIM and unites a number of hitherto unlinked proteins in a new family of known and putative p97 cofactors.

The New VIM Consensus Sequence Is Necessary and Sufficient for p97 Binding

In order to provide experimental evidence for the direct binding of the redefined VIM to p97, we performed in vitro pull-down assays using purified, recombinant p97 and a biotinylated, minimal VIM peptide. We chose a peptide derived from ZNF744 because its VIM is highly conserved from yeast to humans (Fig. 1). The immobilized VIM peptide bound full-length p97 and the truncated variant p97ΔC lacking the six C-terminal residues but not the p97ΔN variant lacking the entire N domain (Fig. 2A), showing that the new VIM is indeed sufficient for p97 binding. To analyze the importance of the most highly conserved residues within the VIM for p97 binding, we monitored binding of full-length ZNF744 to p97 in a yeast two-hybrid assay (Fig. 2B). Whereas wild-type ZNF744 interacted robustly with p97, mutating the VIM at the flanking residues Arg⁶¹⁷ and Arg⁶²⁷-Leu⁶²⁸ to alanine or at the two central residues Ala⁶²²-Ala⁶²³ to leucine completely abrogated the two-hybrid interaction with p97. Similarly, the co-immunoprecipitation of endogenous p97 with ZNF744 ectopically expressed in human HEK293T cells was reduced to background levels by the same mutations (Fig. 2C). Together, these results demonstrate that the ZNF744 residues defining the minimal VIM consensus are required for p97 binding.

A Cryptic VIM Is the Second p97 Binding Site of UBXD1

UBXD1 is an atypical UBX protein inasmuch as its UBX domain lacks key residues critical for p97 binding (24, 33). Instead, UBXD1 interacts with p97 through two independent sites, a central PUB domain binding to the C terminus of p97 and an undefined second site located within the first 150 residues, which binds to the N domain of p97 (24). Our bioinformatic analysis strongly suggested that the previously undetected minimal VIM (residues 52–63 of human UBXD1; Fig. 1) represents the unknown second p97 binding site of UBXD1. We tested this possibility in pull-down assays using a recombinant UBXD1 fragment spanning the first 150 residues, UBXD1(1–150), fused to GST (Fig. 3A). Mutations in the two central alanine residues or the C-terminal residues Arg⁶²-Leu⁶³ of the VIM completely abolished binding of UBXD1(1–150) to p97, demonstrating that these residues within the minimal VIM of UBXD1 are indeed required for p97 binding. In contrast, mutations in residues QM53,54 preceding the two critical alanine residues did not affect p97 binding, confirming that residues other than the conserved consensus residues are not required for VIM function. p97 binding assays using GST fusions of full-length UBXD1 yielded highly consistent results (Fig. 3B). Whereas mutations within the VIM alone had no effect on p97 binding of UBXD1 due to the presence of its intact PUB domain, the additional mutation of five critical residues within the PUB domain (UBXD1^5×PUB) (24, 25) resulted in the loss of p97 binding of the AA and RL but not of the QM variant of UBXD1^5×PUB. Moreover, the addition of an excess of the ZNF744-derived VIM peptide abolished p97 binding to UBXD1(1–150) and full-length UBXD1^5×PUB, both containing their wild-type VIM (data not shown), further supporting the importance of the VIM of UBXD1 for p97 binding.

In order to confirm the observed binding behavior also for untagged UBXD1, we immunoprecipitated recombinant p97 in vitro and analyzed the binding of recombinant, full-length UBXD1 to the immobilized p97 (Fig. 3C). Consistent with the GSH pull-down assays, wild-type UBXD1, UBXD1^5×PUB, and the UBXD1-RL variant mutated in the VIM showed comparable binding to full-length p97, whereas UBXD1^5×PUB-RL did not bind above background (Fig. 3C, top), indicating that the PUB domain and the VIM of UBXD1 are each sufficient for p97 binding. This conclusion was further supported by the finding that wild-type UBXD1 and UBXD1^5×PUB bound to a C-terminally truncated p97 variant unable to interact with the PUB domain (p97ΔC) in a VIM-dependent manner (Fig. 3C, center left) and that wild-type UBXD1 and UBXD1-RL bound to an N-terminally truncated p97 variant unable to interact with the VIM (p97ΔN) in a PUB domain-dependent manner (Fig. 3C, center right).

Finally, the in vivo relevance of the VIM in UBXD1 for p97 binding was investigated in immunoprecipitation experiments using HEK293T cell lysates (Fig. 3D). Again, mutation of the critical consensus VIM residues, but not of QM, resulted in loss of p97 binding for the UBXD1^5×PUB variant but not for the otherwise wild-type UBXD1. Taken together, these data demonstrate that a minimal VIM is indeed the elusive second p97 binding site of UBXD1 and that it is the only other binding site besides the PUB domain.

Mapping of the VIM Binding Site on p97

After establishing that the redefined, minimal VIM consensus sequence is necessary and sufficient for p97 binding, we sought to identify regions on the p97 N domain critical for VIM binding. To that end, we recorded ¹H-¹⁵N HSQC NMR spectra of uniformly ¹⁵N-labeled N domain (residues 1–213), an approach that had previously proven useful for the mapping of N domain interactions (34). The obtained ¹H-¹⁵N HSQC spectrum (Fig. 4A) was superimposable to the previously published spectrum of the N domain (34). Titration of the unlabeled ZNF744-derived VIM peptide to the N domain resulted in saturable, distinct shifts of several cross-peaks (e.g. see insets in Fig. 4A), whereas the majority of peaks were not at all or not significantly shifted (Fig. 4, A and B). These data are indicative of the specific binding of the VIM peptide to the isolated N domain and, in fact, show for the first time that the N domain alone is sufficient for VIM binding, in contrast to previous reports (17, 42). When residues exhibiting strongly shifted cross-peaks were mapped onto the known three-dimensional structure of the N domain, most were found to cluster in two neighboring regions (Fig. 4C). Several residues line the deep, central cleft separating the Nn and Nc subdomains (e.g. Ser³⁷, Val³⁸, Gly⁵⁴, Thr⁵⁶, and Asp¹⁵⁰), whereas others form the extended linker connecting Nn and Nc (Asp¹⁰⁷, Lys¹⁰⁹, Tyr¹¹⁰, Gly¹¹¹, and Lys¹¹²).

Next, we wanted to analyze the relevance of these regions for VIM binding. In order to identify candidate residues for site-directed mutagenesis of full-length p97, we reasoned that VIM-specific contact sites (i) should contain negatively charged and/or polar side chains given the critical importance of the flanking arginine residues for the VIM and (ii) should not extensively overlap with known binding sites of other N domain interactors to facilitate the interpretation of subsequent in vivo experiments. We therefore chose the following sets of residues for replacement with alanine. (i) Asp³⁵-Asn³⁶-Ser³⁷-Val³⁸. Ser³⁷ and Val³⁸ show the strongest chemical shift perturbation within a stretch of shifted residues (Fig. 4B) and face the cleft separating Nn and Nc. Although the neighboring residues Asp³⁵ and Asn³⁶ could not be unambiguously assigned in the NMR spectrum, they were included because they are negatively charged and polar, respectively, and face the central cleft as well. The corresponding p97 variants were termed p97^SV and p97^DNSV, respectively. (ii) We chose Gly⁵⁴-Asp⁵⁵-Thr⁵⁶. Gly⁵⁴ and Thr⁵⁶ belong to the most strongly shifted cross-peaks (Fig. 4B) and are also oriented toward the central cleft. Because substitution of glycine residues can be difficult to interpret, residue Asp⁵⁵ was chosen based on its negative charge and proximity to Thr⁵⁶. The corresponding p97 variant is p97^DT. (iii) Asp¹⁵⁰, a negatively charged residue at the bottom of the subdomain cleft, was mutated to generate p97^D150A. We disregarded the interconnecting loop Asp¹⁰⁷–Lys¹¹² and some neighboring residues, mainly because previous structural studies had identified this region as the major interaction site for other N domain binding modules, including the UBX domain, the UBX-like domain of Npl4, and the BS1 motif of Ufd1 (34, 43, 44).

The recombinant p97 mutant proteins were analyzed in a peptide pull-down assay for their ability to bind to the ZNF744-derived VIM peptide (Fig. 5A). Intriguingly, all four mutant p97 proteins exhibited strongly reduced VIM binding, indicating that the NMR chemical shift mapping indeed had identified regions of the N domain critical for the interaction with the VIM. The interaction with the VIM peptide was virtually abolished for p97^SV and p97^DNSV, whereas p97^DT and particularly p97^D150A showed some residual binding. We also analyzed the p97 mutant proteins in pull-down experiments using immobilized GST-UBXD1(1–150) with its intact VIM (Fig. 5B, top). Although p97^DT and p97^D150A showed no reduction and moderate reduction in UBXD1(1–150) binding, respectively, the interaction was strongly reduced for p97^SV and almost completely lost for p97^DNSV. In this assay, p97^DNSV appeared to be significantly more defective than p97^SV in VIM binding, suggesting that residues Asp³⁵ and/or Asn³⁶ indeed contribute to the interaction. Importantly, all four p97 mutant proteins were proficient in binding to the major N domain-interacting cofactor p47 (Fig. 5B, bottom). Taken together, these results show that the p97 mutant proteins that were engineered on the basis of the NMR chemical shift mapping are specifically impaired in VIM binding.

The VIM of Yeast Vms1 Is Important for Cellular Stress Resistance

In order to test the functional relevance of VIM-mediated cofactor binding in vivo, we chose the yeast S. cerevisiae as a model. We first confirmed that the yeast homolog of ZNF744, Vms1 (Ydr049w), interacts with the yeast p97 homolog, Cdc48, in a VIM-dependent manner (Fig. 6A) (37). Coimmunoprecipitation of Cdc48 with Vms1 was strongly reduced to background levels upon mutation of the central alanine residues 623/624 or of the C-terminal arginine residue 627, whereas mutation of the N-terminal arginine residue 617 reduced but did not abolish Cdc48 binding. Vms1 was recently reported to be involved in mitochondrial stress resistance and ERAD (36, 37). Consequently, yeast mutants lacking Vms1 exhibit a number of stress phenotypes, among them hypersensitivity against the translation inhibitor cycloheximide (Fig. 6B) (45). Importantly, the differential reduction in Cdc48 binding observed for the VIM mutants of VMS1 correlated well with their sensitivity toward cycloheximide. Whereas the vms1^R627A and vms1^AA mutants were almost as sensitive as the Δvms1 knock-out strain, vms1^R617A exhibited hardly any growth defect in the presence of cycloheximide. These results indicate that the physiological function of Vms1 in cellular stress resistance critically depends on Cdc48 binding via the VIM, in agreement with a recent report (37).

A Yeast cdc48 Mutant Defective in VIM Binding Is Stress-sensitive

We next addressed the importance of the VIM binding region mapped on p97 for cellular stress resistance. To that end, we used shuffle strains deleted for the chromosomal CDC48 locus and expressing untagged wild-type or mutant CDC48 from a centromeric plasmid at endogenous levels (26). Strains expressing the cdc48^SV, cdc48^DNSV, and cdc48^D160A mutant alleles (corresponding to the respective p97 variants described above) were viable and did not exhibit any general growth defect. Immunoprecipitation experiments showed that Cdc48^SV and Cdc48^DNSV failed to stably interact with Vms1 in vivo, whereas Cdc48^D160A showed residual binding to Vms1 (Fig. 6C), in excellent agreement with the in vitro analysis of the corresponding p97 variants. Importantly, the Cdc48^SV, Cdc48^DNSV, and Cdc48^D160A mutant proteins were unaffected in their ability to interact with the Cdc48 cofactors Ubx2 and Ufd1 in vivo (Fig. 6D), whereas Cdc48^DT showed reduced binding to both cofactors (data not shown) and was therefore excluded from further analyses. Together, these results show that the DNSV turn facing the subdomain cleft of the Cdc48 N domain is required for the binding of the yeast VIM proteins Vms1 and Ygl108c (data not shown) but not of cofactors interacting through the UBX domain or the BS1 motif. Finally, we tested the cycloheximide sensitivity of the cdc48 shuffle strains (Fig. 6E). Intriguingly, the cdc48^DNSV mutant exhibited a hypersensitivity that was comparable with that of the Δvms1 strain, whereas the cdc48^SV and cdc48^D160A mutants behaved similarly to the wild-type. Thus, binding of Vms1 to Cdc48 via the DNSV turn is critically required in vivo for the cellular stress resistance to cycloheximide.