Main

Ubiquitin (Ub) signals regulate nearly all cellular pathways in eukaryotes1. Malfunctions of Ub signals underlie numerous diseases, including cancer, neurodegeneration, and metabolic and infectious diseases2,3,4,5. Ubiquitylation usually involves the concerted action of E1, E2 and E3 enzymes, which ligate the carboxyl terminus of Ub to a lysine residue on selected protein targets6,7,8,9,10. A large set of Ub receptors decrypt the Ub signals into a cellular response11. These Ub receptors usually contain several Ub-binding domains (UBDs), a responding domain and a targeting domain. Intriguingly, Ub receptors are themselves regulated by coupled monoubiquitylation12,13,14,15,16. Despite their biological importance, researchers' efforts to fully characterize Ub cascades are impeded by (i) deubiquitylating enzymes (DUBs) that rapidly reverse the ubiquitylation signal and (ii) the multiplex connectivity of the Ub pathways, which are comprised of dozens of E2s, hundreds of E3s and thousands of targets (Supplementary Fig. 1).

Consequently, most of the Ub cascades linking specific E1s, E2s and E3s to their cognate targets are still not known. To circumvent these limitations, we developed a genetic system for the screening and characterization of Ub cascades in E. coli, which lack DUBs. We demonstrate the system's applications for identifying novel interactions and components of Ub cascades. The simplicity of E. coli as a model organism should facilitate the use of this approach for structural characterization of interfaces along the cascades and for drug discovery. Specifically, the conversion of noncovalent and transient UBD–Ub interactions into stable ubiquitylation is highly beneficial for identification and characterization of ultraweak interactions. Finally, we apply the method to identify a novel physiological ubiquitylation target downstream of an E3 ligase.

Results

Construction of a selection system for ubiquitylation

We recently established a synthetic system expressing the entire ubiquitylation cascade in E. coli and purified milligram quantities of ubiquitylated proteins in order to determine their structures17,18. Here, we further developed this system to carry out genetic screening of ubiquitylation events in E. coli. In the new system, two fragments of a split reporter gene are tethered to Ub, and a ubiquitylation target is coexpressed along with ubiquitylation apparatus in E. coli (Fig. 1a). Specifically, we fused the N-terminal fragment of murine dihydrofolate reductase (nDHFR) to the N terminus of Ub (in a plasmid denoted pND–Ub) and the C-terminal fragment of DHFR (cDHFR) to the N terminus of the substrate (in a plasmid denoted pCD–Sub). Pending substrate ubiquitylation, the two DHFR fragments were assembled into a functional enzyme conferring antibiotic (trimethoprim (TRIM)) resistance and the ability to grow on selective media19. The DHFR fragments were tethered to Ub and to the substrate with long linkers designed to confer flexible but stable characteristics that facilitated the functional assembly of the reporter (Supplementary Fig. 2). The system has a polycistronic architecture in which ubiquitylation apparatus and the substrate are coexpressed from two or three compatible vectors (Fig. 1a). Synthetic operons are expressed from a constitutive promoter (an unregulated λ-phage left promoter (pL)) suitable for bacterial genetic studies.

Figure 1: Bacterial genetics approach for selection of ubiquitylated proteins.
figure 1

(a) A genetic selection system for discovering ubiquitylated proteins in bacteria. The ubiquitylation apparatus is expressed from two compatible plasmids. Each plasmid harbors different antibiotic resistance and origin of replication, facilitating cotransformation and selection of the vectors regardless of ubiquitylation. pND–Ub denotes the N-terminal fragment of DHFR fused to Ub. pCD–Sub denotes the C-terminal fragment of DHFR fused to a ubiquitylation substrate. A complete system (consists of pND–Ub, pCD–Sub and a cognate set of ubiquitylating enzymes) confers antibiotic resistance and bacterial growth in the presence of trimethoprim (TRIM). Red cross represents bacteria that express an incomplete system and therefore are not resistant to the selective media. K, lysine residue. (b) Vps9 ubiquitylation. 2.5 μl of E. coli W3110 (at OD600 of 0.2), expressing a complete (cDHFR–Vps9, nDHFR–Ub, yeast Ubc4 and Rsp5) or incomplete (ΔE1,ΔE2; ΔUb or ΔE3) ubiquitylation apparatus, were seeded as spots on selective (10 μg/ml TRIM) or nonselective minimal agar plates. Bacterial spots were visualized by a UV video camera. (c) Same as in b, except the Vps9 cognate E3 ligase (Rsp5) was replaced by a noncognate E3 ligase (SIAH2). (d) As shown in b, but the trimethoprim concentration is varied. Binding, bacteria expressing the pND–Ub and pCD–Vps9 without a complete ubiquitylation apparatus (ΔE1,ΔE2). Ubiquitylation, bacteria expressing the pND–Ub and pCD–Vps9 along with a cognate ubiquitylation apparatus.

Full size image

To assess the functionality of the system, the well-characterized Ub receptor Vps9 (refs. 15,16,17) was fused to cDHFR (pCD–Sub) and coexpressed with Rsp5 (E3) and the pND–Ub vector that expressed E1 and yeast Ubc4 (E2). We found that the W3110 strain provided the best genetic background under the described experimental conditions. Bacteria expressing a complete ubiquitylation system for Vps9 grew under both permissive and nonpermissive conditions (Fig. 1b). However, strains lacking Ub or E1 and E2 did not grow on selective media. To demonstrate that the developed system maintains the known E3–substrate specificity, we replaced Rsp5 with SIAH2, a noncognate E3 ligase. Bacteria grew only when the cognate E3 was expressed (Fig. 1c), even though SIAH2 was functional in the selection system, as it promoted self ubiquitylation (Fig. 2).

Figure 2: Identification and characterization of NleG6-3 as E3 ligase.
figure 2

Applying the genetic selection system for identification and characterization of E3 ligases. Bacterial spots were seeded and visualized as in Figure 1. (a) Bacteria that coexpress cDHFR fusion to the E3 ligase Rsp5 (representative of the HECT family) with its cognate ubiquitylation apparatus including Ubc4 are labeled 'complete'. Deletions of E1, E2 or Ub are indicated. (b) Bacteria that coexpress cDHFR fusion to the E3 ligase SIAH2 (representative of the RING family) with its cognate ubiquitylation apparatus, including UBCH5A, are labeled 'complete'. Deletions of E1, E2, Ub or substitution of the cognate E2 with Cdc34 (a noncognate E2) are indicated. The right panel shows the effect of the small-molecule inhibitor, menadione, on SIAH2-dependent bacterial growth. (c) Sequence alignment derived from PSI-BLAST search using human CHIP as probe against the EHEC proteome. (d) Yeast E2s scan for the E3 ligase ECs3488 (NleG6-3). Self ubiquitylation of the potential E3 ligase was examined against the indicated E2s. (e) Self ubiquitylation of NleG6-3 by human UBCH5B and UBCH5C. (f) Homology-based model of NleG6-3 (orange) superimposed on the structure of EHEC NleG2-3 (blue) and Human CHIP–UBCH5B complex (gray and magenta, respectively), showing a zoomed-in view of the predicted interface. (g) Employment of the selection system for mutational analysis of the predicted E2–E3 interface. Growth of bacterial spots of the indicated mutants on the predicted E3-ligase surface coexpressed with E1 and UBCH5B are shown.

Full size image

The split DHFR system was originally developed for identification of noncovalent protein–protein interactions19, suggesting that the selection system may detect noncovalent UBDs–Ub interactions. Since UBDs present a fairly low affinity to Ub14, we tested different antibiotic concentrations to identify suitable growth conditions for Vps9–Ub noncovalent interactions. Although Vps9 presents one of the highest measured UBD–Ub affinities15,16, we observed growth only at antibiotic concentrations of 0.1 μg/ml (Fig. 1d). Since such antibiotic concentration is limited for selection, it seems that the developed system is selective only for ubiquitylation events.

Identification and characterization of E3s

Most E3 ligases undergo self ubiquitylation20. Our selection system provides a straightforward tool for their identification and characterization, as self ubiquitylation is predicted to confer antibiotic resistance. Yeast Rsp5 and human SIAH2, representatives of the HECT and the RING ligases, respectively, were fused to cDHFR. We observed growth phenotypes under nonpermissive conditions only when all the required ubiquitylation components were coexpressed (Fig. 2a,b).

SIAH2 inhibition by menadione may have an important role in cancer therapy21. As our engineered bacteria became addicted to self ubiquitylation of SIAH2, we predicted that SIAH2 inhibition would present a growth-arrest phenotype. Figure 2b (right) shows a growth-arrest phenotype in the presence of menadione only under the addictive conditions. Similarly, Cdc34, a noncognate E2, did not support growth. Our system could thus be employed for the screening of potential drugs. It has been suggested that some E3s use one E2 to attach the first Ub and a different E2 for building the Ub chain22. Our system may facilitate the identification of such E2s.

Identification of a novel E3 ligase and its cognate E2s

To demonstrate our system's ability to identify a novel E3 ligase, we focused on a U-box family from the pathogenic enterohaemorrhagic E. coli (EHEC). We used the well-characterized human ligase CHIP as a probe in a PSI-BLAST search against the EHEC proteome. The search retrieved an uncharacterized sequence, ECs3488, as a potential ligase containing a conserved 35-amino-acid sequence with 22% identity to a helix–loop–helix–beta region of the CHIP U-box domain (Fig. 2c). ECs3488, also named NleG6-3, has been postulated to be an E3 ligase, but its expression and function have never been demonstrated23,24. We cloned ECs3488 as a fusion with cDHFR and screened this fusion against a full yeast library of E2s (Supplementary Table 1), resulting in the identification of Ubc4 and Ubc5 as cognate E2s for the putative E3 ligase (Fig. 2d). We then examined the ligase functionality with the human E2 orthologs UBCH5B and UBCH5C (Fig. 2e).

Based on the structures of NleG2-3 and of the CHIP–UBCH5A complex (PDB 2KKX and 2OXQ)23, we built a structural model for the ECs3488–E2 interface (Fig. 2f) and employed the selection system to assess our model. The results corroborated the structural model, as the ECs3488–E2-binding mutants presented significant growth-arrest phenotypes (Fig. 2g).

Identification and characterization of UBDs

UBDs usually bind mono-Ub with low affinity, ranging from 2 μM to 2,100 μM15,25, posing a challenge to biochemical and biophysical studies. Our selection system stabilizes the dynamic and weak UBD–Ub noncovalent interactions by forming a covalent bond between Ub and Ub receptors (i.e., ubiquitylation) in the bacteria. We assessed the system's ability to sense low-affinity UBDs by tethering the double-sided UIM (dsUIM) of HRS as substrate26 (Fig. 3a). The structures of HRS–STAM (also known as ESCRT-0 complex), and particularly its dsUIM–Ub complex, were determined, and they facilitated detailed molecular assessment27. The E3-independent self ubiquitylation of most Ub receptors and the promiscuous function of Ubc4 and Ubc5 subfamily members were to our advantage16,28. Available lysine residue for ubiquitylation does not necessarily need to be part of the UBD, as was demonstrated for several UIM and UBA proteins15. Alanine residues at each face of the dsUIM were demonstrated to interact with the Ub I44 hydrophobic patch (Fig. 3b). Indeed, we found that the A266Q, A268Q double mutant presented a growth-arrest phenotype. Similarly, I44-patch mutant abolished bacterial growth. The phenotype of Ub–G76R demonstrated that growth is also dependent on ubiquitylation. Furthermore, western blot analysis showed that a covalent bond between Ub and HRS is generated (Fig. 3c). Finally, we demonstrated that bacterial growth is dependent on a functional ubiquitylation apparatus by omitting the E1 and E2 enzymes (Fig. 3a) or by administering the E1 inhibitor, PYR-41 (ref. 29) (Fig. 3d).

Figure 3: The bacterial selection system facilitates the identification and characterization of Ub receptors.
figure 3

UBDs promote self ubiquitylation and can therefore be detected by the bacterial selection system. The cDHFR was fused to various UBDs in the pCD–Sub and coexpressed in the selection system without E3 ligase. Bacterial spots were seeded and visualized as in Figure 1. (a) Ubiquitylation promoted by the HRS–UIM was used to validate the system performance. The effect of mutations at the UIM–Ub interfaces, the carboxyl terminus of Ub or deletions of the indicated individual components are shown. Complete coexpression of pND–Ub, including E1 and yeast Ubc4 and pCD–HRS–UIM. (b) Structure of the HRS dsUIM–Ub complex (PDB code 2D3G). (c) His6–Ub was coexpressed with E1 and Ubc4 along with GST–dsUIM. The protein was purified on GSH beads and subjected to western blot analysis using anti-GST (red) and anti-His-tag (green) antibodies. The yellow band that migrated at the molecular size of the ubiquitylated protein was detected by both antibodies. (d) Bacteria expressing a complete (pND–Ub, E1, yeast Ubc4 and pCD–HRS–UIM) or incomplete ubiquitylation apparatus (−E1), were grown in 96-well plates supplemented with PYR41 and/or TRIM as indicated. The relative growth rates at the log phase are shown with s.d. (n = 9). Norm. OD595, normalized optical density. (e–h) Growth phenotypes of assorted UBDs in the selection system. (e) Yeast Rpn10; (f) human STAM1-UIM; (g) human Alix-V domain; and (h) yeast Hse1-VHS domain. (i) The binding of Ub to the VHS domains of human STAM1, GGA1 and GGA2 was examined using the selection system. Critical STAM1-VHS residues known to bind Ub were mutated, and the growth phenotype was tested.

Full size image

The system's functionality with other structurally different UBDs, including the proteasomal receptor Rpn10, human STAM1-UIM, ALIX-V domains30 and the yeast Hse1-VHS domain25, that together are involved in multivesicular and retroviral budding, was furthered verified (Fig. 3e–h).

Similar to ESCRT-0 (STAM, HRS and yeast Hse1) components, GGA proteins also utilize UBDs to transport ubiquitylated cargo from the Golgi to the multivesicular body. An affinity of 2,100 μM was demonstrated to the GGA3-VHS–Ub complex25. Critical tryptophan and leucine residues were identified in STAM1 and other VHS domains that bind Ub at the I44 patch. Moreover, the VHS domains of GGA1 and GGA2, which naturally lack the critical leucine, do not bind Ub. We demonstrated that despite the weak affinity, the selection system distinguished between these phenotypes (Fig. 3i).

One benefit of a bacteriostatic antibiotic like trimethoprim is that it enables the accumulation of functional DHFR assemblies, due to the ubiquitylation, up to a threshold level that is sufficient to confer resistance while not harming the bacteria. Thus, our system may provide a highly sensitive readout for genetic identification and characterization of potential UBDs without the need to purify them.

ENTH is a UBD

We sought to challenge the system to detect a novel ultraweak-affinity UBD. ENTH domains assume a similar fold to that of VHS31. Moreover, epsin proteins possess a similar architecture to that of the Ub receptors HRS, STAM and GGAs by harboring VHS or ENTH, two Ub binding patches (dsUIM or 2-UIMs or GAT domains) followed by a long, flexible linker containing endocytic-machinery-binding elements including a clathrin-binding box (Supplementary Fig. 3). Therefore, although ENTH probably lacks the critical tryptophan or leucine residues25, we speculated that it too binds Ub. We employed the selection system and demonstrated that cDHFR–ENTH domains of yeast and zebrafish promote growth on selective media in an E3-independent manner (Fig. 4a,b). This suggests that these ENTH domains directly bind Ub–E2. Indeed, biochemical crosslinking assays and purification or detection of ubiquitylated yeast Ent1 derivatives from E. coli17 strongly support the genetic data suggesting that ENTH directly binds Ub (Supplementary Fig. 4).

Figure 4: Structural insight into a predicted ENTH–Ub interface.
figure 4

(a,b) Coexpression of ENTH domains from Danio rerio Epn1 (zebrafish, a) or Saccharomyces cerevisiae Ent1 (yeast, b) with the ubiquitylation apparatus in the bacterial selection system. Bacterial spots were seeded and visualized as in Figure 1. Complete, coexpression of pND–Ub, including E1 and yeast Ubc4 and pCD–ENTH domain. (c,d) Structures of the zebrafish and yeast epsin1 ENTH domains. The 2mFo-DFc sigma-A maps of the fish and the yeast proteins at 2.2 and 1.8 σ, respectively, of the refined models. (e) Superimposition of the fish and the yeast ENTH structures on top of the STAM1-VHS–Ub complex (pdb code 3LDZ). Average Cα r.m.s. deviations are 1.5 Å and 1.7 Å, respectively. (f) Zoomed-in view into crystal structure of the VHS–Ub interface (3LDZ). (g–h) Refined models of the ENTH–Ub complexes. (i,j) Zoomed-in view into the ENTH–Ub interfaces of the models.

Full size image

Structural insight into Ub recognition by ENTH domains

To obtain high-resolution information on the ENTH–Ub interaction, we tried to crystalize the complex. Probably because of the ultraweak affinity (see quantification below), only crystals of apo ENTH or Ub were obtained. We determined the structures of yeast (Sc_ENTH) and zebrafish (Zf_ENTH) domains by molecular replacement to 1.95-Å and 1.41-Å resolutions, respectively (Supplementary Table 2 and Fig. 4). High-quality electron density maps showed that (Fig. 4c,d) although these structures are highly similar, apparent differences can be seen in the loop-tethering helices 3 and 4 and in the angle between the superhelix structure and helix 8 (Supplementary Fig. 5).

To generate a structural model of the ENTH–Ub complex, we superimposed the ENTH domains onto the STAM1-VHS–Ub complex25,32. The models showed that the STAM1 W26 and L30 Ub-binding residues were naturally substituted with S37 and S41 in Zf_ENTH and K36 and I40 in Sc_ENTH (Fig. 4). Intriguingly, these models suggested that Ub R42 and R72 form electrostatic interactions with E42 and D45 or E41 and E44 of Zf_ENTH and Sc_ENTH, respectively. Both models predicted additional interactions that seemed to contribute little to the binding.

Superimposing the ENTH–Ub model onto the structure of ENTH complex with the membrane lipid phosphatidylinositol-4,5-bisphosphate33 showed that ENTH binds Ub and the membrane lipid at opposite sites (Supplementary Fig. 6), suggesting that ENTH can bind Ub while associated with the membrane. Similarly, superimposing the VHS–Ub complex onto the VHS–M6PR tail (the acidic-cluster-dileucine sorting signal of the mannose-6-phosphate receptor) complex34 showed the same phenomenon, suggesting that both VHS and ENTH domains can recognize ubiquitylated transmembrane cargo while associated with the membrane.

Genetic validation of ENTH–Ub interfaces

ENTH domains lack the critical tryptophan or isoleucine, suggesting that cumulative interactions of peripheral residues compensate for the role of ENTH domains in Ub binding. To evaluate the contributions of specific residues to the ENTH–Ub interaction, we introduced point mutations at the predicted interface. We monitored the growth rates by timelapse scanning the bacteria using a simple A4/US-letter scanner with a modified control software35. To quantify the growth rates, we applied a simple Fiji36-based timeseries analyzer procedure to measure a typical stack of 100–200 scans by means of optical density (Fig. 5).

Figure 5: Characterization of the ENTH–Ub binding interface.
figure 5

Structure-based mutants at the predicted ENTH–Ub interface were constructed and characterized. (a) Growth phenotypes of the zebrafish ENTH and Ub mutants in the selection system. Bacteria expressing a complete (pND–Ub, E1, yeast Ubc4 and pCD–ENTH variants) or incomplete apparatus (ΔE1,ΔE2 or ΔUb) were seeded and visualized as in Figure 1. (b) Growth curves of Ub wild type (WT) and mutants derived from the timelapse scanning of the spots (density was analyzed by Fiji). (c) As shown in b, but for ENTH mutants. (d) As shown in b and c, but for reciprocal mutations. (e) Growth phenotypes of the yeast ENTH–Ub interface mutants in the selection system. Bacterial spots were seeded and visualized as in Figure 1. (f) As shown in b, but for yeast ENTH and Ub mutants. (g) The ubiquitylation yield for wild type and indicated mutants of the yeast His6–MBP–ENTH were evaluated. The apo and ubiquitylated proteins were purified on amylose beads as described17, resolved on SDS–PAGE and detected by western blot with anti-His-tag antibody. (h) Quantification of the ubiquitylation yields shown in g. The ubiquitylated/apo ratio is presented as a percentage of the wild-type ratio. Values were averaged from four independent experiments, and s.d. error bars are presented. (i) Surface plasmon resonance (SPR) analysis of the yeast ENTH–Ub binding affinity of wild-type and mutant proteins. Fitting to binding curves was carried out with a single-site-binding model using the OriginLab software. Standard errors derived from three independent measurements are indicated.

Full size image

Alanine mutagenesis or exchanging the charge of the acidic residues of Zf_ENTH demonstrated growth-arrest phenotypes (Fig. 5a,c,d). A similar phenotype, though less severe for the Ub R42E, R72E mutant, was found when the acidic residues of the yeast protein were mutated, suggesting slight structural differences between these complexes. Remarkably, a permutation cycle in which the Zf_ENTH E42 and D45 residues were replaced with arginine residues, and Ub R42 and R72 were replaced with glutamine residues, restored growth (Fig. 5d). This result reflects the accuracy of our structural model and the high sensitivity of the selection system. Moreover, the Ub R42E, R72E mutant could not suppress the Zf_ENTH E42A, D45A mutant, signifying the importance of these electrostatic interactions. The Zf_ENTH S37G and Y83A mutants, predicted to provide lesser contributions to the binding, indeed yielded minor but significant growth phenotypes (Fig. 5a). Taken together, the mutational analysis results strongly corroborate the structural model and demonstrate the power of our system to detect and quantify relatively minor differences in protein–protein interactions along Ub pathways.

We biochemically quantified the yields of ubiquitylation of the Sc_ENTH wild-type and mutant proteins (Fig. 5g,h). The ENTH E41R, E44R and the Ub R42E, R72E double mutants significantly reduced the ubiquitylation yield by about 60–80%.

Finally, to biophysically corroborate the data and to quantify the affinity of the ENTH–Ub complex, we performed surface plasmon resonance (SPR) measurements with immobilized Sc_ENTH and free mono-Ub (Fig. 5i). Ultraweak binding with Kd of 2,300 μM was found for the wild-type complex. This result is compatible with our model and previous measurements of homologous VHS–Ub complexes in HRS, Vps27 and GGA3, which presented affinities of 1,400, 1,500 and 2,100 μM, respectively25. Interestingly, the Ub R42E, R72E or Sc_ENTH E41R, E44R mutants showed saturation-binding curves that could be fit to a single binding model (Fig. 5i and Supplementary Fig. 7) with respective estimated affinities 2.6-fold and 3.5-fold lower than that of the wild type. Notably, for these mutants, the Ub (analyte) concentrations were too low to obtain accurate Kd values, as reflected in the high standard errors. Together, the correlation between the SPR measurements and the genetic and biochemical data provides a rough estimation for the sensitivity of the selection system in monitoring Ub binding.

Ultraweak (∼3,000 μM) protein–protein interactions are significant, as they regulate various biological functions37,38. Typically, K63-tri–Ub chains constitute the main signal for clathrin-dependent membrane-protein trafficking39,40,41. Therefore, Ub receptors decoding this signal usually possess three Ub-binding patches42 (Supplementary Fig. 3). Avidity and/or cooperative of tandem UBDs render selectivity of these Ub receptors. The contribution of the VHS domain to the total affinity and selectivity of VHS–UIM proteins has been thoroughly studied25,32. Therefore, the ultraweak affinity reported here should be considered in the context of full-length epsin, which contains two additional UIMs.

Identification of Sem1 as ubiquitylation target of Rsp5

One of the greatest challenges in the Ub field is to identify association between E3 ligases and their cognate ubiquitylation targets (Supplementary Fig. 1). To demonstrate the potential of the developed system to address this challenge, we constructed a whole-genome yeast fusion library inframe with cDHFR and screened the library against Rsp5. The entire array of yeast-GST-tagged ORFs (GE collection) were amalgamated, and plasmids were isolated as a pool. We PCR amplified the plasmids library and subcloned the products into our selection system. As many Ub receptors may undergo E3-independent ubiquitylation, we expected to obtain false-positive growth of these ORFs, and therefore we employed an assay for E3-independent ubiquitylation. We compared the growth rates of the positive colonies with and without Rsp5 using the scanner as described above. Less than 100 positive colonies were identified (from ten Petri dishes). Most of them showed very similar growth rates in the presence or absence of Rsp5. However, the colony of Sem1 (Fig. 6) demonstrated significantly higher growth rates when Rsp5 was coexpressed.

Figure 6: Sem1 is a ubiquitylation substrate of Rsp5.
figure 6

Screening of pCD–Sub yeast-fusion library revealed Sem1 as a potential ubiquitylation substrate of Rsp5 in the bacterial selection system. (a) Sem1 ubiquitylation. Bacterial spots, expressing a complete (cDHFR–Sem1, nDHFR–Ub, yeast Ubc4, and Rsp5) or incomplete (ΔE1, ΔE2, ΔUb or ΔE3) ubiquitylation apparatus, were seeded as in Figure 1. Growth of spots was monitored by a timelapse scan using an office scanner (Epson) in 30-min intervals. Scan time postseeding is indicated. (b) Growth curves derived from quantification of the scans using 'Time Series Analyzer' in Fiji. Values are average of eight spots with s.d. bars. (c) Detection of Sem1 ubiquitylation in E. coli. Purified His6–MBP–Sem1 from E. coli that coexpress ubiquitylation apparatus17 was resolved on SDS–PAGE and detected by western blot with anti-His-tag antibody. WB, western blot. (d) His6–Sem1 was expressed from Gal-inducible promoter in wild-type or temperature-sensitive rsp5-1 mutant yeast cells. Cultures grew at 25 °C (permissive conditions). Prior induction temperature was shifted to 37 °C (restrictive temperature) or remained as indicated. His6–Sem1 was purified under denatured conditions, resolved and detected as in c.

Full size image

Sem1 and its human ortholog DSS1 (deleted split-hand/split-foot 1 protein) are involved in critical processes including development, proteasome assembly, DNA repair and cancer43,44,45. As DSS1 possesses two conserved UBDs, we carefully tested if Sem1 underwent E3-independent ubiquitylation. We found that Rsp5 significantly promoted Sem1 ubiquitylation (Fig. 6a,b and Supplementary Video 1). Moreover, detection of ubiquitylated His6–MBP–Sem1 from E. coli showed highly similar results (Fig. 6c). We further tested whether Sem1 undergoes Rsp5-dependent ubiquitylation in vivo in yeast. Ubiquitylation was detected in yeast extracts of wild-type and a temperature-sensitive rsp5 allele (rsp5-1). Expression of galactose-dependent His6-Sem1 at permissive and restrictive temperatures showed that Sem1 underwent ubiquitylation by wild-type Rsp5 at both temperatures. However, in rsp5-1, ubiquitylation was detected only at permissive temperature (26 °C; Fig. 6d), suggesting that Rsp5 is a bona fide E3 ligase of Sem1.

Discussion

Recent developments of chemical biology techniques and bacterial expression of the ubiquitylation apparatus have facilitated the production and downstream studies of ubiquitylated proteins17,18,46,47,48,49. However, genetic tools for screening the ubiquitylation pathways are still limited. The selection system described here serves as a prototypical tool for genetic studies of Ub signals. The system presents some limitations including a nonfunctional expression of some E3s and their targets. Whereas the bacterial system facilitates identification of ubiquitylation targets that are negatively regulated by post-translational modifications, it may miss targets that require modification before ubiquitylation. Also, we did not test the system using different orientation fusions. Nevertheless, we anticipate that this tool should help facilitate the identification of genetic associations linking ubiquitylation enzymes to their substrates. It may be of value for drug discovery and for characterizing protein–protein interfaces along Ub cascades. A fairly low percentage of the known human proteins that undergo ubiquitylation have a fully characterized ubiquitylation cascade comprised of specific E1, E2 and E3 enzymes50. We expect that our selection system will facilitate and expedite the uncovering of many unknown associations.

Methods

Plasmid construction.

Selection system for ubiquitylation. Two sets of E. coli-compatible expression vectors were constructed. Ub and different targets were fused to two fragments of the murine DHFR and expressed from a constitutive lambda phage left promoter pL. A full list of the vectors is available in Supplementary Table 1. The pND–Ub vector is based on a modified pZE21 vector, contacting the pLtetO1. The mouse cDNA of the nDHFR fragment (residues 1–108) was PCR-amplified and subcloned between the KpnI and PacI sites (the PacI site was inserted into the vector by PCR) into the pZE21 vector. Then a flexible linker (as indicated in Supplementary Fig. 1) was constructed using a two-step PCR method and digested with PacI and NotI (the NotI site was inserted into the vector by PCR) and cloned into the vector downstream to the pLtetO1–nDHFR. Ubiquitylation apparatus cassettes containing a His6–Ub–E2–E1 were PCR amplified from pGEN plasmids expressing different E2s or Ub mutants17 and subcloned as fusion to the C terminus of the nDHFR–linker1 into the NotI and AvrII endonuclease recognition sites.

The pCD–Sub vector was constructed based on a pCDF-duet vector. The T7 promoter was substituted with pLtetO1. The cDHFR fragment (residues 109–187) was PCR-amplified and subcloned under the pL promoter at the MluI and AscI endonuclease recognition sites (the AscI site was inserted by PCR). A second linker (linker2, as indicated in Supplementary Fig. 1) was fused to the N terminus of MBP and cloned into the AscI site. Substrates were PCR amplified and subcloned into SacII and SpeI sites downstream and inframe with the cDHFR–linker2–MBP. In some vectors the MBP was removed. Some vectors were prepared by complete chemical synthesis assembly51.

Cloning into expression, purification and detection vectors. The ENTH domain of zebrafish Epn1(18–157) and the dsUIM of human HRS(257–276) were subcloned into the pGST-parallel2 vector52 between BamHI and EcoRI endonuclease recognition sites. The ENTH domain of yeast Ent1(1–152) was subcloned into the pCDF-duet vector fused to His6–MBP as previously described (pCOG21 (ref. 17)). For detection of ubiquitylation in bacteria, yeast Sem1 was subcloned into pCDF-duet vector fused to His6–MBP. For yeast expression and purification His6–Sem1 was subcloned into psGREG600 by recombineering that also removed the GFP from the vector. The structure and sequences of all vectors were confirmed by restriction endonuclease and DNA-sequencing analyses.

Site-directed mutagenesis.

Point mutations were introduced using the ExSite approach (Stratagene). The entire vector was amplified using Phusion DNA polymerase. Parental DNA was digested by DpnI, and the DNA was blunt ligated. All mutants were sequenced to ensure that the desired mutations were introduced and that no other mutations occurred.

Ubiquitylation-dependent E. coli growth assay.

E. coli W3110 (from Yagil laboratory at TAU) were cotransformed with the pND–Ub and the pCD–Sub plasmids and plated on LB agar supplemented with 34 μg/ml kanamycin and 25 μg/ml streptomycin. 5 ml of liquid LB medium supplemented with the same antibiotic concentrations was inoculated with a single colony and left to grow overnight at 37 °C. The culture was harvested and washed twice with 5 ml of minimal Davis medium. The optical density (OD600nm) was measured and adjusted to 0.2. 2.5 μl of the diluted cultures from each sample were spotted on agar Davis plates containing 0, 0.5, 5, 10, 20 or 50 μg/ml of TRIM. The plates were incubated for 2–3 days at 30 °C and photographed with a UV camera in identical conditions. Each spotting assay was repeated at least six times.

Selection experiments in solution growth media.

Overnight cultures in LB medium (supplemented with 30 μg/ml kanamycin and 25 μg/ml streptomycin) were harvested and resuspended in Davis minimal medium. Diluted cells (OD600, 0.2) were grown at 30 °C in 96-well plates containing 0.2 ml Davis medium supplemented with 0, 1, 5, 7, 10 and 12 μg/ml TRIM and with or without 100 μM Pyr-41. Growth rates were monitored by measuring the optical density (OD595) using a microplate spectrophotometer. Doubling time was calculated for early logarithmic growth (OD595 between 0.02 and 0.2). All experiments were performed at least nine times (n = 9).

Genetic selection assays for characterization of structural-based mutants.

Data collection: E. coli. W3110 expressing the pND–Ubs, pCD–Subs (and sometimes also a an ampicillin-resistance plasmid that constitutively expresses E3 ligase) grew to logarithmic phase at 37 °C in 5 ml of LB medium supplemented with 23 μg/ml kanamycin, 16 μg/ml streptomycin and 33 μg/ml ampicillin. The culture was harvested and washed once with 5 ml of minimal Davis medium. The bacterial density was adjusted to OD600nm value of 0.3. Culture samples (2.5 μl each) were spotted on Davis agar Petri dishes containing 10 μg/ml trimethoprim. Culture in each experiment was spotted, usually three to four times. Most experiments were repeated at least three to four times (therefore 9 < n > 16). Timelapse (30 or 60 min) scanning took place in 26 °C incubator using a regular A4/US-letter office scanner (Epson Perfection V37)35.

Image analysis. Images were read into Fiji36 as a stack using 'import -> image sequence'. The densities of the spots were measured using the Time Series Analyzer V3 (Balaji J 2007; a Java ARchive ImageJ/Fiji plugin that can be downloaded and installed from http://rsb.info.nih.gov/ij/plugins/time-series.html). Regions of interest (ROIs) were specified (typically as 20 × 20 ovals), and their total intensities (bacterial densities) were integrated and plotted where 'Z-axis' is the image time index. Similarly, the background was measured and subtracted from the collected data. Logistic regressions of growth curves were calculated using Origin. A single parameter that describes growth efficiency was calculated as follows: the growth curve slope at the 'half max density' was extracted and divided by its time index.

Protein purification.

Proteins were purified from E. coli using affinity tags as previously described17. For crystallization purposes, proteins were concentrated to 5–20 mg/ml using centricon (Amicon Ultra) in a final buffer of 50 mM Tris–HCl (pH 8.0), 150 mM NaCl, and 5 mM dithiothreitol (DTT).

Western blot analysis.

Following separation on SDS–PAGE, samples were transferred onto a nitrocellulose membrane and incubated with rabbit anti-His epitope tag antibody (1:20,000 dilution, Rockland; cat. no. 600-431-382) or mouse anti-GST antibody (1:200, Santa Cruz; cat. no. 600-430-200), and infrared dye-coupled goat anti-mouse secondary antibody (1:12,000, LI-COR; cat. no. 926-32210). Blots were scanned using an Odyssey system (LI-COR Biosciences) at 700 and 800 nm.

Crystallization and data collection.

Crystallization and data collection and processing of yeast Ent1-ENTH have been reported53. Purified zebrafish Epn1-ENTH was concentrated to 10 mg/ml and was crystallized in 0.1 M KPO4 pH 7.0 and 19% PEG 3,350 at 20 °C. Crystals were cryoprotected with 25% ethylene glycol and were frozen in liquid nitrogen. Data were collected at the ID29 beamline (ESRF, Grenoble) and processed with HKL2000 package54.

Structure determination and refinement.

Structures were solved by molecular replacement (MR); 1H0A (ref. 33) was used for searching model with PHASER55. To facilitate the MR search, the first 17 residues were removed, and alanine reduction was exerted to the nonconserved residues. Model building and refinement were carried out with PHENIX56, Refmac5 (ref. 57) and COOT58. Crystallographic data and refinement statistics are in Supplementary Table 2.

Crosslinking assays.

Crosslinking was carried out using 0.5 mM disuccinimidyl substrate (DSS) as described by Azem and coworkers59.

Surface plasmon resonance.

Purified His6–MBP–ENTH or mutants were immobilized on an Ni+2 chip at a density of ∼800 U. Untagged Ub (analyte) was injected at a flow rate of 20 ml per min in 150 mM NaCl, 10 mM HEPES pH 7.0 and 0.010% polysorbate 20 at 24 °C; 500 mM imidazole was used for surface regeneration. To avoid immobilization of aggregated analyte, the purified Ub was chromatographed by gel filtration and briefly sonicated immediately before surface plasmon resonance (SPR) experiments. Data were processed with BioEvaluation and fitting carried out as a single-site-binding model with OriginLab. Standard errors derived from at least three samples.

Detection of ubiquitylation in vivo.

pScHis6–Sem1 plasmid was transformed into the SEY6210 (kindly gifted from the Scott Emr Laboratory at Cornell), rsp5::HIS3, pDsRed415-rsp5WT and rsp5-1 (rsp5L733S) MATa ura3-52, his3-200, trp1-901, lys2-801, suc2–9, leu2-3 strains and grown at 26 °C in 50 ml of YPD medium (2% glucose) supplemented with 200 μg/ml G418. At log phase the cultures were harvested and the pastes were washed with DDW and transferred to 50 ml YPD medium (2% galactose) supplemented with 200 μg/ml G418. Each culture was divided into two flasks and grew at permissive (26 °C) and restrictive (37 °C) temperatures for an additional 4 h. The cultures were harvested and lysed in 2 ml cold 1.85 N NaOH, 7.5% β-mercaptoethanol for 10 min on ice. Proteins were precipitated with the addition of half volume of 50% tricarboxylic acid (TCA at final concentration of 25%) and collected by centrifugation (18,000 r.p.m. for 20 min). The pellet was resuspended and washed with 3 ml ice-cold 80% acetone and collected by centrifugation (18,000 r.p.m. for 5 min). Then the pellet was resuspended in 1.5 ml resuspension buffer (6 M guanidine HCl, 100 mM Tris–HCl, 100 mM NaCl, 0.1% triton X-100, pH 8.8) and incubated at 25 °C for 1 h while rotating. The protein fraction was incubated 20 min at 4 °C with 70 μl Ni+2 resin and washed twice with resuspension buffer followed by two washes with buffer 2 (8 M urea, 100 mM Tris–HCl pH 8.8, 100 mM NaCl, 0.1% triton X-100) and two washes with buffer 3 (50 mM Tris–HCl pH 7.5, 150 mM NaCl). The Ni+2 beads were boiled in Laemmli buffer and separated by SDS–PAGE followed by western blot analysis with anti-His6 antibody.

Library construction.

Yeast GST-tagged ORFs (GE collection) were pooled from 384 pin-plated cultures, and plasmids were isolated as a pool. The GST fusion genes were PCR amplified. The PCR products were size fractionated (350 to 3,000 bp) by gel electrophoresis followed by purification and subcloning inframe with cDHFR to the pCD–Sub vector by recombineering. The resultant library was transformed to DH5α cells and kept at −80 °C. Isolated plasmids of the library were transformed to W3110 competent cells that contained the pND–Ub and constitutively expressed E3-ligase plasmids. Following transformation the bacteria were plated on selective media.

Accession codes.

Atomic coordinates and structure factors for the crystal structures of Saccharomyces cerevisiae Ent1-ENTH and Danio rerio Epn1-ENTH domains have been deposited in the Protein Data Bank under ID codes 5LOZ and 5LP0, respectively. Plasmids have been deposited in Addgene (accession codes are listed in Supplementary Table 3).