A eukaryotic-like ubiquitination system in bacterial antiviral defense

Summary

Ubiquitination pathways play crucial roles in protein homeostasis, signaling, and innate immunity^1–3. In these pathways, an enzymatic cascade of E1, E2, and E3 proteins conjugates ubiquitin or a ubiquitin-like protein (Ubl) to target-protein lysine residues⁴. Bacteria encode ancient relatives of E1 and Ubl proteins involved in sulfur metabolism^5,6 but these proteins do not mediate Ubl-target conjugation, leaving open the question of whether bacteria can perform ubiquitination-like protein conjugation. Here, we demonstrate that a bacterial operon associated with phage defense islands encodes a complete ubiquitination pathway. Two structures of a bacterial E1:E2:Ubl complex reveal striking architectural parallels with canonical eukaryotic ubiquitination machinery. The bacterial E1 encodes an N-terminal inactive adenylation domain (IAD) and a C-terminal active adenylation domain (AAD) with a mobile α-helical insertion containing the catalytic cysteine (CYS domain). One structure reveals a pre-reaction state with the bacterial Ubl C-terminus positioned for adenylation, and a second structure mimics an E1-to-E2 transthioesterification state with the E1 CYS domain adjacent to the bound E2. We show that a deubiquitinase (DUB) in the same pathway pre-processes the bacterial Ubl, exposing its C-terminal glycine for adenylation. Finally, we show that the bacterial E1 and E2 collaborate to conjugate Ubl to target-protein lysine residues. Together, these data reveal that bacteria possess bona fide ubiquitination systems with strong mechanistic and architectural parallels to canonical eukaryotic ubiquitination pathways, suggesting that these pathways arose first in bacteria.

The conjugation of ubiquitin and Ubls to target proteins involves three enzymes termed E1, E2, and E3 (Extended Data Figure 1a). An E1 “activating protein” catalyzes the formation of a high-energy acyl adenylate intermediate through the reaction of ATP with the Ubl C-terminus, then a cysteine on the E1 attacks that intermediate to form a thioester link to the Ubl⁴. Next, a cysteine on an E2 “carrier protein” attacks the E1~Ubl thioester to form a second E2~Ubl thioester. Finally, conjugation of the Ubl to a lysine side chain on a target protein is mediated by an E3 “ligase,” which may simply act as an E2-target adapter (as in RING-family E3s) or form a third cysteine~Ubl thioester intermediate before transfer to a target (as in HECT-family E3s). Some ubiquitination pathways circumvent the need for an E3, with their E2 proteins directly recognizing target proteins^7–9. In all Ubl conjugation pathways, specific peptidases termed deubiquitinases or DUBs cleave Ubl-lysine isopeptide linkages in addition to pre-processing many Ubls to expose their reactive C-terminal glycine residue for E1-mediated catalysis¹⁰.

Bacteria encode widespread E1-like adenylation enzymes (ThiF, MoeB) that act on ubiquitin-like proteins (ThiS, MoaD), but these pathways do not encode E2-like proteins and their primary biological roles are in sulfur metabolism rather than Ubl-target conjugation^5,6. Mycobacteria and other Actinobacteria conjugate Pup (prokaryotic ubiquitin-like protein) to targets to mediate degradation, but the mechanism of Pup-target ligation is distinct from the canonical E1-E2-E3 ubiquitination pathway^11–15. Many pathogenic bacteria encode E3 ligases and DUBs that modulate ubiquitin signaling by their eukaryotic host, but these do not constitute complete ubiquitination pathways^16–18.

Over the past two decades, sparsely distributed bacterial operons have been reported that encode distinct combinations of putative E1, E2, Ubl, and DUB proteins^19–22, plus rare examples of RING-type E3 proteins²³. One such operon family, termed Bil (Bacterial ISG15-like gene) encodes putative E1, E2, Ubl, and DUB proteins and protects its host against bacteriophage (phage) infection by specifically modifying a virion structural protein^24,25. These findings suggest that counter to prevailing models in which ubiquitination pathways arose in archaea^26,27, bacteria encode ancient ubiquitination-like pathways that participate in antiviral defense.

Identification of Type II BilABCD operons

We recently showed that bacterial Type II CBASS immune systems encode an E1-E2 fusion protein (Cap2) that resembles the noncanonical eukaryotic E1 Atg7 and its partner E2s Atg3 and Atg10 ²⁸. Cap2 conjugates a cGAS (cyclic GMP-AMP synthase)-like protein to target proteins using a ubiquitination-like mechanism^28,29. We additionally described four bacterial operon families that encode different combinations of E1, E2, JAB-family DUB, and Ubl proteins, one of which encoded a multidomain E1 protein and an uncharacterized protein termed CEHH after four conserved residues (cysteine, glutamate, histidine, histidine)²⁸. In an earlier bioinformatic analysis of prokaryotic ubiquitination-related proteins, this operon family was denoted “6C” and described as encoding three proteins: multidomain E1, CEHH, and a JAB-family DUB¹⁹. AlphaFold structure predictions³⁰ and sequence alignments suggested that CEHH likely represents a diverged E2-like protein, and we further found that most CEHH-containing operons also encode a previously unidentified Ubl (Supplementary Table 1). Together, these observations suggested that the CEHH/6C operons may represent a complete ancestral ubiquitination system.

We performed comprehensive sequence searches using a multidomain E1 protein from Ensifer aridi TW10 and identified 136 complete nonredundant CEHH/6C operons, with most instances in the plant-associated Rhizobiaceae family (Figure 1a, Supplementary Table 1). Forty-one CEHH/6C operons are located in immune-enriched loci identified by the PADLOC server³¹ (Extended Data Figure 1b, Supplementary Table 1), suggesting that they occur in so-called “defense islands” ^31–34. Further, twenty-five CEHH/6C operons are associated with CapH/CapP transcriptional regulators, which are associated with diverse bacterial immune systems and which activate these systems’ expression in response to DNA damage³⁵. Together, these data show that CEHH/6C operons encode a complete ubiquitination-like pathway with predicted E1, E2, Ubl, and DUB proteins, and further suggest that they participate in antiviral defense.

Figure 1. Structure of a bacterial E1-E2-Ubl complex.

(a) Schematic of a Type II BilABCD operon from Ensifer aridi TW10. See Supplementary Table 1 for a full list of Type II BilABCD operons, and Extended Data Figure 1 for comparison with Type I BilABCD operons. (b) Top: Domain schematic of E. aridi E1^BilD, E2^BilB, and Ubl^BilA. The inactive adenylation domain (IAD) of E1^BilD is colored light blue, the active adenylation domain (AAD) yellow, and the α-helical insertion containing the putative catalytic cysteine (CYS domain) brown. Putative catalytic cysteine residues in E1^BilD (C417) and E2^BilB (C138) are noted. Left: E. aridi E1^BilD:E2^BilB:Ubl^BilA complex Form 1 crystal structure, with domains colored as in the schematic. Putative catalytic cysteine residues are shown as yellow spheres. Right: E. aridi E1^BilD:E2^BilB:Ubl^BilA complex Form 2 crystal structure. See Extended Data Figure 3 for comparison to a canonical eukaryotic E1(NAE1-UBA3):E2(Ubc12):Ubl(NEDD8) complex. (c) Closeup view of the Ubl^BilA C-terminus (orange) docked in the E1BilD adenylation active site (yellow, with conserved active-site residues shown as sticks and labeled). The E1 catalytic cysteine residue (C417) is positioned 14 Å away from the Ubl C-terminus. See Extended Data Figure 3c, g for comparison of bacterial and eukaryotic E1 adenylation active sites. (d) Closeup view of the E1^BilD-E2^BilB binding interface, including the structural Zn²⁺ ion coordinated by four cysteine residues in E1^BilD. Extended Data Figure 3d, h for comparison of bacterial and eukaryotic E1-E1 interfaces. (e) Closeup view of the E1^BIlD and E2^BilB catalytic cysteine positions in the Form 2 structure. Compared to Form 1, the E1 CYS domain is rotated upward and the E1^BilD catalytic cysteine (C417) is positioned within 2 Å of the E2 catalytic cysteine (C138), mimicking the structural state adopted during an E1^BilD-to-E2^BilB transthioesterification reaction.

We observed that the gene order in CEHH/6C operons (Ubl–CEHH/E2–DUB–E1) matches that of BilABCD operons (Ubl–E2–DUB–E1; Extended Data Figure 1c)²⁴. Sequence alignments of the E1, E2, and DUB-like proteins in CEHH/6C and BilABCD operons show that in each case, these proteins share identifiable sequence homology but segregate into distinct groups in unrooted evolutionary trees (Extended Data Figure 2a–c). The broad parallels between CEHH and BilABCD operons, combined with these specific differences, prompted us to rename the CEHH/6C operons as Type II BilABCD operons.

Bacterial E1-E2-Ubl complex structure

We cloned and expressed the four proteins from a Type II BilABCD operon from Ensifer aridi TW10 (Ubl^BilA, E2^BilB, DUB^BilC, and E1^BilD) for biochemical and structural analysis. While both E1^BilD and E2^BilB were insoluble when overexpressed in E. coli, they formed a soluble complex when coexpressed with Ubl^BilA. We crystallized and determined two independent X-ray crystal structures of the E1^BilD:E2^BilB:Ubl^BilA complex to 2.5 Å resolution (Form 1) and 2.7 Å resolution (Form 2) (Figure 1b, Extended Data Table 1).

E1^BilD encodes an N-terminal domain of unknown function (Pfam14459) annotated as “prokaryotic E2 family C” and a C-terminal E1-like adenylation domain (Extended Data Figure 1c), and was originally speculated to comprise an E2-E1 fusion protein¹⁹. Our structure shows that E1^BilD adopts an overall architecture strikingly similar to that of canonical eukaryotic E1 proteins (Figure 1b, Figure 2a–b, Extended Data Figure 3). Whereas all known bacterial E1 and E1-like proteins including ThiS, MoeB, and Cap2 form homodimers through their adenylation domains^5,6,28, E1^BilD possesses an N-terminal inactive adenylation domain (IAD) and a C-terminal active adenylation domain (AAD) that form a structurally related pseudo-dimer. Inserted into the “crossover loop” of the E1^BilD AAD is an ~80 amino acid α-helical domain containing a conserved cysteine residue (CYS domain). This domain architecture – IAD, AAD, and CYS – is a hallmark of canonical eukaryotic E1 proteins. The closest structural relatives of E1^BilD are the human ubiquitin/FAT10 E1 UBA6 ³⁶ and the heterodimeric human NEDD8 E1 NAE1-UBA3³⁷ (Figure 2a–b, Table 1, Extended Data Figure 3). Like these eukaryotic E1s, E1^BilD possesses a single active adenylation site in the AAD, with one key arginine residue (Arg9) provided by the IAD (Figure 1c, Extended Data Figure 3c, i). The E1^BilD CYS domain is linked to the AAD via flexible linkers and is highly mobile, showing high B-factors compared to the IAD/AAD and adopting different positions in Form 1 versus Form 2 structures. In the Form 1 structure, E1^BilD adopts an open-like conformation with the CYS domain positioned directly above the adenylation active site and the conserved cysteine residue (Cys417) 14 Å away from the C-terminus of the bound Ubl^BilC (Figure 1c). In Form 2, the CYS domain is rotated upward to contact the bound E2^BilB (see below). E1^BilD coordinates a zinc ion through four cysteine residues, two in the AAD crossover loop and two near the protein’s C-terminus (Figure 1d); similar structural zinc ions are observed in SUMO and NEDD8 E1 proteins (Extended Data Figure 3d,j)^4,37,38. Overall, while E1^BilD lacks accessory domains found in many eukaryotic E1 proteins –including the first catalytic cysteine half-domain (FCCH) or coiled-coil domain (CC) that is often inserted within the IAD, and the C-terminal ubiquitin fold domain (UFD) that mediates E2 binding – the overall structural parallels between E1^BilD and canonical eukaryotic E1 proteins suggest a common ancestry.

Figure 2. E. aridi E1^BilD, E2^BilB, and Ubl^BilA resemble canonical eukaryotic ubiquitination machinery.

(a) Domain schematic of E. aridi E1^BilD and the H. sapiens ubiquitin/FAT10 E1 UBA6 (PDB ID 7SOL)³⁶. IAD: inactive adenylation domain; AAD: active adenylation domain; CYS: catalytic cysteine-containing domain; FCCH: first catalytic cysteine half-domain; SCCH: second catalytic cysteine half-domain (equivalent to CYS); UFD: ubiquitin fold domain. (b) Structures of E. aridi E1^BilD (left) and H. sapiens UBA6 (PDB ID 7SOL)³⁶, with domains colored as in panel (a) and catalytic cysteines shown as spheres and colored yellow. The Cα r.m.s.d. is 3.5 Å over 401 residue pairs spanning the two proteins’ IAD and AAD domains. (c) Structures of E. aridi Ubl^BilA (left), M. musculus ubiquitin (center; PDB ID 4NQL)⁶³, and S. cerevisiae SUMO/Smt3 (right; PDB ID 5JNE)⁵⁹. (d) Structures of E. aridi E2^BilB (left) and the H. sapiens E2 UBE2D2 (right; PDB ID 4DGG)⁶¹. Each protein’s catalytic cysteine residue is shown as sticks and labeled (Cys85 is mutated to serine in the H. sapiens UBE2D2 structure. E. aridi E2^BilB His151 is also shown as sticks. See Extended Data Figure 4a for an equivalent view showing all four conserved residues of the originally identified CEHH motif. For UBE2D2, the C-terminal α-helices not shared by E2^BilB are shown in white.

Table 1. Structural similarity between Ensifer aridi BilABCD proteins and homologs.

Root mean squared deviation (r.m.s.d.) for Cα atoms when comparing each Ensifer aridi protein to homologous eukaryotic proteins.

Protein	Homolog	Cα r.m.s.d.	Reference
E1BilD	H. sapiens UBA6	3.5 Å (401 Cα)	36
E1BilD	H. sapiens NAE1-UBA3	NAE1: 3.6 Å (221 Cα) UBA3: 3.7 Å (250 Cα)	37
UblBilA	M. musculus ubiquitin	1.7 Å (72 Cα)	58
UblBilA	S. cerevisiae Smt3	1.7 Å (73 Cα)	59
UblBilA	E. coli ThiS	3.3 Å (57 Cα)	6
UblBilA	E. coli MoaD	3.3 Å (64 Cα)	5
E2BilB	S. cerevisiae UBC12	3.0 Å (94 Cα)	60
E2BilB	H. sapiens UBE2D2	3.2 Å (99 Cα)	61
DUBBilC	C. subterraneum Rpn11	2.5 Å (110 Cα)	62
DUBBilC	S. pombe Sst2	1.2 Å (79 Cα)	63

In contrast to identified Type I BilABCD operons whose Ubl (BilA) contains two predicted ubiquitin-like β-grasp domains²⁴, E. aridi Ubl^BilA contains a single β-grasp domain (Figure 2c). Ubl^BilA is structurally more similar to eukaryotic ubiquitin and SUMO than it is to the bacterial β-grasp proteins ThisS and MoaD (1.7 Å Cα r.m.s.d. for eukaryotic homologs, 3.3 Å for bacterial homologs; Table 1). While E. aridi Ubl^BilA encodes a single β-grasp domain, Ubl^BilA proteins across Type II BilABCD operons show remarkable structural diversity: we identified examples of Ubls predicted to encode up to three tandem β-grasp domains, or containing N-terminal domains with predicted coiled-coil or disordered regions (Extended Data Figure 2d) This variety in Ubl architecture suggests that ubiquitination by BilABCD operons could disrupt target-protein function through diverse mechanisms.

In our structures of the E1^BilD:E2^BilB:Ubl^BilA complex, Ubl^BilA is bound to E1^BilD equivalently to known Ubl-E1 complexes, with its C-terminus positioned in the E1^BilD adenylation active site. We do not observe electron density for a bound ATP or AMP molecule in the active site. Indeed, the C-terminal alanine residue of Ubl^BilA (Ala98) is positioned in a manner that would physically prevent binding of ATP. In a sequence alignment of all Ubl^BilA homologs from Type II BilABCD operons, we found that these proteins encode a universally-conserved glycine followed by up to nine residues of non-conserved sequence (Extended Data Figure 2d). In E. aridi Ubl^BilA, the conserved glycine (Gly97) is positioned one residue from the C-terminus. Close inspection of our structures reveals that Ubl^BilA residue Gly97 is positioned properly for adenylation, but cannot undergo this reaction because of the presence of the C-terminal residue Ala98 (Extended Data Figure 3c, i). Thus, in this system Ubl^BilA likely requires proteolytic processing by its cognate DUB^BilC to expose the reactive Gly97 residue for catalysis by E1^BilD (see next section).

In our structures, E2^BilB adopts a fold similar to canonical E2 proteins from S. cerevisiae and H. sapiens (Figure 2d, Table 1). The conserved CEHH motif in this protein includes residues Cys138, Glu144, His146, and His151 ¹⁹. Of these, Cys138 is positioned equivalently to canonical E2 proteins’ catalytic cysteine. His151 is adjacent to Cys138, and may participate directly in catalysis. The other two conserved residues (Glu144 and His146) appear to play purely structural roles in E2^BilB, with their side chains hydrogen-bonding to nearby backbone amide and carbonyl groups (Extended Data Figure 4a). In keeping with their likely functional importance, the putative catalytic cysteine (Cys138) and its adjacent histidine (His151) are highly conserved in E2^BilB proteins across both Type I and Type II BilABCD operons, while Glu144 and His146 are not conserved in Type I BilABCD (Extended Data Figure 4a).

In both structures of the E1^BilD:E2^BilB:Ubl^BilA complex, E2^BilB is bound to the E1^BilD AAD near the adenylation active site and CYS domain (Figure 1b). While E1^BilD lacks the C-terminal UFD that participates in E2 binding in many canonical E1 proteins, the binding of E2^BilB to E1^BilD nonetheless closely resembles canonical E1-E2 binding. The interaction involves a loop and α-helix near the C-terminus of E1^BilD (residues 490-504), which is rigidified by the bound Zn²⁺ ion coordinated by Cys340 and Cys343 in the AAD crossover loop, plus Cys491 and Cys493 near the E1^BilD C-terminus (Figure 1d, Extended Data Figure 3e,j). In the E1^BilD:E2^BilB:Ubl^BilA Form 2 structure, the E1^BilD CYS domain is rotated away from the adenylation active site, and its putative catalytic cysteine is positioned adjacent to the E2^BilB putative catalytic cysteine (Cys138; Figure 1e, Extended Data Figure 3f). In the structure, the two residues are apparently linked via a disulfide bond. This state closely mimics the structural state of an E1-to-E2 transthioesterification intermediate, in which the E1 and E2 catalytic cysteine residues approach to within ~2 Å of one another³⁹. Thus, our two structures reveal how the mobile E1^BilD CYS domain likely reacts first with a Ubl^BilA-adenylate intermediate to generate the E1^BilD~Ubl^BilA thioester, then rotates upward to mediate Ubl^BilA handoff to E2^BilB.

In both structures of the E1^BilD:E2^BilB:Ubl^BilA complex, E2^BilB forms a symmetric homodimer either through non-crystallographic symmetry (Form 1) or crystallographic symmetry (Form 2; Extended Data Figure 4b–d). Supporting the idea that E2^BilB forms a homodimer, size exclusion chromatography coupled to multi-angle light scattering (SEC-MALS) shows that a purified E1^BilD (C417A):E2^BilB:Ubl^BilA complex forms a 2:2:2 complex in solution, while complexes with E2^BilB mutated to disrupt the dimer interface form 1:1:1 complexes (Extended Data Figure 4e–f).

Ubl preprocessing by a bacterial DUB

In eukaryotic Ubl pathways, deubiquitinases play two roles: they pre-process Ubls to expose their reactive C-terminal glycine (peptide bond cleavage), and they also cleave Ubl-target conjugates (isopeptide bond cleavage)¹⁰. Our structures and sequence alignments suggest that in many BilABCD systems including that of E. aridi, DUB^BilC is required to pre-process Ubl^BilA to expose the reactive C-terminal glycine. To test the activity of E. aridi DUB^BilC, we generated a model substrate comprising Ubl^BilA fused at its C-terminus to green fluorescent protein (GFP). We found that wild type DUB^BilC, but not two different active-site mutants (Glu33 to Ala (E33A) and Asp106 to Ala(D106A)), showed robust cleavage of Ubl^BilA-GFP (Figure 3a). We isolated the C-terminal product of Ubl^BilA-GFP cleavage and subjected it to N-terminal sequencing by Edman degradation, and found that DUB^BilC cleaves Ubl^BilA one residue upstream of the C-terminus, between Gly97 and Ala98 (Figure 3b–c, Extended Data Figure 5a). Thus, E. aridi DUB^BilC likely pre-processes Ubl^BilA to expose the reactive Gly97 for catalysis.

Figure 3. E. aridi DUB^BilC specifically cleaves Ubl^BilA.

(a) Biochemical analysis of E. aridi DUB^BilC cleaving a Ubl^BilA-GFP fusion protein. DUB^BilC constructs used were wild-type (WT), E33A active site mutant, or D106A active site mutant. For gel source data, see Supplementary Figure 1. This experiment was independently performed three times, with consistent results. (b) Schematic of the cleavage reaction shown in panel (a) and analysis by N-terminal sequencing (Edman degradation) of the C-terminal fragment (marked with an asterisk in panel (a)). See Extended Data Figure 5a for N-terminal sequencing data showing the sequence AGIGS, pinpointing the DUB^BilC cleavage site as C-terminal to Ubl^BilA Gly97. (c) Sequence logo from bacterial Type II Ubl^BilA proteins (Supplementary Table 1). Type II Ubl^BilA homologs possess up to nine residues C-terminal to the highly conserved glycine (Gly97 in E. aridi Ubl^BilA). (d) X-ray crystal structure of DUB^BilC(E33A) (purple) bound to Ubl^BilA (orange). A composite omit map (Extended Data Figure 5b) shows that Ubl^BilA is cleaved at Gly97. See Extended Data Figure 5d for structural comparisons with eukaryotic JAMM-family peptidases. (e) Closeup view of the DUB^BilC(E33A) active site with bound zinc ion (gray) and the Ubl^BilA C-terminus (orange).

We next purified a complex of E. aridi DUB^BilC(E33A) and full-length Ubl^BilA, and determined two X-ray crystal structures at 1.36 Å resolution (Form 1) and 1.68 Å resolution (Form 2) (Figure 3d–e, Extended Data Figure 5b–c, Extended Data Table 1). The structures reveal DUB^BilC as a JAB/JAMM family peptidase, with the closest structural homologs being an archaeal JAMM peptidase and several eukaryotic AMSH enzymes (Extended Data Figure 5d). In both structures, Ubl^BilA is bound to DUB^BilC with its C-terminus positioned in the active site near the catalytic Zn²⁺ ion (Figure 3d–e). Despite using the catalytic mutant DUB^BilC(E33A) for crystallization, we observed that in both structures Ubl^BilA was cleaved at residue Gly97. This observation is likely explained by residual low-level cleavage activity in this mutant (Figure 3a). These structural data support a model in which E. aridi DUB^BilC pre-processes Ubl^BilA by cleaving between Gly97 and Ala98.

BilABCD mediates Ubl-lysine conjugation

Our structural data on E1^BilD, E2^BilB, and Ubl^BilA suggest that these proteins comprise a complete ubiquitination system equivalent to canonical eukaryotic ubiquitination pathways. To test this idea, we coexpressed E1^BilD and E2^BilB with Ubl^BilA truncated at Gly97 (Ubl^BilAΔ97) in E. coli. Using an N-terminal His₆-tag on Ubl^BilA, we purified Ubl^BilA and associated proteins. In native buffer conditions that maintain both covalent and noncovalent complexes with Ubl^BilA, we observed three major bands representing E1^BilD, E2^BilB, and Ubl^BilA, plus minor bands with a wide range of molecular weights (Figure 4a). These minor bands were not present when Ubl^BilA contained its native C-terminal Ala98 residue; when E1 was mutated to eliminate Ubl^BilA adenylation (Arg246 to Ala); when the E1^BilD putative catalytic cysteine (Cys417) was mutated to alanine; or when the E2^BilB putative catalytic cysteine (Cys138) was mutated to alanine (Figure 4a). These data suggest that the observed minor bands represent covalent Ubl^BilA-target protein conjugates that depend on Ubl^BilA adenylation by E1^BilD, followed by sequential thioester formation with E1^BilD and E2^BilB. A similar experiment with E2^BilB dimer mutants shows that these mutants also generate Ubl^BilA-target conjugates (Extended Data Figure 6a), albeit at a lower level than wild type E2^BilB; thus, E2^BilB dimerization is not strictly required for its catalytic activity.

Figure 4. Bacterial BilABCD systems catalyze ubiquitination.

(a) SDS-PAGE analysis of Ni²⁺ affinity-purified His₆-Ubl^BilA and associated proteins in native conditions, after coexpression with E1^BilD and E2^BilB. Bands representing His₆-Ubl^BilA, E1^BilD, and E2^BilB are marked. Ubl^BilA-target conjugates are visible in the second lane. Ubl^BilA FL: full-length, unreactive Ubl^BilA. Ubl^BilA Δ97: truncated at glycine 97 to enable E1-mediated catalysis; E1 R246A: adenylation active site mutant; E1 C417A: catalytic cysteine mutant; E2 C138A: catalytic cysteine mutant. WT: wild type. For gel source data for panels (a)-(c), see Supplementary Figure 1. Experiments shown in panels (a)-(c) were independently performed three times, with consistent results. (b) SDS-PAGE analysis of Ni²⁺ affinity-purified His₆-Ubl^BilA and associated proteins in denaturing conditions, after coexpression with E1^BilD and E2^BilB. Bands representing His₆-Ubl^BilA and His₆-Ubl^BilA-target conjugates are marked. Red asterisk: contaminant. (c) SDS-PAGE and anti His₆-tag western blot analysis of affinity-purified His₆-Ubl^BilA-target conjugates (from denaturing purifications) treated with DUB^BilC. (d) Experimental scheme for identification of Ubl^BilA targets. See Extended Data Figure 6d for purification of His₆-Ubl^BilA(Δ97, V95K)-target complexes. (e) MS/MS spectrum of modified Ubl^BilA(Δ97, V95K) residues 76-95. Y-ions shifted by the dipeptide Ala-Gly (AG) remnant are annotated in color. (f) Extracted Ion Chromatograms (EICs) of transition ions of the mis-cleaved, modified Ubl^BilA(Δ97, V95K) residues 76-95, using a 10 ppm m/z tolerance. RT: retention time. (g) Extracted Ion Chromatograms (EICs) of transition ions of the mis-cleaved, unmodified Ubl^BilA(Δ97, V95K) residues 76-95, using a 10 ppm m/z tolerance. RT: retention time. See Extended Data Figure 6e for EIC of the properly cleaved, unmodified peptide spanning Ubl^BilA(Δ97, V95K) residues 76-92.

We next purified His₆-tagged Ubl^BilAΔ97 in denaturing conditions to eliminate copurification of non-covalently associated proteins. When coexpressed with wild-type E1^BilD and E2^BilB, we again observed that Ubl^BilA copurified with a range of minor bands, which were not present when either the E1 or E2 putative catalytic cysteine residues were mutated to alanine (Figure 4b). These data confirm that the minor bands are likely covalent complexes between Ubl^BilA and E. coli proteins, that are generated by E1^BilD and E2^BilB. We incubated the purified mixture of Ubl^BilA-target conjugates with purified DUB^BilC, then visualized the reactions by SDS-PAGE and western blotting with an antibody that recognizes the His₆-tag on Ubl^BilA. Untreated samples showed a large number of bands, suggesting that each band represents an E. coli protein conjugated to His₆-tagged Ubl^BilA (Figure 4c). After incubation with wild-type DUB^BilC, the minor bands were mostly lost and western blotting showed a single major band corresponding to monomeric Ubl^BilA (Figure 4c). The minor bands were not lost upon incubation with the DUB^BilC catalytic mutants E33A or D106A, confirming that this loss is due to proteolysis by DUB^BilC. Thus, E1^BilD and E2^BilB can generate covalent Ubl^BilA-target conjugates, which DUB^BilC can cleave.

Next, we coexpressed all four proteins (E1^BilD, E2^BilB, His₆-Ubl^BilA, and DUB^BilC) in E. coli and purified Ubl^BilA and associated proteins. When Ubl^BilA was full-length (ending at Ala98), we observed the formation of Ubl^BilA-target conjugates in the presence of wild-type DUB^BilC, but not the DUB^BilC E33A active-site mutant (Extended Data Figure 6b–c). These data demonstrate that DUB^BilC can pre-process Ubl^BilA to expose the reactive Gly97 residue and enable catalysis by E1^BilD and E2^BilB.

Finally, we used mass spectrometry (MS) to verify that Ubl^BilA is conjugated to target proteins through an isopeptide linkage. We generated a His₆-tagged Ubl^BilAΔ97 construct with a Val95 to Lys (V95K) mutation, such that trypsin digestion of Ubl^BilA-target conjugates would leave an alanine-glycine dipeptide linked to a target lysine residue (Figure 4d). We verified that the V95K mutant did not disrupt Ubl^BilA-target conjugate formation (Extended Data Figure 6d), then purified His₆-Ubl^BilA(Δ97, V95K) in denaturing conditions after coexpression with E1^BilD and E2^BilB. We separated the purified proteins by SDS-PAGE, then excised a region of the gel containing ~20-40 kDa proteins (importantly, excluding monomeric Ubl^BilA). We extracted proteins from the gel, cleaved with trypsin, and performed LC-MS/MS with data-dependent acquisition to search for peptides that contained a lysine residue modified with an alanine-glycine dipeptide. We identified a peptide corresponding to residues 76-95 of Ubl^BilA(Δ97, V95K) modified at Lys92, then modified the MS parameters to selectively detect its daughter ions (Figure 4e). All detected daughter ions showed identical chromatographic elution profiles by reverse-phase liquid chromatography (Figure 4f), supporting the identification of this modified peptide. We similarly identified the unmodified Ubl^BilA(Δ97, V95K) 76-95 peptide with an essentially identical chromatographic elution profile (Figure 4g), and the Ubl^BilA(Δ97, V95K) 76-92 peptide that is properly cleaved by trypsin at Lys92 (Extended Data Figure 6e). These data confirm that E1^BilD and E2^BilB can catalyze formation of a bona fide Ubl^BilA-target isopeptide linkage.

Discussion

Bacteria encode widespread E1- and Ubl-like proteins involved in sulfur metabolism, but true ubiquitin-like protein conjugation was not demonstrated in bacteria until the CBASS-associated protein Cap2 was shown to conjugate a cGAS-like protein to target proteins as part of an antiviral immune response^28,29. In a variation on this theme, other CBASS systems encode an E2-like regulator that mediates covalent oligomerization of its cognate cGAS-like protein through a ubiquitination-like mechanism⁴⁰. Here, we reveal that bacterial BilABCD operons encode full ubiquitination systems with E1, E2, DUB, and Ubl proteins that architecturally and mechanistically resemble canonical eukaryotic ubiquitination machinery. We demonstrate that E1^BilD encodes tandem IAD and AAD domains, with a mobile CYS domain capable of shuttling Ubl^BilA from the E1^BilD adenylation site to E2^BilB. We show that all BilABCD proteins more closely resemble canonical eukaryotic ubiquitination machinery than functionally related bacterial proteins. When coexpressed in E. coli, E1^BilD and E2^BilB mediate Ubl^BilA conjugation to both cellular proteins and Ubl^BilA itself. Finally, the DUB protein BilC pre-processes Ubl^BilA to expose its reactive C-terminal glycine, and can also cleave Ubl^BilA-target conjugates. Overall, these data show that bacteria encode ancient relatives of canonical ubiquitination pathways. Remarkably, several other uncharacterized bacterial operon families are predicted to encode distinct combinations of Ubl, E1, E2, RING E3, and DUB proteins^19–22,28. As with Type II CBASS and BilABCD systems, these operon families are predicted to function in antiviral defense^19–22. Thus, the evolutionary pressure of the phage-bacteria arms race has given rise to a profusion of ubiquitination-like protein conjugation pathways that likely interrupt phage infections through a variety of mechanisms.

Our data reveal that when expressed in a heterologous host, E. aridi E1^BilD and E2^BilB can mediate nonspecific Ubl^BilA-target conjugation. In the context of bacteriophage infection, BilABCD could protect its host by non-specifically ubiquitinating abundant phage proteins, potentially inhibiting their self-assembly into functional phage progeny. A more likely scenario is that BilABCD specifically ubiquitinates one or a few phage proteins to interrupt key life-cycle events like DNA replication, virion assembly, or host-cell lysis. Alternatively, BilABCD could cause production of progeny phage that are non-infectious due to ubiquitination of key structural proteins. Indeed, recent work on a Type I BilABCD operon from Collimonas sp. OK412 shows that this system specifically ubiquitinates the central tail fiber protein of phages Secphi27 and Secphi4, and that this activity results in both defective virion assembly and impaired infectivity of progeny phage²⁵. We propose that E2^BilB may directly recognize phage target proteins, providing specificity to this system in the absence of a dedicated E3 protein.

BilABCD operons were originally dubbed “Bacterial ISG15-like” because of shared innate-immune function and structural similarity between Ubl^BilA and the eukaryotic Ubl ISG15 ²⁴. Our data solidify the functional parallels between bacterial BilABCD pathways and canonical eukaryotic ubiquitination pathways, including the innate-immune ISG15 pathway. ISG15 and its E1 UBA7 are found only in vertebrates and likely evolved from UBB/UBC and UBA1, respectively^20,41; this pathway is therefore not a direct descendent of bacterial BilABCD. Nonetheless, these pathways are likely to function similarly; in bacterial immunity, Ubl conjugation in different BilABCD systems likely has diverse effects on infecting phages, including the known effects on virion assembly and infectivity through modification of a key tail fiber protein²⁵, plus potentially aberrant oligomerization or aggregation of modified target proteins mediated by diverse Ubls. The ISG15 pathway has been proposed to function similarly, by modifying viral proteins to interrupt key life-cycle events^3,42,43. While the ISG15 pathway appears to have emerged after the establishment of eukaryotes rather than being inherited from a bacterial or archaeal ancestor²⁰, the functional parallels between BilABCD and ISG15 pathways strongly point to the effectiveness of ubiquitination-like protein conjugation pathways in mediating innate immunity across kingdoms.

Methods

Bioinformatics

To identify bacterial CEHH operons, the Ensifer aridi TW10 E1^BilD protein (IMG accession 2509379665) was used as a query for BLAST searches in the IMG database of bacterial genomes (https://img.jgi.doe.gov/). Hit sequences were filtered for redundancy, aligned using the MAFFT⁴⁴ version 7 web server (https://mafft.cbrc.jp/alignment/server/index.html) and visualized in Jalview⁴⁵. For each non-redundant hit, the genomic neighborhood of the E1^BilD gene was visually inspected in IMG for neighboring genes similar to E. aridi Ubl^BilA, E2^BilB, and DUB^BilC, plus genes similar to CapH (usually annotated as Pfam01381; XRE-family HTH domain) and CapP (usually annotated as Pfam06114; Peptidase_M78)³⁵. In all cases, the bilA-bilB-bilC-bilD gene order was consistent, and in all but one case capH and capP were positioned upstream of bilABCD and oriented in the opposite coding direction (i.e. sharing a common promoter region with bilABCD). Average linkage (UPGMA) trees were calculated by the MAFFT web server and visualized using the Interactive Tree of Life server (https://itol.embl.de/). Protein structure predictions (monomer and multimer) were performed using the AlphaFold_MMseqs2 implementation of ColabFold⁴⁶, which implements AlphaFold 2 ^30,47 predictions using sequence alignments generated by MMseqs2 ⁴⁸.

Protein expression and purification

Ensifer aridi Ubl^BilA (Joint Genome Institute Integrated Microbial Genomes (JGI-IMG) accession 2509379668), E2^BilB (IMG accession 2509379667), DUB^BilC (IMG accession 2509379666), and E1^BilD (IMG accession 2509379665) were individually cloned into E. coli expression vectors encoding either no tag (UC Berkeley Macrolab vector 2A-T, Addgene ID 29665) or an N-terminal TEV protease-cleavable His₆-tag (UC Berkeley Macrolab vector 2B-T, Addgene ID 29666). For coexpression, multigene coexpression cassettes were generated by PCR and cloned into UC Berkeley Macrolab vector 2B-T to generate an N-terminal TEV protease-cleavable His₆-tag on one protein. Targeted mutants were generated by PCR-based mutagenesis.

Vectors were transformed into E. coli Rosetta 2 pLysS (EMD Millipore), and 1 L cultures were grown at 37°C in 2XYT media to an OD₆₀₀ of 0.6 before induction with 0.25 mM IPTG at 20°C for 16–18 hours. Cells were harvested by centrifugation, resuspended in buffer A (25 mM Tris-HCl pH 8.5, 300 mM NaCl, 5 mM MgCl₂, 10% glycerol and 5 mM mercaptoethanol) containing 5 mM imidazole, then lysed by sonication (Branson Sonifier). Lysates were clarified by centrifugation, then supernatants were passed over a Ni-NTA Superflow column (Qiagen) in resuspension buffer. The column was washed in wash buffer (buffer A containing 20 mM imidazole), then eluted in elution buffer (buffer A containing 400 mM imidazole). Eluates were concentrated by ultrafiltration (Amicon Ultra; EMD Millipore), then passed over a Superdex 200 Increase size exclusion column (Cytiva) in size exclusion buffer (25 mM Tris-HCl pH 8.5,300 mM NaCl, 5 mM MgCl₂, 10% glycerol and 1mM DTT). Peak fractions were concentrated by ultrafiltration and stored at 4°C.

For denatured purification of E1^BilD:E2^BilB:His₆-Ubl^BilA, cells were harvested by centrifugation, resuspended in buffer A, then lysed by sonication (Branson Sonifier). Lysates were clarified by centrifugation, then supernatants were passed over a Ni-NTA Superflow column (Qiagen) in resuspension buffer. The column was washed in buffer B (100 mM Tris-HCl pH 8, 8 M Urea and 100 mM NaH₂PO₄), followed by wash buffer (buffer A containing 20 mM imidazole), then eluted in elution buffer (buffer A containing 400 mM imidazole) . Eluates were concentrated by ultrafiltration (Amicon Ultra; EMD Millipore) and stored at 4°C.

For characterization of oligomeric state by size exclusion chromatography coupled to multi-angle light scattering (SEC-MALS), 100 μl of purified protein complexes at a concentration of 5 mg/ml were injected onto a size exclusion column (Superdex 200 Increase 10/300 GL, Cytiva) in size exclusion buffer, then light scattering and differential refractive index (dRI) profiles were collected using miniDAWN TREOS and Optilab T-rEX detectors (Wyatt Technology). SEC-MALS data were analyzed using ASTRA software version 8 and visualized with Prism version 10 (GraphPad Software).

DUB^BilC activity assays

To generate a model substrate for DUB^BilC cleavage, Ubl^BilA was cloned into a vector encoding a C-terminal GFP tag (UC Berkeley Macrolab vector H6-msfGFP, Addgene ID 29725) and purified as above. The Ubl^BilA-GFP fusion protein (10 μg) was mixed with 5 μg of DUB^BilC (wild type or mutants) in 20 μL reaction buffer containing 20 mM HEPES pH 7.5, 100 mM NaCl, 20 mM MgCl₂, 20 μM ZnCl₂ and 1 mM DTT, then incubated 30 minutes at 37°C. Reactions were analyzed by SDS-PAGE with Coomassie blue staining. For protein N-terminal sequencing by Edman degradation, cleavage products were separated by SDS-PAGE, transferred to a Bio-Rad ImmunBlot PVDF membrane, and visualized by staining with Coomassie Blue R-250. The band representing the C-terminal cleavage product of Ubl^BilA-GFP was excised, the membrane was washed with methanol and deionized water, then loaded onto a Shimadzu PPSQ-53A instrument for analysis. Five rounds of Edman degradation were performed, and the “evaluated value” scores for each round were analyzed to obtain the likely N-terminal sequence.

Western blotting

For western blots of His₆-Ubl^BilA conjugates, proteins were transferred to PVDF membranes using a Trans-Blot Turbo RTA Mini 0.2 μm PVDF Transfer Kit (Bio-Rad) according to the manufacturer’s instructions, using a Trans-Blot Turbo Transfer System (Bio-Rad). Membranes were blocked with 5% Non-Fat Dry Milk in TBST, then incubated with mouse anti-His tag primary antibody (Millipore Sigma SAB1305538) at 1:1,000 dilution, followed by HRP-conjugated secondary antibody (HRP Goat anti-mouse IgG, Millipore Sigma AP128P) at 1:30,000 dilution. After washing, HRP signal was detected with Amersham ECL Select Western Blotting Detection Reagent (Cytiva) using a ChemiDoc Imaging System (Bio-Rad) in Protein Blot - Chemiluminescence setting.

Crystallography

For crystallization of the E. aridi DUB^BilC(E33A):Ubl^BilA complex (Form 1), purified protein at 30 mg/mL in crystallization buffer (25 mM Tris-HCl pH 8.5, 200 mM NaCl, 5 mM MgCl₂, and 1 mM TCEP (tris(2-carboxyethyl)phosphine)) was mixed 1:1 with well solution containing 100 mM HEPES pH 7.5, 0.2 M MgCl₂, and 25% PEG 3350 in hanging drop format. Crystals were harvested into cryoprotectant solution containing an additional 10% glycerol and frozen in liquid nitrogen. Diffraction data were collected at the Advanced Photon Source (Argonne National Lab) NE-CAT beamline 24ID-E on April 7, 2023 (collection temperature 100 K; x-ray wavelength 0.97918 Å) (Extended Data Table 1). Data were processed with the RAPD data-processing pipeline (https://github.com/RAPD/RAPD), which uses XDS⁴⁹ for data indexing and reduction, POINTLESS⁵⁰ for space group assignment, and AIMLESS⁵¹ for scaling. The structure was determined by molecular replacement in PHASER⁵² using a predicted structure from AlphaFold 2 ³⁰ as a search model. The model was manually rebuilt in COOT⁵³ and refined in phenix.refine⁵⁴ using positional and individual B-factor refinement. B-factors for all atoms except waters were refined anisotropically. The final model has good geometry, with 98.75% of residues in favored Ramachandran space, 1.25% allowed, and 0% outliers. The overall MolProbity score is 1.18, and the MolProbity clash score is 3.98.

For crystallization of the E. aridi DUB^BilC(E33A):Ubl^BilA complex (Form 2), purified protein at 30 mg/mL in crystallization buffer (25 mM Tris-HCl pH 8.5, 200 mM NaCl, 5 mM MgCl₂, and 1 mM TCEP (tris(2-carboxyethyl)phosphine)) was mixed 1:1 with well solution containing 100 mM MES pH 6.5 and 1M sodium/potassium tartrate. Crystals were harvested into cryoprotectant solution containing an additional 30% glycerol and frozen in liquid nitrogen. Diffraction data were collected at the Advanced Photon Source (Argonne National Lab) NE-CAT beamline 24ID-E on April 7, 2023 (collection temperature 100 K; x-ray wavelength 0.97918 Å) (Extended Data Table 1). Data were processed with the RAPD data-processing pipeline, and the structure was determined by molecular replacement in PHASER using the Form 1 BilC^E33A:BilA structure (with ligands and waters excluded) as a search model. The model was manually rebuilt in COOT and refined in phenix.refine using positional and individual isotropic B-factor refinement. The final model has good geometry, with 99.58% of residues in favored Ramachandran space, 0.42% allowed, and 0% outliers. The overall MolProbity score is 1.08, and the MolProbity clash score is 1.86.

For crystallization of the E. aridi E1^BilD:E2^BilB:Ubl^BilA complex (Form 1), purified protein at 10 mg/mL in crystallization buffer was mixed 1:1 with well solution containing 100 mM HEPES pH 7.5, 100 mM sodium citrate, 5% isopropanol, and 10% PEG 3350 in hanging drop format. Crystals were harvested into cryoprotectant solution containing an additional 20% glycerol and frozen in liquid nitrogen. Diffraction data were collected at the Advanced Photon Source (Argonne National Lab) NE-CAT beamline 24ID-E on April 4, 2023 (collection temperature 100 K; x-ray wavelength 0.97918 Å) (Extended Data Table 1). Data were processed with the RAPD data-processing pipeline, and the structure was determined by molecular replacement in PHASER using predicted structures of each protein (E1^BilD CYS domain excluded) from AlphaFold2 as a search model. One copy of Ubl^BilA, two copies of E2^BilB, and two copies of E1^BilD were located. The model was manually rebuilt in COOT and refined in phenix.refine using positional and individual isotropic B-factor refinement. The final model has good geometry, with 96.23% of residues in favored Ramachandran space, 3.47% allowed, and 0.30% outliers. The overall MolProbity score is 1.64, and the MolProbity clash score is 5.86.

For crystallization of the E. aridi E1^BilD:E2^BilB:Ubl^BilA complex (Form 2), purified protein at 10 mg/mL in crystallization buffer was mixed 1:1 with well solution containing 100 mM imidazole pH 8.0, 200 mM MgCl₂, and 10% PEG 3350 in hanging drop format. Crystals were harvested into cryoprotectant solution containing an additional 20% glycerol and frozen in liquid nitrogen. Diffraction data were collected at the Advanced Photon Source (Argonne National Lab) NE-CAT beamline 24ID-E on April 7, 2023 (collection temperature 100 K; x-ray wavelength 0.97918 Å) (Extended Data Table 1). Data were processed with the RAPD data-processing pipeline, and the structure was determined by molecular replacement in PHASER using predicted structures of each protein (E1^BilD CYS domain excluded) from AlphaFold2 as a search model. The model was manually rebuilt in COOT and refined in phenix.refine using positional and individual isotropic B-factor refinement. The final model has good geometry, with 96.05% of residues in favored Ramachandran space, 3.53% allowed, and 0.42% outliers. The overall MolProbity score is 1.82, and the MolProbity clash score is 5.62.

In-Gel Digestion

For mass spectrometry identification of Ubl^BilA-conjugated proteins, E. aridi E1^BilD and E2^BiLB were coexpressed with His₆-Ubl^BilA(V95K) in E. coli, and purified by Ni-NTA chromatography in denatured solution as above. Purified proteins were separated by SDS-PAGE and visualized with Coomassie blue staining, then mass spectrometry was performed by in-gel trypsin digestion as previously described⁵⁵. In brief, gel slices were cut into ~1 mm x 1 mm x 1 mm cubes using a clean razor blade in a glass dish. Gel slices were reduced in 100 μl of 10 mM DTT for 30 minutes at 37°C. Cysteine alkylation was performed by incubating the samples at room temperature in the dark for 20 minutes following the addition of 6 μl of 0.5 M iodoacetamide (30 mM iodoacetamide final concentration). To digest proteins, 100 μl of 10 ng/μl trypsin (Promega #V511A) in 20 mM ammonium bicarbonate (pH 8) was added to submerge the gel pieces, then incubated on ice for 30 minutes until fully swollen. An additional 20–50 μl of ammonium bicarbonate buffer was added to ensure the gel slices were fully submerged prior to overnight incubation at 37°C. The next day, trypsin digested peptides were extracted from the sample via multiple extractions using 50% acetonitrile/5% formic acid, dried under vacuum and reconstituted in 20 μl of 0.1% trifluoroacetic acid (pH 2).

LC-MS/MS Analysis

Peptides were analyzed by LC-MS/MS on a Vanquish Neo high performance liquid chromatography system coupled to a Q-Exactive Plus mass spectrometer (Thermo Fisher Scientific). Data-dependent analysis (DDA) was performed using a Top 10 approach using a linear gradient of 2 – 30% mobile phase B (Mobile Phase A: 0.1% formic acid in H₂O, Mobile Phase B: 0.1% formic acid in ACN) at a flow rate of 75 μL/min on a 15-cm C₁₈ column maintained at 40°C (Acclaim Pepmap 1 mm I.D., 2 μm particle size, 100 Å pore size; Thermo Fisher Scientific). The LC method for DDA starts at 2% B and is held constant for 1 min, followed by a change to 30% B across 20 min, followed by a change to 70% B across 3 min, and then finally increased to 95% B across 0.1 min. Column washing and equilibration is performed across 4.1 min making the total method time 28.2 min. Parallel reaction monitoring (PRM) was performed in DIA mode utilizing a global inclusion list with a full scan using a linear gradient of 2 – 40% mobile phase B at a flow rate of 1.5 μL/min on an in-house packed 15-cm pulled-tip column emitter (100 μm I.D. with 2.2-μm C₁₈ beads, 120 Å pore size; Sepax Technologies) maintained at 50 °C (PRSO-V2; Sonation GmbH). The LC method starts at 2% B and is held constant for 5 min, followed by a change to 40% B across 25 min, and then finally increased to 95% B across 1.5 min. Column washing and equilibration is performed across 13.5 min making the total method time 45 min. Targeted PRM acquisition of the Ubl^BilA (V95K) C-terminal peptides were performed in DIA mode using a targeted inclusion list for the +2 and +3 charge states for the modified and non-modified peptide species. Initial data review and quality control of PRM acquired transition spectra were analyzed in Skyline (v22.2.0.527). Final extracted transition ion chromatograms were generated using FreeStyle (v1.7, Thermo Fisher Scientific) using a ± 10 ppm m/z tolerance for each transition.

Peptide Sequencing Analysis

Database searching of data-dependent acquired MS spectra was performed using the Trans-Proteomic Pipeline software suite v6.3.2 Arcus (TPP, Seattle Proteome Center)^56,57. The search parameter file and protein database used for the database search is included in Supplementary Table 2. Briefly, database search was performed using COMET and peptides were quantified using XPRESS in label-free mode. A static modification of 57.021464 Da was applied to cysteine residues. Two differential modifications were applied as follows: Oxidation of methionine (15.9949 Da) and the expected AG remnant of Ubl^BilA (V95K) on lysine/N-termini (128.05857 Da). Quality control was performed following data analysis and manual inspection of chromatography and MS/MS spectral assignments. Redundant peptide identifications were removed to generate the final list of unique peptides (Supplementary Table 3).

The protein database included all proteins from the E. coli proteome (Uniprot: ECOLX) plus the following:

>tr|HIS6_Ubl-BilA_delta97-V95K

MKSSHHHHHHENLYFQSNASKDSRKGDNHGGGSGKIEIIVVVNGQPTQVEANPNQPLHVV

RTKALENTQNVAQPPDNWEFKDEAGNLLDVDKKIGDFGFANTVTLFLSLKAGKAG

>tr|E1-BilD

MALANFIDRAATAASQVLTDFHLGDFKAALEKQVVAVAFDDQAISCAEGQATLDLAVRLLA

RLYPVLAILPLDSAASSQAQALERLAKSINRKIGIRRSGKSATVCLVAGATRPSLRCPTFF

IGSDGWAAKLSRTDPVGSGSSLLPYGAGAASCFGAANVFRTIFAAQLTGAESDENIDLSLY

SYNKSRAGDAGPIDPAVDLGETHLVGLGAIAHGALWALARQSGLSGRLHVVDHEAVELSNL

QRYVLAGQAEIGMSKAVLATTALRSTALEVEAHPLKWAEHVARRGDWIFDRVGVALDTAAD

RVAVQGALPRWIANAWTQEHDLGISRHGFDDGQACLCCMYMPSGKSKDEHQLVAEELGIPE

AHEQVKALLQTNAGVPNDFVVRVATAMGVPFEPLAPFVGQPLRSFYQQAICGGLVFQLSDG

SRLVRTVVPMAFQSALAGIMLAAELVKHSAGFPMSPTTSTRVNLLRPLGSHLHDPKAKDSS

GRCICSDEDFISAYRRKYGNGVEPLSNISAEQKRTSPLPRTGRQVCA

>tr|E2-BilB

MPELQTVDPEVSRAKFDREISRFRPYADAYRMQGCFLIEESFPSAFFIFASPKVKPRVIGA

AIEIDFTNYDLRPPSVVFVDPFTRQPIARKDLPLNMLRRPQLPGTPPEMISNLIQQNAVSLT

DFIQANSLQDSPFLCMAGVREYHDNPAHSGDPWLLHRGSGEGCLAFILDKIIKYGTGPVEQL

HIQLQYAVGLLVPPQAIPE

Extended Data

Extended Data Figure 1. Comparison of Type I and Type II BilABCD operons.

(a) Schematic of a typical ubiquitination pathway, with ubiquitin-like protein (Ubl) in orange, E1 in yellow, E2 in blue, optional E3 in light gray, and target in gray. An E1 protein mediates adenylation of the Ubl C-terminus followed by generation of an E1~Ubl thioester linkage. This thioester linkage is transferred to E2, then to a lysine residue on a target (optionally with the help of an E3). (c) Schematics of CEHH/6C operons in defense islands identified by the PADLOC server³¹. See Supplementary Table 1 for additional information. (c) Operon schematics of a Type I BilABCD operon from Collimonas sp. OK412 ²⁴, and a Type II BilABCD (CEHH) operon from E. aridi TW10 (Supplementary Table 1). Noted under each gene are the conserved PFAM domain annotations for that gene.

Extended Data Figure 2. Sequence analysis of BilABCD proteins.

(a) Unrooted average distance tree assembled from E1^BilD proteins from Type I BilABCD operons²⁴ and Type II BilABCD operons (Supplementary Table 1). Branches representing proteins from Type I and Type II operons are marked. Specific E1^BilD proteins from Collimonas sp. OK412 (Type I) and E. aridi TW10 (Type II) are indicated with red stars and labeled. Scale bar: 1 substitution per site. (b) Unrooted average distance tree assembled from E2^BilB proteins from Type I BilABCD operons²⁴ and Type II BilABCD systems (Supplementary Table 1). Branches representing proteins from Type I and Type II operons are marked. Specific E2^BilB proteins from Collimonas sp. OK412 (Type I) and E. aridi TW10 (Type II) are indicated with red stars and labeled. Scale bar: 1 substitution per site. (c) Unrooted average distance tree assembled from DUB^BilC proteins from Type I BilABCD operons²⁴ and Type II BilABCD operons (Supplementary Table 1). Scale bar: 1 substitution per site. Branches representing proteins from Type I and Type II operons are marked. Specific DUB^BilC proteins from Collimonas sp. OK412 (Type I) and E. aridi TW10 (Type II) are indicated with red stars and labeled. Scale bar: 1 substitution per site. (d) Unrooted evolutionary tree of all Ubl^BilA proteins in Type II BilABCD operons (Supplementary Table 1). Specific examples are labeled and their domain architectures (inferred from sequence analysis) noted. LLPS NTD: N-terminal domain predicted to undergo liquid-liquid phase separation; CC NTD: N-terminal domain predicted to form a coiled-coil. Inset: Sequence logo from bacterial Type II Ubl^BilA proteins (Supplementary Table 1). Type II Ubl^BilA homologs possess up to nine residues C-terminal to the highly conserved glycine (G97 in E. aridi BilA).

Extended Data Figure 3. Structural parallels between bacterial and eukaryotic ubiquitination machinery.

(a) Domain architecture of E. aridi E1^BilD, E2^BilB, and Ubl^BilA. IAD: inactive adenylation domain; AAD: active adenylation domain; CYS: E1 catalytic cysteine-containing domain; Ubl: Ubiquitin-like domain. (b) Crystal structure (Form 2) of the E. aridi E1^BilD:E2^BilB:Ubl^BilA complex, with domains colored as in (a). (c) Closeup view of the E1^BilD adenylation active site with bound Ubl^BilA C-terminus. Conserved active site residues are shown as sticks and labeled. (d) View equivalent to (c) showing 2F_o-F_c composite-omit electron density at 1.5σ. (e) Closeup view of the E1^BilD-E2^BilB binding interface. The zinc ion (gray sphere) is coordinated by E1^BilD residues C340, C343, C491, and C493. (f) Closeup view of the catalytic cysteine residues of E1^BilD (C417; brown) and E2^BilB (C138; blue) with 2F_o-F_c composite-omit electron density at 2.0σ. (g) Domain architecture of H. sapiens E1^NAE1-UBA3, E2^UBE2M, and Ubl^NEDD8 (PDB ID 2NVU)³⁷. CC: coiled-coil domain; UFD: ubiquitin-fold domain. (h) Crystal structure of the H. sapiens E1^NAE1-UBA3:E2^UBE2M:Ubl^NEDD8 complex (PDB ID 2NVU)³⁷, with domains colored as in panel (e). (i) Closeup view of the E1^UBA3 adenylation active site with bound Ubl^NEDD8 C-terminus. Conserved active site residues are shown as sticks and labeled, and bound ATP is shown as sticks. (j) Closeup view of the E1^BilD-E2^BilB binding interface. The UBA3 UFD is shown in gray. The zinc ion (gray sphere) is coordinated by UBA3 residues C199, C202, C343, and C346.

Extended Data Figure 4. Type II Bil E2^BilB protein structure and homodimer formation.

(a) Top: Sequence alignment of E. aridi E2^BilB (residues 133–158) and Collimonas sp. OK412 E2^BilB (residues 108–135), with the four residues of the originally identified CEHH motif in 6C/CEHH operon proteins shown in blue highlights. Of the four residues, only C138 and H151 are highly conserved across both Type I and Type II E2^BilB proteins. Bottom: Structure of E. aridi E2^BilB, with the four residues of the CEHH motif shown as sticks and labeled. (b) Crystallographic dimer of E2^BilB in the Form 2 structure. (c) Closeup of the non-crystallographic E2^BilB dimer in the Form 1 structure, oriented equivalently to panel (b). (d) Closeup view of the E2^BilB dimer interface (rotated 90° from panel (b)). Residues involved in the interface from one protomer are colored blue, shown in sticks, and labeled. The same residues from the dimer-related protomer are shown as sticks and colored gray. Mutant 1 and Mutant 2 are two multi-site mutants of E2^BilB used in panels (e)-(f). (e) Size exclusion chromatography coupled to multi-angle light scattering (SEC-MALS) analysis of purified E. aridi E1^BilD (1–507, C417A):E2^BilB (1–181, wild type or mutant):Ubl^BilA (17–97) complexes. Thin lines indicate protein concentration as measured by differential refractive index (dRI; left axis), and thick lines indicate molecular weight in kDa (right axis). Dotted horizontal lines indicate the expected molecular weight of a 1:1:1 complex (85.35 kDa) and a 2:2:2 complex (170.7 kDa). Data for three constructs is shown: wild-type E2 (black/gray), mutant 1 (F36R/I48A/I59K/F83E; light blue/dark blue), and mutant 2 (F36R/I38A/F46K/I48A/I59K/F83E; yellow/brown). (f) SDS-PAGE analysis of purified complexes analyzed by SEC-MALS. For gel source data, see Supplementary Figure 1. This experiment was independently performed three times, with consistent results.

Extended Data Figure 5. Biochemical and structural analysis of E. aridi DUB^BilC.

(a) N-terminal sequencing (Edman degradation) of DUB^BilC-cleaved Ubl^BilA-GFP fusion (C-terminal fragment), showing the evaluated value from each of five cycles of degradation. The inferred N-terminal sequence of the fragment is AGIGS. This analysis was performed once. (b) Closeup view of the DUB^BilC(E33A):Ubl^BilA (Form 1) active site, with proteins colored as in Figure 3d. Active site residues of DUB^BilC and glycine 97 of Ubl^BilA are labeled. (c) View equivalent to panel (b), showing 2Fo-Fc composite omit map density at 1.5 σ. (d) Comparison of E. aridi DUB^BilC:Ubl^BilA (left) to two similar structures, Caldiarchaeum subterraneum Rpn11-homolog bound to ubiquitin-homolog (center)⁶² and Schizosaccharomyces pombe Sst2 bound to a ubiquitin K63-linked ubiquitin (right)⁶³. Overall Cα r.m.s.d. values for DUB^BilC versus its homolog are in Table 1.

Extended Data Figure 6. Identification of Ubl^BilA targets.

(a) SDS-PAGE analysis of Ni²⁺ affinity-purified His₆-Ubl^BilA Δ97 and associated proteins in native conditions, after coexpression with E1^BilD and E2^BilB. Bands representing His₆-Ubl^BilA, E1^BilD, and E2^BilB are marked. E2^BilB WT: wild type; Mutant 1: F36R/I48A/I59K/F83E; Mutant 2: F36R/I38A/F46K/I48A/I59K/F83E. Red asterisk indicates a likely Ubl^BilA-E2^BilB conjugate that is more abundant when E2^BilB is mutated. For gel source data for panels (a)-(d), see Supplementary Figure 1. (b) SDS-PAGE analysis of Ni²⁺ affinity-purified His₆-Ubl^BilA (full-length (FL) or Δ97) and associated proteins in native conditions, after coexpression with E1^BilD (wild type or C417A mutant; indicated as “C-A”), E2^BilB (wild type or C138A mutant; indicated as “C-A”), and DUB^BilC (wild type or E33A mutant). Bands representing His₆-Ubl^BilA, E1^BilD, E2^BilB, and DUB^BilC are marked. (c) Anti-His₆ western blot analysis of the experiment shown in panel (b), showing His₆-Ubl^BilA-target conjugates. (d) SDS-PAGE gel (visualized by Coomassie Blue staining) of His₆-tagged Ubl^BilA Δ97 (wild type or V95K mutant) coexpressed with E1^BilD and E2^BilB, then purified in denaturing conditions. (e) Extracted Ion Chromatograms (EICs) of transition ions of the properly cleaved, unmodified Ubl^BilA(Δ97, V95K) residues 76–92, using a 10 ppm m/z tolerance. RT: retention time.

Extended Data Table 1. Crystallographic data collection and structure determination.

Each dataset was collected from an individual crystal. Values in parentheses are for the highest resolution shell. Accession numbers for final refined coordinates, structure factors, and raw diffraction datasets are provided in the Data Availability statement.

	E. aridi DUBBilC(E33A): UblBilA Form 1	E. aridi DUBBilC(E33A): UblBilA Form 2	E. aridi E1BilD:E2BilB: UblBilA Form 1	E. aridi E1BilD:E2BilB: UblBilA Form 2
Data collection
Space group	I222	F432	P21	P42212
Cell dimensions
a, b, c (Å)	53.34, 64.69, 141.69	214.89, 214.89, 214.89	73.85, 89.66, 131.60	199.93, 199.93, 66.88
α, β, γ (°)	90, 90, 90	90, 90, 90	90, 94.93, 90	90, 90, 90
Resolution (Å)	70.85-1.36(1.38-1.36)	124.07-1.68 (1.71-1.68)	89.66-2.47 (2.53-2.47)	199.93-2.68(2.80-2.68)
R merge	0.058 (0.481)	0.097 (0.956)	0.093 (0.812)	0.101 (3.776)
I/σI	25.0 (4.0)	35.8 (5.2)	11.9 (1.5)	19.2 (0.7)
Completeness (%)	99.9 (98.0)	99.9 (100.0)	98.9 (95.0)	100 (100)
Redundancy	12.7 (12.5)	40.3 (41.1)	3.5 (3.3)	13.7 (14.4)
Refinement
Resolution (Å)	70.85-1.36	50.00-1.68	74.01-2.47	70.68-2.68
No. reflections	53036	48715	60794	38499
Rwork / Rfree	0.1534/0.1799	0.1461/0.1628	0.1963/0.2439	0.2075/0.2446
No. atoms
Protein	3768	3767	20,307	5,452
Ligand/ion	1 (Zn2+)	1 (Zn2+)	2 (Zn2+)	1 (Zn2+)
Water	443	381	129	18
B-factors
Protein	20.90	19.66	69.58	89.39
Ligand/ion	14.37	14.48	57.51	125.06
Water	33.03	36.92	56.99	71.19

Data Availability

Final coordinates and structure factors for all structures have been deposited at the RCSB Protein Data Bank (https://www.rcsb.org) under accession codes 8TYX (DUB^BilC(E33A):Ubl^BilA Form 1), 8TYY (DUB^BilC(E33A):Ubl^BilA Form 2), 8TZ0 (E1^BilD:E2^BilB:Ubl^BilA Form 1), and 8TYZ (E1^BilD:E2^BilB:Ubl^BilA Form 2). Raw diffraction images have been deposited at the SBGrid Data Bank (https://data.sbgrid.org) under accession codes 1039 (DUB^BilC(E33A):Ubl^BilA Form 1), 1040 (DUB^BilC(E33A):Ubl^BilA Form 2), 1041 (E1^BilD:E2^BilB:Ubl^BilA Form 1), and 1042 (E1^BilD:E2^BilB:Ubl^BilA Form 2). Sequence data were downloaded from the IMG database of bacterial genomes (https://img.jgi.doe.gov/). Source data for the graphs shown in Extended Data Figures 4e and 5a accompany this manuscript.