Bacterial usurpation of the OTU deubiquitinase fold

Abstract

The extensive cellular signaling events controlled by posttranslational ubiquitination are tightly regulated through the action of specialized proteases termed deubiquitinases. Among them, the OTU family of deubiquitinases can play very specialized roles in the regulation of discrete subtypes of ubiquitin signals that control specific cellular functions. To exert control over host cellular functions, some pathogenic bacteria have usurped the OTU deubiquitinase fold as a secreted virulence factor that interferes with ubiquitination inside infected cells. Herein, we provide a review of the function of bacterial OTU deubiquitinases during infection, the structural basis for their deubiquitinase activities, and the bioinformatic approaches leading to their identification. Understanding bacterial OTU deubiquitinases holds the potential for discoveries not only in bacterial pathogenesis, but in eukaryotic biology as well.

Introduction:

Ubiquitination is a highly versatile posttranslational modification that regulates all aspects of eukaryotic biology [1–3]. Alterations in the human ubiquitin (Ub) system can lead to the development of distinct diseases [4]. Most cellular proteins can be modified with single Ub molecules (monoubiquitination) or polyubiquitinated by modification with Ub chains, serving distinct signals depending on their linkage types. Polyubiquitin (polyUb) linkages can be formed via any of seven Lys residues or via the N-terminal Met residue (M1) on Ub[5]. Lys48- and Lys63-linked Ub chains are the most common and extensively analyzed; Lys48-linked chains signal for targeted proteasomal degradation and Lys63-linked chains are implicated in non-degradative processes such as the DNA damage response. M1- and K11-linked chains regulate immune signaling and cell cycle progression, respectively, while the remaining “atypical” Lys6-, Lys27-, Lys29-, and Lys33-linked chains are less well defined[6]. Ub-mediated signaling processes are tightly regulated by functions of Ub ligases that assemble Ub modifications [7]. The action of Ub ligases can be counteracted by deubiquitinases (DUBs) which proteolytically recycle Ub back into monomers. The human genome encodes approximately 100 DUBs that can be divided into seven families [8]. The ovarian tumor (OTU) family of DUBs has emerged as a set of regulators for cellular signaling cascades [9], and mechanisms of their polyUb linkage specificity have been well analyzed [10].

Although Ub is almost exclusively encoded by eukaryotes, recent analyses of host-pathogen interactions have revealed bacterial pathogens that have acquired mechanisms to subvert host Ub systems. Many bacterial pathogens encode Ub ligases and DUBs, representing bacterial mimicry of host cellular systems in most cases [11–15], while noncanonical mechanisms for reversible conjugation of Ub to substrates have been identified in Legionella pneumophila [16–18]. OTU DUBs have been found to be encoded in some species of bacteria (Table 1). L. pneumophila, which possesses four OTU DUBs, is an extraordinary pathogen known for its intricate mechanisms to manipulate cellular systems including the Ub system to establish a replicative niche (Fig. 1). Molecular mechanisms of activities and specificities of bacterial OTU DUBs, as well as their roles in bacterial pathogenesis, have been an area of active investigation. In this review, we summarize present knowledge on bacterial OTU DUBs, focusing on their structure and function, as well as on methodology for predicting bacterial OTU domains.

Table 1.

Bacterial OTU DUBs described in this article.

Species	Name	Gene	PDB ID	Accession	References
Escherichia albertii	EschOTU	ESCAB7627_1170	6W9S	EDS93808.1	[43]
Burkholderia ambifaria	BurkOTU	BamIOP4010DRAFT_1320	NA	EDT05193.1	[43]
Chlamydia pneumoniae	ChlaOTU	CPn_0483	NA	AAD18623.1	[23,29,43]
Chlamydia caviae	ChlaOTU	CCA00261	NA	WP_011006230	[23]
Rickettsia massiliae	RickOTU	dnaE2	NA	ABV84894.1	[43]
Wolbachia pipientis	wPipOTU	WP0514	NA	CAQ54622.1	[43]
	wMelOTU	WD_0443	6W9O 6W9R	AAS14166.1	[43,68]
	wMelOTU-2	WD_0633	NA	AAS14332.1	[68]
Legionella pneumophila	LotA/Lem21	lpg2248	7F9X 7W54 7UYG 7UYH	AAU28313.1	[32–36]
	LotB/Ceg23	lpg1621	6KS5	AAU27701.1	[40–44]
	LotC/Lem27	lpg2529	6YK8 7BU0	AAU28589.1	[40,44,55]
	LotD/Ceg7	lpg0227	NA	AAU26334.1	[40,43]
	Lpg2952	lpg2952	NA	AAU28998.1	[40]

Fig. 1. The scheme of LCV biogenesis and functions of the LCV-associated bacterial Ub ligases and DUBs.

Upon infection, L. pneumophila delivers various effector proteins (colored circles) into host cell cytosol to modulate cellular systems. Reversible Ub modification of host proteins is a relevant example of actions of the effector proteins. Just after internalization, the Legionella phagosome is morphologically converted to the ER-like compartment [79–81]. This conversion is thought to be a result of fusion between the phagosome and ER-derived vesicles which are captured by intercepting ER-to-Golgi traffic [47]. At this stage of infection, SidE- and SidC-family effector proteins, having distinctive enzymatic activities as Ub ligases, contribute toward conjugation of Ub (red dots) on LCV-associated substrates. At later stages, the Ub chains can be cleaved by L. pneumophila OTU DUBs, LotA, LotB and LotC, according to their linkage specificities (see text), reducing the level of Ub on the LCV. The function of LotD on the LCV-associated Ub chains has not been analyzed yet. The functional interplay between the bacterial Ub ligases and the DUBs is apparently correlated with the recruitment of key host players in membrane fusion, like Rab GTPases and SNARE proteins, to the LCV. This suggests that the effector-mediated regulation of Ub chain assembly is closely connected to the establishment of the LCV replicative niche, although the exact roles of the bacterial enzymes on this scheme have not been fully elucidated yet. This image was created with BioRender.com.

PART I: Functional aspects of bacterial OTU DUBs

ChlaOTU

Chlamydiae, obligate intracellular bacteria, are known to represent ancient lineages associated with eukaryotes [19]. Phylogenomic analyses revealed that Ub-related gene families are massively expanded in the phylum Chlamydiae [20]. Interestingly, F-box and BTB-box proteins, which are largely distributed among the phylum, are not identified in the family Chlamydiaceae. However, there are several proteins functioning as DUBs, such as the Chlamydia trachomatis ChlaDub1 (CT868) and ChlaDub2 (CT867) [21,22] and the Chlamydia caviae ChlaOTU [23] within this family. Unlike OTU DUBs that belong to the CA clan of proteases, ChlaDub1 and ChlaDub2, which were more recently renamed Cdu1 and Cdu2 [24], are CE clan proteases [25]. These proteins are conserved in most Chlamydia species and are indeed recognized as the first bacterial DUBs whose deubiquitinating and deneddylating activities were demonstrated by suicide probe-based analyses [23]. Cdu1 and Cdu2 both preferentially target Lys63-linked polyUb signals [25,26], while Cdu1 exclusively has additional acetyltransferase activity [27]. It has been reported that the DUB activity of C. trachomatis Cdu1 massively removes Ub from the chlamydial inclusion in infected cells while simultaneously acting to fragment the host Golgi apparatus [27,28]. Cdu1, but not Cdu2, can stabilize inclusion-associated Mcl-1 which is an anti-apoptotic Bcl-2 family member [24]. Cdu1 plays an important role in supporting intracellular growth of C. trachomatis [27,28].

ChlaOTU (Cpn_0483) of Chlamydia pneumoniae was first reported in the year 2000 alongside a group of predicted cysteine proteases that constitute the OTU family [29]. However, the experimental validation of its enzymatic activity was not reported until 2013 [23]. Utilizing the guinea pig adapted species Chlamydia caviae, Frutado et al. analyzed the actin polymerization-triggering bacterial entry and observed only transient accumulation of Ub at Chlamydia entry sites [23]. The authors reasoned that C. caviae encodes a putative DUB responsible for the rapid disappearance of Ub signals at the entry sites. They identified C. caviae ChlaOTU (CCA00261), which is an orthologue to Cpn_0483 and a substrate of the bacterial type III secretion system, as the responsible DUB. They observed that not only Ub but also autophagy adaptor nuclear dot protein 52 (NDP52) is transiently recruited to the bacterial entry sites and can be targeted by ChlaOTU. However, the biological implications of the observed activities of ChlaOTUs remain to be elucidated. CCA00261 was also identified in Chlamydia abortus [30]. Interestingly, C. pneumoniae encodes ChlaOTU but neither Cdu1nor Cdu2, which are present in all Chlamydiaceae species except for C. pneumoniae [30,31].

LotA/Lem21/Lpg2248

So far most bacterial OTU DUBs have been found in L. pneumophila. LotA/Lpg2248 (also known as Lem21) is the first example of the Legionella OTU-like (LOT) proteins among the translocation substrates of the bacterial type IV secretion system (T4SS). Bioinformatic analysis revealed that LotA possesses an unusual architecture with tandemly aligned OTU domains in which two distinctive catalytic cysteines, Cys13 and Cys303, were present [32]. Cellular expression of LotA extensively reduced the level of polyUb in a manner dependent on Cys303, while in vitro analysis against diUb substrates showed only subtle activity except against the Lys6-linked substrate [32]. The stringent specificity against Lys6-linked diUb has been shown to rely on Cys13 [32–35]. The poor activity of Cys303 against diUb substrates suggested that the OTU domain has a unique feature to recognize only longer polyUb chains including Lys48 and Lys63-linked chains, a property which was later demonstrated [33,36].

LotA is delivered to the cytosol of infected host cells and localizes on the Legionella-containing vacuole (LCV) (Fig. 1). The localization is mediated by the C-terminal lipid binding domain and is crucial for its activity to remove Ub signals from the surface of the LCV [32]. Interestingly, an analysis using mouse macrophages indicated that combined disruptions of LotA with SidE family proteins, which have activities as noncanonical Ub ligases [37], diminished the intracellular growth of L. pneumophila more severely than sole disruption of SidE family proteins [32]. This observation suggests a possible functional connection between the SidE Ub ligases and LotA DUB activity.

Unveiling the physiological role of the C13-mediated specific cleavage of Lys6-linked Ub chains would be particularly interesting. DUBs with stringent reactivity against Lys6 polyUb had not been reported neither from bacteria nor from humans regardless of the known importance of Lys6 polyUb as signals for regulation of cellular systems including mitochondrial homeostasis [38,39]. The LCV-associated Lys6 Ub signals were found to be diminished in the presence of bacterially-delivered wild-type LotA but not of LotA C13S mutant [33], suggesting potential roles of the C13-dependent DUB activity of LotA in unknown events related to LCV biogenesis or host immune control. Concomitant with increased abundance of K6 polyUb, an increased localization of the AAA ATPase VCP/p97/Cdc48 was also observed [33]. This suggests that LotA’s unique dual DUB activities may act to guard certain proteins on the LCV from VCP-dependent extraction and subsequent proteasomal degradation.

LotB/Ceg23/Lpg1621

LotB/Ceg23/Lpg1621 is a second identified member of the OTU DUB family in the L. pneumophila T4SS substrates. The catalytic activity of LotB was independently reported by several groups in 2020 [40–44]. It exhibits DUB activity with high specificity toward Lys63-linked chains of diUb and pentaUb in in vitro reactions. LotB was shown to be modified not only with propargyl Ub (Ub-PA) but also with propargyl NEDD8 (NEDD8-PA), suggesting that LotB can bind to both Ub and NEDD8 through the conserved Ile44 hydrophobic patch [40,44]. A fluorescence-based cleavage assay suggested that LotB may also demonstrate proteolytic activity against small Ub-related modifier 1 (SUMO-1) [43].

When ectopically expressed in human cells, LotB showed specific co-localization with an endoplasmic reticulum (ER)-marker Calnexin [44] and with ER-Golgi shuttle protein KDEL [41], but not with mitochondria and Golgi markers [44]. Mass spectrometry (MS)-based proteomic analysis utilizing the catalytically inactive LotB showed that some ER-resident proteins, as well as subunits of many membrane protein complexes like ATP synthase and the Sec61 translocon, are potential LotB interactors [44]. This analysis was further conducted in infection conditions, and some bacterial proteins were found to interact with LotB. However, it has not been fully validated whether these proteins are genuine interactors or not.

LotB possesses C-terminal transmembrane (TM) domains which are not essential for its catalytic activity per se. It was shown that the TM domains are crucial for its specific cellular localization to the ER when ectopically expressed [41,42,44]. The MS analyses identified COPI coatomer as an interactor of ectopically expressed LotB [41,44]. Immunofluorescence analysis further revealed that the coatomer protein β՛-COP co-localized with LotB in cells [41]. The COPI system is known to be involved mainly in the retrograde transport in the ER-Golgi segment of the secretory pathway [45,46]. The ER-derived vesicles carried by COPI- and COPII-mediated transport pathways are thought to be used for biogenesis of the LCV [47] (Fig. 1). The implication is that the function of LotB may be involved in regulation of vesicle transport. Suggestively, ectopic expression of LotB affected the early secretory pathway dependent on its TM domains which were shown to be essential for interacting with the COPI coat proteins [41].

In infected cells, LotB associates with the LCV and regulates the level of Lys63-linked polyUb on the vacuole [42] (Fig. 1). Under physiological conditions, Sec22b, a LCV-localized v-SNARE [48,49] was identified as the protein targeted by the catalytic activity of LotB [41] (Fig. 2A). In this study, it was shown that Sec22b can be modified with polyUb as early as 0.5 h after infection of HEK293T cells in a manner dependent on the T4SS. Sec22b ordinarily resides on ER-derived vesicles [50,51]. Upon infection, ER vesicles can associate with the plasma membrane (PM)-derived bacterial phagosome in the initial process of the LCV formation [52]. This event can provide Sec22b with a chance to be unconventionally paired with the PM-derived t-SNAREs including syntaxin3 (Stx3) [53,54]. In the later stages of infection, LotB mediates cleavage of the Lys63-linked polyUb chains on Sec22b, which results in dissociation of Stx3 from Sec22b on the LCV [41] (Fig. 2A). Thus, LotB plays a role in liberating SNARE proteins from the infection-induced noncanonical pairing, presumably to restore the cognate paring.

Fig. 2. Manipulation of host proteins by the functions of LotB and LotC.

(A) Upon internalization, L. pneumophila acquires plasma membrane-localized t-SNAREs including Stx3 on its early phagosome. It then promotes the recruitment of ER-derived vesicles to the LCV by intercepting vesicle traffic. This results in noncanonical pairing of ER-derived v-SNARE Sec22b with Stx3. L. pneumophila also induces polyubiquitination of Sec22b in a T4SS-dependent manner at the initial stage of infection. LotB is delivered to the cell cytosol via the T4SS and localizes to the LCV depending on its TM domains. The polyUb chains on Sec22b are cleaved by the DUB activity of LotB, leading to the dissociation of Stx3 from Sec22b residing on the LCV. (B) The Legionella E3 Ub ligase SidC induces Rab10 ubiquitination and promotes the recruitment of Rab10 to the LCV. The overexpression of LotC in bacteria leads to reduction in ubiquitinated Rab10 as well as the level of Rab10 on LCVs, demonstrating that LotC can reverse the activity of SidC on the LCV. These images were created with BioRender.com.

LotC/Lem27/Lpg2529

LotC/Lpg2529 (also known as Lem27) was identified as another L. pneumophila OTU DUB by bioinformatic analyses [40,44,55]. The enzymatic activity of LotC was characterized in comparison with that of LotB [44,55]. LotC displays a broader preference for polyUb linkages; it can cleave Lys6-, Lys11-, Lys27-, Lys33- and Lys48-linked diUb [44,55]. Activity-based probe assays showed that LotC was not modified with any examined Ub-like modifier probes including NEDD8-PA [44]. Sequence alignments suggested a possible catalytic activity toward ISG15 [55], although the activity has not been experimentally validated.

In contrast to LotB, LotC did not show specific cellular localization when ectopically expressed [44]. Upon infection, LotC is delivered via the T4SS and localizes to the LCV [55]. The level of Ub associated with the LCV was decreased depending on the catalytic activity of LotC [55] (Fig. 1). The Ub accumulation on the LCV has been thought to be mediated largely by the activities of the LCV-localized SidC and SdcA [56,57]. It was reported that the SidC-family E3 ligases ubiquitinate the small GTPase Rab10 which was shown to play an important role in intracellular replication of L. pneumophila [58] (Fig. 2B). Infection with a L. pneumophila strain overexpressing LotC resulted in reduction of the SidC-induced polyubiquitination of Rab10, demonstrating that the DUB activity of LotC can counteract the activity of the Legionella E3 ligases [55]. The level of Rab10 ubiquitination correlated with Rab10 localization to the LCV [55], suggesting that deubiquitination of Rab10 by LotC can dissociate Rab10 from the vacuole (Fig. 2B). These results led to the implication of the possible interplay of the Legionella E3 ligases and the LotC DUB to modulate remodeling of the bacterial phagosome, in which Rab10 is involved. It was shown, however, that LotC is dispensable for bacterial replication in macrophages [55].

LotD/Ceg7/Lpg0227

Prediction of Legionella DUBs utilizing the MEROPS database [59] identified another member of the Legionella OTU family, Lpg0227 (also known as Ceg7) [40,43]. Here, we propose to designate Lpg0027 as LotD. LotD is closely related to LotA [60] and is likely to have the similar catalytic activity [40]. Enzymatic activity of LotD was experimentally demonstrated [43]. LotD showed robust reactivity with the Ub-PA, but not with Ub-like modifier probes. It has a basal preference for Lys6-, Lys11-, Lys48-, and Lys63-linked diUb chains [43]. The diUb cleavage assay showed additional activity of LotD toward Lys33- and M1-linked chains, while RavD was reported as a sole L. pneumophila DUB which specifically hydrolyses M1-linked polyUb chains [61]. Physiological roles of LotD in eukaryotic cells has not been examined yet.

Lpg2952

With the aim of discovering new DUBs, a comprehensive bioinformatic analysis of evolutionary relationships between cysteine protease families including Legionella effector proteins was conducted through hidden Markov model profiling of the MEROPS database [40]. In this analysis, Lpg2952 was identified as a CA-clan Legionella OTU. Because the L. pneumophila Lpg2952 family member has lost its active site, this protein has not been experimentally validated. As other Legionella species also encode Lpg2952 orthologues, these proteins may have a physiological role in specific host organisms or circumstances.

Other bacterial OTU DUBs

In addition to ChlaOTUs and LOT-class OTUs, characterization of the biochemical properties of EschOTU from Escherichia albertii, wMelOTU and wPipOTU from Wolbachia pipientis, BurkOTU from Burkholderia ambifaria, and RickOTU from Rickettsia massiliae have been conducted [43] (Table 1). It was found that these bacterial OTUs demonstrate a range of catalytic efficiencies toward Ub substrates and exhibited differing preferences among K6-, K11-, K48-, and K63-linked polyUb substrates. However, physiological roles of these OTU DUBs have not been reported yet.

Perspectives on bacterial OTU functions

The Ub system has been established in eukaryotic cells for regulating and maintaining cellular functions. Bacterial pathogens have evolved sophisticated systems utilizing the host Ub system partly for counteracting antimicrobial immunity, including degradation pathways such as selective autophagy [62,63]. Studies on L. pneumophila have revealed that the Ub system is also hijacked to synthesize the specialized LCV compartment as a replicative niche, in which a vast array of host proteins is involved. Bacterial OTU DUBs have just emerged by recent studies, particularly on L. pneumophila effector proteins. Exact functions for many of them are still shrouded in mystery. Implications of OTU DUBs in bacterial virulence have not been clarified yet, as deletion of single OTU genes did not cause detectable intracellular growth defects in laboratory infection models [32,41,42]. If bacterial OTUs are identified as key virulence factors for bacterial pathogenesis, their sequence and structure differences from human OTUs (see below), may make them feasible targets for therapeutic intervention. Many of these OTUs commonly display broad polyUb linkage specificity, including “atypical” chain types that are not well-defined even outside the context of bacterial infection. It is possible – or even likely – that bacterial OTU DUBs identified so far are just the tip of the iceberg. Future studies on bacterial OTU DUBs will therefore shed light not only on bacterial pathogenicity but also on eukaryotic biology.

PART II: Structural studies of bacterial OTU DUBs

Introduction to the OTU fold

The OTU fold represents a structurally-distinct family within the papain-like CA protease clan [64]. OTU DUBs are a highly versatile family that can target a range of ubiquitinated substrates with varying levels of activity and specificity. A wide spectrum of specificities toward distinct polyUb substrates is observed within the 16 human OTU DUBs alone [10]; it therefore should come as no surprise that bacterial OTU domains also demonstrate finely tuned activities. As for other DUB families, proteolytic activity toward Ub is directed by an S1 recognition site that binds and orients the Ub C-terminus into the active site (Fig. 3A). Regions of substructure making up the OTU domain S1 site were recently reclassified as three distinct variable regions (VR1-3) by Schubert et al. as part of an effort to categorize the many adaptations observed in eukaryotic, viral, and bacterial OTUs [43]. Specificity toward polyUb chains can be directed through a variety of mechanisms [65], but generally requires coordination of a second Ub moiety in a linkage-specific manner, often at the S1’ recognition site that orients the Ub-modified amino group into the active site (Fig. 3A). In line with their range of activities, specificities, and evolutionary trajectories, bacterial OTU domains exhibit a gamut of unique structural adaptations that support DUB activity.

Fig. 3. Structural comparison of bacterial S1 Ub-binding sites.

(A) Schematic of Ub-binding regions mapped onto the structure of CCHFV vOTU (PDB 3PHW [82]) as a prototypical OTU fold. Catalytic triad residues forming the active site are shown in ball-and-stick. The S1 Ub-binding site is composed of variable regions (VR) highlighted in blue. (B) As in (A), for the structure of Escherichia albertii EschOTU (PDB 6W9S [43]). Variable regions and a VR-1 sequence permutation are annotated. (C) As in (A), for the AlphaFold2 model of Chlamydia pneumoniae ChlaOTU [66]. Variable regions and a large VR-1 insertion domain are annotated. (D) As in (A), for the structure of Wolbachia pipientis wMelOTU (PDB 6W9R [43]). Variable regions are annotated. All structure figures were generated using PyMOL (www.pymol.org).

EschOTU

The structure of EschOTU from E. albertii has been resolved in complex with a Ub-PA activity-based probe bound at its active site [43]. The structure confirmed an unusual topological arrangement of the catalytic triad residues. OTU domains and other members of the CA protease clan, by original definition, encode their catalytic cysteine residues ahead of the general base histidine by primary sequence. In such an arrangement, the catalytic cysteine near the N-terminus and the histidine nearer to the C-terminus are brought together in 3-dimensional space by the OTU fold. In contrast, EschOTU encodes a reversed topological arrangement of the catalytic cysteine and general base histidine that is more reminiscent of members from the CE protease clan (Fig. 3B) [43]. Despite the altered topology, the OTU fold of EschOTU is preserved as a result of a permutation of the N- and C-termini into what is typically a loop region in VR-1 of the S1 site. This region of the OTU fold appears to be particularly malleable, as it can accommodate a permutation in sequence as well as structural alterations and insertions [43].

ChlaOTU

The AlphaFold2 model for C. pneumoniae ChlaOTU predicts an unexpected OTU structure that could easily explain discrepancies in the literature surrounding its DUB activity [66]. Though ChlaOTU was the first bacterial OTU to be predicted by sequence, identifying the construct boundaries for the OTU domain (typically flanking the catalytic triad residues) was not straightforward because a general base histidine could not be identified. Thus, while a full length ChlaOTU construct was reported to exhibit DUB activity, an OTU-sized (~300 amino acids) construct starting near the predicted catalytic cysteine had no detectable activity [23,43]. Remarkably, the AlphaFold2 model of ChlaOTU predicts an ~600 amino acid insertion into VR-1 that adopts a large helical fold with no recognizable structure homology (Fig 3C) [66,67]. Owing to the sheer size of this VR-1 insertion, the predicted catalytic cysteine and general base histidine residues of ChlaOTU are separated by ~750 amino acids of primary sequence, thereby necessitating a near-full length construct in order to encode a complete OTU fold. Whether and how the large insertion domain predicted by AlphaFold2 impacts the function of ChlaOTU DUB activity will be an interesting area of future research.

wMelOTU

The genome analysis of W. pipientis (wMel), an endosymbiont of Drosophila melanogaster, showed that proteins encoded in the prophage regions WD0443 and WD0633 were found to have OTU-like protease domains [68].

Crystal structures determined by Schubert et al. of the WD0443 (renamed as wMelOTU) alone and bound to Ub-PA revealed several interesting features [43]. Firstly, as observed in a number of other DUBs and Ub-like proteases, the wMelOTU active site is misaligned in the apo structure but in the active conformation while in complex with Ub. In this case, the wMelOTU His-loop (i.e., the loop preceding the general base histidine) is shifted down in the apo structure, resulting in both the misalignment of the catalytic triad as well as blockage of the entry tunnel into the active site. Ub binding into the S1 site appears to select for an active conformation of the His-loop, thus alleviating blockage of the entry tunnel and aligning the catalytic triad for proteolysis. Concomitantly, wMelOTU VR-1 and VR-3 also undergo structural rearrangements upon Ub recognition. In the apo structure, regions of both VR-1 and VR-3 are absent from the electron density, consistent with a high degree of conformational heterogeneity. Upon Ub binding, these regions form β-hairpin structures, acting as molecular arms that embrace the Ub molecule through interactions with the Ub Ile44 and Ile36 hydrophobic patches as well as to each other (Fig. 3D). Along with additional contacts from VR-2, wMelOTU has adopted a highly intricate S1 recognition site that interacts with a large portion of the Ub surface.

LOT-class OTUs

L. pneumophila encodes five LOT-class OTU domains: LotB, LotC, LotD, and two within LotA. With the exception of LotD, all LOT-class DUBs have been structurally characterized [33–36,42,44,55]. An AlphaFold2 model is now available for LotD that allows comparisons to be made across the entire class of DUBs (Fig. 4A) [66]. Interestingly, only the first OTU domain of LotA (denoted LotA_N) encodes the catalytic triad within the typical C…H-Ω-D/N/E arrangement (where Ω represents a large aromatic residue). The remaining LOT-class DUBs LotB, LotC, LotD, and the second OTU domain of LotA (denoted LotA_M) all encode an arrangement observed in the A20 class of human OTUs, in which the acidic member of the catalytic triad is encoded N-terminal to the catalytic cysteine and positioned adjacent to the general base histidine on a separate β-strand. Consistent with their ability to cleave monomeric Ub substrates, the structures of LotB and LotC, as well as the AlphaFold2 model of LotD, all show a competent arrangement of the catalytic triad that is primed for hydrolysis (Fig. 4B) [42–44,55]. In contrast, neither LotA_N or LotA_M are capable of cleaving monomeric Ub substrates and their structures show an incompetent arrangement of the catalytic triad, indicating a requirement for some form of activation upon polyUb binding (Fig. 4B) [33,36]. In the case of LotA_N, this activation mechanism was recently resolved through a structure in complex with Lys6-linked diUb [33]. The complex structure showed that the bulky Phe4 and His68 residues of the proximal, Lys6-linked Ub bound at the LotA_N S1’ site act to select for catalytically-competent conformations of LotA_N loops containing the catalytic cysteine and general base histidine [33]. Thus, only a Lys6-linked polyUb chain is capable of correctly orienting the LotA_N catalytic triad for activity. Such a mechanism of “substrate-assisted catalysis” has also been shown to underlie the M1 polyUb specificity of the human OTU DUB, OTULIN [69].

Fig. 4. Structural analysis of Legionella pneumophila LOT-class DUBs.

(A) Structures of all LOT-class DUBs. Active sites within the core OTU domain are annotated, along with adaptive α-1,2 regions of the A-UBDs inserted into VR-1. (B) Active site views of all LOT-class DUBs. Catalytic triad residues are shown in ball-and-stick, with hydrogen bonds indicated by dashed lines. Residues missing from the crystal structure (either due to insufficient electron density or construct boundaries) are indicated by asterisks. Configurations consistent with active or inactive catalytic triads are indicated. (C) A-UBD structures for all LOT-class DUBs. Topologies of the 4-helix A-UBD sub-structure are indicated with numbered arrows. The distinct α-1,2 regions responsible for Ub binding are highlighted. All structure figures were generated using PyMOL (www.pymol.org).

A defining feature of the LOT class of OTU domains is the insertion of a ~180 residue helical domain within VR-1. Despite their low sequence homology, ranging from 12–27% pairwise identity, all LOT-class insertion domains follow a similar topological arrangement of four α-helices (Fig. 4C). Within this 4-helix fold, the largest divergence in structure is located in the α1–2 region, between helices 1 and 2 (Fig. 4C). In structure studies of LotA_N and LotC, as well as modeling studies of LotB and LotA_M, this α1–2 region makes important contacts to the bound Ub [33,36,42,44,55]. In the case of LotA_N, it even makes key contacts to both Ub moieties of the Lys6-linked diUb [33]. Flexibility between the 4-helix insertion and the OTU domain is also an important component of how Lys6-linked diUb is staged for hydrolysis by LotA_N, and the same may be true for other LOT-class DUBs as well. A DALI structure similarity search identified a related 4-helix, Ub-binding domain within the Orientia tsutsugamushi DUB OtDUB, where it mediates a high affinity interaction with Ub through a unique α1–2 region [70]. For its ability to adopt distinct Ub-binding modes through unique α1–2 substructures, this 4-helix domain was coined an “Adaptive Ub-Binding Domain”, or A-UBD [33]. Whether and how this A-UBD has been integrated into other proteins outside of OtDUB and the LOT-class DUBs remains to be seen.

Perspectives on bacterial OTU structures

Despite the comparatively low number of bacterial OTU structures characterized thus far, a remarkable diversity in structure and mechanism is already evident. The framework of OTU variable regions established by Schubert et al. helps categorize these unique adaptations observed among bacterial OTUs [43]. In particular, bacterial OTUs described thus far all exploit drastic changes within VR-1 to fine tune DUB function, including both small (e.g., wMelOTU) and large insertions (e.g., ChlaOTU and LOT-class OTUs), and even permutation of the OTU fold (e.g., EschOTU). Future work on bacterial OTUs will undoubtedly reveal even more mechanisms for regulating and fine-tuning OTU DUB activities. By establishing common themes of regulation and Ub recognition (e.g., the A-UBD), discoveries made among bacterial OTU structures can be translated into advancing our understanding of Ub biology as a whole.

PART III: Prediction of novel bacterial OTU domains

Initial discovery of a bacterial OTU in Chlamydia pneumoniae

The OTU gene was first described in D. melanogaster in the context of oogenesis [71]. To discover related orthologues across species and identify a function for the OTU domain, a BLAST query into the NCBI non-redundant protein database was used to identify related genes in humans and Caenorhabditis elegans [29]. Additional OTU orthologues were found in plants, fungi, and viruses, which allowed for further BLAST queries that led to the discovery of an OTU domain in the bacterium C. pneumoniae [29]. Sequence alignments of the newly identified OTU family showed strong conservation of a catalytic dyad consisting of a cysteine and histidine, indicating potential cysteine protease activity that was later shown to target Ub [72].

Expansion of the bacterial OTU family

The catalytic dyad and other sequence conservation amongst known OTU orthologues informed additional bioinformatic learning approaches for the prediction of novel bacterial OTU domains. The strong sequence conservation allowed the application of hidden Markov models (HMMs), which use statistical modeling to generate sequence profiles that enable the identification of distantly-related proteins [73]. An HMM approach was first used in the discovery and validation of the tandem OTU domains of LotA from L. pneumophila [32]. Other studies used iterative HMM curation and sequence alignment validation to expand the OTU domains encoded by L. pneumophila to include LotB, LotC, and LotD [40–44,55]. This approach also identified OTU domains within E. albertii, B. ambifaria, R. massiliae, and W. pipientis [43].

Limitations of HMM approaches to identifying bacterial OTUs

The majority of bacterial OTU domain discoveries have arisen from sequence alignment and strong statistical similarity to OTU family motifs such as high conservation of the catalytic residues. However, HMMs typically perform poorly when generalizing across drastic divergences in sequence such as permutations. The OTU family sequence profile is further complicated by the ability of the acidic component of the catalytic triad to be encoded on either the same or opposing β-strand as the general base histidine – a difference in at least 150 amino acids by primary sequence – likely explaining why only a catalytic dyad was first recognized by sequence conservation. E. albertii EschOTU exhibits an extreme deviation from the OTU sequence profile, in which a permutation in the OTU fold inverts the relative positions of the catalytic triad residues in primary sequence [43]. HMMs have proven to be a powerful approach in the identification of bacterial OTU DUBs thus far, but extreme deviations from the OTU sequence profile that arise through either divergent or convergent evolution will be difficult or impossible to detect using this method.

Perspectives on the future of bacterial OTU prediction

Future predictions of bacterial OTU DUBs could benefit from machine learning approaches, as have been applied to predicting bacterial E3 Ub ligases [74]. This approach utilized feature engineering from the protein sequence based on amino acid physiochemical properties. A support vector machine (SVM) was used to train a model that distinguishes non-ligases from known bacterial ligases with a receiver-operator characteristic area under the curve (AUC) of 0.90 [74]. Unlike an HMM-based approach, this machine learning method allows for drastic sequence divergences such as extremely low sequence homology and even permutations. Approaches such as these could be better suited for the discovery of distantly-related examples of bacterial OTU DUBs.

Current bioinformatic approaches for predicting bacterial OTU DUBs rely heavily on homology of the protein sequence, which can exhibit extreme variability or even permutation while retaining the same 3-dimensional fold. An alternative approach from identifying bacterial OTUs by sequence would be to functionally identify them using Ub activity-based probes [75]. This approach has been fruitful in the past and allowed the identification of several non-OTU bacterial DUBs [22,76], but current technologies prevent the selective enrichment of only OTU family members. The recent development of AlphaFold and RoseTTAFold, which generate highly-accurate protein structure predictions based on primary sequence [77,78], could change the future of protein classification and allow for targeted identification of bacterial OTUs through structure. The ability to generate structural models with AlphaFold and RoseTTAFold unlocks the ability to use tertiary structure to train machine learning approaches rather than protein sequence. This could improve predictions by focusing on the OTU fold and local structure of the catalytic triad, thereby alleviating the sole reliance on sequence conservation and focusing more on structure and function. Incorporation of these protein modeling breakthroughs into modern bioinformatic approaches will undoubtedly revolutionize bacterial OTU prediction as well as protein classification as a whole.

Abbreviations:

Ub

ubiquitin

DUB

deubiquitinase

OTU

ovarian tumor

NDP

nuclear dot protein

LOT

Legionella OTU-like

LCV

Legionella-containing vacuole

PA

propargyl

SUMO

small Ub-related modifier

ER

endoplasmic reticulum

MS

Mass spectrometry

TM

transmembrane

PM

plasma membrane

Stx

syntaxin

VR

variable region

HMM

hidden Markov model

SVM

support vector machine

AUC

area under the curve