A plethora of ubiquitin ligases determine the intracellular location and fate of numerous proteins in a substrate-specific manner. However, the mechanisms for these functions are incompletely understood. Most ligases have structurally related RING domains that are critical for ligase activity including the recruitment of ubiquitin conjugating enzymes. Here we probe the function of the RING-CH domain of murine γ-herpesvirus-68 ligase mK3 that functions as an immune evasin by targeting major histocompatibility complex (MHC) class I heavy chains for endoplasmic reticulum-associated degradation (ERAD). Interestingly, mK3 mediates ubiquitin conjugation via ester bonds to S or T residues in addition to conventional isopeptide linkages to K residues. To determine the mechanism of non-K ubiquitination of substrates, we introduced into an mK3 background the RING-CH domains of related viral and cellular MARCH (membrane associated RING-CH) ligases. We found that although a conserved W present in all viral RING-CH domains is critical for mK3 function, sequences outside the RING-CH domain determine whether and which non-lysine substrate residues can be ubiquitinated by mK3. Our findings support the model that viral ligases have evolved a highly effective strategy to optimally orient their RING domain with substrate allowing them to ubiquitinate non-K residues.
Ubiquitin (Ub) conjugation of proteins is one of the most pervasive post-translational modifications, and subtle differences in the extent and linkage of the ubiquitination can have profound effects on the intracellular location and ultimate fate of proteins (1–3). However, the basic process by which proteins are ubiquitinated is highly conserved in evolution and involves the catalytic cascade of three classes of proteins. More specifically, the Ub activating enzyme (E1) forms an ATP-dependent thiolester bond between a cysteine residue in its active site and a C-terminal glycine of an Ub subunit. The Ub is then transferred from the E1 to a cysteine residue of an Ub conjugating enzyme (E2). Finally, an E3 Ub ligase (E3) facilitates the transfer of Ub typically to a ε-NH2 group of a lysine residue of the substrate thus making an isopeptide bond (4). The E3 ligases receive considerable attention because of the fact that they are credited with conferring substrate specificity. There are hundreds of mammalian E3 ligases, most of which contain RING domains defined by characteristically spaced conserved cysteine and histidine residues that form a zinc binding cross-brace structure (5). RING-type E3s are non-enzymatic, but are thought to transfer the Ub from a Ub-charged E2 to their substrate by bringing them into close proximity (6). Importantly, the RING domain of the E3 can influence which E2 is recruited among the more than 30 E2s in mammals. However, the details determining which substrate residues are ubiquitinated and the selective roles of the E3, E2 and other host proteins in this process remain largely undefined.
Recently characterized viral E3 ligases that function in immune evasion are providing several insights into molecular interactions of Ub components and their mammalian substrates. More specifically, the mK3 ligase of γHV68 and the highly homologous kK3 and kK5 ligases (also known as MIR1 and MIR2, respectively) of Kaposi's sarcoma-associated herpesvirus (KSHV) have a variant RING domain termed RING-CH (C4HC3) distinguished by the order of their metal binding cysteine and histidine residues. Furthermore all three of these viral ligases have similar membrane topologies with a RING-CH containing N-terminal domain, two transmembrane domains connected by about a 12 amino acid sequence and a variable C-terminal domain. In addition, all three of these viral ligases include major histocompatibility complex (MHC) class I heavy chains (HCs) as substrates (7–9). Strikingly, all three ligases were recently shown to have the unexpected ability to ubiquitinate non-lysine residues as well as lysine residues on their HC substrates (10–12). What is perhaps most intriguing are the differences between these three viral ligases. kK3 was the first ligase reported capable of ubiquitinating cysteine residues of its substrate, using a thiolester bond analogous to the one used by E1 and E2 Ub components (10). More recent studies suggest that kK5 can also ubiquitinate cysteine residues (11). By contrast, mK3 was the first ligase reported to be capable of ubiquitinating serine or threonine residues on its HC substrate via ester bonds (12).
The kK3/kK5 ligases also differ from mK3 in the substrate degradation pathways they induce. Whereas the kK3 and kK5 ligases induce rapid plasma membrane internalization with substrate degradation in the lysosome, mK3 ligase induces endoplasmic reticulum (ER) dislocation to the cytosol with substrate degradation by the proteasome (7,13–16). In addition, kK3 and particularly kK5 target multiple immune regulatory molecules for degradation including HCs, whereas mK3 ligase predominantly targets only HC for degradation (9,16–20). Related to the above differences, mK3 uniquely binds to the transporter associated with antigen processing (TAP), an ATP-dependent peptide pump that transports MHC I ligands from the cytosol into the ER lumen (21). Indeed, TAP has been proposed to function as an adaptor protein for mK3 orienting it such that it only facilitates the ubiquitination of HC (22). By contrast the molecular basis of substrate discrimination by kK3 and kK5 has not been resolved, although sequences in the HC transmembrane domain and cytoplasmic tail have been implicated (23). In any case, defined differences among these highly homologous viral ligases make them particularly efficacious probes for defining the molecular interactions of host proteins required for the ubiquitination of substrate-specific residues, and the biological consequences of this modification.
Subsequent to the discoveries of the viral ligases, cellular proteins containing a RING-CH and putative transmembrane domains were identified. These proteins are typically called MARCH proteins, an acronym for membrane-associated RING-CH proteins. Ten MARCH proteins were discovered in humans with homologs in mouse (24,25). Although still in earlier stages of characterization, it is already known that many cellular MARCH proteins have diverse and important physiological functions. For example, MARCH I and MARCH VIII proteins regulate CD4 T-cell responses by controlling the level of expression of immune molecules MHC class II and B7.2 on antigen presenting cells (24,26–29). By contrast, MARCH IV and MARCH IX proteins potentially regulate CD8 T-cell responses by controlling the level of expression of MHC class I and intercellular adhesion molecule (ICAM) (24). These four MARCH proteins thus regulate expression of immune molecules similar to the viral MARCH proteins, kK3, kK5 and mK3 making it highly likely the virus stole a progenitor MARCH protein from its host and turned it into an immune evasion protein. In striking contrast MARCH V and MARCH XI appear to be tissue-specific ligases controlling mitochondrial morphology or spermatogenesis, respectively (25,30,31). And MARCH II and MARCH III localize to endosomal vesicles and appear to regulate their recycling by controlling the level of expression of epithelial polarity determinants (24,32,33). Although clearly of considerable interest, much is unknown concerning how MARCH proteins specifically identify and ubiquitinate their physiological targets.
A central question to understanding the function of viral and cellular MARCH proteins is what specifically does the RING-CH domain do? For example, is the RING-CH domain required for the ubiquitination of non-lysine resides? Or more specifically, do subtle differences between the RING-CH domains of mK3 and kK3/kK5 ligases determine whether ester versus thiolester bonds are formed with the substrate. Alternatively, do the RING-CH domains of the cellular MARCH protein also have the capability to ubiquitinate non-lysine residues? The answers to these questions provide key insights into the molecular basis of non-lysine ubiquitination with fundamental implication on how E3s interact with E2s to ubiquitinate specific substrate residues.
Preferential ubiquitination of C-terminal S, T or K residues on HC tails by mK3 ligase
Because mK3 is promiscuous in that it can conjugate S, T or K residues, we questioned whether there were differential positional constraints. To assess the influence of target residue position within the tail of HC on mK3s ability to facilitate its ubiquitination, we engineered single S and K add back mutants on an Ld KCST-less tail background at transmembrane (313) and C-terminal proximal positions (329/337). The Ld KCST-less tail mutant was derived after mutagenesis studies showed that removal of all the K residues or all the K and C residues from the cytoplasmic tail of Ld was insufficient to ablate mK3-mediated ubiquitination (12). As shown in Figure 1A, the positions and conserved amino acid substitutions utilized to generate the Ld KCST-less tail molecule are indicated in red and the single add back S and K residues proximal to the transmembrane or C-termini are designated in blue. These Ld mutants were found to have normal surface expression and TAP binding (data not shown). Lysates from WT3 cells stably co-expressing Ld molecules and mK3 were immunoprecipitated with anti-Ld after which half of each sample was treated with endoglycosidase H (Endo H) to specifically identify ER-resident HCs. Subsequently, all samples were blotted with the anti-Ub antibody P4D1, whose specificity has been well established in the literature and performs identically to anti-hemagglutinin antibody in our assays utilizing HA-tagged Ub (20,34–36). Interestingly, the ability of individually added back S residues to restore mK3-induced polyubiquitination occurred exclusively at position 329 but not 313 (Figure 1B). Importantly, consistent with all previous studies on mK3 and recent Tandem mass spectrometry (MS)/MS data (Wang et al. unpublished data) the minimum form of substrate ubiquitination observed is Ub2(12,13,22,37,38). These results extend our previous findings that a single threonine residue positioned near the C-terminus of an Ld KCST-less tail molecule is sufficient to allow mK3-facilitated ubiquitination (12). Importantly, we also found that adding back single K residue at position 337 but not at position 313 was sufficient to restore polyubiquitination of the Ld KCST-less tail (Figure 1B). In addition, regardless of whether single T or K residues were added back to the Ld KCST-less tail molecule at positions 329 or 337 mK3-induced ubiquitination was restored to identical levels (data not shown).
These results clearly indicate that a location proximal to the C-terminus of the cytoplasmic tail of HC is essential for mK3-facilitated ubiquitination of S, T and K target residues. These results suggest that the determining factor for which HC tail residues are ubiquitinated by mK3 is either RING-CH domain proximity or neighboring sequence.
Ubiquitination of S or T residues of HC by mK3 ligase is direct, robust and completely independent of HC tail sequence
Extensive mutagenesis and chemical treatment data has demonstrated that mK3 is capable of ubiquitinating S or T residues in the tail of its HC substrate via ester linkages (12). To unequivocally demonstrate that ubiquitination of S and T residues is direct and independent of the surrounding substrate sequence, the wild-type (wt) cytoplasmic tail of the mouse HC Ld molecule was replaced by a completely generic tail composed of 20 glycine residues (Figure 2A). In addition, Ld poly (G) tail constructs were made in which the second and third to last positions were changed to either a di-K, -C, -S or -T motif [poly (G)-KK, -CC, -SS, -TT]. These Ld mutants were found to have normal surface expression and TAP binding (data not shown). Lysates from WT3 cells stably co-expressing generic tailed Ld molecules and mK3 were immunoprecipitated with anti-Ld, treated with Endo H and blotted with anti-Ub antibody. The results were unambiguous; polyubiquitinated species of HC were readily detectable as long as viable target residues (SS, TT or KK) were present in their tails (Figure 2B).
From these data we conclude that mK3 facilitated coupling of Ub to HC by ester bonds as well as amide bonds is sequence, and likely secondary structure, independent. Furthermore, these findings show that ubiquitination of S or T residues on HC by mK3 is robust and direct. Currently, it is poorly understood how Ub conjugation sites are selected (39). However, there are examples of both Ub acceptor site specificity and promiscuity within various substrates (40–43). In any case, the findings reported here provide the most compelling evidence thus far that substrate specificity of mK3 is not determined by its recognition of an HC-specific sequence, but rather by proximity imposed by its adaptor protein TAP.
Cellular RING-CH domains in the context of mK3 can support low levels of ubiquitination via isopeptide as well as ester bonds
Given the importance of the RING domain in E3 ligase activity and specific recruitment of E2 conjugases, we wanted to investigate the importance of the mK3 RING domain and juxtaposed sequences for ester or amide bond formation. To test the specific contribution of the RING domain, we evaluated whether the RING domain of cellular ligases could support ubiquitination in the context of mK3 sequences outside the RING [i.e. RING swaps (RS)]. For these comparisons, RS constructs were made using the RING domains of Hrd1 and gp78, two well-characterized mammalian ligases implicated in endoplasmic reticulum-associated degradation (ERAD), as well as the RING domains of five representative MARCH proteins. Selection of viral and cellular RINGs for these experiments was based on the amino acid sequence alignments of various cellular and viral RINGs using the Tree based Consistency Objective Function For AlignmEnt Evaluation (T-COFFEE) algorithm (shown in Figure 3A) (44). This alignment shows the conserved zinc ion coordinating motifs that are found in all cellular and viral RING domains and are required for their E3 ligase activity (blue shading; Figure 3A). The aligned RING-CH sequences were employed to generate a bootstrap consensus tree inferred from 5000 replicates using the neighbor-joining method (Figure 3B) (45,46). A striking feature of this dendrogram is that the viral RING-CH domains form a separate clade divergent from all of the cellular RING-CH domains. Significantly, a more comprehensive phylogenetic analysis confirmed that the viral RING-CH domains are divergent from all recognized human and murine RING domains as well as several RING-CH domains from other vertebrates including rat, bull and zebra fish (data not shown). This observation suggests that the viral RING-CH domains have diverged from cellular RING-CH domains to provide a necessary or beneficial trait for viral subversion of the host ubiquitination machinery.
Initially, we tested the ubiquitination of wt MHC I heavy chain (wt HC) by chimeric constructs gp78RS and Hrd1RS that have the H2 RINGs of gp78 or Hrd1 replacing the RING-CH domains of mK3. Interestingly, gp78RS and Hrd1RS did not facilitate the ubiquitination of Endo H sensitive HC molecules (Figure 4A). However, ubiquitination of post-ER (Endo H resistant) HC was detected with both gp78RS and Hrd1RS, but not wt mK3 ligase. We have found that in the presence of a non-catalytic RING-CH mutation of mK3 (mK3 RM) (Figure 4B,C) or in the absence of viral E3 ligase (Figure 6B), a similar pattern of lysine-dependant Endo H resistant ubiquitination of HC molecules is consistently observed. Therefore, this post-ER ubiquitination of HC is likely because of the cellular E3 ligase/s that normally target HC as a substrate to maintain homeostasis of MHC I proteins. Importantly, failure to ubiquitinate Endo H sensitive HC by gp78 RS or Hrd1 RS was not because of an inability to bind TAP (data not shown), a binding mediated by the C-terminal domain of mK3 (37). In any case, the RING-H2 domains of gp78 and Hrd1 were unable to support the ubiquitination of nascent HC required to induce ERAD.
Subsequently, we evaluated the ability of mouse MARCH RING-CH domains in the context of mK3 to facilitate the ubiquitination of wt Ld molecules. To provide representative coverage of the mouse RING-CH domains, MARCH II, V, VI, VIII and IX were selected as RING donors for the chimeras in the following experiments (see Figure 3A,B). Based on the presence of Ub2 forms of HC, all of the MARCH RING-CH domains tested could function in the context of mK3 (Figure 4B). However, densitometric analysis of lanes in which Endo H sensitive Ld-Ub3 species were detectable showed that the ratio of Ub3/Ub2 Endo H sensitive HCs modified by wt mK3, MARCH V and MARCH VIII swaps (Figure 4B lanes 1, 7 and 11) was 0.799, 0.459 and 0.0831, respectively. Indicating that although all of the chimeric MARCH swaps were less efficient than wt mK3 at facilitating the formation of polyubiquitin chains on HCs, the MARCH V RING swap was impaired to a lesser degree.
We next tested whether cellular RING-CH domains in the context of mK3 could mediate the formation of ester bonds with HC substrates. To explore this possibility, we evaluated the ability of the MARCH swap chimeras to facilitate the ubiquitination of Ld KCST-less tail molecules with a S residue added back at position 329 (Ld KCST-less tail 329S). Interestingly, the MARCH V, VI and VIII RS facilitated weak but detectable polyubiquitination of this hydroxylated amino acid in the HC substrate (Figure 4C). However, polyubiquitination of 329S HC greater than Ub2 was only observed in the presence of the MARCH V RING swap. A potential explanation for the reduced capacity of the MARCH RS to facilitate the ubiquitination of the Ld KCST-less tail 329S molecules compared to wt Ld may be because of the availability of only a single viable target residue. To explore this possibility, we replaced position 337 in the Ld KCST-less tail molecule with a conventional lysine residue (Ld KCST-less tail 337K) and assessed for its ubiquitination in the presence of the MARCH RING swap chimeras. We found that the MARCH RS could facilitate the ubiquitination of the Ld KCST-less tail 337K molecules in a manner very similar to that of wt Ld (data not shown). This suggests that the formation of ester linkages via hydroxylated residues by MARCH RS is less efficient than the formation of traditional amide linkages via K residues.
These data provide evidence that ubiquitination of non-K residues can be weakly facilitated by certain cellular RING-CH domains in the context of a viral E3 ligase. This implies that although the process is inefficient, the necessary Ub components, including the E2, are present in mammalian cells that are necessary for the conjugation of Ub to S residues in substrates. The ability of the MARCH RS chimeras to inefficiently confer ester as well as amide Ub linkages raised the question of why only rare intact ligases have the ability to ubiquitinate non-K residues. To address this question we next compared mK3 chimeras with RING-CH domains from KSHV ligases that facilitate ubiquitination of thiolester and not ester bonds (10,11).
Do differences in the viral RING-CH domains of γHV68 versus KSHV determine whether an ester or thiolester Ub linkage is formed with the HC substrate?
To test this we replaced the RING-CH domain of mK3 with equivalent domains from either kK3 or kK5 to evaluate the role of these domains in facilitating the ubiquitination of non-lysine residues. For these assays we stably expressed each of the di-K, -C, -S, -T and entirely poly (G) tail Ld constructs described earlier (Figure 2A) in the presence of either the kK3 or kK5 RING swap mutants. Although wt kK3 does not have the ability to facilitate the ubiquitination of either S or T residues on the cytoplasmic tail of its HC substrate, its RING-CH domain in the context of mK3 now supports the ubiquitination of both these hydroxylated residues (Figure 5A). In a parallel experiment, we also observed that in contrast to wt kK5, the kK5 RING in the context of mK3 also gains the ability to facilitate the ubiquitination of S and T residues (Figure 5C).
Using the single S and K add back mutants on an Ld KCST-less tail background described earlier (Figure 1A), we next tested whether the RING-CH domains of either kK3 or kK5 in the context of mK3 would alter the substrate target residue positional constraints demonstrated by wt mK3 (Figure 1B). Interestingly, we found that like wt mK3, both the kK3 RS (Figure 5B) and kK5 RS (Figure 5D) chimeras only facilitated the ubiquitination of target residues proximal to the C-terminus of the cytoplasmic tail.
These data demonstrate that the RING-CH does not function autonomously in conferring substrate residue positional constraints or in determining the capacity to form ester versus thiolester bonds. It should also be noted the viral RS mK3 chimeras gave substantially more robust ubiquitination of HC compared with the MARCH RS chimeras. This impairment could potentially be accounted for by sequences within the RING-CH or the need for complementing sequence outside the RING-CH to support efficient Ub conjugation. These two possibilities were tested below.
RING-CH function is complemented by membrane proximal sequences
The cytosolic N-terminus of mK3 is comprised of a sequence of seven amino acids preceding the RING-CH domain both of which are followed by what we propose to be designated the domain in between the RING-CH and the first transmembrane or DIRT. To assess the influence of the DIRT on whether target residues can be utilized by mK3, we replaced the DIRT of mK3 with the corresponding domains from either kK3 or kK5 (Figure 6A). We evaluated which the kK3 and kK5 DIRTs could support the ubiquitination of wt Ld in the context of mK3 by examining steady-state levels of ubiquitinated HC in their presence (Figure 6B). Only the kK3 but not the kK5 DIRT swap was capable of facilitating the ubiquitination of Endo H sensitive wt Ld molecules. However, the blot of the kK5 DIRT construct did show ubiquitination of post-ER HC like the non-catalytic mK3 variants. As mentioned earlier, post-ER ubiquitination likely reflected physiologic turnover as mediated by endogenous ligases. One possible explanation for failure of the kK5 DIRT construct to ubiquitinate nascent HC would be a failure to associate with TAP. However, this possibility was eliminated by demonstrating that both the kK3 and kK5 DIRT constructs were normally associated with tapasin that bridges HC to TAP (Figure 6B, right upper panel). These data provide direct evidence that sequences outside the RING-CH domain of mK3 play a necessary role in the ubiquitination of nascent HC.
The first conserved W in the RING-CH domain of viral ligases is required for efficient mK3-mediated polyubiquitination of HC and ERAD of HC
Interestingly, there is a conserved W residue (light red highlight; Figure 3A) located between the first and second zinc coordinating residues of all viral RING-CH domains (with the exception of Herpesvirus saimiri which lacks both W residues in its RING) which is absent in almost all cellular RING-CH domains. In the solution-based model of the RING-CH structure, this W residue has a location that could potentially alter E2 association. More specifically, chemical shift perturbation nuclear magnetic resonance spectroscopy (NMR) studies have defined the RING-CH residues of kK3 that interact with E2 conjugase UbcH13 and are underlined in red (Figure 3A) (47). To determine if this conserved viral W affects ubiquitination of HC, we made a mutant in which the first W in the RING-CH domain of mK3 was changed to an R (mK3 W9R). We first evaluated whether the W9R mutant facilitated comparable amounts of polyubiquitination as wt mK3 by comparing their respective steady-state levels of HC ubiquitination in the presence of a proteasome inhibitor. As shown in the left upper panel of Figure 7A, the polyubiquitination of HC by the W9R mutant appears to be impaired. More specifically compared with wt mK3, less Ub2 and nearly undetectable amounts of Ub3 and Ub4 modified HC species are generated in cells expressing the W9R mutant. In addition, we found that in the absence of proteasome inhibition the steady-state level of unmodified HC in the presence of the W9R mutant was much greater than observed with wt mK3 (data not shown). To further investigate whether the polyubiquitination induced by the W9R mutant resulted in ERAD of HC, a pulse-chase experiment was carried out. Interestingly, metabolic labeling revealed that the W9R mutant induced little, if any, turnover of HC compared to wt and a catalytically inert mK3 ligase (Figure 7B). Based on the above-mentioned mapping of the RING-CH/E2 interaction site of kK3 (47), it is attractive to speculate that the W9R mutation alters E2 binding to mK3 ligase. Indeed, comparative protein modeling of wt mK3 and mK3 W9R reveals a striking difference in the topology of their predicted E2 binding faces (yellow highlight; Figure 8A,B). Conceptually, such an alteration could result in either a different E2 binding or the repositioning of the same E2 perhaps effecting chain topology and/or efficiency of chain formation. Alternatively, if multiple E2 conjugases are required for mK3-mediated Ub chain assembly [a la KSHV viral ligase kK3 (48)], the W9R mutation could prevent interaction with a second E2 that constructs chains beyond Ub2. To address the possibility of the W9R mutant facilitating the generation of polyubiquitin chains with a different topology (linkage) on its substrate than wt mK3, a semi-permeabilized cell system (described below) and a panel of HA-tagged mutant Ub molecules were employed. These analyses revealed that both W9R and wt mK3 ligases facilitated the generation of polyubiquitin chains with the same topology (linkage) (Wang unpublished data).
To compare the efficiency of Ub chain assembly catalyzed by W9R versus wt mK3, cells were semi-permeabilized to remove endogenous Ub and recombinant HA-tagged Ub (HA-Ub) was then added. After 45 min of incubation in the presence of proteasome inhibitor, HC from semi-permeabilized cells expressing W9R or wt mK3 were precipitated, electrophoresed and blotted with anti-HA (to detect newly ubiquitinated forms) or anti-Ub (to detect steady-state ubiquitinated forms). As shown in Figure 7C left panel, semi-permeabilized cells expressing wt mK3 displayed high levels of HA-Ub2 modified HC as well as some higher multimers. By contrast, semi-permeabilized cells expressing the W9R mutant displayed little if any HA-Ub modified HC. Thus, the W9R mutant is severely impaired in de novo Ub chain assembly. However, steady-state levels of HC ubiquitination as shown in the Ub blot of semi-permeabilized (Figure 7C,D, right panel bands without asterisks) cells expressing W9R or wt mK3 were very similar. These combined findings thus clearly demonstrate that Ub chain assembly facilitated by W9R is impaired, causing fewer Ub modified HC species to be found in the ER when compared with HC levels in the presence of wt mK3 (Figure 7A).
It should be noted that the W9R mutant and wt mK3 were found to have a comparable Ub phenotype on HC tails with a single S or K ubiquitination site. Indeed, we found that Ub chain assembly on Ld KCST-less tail A329S molecules in the presence of W9R was kinetically impaired in an almost identical manner as wt HCs (Figure 7D). Thus, the first conserved W in the RING-CH domain of mK3 is not a determining factor for ester Ub ligation. However, it is clearly important for ligation of Ub by mK3 and likely other viral ligases. Further studies of the W9R mutation are likely to prove insightful for dissection of the interaction of mK3 with cognate E2s and HC substrates.
Viral E3 ligases provide novel strategies to resolve several outstanding enigmas regarding the precise mechanisms of substrate ubiquitination and its consequences. The reason for this stems from their potency and known substrate specificity, both consequences of their function as immune evasion proteins of the virus. More specifically, the mK3 ligase encoded by murine herpesvirus γHV68 and the highly homologous kK3 and kk5 ligases of KSHV all include MHC class I HCs as substrates that are targeted to Ub-dependent degradation pathways. Indeed, the targeted degradation of HC molecules by these viral ligases allows cells infected by these herpesviruses to persist in their host by escaping detection by immune CD8 T cells. What has also become clear from recent studies is that these viral ligases are highly efficacious molecular probes for degradation pathways. For example, elegant studies of the kK3 ligase have elucidated an Ub-dependent pathway of rapid substrate endocytosis of plasma membrane proteins resulting in their degradation in the lysosome (7,14–16). By contrast, studies of mK3 ligase have elucidated an Ub-dependent pathway of ERAD (13). There is little doubt that these viral ligases exploit physiological pathways as evidenced by their defined interactions with host proteins as well as the fact they have mammalian cellular homologs (the MARCH proteins).
Substrate specificity is a function of the E3 ligase. However, the molecular interactions conferring substrate specificity are in most cases unknown. Although E3 ligases can potentially bind directly to substrate-specific sequence motifs, this is unlikely to be a universal mechanism because most E3 ligases detect multiple substrates that lack an obvious shared sequence or structural motifs. For example, in addition to HC, viral ligase kK5 targets MICA (MHC class I-related chain A), MICB (MHC class I-related chain B), ICAM-1, CD31/PECAM (platelet/endothelial cell adhesion molecule-1), interferon-γ (IFN-γ) receptor 1 as well as VE-cadherin for lysosomal degradation (17–20,49,50). Furthermore, the viral ligase mK3 specifically ubiquitinates the cytoplasmic tails of HC molecules (12). However, the tails and the transmembrane domains of HC molecules are highly polymorphic and share no obvious conserved sequence motifs. Indeed we show here that an HC with a generic (poly G) tail can be ubiquitinated as long as it has K, S or T residues located near the C-terminus. This finding is in striking support of our previously proposed proximity model, whereby mK3 binds TAP that functions as an adaptor protein to confer HC specificity to mK3-mediated ubiquitination (see model in Figure 9). More specifically, this model speculates that when mK3 binds to TAP it orients its RING-CH domain such that it only ubiquitinates HC [although there can be some collateral damage to TAP in certain cells (51)]. Because all HC molecules bind to TAP before peptide acquisition, mK3 can effectively prevent antigen presentation and thereby CD8 T-cell recognition of infected cells. It seems likely to us that this may be a common paradigm and other E3s will similarly use adaptor molecules to orient their RING domains to confer substrate specificity.
Another remarkable feature of viral ligases is their ability to ubiquitinate non-K residues of their substrates. Whereas kK3 and kK5 can ubiquitinate cysteine residues of HC substrates, mK3 can ubiquitinate S/T residues of HC substrates (10–12). The molecular basis of this non-K ubiquitination is very intriguing as is the question of its physiological relevance. Central to both of these issues is the precise function of the RING-CH domain in substrate ubiquitination. For example, do the viral RING-CH domains possess unique structures that determine their ability to ubiquitinate non-K residues perhaps, or do the RING-CH domains of cellular E3 have the same capacity to ubiquitinate non-K residues? To address these questions, we initially probed the ubiquitination potential of the RING-CH domains of five representative MARCH proteins in the context of mK3 ligase, i.e. RS. Interestingly, all five of the MARCH RING-CH domains in the context of mK3 were capable of ubiquitinating a wt HC tail, but to various degrees and all less than intact mK3. What was particularly striking was an impaired ability to polyubiquitinate beyond Ub2. The overall lack of efficient polyubiquitination by these RING-CH swaps suggests two non-mutually exclusive explanations. One that the structures and functions of the viral versus cellular RING-CH are different and/or that sequences outside the RING are important to support RING activity. Here we present evidence that both of these explanations are correct, thereby suggesting a model for E3/E2 substrate interaction.
Although the viral and cellular ligases with RING-CH domains share the absolutely conserved core C4HC3 zinc coordinating motif, they have sufficient sequence differences to make viral versus cellular RING-CH type ligases cluster into distinct phylogenetic branches (Figure 3B). Indeed an immediately notable difference in the comparisons of the RING-CH sequences is the amino acid in the second position. Although all three viral ligases compared here have a W residue in the second position of their RING-CH domains (position W9 in mK3), the cellular RING-CHs tested mostly have an R residue. The single exception is the RING-CH domain of MARCH V that like the viral homologs has a W. Interestingly, the MARCH V RING-CH in the context of mK3 was the most effective of the MARCH RING-CH swaps at polyubiquitinating HC. As another reflection of its superior polyubiquitination potential, the MARCH V RING-CH swap also displayed the keenest ability to ubiquitinate a single S residue in the HC tail, albeit considerably weaker than wt mK3 ligase (Figure 4B).
Importantly, a recent solution structure showed the W residue in the second position of the RING-CH domain of viral ligase kK3 forms a novel hydrophobic pocket with a second W residue between the sixth and seventh zinc coordinating residues that is highly conserved among all viral and cellular RING-type E3s (shown in Figure 8A) (47). It should be noted that this highly conserved W residue between the sixth and seventh zinc coordinating residues is involved in E2 recruitment as shown in previously published studies of c-Cbl and kK3 (14,52). To define the functional importance of W in the second RING-CH position of viral ligases and its possible importance for non-K ubiquitinations, we mutated the W to R in mK3. This mutant, mK3 W9R, resulted in a dramatic kinetic impairment of Ub chain assembly compared to wt mK3 ligase. Remarkably, this kinetic impairment rendered the mK3 W9R mutant functionally inert in that it was no longer capable of destabilizing HC as shown in a pulse/chase comparison with wt mK3 (Figure 7B). Based on the model shown in Figure 8B, the W9R mutation of mK3 is predicted to drastically reconfigure the topology of the aforementioned hydrophobic pocket implicated in E2 interaction of other E3 ligases. Thus, it seems likely that the mK3 W9R mutant interacts poorly with its cognate E2 or associates with a different E2 than wt mK3, possibly resulting in aberrant Ub chain assembly. Alternatively, mK3 may recruit two E2 conjugases similar to kK3 which recruits UbcH5b/c and Ubc13 that are required for targeting HC degradation in the lysosome (48). However, in the case of mK3 we would speculate that the first E2 might transfer a premade Ub2 chain to HC and the second E2 might facilitate the further polyubiquitination needed to induce ERAD. If this model is correct, the W9R mutation may have affected only interaction with the second E2. In any case, we show here for the first time the W residue in the second RING-CH position (between the first and second zinc coordinating residues) is critical for mK3 function, but it is not a critical factor determining the ability to ubiquitinate non-K residues.
Consistent with the similarity among viral RING-CH domains, chimeric constructs with the kK3 or kK5 RING-CH domains in the context of mK3 were clearly capable of robustly polyubiquitinating HC substrates via K, S or T residues (Figure 5). This finding was of considerable interest because intact kK3 and kK5 can efficiently only ubiquitinate K or C residues, but not S/T residues. Another interesting result was that in contrast to polyubiquitination of HC by intact kK3 and kK5 (14,53), Ub2 is the minimal form of ubiquitinated HC observed in the presence of the kK3 and kK5 RS. In addition, we also found that the kK3 RS and kK5 RS have the same target residue positional preferences as wt mK3. Thus, our findings demonstrate unequivocally that the RING-CH domain does not function autonomously in its ability to make ester versus thiolester Ub linkages, conferring substrate residue positional constraints, or in determining the mechanism of Ub chain assembly on substrates. In further support of these conclusions, we show here that the DIRT is incredibly important for mK3 ligase activity. More specifically, we found that in the context of mK3 the DIRT from kK3 but not kK5 supported ubiquitination of HC substrates. Consistent with its ability to support ubiquitination, the sequence of the kK3 DIRT is considerably more similar to that of mK3 than it is to the sequence of the kK5 DIRT. Thus consistent with recent studies of other E3 ligases, the RING domain of mK3 does not function autonomously to ubiquitinate K or S/T residues of HC substrates (24,54–57).
Cumulative findings reported here suggest a model by which mK3 ubiquitinates substrates, with implications on its ability to form ester bonds. Clearly the RING-CH domain itself does not determine the ability to form ester versus thiolester bonds. The importance of the DIRT between the RING-CH and transmembrane (TM) is also of critical importance and could be directly involved in E2 binding as published with other ligases. Alternatively, the DIRT could optimally orient the RING domain for substrate interaction. Indeed an orientation model was recently proposed for the interaction of the kK3 and kK5 viral ligases with HC whereby their respective DIRTs differentially interact with the membrane proximal sequences in their C-terminal domains (11). It was further speculated that the differential juxtaposition of these two membrane proximal sequences in kK3 versus kK5 differentially affects the location of residues on the HC tail that were preferentially ubiquitinated by kK3 versus kK5. This model has similarities to our proximity model, whereby TAP binding with the C-terminal domain of mK3 orients the RING-CH domain and thereby impacts on the location of optimally ubiquitinated residues. It should also be noted that membrane proximal regions of the C-terminal domain of kK3 and kK5 contain putative protein-interaction domains, suggesting they might also interact with adapter proteins that could also influence substrate specificity (16). In any case, it is the location on the tail of the ubiquitinated residue (S/T or K) and not surrounding sequences that is important for mK3 activity, thus highly accentuating the importance of RING orientation/proximity to its substrate. Based on these combined findings, we propose that what is unique about mK3 and perhaps other viral ligases is their optimal positioning of the RING-CH domain as rendered by its interaction with the DIRT and host adaptor proteins. It is attractive to speculate this putative optimized activity of the viral ligase is reflective of its divergence from host ligases to more proficiently ubiquitinate substrates and thereby function as an immune evasin. If this model is correct, a specific E2 uniquely capable of forming ester or thiolester bonds need not be recruited. In support of this conclusion, in comparisons of the numerous mutant and chimeric mK3 constructs we have tested, the ones found most active at ubiquitinating wt HC were also capable of ubiquitinating HC constructs with optimally placed S, T or K tail residues. Thus, we propose that the data is best explained by viral ligases having evolved a strategy for establishing an optimal topological orientation of their RING-CH domains with substrate thereby allowing the ubiquitination of S/T or K residues.
The selective pressures on viral RING-CH E3 ligases have favored greater target residue flexibility compared to most cellular ligases; however, it should be noted that there is increasing evidence for the physiological importance of substrate ubiquitination via thiolester/ester bonds (e.g. TCRa, CD4, Bid) (58–60). Most notably, in yeast and mammals export of Pex5p, the peroxisomal import receptor, requires ubiquitination of a conserved C residue of Pex5p (61–63). Interestingly, this thiolester bond formation in mammals appears to be mediated by members of the UBE2D family that are thought to be responsible for a large fraction of the total ubiquitinated proteins (64,65). Thus, these findings are highly complementary to our findings in suggesting general applications of non-K ubiquitination and the fact that they need not be mediated by specialized E2s.
Materials and Methods
Cell lines and flow cytometry
Murine B6/WT3 (WT3, H-2b) has been described previously (37). 293T cells (66) were used for production of ecotropic retrovirus. All cells were maintained in complete RPMI-1640 with 10% fetal calf serum (FCS) (HyClone) as described (21). Retrovirus-containing supernatants were produced using Vpack vector system (Stratagene) with transient transfection of 293T cells. Cells transduced by pMIN-containing virus were enriched by geneticin selection, whereas green fluorescent protein (GFP)+ cells from pMIG transduced lines were enriched by cell sorting. Where indicated, cells were cultured for 24 h with 125 U/mL of mouse gamma interferon (Biosource) before harvesting with trypsin ethylenediaminetetraacetic acid (EDTA). All flow cytometric analyses were performed as previously described (21).
Two retroviral expression vectors, pMSCV.IRES.GFP (pMIG) and pMSCV.IRES.neo (pMIN) (37), were used to express mK3 and Ld constructs, respectively. mK3 sequence was obtained by polymerase chain reaction (PCR) amplification of the K3 gene from a γHV68 subclone (67). Both mK3 and Ld mutants were generated by site-directed mutagenesis (Stratagene). The Ld with a poly (G) 20 mer tail was a gift from Dr Lonnie Lybarger. Cellular and viral RING/DIRT chimeric constructs were made by overlapping PCR (68). The correct sequences for all of the constructs were confirmed by DNA sequencing.
Rabbit anti-mK3, hamster anti-tapasin, Ub antibodies, β-actin (AC-74) antibodies and mAbs 30-5-7 and 64-3-7 to folded and open forms of MHC class I Ld have been previously described (21). Anti-HA antibody (16B12) was purchased from Covance.
Immunoprecipitation and immunoblots
Immunoprecipitation (IP) and immunoblot were conducted as previously described (21). Briefly, for co-immunoprecipitation (co-IP), cells were lysed in PBS buffer containing 1% digitonin (Wako), 20 mm iodoacetamide (IAA, Sigma) and protease inhibitors (Complete mini, Roche). For IP, cells were lysed in 1% Nonidet P-40 (NP-40), PBS containing 20 mm IAA (PBS/IAA) and 0.4 mm phenylmethylsulphonyl fluoride (PMSF) (Sigma). Postnuclear lysates were incubated with protein A–Sepharose beads (Sigma) and antibodies. After washing beads 4× with PBS/IAA buffer containing either 0.1% digitonin for co-IP samples or 0.15% NP-40 for IP samples, immunoprecipitates were eluted from protein A beads by boiling for 3 min in lithium dodecyl sulphate (LDS) sample buffer (Invitrogen) or 10 mm Tris–HCl, pH 6.8, with 0.5% SDS if Endo H treatment followed, in which elutes were mixed with an equal volume of 100 mm sodium acetate, pH 5.4 and incubated with 1 mU of Endo H (ICN) at 37°C for 2 h. Immunoblotting was performed following SDS–PAGE separation of precipitated proteins or cell lysates as previously described (69). Specific proteins were visualized by chemiluminescence using the enhanced chemiluminescence (ECL) system (GE Healthcare).
Metabolic labeling and pulse chase
After 30 min of preincubation in cysteine- and methionine-free medium (DMEM with 5% dialyzed FCS), cells were pulse labeled with Express 35S-Cys/Met labeling mix (Perkin Elmer Life Sciences) at 150 μCi/mL for 15 min. Chase was initiated by the addition of an excess of unlabeled Cys/Met (5 mm each). IP was performed as described above. Samples were subjected to SDS–PAGE, and gels were treated with Amplify (GE Healthcare), dried and exposed to BioMax-MR film (Kodak).
MHC I heavy chain ubiquitination in semi-permeabilized cell system
A semi-permeabilized cell system (70) with modification was utilized to determine the efficiency of MHC I HC ubiquitination mediated by mK3. Briefly, cells co-expressing mK3 and Ld were permeabilized in 0.02% digitonin permeabilization buffer (PB) buffer (25 mm HEPES, pH 7.4, 115 mm KOAc, 5 mm NaOAc, 2.5 mm MgCl2 and 0.5 mm EGTA) on ice for 20 min. Following centrifugation at 18000×g for 15 min at 4°C and washing with PB buffer, the cells were resuspended and incubated in reaction buffer [25 mm Tris–HCl, pH 7.4 supplemented with ATP regenerating reagent, 50 μm MG132, 1 mm PMSF, 2 mm ATP, 20 μm HA-Ub (Boston Biochem) and 90 nm human recombinant E1 or 150 μg rabbit Fraction II (FII) (Boston Biochem)] at 37°C for 45 min. After incubation, cells were pelleted by centrifugation and then subjected to IP of HC and immunoblot as described above.
Alignments and phylogenetic analysis
The T-COFFEE algorithm was utilized to perform multiple sequence alignments (44). Phylogenetic analyses of aligned sequences were conducted in Molecular Evolutionary Genetics Analysis (MEGA4) (71). The evolutionary history of the taxa analyzed was inferred using the neighbor-joining method, and a bootstrap consensus tree was inferred from 5000 replicates (45,46). Evolutionary distances were computed using the Dayhoff matrix based method (72). All positions containing alignment gaps and missing data were eliminated only in pairwise sequence comparisons. Accession numbers for analyzed sequences: gp78(NM 001144), Hrd1(NM 032431), MARCH I(NM 175188), MARCH II(NM 145486), MARCH III(NM 177115), MARCH IV(NM 001045533), MARCH V(BC132457), MARCH VI(NM 172606), MARCH VII(NM 020575), MARCH VIII(NM 027920), MARCH IX(NM 001033262), MARCH XI(NM 177597), γHV68 mK3(U97553), KSHV kK3(KSU83350), KSHV kK5(AAB62655), BHV4 pBo5(AAK07931), SFV gp153R(AAF18029), MYXV m153R(AF229033), SPV C7L(AAC37864), GTPV gp008(AY077836) and SPPV 08(AY077834).
We thank Dr Lonnie Lybarger for kindly providing the pMIN Ld poly (G) tail construct that was used to generate all of the Ld poly (G) tail variants used in this study. This work was supported by National Institutes of Health grants AI19687 (to T. H. Hansen and X. Wang) and AI07163 (to R. A. Herr).