HIV-1 Gag engages components of the ESCRT (endosomal sorting complex required for transport) pathway via so-called L (late-assembly) domains to promote virus budding. Specifically, the PTAP (Pro-Thr-Ala-Pro)-type primary L domain of HIV-1 recruits ESCRT-I by binding to Tsg101 (tumour susceptibility gene 101), and an auxiliary LYPXnL (Leu-Tyr-Pro-Xaan-Leu)-type L domain recruits the ESCRT-III-binding partner Alix [ALG-2 (apoptosis-linked gene 2)-interacting protein X]. The structurally related CHMPs (charged multivesicular body proteins), which form ESCRT-III, are kept in an inactive state through intramolecular interactions, and become potent inhibitors of HIV-1 budding upon removal of an autoinhibitory region. In the absence of the primary L domain, HIV-1 budding is strongly impaired, but can be efficiently rescued through the overexpression of Alix. This effect of Alix depends on its ability to interact with CHMP4, suggesting that it is the recruitment of CHMPs that ultimately drives virus release. Surprisingly, HIV-1 budding defects can also be efficiently corrected by overexpressing Nedd (neural-precursor-cell-expressed developmentally down-regulated) 4-2s, a member of a family of ubiquitin ligases previously implicated in the function of PPXY (Pro-Pro-Xaa-Tyr)-type L domains, which are absent from HIV-1. At least under certain circumstances, Nedd4-2s stimulates the activity of PTAP-type L domains, raising the possibility that the ubiquitin ligase regulates the activity of ESCRT-I.
The ESCRT (endosomal sorting complex required for transport) pathway consists of a network of class E Vps (vacuolar protein sorting) proteins that is required for the biogenesis of MVBs (multivesicular bodies) and for the sorting of ubiquitinated cargo into these endocytic organelles [1–3]. The majority of class E Vps proteins are components of complexes termed ESCRT-I, -II and -III [3,4]. These complexes appear to act in a sequential manner to mediate the sorting of transmembrane proteins into cellular vesicles that bud into the lumen of MVBs. ESCRT-I recognizes ubiquitinated protein cargo via its Tsg101 (tumour susceptibility gene 101) subunit and activates ESCRT-II, which then triggers the assembly of ESCRT-III on the cytosolic side of endosomal membranes [5–7]. In mammals, ESCRT-III is formed by approximately ten structurally related CHMPs (charged multivesicular body proteins). These interact with the ATPase Vps4 and its activator Vta1 (Vps20-associated 1), class E Vps proteins that mediate the recycling of ESCRTs [8–15]. Furthermore, the CHMP4A, CHMP4B and CHMP4C components of ESCRT-III interact with Alix [ALG-2 (apoptosis-linked gene 2)-interacting protein X], a homologue of the yeast class E Vps protein Bro1 [10,11,16–19].
To promote virus release from the plasma membrane, the C-terminal p6 domain of HIV-1 Gag binds directly to two components of this cellular budding machinery through so-called L (late-assembly) domains [20–23]. The primary HIV-1 L domain consists of a PTAP (Pro-Thr-Ala-Pro) motif that binds ESCRT-I component Tsg101, and this interaction is critical for the correct proteolytic processing of Gag and for virus particle release [24–26]. The PTAP–Tsg101 interaction is even more important for the release of infectious virus, which is down nearly 500-fold if the PTAP motif is disrupted . The p6 domain of HIV-1 Gag also contains an auxiliary L domain of the LYPXnL (Leu-Tyr-Pro-Xaan-Leu)-type that serves as a docking site for Alix [10,16]. All types of L domain are strongly inhibited by ATPase-defective Vps4 and by mutant ESCRT-III components, indicating that all require an intact ESCRT pathway to function [10,11,16,24,28,29].
ESCRT-III components are regulated through autoinhibition
All ESCRT-III components become potent inhibitors of HIV-1 budding if fused to a bulky tag such as GFP (green fluorescent protein) [10,11,16]. This phenomenon raised the possibility that the presence of a large tag may irreversibly activate CHMPs by interfering with autoinhibitory interactions. Furthermore, the presence of highly basic N-terminal halves and highly acidic C-terminal halves in all CHMPs suggested a model in which autoinhibition is mediated by electrostatic intramolecular interactions [30,31]. As predicted by this model, progressive truncations into the C-terminal acidic domains of CHMPs triggered an increasingly robust anti-HIV budding activity [28,29]. Indeed, the basic N-terminal half of CHMP3 was sufficient to inhibit HIV-1 budding with maximal efficiency [28,32]. This observation suggested that the acidic C-terminal halves of CHMPs function as autoinhibitory modules that repress the wild-type proteins through interactions with their basic N-terminal halves. Consistent with this notion, it has been shown that removing C-terminal residues from CHMPs triggers their membrane association and assembly into polymers [29,32], and structural studies indicate that at least CHMP3 exists in a closed and open (activated) conformation in solution .
The model of CHMP regulation outlined above also predicted a physical interaction between the N-terminal basic domains of the CHMPs and their cognate autoinhibitory regions. Indeed, the isolated basic domain of CHMP3 interacted avidly with a fragment from the acidic domain in GST (glutathione transferase) pull-down assays . Furthermore, the apparent affinity of the CHMP3 truncation mutants for the acidic domain fragment correlated with their potential to act as inhibitors of HIV-1 budding . A likely explanation is that weakly dominant-negative mutants still retained part of the autoinhibitory domain, preventing efficient binding to autoinhibitory sequences in trans. It is also noteworthy that the N-terminal basic domains of different CHMPs interacted preferentially with their cognate acidic domains , arguing against the possibility that only electrostatic interactions are involved.
It was also tested whether native CHMPs can be triggered to inhibit HIV-1 budding through the overexpression of a cellular binding partner. To this end, the ubiquitin isopeptidase AMSH [associated molecule with the SH3 (Src homology 3) domain of STAM (signal-transducing adaptor molecule)] was used, which interacts tightly with the very C-terminus of the autoinhibitory domain of CHMP3 [28,33]. Indeed, when overexpressed together with native CHMP3, AMSH strongly inhibited HIV-1 budding, even though it had little effect on its own . These observations are consistent with the interpretation that the intermolecular interaction with AMSH interfered with the intramolecular interaction required for the autoinhibition of CHMP3 (Figure 1).
Model of CHMP3 activation by AMSH
Structural basis for the autoinhibition of ESCRT-III component CHMP3
Chemical cross-linking followed by gel-filtration chromatography has indicated that full-length human CHMP3 is monomeric in solution . In contrast, CHMP3 lacking most of the C-terminal acidic domain formed dimers under the same cross-linking conditions. This observation provided further support for the model that the ability of CHMPs to interact with each other is controlled by autoinhibitory interactions between their differentially charged N- and C-terminal halves.
The crystal structure of CHMP3-(9–183), a C-terminally truncated form of CHMP3, showed a 70 Å (1 Å = 0.1 nm) long helical hairpin that, together with two short helices, forms a flat four-helical bundle . The crystal structure also revealed two potential dimerization interfaces, and a mutagenic analysis of these interfaces suggested that both are employed in solution. The electrostatic potential map of CHMP3-(9–183) revealed that one side of the flat molecule displays an extensive basic surface charge, with most of the exposed basic residues conserved between human and yeast CHMP3. To elucidate the function of this basic patch, conserved basic residues were mutated in the context of a CHMP3-(1–150) fusion protein, which exhibited a profound dominant-negative effect on HIV-1 budding . Interestingly, the substitution of various combinations of basic residues in CHMP3-(1–150) profoundly diminished the dominant-negative effect. Furthermore, the plasma membrane localization of the CHMP3-(1–150) fusion protein was invariably abolished by the mutations. These results implicated the extensive basic surface cluster of CHMP3 in lipid bilayer interactions.
The function of Alix in HIV-1 budding depends on its interaction with ESCRT-III component CHMP4
Although HIV-1 mutants that lack a functional Tsg101-binding site exhibit severe replication defects in most T-cell lines, the Tsg101- and Alix-binding sites appear to have redundant roles in promoting HIV-1 replication in Jurkat cells. This observation suggested that Alix may, under some circumstances, be able to substitute for Tsg101 in HIV-1 budding. To examine this possibility, Alix was overexpressed to determine whether it can rescue HIV-1 mutants that lack a functional Tsg101-binding site. Remarkably, the profound budding defect of such mutants could be almost completely corrected in this manner [18,34]. The rescue of HIV-1 L domain mutants by Alix was blocked by a dominant-negative ESCRT-III component, indicating that the effect of Alix depended on an intact class E Vps vesicle-formation pathway . As expected, the rescue of HIV-1 budding by Alix was fully dependent on the integrity of the LYPXnL motif near the C-terminus of Gag [18,34], which had previously been identified as the Alix-binding site in p6. These observations confirmed the functional relevance of the interaction between Alix and p6, and provided the first direct evidence that Alix can promote virus budding.
The ability of Alix to rescue Tsg101-binding site mutants also yielded a convenient assay for its function in viral budding. Alix has a modular organization, with an N-terminal Bro1 domain that binds CHMP4 [18,35,36]. To determine the role of the interaction with CHMP4, individual residues in the Alix Bro1 domain were mutated, based on the crystal structures of a yeast homologue of Alix  or of Alix itself . Interestingly, all single amino acid substitutions in the Alix Bro1 domain that abolished the interaction with CHMP4 also eliminated the ability of Alix to rescue HIV-1 budding [18,34]. Thus the interaction of Alix with the ESCRT-III component CHMP4 is required for its function in viral budding. It is noteworthy in this regard that the L domain of EIAV (equine infectious anaemia virus), which binds Alix, can be functionally replaced by fusing a domain of ESCRT-I component Vps28 to EIAV Gag that directly binds ESCRT-III component CHMP6 . Thus the recruitment of ESCRT-III components may be what ultimately drives virus budding.
It recently emerged that the Bro1 domain of Alix also interacts with the NC (nucleocapsid) domain of HIV-1 Gag, which mediates the encapsidation of the viral RNA genome through conserved zinc fingers . Although the interaction between NC and Alix depends on these zinc fingers, it does not require the presence of nucleic acid . Remarkably, overexpression of the isolated Bro1 domain greatly enhanced the release of a minimal HIV-1 Gag construct that retained NC, but lacked the Alix-binding site in p6 . Thus even the isolated Bro1 domain of Alix may possess a membrane-remodelling activity.
Alix also has a central V domain that binds to viral LYPXnL-type L domains, and a C-terminal PRD (proline-rich domain) of approx. 150 amino acids that binds to Tsg101, endophilin, CIN85 (Cbl-interacting protein of 85 kDa) and ALG-2 [35,39]. As expected, the LYPXnL motif-binding site in the V domain is required for the function of Alix in virus budding [18,40]. However, the binding sites for Tsg101, endophilin and CIN85 in the PRD all turned out to be dispensable [18,34]. Nevertheless, the PRD is essential, as even short C-terminal truncations eliminated the ability of Alix to promote HIV-1 budding [18,34]. It is thus possible that additional interaction partners for the Alix PRD that play a role in virus release remain to be identified.
The requirement for ESCRT-I recruitment in HIV-1 budding can be bypassed by a Nedd (neural-precursor-cell-expressed developmentally down-regulated) 4-like ubiquitin ligase
HIV-1 Gag lacks PPXY (Pro-Pro-Xaa-Tyr)-type L domains, which are generally used by simpler retroviruses to promote budding, and are thought to engage Nedd4-type ubiquitin ligases [20,22,23]. It therefore came as a surprise that the overexpression of a particular isoform of Nedd4-2 (also called Nedd4L) allowed an essentially complete rescue of the budding defects of HIV-1 mutants unable to recruit ESCRT-I [27,41]. The various Nedd4 family members all contain an N-terminal C2 domain involved in membrane targeting, followed by up to four WW (Trp-Trp) domains that bind substrates via PPXY motifs, and a HECT (homologous with E6-associated protein C-terminus) domain that mediates the conjugation of ubiquitin to substrates. However, Nedd4-2 is unique because it is mostly expressed without a C2 domain as a consequence of alternative proximal exon usage, yielding an isoform that has been called Nedd4-2s.
Remarkably, the overexpression of the Nedd4-2s isoform increased the release of infectious virus by an HIV-1 mutant lacking a Tsg101-binding site nearly 50-fold, and was clearly more effective than Alix in correcting the characteristic processing defect of the L domain mutant . In contrast, several other Nedd4 family members, including a Nedd4-2 isoform with an intact C2 domain, showed little or no activity. In contrast with Alix, Nedd4-2s could rescue HIV-1 release even in the absence of all known L domains [27,41]. The effect of Nedd4-2s appeared to be specific for HIV-1, because another retrovirus lacking all known L domains was not rescued . Thus the effect of Nedd4-2s appears to depend on a specific determinant in HIV-1 Gag [27,41].
Nedd4-2s retains a small C-terminal portion of the C2 domain, which turns out to be essential for the rescue of HIV-1 budding . C2 domains often function as membrane-targeting modules, but can also mediate protein–protein interactions. Indeed, it appears that the truncated C2 domain of Nedd4-2s is required for the association of the ubiquitin ligase with HIV-1 Gag, because intact Nedd4-2s was incorporated into HIV-1 particles, but a mutant lacking the residual C2 domain was not.
Substrate recognition by Nedd4 family ubiquitin ligases is thought to occur through their multiple WW domains. However, all four WW domains of Nedd4-2s could be disrupted simultaneously without affecting its ability to rescue HIV-1 budding . Thus WW domain-mediated protein–protein interactions do not play a critical role in the rescue of HIV-1 budding by Nedd4-2s. Nevertheless, the effect of Nedd4-2s on HIV-1 budding and Gag processing requires a catalytically active HECT domain, implying that the ubiquitination of an unknown substrate is involved [27,41]. It has recently been reported that a Nedd4 family ubiquitin ligase can promote virus-like particle release through PPXY-type L domains in the absence of viral protein ubiquitination . Also, there is evidence that Nedd4-2s can enhance retroviral budding by stimulating the activity of ESCRT-I [27,41]. Indeed, Nedd4-2s can induce the ubiquitination of ESCRT-I component MVB12B . It is thus tempt-ing to speculate that the enhancement of HIV-1 budding by Nedd4-2s involves the regulatory ubiquitination of ESCRT-I or of another component of the ESCRT pathway.
The evidence available strongly suggests that the assembly of ESCRT-III is regulated via autoinhibitory interactions. It appears that viral L domains ultimately function by inducing the assembly of ESCRT-III components, which may explain why Alix needs to interact with CHMP4 to promote HIV-1 budding. Intriguingly, HIV-1 budding is also strongly enhanced by a Nedd4-like ubiquitin ligase that may activate the ESCRT pathway.
ESCRTs: from Cell Biology to Pathogenesis: Biochemical Society Focused Meeting held at Robinson College, Cambridge, U.K., 26–28 August 2008. Organized and Edited by Katherine Bowers (University College London, U.K.), Juan Martin-Serrano (King's College London, U.K.) and Paul Whitley (Bath, U.K.).
apoptosis-linked gene 2
ALG-2-interacting protein X
associated molecule with the SH3 (Src homology 3) domain of STAM (signal-transducing adaptor molecule)
charged multivesicular body protein
Cbl-interacting protein of 85 kDa
equine infectious anaemia virus
endosomal sorting complex required for transport
homologous with E6-associated protein C-terminus
neural-precursor-cell-expressed developmentally down-regulated
tumour susceptibility gene 101
vacuolar protein sorting
Work is supported by the National Institutes of Health [grant number AI29873].