Evidence for Cooperative and Domain-specific Binding of the Signal Transducing Adaptor Molecule 2 (STAM2) to Lys63-linked Diubiquitin*

Background: The STAM2/Hrs complex is part of the ESCRT-0 (endosomal sorting complexes required for transport) machinery responsible for cargo sorting to the lysosome. Results: Binding of the VHS-UIM construct of STAM2 to Lys63-linked diubiquitin is cooperative with a specific structural organization. Conclusion: Spatial arrangement of the VHS-UIM/Lys63-linked diubiquitin complex probably influences the sorting of Lys63 polyubiquitinated proteins. Significance: Characterization of the VHS-UIM/Lys63-linked diubiquitin complex is crucial for understanding lysosomal degradation. As the upstream component of the ESCRT (endosomal sorting complexes required for transport) machinery, the ESCRT-0 complex is responsible for directing ubiquitinated membrane proteins to the multivesicular body pathway. ESCRT-0 is formed by two subunits known as Hrs (hepatocyte growth factor-regulated substrate) and STAM (signal transducing adaptor molecule), both of which harbor multiple ubiquitin-binding domains (UBDs). In particular, STAM2 possesses two UBDs, the VHS (Vps27/Hrs/Stam) and UIM (ubiquitin interacting motif) domains, connected by a 20-amino acid flexible linker. In the present study, we report the interactions of the UIM domain and VHS-UIM construct of STAM2 with monoubiquitin (Ub), Lys48- and Lys63-linked diubiquitins. Our results demonstrate that the UIM domain alone binds monoubiquitin, Lys48- and Lys63-linked diubiquitins with the same affinity and in the same binding mode. Interestingly, binding of VHS-UIM to Lys63-linked diubiquitin is not only avid, but also cooperative. We also show that the distal domain of Lys63-linked diubiquitin stabilizes the helical structure of the UIM domain and that the corresponding complex adopts a specific structural organization responsible for its greater affinity. In contrast, binding of VHS-UIM to Lys48-linked diubiquitin and monoubiquitin is not cooperative and does not show any avidity. These results may explain the better sorting efficiency of some cargoes polyubiquitinated with Lys63-linked chains over monoubiquitinated cargoes or those tagged with Lys48-linked chains.

The turnover of many membrane proteins is determined through the endocytic pathway, which can either result in their recycling or lysosomal degradation (1)(2)(3). In the case of lysosomal degradation, membrane proteins are sorted into distinctive endosomes known as multivesicular bodies (MVBs). 3 These MVBs fuse with lysosomes, resulting in the degradation of their cargoes. Prior to entering the MVB pathway, cargoes have to be properly tagged by a process called ubiquitination. Ubiquitination occurs through the attachment of a single ubiquitin (Ub), a 76-amino acid protein, to a lysine (monoubiquitination) (4) or several lysines (multi-monoubiquitination) (5) of a target protein, as well as by the attachment of polyubiquitin (poly-Ub) chains (6). Interestingly, depending on the type of ubiquitin chain linkages, tagged proteins are committed to different pathways (7). Although monoubiquitination is a sufficient signal to direct cargoes through the MVB pathway (8,9), in many cases, polyubiquitination by Lys 63 -linked chains is a more efficient signal for cargo sorting (10 -12). Most proteins that are not ubiquitinated follow other pathways and are recycled from the endosome to other cellular compartments (13).
The machinery responsible for committing ubiquitinated cargoes to the MVB pathway is the ESCRT machinery (14), whose most upstream component is ESCRT-0 that is composed of the STAM/Hrs complex (15) in mammalian cells (Vps27/ Hse1 complex in yeast) (16,17). Recognition of the ubiquitin signal is achieved by modular motifs called ubiquitin-binding domains (UBDs) (18). Both STAM and Hrs harbor two UBDs. The VHS (19) and UIM (ubiquitin-interacting motif) (20) domains of STAM, which are connected by a 20-amino acid linker, are key players in ubiquitin recognition and cargo sorting. Deletion of either the VHS or UIM domains of STAM gives rise to a partial loss of function of ESCRT-0 (19,21). These results highlight the importance of both the VHS and UIM domains, and it has recently been proposed that ESCRT-0 as well as the VHS-UIM domains of STAM1 bind avidly to long polyubiquitin chains and show a moderate degree of selectivity for Lys 63 -over Lys 48 -linked chains (21). However, the molecular details of the recognition of Lys 63 -linked diubiquitin (Lys 63 -Ub 2 ) chains by the ESCRT machinery remain poorly understood, and structural studies are required to decipher the underlying mechanisms. Because ESCRT-0 appears to be primarily responsible for cargo clustering (22), we are seeking to understand if any particular structural organization forms the basis for the polyubiquitin chain selectivity of ESCRT-0 and, more specifically, of the VHS-UIM fragment of STAM2. To address this question, we compared interactions of the UIM domain and the VHS-UIM construct of STAM2 with monoubiquitin (mono-Ub), Lys 48 -and Lys 63 -Ub 2 . Here we show that the UIM domain binds mono-Ub, Lys 48 -and Lys 63 -Ub 2 with the same affinity and binding mode, in striking contrast with our previous findings for the VHS domain (23). However, the situation is somewhat different when the UIM and the VHS domains are connected. Our results indicate that interaction of the VHS-UIM construct with Lys 63 -Ub 2 , but not with mono-Ub or Lys 48 -Ub 2 , is cooperative. In addition, site-directed paramagnetic spin labeling data allow modeling of the VHS-UIM/Lys 63 -Ub 2 complex and clearly demonstrate that each domain of STAM2 VHS-UIM interacts with a specific Ub unit in Lys 63 -Ub 2 .
NMR Experiments-To be consistent with our previous study related to VHS (23), the ensemble of NMR experiments was acquired at 288 K where the NMR samples were exchanged into a buffer containing 20 mM sodium phosphate at pH 6.8, 10% D 2 O, 0.02% (w/v) NaN 3 . Assignment and spin relaxation experiments for UIM and VHS-UIM have been carried out on a Varian Inova Unity 600 equipped with a triple-resonance probe and on a Bruker Avance 600 equipped with TXI cryoprobe or TCI cryoprobe.
Protein Resonance Assignment-Backbone resonance assignments for VHS-UIM were obtained using a combination of the following experiments: [ 15  Relaxation Measurements and Analysis-Relaxation measurements including 15 N longitudinal (R 1 ) and transverse (R 2 ) relaxation rates and the 15 N-1 H cross-relaxation rates, via steady-state 15 N{ 1 H}NOE, were performed as previously described (32,33). NMR spectra were recorded with spectral widths of 2000 Hz in the 15 (34,35). The relaxation delay was set to 6 s to allow the bulk water magnetization to return as close as possible to its equilibrium value. All NMR data were processed with NMRpipe (29) and SPARKY (36), and the relaxation rates were extracted using RelaxFit (32). For complexes, the relaxation experiments were performed with a concentration of 200 M for the protein and a saturating concentration of its corresponding ligand.
NMR Titration Studies-Interaction surfaces on 15 N-mono-Ub, 15 N-Ub 2 , 15 N-UIM, and 15 N-VHS-UIM were characterized by means of chemical shift perturbations (CSPs) where a series of 1 H, 15 N-HSQC experiments were recorded upon addition of the (unlabeled) binding partner. To avoid possible aggregation of the proteins, we started from a 15 N-labeled protein at 200 M concentration and added an increasing volume of a concentrated stock of unlabeled ligand protein until reaching saturation. To derive the corresponding binding constant, spectral perturbations were quantified as the combined amide CSPs: ⌬␦ ϭ [(⌬␦ H ) 2 ϩ (⌬␦ N /5) 2 ] 1/2 . The equations used to derive the dissociation constants for the 1:1 or 2:1 stoichiometry models can be found in Ref. 27; the formula used to calculate the effective concentration can be found in the supplemental Materials.
Site-directed Spin Labeling-A paramagnetic spin label, 1-oxy-2,2,5,5-tetramethyl-3-pyrroline-3-methyl)methanesulfonate (MTSL), was covalently attached to Lys 63 -Ub 2 or Lys 48 -Ub 2 on the proximal or the distal Ub T12C at position 12 through the cysteine side chain, as described previously (28). The paramagnetic relaxation enhancement (PRE) effect was measured as a ratio of protein signal intensities in the 1 H, 15 N NMR spectra obtained when MTSL was in the oxidized

STAM2/Lys 63 -linked Diubiquitin Interaction
(paramagnetic) and the reduced (diamagnetic) states (28). The oxidized state spectrum was obtained for 15 N-VHS-UIM in the presence of unlabeled Lys 63 -Ub 2 or Lys 48 -Ub 2 with MTSL attached to the proximal or the distal Ub; the reduced state spectrum was obtained after adding 3 molar eq of ascorbate to that sample. The position of the unpaired electron in MTSL with respect to the protein was reconstructed by fitting the observed PREs using program SLfit (37). The corresponding equations are summarized in supplemental Materials.
Homology Modeling of UIM-The three-dimensional structures of the UIM domain alone and the UIM part in the VHS-UIM construct were obtained by homology modeling. As seen from the pairwise alignment (supplemental Fig. S2A), amino acid sequences of STAM2-UIM and Vps27-UIM1 share 55% identity and 70% similarity. We used the UIM1 domain (38) of Vps27 (PDB code 1Q0V) to model the structure of the UIM domain as well as the UIM part of the VHS-UIM construct. The NMR structure of the VHS domain of STAM2 (PDB code 1X5B) was used to model the VHS part of VHS-UIM. Models were generated by using the Modeler program (39). After alignment of the query and template sequences with Align2D, they were used as input in Modeler. A total of 10 structures were generated for UIM as well as VHS-UIM (supplemental Fig.  S2B).

Mapping the Interaction of UIM with Mono-Ub, Lys 48 -and
Lys 63 -Ub 2 -To compare Ub/poly-Ub binding properties of the isolated UIM domain and the tandem VHS-UIM construct, we first mapped the interactions between UIM and mono-Ub or Ub 2 by means of NMR CSPs. To obtain information from both sides of the complex, we monitored signal shifts in 1 H, 15 N-HSQC spectra of 15 N-labeled UIM, mono-Ub, or Ub 2 upon addition of the corresponding unlabeled binding partner. As supported by the 1 H, 15 N-TROSY and three-dimensional 15 Ndispersed NOESY spectra (see supplemental Figs. S1 and S2), the UIM domain forms an ␣-helix. Chemical shift index (40) predicts an ␣-helical region from Lys 172 to Glu 187 . Upon titration of mono-Ub into 15 N-UIM, several residues experienced significant CSPs, whereas some showed a strong decrease (up to 70% already at the beginning of titration) in signal intensity indicative of intermediate exchange (Fig. 1A). Specifically, strong signal attenuations were detected for the hydrophobic residues Ile 177 , Ala 178 , Ala 180 , Ile 181 , and Leu 185 . These residues form a stretch along one face of the ␣-helix of UIM, flanked on both sides by negatively charged residues Glu 173 , Asp 174 , Asp 176 , and Glu 187 , which show significant CSPs (Fig. 1B). These perturbed residues are in good agreement with those perturbed on other UIM domains upon binding to mono-Ub (38,41). On the mono-Ub side, the CSPs cluster around the canonical hydrophobic patch of Ub, specifically including residues Leu 8 , Ile 44 , Ala 46 , Leu 50 , Val 70 , and Leu 73 , and also including Arg 42 , Lys 48 , Gly 47 , His 68 , and Arg 72 (Fig. 1, C and D).
We performed similar studies to examine and compare UIM binding to Lys 48 -Ub 2 and Lys 63 -Ub 2 . To enable studies of UIM interactions with each individual Ub unit in the chain, the Ub 2 constructs were segmentally 15 N-labeled (see "Experimental Procedures") on either the proximal Ub (that carries the free C terminus) or the distal Ub (whose C terminus is linked to Lys 48 or Lys 63 on the proximal Ub). Our NMR data (Fig. 1, E and F) show that the UIM surface perturbed by either Lys 48 -or Lys 63 -Ub 2 is essentially the same for both chains and similar to the one involved in the mono-Ub binding (Fig. 1, A and B). Likewise, there is a striking similarity between the UIM-binding surfaces on the two Ub units in each chain as well as between the chains (Fig. 1, G and H). Moreover, the perturbed residues in the Ub units in these Ub 2 s are essentially identical to those involved in mono-Ub binding to the UIM (Fig. 1, C and D). These results suggest that the interaction of UIM with the chains occurs via its binding to each individual Ub unit separately and regardless of the chain linkage.
UIM Adopts the Same Mode of Binding with Mono-Ub and Ub 2 -To verify the aforementioned prediction, we set out to quantify the stoichiometry and the strength of these interactions. To monitor the stoichiometry of the complexes studied here, we used transverse spin relaxation rate ( 15 N R 2 ) measurements, as R 2 reflects the rate of overall tumbling (hence the size) of the molecular object under investigation. The molecular mass dependence of R 2 was utilized as a "molecular mass ruler" (calibrated using R 2 data for Ub and Ub 2 ) to assess the mass of a molecular complex and hence the stoichiometry (supplemental Fig. S3B). Measurements for the target protein (UIM, mono-Ub, or Ub 2 ) in the corresponding complexes were carried out under saturating conditions of its binding partner. Residues experiencing fast local dynamics or conformational exchange were removed from the analysis. For mono-Ub, the average R 2 for residues in the secondary structure increased from 7.3 Ϯ 0.4 to 11.1 Ϯ 0.8 s Ϫ1 upon binding to UIM. The R 2 value of UIM was 5.1 Ϯ 0.1 s Ϫ1 in the free state, and increased to 11.5 Ϯ 0.5 s Ϫ1 in the mono-Ub-bound state (see supplemental Fig. S3B). These R 2 values for the UIM/mono-Ub complex correspond to a molecular mass range from 12.0 to 14.5 kDa, in good agreement with the expected molecular mass of 12 kDa for a 1:1 UIM/mono-Ub equilibrium. For the Ub 2 /UIM complexes, we measured an average R 2 value of 21.3 Ϯ 1.4 and 20.6 Ϯ 1.6 s Ϫ1 for the distal Ub in Lys 48 -and Lys 63 -Ub 2 , respectively, compared with the value of 13.3 Ϯ 0.9 s Ϫ1 measured for the same Ub unit in the free form of Lys 63 -Ub 2 . These R 2 values for UIMbound Ub 2 correspond to a molecular mass range from 24 to 28.5 kDa, pointing to a 2:1 (UIM:Ub 2 ) binding stoichiometry (the expected molecular mass is 24 kDa). Thus, our 15 N relaxation data clearly support the model where one UIM binds to one mono-Ub molecule, whereas two UIMs can bind to each Lys 48 -and Lys 63 -Ub 2 .
To quantify the strength of the binding interactions, we derived the dissociation constants from the NMR titration data for each interacting partner in the complexes under investigation. For the UIM/mono-Ub binding equilibrium, the dissociation constant was derived using a 1:1 stoichiometry model, which gave an average K d of 287 Ϯ 36 M (Table 1 and  It is worth noting that despite the higher backbone flexibility on the ps-ns time scale, the UIM part retains its helical fold, as evident from the various spectroscopic data presented above. In the 1 H, 15 N-HSQC spectrum of VHS-UIM, the VHS and UIM parts overlap with their respective spectra as isolated units. This fact provides clear evidence that the VHS and UIM domains do not interact directly in the VHS-UIM construct (see supplemental Fig. S8). Therefore, any perturbation observed in VHS-UIM spectra upon addition of mono-Ub or (poly)Ub is caused by the interaction of VHS-UIM with the ligand.
Our NMR mapping data suggest a specific interaction between VHS-UIM and mono-Ub, where both the VHS and UIM domains are involved in binding, each to a separate ubiquitin molecule. On the VHS-UIM side, the largest perturbations stretch along the ␣2 and ␣4 helices of the VHS domain, which comprise residues Thr 29 -Asp 38 and Leu 75 -Asn 82 , respectively ( Fig. 2A). The UIM domain was also affected by Ub binding, through residues located on one side of the ␣-helix; these include Ala 178 , Ile 181 , Glu 182 , Leu 185 , and Gln 186 that experienced strong signal attenuations (hence intermediate exchange) and Asp 174 , Asp 176 , and Leu 183 that showed significant CSPs. On the mono-Ub side, strong perturbations clustered around the hydrophobic residues Ile 44 , Val 70 , and Ile 13 (see Fig. 2B). Surprisingly, Gly 75 and Gly 76 were significantly perturbed compared with the interactions of mono-Ub with the individual UIM (see above) or VHS domains (23). This suggests additional interactions, involving the C terminus of Ub and possibly with the VHS-UIM linker. The dissociation constant for the VHS-UIM/mono-Ub equilibrium was derived by fitting the CSPs data for each individual residue in the course of titration (supplemental Fig. S9). From the Ub side, the dissociation constant was 102 Ϯ 38 M, and a similar average value of 117 Ϯ 33 M was obtained for the VHS-UIM residues, thus giving a total average K d of 112 Ϯ 35 M (Table 2).
VHS-UIM Construct Binds Differently to Lys 48 -Ub 2 and Lys 63 -Ub 2 -According to the aforementioned results, the isolated UIM has the same dissociation constant and the same mode of binding when forming a complex with mono-Ub, Lys 48 -Ub 2 , or Lys 63 -Ub 2 . By contrast, we have previously reported a different behavior for the isolated VHS domain of STAM2, which binds Lys 48 -Ub 2 and Lys 63 -Ub 2 differently (23). In that case, up to two VHS molecules can bind to Lys 63 -Ub 2 , whereas only one VHS domain can bind to Lys 48 -Ub 2 . This raises an important question: how do the two UBDs, UIM and VHS, interact with Ub 2 chains when they are in tandem?
To gain insights into the interaction of VHS-UIM with Ub 2 , we monitored CSPs on distal and proximal Ub of both Lys 48and Lys 63 -Ub 2 , as well as on VHS-UIM. Any change in the chemical shift (i.e. resonance frequency) reflects a change in the electronic environment of the nucleus under observation. Therefore, similarities in the directions of the signal shifts in the 1 H-15 N spectra upon binding would suggest similarities in the changes of the electronic environment of 1 H nuclei and (separately) of 15 N nuclei in the corresponding complexes, thus similar changes of the chemical environment. On the VHS-UIM side, the CSP pattern for the VHS-UIM/Ub 2 complex is similar to the one seen upon VHS-UIM binding to mono-Ub (Fig. 2, D, E, and I). It is noteworthy that several residues experienced strong signal attenuation during titration, but there were no additional perturbed sites compared with VHS-UIM/mono-Ub binding. Strong signal attenuations were also detected in several residues located around Ile 44 and Val 70 on both Ub units of Lys 48 -and Lys 63 -Ub 2 (Fig. 2, G and H). As seen in Fig. 2, C and G, the perturbation patterns for the proximal and distal Ubs of Lys 48 -Ub 2 show no significant differences from each other or from the perturbations observed in mono-Ub (Fig. 2B) upon binding to VHS-UIM. By contrast, there is a striking difference  between the perturbations in the distal and the proximal Ubs of Lys 63 -Ub 2 (Fig. 2, F and H, and supplemental Fig. S10). It is noteworthy that the CSP profile obtained for the distal Ub in Lys 63 -Ub 2 is in many aspects similar to the one observed upon binding to the isolated UIM (Fig. 1C). This is especially true for residues Leu 8 , Ile 61 , and around Ile 30 , which display significant signal shifts upon binding to UIM but not VHS (23). On the other hand, the CSP pattern observed in the proximal Ub is quite similar to the one for the mono-Ub/VHS complex (23). These results suggest that in the Lys 63 -Ub 2 /VHS-UIM complex, the distal Ub binds the UIM and the proximal Ub binds the VHS domain, whereas in the case of Lys 48 -Ub 2 the binding preferences are not that pronounced.
To further support these observations, the trajectories of the Ub 2 signal shifts (in 1 H-15 N spectra) upon addition of VHS-UIM were compared with those during titration with the isolated VHS or UIM domain (supplemental Fig. S6). Indeed, the 1 H-15 N signals belonging to residues from the distal Ub of Lys 63 -Ub 2 shifted in the same direction upon binding to UIM and VHS-UIM (supplemental Fig. S6). Likewise, the trajectories of the signal shifts in the proximal Ub of Lys 63 -Ub 2 were very similar when binding to VHS-UIM and VHS. The direction of the shifts of the 1 H-15 N signals reflects the ratio of 1 H and 15 N CSPs, which is unique for each residue (N-H bond) and for the changes in its electronic environment induced by binding. Thus, similar directions of the NMR signal shifts indicate similarity in the local intermolecular contacts, further supporting the model in which VHS-UIM interacts with Lys 63 -Ub 2 through specific pairwise contacts between the UIM and distal Ub and the VHS domain and proximal Ub. Unlike Lys 63 -Ub 2 , comparison of the signal shift trajectories for Lys 48 -Ub 2 upon binding to VHS-UIM did not reveal any clear similarities with the shifts induced by the isolated VHS or UIM. However, one has to bear in mind that these conclusions are based largely on empirical results from numerous observations, and have not been clearly quantified by a direct comparison between alterations in the local structure and dynamics at the interface (including solvent accessibility) and the experimental or theoretical/computed chemical shifts.
We then performed 15 N spin relaxation experiments to determine the stoichiometry of the VHS-UIM/Ub 2 complex. In the case of VHS-UIM saturated with Lys 63 -Ub 2 , the measurements yielded a 15 N R 2 value of 31.5 Ϯ 2.5 s Ϫ1 (for residues belonging to VHS) compared with a value of 17.0 Ϯ 1.2 s Ϫ1 in the free state (see above). This increase in R 2 indicates an increase in the overall size (the total molecular mass), consistent with the formation of the VHS-UIM/Lys 63 -Ub 2 complex. This R 2 value corresponds to a molecular mass range of 38 -48 kDa, comparable with the sum of Ub 2 and VHS-UIM masses (ϳ39.5 kDa), thus indicating a 1:1 stoichiometry. After diluting the sample by ϳ2-fold, from 200 to 120 M concentration of VHS-UIM, we performed 15 N spin relaxation experiments. The average R 2 for the well structured part of VHS was 28.4 Ϯ 2.7 s Ϫ1 . This value is somewhat lower than (although within the error bars from) the value determined for the higher concentration, which likely reflects a lesser fraction of the complex due to lower concentrations of both binding partners. This R 2 corresponds to a molecular mass of 35.7-44 kDa and is in good agreement with a 1:1 stoichiometry. We were not able to extract a reliable transverse relaxation rate for the VHS-UIM/Lys 48 -Ub 2 complex due to a significantly faster relaxation decay. This is indicative of a larger size (molecular mass) of the complex and suggests that more than one Lys 48 -Ub 2 molecule can bind to VHS-UIM, perhaps with different populations. These results agree with the CSP mapping data shown above, which indicate a domain-specific "preferred" orientation of the VHS-UIM interactions in complex with Lys 63 -Ub 2 , and absence of clear interdomain interaction preferences in the case of Lys 48 -Ub 2 .

Lys 63 -Ub 2 Exhibits Preferred Binding Mode with VHS-UIM-
To further investigate the binding modes in the Lys 48 -Ub 2 / VHS-UIM and Lys 63 -Ub 2 /VHS-UIM complexes, we used MTSL as a paramagnetic spin label. This approach is used to derive long-distance constraints and identify low-populated states (42,43). The paramagnetic tag causes strong signal attenuation in nuclei that are close in space (Ͻ25 Å) to the unpaired electron of MTSL via the PRE effect. MTSL was attached to a cysteine in Lys 48 -or Lys 63 -Ub 2 introduced through a T12C mutation in either the distal or proximal Ub. The PRE effects were quantified from 1 H, 15 N-HSQC spectra recorded for 15 N-VHS-UIM in complex with Ub 2 carrying the MTSL on the distal or proximal Ub (see "Experimental Procedures"). As shown in Fig. 3, A-D, the PREs arising from MTSL attached to Lys 48or Lys 63 -Ub 2 result in strikingly different patterns of signal attenuations in VHS-UIM.
As obvious from Fig. 3, A and B, the attenuations in VHS-UIM induced by MTSL attached to the proximal or distal Ub of Lys 63 -Ub 2 are strikingly different, especially in the linker region around Ala 160 . Indeed, the linker was affected by MTSL attached to the proximal Ub, whereas MTSL on the distal Ub did not induce any significant perturbation in the same region. Interestingly, when MTSL was attached to the proximal Ub, the signal attenuation pattern in the VHS region was similar to the one seen for the VHS/mono-Ub interaction (23), where residues around Thr 30 were strongly attenuated, whereas the MTSL did not significantly affect the UIM domain (Fig. 3A). Conversely, when MTSL was attached to the distal Ub of Lys 63 -Ub 2 , the main attenuations occurred on the UIM as well as the C terminus of VHS-UIM (Fig. 3B). These observations confirm the above results and point to a particular binding mode where the proximal Ub of Lys 63 -Ub 2 predominantly binds the VHS domain, whereas the distal Ub binds the UIM domain. It should be borne in mind, however, that we cannot exclude that a small fraction of the Lys 63 -Ub 2 /VHS-UIM complexes adopts a different structural organization. Indeed, MTSL attached to the distal or the proximal Ub of Lys 63 -Ub 2 also induced modest attenuations in VHS or UIM, respectively.
A dramatically different pattern of PRE-induced signal attenuations appeared when MTSL was attached to any of the two Ubs in Lys 48 -Ub 2 . In this case, both the VHS and UIM domains were affected by MTSL attached to the distal or the proximal Ub (Fig. 3, C and D). In particular, for MTSL attached to the proximal Ub of Lys 48 -Ub 2 , the PREs induced in the VHS part of VHS-UIM (Fig. 3C) were similar to those seen for the VHS/ Lys 48 -Ub 2 interaction (23). These observations suggest a more complicated molecular organization of the VHS-UIM/Lys 48 -Ub 2 complex than in the VHS-UIM/Lys 63 -Ub 2 complex.
Modeling VHS-UIM/Lys 63 -Ub 2 Interaction-Taking into account all the aforementioned results, we asked the following questions: "how does Lys 63 -Ub 2 accommodate VHS-UIM?" and "is the intervening linker between VHS and UIM flexible enough to allow this interaction." To address these questions, we attempted to model the Lys 63 -Ub 2 /VHS-UIM complex using our PRE data (Fig. 3, A and B). The position of MTSL attached to the proximal Ub with respect to the VHS domain was reconstructed by taking into account the PREs in VHS from Fig. 3A. Likewise, the position of MTSL attached to the distal a Data were fit using a single-site binding equation. b Data were fit using sequential binding model (see supplemental materials) assuming two consecutive binding events, characterized with two binding constants, K d1 and K d2 . Ub with respect to the UIM domain was reconstructed by considering the PREs in the UIM from Fig. 3B. Due to high flexibility of the linker, the PREs in that region were not taken into account.
The modeling was performed using molecular docking with NMR constraints implemented in the Haddock 2.0 program (44,45) (see supplemental Materials). The docking was driven by using a combination of distance restraints derived from PRE data and homology modeling complemented by ambiguous restraints from the CSP data. Based on the results presented above, our model assumed that in the VHS-UIM complex with Lys 63 -Ub 2 , the VHS domain contacts the proximal Ub, whereas the UIM domain contacts the distal Ub. The existing structure of the VHS/mono-Ub complex (23) was used as the starting structure for the VHS/proximal Ub interaction. For the UIM/Ub contact, the structure of the Vps27-UIM/Ub complex (38) was used. The resulting structures were subjected to clustering and details of the docking results are summarized in supplemental Table S1 for the 10 best structures of the best cluster; these structures are presented in Fig. 3, E and F. To evaluate the goodness of our model, the distances between the oxygen of MTSL and each amide proton in the VHS or UIM domains of the modeled complex were back-calculated. As can be seen from supplemental Fig. S11, the back-calculated distances are in good agreement with those derived from our PRE data. It is noteworthy that deviations from the experimental distances seen for the VHS and UIM domains are likely due to (i) the decrease of accuracy of PRE data for distances higher than a given threshold (25 Å), (ii) the fact that the C terminus of UIM is highly flexible (supplemental Fig. S11D), and (iii) the possibly of the presence of low-populated structures that could also affect the helix ␣7 of VHS (supplemental Fig. S11C). Also interesting is the fact that the linker region can approach the MTSL attached to the proximal Ub of Lys 63 -Ub 2 at a distance as close as 7.5 Å (Ala 162 ) and 8.8 Å (Asn 164 ) (Fig. 3E). This observation is in fact in agreement with the moderate CSPs seen on the linker region of VHS-UIM (Fig. 2D), as well as the strong decrease in intensity in the linker due to the MTSL attached on the proximal Ub of Lys 63 -Ub 2 .
Lys 63 -Ub 2 , Unlike Lys 48 -Ub 2 , Binds VHS-UIM Cooperatively-Following the aforementioned results, we wondered if the difference in structural organization between the two VHS-UIM/ Ub 2 complexes is reflected in the titration curves. To address this, we analyzed the NMR titration data collected from the VHS-UIM and Ub 2 sides. As seen in Fig. 4, the binding isotherms for the VHS-UIM/Lys 63 -Ub 2 complex have a sigmoidal shape, in stark contrast with the hyperbolic-shape binding isotherms for the VHS-UIM/Lys 48 -Ub 2 complex.
For the VHS-UIM/Lys 48 -Ub 2 equilibrium, looking at the VHS-UIM side (see Fig. 4F 4, D and E). These values are in excellent agreement with those determined from residues in the VHS domain and suggest that of the two UBDs in VHS-UIM, Lys 48 -Ub 2 predominantly binds to the VHS domain. Recall that Lys 48 -Ub 2 can bind two isolated UIM domains but it can accommodate only one VHS molecule (23). Together with the observation that the titration curves in Fig. 4, D and E (also Fig. 4F), saturate at the [VHS-UIM]/[Ub 2 ] molar ratio close to 1 (pointing to a 1:1 stoichiometry of binding), these results further support the conclusion that the VHS domain is the predominant Lys 48 -Ub 2 -binding domain in VHS-UIM.
The sigmoidal shape of the titration curves in Fig. 4, A-C, suggests that Lys 63 -Ub 2 binding to VHS-UIM is cooperative. Note that specific perturbations observed in both VHS and UIM indicate that both UBDs in VHS-UIM are involved in Lys 63 -Ub 2 binding, consistent with the cooperative character of this interaction. To further examine the binding process, we considered a stepwise sequential model involving two binding events (see supplemental Materials) or a multivalent binding model (46) characterized by the binding of a single Ub to a single UBD (K d,mono ) and a multivalent binding of VHS-UIM to Lys 63 -Ub 2 (K D,mv ). The various K d values obtained from each side of the complex are summarized in Table 2 for the sequential model. The average dissociation constants for the VHS-UIM/Lys 63 -Ub 2 complex are K d1 ϭ 41 Ϯ 11 and K d2 ϭ 7 Ϯ 4 M (for the first and second binding events, respectively). When using the multivalent binding model, the residuals of the fit were twice as high compared with the sequential model, using a three-parameter fit for both models. Moreover, the multivalent binding model could not reproduce the sigmoidal shape of our titration data, indicating that the multivalent binding model is not suitable in our case. In case of a two-step binding model, we can hypothesize that binding of either VHS to the proximal Ub or UIM to the distal Ub of Lys 63 -Ub 2 increases the probability that the other UBD of the same VHS-UIM molecule binds to the second Ub unit of the same Ub 2 chain. For such a molecular model, the effective concentration or probability density for the end to end length of the linker vector can be modeled as a polymer chain (47, 48) (see supplemental Materials). By considering a 20-amino acid linker and a 23-Å end to end linker length in the bound state, we estimated an effective concentration of ϳ10 mM, comparable with the multivalent binding of RAP80-tUIM to Lys 63 -Ub 2 (46). In addition to stepwise binding and high local effective concentration, we are interested in determining the molecular origins of this cooperative effect. To gain insights into this issue, we performed heteronuclear 15 N{ 1 H}-NOE experiments on VHS-UIM upon saturation by Lys 63 -Ub 2 (supplemental Fig. S12). Compared with the free state of VHS-UIM, we observed that NOE values increase in the linker and UIM regions upon binding to Lys 63 -Ub 2 , suggesting that the binding of UIM to the distal unit of Lys 63 -Ub 2 may induce stabilization of its helical structure and a higher rigidity. This conclusion is further supported by the fact that the UIM resonances in the 1 H, 15 N-HSQC spectra became more widely dispersed upon addition of Lys 63 -Ub 2 .

DISCUSSION
The human STAM2 protein is involved in lysosomal degradation and associates with Hrs to form the ESCRT-0 component of the ESCRT machinery. STAM2 possesses two UBDs:

STAM2/Lys 63 -linked Diubiquitin Interaction
the UIM and VHS domains. Deletion of one of these domains or mutation alters the transport of ubiquitinated cargoes (19,21,49,50). Additionally, ESCRT-0 binds polyubiquitin chains with high avidity and seems to have a preference for Lys 63 -linked chains (21). For instance, Lys 63 -linked ubiquitination is required for MVB sorting of Gap1, CPS, and probably other cargoes (12). Because it has been reported that the binding of ESCRT-0 showed a modest increase in affinity for Lys 63 -Ub 2 over Lys 48 -Ub 2 and "linear," head-to-tail linked chains (NC-Ub 2 ) (21), it is likely that other factors are responsible for the specificity of Lys 63 -linked chains toward cargo sorting. We previously characterized the interaction of an isolated VHS domain with mono-Ub, Lys 48 -and Lys 63 -Ub 2 and reported that Lys 63 -and Lys 48 -Ub 2 chains bind VHS via different binding modes (23).
In the present study, we first examined whether the UIM domain of STAM2 also exhibits such structural features when binding to Lys 48 -and Lys 63 -Ub 2 chains. A UIM was originally identified in the S5a subunit of the 26 S proteasome and is often found in proteins involved in the proteasomal and lysosomal degradation (51). UIMs fold into a single ␣-helix and are embedded as independent domains into various modular proteins. Our NMR mapping revealed that the Ub-binding surface is located on one face of the helical region of STAM2 UIM (supplemental Fig. S13B) and involves a set of conserved hydrophobic residues (supplemental Fig. S13A). These findings are in good agreement with the Vps27-UIM1/mono-Ub and STAM-UIM/mono-Ub structures (38,41). Using NMR titration experiments, we determined the STAM2-UIM/mono-Ub dissociation constant of 287 Ϯ 33 M, which implies that the UIM domain of STAM2 binds to mono-Ub weaker than VHS (23). Furthermore, our study revealed that Lys 48 -and Lys 63 -Ub 2 show comparable affinities for UIM (Table 1) and can accommodate up to two STAM2-UIM molecules per chain. A similar structural organization has already been seen in the case of the UIM2 domain of S5a (52). According to our results, the UIM domain of STAM2 does not discriminate between mono-Ub, Lys 48 -and Lys 63 -Ub 2 .
Our data suggest that the UIM adopts the same mode of binding to mono-Ub, Lys 48 -and Lys 63 -Ub 2 chains. Several lines of evidence support this conclusion. First, the trajectories of the NMR signals in the course of titration with UIM are similar for mono-Ub and the distal and proximal Ubs of Lys 48 -Ub 2 and Lys 63 -Ub 2 chains (see supplemental Fig. S6). Second, the perturbation maps (i.e. residues showing strong CSPs and/or signal attenuations as a result of UIM binding) are similar for mono-Ub, Lys 48 -and Lys 63 -Ub 2 chains. Based on the present results on UIM binding and previous work on VHS (23), we can rule out the possibility that the individual UIM or VHS domains are solely responsible for the better affinity of STAM2 for Lys 63 -Ub 2 .
To determine the impact of having two UBDs in tandem (as in STAM2) on binding to ubiquitin and polyubiquitin chains, we characterized the interaction of the VHS-UIM construct from STAM2 with mono-Ub, Lys 48 -and Lys 63 -Ub 2 chains. Our titration experiments suggest that both UBDs in VHS-UIM bind mono-Ub independently and with affinities similar to those observed for the isolated VHS and UIM domains (see Lange et al. (23) and Table 2). Note that VHS binds Ub stronger than UIM does. Switching from mono-Ub to Ub 2 , the VHS-UIM/Lys 48 -Ub 2 interaction gives a somewhat more complicated picture of the equilibrium, where various intermolecular arrangements (and perhaps with different populations) can exist. The VHS-UIM construct binds Lys 48 -Ub 2 with the same affinity as mono-Ub (average K d ϭ 115 Ϯ 41 M), and thus does not bind avidly to Lys 48 -Ub 2 chains. As shown above, the isolated VHS and UIM exhibit different stoichiometry in binding to Lys 48 -Ub 2 . Then one might hypothesize that, due to the steric hindrance introduced by the VHS domain, it is unlikely that the UIM domain of VHS-UIM could bind another Ub unit of the same Lys 48 -Ub 2 chain. An attempt to model the VHS-UIM/Lys 48 -Ub 2 complex with a 1:1 stoichiometry supports this hypothesis. Indeed, the resulting structure is in disagreement with our CSP data and would induce a severe bending of the VHS-UIM linker (supplemental Fig. S14). In addition, it should be borne in mind that Lys 48 -Ub 2 is in dynamic equilibrium between a closed conformation, in which the functionally important hydrophobic patch residues of each Ub unit are sequestered at the Ub/Ub interface, and one or more open conformations (24,37). VHS-UIM binding would require opening of the interface to allow contacts between the hydrophobic residues of the two Ub units and VHS-UIM. In such a situation, the molecular recognition process would likely proceed through a conformational selection mechanism, namely selection of the right (open) conformation of Lys 48 -Ub 2 . At neutral pH the closed state is predominantly populated (ϳ85%), and the time of interconversion between the closed and open conformations, estimated by NMR to be ϳ10 ns, is comparable with the overall tumbling time (37). Analysis of our relaxation data indicates that UIM and VHS tumble essentially independently from each other with respective rotational correlation times of 3.8 Ϯ 0.5 and 15.0 Ϯ 0.4 ns, respectively. Thus, slower tumbling of the VHS domain could be a limiting factor in the recognition and binding to the open state of Lys 48 -Ub 2 , possibly explaining the less efficient sorting of Lys 48 -polyubiquitinated cargoes by ESCRT-0.
A strikingly different picture arises in the case of VHS-UIM interaction with Lys 63 -Ub 2 , where the VHS and the UIM domains preferentially bind to the proximal and the distal Ub of Lys 63 -Ub 2 , respectively. Lys 63 -Ub 2 adopts an extended conformation (27,53) with the hydrophobic patches exposed to the solvent; this allows the interaction with the VHS and UIM domains without the need to compete with the Ub/Ub interaction as in Lys 48 -Ub 2 . Interestingly, the titration curves corresponding to the interaction of VHS-UIM with Lys 63 -Ub 2 have a sigmoidal shape. This is likely to result from two binding events and a cooperative effect, thus demonstrating a clear difference in the binding of mono-Ub, Lys 48 -Ub 2 , and Lys 63 -Ub 2 to VHS-UIM. We derived two dissociation constants (average K d1 ϭ 41 Ϯ 11 M K d2 ϭ 7 Ϯ 4 M) that reflect an increase in the affinity of VHS-UIM for Lys 63 -Ub 2 compared with Lys 48 -Ub 2 . Assuming that VHS binds first and UIM second, it appears that the binding of VHS to Lys 63 -Ub 2 is only slightly stronger than to Lys 48 -Ub 2 or mono-Ub, whereas the UIM binds ϳ40 times stronger. This agrees with the notion that the presence of multiple UBDs (in tandem) can greatly enhance the affinity for binding partners and/or induce a preference for a given polyubiquitin chain (54). From a mechanistic point of view, the binding can be envisioned to occur stepwise by first binding one of the UBDs to a Ub unit. Upon binding, the first UBD becomes part of the complex, and the binding of the second UBD becomes intramolecular. One can assume that one of the roles of the flexible linker is to keep the UBD in proximity, thus increasing its effective local concentration to ϳ10 mM. This value is of the same order of magnitude as the one calculated for the RAP80-tUIM/Lys 63 -Ub 2 (ϳ12 mM) or S5A/Lys 48 -Ub 2 (ϳ3 mM) complexes (46) and reflects the dependence of the effective local concentration on the length of the inter-UBD linker. In addition to stepwise binding and increasing the local effective concentration, one has to take into account the cooperative nature of the VHS-UIM/Lys 63 -Ub 2 interaction. A possible mechanism to explain the cooperativity is that the binding of VHS to the first Ub unit would position UIM favorably for interaction with a nearby Ub, leading to the cooperative stabilization of the UIMs helix. The difference in tumbling between the VHS and UIM in VHS-UIM could have important consequences regarding molecular events, as the initial binding of VHS to 1 Ub unit can be followed by a rapid reorientation (on the nanosecond time scale), binding, and subsequent stabilization of the UIM helix. Binding of the first UBD to a Ub unit will result in a loss of its translational and rotational degrees of freedom (55), hence reducing the conformational entropy. However, the flexibility of the linker as well as the cooperative stabilization of the UIM helix can overcome the entropic penalty. One has to keep in mind that flexibility has an important influence on the thermodynamics of binding and may both favor and disfavor association (56). In the light of our results, the differences between the binding of VHS-UIM to mono-Ub, Lys 48 -Ub 2 , and Lys 63 -Ub 2 could be one explanation for the bet-ter sorting efficiency of Lys 63 -compared with Lys 48 -polyubiquitinated targets in the context of the full ESCRT-0.
A polypeptide chain linking two successive UBDs as well as the nature of the UBDs could favor a specific ubiquitin chain linkage. Among the few existing structures reporting the interaction of multiple UBDs with polyubiquitin chains, S5a, which is involved in proteasomal degradation, contains two UIMs and binds preferentially to Lys 48 -Ub 2 . Moreover, the UIMs of S5a bind the two Ub units simultaneously and with UIM1 showing a 3:1 preference for the distal Ub of Lys 48 -Ub 2 possibly reflecting specific interactions with the Ub-Ub linker (57). Rap80 is another protein that harbors two UIM domains linked in tandem (tUIM). The tUIM fragment specifically recognizes Lys 63 -Ub 2 chains where the length of the linker is critical for binding with high affinity. By engineering different linker lengths, Sims and Cohen (54) demonstrated that the linker length and composition modulated the affinity of RAP80-tUIM for Lys 63 -Ub 2 in a periodic fashion. Changes in the length of the helical linker not only modify the distance between the two UIMs, but also the relative orientation of their Ub-binding surfaces. Thus, the Rap80 linker defines selectivity through domain positioning rather than specific contact with the Ub-Ub linkage. Another example is given by NEMO (NFessential modulator) that forms a heterodimer, which preferentially binds to NC-Ub 2 via its UBAN domain (ubiquitin binding in ABIN and NEMO) (58,59). The NEMO dimer accommodates two Ub 2 s, whereas both NEMO protomers contribute to the binding to each Ub 2 . In contrast to Rap80, NEMO specifically recognizes the linker region of head to tail chains and can distinguish between NCand Lys 63 -linked chains. These examples demonstrate that an array of homologous UBDs can define chain linkage selectivity. Therefore, not only the combination of UBDs but also the length of the intervening linker and the arrangement of the UBDs could form the basis for linkage-specific polyubiquitin recognition. Additionally, the combination of heterologous UBDs of different sizes, like in STAM2, could influence the discrimination between different polyubiquitin chain linkages. Other factors like oligomerization, compartmentalization, and concentration of UBD-containing proteins could contribute to and possibly enhance/complicate the discrimination between different Ub chains.
In the context of the full ESCRT machinery, STAM2 and Hrs form a heterodimer/tetramer (16,60) through their GAT domain to form the ESCRT-0 component. The fact that Lys 63linked polyubiquitin chains now emerge as a specific signal for protein sorting into the MVB pathway (12,61,62) could be linked to different factors, one of them being the higher affinity and cooperativity of the STAM2 VHS-UIM domains for Lys 63 -Ub 2 chains. From the general view of the ESCRT-0 complex, two different models have been proposed. In one of them, cargoes are passed sequentially from ESCRT-0 to ESCRT-I, -II, and -III. If one considers that ESCRT-0 originally possesses four UBDs being part of STAM (VHS and UIM) and Hrs (VHS and DUIM), one can wonder how cargoes carrying Lys 63 -Ub 2 chains can be transferred to the ESCRT-I complex, with a UEV domain of Tsg101 that binds to mono-Ub with a K d as high as 510 M (63). In contrast, our data tend to support a second model where UBDs within the different ESCRT complexes cooperate to increase the binding affinity for ubiquitinated cargoes and work as a "supercomplex" (64), with ESCRT-0 being the main complex to cluster ubiquitinated cargoes (65).