Quantitative Membrane Proteomics Reveals New Cellular Targets of Viral Immune Modulators

Immunomodulators of pathogens frequently affect multiple cellular targets, thus preventing recognition by different immune cells. For instance, the K5 modulator of immune recognition (MIR2) from Kaposi sarcoma–associated herpesvirus prevents activation of cytotoxic T cells, natural killer cells, and natural killer T cells by downregulating major histocompatibility complex (MHC) class I molecules, the MHC-like molecule CD1, the cell adhesion molecules ICAM-1 and PECAM, and the co-stimulatory molecule B7.2. K5 belongs to a family of viral- and cellular-membrane-spanning RING ubiquitin ligases. While a limited number of transmembrane proteins have been shown to be targeted for degradation by this family, it is unknown whether additional targets exist. We now describe a quantitative proteomics approach to identify novel targets of this protein family. Using stable isotope labeling by amino acids, we compared the proteome of plasma, Golgi, and endoplasmic reticulum membranes in the presence and absence of K5. Mass spectrometric protein identification revealed four proteins that were consistently underrepresented in the plasma membrane of K5 expression cells: MHC I (as expected), bone marrow stromal antigen 2 (BST-2, CD316), activated leukocyte cell adhesion molecule (ALCAM, CD166) and Syntaxin-4. Downregulation of each of these proteins was independently confirmed by immunoblotting with specific antibodies. We further demonstrate that ALCAM is a bona fide target of both K5 and the myxomavirus homolog M153R. Upon exiting the endoplasmic reticulum, ALCAM is ubiquitinated in the presence of wild-type, but not RING-deficient or acidic motif–deficient, K5, and is targeted for lysosomal degradation via the multivesicular body pathway. Since ALCAM is the ligand for CD6, a member of the immunological synapse of T cells, its removal by viral immune modulators implies a role for CD6 in the recognition of pathogens by T cells. The unbiased global proteome analysis therefore revealed novel immunomodulatory functions of pathogen proteins.


Introduction
Viral and bacterial virulence often correlates with the pathogen's ability to modulate the host's immune response. One pathway that is frequently targeted by intracellular pathogens, particularly viruses, is antigen presentation by major histocompatibility complex (MHC) class I molecules. By interfering with MHC I expression, transport, or peptide loading, viruses become invisible to cytotoxic T cells. Interestingly, many viral gene products that interfere with MHC I also target other immunologically relevant host cell proteins. For instance, US3 and US11 of human cytomegalovirus affect surface expression of both MHC I and MHC II [1,2]. The murine cytomegalovirus protein m152 causes mislocalization of both MHC I and Rae-1, an unrelated molecule that is a ligand for the activating receptor NKG2D found on natural killer cells and T cells [3,4]. Finally, the HIV-1 proteins nef and vpu both downregulate MHC I and the T cell co-receptor CD4 [5][6][7][8].
This multi-functionality is likely advantageous for the virus since it interferes with recognition by different immune cells and might counteract negative side effects of MHC I downregulation such as activation of natural killer cells. These observations, however, also raise the question of whether the currently known host cell proteins are the only targets of a given viral immune regulator or whether additional targets exist. To address this question experimentally, we selected the K5 protein of Kaposi sarcomaassociated herpesvirus (KSHV), a particularly promiscuous member of a family of viral immune modulators. KSHV encodes two homologous members of the K3 family of viral immune modulators, K5 and K3 [9,10]. These proteins display an amino-terminal RING-CH domain facing the cytoplasm followed by two membrane-traversing domains, resulting in a type III transmembrane topology [11]. K3-type immune modulators are also found in several poxviral genomes [12,13] and are most likely derived from host genes since many eukaryotic organisms, including humans, contain similar proteins, termed membrane-associated RING-CH (MARCH) proteins, or c-MIR [14,15]. Both viral and mammalian members of this family act as RING-type ubiquitin ligases (RING-E3s) and mediate the ubiquitination of lysines or cysteines in the cytoplasmic tail of transmembrane proteins [16,17]. In most cases, ubiquitination occurs post-endoplasmic reticulum (ER), and ubiquitinated transmembrane proteins are endocytosed, sorted to multivesicular bodies (MVBs), and degraded in lysosomes [9,10,12,14,18]. ERassociated proteasomal degradation has also been observed for the K3 protein of murine herpesvirus 68 [16] and for some substrates of KSHV-K5 [19].
All viral K3 family proteins studied so far, and two of the human MARCH proteins, target MHC I for degradation. Several K3 family proteins target other immunoreceptors in addition to MHC I. For example, K5 also downregulates CD1, ICAM-1, PECAM, and B7.2 [19][20][21][22], and M153R also targets CD4 [12]. While the rules for substrate selection have yet to be established, it seems evident that this protein family has evolved to target subsets of transmembrane proteins. These proteins are thus an excellent model for viral immune modulators that have more than one cellular target. In addition, these proteins are representative for other ubiquitin ligases that target certain groups of proteins, not all of which might be known [23].
K5 was initially identified by screening the KSHV genome for gene products inhibiting MHC I expression [9]. Further targets for K5 were then revealed by monitoring the expression of a small set of cell surface proteins, particularly those involved in T cell recognition of infected cells [19][20][21]. These experiments, however, did not rule out that expression of other cellular proteins is inhibited by K5. Ideally, to determine whether additional targets exist, the entire cell proteome would be studied in the presence or absence of K5.
A method to absolutely quantify the entire protein complement of a cell has yet to be developed; however, recent advances in quantitative proteomics using stable isotope labeling allow the relative quantification of proteins in different samples by mass spectrometry (MS) [24].
In cell culture, the preferred method of comparative quantitative proteomics is to metabolically incorporate heavy or light forms of specific amino acids. This technique is known as stable isotope labeling with amino acids in cell culture (SILAC) [25,26]. SILAC incorporates naturally occurring amino acids (C 12 and N 14 ) into one sample and isotopically labeled amino acids (C 13 and N 15 ) into another sample. The ratio of the ''light'' to ''heavy'' peptide can then be determined upon MS analysis of the combined samples. SILAC has been successfully used in yeast, plant, and mammalian cells to measure relative protein abundance [27][28][29].
K5 is a transmembrane protein that localizes to the ER, Golgi, and plasma membranes [11,19]. Since K5 is known to degrade other transmembrane proteins, we focused on identifying proteins that were reduced upon K5 expression in ER, Golgi, and plasma membrane fractions. In three independent repeat experiments we observed that MHC I, activated leukocyte cell adhesion molecule (ALCAM), bone marrow stromal antigen 2 (BST-2), and Syntaxin-4 were consistently reduced in plasma membranes of K5-expressing cells, whereas only K5 itself was increased in any fraction. While reduced MHC I expression was expected, none of the other proteins were previously known to be affected by K5. Importantly, downregulation of each of these proteins by K5 was confirmed in independent experiments and by independent methods, thus validating the quantitative proteomics approach as a new method for identifying cellular proteins affected by viral immunomodulators.

Membrane Preparation and Mass Spectrometric Analysis
Since all previously described targets of K5 were type I transmembrane glycoproteins, we hypothesized that novel targets of K5 were most likely to be found in the membrane fraction. Most targets of K5 are ubiquitinated in a post-ER compartment and then sorted to endosomes in a clathrindependent process [30]. We recently showed, however, that newly synthesized CD31/PECAM is degraded by K5 prior to ER exit by ER-associated proteasomal degradation [19]. It is thus conceivable that novel targets could be eliminated from either the plasma membrane or from intracellular membranes, particularly within the exocytic pathway. For these reasons, we wanted to measure changes in the proteome of the plasma membrane, the Golgi membrane, and the ER membrane.
To measure changes in the membrane proteome we used stable-isotope-labeled amino acids coupled with mass spectrometric analysis. Since isotope labeling incorporates differentially weighted tags directly into proteins, this method allows the fractionation of membranes after combination of samples, thus minimizing variation due to sample preparation. The weight differential is then used to discriminate between peptides derived from each sample and to determine relative protein abundance by mass spectrometric analysis. Previous work has demonstrated that isotopically labeled amino acid analogs are indistinguishable from the natural product with respect to supporting cell growth in tissue

Synopsis
Viral immune modulators often target multiple cellular proteins for destruction. Presumably, this strategy enables viral pathogens to optimize evasion of multiple immune responses. To systematically identify such host cell targets in an unbiased fashion, Bartee et al. applied recently developed quantitative proteomics methods to identify novel targets for K5. K5 belongs to a family of viral ubiquitin ligases found in gamma-herpesviruses and poxviruses that target multiple cellular transmembrane proteins for destruction. Using stable isotope labeling combined with tandem mass spectrometry, the authors compared the abundance of proteins in membrane preparations from cells that expressed K5 to that in cells without K5. In their experiments, three novel membrane proteins (BST-2, Syntaxin-4, and ALCAM) were consistently found in lower abundance in K5-expressing cells. Importantly, the authors were able to confirm the K5-dependent downregulation of all of these proteins in independent experiments and by independent methods. ALCAM was chosen for a more in-depth analysis to firmly demonstrate that this protein is downregulated by K5 in a manner similar to other known targets. This proof-of-principle study demonstrates that novel targets of viral immune modulators can be identified with quantitative proteomics.
culture [25]. For our experiments, we added both N 15 / 6 C 13labeled leucine and 2 N 15 / 6 C 13 -labeled lysine to HeLa cells. Since trypsin cleaves after lysines, and almost 70% of predicted tryptic peptides in the human genome contain at least one leucine, this double labeling procedure ensures that most tryptic peptides will contain at least one isotopic amino acid [26]. HeLa cells were grown in labeling medium for 5 d prior to expression of K5 via adenovirus transduction. Due to rapid doubling of HeLa cells, this time frame allowed for the complete incorporation of isotopic amino acids (unpublished data). Control cells were grown in medium containing natural amino acids and transduced with the same multiplicity of infection of control adenovirus. K5-mediated downregulation of MHC I was confirmed by flow cytometry 24 h postinfection (unpublished data). To minimize variation due to biochemical procedures, plasma membrane, ER, and Golgi fractions were generated by sucrose gradient separation after lysis of the combined labeled and unlabeled cells (Figure 1). To further control for technical and biological variation, we repeated each experiment three times. Tryptic peptides from the resulting fractions were separated by two sequential chromatography steps (strong cation and reverse phase) prior to analysis by MS using an LCQ ion-trap instrument in datadependent MS/MS mode. Tandem mass spectra of dissociated peptides were used to search the human subset of the UniProt database, which was spiked with the viral K5 sequence using SEQUEST [31]. To discriminate true assignments of MS/MS spectra to peptide sequences from false assignments, we applied PeptideProphet software using a score of 0.85 as the cutoff value for each identified peptide [32].
For each fraction we identified 500-700 unique proteins by at least one peptide in at least one experiment (Figure 2A). Of these, 100-150 proteins were identified in all three experi-ments, a finding that is consistent with previous reports [33]. To validate the membrane preparations, we analyzed proteins that were identified in all three replicate experiments of each fraction with respect to their subcellular localization ( Figure  2B). The plasma membrane fraction contained a large percentage of predicted cell surface proteins. The ER and the Golgi fraction contained both bona fide ER-and Golgiresident proteins and many plasma membrane or transmembrane proteins, presumably en route to the cell surface. In addition, all fractions contained significant numbers of proteins predicted to localize to other intracellular organelles such as the nucleus, cytoplasm, or mitochondria. These proteins are likely derived from contaminating protein fractions. Since K5 is not expected to affect proteins in these compartments, these proteins serve as internal controls.

Differentially Expressed Proteins
To identify proteins whose expression was altered by K5, we analyzed proteins identified in all three experiments using the program ASAPRatio [34]. ASAPRatio calculates expression ratios based on peak intensities such as those shown in Figure 2C. Since a 1.5-fold cutoff was previously shown to reflect a significantly different protein level [26], we considered proteins differentially expressed if they displayed higher than a 1.5-fold change in either direction. However, instead of limiting our analysis to proteins identified by multiple peptides in a single experiment as done previously [26], we accounted for biological variation by focusing on proteins that changed in the same direction in all three experiments, regardless of the number of peptides identified in either experiment. Using these criteria, only one protein changed in the ER and Golgi fractions. This protein was K5 itself, which was scored as  Approximately 500-700 proteins were identified by at least one peptide in each fraction; however, fewer than 150 proteins per fraction were identified in all three replicates. Of these, only five proteins changed more than 1.5-fold in the plasma membrane (PM) fraction, while only one protein changed more than 1.5-fold in all three replicates of either the Golgi or ER fractions. (B) Proteins present in all three replicates of each fraction were analyzed for their predicted subcellular distribution using annotations in the Swiss-Prot proteomics database. (C) Comparison of signal intensities obtained for selected peptides from the indicated proteins. The red line indicates the actual raw data recovered from the mass spectrometer; the blue line is a hypothetical best fit line drawn to help analyze elution time of the light and heavy peptides. The orange arrows indicate the time at which the initial MS/MS scan was initiated. Peptides derived from unaffected proteins, such as CD44, display a similar intensity for both the light and heavy isotopes. Peptides derived from differentially expressed proteins, such as BST-2, show clearly different intensity peaks for the light and heavy peptides. Note the slightly different scale for intensity of each peptide. DOI: 10 ''upregulated'' since it was present only in the transduced samples. K5 was also scored as upregulated in the plasma membrane fraction of K5-expressing cells. Additionally, four proteins were downregulated in all three plasma membrane samples ( Figure 2A; Table 1). Among this list were peptides mapping to the HLA-A01 allele, which were present in significantly lower amounts in the plasma membrane fraction, but not in the ER or Golgi fractions. The lack of statistically significant MHC I downregulation in the ER fraction is consistent with the observation that MHC I assembly in the ER is not affected by K5 [9,10]. Since both internal controls-transduced K5 as well as the K5 target protein MHC I-were successfully identified as displaying altered ratios, we anticipated that the three remaining proteins in the plasma membrane fraction had a high chance of being true positive results.
Each of these three proteins was not previously identified as downregulated by K5: BST-2 (also known as HM1.24 and CD316), Syntaxin-4, and ALCAM (CD166). To independently confirm the downregulation of these proteins, we obtained specific antibodies for each and analyzed their expression in the presence or absence of K5.

BST-2
BST-2 (Hs.118110) is a protein of largely unknown function that has been implicated in normal as well as malignant B cell differentiation [35,36]. BST-2 displays a very unusual topology in that it contains both an amino-terminal transmembrane domain and a carboxy-terminal GPI linker [37]. Despite this topological difference and lack of relationship to other K5 targets, BST-2 was consistently scored as the most strongly underrepresented protein in K5-expressing cells (Table 1). A dramatic reduction of BST-2 expression was confirmed in immunoblot of Ad-K5-infected HeLa cells, whereas BST-2 was unaffected in Mock-or Ad-WT-infected cells ( Figure 3A). BST-2 expression was also strongly reduced upon adenovirus-mediated expression of the human K5 homolog MARCH-VIII, but was less affected by the HIV protein vpu, which recruits a cellular ubiquitin ligase. To determine whether downregulation of BST-2 required the ubiquitin ligase activity of K5, we transfected a catalytically inactive mutant of K5 (K5-RING) that lacks two critical cysteines in the K5-RING domain [38], and monitored BST-2 expression by immunoblot. While transiently transfected K5 was still able to downregulate BST-2, the K5-RING construct showed no effect despite significantly higher levels of expression ( Figure 3B). Whether this reduction of BST-2 levels is a direct consequence of K5 ubiquitinating BST-2, or an indirect consequence of the downregulation of other proteins by K5, has not yet been established. However, these experiments clearly confirm that BST-2 protein levels are reduced in K5-expressing cells and demonstrate that this reduction requires catalytically active K5.

Syntaxin-4
The second protein identified as downregulated by K5 was Syntaxin-4 (Hs.83734). Syntaxins belong to the target-membrane-associated soluble N-ethylmaleimide fusion protein attachment protein receptor (t-SNARE) family and are responsible for correctly targeting and fusing intracellular vesicles during transport [39]. Syntaxin-4 is a plasma membrane-localized t-SNARE and mediates docking of transport vesicles at the cell surface. Interestingly, Syntaxin-6, a Golgi-localized t-SNARE, was previously observed to associate with MARCH-II and MARCH-III, human homologs of K5 [40,41]. To confirm the reduction of Syntaxin-4 protein in K5-expressing cells and to determine whether K5 or any of the MARCH proteins associate with Syntaxin-4, we performed immunoblots and immunofluorescence experiments. Reduced levels of Syntaxin-4 were observed in K5-expressing HeLa cells, thus confirming the mass spectrometric results ( Figure 4A). Interestingly, Syntaxin-4 levels were also reduced upon expression of MARCH-VIII, but not upon expression of vpu, suggesting that Syntaxin-4 is specifically targeted by transmembrane RING-CH proteins. In immune fluorescence analysis of control cells, Syntaxin-4 was observed at the plasma membrane. Expression of K5 led to a decrease in the overall intensity of Syntaxin-4, but there was no obvious co-localization ( Figure 4B). In contrast, significant co-localization was observed upon transfection of all MARCH proteins, except for MARCH-III and a naturally occurring truncated splicing product to MARCH-IX (''ÀRING'' in Figure 4B) [14]. Moreover, Syntaxin-4 seemed to be enriched in the subcellular compartments corresponding to the subcellular localization previously shown for the MARCH proteins [14]. Thus, MARCH-IV and MARCH-VIII relocated Syntaxin-4 to the Golgi and endosomes, respectively. Since these results are similar to the previously reported binding of Syntaxin-6 to MARCH-II and MARCH-III [40,41], it seems highly likely that Syntaxin-4 also interacts with the MARCH proteins with which it co-localizes.
The lack of co-localization of Syntaxin-4 with K5 could be due to K5 degrading Syntaxin-4. However, Syntaxin-4 levels are also reduced in MARCH-VIII-expressing cells, yet MARCH-VIII clearly associates with Syntaxin-4. Alternatively, it could be that only the minor fraction of K5 that traffics to the plasma membrane [19] interacts with Syntaxin-4, whereas the major, ER-localized population of K5 molecules is unable to do so. To determine whether ER exit was the determining factor for co-localization with Syntaxin-4, we generated truncated versions of MARCH proteins that were unable to leave the ER. Previous work has demonstrated that a carboxyterminal PDZ motif in MARCH proteins is required for their forward transport since deletion of the final four amino acids resulted in ER retention of MARCH-II and MARCH-III [40,41]. While we could not confirm such a role for the carboxy-terminal 45 amino acids in MARCH-VIII, we did observe that MARCH-VIII truncated by 62 or 74 amino acids was unable to exit the ER ( Figure 4C). Importantly, ERresident MARCH-VIII mutants did not interact with Syntaxin-4. This raised the possibility that Syntaxin-4 regulates ER exit of MARCH-VIII by binding within the 16 residues that differ between the D46 and D62 mutants. Small interfering RNA knockdown of Syntaxin-4, however, had no effect on MARCH-VIII localization (unpublished data), so we consider this hypothesis unlikely. Therefore, we conclude that Syntaxin-4 associates only with MARCH proteins that have left the ER. These data also render it highly likely that Syntaxin-4 associates, at least transiently, with the minor fraction of surface-expressed K5. This K5 fraction is not visible by immune fluorescence analysis, but can be detected by cell surface biotinylation [19]. Our data further suggest that K5 and MARCH proteins not only interact with, but also degrade syntaxins. Since syntaxins play a major role in vesicular traffic, the reduction of syntaxin levels could mislocalize transmembrane proteins, which might facilitate their targeting for ubiquitination. Further work will be required to address the role of syntaxins in the function of this protein family.

ALCAM (CD166)
ALCAM (CD166, Hs.591293) is a type I transmembrane glycoprotein of the Ig superfamily that is the ligand for the scavenger receptor CD6 on T cells [42]. The ALCAM-CD6 interaction is part of the immunological synapse and is considered a co-stimulatory signal involved in lymphocyte activation and thymocyte development [43][44][45]. Using transfection of a hemagglutinin epitope (HA)-tagged construct, we confirmed that K5 very efficiently eliminated ALCAM-HA upon co-transfection ( Figure 5A). Additionally, downregulation of ALCAM from the cell surface was confirmed by flow cytometry of HeLa cells transduced with Ad-K5. In contrast to ALCAM, K5 did not alter surface expression of CD9 or CD29 ( Figure 5B). The co-stimulatory function of ALCAM renders this molecule an attractive target for viral immune modulators. To determine whether other viral K3 family proteins target ALCAM, we transfected KSHV-K3 and the myxomavirus homolog M153R and monitored ALCAM expression by flow cytometry. Interestingly, M153R clearly reduced ALCAM surface levels, whereas the K3 protein, the most closely related K5 homolog, did not affect ALCAM ( Figure 5C). This result is consistent with previous observations that K3 targets are limited to classical and nonclassical MHC I-like molecules [9,10,20], whereas both K5 and M153 display broader substrate specificity [12,21,22]. Within the MARCH family, MARCH-IV and MARCH-IX downregulated ALCAM to a significant degree, while a modest effect was observed with MARCH-VIII ( Figure 5C). The RING and transmembrane domains of MARCH-IV and MARCH-IX are highly homologous to each other, and this result is consistent with previous observations showing that the range of substrates for these two MARCH proteins overlaps with that of K5 [14].
The catalytically inactive K5-RING construct [38], as well as a K5 construct lacking acidic clusters in its carboxy-terminal domain (K5DE12) [19,46], did not affect surface expression of ALCAM ( Figure 5C). These data suggest that downregulation of ALCAM requires the ubiquitin ligase activity and proper subcellular targeting of K5.
To investigate whether ALCAM downregulation occurs in KSHV-infected endothelial cells, we took advantage of the fact that K5 is not expressed in latently infected dermal microvascular endothelial cells (DMVECs), but can be induced upon addition of phorbol 12-myristate 13-acetate (PMA), which activates the lytic cycle [19]. As shown previously, PMA treatment reduced expression of MHC I on the cell surface of KSHV-infected DMVECs, but not uninfected DMVECs ( Figure 5D). The mean florescence of ALCAM was unchanged in latently infected DMVECs, but was reduced approximately 2-fold upon PMA treatment. In contrast, levels of CD81, a multiple transmembrane-spanning plasma membrane protein that is elevated in latently infected cells, did not change upon PMA treatment. Since PMA treatment does not induce lytic gene expression, and hence K5 expression, in all of the KSHV-infected cells [47], the actual levels of downregulation of both MHC I and ALCAM might be considerably higher in individual cells. The correlation of ALCAM downregulation and K5 expression, however, renders it highly likely that virally expressed K5 is causing this effect.
Taken together, these data suggested that ALCAM is a bona fide substrate for the viral ubiquitin ligase K5. To confirm that ALCAM downregulation involved K5-dependent conjugation of ubiquitin, we transfected HeLa cells with ALCAM-HA either in the presence or absence of Ad-K5. Ad-vpu, Ad-K5DE12, and Ad-MARCH-VIII were used as controls. Then, 24 h post-transfection, cells were lysed and ALCAM precipitated using an antibody against the HA epitope. Samples were resolved on a SDS-PAGE gel, and the presence of ubiquitinated ALCAM-HA was analyzed by immunoblotting with the anti-ubiquitin antibody P4D1. In cells transfected with ALCAM-HA alone, ubiquitinated ALCAM-HA was undetectable ( Figure 6A). In the presence of K5, however, substantially increased levels of ubiquitinated ALCAM-HA were observed, consistent with K5-mediated ubiquitination of ALCAM. Acidic motif-deleted K5 and HIV-vpu had no effect, whereas MARCH-VIII slightly increased ALCAM-HA ubiquitination, consistent with the slight downregulation of endogenous ALCAM observed by flow cytometry.
Depending on the subcellular localization of the ubiquitination reaction, K5 targets are degraded in lysosomes or by the proteasome [19]. To determine which degradation system was responsible for ALCAM degradation, we followed the maturation and degradation of ALCAM-HA using pulsechase analysis in the presence of proteasomal inhibitors or inhibitors of endosomal acidification. In the absence of K5, ALCAM-HA acquired complete resistance to endoglycosidase H within 2 h post-labeling, indicating its intracellular transport through the Golgi complex ( Figure 6B). After chasing for 8 h, ALCAM-HA levels remained high in control cells, showing only a minor reduction due to endogenous turnover. Expression of K5 did not alter ALCAM biosynthesis or trafficking through the Golgi; however, amounts of endoglycosidase H-resistant ALCAM were significantly reduced, consistent with a post-Golgi degradation mechanism. K5-dependent degradation of ALCAM was inhibited by inhibiting lysosomal acidification with concanamycin A but not by the proteasomal inhibitor MG132. A similar effect was seen with fluorescent microscopy analysis (unpublished data). To determine whether ALCAM is targeted to lysosomes via the MVB pathway, we co-transfected K5 with a dominant negative version of the AAA-ATPase Vps4 (Vps4mut), which prevents MVB formation [48]. Expression of dominant negative Vps4 partially restored surface expression of both MHC I and ALCAM measured using flow cytometry, whereas wild-type Vps4 had only a minor effect ( Figure 6C). Taken together, these data indicate that K5 ubiquitinates ALCAM in a post-ER compartment, thus triggering ubiquitin-mediated endocytosis and targeting of ALCAM for lysosomal destruction via the MVB pathway. Since this mechanism is similar to that of other K5 targets, we conclude that ALCAM is a novel substrate for K5 as well as a target for M153R. Thus, we present a novel approach for the systematic identification of substrates for transmembrane ubiquitin ligases as well as viral immune modulators targeting membrane proteins.

Discussion
The goal of this work was to systematically search for novel targets of viral immune modulators. To this end, we applied SILAC, a quantitative proteomics method based on metabolic isotope labeling. Until now, the SILAC method was mainly used for global comparative proteomics experiments expected to yield a relatively large number of changes, such as the comparison of cell surface proteins from T helper 1 versus T helper 2 cells [49], differential protein secretion from cancer versus normal cells [50], proteins associated with stimulation of a given receptor [51], or changes in substrates trapped by a mutant protease complex upon stress activation [52]. To our knowledge, however, this is the first time SILAC has been used in an experiment where only a few protein changes were expected. A major challenge for this approach was to identify true changes among the high rate of false positives that are caused by false peptide identification, inaccurate measurements of peptide intensity, or biological variation arising during the labeling procedure. For these reasons, protein quantitation in SILAC experiments is usually limited to proteins that are identified by two or more peptides (e.g., [52]). This allows confirmation of each peptide by a second peptide from the same protein, thus raising confidence that calculated protein ratios are accurate. This approach, however, severely limits proteome coverage, particularly for low abundance proteins that are often identified by only one peptide. For this reason we sought to include all identified proteins in our analysis regardless of how many peptides were used in their identification. Inclusion of single peptide identifications raised the background for individual experiments to the point that it was impossible to discriminate between true positives and false positives purely by a fold-change cutoff (unpublished data). By analyzing three biological replicate experiments, we were able to identify true positives. The use of three replicates sets a high threshold and might exclude potential substrates. However, upon lowering the stringency (either two out of three experiments or several peptides in one experiment), we were unable to confirm additional targets from a few selected candidates, suggesting a dramatic increase in the false positive rate (unpublished data). In contrast, using three replicates resulted in 100% specificity.
The rate of false negatives-the proteins that were affected by K5 but were eliminated by these selection criteria-is difficult to gauge since only a few of the known K5 targets (MHC I and ICAM-1) are present in HeLa cells. No peptides derived from ICAM-1 were ever identified, probably since ICAM-1 is expressed at low levels in this cell type unless induced by pro-inflammatory stimuli. A better estimate of the false negative rate could potentially be obtained by analyzing other cell types, such as B cells and endothelial cells that express high levels of multiple K5 targets. The sensitivity of the method is therefore presently unknown. However, since only a fraction of the membrane proteome was identified here, it is likely that additional targets for K5 will be revealed by improved protein identification, improved analysis methods, or improved enrichment of the target proteome.
Using specific antibodies as well as epitope tagging, we confirmed the K5-dependent reduction in protein abundance for BST-2, Syntaxin-4, and ALCAM in independent experiments. Of these proteins, only BST-2 was unrelated to protein families previously shown to interact with or be targeted by the MARCH family or K3 family proteins. Downregulation of BST-2 required the ubiquitin ligase function of K5, suggesting that the reduced expression of BST-2 is linked to the enzymatic function of K5. However, we have not ruled out the possibility that BST-2 levels are indirectly affected by another protein that is targeted by K5. Thus, one of the limitations of the SILAC approach is that it does not discriminate between proteins that are targeted directly and those that are targeted indirectly by the transmembrane ubiquitin ligases. This limitation could possibly be overcome by combining the SILAC analysis with proteomic analysis of ubiquitinated proteins [53,54]. Further experiments will be needed to determine whether BST-2 is a bona fide target of K5 or related molecules. If so, it could be that BST-2 plays an as yet unknown role in immune defense.
The identification of Syntaxin-4 by the SILAC method was surprising for several reasons. First, the K5-induced changes in Syntaxin-4 levels were relatively modest compared to those for MHC I, ALCAM, or BST-2. This demonstrates that even moderately changing proteins can be identified by SILAC, provided these changes can be measured reproducibly. Second, previous observations with MARCH proteins indicated that syntaxins interact with this protein family, but evidence for degradation was not reported. Third, Syntaxin-4 is quite different from the typical K5 substrates, which mostly represent type I glycoproteins of the Ig superfamily. In contrast, Syntaxin-4 is a type II transmembrane protein with a carboxy-terminal tail-anchor and multiple coiled-coil domains. As discussed above for BST-2, it might therefore be that K5 mediates degradation of Syntaxin-4 through an indirect mechanism. However, the co-localization of Syntaxin-4 with all MARCH proteins that locate to post-ER compartments renders it likely that Syntaxin-4 interaction is a general feature of the entire protein family, including K5. Syntaxin-4 thus joins a as well as the overexpressed E3 (using a FITC-conjugated anti-FLAG antibody, shown as green). Co-localization of Syntaxin-4 and the ubiquitin ligases is revealed as yellow co-staining. While K5 did not co-localize with Syntaxin-4, several MARCH family proteins relocalized Syntaxin-4 to a MARCHcontaining compartment. A naturally occurring splice variant of MARCH-IX that lacks a complete RING domain and fails to exit the ER was unable to relocalize Syntaxin-4.
(C) C-terminal truncation mutants of MARCH-VIII that failed to exit the ER do not co-localize with Syntaxin-4. DOI: 10.1371/journal.ppat.0020107.g004 PLoS Pathogens | www.plospathogens.org October 2006 | Volume 2 | Issue 10 | e107 0983 growing list of vesicular trafficking regulators interacting with MARCH family or K3 family proteins. In addition to the above-mentioned Golgi t-SNARE Syntaxin-6 that interacts with MARCH-II and MARCH-III, this list includes the COP-I adaptor protein PACS-2 [19] and the ER/cis-Golgilocated adaptor protein Vap-A (R. Means and S. Lang, personal communication). Either such interactions could locate the RING-CH proteins to specific vesicular compartments, or the RING-CH protein could interfere with the function of the vesicular regulators. In the case of Syntaxin-4, it is conceivable that downregulation of Syntaxin-4 reduces the exocytosis of certain proteins. For instance, Syntaxin-4 is required for the secretion of insulin in pancreatic beta cells [55]. One could therefore speculate that, by degrading Syntaxin-4, KSHV prevents the secretion of immune stimulatory chemokines or cytokines. Interestingly, another membrane-associated RING ubiquitin ligase, termed Staring, was previously shown to interact with, ubiquitinate, and degrade Syntaxin-1 [56]. It is therefore conceivable that the degradation of Syntaxin-4 by endogenous MARCH proteins plays a role in the turnover of Syntaxin-4, thus regulating exocytosis.
The final novel protein identified by our SILAC screen was ALCAM (CD166). Similar to other proteins downregulated by K5, ALCAM is a type I glycoprotein of the Ig superfamily. Several lines of evidence suggest that ALCAM is a bona fide target of K5. ALCAM downregulation required a functional RING-CH domain and acidic motif in K5. Similar to other K5 target molecules, ALCAM was ubiquitinated and targeted to lysosomes via the MVB pathway. Most likely, K5 ubiquitinates ALCAM at the plasma membrane, resulting in ubiquitinmediated endocytosis and lysosomal destruction.
In addition, we observed that the myxomavirus protein M153R downregulated ALCAM, likely by a similar mechanism, although this has not been experimentally verified. Previously, it was shown that M153R prevents T cell recognition of myxomavirus-infected cells [13]. In addition, Coscoy and Ganem demonstrated that K5-transfected B cells do not stimulate T cells [21]. While the predominant reason for this lack of T cell recognition is likely the downregulation of MHC I from the cell surface, Coscoy and Ganem showed that the elimination of B7.2 and ICAM-1 further reduces T cell receptor (TCR) signaling [21]. It was thus concluded that the combined effect of preventing both the primary signal via the TCR and the co-stimulatory signals via LFA-1 and CD28, the respective ligands of ICAM-1 and B7.2, minimizes T cell activation by KSHV-infected B cells. Similar to LFA-1 and CD28, ALCAM's ligand, CD6, becomes an integral part of the immunological synapse. Specifically, LFA-1 forms a ring around the synapse, whereas CD28 and CD6 co-localize with the TCR in the center of the synapse [43]. It is generally believed that this supramolecular structure is required for the sustained TCR engagement needed to generate the full repertoire of T cell responses, including signal transduction, cytokine generation, and cell proliferation [57]. Although the function of CD6 has not been studied as extensively as that of LFA-1 and CD28, there is accumulating evidence that the CD6-ALCAM interaction contributes to formation and function of the immunological synapse. It has been shown that antibodies to CD6 activate T cells [58] and that CD6blocking antibodies or recombinant ALCAM-Fc proteins inhibit T cell proliferation induced by dendritic cells [45]. Taken together, these observations strongly support the notion that, by downregulating ALCAM, K5 and M153R further optimize the ability of their respective viruses to prevent T cell activation. Moreover, the finding that ALCAM is eliminated by immune modulators from both herpesviruses and poxviruses suggests that the ALCAM-CD6 ligand receptor pair plays a previously unappreciated role in antiviral immunity.
In summary, our results validate the use of quantitative proteomics to identify new substrates of viral and cellular immune modulators. In addition to identifying this specific class of transmembrane ubiquitin ligases, it is conceivable that this approach can be used to identify additional targets for any viral proteins that are known to reduce the abundance of their host cell target molecules. Since many viral proteins are thought to be multi-functional, such a systematic, unbiased search for novel host cell targets could help to reveal new functions of viral proteins. This proteomics approach can also be used to find novel targets for cellular proteins that degrade a specific set of targets. In particular, novel substrates for cellular ubiquitin ligases can potentially be found this way. Since ubiquitin ligases are one of the most abundant gene families in the human genome, it is likely that almost every protein in the human genome is regulated by one or more ubiquitin ligases. Identifying the ligase-substrate relationship will be a daunting but important task in unraveling the regulatory networks that govern cellular functions. Novel approaches, as the one described here, are likely to aid this process.
A plasmid containing the ALCAM cDNA was a gift from G. Swart (Radboud University, Netherlands). ALCAM-HA was generated from this plasmid by PCR amplification of the ALCAM open reading frame. A C-terminal HA tag was added during the amplification. The PCR product was then digested using suitable enzymes and ligated into the pUHD10-1 vector. MARCH-VIII truncation mutants (D46, D62, and D74) were generated by PCR amplification of the given region. An N-terminal FLAG tag was added to each construct during amplification. Each PCR was then digested using suitable enzymes and ligated into the mammalian-expressing vector pUHD10-1. An adenoviral vector expressing MARCH-VIII was generated by cloning the full-length open reading frame from MARCH-VIII into the pShuttle vector (Stratagene, http://www.stratagene.com). Ad-MARCH-VIII virus was generated by linearization, transfection, and amplification according to the manufacturer's directions.
Preparation of samples for MS/MS analysis. Cells grown in labeling medium were infected with either Ad-K5 (heavy-labeled cells) or Ad-WT (light-labeled cells) at an MOI of 25. Then 24 h postinfection, cells were harvested by scraping, washed twice in PBS, resuspended in PBS containing 5 mM EDTA, and lysed by douncing. Unlysed cells and debris were cleared from the lysate by centrifugation for 5 min at 3,000 3 g. The cleared lysates were separated into membrane and soluble fractions by centrifugation for 30 min at 45,000 3 g. The membrane fraction was resuspended in PBS by sonication and separated over a discontinous sucrose gradient (2 M, 1.6 M, 1.25 M, 1.2 M, and 0.8 M) by centrifugation for 2.5 h at 25,000 rpm (Sorvall SW-28 rotor, http://www.sorvall.com). The bands corresponding to the plasma membrane (0.8-1.2 M interphase), Golgi (1.25-1.2 M interphase), and ER (1.6-1.25 M interphase) fractions were removed, diluted 53 in Tris-EDTA (pH 8.0), and centrifuged for 30 min at 45,000 3 g to pellet the proteins contained in each fraction. Pellets were washed for 30 min in 50 mM sodium bicarbonate (pH 11.5) and centrifuged for 30 min at 45,000 3 g, followed by a second wash in 50 mM ammonium bicarbonate (pH 8.5) and further centrifugation for 30 min at 45,000 3 g. Final pellets were resuspended in 8.0 M deionized urea and 50 mM ammonium bicarbonate (pH 8.5) and protein levels quantitated using the Bio-Rad Protein Assay (Bio-Rad, http://www. biorad.com). Samples were reduced with DTT (Sigma) and alkylated with iodoacetamide (Sigma) prior to overnight digestion with trypsin (Promega, http://www.promega.com).
Proteins were identified by analyzing tandem mass spectra with the Sequest algorithm (Thermo Electron) as described by Yates et al. [61] using the human subset of the UniProt/Swiss-Prot protein database (UniProt release 5.1, http://www.expasy.org/sprot). The search results were further analyzed using PeptideProphet [32]. SILAC ratios were determined using the ASAPRatio algorithm [34]. Multiple peptides derived from a single protein were included if PeptideProphet probability was greater than or equal to 0.85. All positive results were manually verified. The putative subcellular localization of proteins was determined using the Swiss-Prot database.
Immune fluorescence and flow cytometry. Cells (1.5 3 10 4 ) were plated on 15-mm coverslips (Fisher Scientific) and allowed to adhere overnight prior to transfection. Following transfection, cells were washed with PBS, fixed with 2% paraformaldehyde for 20 min at room temperature, and permeabilized with 0.2% Triton X-100 for 3 min at room temperature. Nonspecific binding sites were blocked with 3% BSA and 0.5% fish gelatin in PBS for 30 min at 37 8C. The fixed cells were incubated overnight at 37 8C with primary antibody diluted in blocking solution. Secondary and conjugated antibodies were diluted in blocking solution and incubated with the cells for at least 30 min at 37 8C. Cells were washed six times with PBS between all antibody treatments. Slides were fixed a second time in 2% paraformaldehyde after the final antibody treatment and washed twice with PBS. Coverslips were then mounted on slides and covered with Vectashield H-1200 þ DAPI (Vector Laboratories, http://www. vectorlabs.com).
For flow cytometry, cells were removed from tissue culture dishes with 0.05% trypsin-EDTA (Invitrogen), washed with ice-cold PBS, and incubated with appropriate antibody for 30 min at 4 8C. The cells were washed with ice-cold PBS and either resuspended in ice-cold PBS or incubated with PE-conjugated anti-mouse secondary antibody (Dako, http://www.dako.com) and washed again before analysis with a BD Biosciences FACSCalibur flow cytometer.
Metabolic labeling, immunoprecipitation, and Western blotting. HeLa cells were grown to 80% confluency in 100-mm tissue culture dishes and were transfected as above. At 24 h post-infection, cells were incubated in serum-free and methionine-free medium for 30 min, and metabolically labeled with 35 S-cysteine/ 35 S-methionine (300 lCi/plate; Amersham, http://www.amersham.com) for 20 min. After labeling, cells were washed twice with PBS and the label was chased for the indicated time in DMEM containing excess cold methionine/ cysteine. Following chase, cells were lysed in PBS containing 1% NP-40 and protease inhibitors (Roche, http://www.roche.com). The cell lysate was pre-cleared with protein A/G agarose beads overnight and incubated with 3 lg of antibody for 1 h, followed by 1 h with protein A/G beads. Immunoprecipated proteins were washed five times with 1% NP-40 in PBS. All samples were boiled in Laemmli buffer and analyzed by SDS-PAGE gel electrophoresis. Gels were fixed, dried, and exposed to Kodak (http://www.kodak.com) BioMax MR film. Western blotting was accomplished using the Western-Breeze Chemiluminescent Detection System (Invitrogen) following semi-dry transfer to PVDF membranes (Millipore, http:// www.millipore.com).