Determinants of the Higher Order Association of the Restriction Factor TRIM5α and Other Tripartite Motif (TRIM) Proteins*

Many tripartite motif (TRIM) proteins self-associate, forming dimers and higher order complexes. For example, dimers of TRIM5α, a host factor that restricts retrovirus infection, assemble into higher order arrays on the surface of the viral capsid, resulting in an increase in avidity. Here we show that the higher order association of different TRIM proteins exhibits a wide range of efficiencies. Homologous association (self-association) was more efficient than the heterologous association of different TRIM proteins, indicating that specificity determinants of higher order self-association exist. To investigate the structural determinants of higher order self-association, we studied TRIM mutants and chimeras. These studies revealed the following: 1) the RING domain contributes to the efficiency of higher order self-association, which enhances the binding of TRIM5α to the human immunodeficiency virus (HIV-1) capsid; 2) the RING and B-box 2 domains work together as a homologous unit to promote higher order association of dimers; 3) dimerization is probably required for efficient higher order self-association; 4) the Linker 2 region contributes to higher order self-association, independently of effects of Linker 2 changes on TRIM dimerization; and 5) for efficiently self-associating TRIM proteins, the B30.2(SPRY) domain is not required for higher order self-association. These results support a model in which both ends of the core TRIM dimer (RING-B-box 2 at one end and Linker 2 at the other) contribute to the formation of higher order arrays.

Retroviruses require host factors to negotiate successfully the different stages of their replication cycle. Host species that encounter retroviruses have developed intracellular factors that can counteract the establishment of a permanent viral infection (1)(2)(3)(4)(5). One such restriction factor, TRIM5␣, mediates a potent block to the infection of numerous retroviruses soon after virus entry into the host cell (4). TRIM5␣ proteins exhibit species-specific differences in the range of retroviruses that are susceptible to restriction (6). For example, TRIM5␣ pro-teins from rhesus macaques and most other Old World monkeys potently block infection of human immunodeficiency virus (HIV-1), whereas human TRIM5␣ only modestly inhibits HIV-1 infection; however, human TRIM5␣ potently inhibits N-tropic murine leukemia virus (N-MLV) 2 infection (4,(7)(8)(9)(10). TRIM5 proteins are thought to recognize incoming viral capsids through direct binding. Using either crude cell lysates or purified proteins, TRIM5␣ has been shown to bind specifically to in vitro assembled HIV-1 capsid-nucleocapsid (CA-NC) complexes that resemble authentic viral cores (11)(12)(13). TRIM5␣ binding to the viral core leads to the restriction of retroviral infection by an incompletely understood process. The amount of particulate capsids is consistently decreased in the cytosol of TRIM5␣-expressing cells during infection with sensitive viruses, suggesting that TRIM5␣ blocks infection by promoting premature disassembly of viral capsids (13). Proteasome inhibitors have been reported to rescue HIV-1 reverse transcription, but not HIV-1 infection, in the presence of a restricting TRIM5␣ protein (14,15). Moreover, degradation of TRIM5␣ in cells exposed to saturating levels of susceptible viruses has been reported (16). Thus, the proteasome may be involved in TRIM5␣-mediated retroviral restriction in as yet unclear ways.
TRIM5␣ is a member of the tripartite motif (TRIM) protein family (17). TRIM5␣ contains RING, B-box 2, and coiled-coil domains common to all TRIM proteins and an additional C-terminal B30.2(SPRY) domain (17,18). The interaction between TRIM5␣ and retroviral capsids requires an intact B30.2(SPRY) domain. Sequence variation in the B30.2(SPRY) domain has been shown to determine the viral specificity of TRIM5␣ restriction (19 -24). The observed patterns of capsid and B30.2(SPRY) sequence determinants of restriction support a model in which the TRIM5␣ B30.2(SPRY) domain directly contacts the targeted capsid. Two other TRIM5 structural domains, the coiled-coil and B-box 2 domains, enhance TRIM5-capsid interactions by promoting cooperative binding. The coiled-coil domain is essential for TRIM5 dimerization (25), whereas the B-box 2 domain promotes higher order selfassociation between preformed TRIM5␣ dimers (26,27). These processes could increase the binding avidity of TRIM5␣ com-plexes for the retroviral capsid. Particularly when the interaction of the B30.2(SPRY) domain with the retroviral capsid is weak, higher order self-association of TRIM5␣ is essential for potent restriction (26,28). A strong correlation between inhibition of HIV-1 infection and higher order TRIM5␣ self-association has been reported for a panel of mutants in a B-box 2 domain surface patch (27); in addition, the observed correlation between higher order self-association and binding to HIV-1 CA-NC complexes supports the notion that TRIM5␣ self-association promotes retroviral restriction by contributing to the avidity of TRIM5␣ complexes for the capsid (27).
The propensity to form oligomers is a property common to most TRIM proteins and therefore is likely to be important for their respective biological functions (17). Many TRIM proteins form discreet nuclear or cytoplasmic aggregates ("bodies") when overexpressed (17). Coiled-coil domain-mediated homodimerization probably contributes to the formation of these bodies; in addition, changes in the RING and B-box domains have been shown to alter the subcellular localization or compartmentalization of some TRIM bodies (17). Preformed TRIM5␣ cytoplasmic bodies are not required for retroviral restriction (29,30). TRIM5␣ has been observed to rapidly assemble around incoming HIV-1 viral complexes (31), and some TRIM5␣ mutants that do not form cytoplasmic bodies fail to block HIV-1 infection (32). Although the ability of TRIM5␣ to multimerize and form higher order structures clearly contributes to capsid binding and restriction, the relationship of these processes to cytoplasmic body formation is not presently understood.
In this study, we aim to understand further the mechanism by which the TRIM5␣ protein recognizes and disrupts the viral core structure by characterizing its higher order self-association. Our results show that, in addition to the B-box 2 domain, the RING domain and the Linker 2 region are also required for TRIM5␣ higher order self-association. The RING and B-box 2 domains appear to function as a unit in mediating the interactions required for higher order self-association. Several other TRIM dimers examined also self-associated into higher order forms, but the efficiency of this process varied for the different TRIM proteins. The TRIM proteins examined exhibited similar domain requirements for higher order self-association. The higher order self-association of TRIM5␣ dimers could assist the formation of an ordered lattice structure that binds the hexagonally arrayed viral capsid, eventually leading to capsid disruption.

EXPERIMENTAL PROCEDURES
Plasmid Constructs-Expressor plasmids for rhesus monkey TRIM5␣ (TRIM5␣ rh ) and human TRIM4, TRIM6, TRIM21, TRIM22, and TRIM34 were constructed in the pLPCX vector (Clontech) and have been described previously (33). All FLAGtagged proteins have the epitope tag at the N terminus. All HA-tagged proteins have the epitope tag at the C terminus, except for the TRIM6 Linker 2 truncation mutants, in which the HA tag was placed at the N terminus of the proteins to preserve the C-terminal structure. The TRIM5␣ rh ⌬RING mutant contains residues 60 -497 of TRIM5␣ rh , whereas the ⌬RING-L1 mutant contains residues 97-497. The RBCC and RBCC-L2 mutants of TRIM5␣ rh , TRIM6, and TRIM34 have been described previously (25,33). The termini of these deletion mutants were determined based on sequence alignment and the proposed domain boundaries for TRIM proteins (34). All deletion mutants were made by PCR amplification. The R121E change was introduced by QuikChange mutagenesis (Stratagene). TRIM4-TRIM5 chimeras were constructed either by overlapping PCR extension or QuikChange mutagenesis. For domain replacement, the TRIM5␣ rh RING sequence 1-59 was exchanged with the TRIM4 RING sequence 1-53; the TRIM5␣ rh B-box 2 sequence 97-129 was exchanged with the TRIM4 B-box 2 sequence 87-119. In the RING-B-box 2 chimeras, TRIM5␣ rh residues 1-129 were exchanged with TRIM4 residues 1-119. The TRIM4 -5SPRY chimera includes residues 1-280 of TRIM4 and 297-497 of TRIM5␣ rh .
Cross-linking of TRIM Proteins-Cell lysates prepared in 1% Nonidet P-40, PBS, protease inhibitor mixture were incubated with varying concentrations of glutaraldehyde (Sigma) at room temperature for 5 min, followed by the addition of excess glycine to quench the reaction (35). The cross-linked lysates were then subjected to SDS-PAGE and Western blotting with horseradish peroxidase (HRP)-conjugated anti-HA (1:1000; Roche Applied Science) or anti-FLAG antibody (1:1000; Sigma).
Assay for Higher Order Self-association-The ability of TRIM dimers to form higher order complexes was assessed by a coprecipitation assay. 293T cells were transiently transfected with the empty pLPCX vector or pLPCX vectors encoding HAor FLAG-tagged TRIM variants in 6-well plates. Forty-eight hours after transfection, cells were lysed in 1 ml of lysis buffer (1% Nonidet P-40, PBS, protease inhibitor mixture). The lysates were cleared of insoluble materials and aggregates by centrifugation at 13,200 rpm for 1 h at 4°C. Samples of the cleared lysates were taken for analysis of the level of TRIM proteins. When needed to achieve comparable levels of input TRIM proteins in the coprecipitation assay, selected lysates containing the TRIM variants were diluted with lysates of 293T cells transiently transfected with the empty pLPCX vector. The lysates containing different TRIM variants were then mixed and incubated with 20 l (packed volume) of Protein A-Sepharose beads (Amersham Biosciences) for 3 h at 4°C to remove proteins that bind nonspecifically to the beads. Samples of the precleared lysate mixture were taken at this point for analysis of the level of input TRIM proteins. The remaining lysates were incubated with 20 l of fresh beads and 1 l (ϳ5-6 g) of anti-FLAG antibody (Sigma) overnight at 4°C on a rocker. The immunoprecipitates were washed three times with buffer I (300 mM NaCl, 50 mM Tris-HCl, 1% Nonidet P-40) at 4°C for 10 min each on a rocker and once with buffer II (150 mM NaCl, 10 mM Tris-HCl) for 10 min at 4°C. The beads were then treated with 2ϫ SDS sample buffer (125 mM Tris-HCl, 2% SDS, 16% glycerol, 3% ␤-mercaptoethanol, 0.01% bromphenol blue) and boiled for 5 min to release the precipitated proteins.
Western Blotting and Quantitation of Co-immunoprecipitation Efficiency-Input and precipitated proteins from the coimmunoprecipitation (co-IP) experiments were analyzed by SDS-PAGE and Western blotting with HRP-conjugated anti-HA antibody (1:1000; Roche Applied Science) or anti-FLAG antibody (1:1000; Sigma). Immunoreactive proteins were visualized by ECL Plus detection reagents (GE Healthcare) on Eastman Kodak Co. autoradiography film. For densitometric analysis, films at low exposure levels were scanned digitally, and the relative intensities of the protein bands were quantitated by ImageJ 1.42q software (National Institutes of Health). The co-IP index is defined as the intensity of HA-tagged protein in the pellet divided by the sum of the intensities of the HAtagged protein in the input and the FLAG-tagged protein in the pellet; the co-IP indices of TRIM protein variants are reported relative to the values for the respective wild-type TRIM protein.
HIV-1 Capsid-binding Assay-Purification of recombinant HIV-1 CA-NC protein from Escherichia coli was carried out as described previously (36). For a source of TRIM5 proteins, 293T cells transiently transfected with plasmids expressing TRIM5 variants were lysed by freeze-thawing in hypotonic lysis buffer (10 mM Tris, pH 7.4, 1.5 mM MgCl 2 , 10 mM KCl, 0.5 mM DTT) containing protease inhibitor (Roche Applied Science). The lysates were then mixed with the in vitro assembled HIV-1 CA-NC complexes, and the binding assay was carried out as described previously (13,37).
Creation of Chicken DF1 Cell Lines Stably Expressing TRIM5␣ Proteins-Recombinant viruses were produced in 293T cells by co-transfecting the pLPCX plasmids expressing TRIM5 variants with the pVPack-GP and pVPack-VSV-G packaging plasmids (Stratagene). The pVPack-VSV-G plasmid encodes the vesicular stomatitis virus G envelope glycoprotein, which allows efficient entry into a wide range of vertebrate cells (38). The resulting virus particles were used to transduce 2 ϫ 10 5 DF1 chicken fibroblast cells (ATCC) in 6-well plates. The transduced DF1 cells were selected in 5 g/ml puromycin (Sigma). Expression of the TRIM5␣ proteins was confirmed by Western blotting with an anti-TRIM5␣ polyclonal antibody (1:2000; Imgenex), followed by an HRP-conjugated anti-rabbit secondary antibody (1:1000; Sigma).
Infection of Cells with Viruses Expressing Green Fluorescent Protein (GFP)-Recombinant HIV-1 and N-MLV viruses expressing GFP were made as described previously (4,9). For infection, 6 ϫ 10 4 DF1 cells or 2 ϫ 10 4 Cf2Th cells were seeded in 24-well plates and incubated with the viruses for 60 h. Cells were then washed with PBS, fixed with 3.7% formaldehyde, and subjected to fluorescence-activated cell sorting (FACS) analysis with a FACScan (BD Biosciences).

The RING Domain Is Required for TRIM5␣ Higher Order
Self-association-The domain arrangement of TRIM5␣ and the other TRIM proteins used in this study is shown in Fig. 1A. Deletion or disruption of the RING domain reduces but does not eliminate the antiviral activity of TRIM5␣ (4,39,40). Here we examined the contribution of the RING domain to rhesus monkey TRIM5␣ (TRIM5␣ rh ) higher order self-association.
Two TRIM5␣ rh RING domain deletion mutants, ⌬RING and ⌬RING-L1, were assayed for their ability to form higher order complexes with wild-type TRIM5␣ rh . The ⌬RING mutant contains residues 60 -497 of TRIM5␣ rh and has only the RING domain deleted. The ⌬RING-L1 mutant contains residues 97-497 of TRIM5␣ rh and lacks both the RING domain and the adjacent Linker 1 region. Both deletion mutants were able to form dimers when transiently expressed in 293T cells (Fig. 1B). The mutant dimers were then mixed in vitro with preformed wild-type TRIM5␣ rh dimers, and the mixture was subjected to co-immunoprecipitation to detect higher order association as described (26). This assay specifically detects dimer-dimer interactions because several TRIM5 mutants that efficiently coimmunoprecipitate when expressed in the same cells (and therefore are capable of forming heterodimers) were not able to co-immunoprecipitate in this assay (26); these observations suggest that TRIM5 proteins form tight dimers with a low dissociation constant. As shown in Fig. 1C, neither of the RING deletion mutants was efficiently coprecipitated with wild-type TRIM5␣ rh , although a weak interaction between the ⌬RING-L1 mutant and wild-type TRIM5␣ rh proteins was reproducibly detected. We conclude that the RING domain is required for efficient TRIM5␣ higher order self-association, in addition to the B-box 2 domain, whose requirement has been demonstrated previously (26). Of note, the ⌬RING-L1 mutant and wild-type TRIM5␣ rh can be efficiently coprecipitated when they are co-expressed in the same cell (39), indicating that the RING domain is specifically required for TRIM5␣ higher order self-association but not for dimerization.
Higher order self-association of TRIM5␣ has been shown to contribute to binding of the HIV-1 capsid (26). Consistent with a role of the RING domain in TRIM5␣ higher order self-association, both RING deletion mutants demonstrated significantly reduced affinities for in vitro assembled HIV-1 CA-NC complexes, which mimic authentic viral cores ( Fig. 1D) (36).
To assess the abilities of the two TRIM5␣ rh RING domain deletion mutants to restrict retrovirus infection, chicken DF1 cell lines stably expressing the wild-type and mutant TRIM5␣ rh proteins were established (Fig. 1E). Because no orthologs of TRIM5 have been identified in the chicken genome (41), endogenous TRIM5 or TRIM5-like proteins cannot influence the measurement of the restriction activity of the mutant proteins in these cells. Both mutants modestly restricted the infection of HIV-1 in chicken DF1 cells but lost the ability to inhibit N-MLV infection, which is consistent with the results obtained in other cell types (Fig. 1, F and G) (39,40). Defects in higher order self-association and the consequent decrease in capsid-binding ability may contribute to the observed decrease in retrovirus restriction activity of these mutants.
Higher Order Self-association of Different TRIM Proteins-Most TRIM family proteins exhibit a strong tendency to selfassociate into dimers, a process that is dependent on the coiledcoil domain (17). We examined whether TRIM proteins other than TRIM5␣ can also form higher order complexes beyond dimers. Three TRIM proteins, TRIM6, TRIM22, and TRIM34, that are the products of TRIM5 paralogs, and an unrelated TRIM protein, TRIM4, were chosen for this study. A previous cross-linking analysis demonstrated that all of these TRIM proteins efficiently dimerize (33). When preformed TRIM dimers were mixed and used in the coprecipitation assay, all of the TRIM proteins except TRIM22 exhibited some degree of higher order self-association (Fig. 2). TRIM4 and TRIM6 selfassociated at much higher efficiencies than TRIM5␣ and TRIM34 (Fig. 2). Of note, even after boiling in high concentrations of SDS and reducing agent, TRIM34 exhibits high molec-ular weight aggregates on a gel (Fig. 2). The formation of selfinteracting TRIM34 aggregates in cell lysates may compromise the ability of TRIM34 to associate efficiently with heterologous TRIM34 in the co-immunoprecipitation assay.
Heterologous Association of TRIM5␣ and Other TRIM Proteins-We examined the ability of TRIM5␣ rh to form heterologous higher order complexes with other TRIM proteins. Only the closely related TRIM proteins, TRIM6 and TRIM34, efficiently coprecipitated TRIM5␣ rh (Fig. 3A). Despite a high efficiency of self-association, TRIM4 minimally associated with TRIM5␣. Neither TRIM21 nor TRIM22 detectably associated with TRIM5␣ rh (Fig. 3A). The heterologous higher order associations between TRIM6/34 and TRIM5 were less efficient than TRIM5 self-association (Fig. 3B). These results indicate that, although some close TRIM relatives can form heterolo-FIGURE 1. The RING domain is required for TRIM5␣ higher order self-association. A, the arrangement of the domains of TRIM5␣ and the other TRIM proteins used in this study is shown. B, cell lysates from 293T cells transiently transfected with plasmids expressing HA-tagged wild-type TRIM5␣ rh or the ⌬RING and ⌬RING-L1 mutants were cross-linked with increasing amounts of glutaraldehyde (GA; final concentrations 0, 0.4, 1, 2, and 4 mM). The cross-linked products were resolved by SDS-PAGE and visualized by Western blotting with an anti-HA antibody. The positions of the molecular weight markers are indicated to the left of the blot. C, coprecipitation of HA-tagged TRIM5␣ rh variants with FLAG-tagged TRIM5␣ rh dimers. 293T cells were transiently transfected with the empty pLPCX vector or vectors expressing the indicated TRIM5␣ rh variants. Cytosolic extracts containing the wild-type and mutant TRIM5␣ rh dimers were prepared separately, mixed in a 1:1 ratio, and used for precipitation with an anti-FLAG antibody. The amount of HA-and FLAG-tagged proteins in the lysates (Input) and immunoprecipitates (Pellet) was analyzed by Western blotting with HRP-conjugated anti-HA and anti-FLAG antibodies. D, HIV-1 CA-NC binding assay. Cell lysates from 293T cells transiently expressing wild-type TRIM5␣ rh and TRIM5␣ rh ⌬RING and ⌬RING-L1 mutants were used in the HIV-1 CA-NC binding assay. The TRIM5␣ rh ⌬RING and ⌬RING-L1 mutants are designated ⌬R and ⌬R-L1, respectively. Two different input concentrations of the ⌬RING-L1 mutant were used in the assay. The top panel shows the input TRIM5␣ rh proteins; the middle panel shows TRIM5 proteins that co-sediment through the 70% sucrose cushion with the assembled HIV-1 CA-NC complexes. The HIV-1 CA-NC complexes in the pellets were detected by Western blotting with an antibody directed against the p24 CA protein (bottom panel). E, expression levels of the wild-type TRIM5␣ rh and ⌬RING and ⌬RING-L1 mutants in stable chicken DF1 cell lines. Western blotting was done with a polyclonal anti-TRIM5␣ antibody (Imgenex) and an anti-␤-actin antibody as control. F and G, DF1 cells stably expressing the empty vector pLPCX, wild-type TRIM5␣ rh , ⌬RING, and ⌬RING-L1 were incubated with various amounts of HIV-1-GFP (F) and N-MLV-GFP viruses (G). Infected GFPpositive cells were counted by FACS. 293T cells were transiently transfected with the empty pLPCX vector or vectors expressing the indicated HA-or FLAG-tagged TRIM proteins. Cytosolic extracts were prepared separately, mixed in a 1:1 ratio, and used for precipitation with an anti-FLAG antibody. TRIM34 typically forms gel-stable higher order complexes, even in the presence of SDS and ␤-mercaptoethanol.
gous higher order complexes with TRIM5, higher order association preferentially occurs among homologous TRIM5 dimers.
TRIM6 and TRIM34 Domain Requirements for Higher Order Self-association and Association with TRIM5-It has been shown previously that alteration of a surface-exposed residue, arginine 121 in the TRIM5␣ rh B-box 2 domain, to glutamic acid diminished higher order self-association and the ability of the protein to restrict retroviral infection (26). Sequence alignments indicate that this arginine is conserved in both human TRIM6 and TRIM34 proteins. To investigate the contribution of this B-box 2 domain residue to the higher order self-association of TRIM6 and TRIM34, an R121E change was introduced into both proteins, and the association with wild-type TRIM dimers was examined. Similar to the results obtained with TRIM5, the R121E change significantly reduced the higher order self-association of TRIM6 (Fig. 4A). The higher order self-association of TRIM34 detected by co-immunoprecipitation was slightly decreased by the R121E change, although the weakness of TRIM34 self-association makes it difficult to judge the magnitude of this effect (Fig. 4A).
The heterologous higher order associations between TRIM6/34 and TRIM5 were markedly attenuated by the R121E change (Fig. 4B). When this change was present on either of the interacting dimers or on both dimers, a drastic decrease in the association efficiency was observed (Fig. 4B). We conclude that the B-box 2 domain contributes to the higher order association of both TRIM6 and TRIM34.
The B30.2(SPRY) domain is not required for TRIM5 higher order self-association (26). We tested whether the B30.2(SPRY) domain is required for the self-association of TRIM6/34 or for their heterologous association with TRIM5. Deletion mutants that lack the B30.2(SPRY) domain (6RBCC-L2, 34RBCC-L2) or the B30.2(SPRY) and Linker 2 regions (6RBCC, 34RBCC) were used in the co-immunoprecipitation assay with wild-type TRIM5, TRIM6, or TRIM34. Deletion of the B30.2(SPRY) domain of TRIM6 had minimal effect on its self-association, whereas deletion of both the B30.2(SPRY) and Linker 2 regions significantly reduced the association efficiency (Fig. 4C). Unlike the case for TRIM5 and TRIM6, deletion of the B30.2(SPRY) domain of TRIM34 further reduced its already weak self-association. Interestingly, the smear of high molecular weight bands uniquely associated with the wild-type TRIM34 disappeared upon deletion of the B30.2(SPRY) domain (compare the pellet for FLAG-tagged 34, 34R-L2, and 34RBCC proteins in Fig. 4C). Apparently, the B30.2(SPRY) domain is important for the higher order self-association of TRIM34 but not TRIM6. For the heterologous higher order association with TRIM5, deletion of the B30.2(SPRY) domain of both TRIM6 and TRIM34 impaired the association efficiency (Fig. 4C).
Contribution of TRIM5␣ Dimerization to Higher Order Self-association-The coiled-coil domain and adjacent Linker 2 region are essential for TRIM5 dimerization (25). Study of a panel of TRIM5␣ rh mutants with changes in these regions indicated that the vast majority were capable of both dimerization and higher order self-association (Fig. 5A). Calculation of the co-immunoprecipitation indices indicated that most of the mutants exhibited higher order self-association at levels within 2-fold of that exhibited by wild-type TRIM5␣ rh . The only mutant (L194K) that failed to dimerize exhibited significantly reduced efficiency in higher order association with wild-type TRIM5␣ rh (Fig. 5A). One other mutant, D247K/L248D/L249D in the Linker 2 region, dimerized efficiently but reproducibly exhibited a 3-fold decrease in the efficiency of higher order association with wild-type TRIM5␣ rh (Fig. 5A). Based on these results, it appears that the TRIM5 coiled coil contributes to higher order self-association by promoting the initial dimerization of the protein. Linker 2 sequences may contribute specifically to the efficiency of higher order self-association.
To test further the hypothesis that TRIM5 dimerization is important for higher order self-association, we compared the efficiency of these processes for C-terminal deletion mutants lacking either the B30.2(SPRY) domain (RBCC-L2) or the Linker 2 region and B30.2(SPRY) domain (RBCC) (Fig. 5B). Previous studies have shown that the TRIM5␣ RBCC-L2 protein efficiently dimerizes, whereas the RBCC protein does not (25). Compared with the wild-type TRIM5␣ rh and RBCC-L2 proteins, the RBCC protein coprecipitated the wild-type TRIM5␣ dimers poorly (Fig. 5B). These results are consistent with dimerization playing an important role in TRIM5 higher order self-association. A, coprecipitation of HA-tagged TRIM5␣ rh with other FLAG-tagged TRIM proteins. The procedure for the coprecipitation is the same as that described in the legend to Fig. 2. B, quantitative comparison of TRIM5␣ rh higher order self-association and heterologous higher order association with TRIM6/34. Cell lysates made from 293T cells transiently transfected with plasmids encoding FLAG-tagged TRIM5␣ rh , TRIM6, or TRIM34 were serially diluted 1:2 or 1:4 with lysates from 293T cells transfected with the empty vector pLPCX. Undiluted (1) and diluted (1:2 or 1:4 ( 1 ⁄2 or 1 ⁄4, respectively)) lysates were then mixed with cell lysates of 293T cells transiently transfected with HA-tagged TRIM5␣ rh in a 1:1 ratio, and the mixture was precipitated with an anti-FLAG antibody. The co-IP index is shown for each of the co-immunoprecipitation reactions, except for the 1:4-diluted TRIM6 lysates, for which the input level was too low to yield a meaningful readout. AUGUST 12, 2011 • VOLUME 286 • NUMBER 32

JOURNAL OF BIOLOGICAL CHEMISTRY 27963
Contribution of the Linker 2 Region to Higher Order Self-association of TRIM Proteins-Based on the results obtained with the D247K/L248D/L249D mutant, we wished to examine the specific contribution of the Linker 2 region to higher order . Domain requirements for higher order self-association and heterologous association of TRIM6 and TRIM34. A, effect of the B-box 2 domain change, R121E, on higher order self-association. FLAG-tagged wildtype TRIM6 and TRIM34 dimers were used to coprecipitate HA-tagged wildtype or R121E (RE) TRIM6 and TRIM34 dimers. B, effect of the R121E change in the B-box 2 domain on higher order heterologous association. A minus sign, wt, and RE indicate lysates made from 293T cells transiently transfected with the empty pLPCX vector, a vector encoding FLAG-tagged wild-type TRIM5␣ rh , and a vector encoding FLAG-tagged TRIM5␣ rh with the R121E change, respectively. C, effects of C-terminal deletions on higher order selfassociation of TRIM6/34 and heterologous association with TRIM5␣ rh . The ability of the indicated FLAG-tagged TRIM6 and TRIM34 variants to coprecipitate the wild-type HA-tagged TRIM6, TRIM34, and TRIM5␣ rh proteins was tested. A schematic diagram of TRIM protein structural domains is shown. The 6RBCC-L2 and 34RBCC-L2 proteins (6R-L2 and 34R-L2, respectively) have a deletion of the B30.2(SPRY) domain only; the 6RBCC and 34RBCC proteins have a deletion of both the B30.2(SPRY) domain and the Linker 2 region. FIGURE 5. Contribution of TRIM5␣ dimerization to higher order self-association. A, coprecipitation of HA-tagged TRIM5␣ rh coiled-coil domain mutants with FLAG-tagged wild-type TRIM5␣ rh dimers. 293T cells were transiently transfected with plasmids encoding HA-tagged wild-type TRIM5␣ rh or TRIM5␣ rh with alterations in the coiled-coil domain. Cytosolic extracts from these cells were mixed in a 1:1 ratio with lysates from 293T cells transiently transfected with the empty vector pLPCX (Ϫ) or the plasmid expressing FLAG-tagged wild-type TRIM5␣ rh (ϩ). The mixture was precipitated with an anti-FLAG antibody. The abilities of these coiled-coil domain mutants to dimerize and their co-IP indices are indicated below the Western blots. B, coprecipitation of HA-tagged wild-type TRIM5␣ rh by FLAG-tagged C-terminal deletion mutants of TRIM5. RBCC-L2 is TRIM5␣ rh with a deletion of the B30.2(SPRY) domain; RBCC is TRIM5␣ rh with a deletion of both the B30.2(SPRY) domain and the Linker 2 region. self-association. However, the Linker 2 region contributes to TRIM5 dimerization (25). Therefore, we utilized the TRIM6 protein to study the specific effects of Linker 2 changes on higher order self-association because TRIM6 does not require a Linker 2 region for dimerization (25). The above results demonstrate that the wild-type and 6RBCC-L2 TRIM6 proteins (designated 6 and 6R-L2 in Fig. 4C), both of which contain the Linker 2 region, associated with wild-type TRIM6 to form higher order complexes. By contrast, the 6RBCC variant of TRIM6, which lacks the Linker 2 region, did not.
Upon cross-linking with glutaraldehyde, both the full-length TRIM6 and 6RBCC-L2 proteins formed dimers and higher order complexes, the latter appearing as an intense smear in the high molecular weight range (Fig. 6A). By contrast, 6RBCC formed only dimers, with no distinct higher order bands. These observations are consistent with the ability of 6RBCC to form dimers but not to assemble into higher order complexes. Together with the coprecipitation experiments, these crosslinking studies indicate that an intact Linker 2 region is required for TRIM6 higher order self-association but not dimerization.
To map the determinant in the TRIM6 Linker 2 region for higher order self-association, we created a series of deletion mutants and examined their higher order association with the wild-type TRIM6 protein. The design of these deletion mutants was guided by secondary structure predictions that suggest the presence of several tandem ␣ helices in this region; the deletions were planned to keep the individual predicted helices FIGURE 6. The Linker 2 region is required for higher order self-association. A, cross-linking of FLAG-tagged wild-type TRIM6 and its C-terminal deletion mutants with glutaraldehyde (GA; final concentrations 0, 1, 2, 4, and 8 mM). The 6RBCC-L2 mutant (here abbreviated 6R-L2) is TRIM6 with a deletion of the B30.2(SPRY) domain; 6RBCC is TRIM6 with a deletion of both the B30.2(SPRY) domain and the Linker 2 region. The cross-linked products were resolved by SDS-PAGE and visualized by Western blotting with an anti-FLAG antibody. B, the C termini of the TRIM6 truncation mutants studied are indicated on a detailed diagram of the TRIM6 Linker 2 region. The TRIM6 and TRIM5 Linker 2 regions are aligned, and identical and similar amino acid residues are shaded black and gray, respectively. The residue numbers for TRIM6 and TRIM5␣ rh are identical in this region and are noted above the alignment. C, dimerization of TRIM6 Linker 2 deletion mutants. Cell lysates of 293T cells transiently transfected with plasmids encoding N-terminally HA-tagged TRIM6 Linker 2 deletion mutants were cross-linked with increasing amounts of glutaraldehyde (final concentrations 0, 0.4, 1, and 2 mM). The cross-linked products were resolved by SDS-PAGE and visualized by Western blotting with an anti-HA antibody. The positions of the molecular weight markers are indicated to the left of the blots. D, coprecipitation of HA-tagged TRIM6 Linker 2 truncation mutants (6R-L2 1-296, 1-282, 1-265, 1-253 and 6RBCC 1-233) with FLAG-tagged wild-type TRIM6. The lysates of deletion mutants were undiluted (1) or diluted 1:2 or 1:4 with lysates of 293T cells transfected with the empty vector pLPCX to achieve equivalent levels of input proteins. The coprecipitation was carried out as described in the legend to Fig. 1. The co-IP indices are indicated below the Western blots. intact (Fig. 6B). All of the TRIM6 deletion mutants were able to dimerize (Fig. 6C). Compared with the 6RBCC-L2 protein (residues 1-296), all of the Linker 2 deletion mutants efficiently associated with the wild-type TRIM6 protein (Fig. 6D). Quantitative analysis indicated that progressive truncation of the Linker 2 region up to residue 253 resulted in a gradual, modest decrease in the efficiency of association with wild-type TRIM6 (Fig. 6D, left). However, the 6RBCC protein (residues 1-233) exhibited a significant decrease in higher order association with the wild-type TRIM6 protein, compared with the 1-253 mutant (Fig. 6D, right). Thus, sequences important for higher order TRIM6 association reside between residues 233 and 253, in the Linker 2 region immediately adjacent to the coiled coil.
Characterization of the Domain-Domain Interactions Involved in Higher Order Self-association-TRIM5␣ and TRIM4 dimers both self-associate to form higher order complexes, but the two proteins do not interact with each other (Fig.  3A) (33). The determinants of this specificity were investigated by using a series of TRIM4-TRIM5 chimeras.
Both the RING and the B-box 2 domains are essential for TRIM5 higher order self-association (see above  S1). The ability of these chimeric proteins to associate with each other in the co-immunoprecipitation assay was examined in all possible combinations, except when no interaction was expected due to a lack of any common domains (Fig. 7, B and C).
Neither mutant 5R nor 5B formed higher order complexes with wild-type TRIM5␣ rh (Fig. 7B). Similarly, mutants 4R and 4B did not interact with TRIM4 dimers (Fig. 7C). Thus, neither the RING nor the B-box 2 domain is sufficient for higher order self-association.
Replacing the RING or the B-box 2 domain of TRIM5 with the corresponding domains from TRIM4 (4R and 4B, respectively) significantly impaired the interactions between these chimeras and wild-type TRIM5␣ rh , indicating that each of these domains is required for TRIM5 higher order self-association (Fig. 7B). Mutant 4B exhibited a weak self-association, whereas mutant 4R was highly defective in associating with any of the proteins examined (Fig. 7, B and C). Replacing the RING domain of TRIM4 with the TRIM5 RING domain (5R) significantly reduced its higher order association with wild-type TRIM4 as well as its self-association (Fig. 7C). By contrast, mutant 5B, in which the B-box 2 domain of TRIM4 is replaced by the TRIM5 B-box 2 domain, retained a substantial level of higher order interaction with wild-type TRIM4; the self-association of mutant 5B was also highly efficient (Fig. 7C). Apparently, with respect to the higher order association of both TRIM5 and TRIM4, replacing the B-box 2 domain is relatively better tolerated than replacing the RING domain.
Two mutant pairs have the same combination of heterologous RING and B-box 2 domains; 5R and 4B both have a RING domain from TRIM5 and a B-box 2 domain from TRIM4, whereas 5B and 4R both have a TRIM4 RING domain and a TRIM5 B-box 2 domain. The C-terminal regions of these mutant pairs differ in origin. As can be seen in Fig. 7C, these paired mutants did not detectably associate with each other. Thus, matched RING and B-box 2 domains on the potential partner proteins are not sufficient for higher order association. On the other hand, two mutant pairs (mutants 4R and 4B and mutants 5R and 5B) have mismatched RING and B-box 2 domains on the potential partner proteins but share common C-terminal domains. In both cases, weak but reproducible higher order association was detected (Fig. 7, B and C). Apparently, the TRIM C-terminal domains make some contribution to higher order association. This conclusion is consistent with FIGURE 7. Higher order association between wild-type TRIM5␣ rh , TRIM4 hu , and chimeras with individual replacement of the RING or the B-box 2 domain. A, schematic diagram of wild-type TRIM5␣ rh , TRIM4 hu , and the chimeras. B, coprecipitation of wild-type TRIM5␣ rh , TRIM5␣ rh with the TRIM4 RING or B-box 2 domain (4R and 4B, respectively), and TRIM4 with the TRIM5␣ rh RING or B-box 2 domains (5R and 5B, respectively). C, coprecipitation of wild-type TRIM4 hu or TRIM4 hu with the TRIM5␣ rh RING or B-box 2 domain (5R and 5B, respectively) and TRIM5 with the TRIM4 RING or B-box 2 domain (4R and 4B, respectively). The co-IP indices are indicated below the Western blots. the evidence presented above for a specific contribution of the Linker 2 region to TRIM6 higher order self-association.
RING and B-box 2 Domains Function as One Motif in Mediating Higher Order Self-association-The above results indicate that when the TRIM5 and TRIM4 RING domains were individually replaced with a heterologous RING domain (4R and 5R, respectively), higher order self-association and the association with the parental proteins were completely lost or significantly reduced. Heterologous replacement of the B-box 2 domain (4B and 5B) was better tolerated but still resulted in a significant reduction in association efficiency. When the potentially interacting partners shared identical RING and B-box 2 domains from different origins, no higher order self-association was evident. We hypothesized that homologous RING and B-box 2 domains may need to function together as a single motif to promote efficient higher order self-association.
To test this hypothesis, the RING and B-box 2 domains of TRIM4 and TRIM5 were together exchanged between the two proteins, and higher order self-association and association with the wild-type parental proteins were examined. Chimera 4RB is TRIM5␣ rh with the TRIM4 RING and B-box 2 domains, whereas 5RB is TRIM4 with the TRIM5 RING and B-box 2 domains (Fig. 8A). Both chimeras self-associated with high efficiency (Fig. 8B), compared with the chimeras with individual RING or B-box 2 domain replacements (4RB versus 4R or 4B; 5RB versus 5R or 5B). The 4RB and 5RB chimeras also exhibited improved association with the wild-type parental proteins. For example, mutants 4R and 4B did not detectably interact with TRIM4, but mutant 4RB exhibited substantial ability to associate with TRIM4 (compare Figs. 7C and 8B). Similarly, neither mutant 5R nor 5B interacted with TRIM5␣ rh , but mutant 5RB exhibited some weak interaction with wild-type TRIM5 (compare Figs. 7B and 8B). The C-terminal half of the molecule may also contribute to this interaction because 5RB exhibited a modest level of interaction with TRIM4. These C-terminal sequences probably reside in the coiled-coil domain and Linker 2 region because the TRIM4 B30.2(SPRY) domain can be replaced by that of TRIM5␣ without diminishing either homologous self-association or heterologous higher order association with TRIM4 (see the 4 -5SPRY protein in Fig. 8, A and B). These results support a model in which homologous RING and B-box 2 domains function as a unit in higher order association of TRIM proteins. The results also support the existence of an independent contribution to higher order self-association by TRIM sequences in the coiled-coil and/or Linker 2 regions.

DISCUSSION
Although TRIM proteins are functionally diverse, they share a core domain structure (RING-Linker 1-B-box-coiled coil-Linker 2) and the tendency to self-associate into large assemblies when overexpressed (17). The coiled-coil domain is essential for the lower order oligomerization of TRIM proteins, many of which are dimers. We showed that preformed TRIM4, TRIM5, TRIM6, and TRIM34 dimers coprecipitated with the homologous dimers, suggesting that these TRIM proteins efficiently form higher order complexes. Because detection in the coprecipitation assay requires that complexed TRIM proteins remain associated during several wash steps, we cannot rule out that TRIM22 also forms higher order complexes that are not stable under the conditions of our assay.
TRIM5 formed higher order complexes with the closely related TRIM6 and TRIM34 proteins but not with more distantly related TRIM proteins. Because TRIM5 can restrict retroviral infection in cells that lack TRIM6 and TRIM34, the biological relevance of this association between paralogous TRIM proteins is unknown. Quantitative studies indicated that the heterologous associations between TRIM5 and its relatives, TRIM6 and TRIM34, were weaker than TRIM5 higher order self-association (Fig. 3B). TRIM6 also associated with itself significantly more efficiently than with TRIM5 (Fig. 4C). These observations suggest that the higher order self-association of TRIM proteins is probably determined by specific structural elements on each TRIM dimer.
Previous studies implicated a B-box 2 surface patch consisting of hydrophobic and basic amino acid residues in TRIM5 higher order self-association (27). We found that alteration of a conserved arginine residue in this B-box 2 patch decreased the ability of TRIM6 to self-associate and to associate with TRIM5. Although modification of the equivalent residue only modestly decreased the weak TRIM34 higher order self-association, heterologous association of TRIM34 and TRIM5 was significantly reduced by this change. These results suggest that a FIGURE 8. Higher order associations of the TRIM5␣ rh /TRIM4 hu chimeras with replacements of both the RING and the B-box 2 domains. A, schematic diagram of wild-type TRIM5␣ rh and TRIM4 hu and the chimeras. B, higher order association of the RING-B-box 2 (RB) chimeras, the B30.2(SPRY) domain chimera (4 -5SPRY), and the wild-type parental TRIM5 and TRIM4 proteins. The 4RB chimera is a TRIM5␣ rh protein in which both the RING and B-box 2 domains have been replaced by those of TRIM4. The 5RB chimera is a TRIM4 protein in which both the RING and B-box 2 domains have been replaced by those of TRIM5. The 4 -5SPRY chimera is a TRIM4 protein with the B30.2(SPRY) domain replaced by that of TRIM5. The co-IP indices of the chimeras were normalized to those associated with TRIM4 self-association and/or TRIM5 selfassociation and are indicated below the Western blots. related structural region on the TRIM5, TRIM6, and probably TRIM34 B-box 2 domain surfaces contributes to higher order association.
In this study, we found that the RING domain is also essential for efficient higher order self-association, which in turn promotes capsid-binding avidity. The partial attenuation of TRIM5␣ restriction activity by deletion or disruption of the RING domain may result, at least in part, from this decrease in higher order association. In general, replacement of the TRIM5␣ rh RING with a heterologous RING domain caused more severe defects in higher order self-association than replacement of the B-box 2 domain. The RING domain mutants are not globally misfolded because they dimerize and, in some cases (⌬RING, ⌬RING-L1 and 4R), partially inhibit HIV-1 infection (39,42) (Fig. 1F and supplemental Fig. S2).
In most cases, individual replacement of either the RING domain or the B-box 2 domain with a heterologous TRIM domain resulted in a significant reduction in higher order association, even when the RING and B-box 2 domains on the interacting protein partners were identical. When the replacement RING and B-box 2 domains were both derived from the same TRIM protein, the resulting chimeras exhibited much more efficient higher order self-association than most of the chimeras with the RING or B-box 2 domains individually replaced. This observation suggests that, at least with respect to higher order self-association, the RING and B-box 2 domains may function as a unit. Studies of the antigenic determinants of TRIM21 targeted in Sjögren syndrome have suggested that the RING-Linker 1-B-box 2 regions are intimately associated (43)(44)(45).
Our results suggest that TRIM sequences C-terminal to the RING and B-box 2 domains contribute to higher order selfassociation. Some C-terminal TRIM sequences indirectly contribute to higher order self-association through effects on dimerization. TRIM dimerization mediated by the coiled coil (and, for some TRIM proteins, the adjacent Linker 2 region as well) probably represents a prerequisite for higher order selfassociation. An alteration of the coiled coil (L194K) that dis-rupted TRIM5 dimerization eliminated higher order self-association. Likewise, deletion of the Linker 2 region, which is required for TRIM5 dimerization (25), also disrupted the ability of the protein to form higher order oligomers. Recently, a chimeric protein (TRIM5-21R) consisting of rhesus monkey TRIM5 with the RING domain of TRIM21 has been shown to assemble spontaneously into hexagonal arrays (see below); of interest, only the dimeric form exhibited efficient self-assembly, whereas the monomeric form assembled with much lower efficiency (46). Gel filtration analysis has shown that TRIM protein mutants containing deletions in the coiled-coil domain elute in fractions corresponding to monomers but not higher order oligomers (17). Together, these observations support a model in which TRIM dimerization is required for efficient higher order self-association.
C-terminal TRIM sequences also appear to contribute directly to higher order association, independent of any effects on TRIM dimerization. The higher order association of TRIM4 with the 5RB chimera, in which both the RING and the B-box 2 domains of TRIM4 are replaced by those of TRIM5, was more efficient than that between TRIM4 and TRIM5. In this case, sequences C-terminal to the B-box 2 domain in the RB chimera must contribute to the observed improvement in higher order association with TRIM4, relative to that seen for TRIM5. The 4 -5SPRY chimera, with the TRIM5␣ B30.2(SPRY) domain replacing that of TRIM4, efficiently forms higher order complexes with TRIM4; therefore, as shown for TRIM5␣ and TRIM6, the B30.2(SPRY) domain minimally contributes to the higher order self-association of TRIM4. Although deletion of the B30.2(SPRY) domain reduced the efficiency of some of the weak heterologous TRIM higher order associations, this domain appears to be dispensable for the more robust higher order self-association of TRIM5, TRIM6, and TRIM4.
We examined the TRIM5 and TRIM6 Linker 2 regions as potential contributors to higher order self-association. We were able to take advantage of the fact that TRIM6 dimerization does not depend upon Linker 2 sequences to demonstrate a Linker 2 requirement for TRIM6 higher order self-association. Study of a series of deletion mutants indicated that Linker 2 sequences near the coiled coil are important for TRIM6 higher order self-association. These Linker 2 sequences are well conserved in TRIM5 (Fig. 6B). Given the similar domain requirements for the higher order self-association of TRIM5 and TRIM6 and the ability of the RBCC-L2 segment of TRIM6 to replace functionally the TRIM5␣ rh RBCC-L2 segment (34), the Linker 2 region of TRIM5 probably contributes to its higher order self-association. Indeed, we identified a TRIM5␣ rh mutant (D247K/L248D/L249D) with changes in the conserved coiled-coil-proximal region of Linker 2 that dimerized efficiently but exhibited decreased higher order association with the wild-type TRIM5␣ rh protein. Thus, Linker 2 sequences apparently contribute to the higher order association of TRIM5 and TRIM6 dimers. Recently, changes in a more C-terminal segment of the TRIM5␣ rh Linker 2 region have been reported to decrease the formation of cytoplasmic bodies, one form of TRIM5 multimeric assembly (32).
A TRIM5␣ rh variant with a TRIM21 RING domain (TRIM5-21R) has recently been reported to assemble spontaneously into two-dimensional hexagonal arrays (46). This assembly occurred even when the B30.2(SPRY) domain was deleted and apparently required arginine 121 in the B-box 2 domain. As discussed above, the hexagonal assembly was much more efficient for TRIM5-21R dimers than monomers. These requirements are in good accordance with those necessary for higher order self-association in our co-immunoprecipitation assay. Therefore, the insights provided in this work should help to refine models for TRIM5 interaction with the retroviral capsid, which promotes the assembly of hexagonal arrays of TRIM5-21R indistinguishable from those formed spontaneously on carbon-coated grids (46). The low resolution of the available cryoelectron microscopic images of the TRIM5-21R lattices precludes precise definition of the architecture of the TRIM5-21R dimers on the hexagonal array. However, the symmetry and dimensions of the two-dimensional lattice and the expected size of the TRIM5-21R dimer have been used to suggest a working model (46). In this model, each edge of the hexagon is composed of two TRIM5-21R dimers; each end of the TRIM5-21R dimer coincides with either a 3-fold or 2-fold axis of symmetry. The assumption that TRIM5-21R dimers are in a parallel orientation provides a natural explanation for the different interactions required to achieve 3-fold and 2-fold symmetry at each dimer terminus. Our results support a model in which the RING-B-box 2 domains at one end of the TRIM5 dimer and the Linker 2 region at the other end mediate the contacts with adjacent dimers (Fig. 9). It is currently unknown whether trimeric or dimeric contacts in the TRIM5-21R hexagonal array are mediated by the RING-B-box 2 domains or the Linker 2 region. The Linker 2 region implicated in TRIM higher order self-association in this study is immediately C-terminal to the coiled coil and is known to contribute to the stability of dimers formed by some TRIM proteins, including TRIM5 (25). An element capable of stabilizing parallel dimers could readily rearrange to support intermolecular interactions between dimers; either dimeric or trimeric interactions could theoretically be supported by Linker 2 sequences in this setting (Fig. 9). Further work should enhance models of TRIM5 assembly.
The ability of TRIM5 dimers to self-associate and form hexagonal lattices would allow them to recognize the hexagonal lattices of the retroviral capsid with enhanced efficiency. We have previously demonstrated that, when B30.2(SPRY) domain-mediated direct binding to the retroviral capsid is weak, this cooperative effect of higher order self-association can dramatically increase the binding avidity and allow restriction of infection (26,28). Higher order self-association provides an opportunity for "pattern recognition" of the hexagonal capsid surface, thus extending the range of retroviruses potentially targeted by a given TRIM5␣ protein. Our characterization of higher order self-association provides further insights on how TRIM5 proteins might assemble on the viral capsid to recognize and/or disrupt the capsid structure. The nature and functional purpose of the arrays formed by other TRIM proteins are also of interest for future study.