TRIM37 prevents ectopic spindle pole assembly by peptide motif recognition and substrate-dependent oligomerization

Abstract

Tightly controlled duplication of centrosomes, the primary microtubule-organizing centers of animal cells, ensures bipolarity of the mitotic spindle and accurate chromosome segregation. The RBCC (RING-B-box-coiled coil) ubiquitin ligase TRIM37, whose loss is associated with elevated chromosome missegregation and the tumor-prone human developmental disorder Mulibrey nanism, prevents the formation of ectopic spindle poles assembling around structured condensates that contain the centrosomal protein centrobin. Here, we show that TRIM37’s TRAF domain, which is unique in the extended TRIM family, engages peptide motifs in centrobin to suppress condensate formation. TRIM family proteins form anti-parallel coiled-coil dimers with RING–B-box domains at each end. Oligomerization resulting from RING–RING interactions and conformational regulation through B-box-2–B-box-2 interfaces are essential for TRIM37 to suppress centrobin condensate formation. These results indicate that, similar to anti-viral TRIM ligases, TRIM37 activation is coupled to detection of oligomerized substrates, facilitated by recognition of specific motifs in the substrate, to enforce ubiquitination-mediated clearance of ectopic centrosomal protein assemblies.

Centrosomes are the major microtubule-organizing centers of animal cells. Centrosomes are composed of a centriolar core that recruits a pericentriolar material matrix (PCM matrix) to nucleate and anchor microtubules. Centrosome duplication is tightly controlled so that mitotic cells have exactly two centrosomes that catalyze microtubule assembly to form the two poles of the bipolar spindle ^1,2. Centrioles are ~2 gigadalton macromolecular assemblies consisting of more than 100 proteins organized into multiple substructures ^3–5. Given the complex interconnected nature of the centrosomal protein network, a key question is how rogue collections of centrosomal proteins are prevented from assembling in the cytoplasm to form ectopic structures with microtubule-organizing capacity.

Recent work identified the ubiquitin ligase TRIM37, a TRIpartite Motif (TRIM) family member, as essential for preventing the formation of ectopic centrosomal protein assemblies ^6–11. TRIM family ligases contain an RBCC domain comprised of a RING domain, B-box, and an anti-parallel coiled-coil ^12–14. Unique among TRIM proteins, TRIM37 harbors a TRAF domain predicted to bind to peptide ligands ^15,16. In cells with centrioles, TRIM37 loss induces formation of a large and highly ordered assembly (termed a “condensate”) containing the centrosomal protein centrobin (encoded by CNTROB), alongside an array of smaller foci containing the centriolar protein centrin. In ~25% of mitotic cells, the centrobin condensate forms an ectopic spindle pole, causing transient or persistent multipolarity and increased chromosome missegregation ^6,7,9. CNTROB deletion suppresses ectopic spindle pole formation in TRIM37Δ cells ⁹. Thus, ectopic centrobin-scaffolded condensates function as aneuploidy generators and may contribute to the high prevalence of tumors in Mulibrey nanism, a TRIM37 loss-of-function disorder ^17,18.

How TRIM37 recognizes nascent centrobin assemblies and promotes their disassembly remains unclear. Here, we identify short peptide motifs in a predicted disordered C-terminal region of centrobin that engage TRIM37’s TRAF domain, enabling binding and suppression of ectopic assembly in vivo. TRIM37 is predicted to form anti-parallel coiled-coil dimers that place their RING-B-box-2 domains on opposing ends ^12,19. As RING domain dimerization is typically required for ubiquitin ligase activation ¹³, the RING-B-box-2 region must oligomerize, potentially facilitated by binding to oligomeric substrates, for efficient substrate ubiquitination. Using engineered recombinant miniTRIM37s, we show that the RING domain, but not the B-box-2, is the primary oligomerization interface, while the B-box-2 interface controls oligomer conformation. In vivo, RING and B-box interfaces are both essential to recognize centrobin assemblies and prevent condensate formation. These findings suggest that, like the anti-viral TRIM5α ligase ²⁰, TRIM37 combines motif binding with substrate-induced oligomerization and conformation-based detection of substrate oligomerization to effect ubiquitination-mediated clearance of ectopic centrosomal protein assemblies.

RESULTS

TRIM37 can prevent centrobin condensates from forming in the cytosol

In TRIM37Δ RPE1 cells and fibroblasts isolated from Mulibrey nanism patients, which lack TRIM37, most cells develop a single highly-ordered non-centrosomal condensate containing the centriolar protein centrobin. In ~25% of mitotic cells, the centrobin-containing condensate acquires the ability to nucleate microtubules during mitotic entry and forms an ectopic spindle pole that induces transient or persistent multipolarity and elevates chromosome missegregation (Fig. 1A; ^6,9). Consistent with an essential role in condensate formation, knockout of the gene encoding centrobin (CNTROB) in TRIM37Δ cells eliminated detectable centrobin signal and ectopic spindle poles ^6,9. Prior immunofluorescence suggested that PLK4 also localizes to condensates ^9,10, but PLK4 knockout was not feasible because it leads to centriole loss, delays mitosis, and triggers a p53-dependent arrest ⁹. To determine whether PLK4 is a condensate component, we generated a TRIM37Δ cell line that inducibly expresses Cas9 and has p53 knocked down (TP53sh). PLK4 knockout in this system revealed that the signal at condensates is not due to PLK4 presence but to antibody cross-reactivity with another protein in the condensate (Extended Data Fig. 1A–C). Thus, centrobin-scaffolded condensates do not contain PLK4.

Figure 1. TRIM37 can prevent the formation of centrobin condensates in the cytosol as well as at centrosomes.

(A) Images and schematic summary of phenotypes in interphase (left) and mitotic (right) wildtype and TRIM37Δ RPE1 cells. TRIM37 loss leads to the formation of a centrobin-containing condensate. During mitosis, centrobin condensates frequently form ectopic spindle poles. (B) Images and schematic summary of a TRIM37Δ RPE1 cell expressing mRuby-centrobin; mNG-tagged CEP192 marks centrosomes and SiR-DNA was added to visualize chromosomes. TRIM37Δ cells typically have a single condensate; following mitosis, the left daughter inherits the condensate, while the right daughter (dashed box in 54 min panel) does not. The daughter cell lacking a condensate is shown (row below), with the region containing the centrosome magnified on the right; top row is a merge of the CEP192 centrosome marker and centrobin; bottom row shows only centrobin. Centrosomes (yellow arrows) and a newly forming condensate (cyan arrowheads) are indicated. Times are relative to metaphase in the mother cell. A similar phenomenon was observed in 4 cells. (C) (left) Schematic of the approach (see also Extended Data Fig. 1D,E); (middle) immunofluorescence images of an inducible TRIM37 knockout (iTRIM37KO) cell line without and with doxycycline-induced knockout. Cells were stained for centrobin and the centriolar marker CPAP. Centrosomes are magnified to the left of the lower magnification view and, when present, centrobin condensates to the right. (right) Graph plots percentage of cells with centrobin condensates (ns=not significant, p=0.159; **** indicates p<0.0001). (D) (top left) Scheme for analyzing the effect of TRIM37 knockout in cells lacking centrosomes. Images of cells treated as indicated and labeled for centrobin and CPAP are shown. The absence of focal CPAP staining indicates absence of centrioles (see also Extended Data Fig. 1F). Magnified view of the condensate shows centrobin staining on top (magenta) and CPAP signal, which is absent, on the bottom. (right) Graph plotting percentage of cells with condensates for the indicated conditions. (ns=not significant, p=0.811; **** indicates p<0.0001). In C and D, points represent biological replicates, bar heights represent the weighted mean, and n values are total number of cells analyzed. Error bars are 95% confidence intervals. One-way ANOVA and Dunnett’s multiple comparisons test were used to calculate p values. Scale bars, 5μm (low magnification views) and 1 μm (high magnification insets).

Centrobin condensates form when TRIM37 is deleted. Complementing TRIM37Δ cells with epitope-tagged wild-type TRIM37 prevents condensate formation, whereas complementation with a ligase-mutant (C18R) TRIM37 does not, demonstrating that TRIM37’s ubiquitin ligase activity is essential for suppression ^9,10. Although fluorescently tagged ligase-mutant TRIM37 (Lig^mut TRIM37-mNG) cannot prevent condensates, it stably binds to them and allows real-time tracking of condensate dynamics ^9,10. When a cell with a single condensate divides, one daughter cell inherits the condensate, while the other does not (Fig. 1B). In cells born without a condensate, Lig^mut TRIM37-mNG hyper-accumulates around the centrosome before budding off, suggesting that new condensates form by budding from centrosomes ⁹. To confirm that this is also true in the absence of ligase-mutant TRIM37, we imaged mRuby-centrobin in cells expressing CEP192-mNG to mark the centrosomes. In TRIM37Δ daughter cells lacking condensates, mRuby-centrobin hyper-accumulated around the centrosome before budding off to form a new condensate (Fig. 1B). This raises the question of why only a single condensate forms, and whether condensates can spontaneously emerge in the cytoplasm or require ‘nucleation’ by a centrosome or pre-existing condensate.

To determine whether cells lacking a centriole or condensate can spontaneously form a condensate, we engineered and validated an inducible TRIM37 knockout (Fig. 1C; Extended Data Fig. 1D,E). We first depleted centrosomes from cells by treatment with the PLK4 inhibitor centrinone (Extended Data Fig. 1F; ²¹) and then induced TRIM37 knockout with doxycycline (Fig. 1D). If centrosomes nucleate condensates, we would expect either no condensate formation or the simultaneous emergence of multiple condensates. In contrast to this expectation, inducibly deleting TRIM37 after centrosome loss resulted in centrobin condensate formation at a frequency similar to that in constitutive TRIM37Δ cells with centrosomes (Fig. 1D). Notably, even in cells lacking centrosomes, only one condensate typically formed. Quantification showed that in cells with TRIM37 constitutively knocked out (TRIM37), 97% of condensate-positive cells had only one condensate (252/259 cells). Similarly, after inducibly knocking out TRIM37 in cells pretreated with centrinone, 94% (77/82) of condensate-positive cells had only one condensate (77/82 cells). These results suggest that condensates form through nucleated assembly, in which incoming centrobin monomers preferentially bind to existing oligomers rather than other monomers, most probably due to their ability to make multiple contacts simultaneously when interacting with an oligomer. This is akin to actin filament formation, in which actin dimers form only one interface when they interact with another actin monomer making the resulting dimers unstable. However, in the rare cases in which a trimer is able to form, it serves as a nucleus because subsequent incoming monomers can make lateral and longitudinal contacts simultaneously ²². Consistent with this idea, prior work has shown that centrobin condensates have a complex repetitive structure, typically appearing as hexagonally-packed sheets or linear striated structures ^6,9.

These results suggest that TRIM37 suppresses the formation of oligomeric centrobin-containing assemblies both at centrosomes and in the cytoplasm. In the absence of TRIM37, centrobin condensates form. While condensates preferentially assemble at centrosomes—likely building on an existing centrobin scaffold—condensates can also nucleate spontaneously in the cytoplasm.

The TRIM37 TRAF domain is required for centrobin binding and condensate clearance

TRIM family ligases are defined by an RBCC domain, which includes a RING domain, B-box, and an anti-parallel coiled-coil ^12,13. AlphaFold modeling predicts that, like other TRIM proteins, the TRIM37 coiled-coil forms a T-shaped anti-parallel dimer that places its two RING-B-box domains on opposite sides (Fig. 2A; Extended Data Fig. 2A,B). Unique among TRIM family proteins, TRIM37 harbors a TRAF domain (Fig. 2A; Extended Data Fig. 2C), a domain first identified in TNF receptor-associated factors that bind short peptide sequences, with different TRAF domains exhibiting binding specificity for different peptide motifs ^15,16. AlphaFold predicts that the two TRAF domains are positioned on the stem of the T, just below the extended antiparallel coiled-coil (Fig. 2A). We hypothesized that the TRAF domains recognize specific peptide motifs in targets such as centrobin to position the substrate for ubiquitination by the RING-E2 complex. To test this model, we designed a mutant based on structural homology to the USP7 TRAF domain, which binds peptide ligands in p53 ^23,24. Specifically, we disrupted TRAF domain–peptide interactions by mutating a critical exposed tryptophan residue to alanine (Fig. 2B; TRAF^mut, W373>A; ¹⁰). After confirming expression of transgene-encoded WT, Lig^mut, and TRAF^mut TRIM37 proteins in TRIM37Δ cells (Fig. 2B), we analyzed the frequency of centrobin condensate formation. Unlike expression of WT TRIM37, which fully suppressed condensate formation, expressing TRAF^mut or Lig^mut TRIM37 failed to do so (Fig. 2C–E). Thus, TRIM37’s TRAF domain is essential for suppressing condensate formation, likely by engaging peptide ligands.

Figure 2. The TRIM37 TRAF domain is required for binding to centrobin and to prevent centrobin condensate formation in vivo.

(A) AlphaFold model of a TRIM37 dimer. TRAF domains are below the mid-point of the anti-parallel coiled-coil and the RING and B-box-2-containing regions are positioned on opposite ends. One monomer is in grey and the second in rainbow from N- to C-terminus. See also Extended Data Fig. 2A,B. (B) (left) Schematic of TRIM37 with an AlphaFold model of the TRAF domain highlighting the tryptophan residue (W373) mutated to prevent peptide ligand engagement (see also Extended Data Fig. 2C). (right) Immunoblot of TRIM37Δ RPE1 cells expressing the indicated FLAG-tagged TRIM37 variants; α-tubulin serves as a loading control. Numbers below are the percentage of cells in the lentivirally transduced pools that express the transgene. Experiment repeated twice with similar results. (C) & (D) Representative images of control and TRIM37Δ RPE1 cells (C) and of TRIM37Δ cells expressing the indicated TRIM37 transgenes (D). Cells were labeled for centrobin and DNA; centrosomes (yellow arrows) and condensates (cyan arrowheads) are indicated. Centrosomes are magnified to the left of the lower magnification view and, when present, centrobin condensates to the right. Experiments performed three or more times with similar results. (E) Graph plots percentage of cells with centrobin condensates for the indicated conditions. n is the number of cells analyzed. One-way ANOVA and Dunnett’s multiple comparisons test were used to calculate p values (ns=not significant, p=0.97; **** indicates p<0.0001). Bar heights represent the weighted mean; error bars are the 95% confidence interval. Points represent biological replicates. n values are the total number of cells analyzed. (F) (left) Schematic shows how the interaction of TRIM37 with centrobin was tested following co-expression in FreeStyle 293F cells. TRIM37 ligase-activity was mutated to enable robust expression and binding assessment. The centrobin fragment used (1-767) is soluble and contains the TRIM37-binding region (see Fig. 3A & Fig 4A). (right) Immunoblot of 5xMyc-tagged Centrobin (1-767) and TRIM37 in cell lysates and following anti-Myc immunoprecipitation. α-tubulin serves as a lysate loading control. Scale bars, 5μm (low magnification views) and 1 μm (high magnification insets).

To determine whether the TRIM37 TRAF domain is important for its binding to centrobin, we employed a human cell co-expression approach (Fig. 2F; ⁹). Since full-length centrobin is largely insoluble, we assessed binding to a soluble truncated version of centrobin (aa 1-767), which lacks the C-terminal 125 amino acids. Because TRIM37 negatively autoregulates its own abundance ¹⁰, we used Lig^mut TRIM37 (C18R), which is present at higher levels than WT TRIM37, to analyze centrobin binding. Lig^mut TRIM37(C18R) bound centrobin (1-767) robustly, whereas Lig & TRAF^mut TRIM37(C18R;W373A) showed significantly reduced binding (Fig. 2F). We conclude that the TRAF domain-peptide ligand interface is necessary for TRIM37 to bind to centrobin and suppress centrobin condensate formation.

The TRIM37 TRAF domain recognizes specific peptide motifs in centrobin

To define how the TRIM37 TRAF domain recognizes centrobin, we used a co-expression binding assay map the TRIM37-binding region within centrobin to aa 576-767 (Fig. 3A). AlphaFold modeling predicted three potential TRAF-binding motifs within this region, with motif 1 having the highest confidence and motif 3 the lowest; motifs 2.1 and 2.2 overlap (Fig. 3B; Extended Data Fig. 3A). To test the role of these motifs in TRAF domain-mediated recognition, we engineered a TBM^mut (TRIM37 Binding Motif mutant) centrobin by altering 18 predicted interface residues in all motifs (Fig. 3B,C). This mutation largely eliminated TRIM37 binding, confirming the motifs’ essential role in the recognition of centrobin by TRIM37 (Fig. 3C). To directly assess motif binding, we performed fluorescence polarization-based peptide binding assays using purified recombinant WT or mutant (W373>A) TRIM37 TRAF domains (Fig. 3D; Extended Data Fig. 3B). This confirmed specific binding of motif 1 and weaker specific binding of motifs 2.1 and 2.2; no specific binding was observed with motif 3. Although these TRAF-peptide interactions are relatively low affinity, they fall within the expected for TRAF-peptide ligands ^23,24. Given the close proximity of the two TRAF domains in the anti-parallel TRIM37 dimer (Fig. 2A), avidity effects from multiple low-affinity interactions may enhance centrobin recognition.

Figure 3. The TRIM37 TRAF domain recognizes specific peptide motifs in centrobin.

(A) (left) Schematic of full-length (FL) and engineered centrobin fragments (cc, predicted coiled-coils). (right) Immunoblot analyzing TRIM37 binding to centrobin fragments, performed as in Fig. 2F. Input here is crude extract, prior to centrifugation. For FL centrobin, which is largely insoluble, the clarified supernatant used for the anti-Myc immunoprecipitation is depleted of the TRIM37 ligase-mutant, which pellets with the insoluble centrobin assemblies, explaining the absence of a TRIM37 band in the IP. Experiment repeated twice with similar results. (B) (left) Sequence of the TRIM37-binding region of centrobin, highlighting potential TRAF domain-binding motifs and mutations to disrupt them; (right) AlphaFold models of centrobin motifs (colored by AlphaFold prediction confidence score) interfacing with the TRIM37 TRAF domain shown in gray (see also Extended Data Fig. 3A). (C) Immunoblot analyzing TRIM37 binding to wildtype centrobin (aa 1-767) or a variant in which the putative TRAF-binding motifs were mutated as in B. Experiment, conducted as in Fig. 2F, was repeated three times with similar results. (D) (left) Coomassie-stained gel showing purified recombinant WT versus W373A-mutant TRAF domains. Lanes are from the same gel; an intervening lane was removed (white line). (right) Binding of the indicated fluorescent peptides to the purified TRAF domains was monitored using fluorescence polarization (see also Extended Data Fig. 3B). Points are the mean of triplicate technical replicates (n=3), error bars are the SD. Data are representative of two independent experiments. (E) (left) Representative images after expression of wildtype or TRIM37 binding motif-mutant centrobin in TRIM37Δ; CNTROBΔ or TRIM37 WT; CNTROBΔ cells (see also Extended Data Fig. 3C–E). (right) Graph shows the percentage of cells with centrobin condensates for the indicated conditions. One-way ANOVA and Šidák’s multiple comparisons test were used to calculate p values (ns=not significant, p=0.668; **** indicates p<0.0001). Bar heights represent the weighted mean; error bars are the 95% confidence interval. Points represent biological replicates. n values in bars are the total number of cells analyzed. Centrobin transgenes included an mRuby tag for visualization. Scale bars, 5μm (low magnification views) and 1 μm (high magnification insets).

TRIM37 suppresses centrobin condensate formation via TRAF domain-binding motifs

If TRIM37 recognizes centrobin via its TRAF-binding motifs in vivo, then mutating these motifs should prevent TRIM37 from targeting centrobin, leading to condensate formation even in the presence of WT TRIM37. To test this, we used CRISPR/Cas9 to generate stable CNTROBΔ and CNTROBΔ;TRIM37Δ RPE1 cell lines (Fig. 3E; Extended Data Fig. 3C–E) and introduced lentiviral transgenes expressing either WT centrobin or mutant centrobin with its TRAF binding motifs mutated (TBM^mut), fused to mRuby (Fig. 3E). In TRIM37Δ cells, both WT and TBM^mut centrobin formed condensates (Fig. 3E), confirming that TRAF binding motif mutations do not inherently prevent condensate formation. However, in the presence of WT TRIM37, only TBM^mut centrobin formed condensates (Fig. 3E). These data indicate that TRAF domain-mediated binding of TRIM37 to specific motifs in centrobin is a critical step in preventing centrobin condensate formation. We additionally mutated only motif 1, which exhibited the most robust specific binding in the purified TRAF domain-peptide binding assay (Extended Data Fig. 4A,B). While mutating motif 1 increased condensate frequency compared to WT centrobin, the motif 1 centrobin mutant exhibited a much lower frequency of condensates compared to TBM^mut centrobin (9% vs 69%; Extended Data Fig. 4A,B), indicating that multiple binding motifs contribute to TRIM37-mediated centrobin recognition.

Collectively, these results establish that the TRAF domain of TRIM37 plays a critical role in preventing the formation of centrobin condensates and define molecular interfaces through which the TRAF domain mediates centrobin recognition.

Centrobin oligomerization is important for its ubiquitination by TRIM37

TRIM37 suppresses the formation of centrobin condensates, which can function as non-centrosomal spindle poles ^6,9. While TRIM37 prevents the formation of large ectopic centrobin assemblies, it allows centrobin to function at centrioles and in ciliogenesis ^25,26, suggesting that TRIM37 may selectively recognize and ubiquitinate oligomerized centrobin rather than unassembled centrobin in the cytosol. To determine whether centrobin ubiquitination by TRIM37 depends on its oligomerization, we first assessed the requirements for centrobin oligomerization using centrifugation. Our prior work showed that full-length centrobin expressed in human cells oligomerizes readily and fractionates into the pellet following centrifugation ⁹. In addition, when TRIM37 Lig^mut is co-expressed with full-length centrobin, TRIM37 Lig^mut co-sediments with the centrobin assemblies, mimicking the stable association of TRIM37 Lig^mut with centrobin-scaffolded condensates observed in vivo ⁹. To determine the regions of centrobin required for its oligomerization, we analyzed the sedimentation of full-length centrobin and its fragments (Fig. 4A). While full-length centrobin largely partitioned into the pellet, all truncated centrobin fragments remained in the supernatant, indicating that both N-terminal and C-terminal regions are required for oligomerization (Fig. 4A). We conclude that centrobin oligomerization requires both N-terminal and C-terminal regions that are distinct from the TRIM37-binding region.

Figure 4. Centrobin oligomers are targeted for ubiquitination by TRIM37, whose oligomerization requires the RING domain.

(A) (left) Schematic of full-length centrobin and analyzed N-and C-terminal truncations highlighting the TRIM37 binding region (green); (right) immunoblot showing the results of a sedimentation analysis of the indicated centrobin truncations expressed in Freestyle 293F cells (S, supernatant; P, pellet). α-tubulin serves as a loading and solubility control. The experiment was performed twice with similar results. (B) Immunoblot showing centrobin fragment ubiquitination by WT TRIM37 following co-expression in FreeStyle 293F cells followed by immunoprecipitation. Endogenous ubiquitin was detected in the immunoprecipitates using an anti-ubiquitin antibody. The experiment was performed twice with similar results. (C) Schematic model of substrate oligomerization-directed activation of TRIM37 ligase activity. (D) Schematics of mini-TRIM37 variants engineered to examine the roles of the RING and B-box-2 domains in oligomerization. (E) SEC-MALS analysis of purified recombinant miniTRIM37s. The numbers above indicate the predicted molecular weights from the primary sequences and the native molecular weights of the entities in the major UV peaks measured by MALS.

Next, we co-expressed WT TRIM37 with either full-length centrobin or two centrobin fragments, that both contain the TRIM37-binding region and analyzed their ubiquitination (Fig. 4B). While WT TRIM37 robustly ubiquitinated the small pool of full-length centrobin in the supernatant, neither of the soluble centrobin fragments were ubiquitinated. As these two fragments both bind to TRIM37 and together span the entire centrobin sequence, these data suggest that centrobin oligomerization is important for its ubiquitination by TRIM37.

The TRIM37 RING–B-box-2 oligomerizes primarily via a RING-RING interface

To investigate how oligomeric centrobin assemblies activate TRIM37-mediated ubiquitination, we focused on how TRIM37 dimers might oligomerize. Prior work on TRIM proteins ^12,13 supported by AlphaFold modeling (Fig. 2A) suggests that TRIM37 forms an antiparallel dimer that positions its two RING-B-box-2 domains on opposite ends of a T-shaped dimer. Since RING domain dimerization is important for activating the bound E2-ubiquitin complex for substrate ubiquitination ¹³, we hypothesized that binding to an oligomeric substrate promotes TRIM37 dimer oligomerization, leading to ligase activation and substrate ubiquitination (Fig. 4C). Such an activation mechanism has been established for TRIM5α, which detects viral nucleocapsid shells and targets them for ubiquitin-mediated degradation ^20,27. Like TRIM37, TRIM5α has a RING–B-box-2-coiled coil (RBCC) domain; however, instead of a TRAF domain, it has a SPRY domain that recognizes the viral nucleocapsid. TRIM5α oligomerization, driven by capsid binding, is important for its activation. The ability of TRIM5α dimers to oligomerize in a manner that matches the geometry of the viral nucleocapsid requires an interaction interface on its B-box-2 domains, with a well-characterized point mutation in this interface disrupting oligomerization and the ability to restrict HIV infection ^28–30.

To determine whether TRIM37 oligomerizes similarly, we adapted an approach used for TRIM5α, which employed miniTRIMs containing either a RING-B-box-2 or B-box-2 domain fused to a short hairpin coiled-coil from T. thermophilus seryl-tRNA synthetase (PDB ID 1SER, ³¹; ³⁰). This approach allowed analysis of the RING and B-box-2 interfaces that mediate oligomerization independently of the extended anti-parallel coiled-coil of the native dimer. Both miniTRIM5αs were dimers, with a point mutation in B-box-2 disrupting dimerization ³⁰. Inspired by this prior work, we designed, expressed and purified two miniTRIM37 constructs: (1) RING–B-box-2–hairpin coiled-coil (RING–B-box-2–hcc), and (2) B-box-2–hairpin coiled-coil (B-box-2–hcc) (Fig. 4D; Extended Data Fig, 4C). Using size exclusion chromatography coupled to multi-angle light scattering (SEC-MALS), we found that, like the miniTRIM5α RING–B-box-2–hcc, miniTRIM37 RING–B-box-2–hcc was dimeric (Fig. 4E). However, in contrast to miniTRIM5α B-box-2–hcc ³⁰, miniTRIM37 B-box-2–hcc was monomeric (Fig. 4E), indicating that the TRIM37 B-box-2 domain alone is insufficient to support oligomerization. These results suggest that TRIM37 dimer oligomerization is primarily driven by RING-RING interactions, rather than by B-box-2-mediated interfaces.

The TRIM37 B-box 2 interface is dispensable for dimerization but impacts oligomer conformation

Consistent with our biochemical results, AlphaFold modeling of the TRIM37 RING–B-box-2–hcc dimer predicted an extensive RING–RING dimer interface (Fig. 5A; Extended Data Fig. 4D), with a buried surface area of ~ 1070 Ų per protomer (calculated by PDBePISA: https://www.ebi.ac.uk/pdbe/pisa/papers/pisa-web.pdf). The interface is predominantly hydrophobic, with I9, F13, L27, V64, V72, and L79 making symmetric contacts across the dimer interface (Fig. 5B). The predicted RING–RING interface matches almost perfectly a prior crystal structure of the TRIM37 RING domain dimer (PDB 3LRQ; <0.5 Å Cα rmsd in a single chain, and a near-exact dimer packing match). Although the B-box-2 domain is insufficient for oligomerization, modeling predicts a secondary, smaller B-box-2-mediated dimer interface analogous to the TRIM5α B-box-2 dimer (Fig. 5B; ³⁰). The predicted B-box-2 interface is anchored by three residues: H115 makes symmetric pi-stacking interactions across the dimer interface, and L119 and W120 form additional hydrophobic interactions. The crystal structure of the TRIM5α B-box-2 dimer revealed a three-layer interface: Layer 1 involves electrostatic contacts between residues E102 and K103 across the dimer interface; Layer 2 involves hydrophobic and pi-stacking interactions between residues W117 and L118, and Layer 3 involves electrostatic interactions between residues E120, R121, and T130 ³⁰. The TRIM37 B-box interface has residues that are positioned similarly to one of the Layer 2 residues (H115 in TRIM37) and two of the Layer 3 residues (L119 and W120 in TRIM37).

Figure 5. Functional analysis of predicted RING and B-box 2 interfaces of TRIM37.

(A) AlphaFold model of the miniTRIM37 RING–B-box-2–hcc dimer, highlighting the predicted RING–RING and B-box-2–B-box-2 interfaces. (B) Detailed view of the 2 predicted interfaces, highlighting key residues. Mutations were engineered in the residues marked in red to disrupt key interface contacts. (C) SEC-MALS analysis of the indicated miniTRIM37 RING–B-box-2–hcc variants. Numbers next to the predicted molecular weights in the table below are the native molecular weights of the entities in the major UV peaks measured by MALS. The SEC-MALS analysis was repeated twice with identical results. (D) Immunoblot of transgene-mediated expression of the indicated engineered TRIM37 variants in TRIM37Δ cells. The lanes shown are from the same gel and immunoblot; one intervening lane was removed, as indicated by the white line. α-tubulin serves as a loading control. The experiment was performed twice with similar results. (E) Immunofluorescence images of TRIM37Δ cells expressing the indicated transgene variants after fixing and staining for centrobin and DNA. Centrosomes (yellow arrows) and centrobin condensates (cyan arrowheads) are marked on the images. (F) Graph plots the percentage of cells with centrobin condensates for the indicated conditions. One-way ANOVA and Dunnett’s multiple comparisons test were used to calculate p values (**** indicates p<0.0001). Bar heights represent the weighted mean. Error bars are the 95% confidence interval. Points represent biological replicates. n values in the bars are the total number of cells analyzed. (G) Analysis of TRIM37 autoregulation following expression in FreeStyle 293F cells. The experiment was performed twice with similar results. (H) Western blot analyzing centrobin ubiquitination by the indicated TRIM37 variants. HA-ubiquitin was co-transfected along with FL centrobin and the indicated TRIM37 variants. Ubiquitination assessed is on the small pool of soluble FL centrobin following immunoprecipitation. The experiment was performed twice with similar results.

To evaluate the contribution of RING–RING and B-box-2–B-box-2 interfaces to miniTRIM37 RING–B-box-2–hcc dimerization, we designed two mutants. To disrupt the RING–RING interface, we mutated four key hydrophobic residues to charged residues (F13D, V64D, W68E, and L79D; Fig. 5B); these residues are distant from the E2-binding interface ^32,33 and are not expected to affect E2 binding. To disrupt the B-box-2 interface, we mutated H115 and L119 to alanine (Fig. 5B). SEC-MALS analysis revealed that mutation of the RING–RING interface disrupted dimer formation (Fig. 5C). By contrast, the B-box-2 interface mutant remained dimeric, with a native molecular weight identical to WT miniTRIM37 RING–B-box-2–hcc (Fig. 5C), indicating that the B-box-2 domain is dispensable for dimerization. These results confirm that TRIM37 cross-dimer interactions are primarily mediated by the RING domain. Interestingly, the miniTRIM37 RING–B-box-2–hcc B-box-2 mutant eluted at a higher volume in size exclusion chromatography (Fig. 5C), suggesting that it adopts a more compact conformation than the WT dimer. Introducing the same B-box-2 interface mutations into the monomeric miniTRIM37 B-box-2–hcc did not alter its elution volume (Extended Data Fig. 4E). Thus, while the B-box-2 interface is not required for oligomerization, it potentially affects the conformation of TRIM37 oligomers. These findings indicate that, while the B-box-2 domain is not required for TRIM37 oligomerization, it likely modulates oligomer conformation, distinguishing TRIM37 from TRIM5, where the B-box-2 interface is essential for oligomerization.

RING and B-box-2 interface mutants both compromise TRIM37 function

To test the functional importance of the RING and B-box-2 interfaces in vivo, we introduced transgenes expressing TRIM37 variants into TRIM37Δ RPE1 cells. After confirming expression (Fig. 5D), we compared the frequency of centrobin condensate formation. Both the RING and B-box-2 interface mutations impaired TRIM37’s ability to prevent centrobin condensate formation in vivo (Fig. 5E,F). Thus, while the B-box-2 interface is not required for oligomerization, it is still critical for TRIM37 function.

We next analyzed the impact of mutating the RING and B-box-2 interface mutations on the autoregulation of TRIM37 by its ligase activity. Our prior work showed that the C18R mutation, which impairs ligase activity, stabilizes TRIM37 and significantly elevates its expression relative to WT TRIM37 (Fig. 5G,H; Extended Data Fig. 5A; ¹⁰). Similarly, the RING interface mutant showed elevated expression, comparable to the ligase mutant. By contrast, the B-box-2 interface and TRAF mutants had a more minor effect (Fig. 5G,H; Extended Data Fig. 5A). These findings suggest that TRIM37 autoregulation depends primarily on the RING interface and ligase activity, with less reliance on the B-box-2 interface and TRAF domain.

Finally, we analyzed full-length centrobin ubiquitination by TRIM37 variants using co-expression analysis in human cells (Fig. 5H). Consistent with prior work, the results revealed robust ubiquitination by wild-type TRIM37 and no ubiquitination by the ligase mutant. The RING interface mutant behaved similarly to the ligase mutant. Ubiquitination was significantly reduced compared to wild-type TRIM37 by the B-box-2 interface mutant and was more substantially reduced by the TRAF domain mutant (Fig. 5H). In support of this data, immunoblotting of crude cell extracts to assess centrobin levels, which reflects how efficiently it is targeted for ubiquitination by TRIM37, revealed significantly higher centrobin levels when it was co-expressed with ligase-mutant, TRAF-mutant, or RING-mutant TRIM37, relative to WT TRIM37; centrobin levels were more mildly elevated when it was co-expressed with the B-box-2 mutant (Fig. 5H, Extended Data Fig. 5A).

Overall, these findings indicate that TRAF domain-mediated recognition of specific motifs, RING domain-mediated oligomerization, and the B-box-2 interface collectively contribute to preventing the formation of centrobin condensates. While the B-box-2 interface is not essential for oligomerization and only modestly affects ubiquitin ligase activity, it appears to modulate TRIM37 oligomer conformation, potentially allowing TRIM37 to geometrically recognize ectopic centrobin assemblies and initiate their ubiquitination-mediated clearance in vivo.

Detection of centrobin condensates in vivo requires the TRIM37 B-box-2 interface

To investigate the role of the B-box-2 interface in TRIM37 localization to centrobin condensates in vivo, we took advantage of the fact that ligase-mutant TRIM37 binds stably to the condensates but cannot disassemble them ⁹. We expressed mutant TRIM37 variants combining the ligase mutation (C18R) with the TRAF mutant or with the mutations disrupting the B-box-2 or RING interfaces in TRIM37Δ cells and monitored their localization. Using an epitope tag to detect the transgene-encoded TRIM37 variants, we quantified the ratio of each variant to the centrobin signal at condensates. This analysis revealed that, while ligase-mutant TRIM37 robustly concentrated on centrobin condensates, the ligase-and-TRAF double mutant failed to do so (Fig. 6A–C). Similarly, combining the ligase mutant with either the B-box-2 or RING interface mutants also impaired localization to condensates (Fig. 6A–C). Consistent with these findings, single mutations in the B-box-2 interface, RING interface, or TRAF domain also reduced TRIM37’s association with centrobin condensates (Extended Data Fig. 5B,C). These data indicate that the B-box-2 interface is essential for recognizing oligomerized substrates in vivo. We propose that the B-box-2 interface enables TRIM37 to geometrically sense oligomeric centrobin assemblies and target them for ubiquitination by stabilizing the substrate-bound active conformation of TRIM37 oligomers (Fig. 6D).

Figure 6. The RING and B-box-2 interfaces, along with the TRAF domain, facilitate detection of centrobin oligomers in vivo.

(A) (left) Schematics highlighting TRIM37 variants expressed from transgenes in TRIM37Δ cells. (right) Immunofluorescence images of fixed TRIM37Δ cells labeled for centrobin (magenta) and for the FLAG epitope fused to the expressed transgene-encoded TRIM37 variants (green). Condensates, when present, are highlighted by cyan arrowheads and are shown in a magnified view with separated channels on the right. (B) Graph plotting the ratio of FLAG to centrobin signal at condensates in cells expressing the indicated TRIM37 variants. Plotted values were normalized by dividing by the median value in the TRIM37 ligase-mutant. The mean and standard deviation are plotted over individual values. n values for each condition: Lig^mut, n=50; Lig^mut & TRAF^mut, n=32; Lig^mut & RING^mut, n=50; Lig^mut & B-box-2^mut, n=28. p-values are from unpaired, two-sided t-tests; **** indicates p<0.0001. (C) Immunoblot of transgene-mediated expression of the indicated engineered TRIM37 variants in TRIM37Δ cells. α-tubulin serves as a loading control. This experiment was performed twice with similar results. (D) Model for the substrate-dependent oligomerization and activation of TRIM37 dimers. Model highlights how TRIM37 selectively targets oligomers resulting in ligase activation (i), and summarizes the effect of TRIM37 TRAF, RING, and B-box-2 mutations on target binding and ligase activation(ii).

DISCUSSION

Centrosomes are multi-gigadalton macromolecular assemblies composed of approximately 150 proteins organized into various substructures ^3–5. Many of these interconnected structures (centriolar microtubules, cartwheel, inner scaffold, pericentriolar material) are polymeric. How the dimensions of these polymeric structures are specified and how ectopic polymeric assemblies containing centrosomal proteins are prevented from forming in the cytoplasm are important questions. While TRIM37 was initially thought to act at peroxisomes ^34,35, recent work has suggested that its major role is to restrict the growth of several ectopic polymeric assemblies of centrosomal proteins ^6,9,10. In this work, we address the molecular mechanism by which TRIM37 recognizes and prevents assembly of ectopic spindle poles that form on structured condensates of the centrosomal protein centrobin. These ectopic poles elevate chromosome missegregation during cell division and likely contribute to the high tumor incidence as well as other phenotypes in Mulibrey nanism patients ^6,9. Our prior work showed that deleting the gene encoding centrobin suppresses ectopic spindle poles in cells lacking TRIM37 function ⁹. Thus, a key question was how TRIM37 recognizes and selectively promotes the clearance of ectopic oligomeric centrobin condensates, while sparing centrobin monomers in the cytoplasm and endogenous centrobin assemblies at centrioles, which are important for ciliogenesis ^25,26. Our results suggest a mechanism in which substrate-guided oligomerization of TRIM37 dimers increases the avidity of binding to ectopic assemblies and activates ligase activity to promote their ubiquitination and clearance (Fig. 6D).

A mechanism for TRIM37-mediated selective suppression of centrobin oligomers

Formation of centrobin condensates in cells lacking TRIM37 occurs via the accumulation of material at centrosomes that eventually buds off to form an acentrosomal condensate; these condensates in turn mature at an appreciable frequency into ectopic spindle poles that elevate chromosome missegregation and likely contribute to the tumor-prone nature of the TRIM37 loss-of-function disorder Mulibrey nanism. Using the PLK4 inhibitor centrinone and an inducible TRIM37 knockout, we show that centrosomes are not required for condensate formation in TRIM37Δ cells; condensates can initiate and expand, likely via centrobin oligomerization, without the need for centrosomes as a nucleating structure. Thus, TRIM37 is able to patrol the cytosol to detect and prevent formation of ectopic centrobin-containing assemblies.

Using in vivo replacement of TRIM37 and centrobin in combination with in vitro biochemical analysis and structural modeling, we propose a mechanism that enables TRIM37 to detect and clear nascent centrobin oligomers without targeting centrobin monomers that are recruited from the cytoplasm to function at centrioles and enable ciliogenesis (Fig. 6D). The first element required for the role of TRIM37 in this clearance is its TRAF domain, which follows the coiled-coil and is unique in the extended TRIM family. Consistent with the function of other TRAF domains, which exhibit binding specificity for different peptide motifs ^15,16, our results implicate the TRIM37 TRAF domain in specific binding to peptide motifs in centrobin. Our results further show that disrupting either the peptide binding interface of the TRAF domain or the peptide motifs in centrobin prevents TRIM37 from detecting and clearing oligomeric centrobin condensates. Thus, analogous to the similarly positioned SPRY domain of TRIM5α, which binds the subunits of viral nucleocapsids in the cytosol ^36,37, the TRIM37 TRAF domain detects motifs in centrobin. Comparison of the in vivo localization of ligase-mutant TRIM37, which strongly concentrates on centrobin condensates, to that of ligase-and-TRAF-mutant TRIM37, which does not localize to centrobin condensates, provided strong support for the conclusion that the TRAF domain provides key specificity in the recognition of centrobin assemblies.

As TRAF domain-mediated motif recognition is low-affinity and would be unable to distinguish centrobin monomers from oligomers, additional elements of TRIM37 must also contribute to the specific recognition and clearance of centrobin oligomers. Based on the fact that TRIM family proteins are dimerized by an anti-parallel coiled-coil that places their two E3 ligase RING domains on opposite ends, and that many RING ligases of the TRIM family must dimerize to fully activate the ligase ¹³, we hypothesized that the oligomeric nature of the substrate facilitates the inter-dimer association of RING domains to promote ligase activation. In the case of TRIM5α, which employs such a mechanism ^20,27, the B-box-2 interface is critical for the inter-dimer association of RING domains and ligase activation ³⁰. By contrast, TRIM37 primarily employs a RING–RING interface for inter-dimer oligomerization, which represents the second critical element required for TRIM37 dimers to clear of ectopic centrobin assemblies. Notably, while the B-box-2 interface of TRIM37 is dispensable for inter-dimer oligomerization in vitro, its disruption resulted in a more compact conformation of the cross-dimer and prevented detection and clearance of centrobin condensates in vivo. Thus, the B-box-2 interface represents the third critical element required for TRIM37 to clear centrobin oligomers. We speculate that this requirement arises because TRIM37 dimers are present at relatively low concentrations in vivo, where the geometric properties that the B-box-2 interaction confers to the inter-dimer interface are required for efficient substrate-templated TRIM37 oligomerization (Fig. 6D). Thus, analogous to the way that trimerization of the TRIM5α B-box-2 is thought to match the lattice arrangement of nucleocapsid subunits to mediate viral recognition ^30,36, the TRIM37 B-box-2 may function to recognize the specific geometry of ectopic centrobin assemblies. Our data suggest that TRAF domain-mediated substrate recognition, RING dimerization, and B-box-2-mediated sensing of oligomer geometry collectively ensure that centrobin oligomers are selectively targeted for ubiquitination and degradation. Consistent with the importance of these TRIM37 domains in clearing centrobin condensates, we note that in addition to mutations that introduce early stop codons or delete the TRIM37 gene, Mulibrey nanism patients have been identified with mutations in the B-box-2 (C108S) and with a 17 amino acid deletion in the TRAF domain (OMIM; https://www.omim.org/; ³⁸).

In contrast to the requirements to clear centrobin oligomers, for TRIM37 to autoregulate its own levels, the most significant elements are the RING interface and ligase activity; the TRAF interface and the B-box-2 interface only make a modest contribution. Thus, at sufficient concentrations, the RING interface alone is capable of activating the ligase and triggering self-destruction. By destroying the TRIM37 that oligomerizes in the absence of substate templating, this mechanism may serve to tune the concentration of TRIM37, to ensure that only ectopic centrobin oligomers, and not cytoplasmic centrobin, or centriole-associated centrobin, become TRIM37 targets.

In addition to centrobin, the centrosomal protein CEP192 is a TRIM37 target and has received significant interest because its reduction following elevated TRIM37 expression due to genomic amplification in specific cancers leads to synthetic lethality with PLK4 inhibition ^10,11. TRIM37 also binds PLK4, can ubiquitinate it, and prevents formation of PLK4-dependent foci that accelerate spindle formation following centrosome removal ^8–10. Addressing how TRIM37 regulates these other targets will be influenced by the in-depth analysis of centrobin that we present here. We note that loss of TRIM37 also leads to the formation of centrin-containing foci whose composition and functional significance are unclear ^6,9. While the key scaffold of these centrin-containing foci has not yet been defined, their existence highlights that TRIM37 likely has additional centrosomal protein substrates.

Both prior work and the analysis presented here cement the idea that TRIM37 acts as a guardian of the centrosome. Given its mutation in the human tumor-prone developmental disorder Milibrey nanism, and its elevation leading to synthetic lethality with PLK4 inhibition in specific cancers, the work presented here will aid in understanding the full spectrum of TRIM37 roles in vivo. An important open question is whether the types of assemblies formed in the absence of TRIM37 are physiologically relevant in specific contexts where TRIM37 is downregulated, for example as part of a developmental program. Here, we note that the CNTROB gene was identified as a natural mutant in rats that impacts the formation of complex structures important for spermatogenesis ³⁹ , and centrobin was later shown to be important for ciliogenesis ^25,26,40,41. Selectively perturbing oligomerization of TRIM37 substrates such as centrobin, independently of their control by TRIM37, may enable testing whether the assemblies observed in the absence of TRIM37 are physiologically significant.

MATERIALS AND METHODS

Cell line construction and culture

Cell lines used in this study are listed in Table S1. RPE1 cells were grown in F12/DMEM with 10% FBS, 100 μg/ml streptomycin, and 100 μg/ml penicillin. Lenti-X 293T cells were grown in DMEM with 10% tet system approved FBS (Thermo Fisher), 100 μg/ml streptomycin, and 100 μg/ml penicillin. Freestyle 293F cells were grown in Freestyle 293 expression medium (Thermo Fisher). Cells were cultured at 37°C in 5% CO₂ except for Freestyle 293-F cells, which were cultured at 37°C in 8% CO₂.

The RPE1 inducible knockout of TRIM37 was generated by sequential lentiviral integration of inducible Cas9 (Edit-R Inducible Lentiviral Cas9; Dharmacon) and of a gRNA targeting exon 5 of TRIM37 (pOD4913; 5’-TCAGTGTGCACTTTGGGGAG-3’). Cas9 expression was induced by adding doxycycline to a final concentration of 1 μg/ml and the efficiency of indel generation at the TRIM37 locus was assessed using TIDE ⁴². To deplete centrosomes via PLK4 inhibition, centrinone was added at a final concentration of 100 or 150 nM.

RPE1 cell lines with CNTROB deleted were generated by treating the parent cell lines ODCL188 (TRIM37 WT; USP28Δ) and ODCL189 (TRIM37Δ), which contained inducible Cas9 and a gRNA targeting exon 5 of CNTROB (5’-GCTACAGCAACAATTAGCCG-3’), with doxycycline. Using limiting dilution the clonal cell lines ODCL0600 (CNTROBΔ; TRIM37 WT; USP28Δ) and ODCL0603 (CNTROBΔ; TRIM37Δ), which were confirmed by immunolabeling, immunoblotting, and genomic sequencing, were obtained. To generate the polyclonal inducible PLK4 KO cell line with p53 knocked down (ODCL420), a shRNA targeting p53 (pOD4894) was stably integrated using lentiviral delivery and the cell population was selected using the MDM2 inhibitor RG-7112 (Selleckchem).

Cell lines containing stably integrated FLAG-tagged TRIM37 transgenes (wild-type, C18R, W373A, Ring^Mut, B-Box-2^Mut, C18R-W373A, C18R-Ring^Mut, C18R-B-Box-2^Mut, and Ring^Mut-B-Box-2^Mut; see Table S1) were generated by lentiviral delivery into the parent cell line ODCL61 (TRIM37Δ). Cell lines containing stably integrated transgenes encoding mRuby-centrobin (wild-type, TBM^Mut, Motif-1^Mut; see Table S1) were generated by lentiviral delivery into the parent cell lines ODCL600 (CNTROBΔ; USP28Δ) and ODCL602 (CNTROBΔ; TRIM37Δ).

For all lentiviral integrations, viral particles were prepared by transfecting the lentiviral construct into Lenti-X 293T cells using Lenti-X^™ Packaging Single Shots (Takara Bio USA, 631276). 72 hours after transfection, the virus-containing culture supernatant was collected and added to the growth medium of the cells to be transfected in combination with addition of polybrene (EMD Millipore) to 8 μg/ml. Polyclonal cell lines were selected using appropriate antibiotics (neomycin, 400 μg ml⁻¹; puromycin, 10 μg ml⁻¹ for RPE1 cells).

Expression in Freestyle 293-F cells

For immunoprecipitation assays, plasmids containing centrobin constructs with a 5xMyc tag and TRIM37 constructs with a 3xFLAG tag were transfected into FreeStyle 293-F cells (Thermo Fisher) using FreeStyle MAX Reagent and OptiPRO SFM according to manufacturer guidelines (Thermo Fisher). Each transfected sample was a 20ml culture (10⁶ cells/ml). Equal amounts (12.5 μg) of each expression plasmid were used for co-transfections. Following transfection, FreeStyle 293-F cells were incubated for 48 hours on an orbital shaker platform (125 rpm) at 37°C in 8% CO₂. 10 ml of each sample were collected and washed with Dulbecco’s phosphate buffered saline (DPBS; Thermo Fisher). Cells were collected by pelleting and resuspended in 1 ml of lysis buffer (20 mM Tris/HCl, pH 7.5, 150 mM NaCl, 1% Triton X-100, 5 mM EGTA, 1 mM DTT, 2 mM MgCl and one EDTA-free protease inhibitor cocktail tablet (Roche)). Cells were sonicated in a water bath sonicator at 4°C for 6 minutes to generate a crude lysate. The crude lysate was centrifuged at 15,500 xg for 15 minutes at 4°C in a microfuge to generate supernatant and pellet fractions. Immunoprecipitations were performed by adding 20 μl of anti-Myc magnetic beads (Pierce; see Table S3) to 1 ml of supernatant. After incubating for 2 hours on a rotator at 4°C, beads were washed 5X in 900μl of lysis buffer and resuspended in 60 μl of 4x Laemmli sample buffer. For coimmunoprecipitation assays detecting ubiquitination, FreeStyle 293-F cells were transfected and incubated with equal amounts (8.5 μg) of three DNA constructs encoding: Myc-centrobin, FLAG-TRIM37, and haemagglutinin (HA)-tagged ubiquitin. For these IPs, 5 mM N-ethylmaleimide was added to the lysis buffer.

Immunofluorescence analysis

For immunofluorescence, 8,000 - 10,000 cells were seeded per well into 96-well plates one day prior to fixation. Cells were fixed in 100 μl of −20 °C methanol for 7 min. Cells were washed three times with wash buffer (phosphate-buffered saline (PBS) containing 0.1% Triton X-100) and blocked with blocking buffer (wash buffer containing 2% bovine serum albumin (BSA) and 0.1% sodium azide) overnight at 4°C or for 2 hours at RT. After blocking, cells were incubated for 2h with primary antibody (see Table S3) in blocking buffer at room temperature, followed by three washes with wash buffer. Cells were incubated for 1.5-2 h at RT with the secondary antibody, stained with Hoechst 33342 dye in blocking buffer, and washed three times with wash buffer before imaging.

A directly labeled mouse polyclonal anti-centrobin antibody (see Table S3) was generated by using a Mix-n-Stain^™ CF568 dye antibody labeling kit (Biotium). Prior to adding the directly-labeled antibody, cells were blocked with normal mouse serum (Jackson Immunoresearch; 1:20 in wash buffer) for 2h, after which cells were incubated at room temperature with the directly-labeled antibody at a final concentration of 0.5 μg/ml in blocking buffer.

Images were acquired with a CQ1 spinning disk confocal system (Yokogawa Electric) equipped with a 40X (numerical aperture (NA) 0.95) U-PlanApo objective and a 2,560 × 2,160 pixel sCMOS camera (Andor). Image acquisition and data analysis were performed using CQ1 software and ImageJ, respectively. Images were also acquired on a Nikon ECLIPSE Ti2 spinning disk confocal system equipped with 40X dry (NA 0.95) or a 60X oil (NA 1.42) objectives and a 1,024 × 1,024 pixel iXon 888 EMCCD camera (Andor). Image acquisition and data analysis were performed using Image J and Nikon NIS-Elements software.

Live-cell imaging

Samples were prepared 24 hours prior to imaging by seeding 4,000 cells per well into 96-well polystyrene plates. SiR-DNA was added for two hours prior to imaging at a final concentration of 0.5 μM per well. Live-cell imaging was performed with a CQ1 spinning-disk confocal microscope (Yokogawa Electric Corporation), using a 40X 0.95 NA U-Plan Apo objective at 37°C and 5% CO₂. For imaging, 5 x 2 μm Z sections were acquired in 5 fields every 6 minutes for 14 hours in the following channels: far red/SiR-DNA (20% laser power, 200 ms exposure), red/mRuby::centrobin (50% laser power, 200 ms exposure), green/CEP192-mNG (20% laser power, 200 ms exposure).

Quantification of centrobin and TRIM37::3xFLAG at condensates

Integrated fluorescence intensity was measured in a box fit around the condensate in the centrobin (red) channel; the same box was transferred to the FLAG (far red) channel. The local per-pixel background was measured in a 1-pixel wide box around the selection. The normalized ratio of FLAG-to-centrobin signal was calculated by dividing the background-subtracted FLAG signal by the background-subtracted centrobin signal and dividing this ratio by the median value of the same ratio for ligase-mutant TRIM37.

Design and construction of miniTRIM37 RING–B-box-2–hcc and B-box-2–hcc expression plasmids

miniTRIM37 RING–B-box-2–hcc and B-box-2–hcc constructs were designed using a strategy similar to that employed previously for TRIM5α ³⁰. Specifically, miniTRIM37 RING–B-box-2–hcc was designed by fusing residues 1-158 of human TRIM37 (containing the RING and B-box-2 domains, followed by a short segment of coiled-coil) to a serine tRNA synthetase hairpin sequence (resides 549-578, extracted from PDB 1D 1SER), followed by residues 214-254 of TRIM37; the deletion of residues 159-213 removes a majority of the antiparallel coiled coil. The annotated sequence of miniTRIM37 RING–B-box-2–hcc is shown in Extended Data Fig. 4C. DNA sequences encoding WT, RING-mutant (F13D, V64D, W68E, and L79D) and B-box-2 mutant (H115A and L119D) miniTRIM37 RING–B-box-2–hcc with extensions for ligation-independent V2 cloning (https://qb3.berkeley.edu/facility/qb3-macrolab/projects/lic-cloning-protocol/) were ordered as gBlocks from IDT technologies. For miniTRIM37 B-box-2–hcc constructs, PCR amplification using the oligos in parentheses (Forward 5’ TTTAAGAAGGAGATATAGATCATGGAGGAGAACGAGAAAGATAAGTGC 3’ and reverse: 5’ TTATGGAGTTGGGATCTTATTAGTGGTGATGGTGGTGGTGACC3’; underlined sequences are for ligation-independent cloning) was performed using the RING–B-box-2–hcc gBlocks as template to add a start codon and delete the first 87 amino acids. The annotated sequences of miniTRIM37 B-box-2–hcc is shown in Extended Data Fig. 4C. miniTRIM37 RING–B-box-2–hcc gBlocks and B-box-2–hcc PCR-amplified DNA sequences were cloned into the pLICTr-NTA vector utilizing ligation-independent cloning and verified using sequencing. The Alphafold dimer prediction of the engineered miniTRIM37 RING–B-box-2–hcc is shown in Fig. 5A.

Expression and Purification of miniTRIM37 RING–B-box-2–hcc and B-box-2–hcc Proteins

miniTRIM37 RING–B-box-2–hcc and B-box-2–hcc plasmids were transformed into Rosetta (DE3) E. coli cells employing a standard bacterial transformation protocol. For expression of each variant, 2 liters of 2XYT growth medium were inoculated with overnight starter culture grown at 37°C in carbenicillin and chloramphenicol. Cultures were incubated at 37°C until the OD₆₀₀ reached ~0.9-1.0. Cultures were cooled on ice for 10 minutes to lower the temperature and protein expression was induced by adding IPTG to 1 mM. Induced cultures were returned to the incubator at 20°C for 4 hours, harvested and lysed via sonication in resuspension buffer (20 mM Tris-HCl pH 7.5, 300 mM NaCl, 10 mM Imidazole, 2 mM β-mercapethanol, and 10% glycerol) supplemented with a protease inhibitor mix (PMSF, leupeptin, pepstatin, and aprotinin). Lysates were clarified by centrifugation at 35,000 xg at 4°C and supernatants loaded onto a 2 mL Ni-NTA column, washed with 10 column volumes of washing buffer (20 mM Tris-HCl pH 7.5, 300 mM NaCl, 20 mM Imidazole, and 10% glycerol), and eluted using 20 mM Tris-HCl pH 7.5, 150 mM NaCl, 400 mM Imidazole, and 10% glycerol. Eluates were concentrated at 4°C using Amicron Ultra spinning concentrators (10 kDa MW cutoff) and fractionated on a Superdex 200 gel filtration column in 50 mM Tris-HCl pH 8, 150 mM NaCl, 1 mM DTT, and 1 mM sodium azide. Protein purity was assessed via SDS-PAGE and fractions were concentrated to 1 mg/mL for SEC-MALS analysis.

SEC-MALS analysis of miniTRIM37 RING–B-box-2–hcc and B-box-2–hcc proteins

Prior to analysis of purified proteins, a Superdex 200 column coupled to a MALS machine (Wyatt Technology) was equilibrated overnight at room temperature in the buffer used for gel filtration. 120 μL of each purified protein (~1 mg/ml) was auto-loaded onto the Superdex 200 column. Elution and light scattering of miniTRIM37 RING–B-box-2–hcc and B-box-2–hcc proteins were monitored using Wyatt Technology HPLC software. After each run, baselines of UV absorbance and light scattering were established and protein peaks defined to obtain the predicted molecular weights. SEC-MALS data was exported and graphed using GraphPad Prism software.

Purification of recombinant TRIM37 TRAF domain

The TRIM37 TRAF domain coding sequence (residues 274-407) was amplified and cloned into the UC Berkeley Macrolab vector 2CT (Addgene number: 29706) in order to express N-terminal TEV protease-cleavable His₆-MBP-tagged fusions. The TRAF W373A point mutant (TRAF^mut) was generated using PCR-based site-directed mutagenesis and constructs were verified by sequencing.

The MBP-TRIM37 TRAF and MBP-TRIM37 TRAF^mut expression constructs were transformed, and proteins were expressed in E. coli Rosetta2 pLysS (EMD Millipore). The cells were grown to an OD₆₀₀ of 0.6-0.8, followed by induction by addition of IPTG to 0.33 mM. Protein expression was carried out at 20°C for 16-18 hours. Cells were harvested by centrifugation and resuspended in ice-cold resuspension buffer (50 mM Tris, pH 7.5, 300 mM NaCl, 10 mM imidazole, 10% glycerol, 2 mM β-mercaptoethanol). Resuspended cells were lysed using sonication, and the lysate was clarified by centrifugation. Proteins were purified by nickel affinity chromatography, and the eluted proteins were concentrated and further purified using size exclusion chromatography (Superdex 200 Increase 10/300 GL, Cytiva) in SEC buffer (20 mM Tris, 150 mM NaCl, and 1 mM DTT). Peak fractions were pooled and concentrated for use in binding assays.

Fluorescence polarization-based peptide motif binding analysis

FITC-Ahx-labeled centrobin peptides, whose sequences are shown in Fig. 3D, were synthesized (BioMatik) and resuspended in DMSO at a concentration of 5 mM. Peptides were diluted in binding buffer (20 mM Tris, pH 7.4, 150 mM NaCl, 1 mM DTT) to a final concentration of 100 nM. Each 20 μL reaction mixture contained 100 nM peptide and specified varying concentrations of TRIM37 TRAF or TRIM37 TRAF^mut in the binding buffer. The reaction mixtures were incubated at room temperature for 20 minutes, and fluorescence polarization was measured using a TECAN Infinite M1000 PRO fluorescence plate reader in 384-well plates. The measurements were performed in triplicate and the binding data were analyzed with GraphPad Prism v9 using a single-site binding model.

AlphaFold2 modeling

Protein structure and interaction predictions were performed using the AlphaFold2 ColabFold notebook (Jumper et al., 2021; Mirdita et al., 2022). Residues 1-451 from TRIM37 were used to predict the homodimeric model shown in Fig. 2A. To identify potential binding motifs in centrobin for the TRIM37 TRAF domain, we tested the disordered N-terminal region (residues 1-185) and the C-terminal region (residues 569-903) of centrobin, along with a single chain of TRIM37 (residues 274-451). While we did not observe confident predictions for interactions between the N-terminal region of centrobin and TRIM37, modeling of the C-terminal region provided strong interaction models. Manual inspection of the five resulting models revealed the four distinct interaction interface motifs on centrobin shown in Figure 3B and Extended Data Figures 3A,B. Structural analysis and visualizations were performed using PyMOL ⁴³ and ChimeraX ⁴⁴.

Extended Data

Extended Data Figure 1. Analysis of PLK4 localization to centrobin condensates and validation of the inducible TRIM37 knockout.

(A) Schematic of the PLK4 kinase highlighting its key domains and the location of the antigen used to generate ODAb207 in rabbits. Following serum recirculation on the antigen column, antibodies were eluted first using low pH (Acid elution) and second using high pH (Base elution). The pH of eluted fractions was rapidly neutralized and the Acid and Base elutions were separately dialyzed into a storage buffer. (B) Analysis of ODAb207 staining following inducible knockout of PLK4 in TRIM37Δ p53sh RPE1 cells. Knockout efficiency was confirmed by the absence of centrioles (not shown). Two examples are shown for the Acid and Base ODAb207 elutions. For each elution, the top example shows labeling (strong or weak) with ODAb207 of condensates marked by centrobin, while the bottom example shows no labeling. The Base but not the Acid elution of ODAb207 robustly labeled centrobin condensates. Scale bars, 5 μm (low magnification) and 1 μm (higher magnification). (C) Graphical summary of ODAb207 staining of centrobin condensates. Only condensates in acentriolar cells, indicative of PLK4 knockout, were analyzed. Condensate labeling was prominent with the Base elution of ODAb207, even 9 days after knockout induction; by contrast, little-to-no labeling was observed with the Acid elution of ODAb207. n is the number of cells analyzed. The analysis of the inducible PLK4 knockout and the predominant labeling of centrobin-containing condensates with the Base but not Acid elution of ODAb207 indicate that the observed staining does not represent PLK4 and instead represents cross-reactivity with a condensate component. (D) Sequencing traces of control (ODCL192) and iTRIM37 KO (ODCL623) cells following 3 days of doxycycline induction of Cas9. The sequence of the TRIM37 gRNA is indicated above the control sequencing trace. The frequency of indels was estimated using TIDE analysis (see Methods) and is derived by subtracting the percentage of 0 bp changes (10.4%) from 100%. (E) Immunoblot of the cell lines shown in (D) highlighting significant reduction in TRIM37 protein after knockout induction. α-tubulin serves as a loading control. (F) Quantification of centriole depletion following 5-day centrinone treatment of iTRIM37 KO cells. Centrione was used to first deplete centrioles prior to inducing TRIM37 knockout. Bar heights are the weighted mean and error bars are the 95%CI. Points represent biological replicates. n values in the bars are the total number of cells analyzed; p-value is from an unpaired, two-sided t-test; ****: p<0.0001.

Extended Data Figure 2. Error plots of Alphafold models of TRIM37 dimer and TRAF domain.

(A) Predicted Aligned Error (PAE) plot of the TRIM37 dimer model. (B) TRIM37 dimer model colored by confidence (pLDDTscore; left) and by chain (right). (C) TRIM37 TRAF domain model colored by confidence (pLDDT score).

Extended Data Figure 3. Error plots of TRAF–centrobin motif Alphafold models and validation of CNTROB knockouts.

(A) PAE plots of TRIM37 TRAF–centrobin motif models. The regions of TRIM37 and centrobin used to generate the models are noted above. (B) TRAF-centrobin motif models (same as in Fig. 3B) with binding affinities of FITC-coupled motif peptides estimated from the curves shown in Fig. 3D. (C) Sequence trace of a control cell line highlighting the location of the CNTROB gRNA. (D) Sequencing traces of CNTROBΔ cell lines with TRIM37 WT or TRIM37Δ. For each cell line, the top trace is with a forward sequencing primer and the bottom trace with a reverse sequencing primer (presented as the complementary strand). The changes in each allele are annotated on the traces. (E) Immunoblot verifying loss of centrobin protein in CNTROBΔ cells, with or without TRIM37 present. α-tubulin serves as a loading control. This experiment was performed twice with similar results.

Extended Data Figure 4. Analysis of Motif 1-mutant centrobin and of miniTRIM37s.

(A) Representative images of CNTROBΔ cells, either without (left) or with (right) TRIM37 present, in which either mRuby-tagged WT (top) or Motif 1-mutant (bottom) centrobin was expressed from a lentivirally-delivered transgene. For TRIM37 WT cells expressing the Motif 1-mutant, example images are shown of cells with and without a condensate. Scale bars, 5 μm (low magnification) and 1 μm (high magnification). (B) Graph plotting the percentage of cells with centrobin condensates for the indicated conditions. p values are from unpaired, two-way t-tests (ns=not significant, p=0.712; ** indicates p<0.0066). Bar heights represent the weighted mean. Points represent individual biological replicates. Error bars are the 95% confidence interval. n values in the bars are the total number of cells analyzed. (C) (top) Protein sequences of miniTRIM37 RING–B-box-2–hcc and B-box-2–hcc. (bottom) Coomassie-stained gels of purified recombinant miniTRIM37 RING–B-box-2–hcc and B-box-2–hcc proteins (schematized above the gels; the same schematics are shown in Fig. 4D). The RING mutant migrates higher than the wild-type protein, likely due to the negative charges introduced by the mutations made to disrupt the RING–RING interface. The identity of the RING mutant band was confirmed using mass spectrometry. (D) (top) miniTRIM37 RING–B-box-2–hcc dimer model, also shown in Fig. 5A, and (bottom) PAE plot of the miniTRIM37RING–B-box-2–hcc dimer model. The color-coded two chains used to construct the model are shown on the left and bottom sides of the PAE plot. (E) SEC-MALS analysis of miniTRIM37 B-box-2–hcc WT and mutant purified proteins. Both proteins elute at the same time from the column and have the same native molecular weight, corresponding to the predicted monomer molecular weight.

Extended Data Figure 5. Analysis of TRIM37 single mutant’s localization to centrobin condensates.

(A) Graph quantifying the levels of TRIM37 variants and centrobin relative to WT in crude extract. This experiment was repeated two times, values for both experiments are shown.The blot for one of the two experiments is shown in Fig. 5H. Graph in the box to the right has been rescaled to facilitate comparison of the levels for WT and B-box-2mut TRIM37. (B) (left) Schematics highlighting TRIM37 variants expressed from transgenes in TRIM37Δ cells; (right) representative images of cells expressing the TRIM37 variants that were fixed and labeled for centrobin and for the FLAG epitope fused to the transgene-encoded TRIM37 variants. Scale bars, 5 μm (low magnification) and 1 μm (high magnification). (C) Graph plotting the ratio of FLAG to centrobin signal at condensates in cells expressing the indicated TRIM37 variants. Plotted values were normalized by dividing by the median value in theTRIM37 ligase-mutant. The mean and standard deviation are plotted over individual values. n values for each condition were: Ligmut, n=50; TRAFmut, n=50; RINGmut, n=50; B-box-2mut, n=50. The data shown for the ligase-mutant is the same as in Fig. 6B. p-values are from unpaired, two-sided t-tests; ****:p<0.0001.

Data Availability Statement:

All data supporting the findings of this study, and source data for all graphs and blots, are available within the article and its supplementary information files. Additional raw data generated during the current study are available from the corresponding author upon reasonable request.