Redirecting the specificity of tripartite motif containing-21 scaffolds using a novel discovery and design approach

Abstract

Hijacking the ubiquitin proteasome system to elicit targeted protein degradation (TPD) has emerged as a promising therapeutic strategy to target and destroy intracellular proteins at the post-translational level. Small molecule–based TPD approaches, such as proteolysis-targeting chimeras (PROTACs) and molecular glues, have shown potential, with several agents currently in clinical trials. Biological PROTACs (bioPROTACs), which are engineered fusion proteins comprised of a target-binding domain and an E3 ubiquitin ligase, have emerged as a complementary approach for TPD. Here, we describe a new method for the evolution and design of bioPROTACs. Specifically, engineered binding scaffolds based on the third fibronectin type III domain of human tenascin-C (Tn3) were installed into the E3 ligase tripartite motif containing-21 (TRIM21) to redirect its degradation specificity. This was achieved via selection of naïve yeast-displayed Tn3 libraries against two different oncogenic proteins associated with B-cell lymphomas, mucosa-associated lymphoid tissue lymphoma translocation protein 1 (MALT1) and embryonic ectoderm development protein (EED), and replacing the native substrate-binding domain of TRIM21 with our evolved Tn3 domains. The resulting TRIM21–Tn3 fusion proteins retained the binding properties of the Tn3 as well as the E3 ligase activity of TRIM21. Moreover, we demonstrated that TRIM21–Tn3 fusion proteins efficiently degraded their respective target proteins through the ubiquitin proteasome system in cellular models. We explored the effects of binding domain avidity and E3 ligase utilization to gain insight into the requirements for effective bioPROTAC design. Overall, this study presents a versatile engineering approach that could be used to design and engineer TRIM21-based bioPROTACs against therapeutic targets.

Targeted protein degradation (TPD) has become a powerful proteome editing tool in which a protein of interest (POI) can be specifically targeted and marked for degradation by natural protein degradation mechanisms. TPD offers several advantages over other protein blocking strategies, including complete ablation of a protein’s functions, sustained inhibition (barring protein resynthesis), and the ability to target previously undruggable proteins (1, 2). In addition, unlike approaches that function at the level of DNA or RNA, such as antisense oligonucleotides and RNA interference, TPD allows for protein modification at the post-translational level and thus has the potential to dissect protein function and discriminate between protein forms at a higher resolution (3, 4).

Although recent approaches such as lysosome-targeting chimera (5, 6), antibody-based proteolysis-targeting chimeras (PROTACs) (7), autophagy-targeting chimera (8, 9), autophagosome tethering compound (10, 11), and others (2, 12, 13, 14, 15, 16, 17, 18, 19) have co-opted the lysosomal degradation pathway for POI degradation, the majority of TPD approaches to date have focused on utilizing the ubiquitin proteasome system (UPS) machinery. The UPS functions through the concerted actions of the three enzymes, the ubiquitin-activating enzyme (E1), ubiquitin-conjugating enzyme (E2), and ubiquitin ligase (E3), which tag accessible lysines on the substrate with polyubiquitin chains. This polyubiquitination serves as the canonical recognition signal for the 26S proteasome and thus marks the substrate for degradation. Since substrate specificity is conferred by the E3 ligase, this class of enzymes has been the primary focus for TPD approaches via the UPS (3, 4, 20).

One of the most prominent TPD approaches utilizing the UPS has been the development of PROTACs, which are heterobifunctional small molecules comprised of a POI-targeted moiety linked to an E3 ligase–targeted moiety, connected by a flexible linker (21, 22, 23). This approach brings together the POI and E3 ligase, leading to subsequent polyubiquitination and degradation of the POI. PROTACs have shown great promise in preclinical studies, and several candidates are being evaluated in clinical trials for treatment of both solid tumors and blood cancers (24, 25). Other PROTAC and small-molecule approaches that hijack the UPS for TPD have also been reported (16, 26, 27), including molecular glues (28, 29, 30), degradation tags (31), and Specific Nongenetic Inhibitors of apoptosis protein-dependent Protein ERasers (32). However, these approaches can be limited by poor oral bioavailability and off-target toxicities resulting from nonspecific small-molecule interactions. Moreover, these approaches have been limited to targeting POIs with known small-molecule ligands, and screening and synthesis processes are often complex and expensive (33).

Biological PROTACs (bioPROTACs), which are engineered fusion proteins comprised of a POI-binding domain and an E3 ligase (or a component of an E3 ligase complex), have emerged as a complementary approach to small-molecule PROTAC approaches (2). Although early designs focused on fusing a POI’s known binding partner to an E3 ligase (34, 35), recent approaches have involved replacing the substrate-binding domain of the E3 ligase with a synthetic binding domain such as a fibronectin type III (FN3), designed ankyrin repeat protein (DARPin), nanobody, alpha repeat protein (αRep), or single-chain antibody fragment (scFv) (1, 2, 3). bioPROTACs offer several advantages compared with small molecule–based TPD approaches because of their ability to target previously undruggable POIs and their precise recognition of specific protein isoforms or mutants as well as proteins with post-translational modifications. Moreover, bioPROTACs allow for the utilization of a much broader range of E3 ligases compared with PROTACs and other small molecule–based approaches that require an available ligand against the E3 ligase (1, 2). bioPROTACs have been employed for degradation of a range of oncoproteins, including KRAS (36, 37, 38, 39) and extracellular signal–regulated kinase (40). These molecules have demonstrated robust degradation both in vitro and in vivo, leading to robust antitumor activity in cancer models (36, 37, 38, 39, 40). bioPROTACs have also proven to be highly modular, as successful approaches have incorporated a diverse set of E3 ligases, including von Hippel-Lindau (VHL), carboxy terminus of Hsc70 interacting protein (CHIP), and speckle-type POZ protein (SPOP) (1, 3, 4, 36). However, despite this broadened usage of E3 ligases for TPD, only a small fraction of E3 ligases has been explored to date, mainly those from the Cullin-RING E3 ubiquitin ligase family.

Another family of E3 ligases that can be leveraged for TPD is the tripartite motif (TRIM) family of proteins, and we specifically focused our attention on the E3 ligase tripartite motif containing-21 (TRIM21). TRIM21 is a cytosolic ubiquitin ligase and Fc receptor that primarily serves to detect and degrade antibody-coated viruses that have evaded extracellular neutralization and penetrated the cell membrane. Like many of the other TRIM family members, TRIM21 is comprised of N-terminal RING, B-box, and coiled-coil (collectively denoted RBCC) domains and a C-terminal PRYSPRY domain. Both the RING and B-box domains are involved in the E3 ligase activity of TRIM21, whereas the coiled-coil domain induces TRIM21 homodimerization. The PRYSPRY domain of TRIM21 mediates high affinity binding of antibody Fc fragments, and it has been shown that TRIM21 engages the Fc as a dimer with subnanomolar affinity, as measured by fluorescence anisotropy (41, 42). Thus, the combined functions of the RBCC and PRYSPRY domains of TRIM21 orchestrate proteasomal degradation of antibody-coated viruses and initiate downstream signaling pathways (41, 42, 43, 44, 45, 46, 47, 48). Although it has been reported that TRIM21-mediated degradation occurs through the autoubiquitination of TRIM21 and subsequent degradation of both TRIM21 and the target (42, 46, 49, 50, 51, 52, 53), it has recently been demonstrated that TRIM21 also catalyzes polyubiquitination of the target protein, and that this target ubiquitination is required for degradation (54, 55).

The TRIM21 protein has been harnessed for TPD through systems such as TRIM-Away, which coordinates protein degradation through an intermediary antibody against a POI (56, 57, 58), TRIMbody-Away, optoTRIM-Away, and miniTRIM-Away, which fused either full-length TRIM21 or a truncated TRIM21 to a nanobody against a GFP (49, 59). Here, we sought to redirect the TRIM21 substrate specificity to orchestrate degradation of disease-associated intracellular POIs by replacing the PRYSPRY-binding domain with a versatile binding protein scaffold based upon the third FN3 domain of human tenascin-C, termed Tn3 (60, 61, 62, 63). This workflow was enabled by development of high-throughput fluorescence-based screening assays to evaluate bioPROTAC degradation activity. We validated our approach by developing Tn3 domains against two oncogenic targets, MALT1 (mucosa-associated lymphoid tissue lymphoma translocation protein 1) and EED (embryonic ectoderm development protein). Constitutive activation of MALT1, which drives chronic NF-κB signaling, has been associated with activated B-cell-like diffuse large B-cell lymphoma, MALT lymphoma, and solid tumors such as pancreatic and prostate cancers (64, 65, 66). Given its promise as a therapeutic target, there are ongoing efforts to inhibit MALT1, including use of PROTAC strategies (67, 68, 69, 70, 71, 72, 73). EED is a core component of the polycomb repressive complex 2 (PRC2), which regulates gene expression through methylation of histone 3 at lysine 27. Activating mutations and overexpression of PRC2 components have been linked to germinal center B-cell-like diffuse large cell B-cell lymphoma and follicular lymphoma as well as breast, colorectal, and prostate cancers (74, 75, 76, 77). Recent efforts in targeting PRC2 to treat cancer include both inhibitors and PROTACs against the complex, and these agents have demonstrated preclinical and clinical success (77, 78, 79, 80, 81, 82, 83).

This study discovered Tn3-binding domains against MALT1 and EED via yeast surface display–mediated directed evolution (84) and fused these Tn3 domains to TRIM21 in place of the PRYSPRY domain, forming TRIM21–Tn3 fusion proteins (Fig. 1, A and B). We demonstrated that these fusion proteins engaged their respective POIs (MALT1 and EED) and retained ubiquitin ligase activity. TRIM21–Tn3 fusion proteins led to up to 75% degradation of the targeted POI when overexpressed, and investigation of the design landscape offered insight into the molecular requirements for TRIM21-mediated TPD. We also developed high-throughput fluorescence-based assays to quantify TPD that can be generally used to screen degradation activity for a variety of bioPROTAC designs. Finally, we showed that MALT1-specific TRIM21–Tn3 fusion proteins induce degradation of endogenous MALT1, motivating further work that will explore the translational potential of TRIM21-based bioPROTACs. The design and engineering approaches we present, as well as the characterization workflow we have established, will be valuable tools in the future development of TRIM21 bioPROTACs and more generally for biologics-based degraders.

Figure 1.

Design and characterization of TRIM21–Tn3 fusion proteins.A, linear representation of the natural TRIM21 protein (top) and an engineered TRIM21–Tn3 fusion protein (bottom). To construct TRIM21–Tn3 fusion proteins, Tn3 domains were fused to the C terminus of TRIM21ΔPRYSPRY (residues 1–286) via a flexible 15-amino acid (Gly₄Ser)₃ linker. B, schematic depicting targeted protein degradation (TPD) of a protein of interest (POI) using an engineered dimeric TRIM21–Tn3. C, equilibrium biolayer interferometry (BLI)–based titrations and corresponding equilibrium K_D values of TRIM21–HR4, HR4–Fc, and TRIM21 binding to the immobilized extracellular domain of HER4 (residues 26–649). D, flow cytometry–based binding titrations of TRIM21–HR4, HR4–Fc, and TRIM21 to HER4-expressing AD-293 cells and corresponding equilibrium K_D values. Error bars represent mean ± SD (n = 3). HER4, human epidermal growth factor receptor 4; TRIM21, tripartite motif containing-21.

Results

Design of TRIM21–Tn3 fusion proteins

We sought to engineer TRIM21 to target and degrade specific disease-associated intracellular POIs by leveraging TRIM21’s endogenous E3 ligase activity. To engineer TRIM21 to bind a desired POI, we replaced the native PRYSPRY-binding domain, which engages the Fc fragment of antibodies, with an engineered Tn3 (61, 62, 63) (Figs. 1A and B and S1 and Table S1). Fibronectin domains, specifically the 10th FN3 domain of human fibronectin, have been successfully employed as alternative binding scaffolds because they contain three surface-exposed loops at one end of the molecule that resemble antibody complementarity-determining regions. These loops can be engineered as versatile binding agents, and because of their small size, a large percentage of their protein surface can be modified (85, 86, 87, 88). In addition, the lack of disulfide bonds, including in the CM4 Tn3 mutant (63) that we engineered in this study, improves stability in cytosolic environments (50, 62, 63, 89). We hypothesized that by engineering Tn3 domains as novel binding scaffolds against a POI and using them to replace the TRIM21 PRYSPRY domain, we could elicit degradation of the targeted POI by capitalizing on TRIM21’s E3 ligase activity to mediate proteolysis through the UPS (Fig. 1B).

As proof of concept, we first sought to establish that TRIM21–Tn3 fusion proteins retained the binding properties of the constituent Tn3 domain as well as the E3 ligase properties of TRIM21. Therefore, we replaced the PRYSPRY domain of TRIM21 with a human epidermal growth factor receptor 4 (HER4)–specific Tn3 domain (HR4), which had previously been isolated from yeast-displayed Tn3 libraries. The Tn3 domain was fused to the C terminus of the TRIM21 protein lacking the PRYSPRY domain (termed TRIM21ΔPRYSPRY, residues 1–286) using a flexible (Gly₄Ser)₃ linker (Figs. 1A and S1). The resulting TRIM21–Tn3 fusion protein was expressed recombinantly through transient transfection of human embryonic kidney (HEK) 293F cells and purified from cell lysates via nickel–nitrilotriacetic acid (Ni–NTA) affinity chromatography followed by size-exclusion chromatography (SEC).

We tested the binding of our TRIM21–HR4 protein to HER4 via biolayer interferometry (BLI) (Fig. 1C) and also via binding studies on engineered HER4-expressing AD-293 cells (Fig. 1D). BLI studies revealed that TRIM21–HR4 bound to HER4 with an equilibrium dissociation constant (K_D) of 46 nM, which was approximately 10-fold weaker (because of both acceleration of k_off and deceleration of k_on) than that of a dimeric Fc-fused HR4 Tn3 construct (HR4–Fc), which exhibited a K_D of 4.4 nM (Fig. 1C and Table S2). This could be due to the difference in orientation of the Tn3 domain, which is located at the C terminus in TRIM21–HR4, whereas it is located at the N terminus in HR4–Fc. However, in a more physiologically relevant HER4-expressing cell binding assay, there was no measurable difference in K_D between TRIM21–HR4 and HR4–Fc, with both having a K_D of approximately 5 nM (Fig. 1D). As expected, there was little to no binding to HER4 above background for wildtype TRIM21, as measured by both BLI and cell binding.

To confirm that TRIM21–Tn3 fusion proteins could still be properly ubiquitinated (and thus susceptible to UPS degradation), we performed an in vitro ubiquitination assay using purified UPS components (E1, E2, ubiquitin, and ATP) alongside either unmodified TRIM21 or TRIM21–HR4 as the E3 ligase. Since TRIM21 is known to self-ubiquitinate in the context of in vitro ubiquitination assays performed in the absence of substrate (42, 46, 49, 50, 51), no substrate was added to the reaction. We chose to simultaneously add two E2 enzymes, Ube2W and Ube2N/V2, because it has been reported that TRIM21 first recruits Ube2W to modify itself with an N-terminal monoubiquitin, and then heterodimeric Ube2N/V2 catalyzes the formation of an anchored K63-linked polyubiquitin chain (4, 5, 6, 7); thus both E2 enzymes are crucial for polyubiquitination of TRIM21-containing proteins. Ubiquitination was evaluated by SDS-PAGE analysis (Fig. S2A) and by immunoblotting with an antiubiquitin antibody (Fig. S2B). In the presence of E1, E2, ubiquitin, ATP, and either TRIM21 or TRIM21-HR4, higher molecular weight species were observed by both SDS-PAGE and immunoblotting, indicative of proper ubiquitination, indicates that TRIM21–Tn3 fusion proteins retain the native ubiquitination activity of TRIM21. Moreover, as expected, ubiquitination was ATP dependent, as only the unanchored ubiquitin band was present, and no higher molecular weight species were observed in the absence of ATP (Fig. S2B). We noted that in the case of TRIM21-HR4, there was a larger presence of unanchored ubiquitin chains (that present as lower molecular weight species compared with TRIM21-HR4), which may reflect impaired autoubiquitination of TRIM21 in the context of this TRIM21–Tn3 fusion protein.

Isolation of Tn3 domains against MALT1 and EED from yeast-displayed Tn3 libraries

To engineer Tn3 domains that engage the cancer-associated targets MALT1 and EED, we designed and screened Tn3 libraries based on a CM4 Tn3 mutant that was engineered for enhanced thermal stability (63) (Fig. S3, A and B). Building upon FN3 domain engineering work pioneered by Koide et al. (85, 86, 87, 88), including the design of loop libraries as well as libraries that diversify alternative surfaces on the FN3 scaffold, we developed seven distinct yeast surface–displayed Tn3 libraries. This included four loop libraries, in which sequence diversity was introduced into the BC, DE, and FG loops, and loop length diversity was also introduced into the BC and FG loops; two concave libraries, in which sequence diversity was introduced into the CD and FG loops as well as several residues within a concave beta-sheet region on one side of the Tn3 scaffold, and loop length was diversified within the CD loop; and a flat library, in which sequence diversity was introduced into a flat beta-sheet region opposite the concave beta sheets (Fig. S3B).

To isolate Tn3 domains that specifically bind MALT1 or EED, the seven yeast-displayed Tn3 libraries were separately selected against recombinant human MALT1 and EED. We conducted four rounds of selection: the first two rounds implemented selections via magnetic-activated cell sorting (MACS); and the third and fourth rounds were carried out via fluorescence-activated cell sorting (FACS). Libraries were pooled into two groups to enable high-throughput screening with group 1 representing the four loop libraries and group 2 representing the two concave libraries and the flat library. Following selections, the naïve libraries (R0) as well as the libraries from each round (postsorting) were stained with 50 nM tetrameric target antigen (Fig. 2A), revealing 14% and 44% library group 1 and group 2 binding, respectively, for MALT1 and 44% and 29% library group 1 and group 2 binding, respectively, for EED. The fact that MALT1 showed stronger enrichment for library group 2 binding, whereas EED showed stronger enrichment for library group 1 binding (Fig. 2A), indicated that the optimal Tn3 library design strategy may vary between protein targets.

Figure 2.

Discovery of Tn3 binders against human MALT1 and EED.A, progression of MALT1 (left) or EED (right) (50 nM tetramer) binding to two groups of yeast-displayed Tn3 libraries over four rounds of selection. Library group 1 consisted of four pooled loop libraries, and library group 2 consisted of two concave and one flat library (Fig. S3B). The percentage on the right-hand side of each plot represents the percentage of total yeast binding to the target protein. B and C, on-yeast MALT1 (B) or EED (C) binding titrations of individual clones isolated from Tn3 library selections, as measured by flow cytometry analysis. Signal was normalized to expression (detected with a V5 tag) for each Tn3 variant. EED, embryonic ectoderm development protein; MALT1, mucosa-associated lymphoid tissue lymphoma translocation protein 1; Tn3, third fibronectin type III domain of human tenascin-C.

Individual MALT1- and EED-binding clones from each library group were isolated, and the lead binders were sequenced. For MALT1, nine specific clones were isolated, denoted MA8, MA11, MC1, MC5, MC8, MC10, MC11, ME2, and MH4 (Table S1). Clones ME2 and MH4 were isolated from library group 2, specifically the flat library, whereas all other clones originated from library group 1 (Fig. S3B). Interestingly, although stronger enrichment for MALT1 binding was observed for library group 2 (Fig. 2A), there was a much higher percentage of streptavidin binders in group 2; thus, the majority of specific MALT1-targeted clones originated from library group 1. For EED, six specific clones were isolated: EA7, ED12, EG9, EH3, EH5, and EH8 (Table S1). Clones EA7 and ED12 were isolated from library group 1, whereas all other clones were isolated from library group 2, specifically the flat library (Fig. S3B). For EED, more streptavidin-binding clones were isolated from library group 1; thus, more EED-binding clones were isolated from library group 2. Interestingly, in the case of both MALT1 and EED, no clones were isolated from the concave libraries.

To further characterize the individual binding clones, we performed yeast surface titrations of recombinant MALT1 or EED against the displayed Tn3 clones (Figs. 2, B and C and S3, C and D). We validated that all nine MALT1-targeted clones showed specific interaction with the target POI (Fig. S3C). We then titrated MALT1 against a representative pool of MALT1-binding clones and found that MC1 and MC11 had stronger binding toward MALT1 (K_D of approximately 60 nM) compared with MA8, ME2, and MH4 (K_D values of 308 nM, 168 nM, and 264 nM, respectively), although the maximum response (E_Max) for MC1 and MC11 was lower (Fig. 2B and Table S3). These five clones were therefore selected to engineer as TRIM21–Tn3 fusion proteins to induce targeted degradation of MALT1. After validating specific binding of the six EED-targeted clones (Fig. S3D), we titrated EED against the highest affinity clones: EA7 (K_D = 295 nM), EG9 (K_D = 221 nM), and EH8 (K_D = 178 nM); exhibited the strongest affinities (Fig. 2C and Table S3). These three clones, as well as clone ED12, were chosen to be reformatted as TRIM21–Tn3 fusion proteins.

Development of TRIM21–Tn3 fusion proteins that bind MALT1 and EED

The lead MALT1- and EED-binding Tn3 clones were then formatted as TRIM21–Tn3 fusion proteins to serve as targeted degraders for their respective disease-linked POIs. Each Tn3 domain was fused to the C terminus of TRIM21ΔPRYSPRY (residues 1–286) via a flexible (Gly₄Ser)₃ linker (Figs. 1A and S1 and Table S1). A C-terminal FLAG tag was also included in each of the constructs for immunoblotting and immunoprecipitation purposes. TRIM21–Tn3 variants were expressed recombinantly through transient transfection of HEK 293F cells and purified from cell lysates via affinity chromatography followed by SEC.

To ensure proper ubiquitination of MALT1- and EED-targeted TRIM21–Tn3 fusion proteins, in vitro ubiquitination reactions were performed for both TRIM21–ME2 and TRIM21–EG9 (Fig. S2, C–F). Higher molecular weight species were observed for both TRIM21–ME2 and TRIM21–EG9 in the presence of E1, E2, ubiquitin, and ATP, indicative of proper ubiquitination. Interestingly, in contrast with TRIM21–HR4, increased autoubiquitination was observed for the TRIM21–ME2 and TRIM21–EG9 fusion proteins compared with the wildtype TRIM21, indicating that distinct Tn3 domains may have differential effects on TRIM21 ubiquitination activity.

To confirm that TRIM21–Tn3 fusion proteins still bound to either MALT1 or EED, binding of recombinant TRIM21–Tn3 variants to MALT1 or EED was evaluated by BLI (Fig. 3, A and B). TRIM21–MC11 showed little to no MALT1 binding above the EED-targeted control (TRIM21–EG9), whereas TRIM21–ME2 and TRIM21–MH4 bound with K_D values of 753 nM and 148 nM, respectively (Fig. 3A and Table S2). The predicted K_D for TRIM21–MC1 was 475 nM, although the E_Max was lower compared with the other clones (Table S2). Because of the fact that recombinant TRIM21–MA8 could not be produced in sufficient amounts, binding to MALT1 was not measured by BLI. With respect to EED clones, TRIM21–EA7 and TRIM21–ED12 exhibited little to no EED binding above the MALT1-targeted control (TRIM21–ME2), whereas TRIM21–EG9 and TRIM21–EH8 bound EED with K_D values of 70 nM and 37 nM, respectively (Fig. 3B and Table S2).

Figure 3.

TRIM21–Tn3fusion proteins bind to target POIs.A and B, equilibrium biolayer interferometry–based titrations and corresponding equilibrium K_D values of MALT1 (A) or EED (B) binding to immobilized TRIM21–Tn3 variants. C and D, immunoprecipitation studies conducted on HBL-1 (C) or Karpas 422 (D) cell lysates. Immunoprecipitation was performed by subjecting lysate to anti-FLAG beads that had been preincubated with the indicated recombinant FLAG-tagged TRIM21–Tn3 variant. An equivalent amount of total protein was added for each condition, and immunoblots are representative of two replicate experiments. Multiple bands are observed for EED (D) because of the presence of different EED isoforms (92). EED, embryonic ectoderm development protein; MALT1, mucosa-associated lymphoid tissue lymphoma translocation protein 1; POI, protein of interest; TRIM21, tripartite motif containing-21; Tn3, third fibronectin type III domain of human tenascin-C.

Importantly, only Tn3 clones isolated from the flat library (ME2, MH4, EG9, and EH8) showed significant binding to their target POI when formatted as TRIM21–Tn3 fusion proteins. To explore this further, we generated models of the composite structures of TRIM21–Tn3 fusion proteins, illustrating the mutated amino acid positions for each library design (Fig. S4). It was observed that the engineered regions in the loop libraries were located near the N terminus of the Tn3 domain (where it is fused to TRIM21ΔPRYSPRY), suggesting that steric hindrance may prevent binding of target to Tn3 domains isolated from the loop libraries when expressed recombinantly in the TRIM21–Tn3 format (Fig. S4A). This was not observed in the yeast display format, as in this system, the Tn3 domain was tethered to Aga2p at its C terminus, leaving the N terminus free (Fig. S3A). By contrast, for TRIM21–Tn3 fusion proteins created with Tn3 domains from the concave or flat libraries, the majority of the mutated residues are accessible in the TRIM21–Tn3 format (Fig. S4, B and C), although no binders were isolated from the concave libraries. These findings underscore the importance of construct topology and library design in development of TRIM21–Tn3 fusion proteins.

TRIM21–Tn3 fusion proteins bind to native MALT1 and EED

Since Tn3 libraries were selected against recombinant MALT1 or EED, we also wanted to evaluate whether TRIM21–Tn3 fusion proteins would bind to natively expressed target. To confirm binding to natural MALT1 and EED, we performed immunoprecipitation assays on lysates from HBL-1 and Karpas 422 cells, respectively. HBL-1 is a human-activated B-cell-like diffuse large B-cell lymphoma cell line, which is known to natively express MALT1 (90). Karpas 422 is a human B-cell non-Hodgkin lymphoma cell line that is known to express EED as well as a mutated version (Y641N) of EZH2, another component of the PRC2 (91). FLAG-tagged TRIM21–Tn3 variants, TRIM21–MC1, TRIM21–ME2, and TRIM21–MH4, showed robust association with MALT1, whereas TRIM21–MC11 and the EED-targeted control TRIM21–EG9 did not (Fig. 3C), consistent with BLI observations (Fig. 3A). Pulldown with TRIM21–MA8 was not performed because of insufficient yield from the recombinant production of TRIM21–MA8. In similar fashion, TRIM21–EG9 and TRIM21–EH8 showed robust association with EED, whereas TRIM21–EA7 showed weak association for EED and TRIM21–ED12 and the MALT1-targeted control TRIM21–ME2 showed no association with EED (Fig. 3D). These results were again consistent with BLI observations (Fig. 3B). Multiple bands were observed in the EED blot because of the presence of different EED isoforms (92), and therefore, the isolated binders were determined not to be isoform specific. The results from both MALT1 and EED immunoprecipitation studies confirm that recombinant TRIM21–Tn3 bind the native forms of their target POIs. Collectively, our ubiquitination and binding studies demonstrated the functionality of TRIM21–Tn3 fusion proteins as a potential new platform for TPD.

TRIM21–Tn3 fusion proteins degrade TurboGFP-tagged MALT1 and EED

To demonstrate that TRIM21–Tn3 fusion proteins can degrade their target proteins in vitro, we cotransfected HEK 293F cells with a plasmid encoding TurboGFP (tGFP)-tagged POI and a plasmid encoding the TRIM21–Tn3 fusion. Degradation was quantified by flow cytometry 72 h after transfection (Fig. 4A). tGFP fluorescence signal was calculated based on the full population of live cells (including both transfected and nontransfected cells). As illustrated in Figure 4B, TRIM21–ME2 and TRIM21–MH4 led to significant degradation of MALT1-tGFP as compared with the control anti-EED TRIM21–EG9, with an average threefold decrease and twofold decrease in tGFP signal for TRIM21–ME2 and TRIM21–MH4, respectively. TRIM21–MC1 and TRIM21–MC11 led to an average 1.3-fold decrease in tGFP signal but did not reach significance. TRIM21–MA8 did not induce MALT1-tGFP degradation compared with the control TRIM21–EG9. Similarly, TRIM21–EA7, TRIM21–ED12, TRIM21–EG9, and TRIM21–EH8 significantly reduced levels of EED-tGFP as compared with the control anti-MALT1 TRIM21–ME2 (Fig. 4C). TRIM21–EG9 demonstrated the most efficient degradation with an average sixfold decrease in tGFP signal, both TRIM21–ED12 and TRIM21–EH8 led to an average 2.5-fold decrease in tGFP signal, and TRIM21–EA7 induced an average 1.6-fold decrease in tGFP signal. Interestingly, although there was no observable binding of TRIM21–ED12 to EED, as measured by BLI and immunoprecipitation (Fig. 3, B and D), degradation effects were still observed.

Figure 4.

TRIM21–Tn3 fusion proteins degrade tGFP-tagged POIs.A, schematic showing engagement and degradation of a tGFP-tagged POI by a TRIM21–Tn3 fusion protein. Plasmids encoding the tGFP-tagged POI and the TRIM21–Tn3 variant were cotransfected into HEK 293F cells, and degradation was evaluated after 72 h by flow cytometry. B and C, flow cytometry analysis of HEK 293F cells cotransfected with a plasmid encoding MALT1-tGFP (B) or EED-tGFP (C) as well as a plasmid encoding the indicated TRIM21–Tn3 variant. The background-subtracted tGFP MFI for each sample was normalized to the average background-subtracted tGFP MFI of the control sample. Data represent mean ± SD of biological triplicates, and statistical significance was determined by one-way ANOVA (Dunnett’s multiple-comparison test compared with the control sample). Statistical data are shown in Table S4. ∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, and ∗∗∗∗p ≤ 0.0001. HEK, human embryonic kidney cell line; MFI, mean fluorescence intensity; POI, protein of interest; tGFP, TurboGFP; Tn3, third fibronectin type III domain of human tenascin-C; TRIM21, tripartite motif containing-21.

To confirm the findings of our flow cytometry studies, we verified degradation via immunoblot analysis of cell lysates from flow cytometry experiments. We observed that TRIM21–ME2 and TRIM21–MH4 promoted the strongest degradation of MALT1-tGFP relative to the EED-targeted control TRIM21–EG9, in agreement with flow cytometry measurements (Figs. 5, A and C and S5A and Table S4). TRIM21–ME2 mediated an average fourfold decrease in normalized MALT1-tGFP band intensity, and TRIM21–MH4 led to an average threefold decrease in normalized band intensity. TRIM21–MC1, TRIM21–MC11, and TRIM21–MA8 did not alter MALT1-tGFP levels relative to the control TRIM21–EG9. Probing with an anti-FLAG antibody revealed that TRIM21–MA8 expressed significantly worse than other TRIM21–Tn3 variants (Fig. 5C), consistent with observations from recombinant expression. We noted expression of native MALT1 (but not EED) in HEK 293F cells, albeit at much lower levels compared with the transfected MALT1-tGFP (Fig. 5A), enabling quantification of endogenous as well as overexpressed MALT1. We found that our most active degrader (TRIM21–ME2) eliminated endogenous MALT1 in this HEK 293F cell system just as effectively as it degraded MALT1-tGFP, leading to an average 3.6-fold reduction in band intensity relative to the EED-targeted control TRIM21–EG9, which was found to be statistically significant (Fig. S5C and Table S4). Moreover, TRIM21–MA8 and TRIM21–MC11, which did not induce degradation of overexpressed MALT1-tGFP (Fig. 5C), also did not degrade endogenous MALT1 (Fig. S5C). Interestingly, variation in degradation efficiency was observed in some of the clones with intermediate performance: TRIM21–MC1 significantly degraded endogenous but not overexpressed MALT1; and TRIM21–MH4 induced statistically significant degradation of overexpressed but not endogenous MALT1 (Fig. 5C and S5C and Table S4). This variability in performance could be attributed to the fact that the HEK 293F system contains little endogenous MALT1 and is engineered to massively overexpress tGFP-tagged MALT1, which competes for binding to our TRIM21–Tn3 variants. It is possible that degradation efficacy could be sensitive to affinity as well as expression of TRIM21–Tn3 fusion proteins and thus may differ for some TRIM21–Tn3 variants depending on target abundance.

Figure 5.

Immunoblot analysis of MALT1 and EED degradation.A and B, immunoblot analysis of cell lysate from HEK 293F cells cotransfected with a plasmid encoding MALT1-tGFP (A) or EED-tGFP (B) as well as a plasmid encoding the indicated TRIM21–Tn3 variant. Cells were harvested and lysed 72 h after transfection. Blots were probed with an anti-MALT1 antibody (A) or an anti-EED antibody (B) as well as an anti-FLAG antibody to detect TRIM21–Tn3. An equivalent amount of total protein was added to each lane, and blots were also probed with an anti-β-actin antibody to confirm equivalent loading. Immunoblots are representative of two replicate experiments. Immunoblot images for biological replicates are shown in Fig. S5, A and B. Bands for MALT1-tGFP and MALT1 are present in A because HEK 293F cells natively express MALT1 (93, 94). C and D, relative quantitation of total MALT1 (C) or EED (D) levels across two biological replicates (Fig. S5A) by densitometry using ImageJ software. Band intensity was normalized to the band intensity of β-actin and then divided by the normalized band intensity of the control sample for each replicate. Data represent the same sample sets analyzed by flow cytometry in Figure 4, B and C. EED, embryonic ectoderm development protein; HEK, human embryonic kidney cell line; MALT1, mucosa-associated lymphoid tissue lymphoma translocation protein 1; tGFP, TurboGFP; Tn3, third fibronectin type III domain of human tenascin-C; TRIM21, tripartite motif containing-21.

With respect to EED targeting, immunoblot analysis confirmed that TRIM21–EG9 and TRIM21–EH8 led to the most extensive EED-tGFP degradation relative to the MALT1-targeted control TRIM21–ME2, consistent with flow cytometry results (Fig. 5, B and D and S5B and Table S4). TRIM21–EG9 led to an average 1.7-fold decrease in normalized EED-tGFP band intensity, and TRIM21–EH8 induced an average 1.4-fold decrease in normalized band intensity. Weak degradation was observed for both TRIM21–EA7 and TRIM21–ED12 by immunoblot analysis, although the anti-FLAG blot revealed that these fusion proteins expressed less prominently than TRIM21–EG9 and TRIM21–EH8.

In the case of both MALT1 and EED, degradation was most actively induced by the TRIM21–Tn3 variants with the highest binding affinities for their target (TRIM21–ME2 and TRIM21–MH4 for MALT1 and TRIM21–EG9 and TRIM21–EH8 for EED) (Fig. 3). Based on these aforementioned degradation studies, we proceeded with the MALT1-targeted TRIM21–ME2 variant and the EED-targeted TRIM21–EG9 variant for further characterization.

To further characterize degradation of tGFP-tagged POI while also tracking expression of the TRIM21–Tn3 variant, mCherry was fused to TRIM21–Tn3 variants with an interceding T2A ribosomal skip peptide (93). The T2A ribosomal skip peptide was included because it has been reported that TRIM21 is degraded along with the target protein, and thus, we wanted to uncouple degradation of the TRIM21–Tn3 fusion from loss of fluorescent signal (42, 46, 49, 50, 51, 52, 53). HEK 293F cells were then cotransfected with either MALT1-tGFP or EED-tGFP and either mCherry-tagged TRIM21–ME2 or mCherry-tagged TRIM21–EG9. After 72 h, degradation was evaluated by plate reader analysis (Fig. 6, B and D), and images were also taken using fluorescence microscopy (Fig. 6, A and C). Normalizing for expression, TRIM21–ME2 induced significant degradation of MALT1-tGFP relative to the control TRIM21–EG9 (eightfold decrease in fluorescence), as determined by plate reader analysis (Fig. 6B) and confirmed using fluorescence microscopy (Fig. 6A). This finding was consistent with observations from flow cytometry and immunoblot analysis of MALT-tGFP degradation (Figs. 4B and 5, A and C). In terms of EED-tGFP degradation, normalizing for expression, TRIM21–EG9 induced a significant decrease in fluorescence relative to the control TRIM21–ME2 (1.7-fold decrease in fluorescence), as measured by plate reader analysis (Fig. 6D) and confirmed by fluorescence microscopy (Fig. 6C). The observed extent of degradation matched findings from flow cytometry and immunoblot analysis of EED-tGFP degradation (Figs 4C and 5, B and D). Taken together, these expression-normalized degradation studies demonstrate that TRIM21–ME2 and TRIM21–EG9 orchestrate robust degradation of their target and that plate reader analysis can be implemented as a high-throughput screening strategy for bioPROTAC discovery.

Figure 6.

Fluorescence microscopy and plate reader analysis of MALT1 and EED degradation. HEK 293F cells were cotransfected with a plasmid encoding MALT1-tGFP (A and B) or EED-tGFP (C and D) as well as a plasmid encoding an mCherry tag fused to the indicated TRIM21–Tn3 variant, separated by a T2A peptide in order to decouple degradation from expression. A and C, after 72 h, 20× images were acquired with a Zeiss Axio Observer with AxioVision 5 software, and composite images were generated with ImageJ. Images are representative of biological triplicates. Scale bar represents 100 μm. B and D, mean tGFP (488 ± 9 nm excitation; 510 ± 9 nm emission) and mCherry fluorescence (580 ± 20 nm excitation; 630 ± 20 nm emission) were also measured after 72 h using a BioTek SynergyMX plate reader (BioTek Instruments, Inc) that analyzed a 3 × 3 grid distributed across the well surface area. Background-subtracted tGFP fluorescence of each well was normalized by the mCherry fluorescence and divided by the mean mCherry-normalized tGFP fluorescence of the appropriate control sample. Data represent mean ± SD of biological triplicates. Statistical significance was determined by a two-tailed unpaired Student’s t test, and statistical data are shown in Table S4. ∗∗∗p ≤ 0.001, and ∗∗∗∗p ≤ 0.0001. EED, embryonic ectoderm development protein; HEK, human embryonic kidney cell line; MALT1, mucosa-associated lymphoid tissue lymphoma translocation protein 1; tGFP, TurboGFP; Tn3, third fibronectin type III domain of human tenascin-C; TRIM21, tripartite motif containing-21.

TRIM21–Tn3 fusion proteins with tandem Tn3 domains can enhance POI degradation

To determine whether avidity effects resulting from the use of multiple Tn3 domains within a TRIM21–Tn3 fusion would impact degradation, we engineered TRIM21–Tn3 fusion proteins that contained two tandem Tn3 domains on each TRIM21 monomer (Fig. S7, A and B). Both Tn3 domains were fused to C terminus of TRIM21ΔPRYSPRY using a flexible (Gly₄Ser)₃ linker, and a (Gly₄Ser)₃ linker was also introduced between the two Tn3 domains (Fig. S7A). For MALT1-targeted fusion proteins, we employed the two lead Tn3 clones, ME2 and MH4, implementing all four orientation permutations (ME2–ME2, ME2–MH4, MH4–MH4, and MH4–ME2). Similarly, for EED-targeted fusion proteins, we used the two lead Tn3 clones, EG9 and EH8, designing all four orientation permutations (EG9–EG9, EG9–EH8, EH8–EH8, and EH8–EH9). A TRIM21–Tn3 variant comprising a nonspecific Tn3 was used as a control for all experiments. HEK 293F cells were cotransfected with a plasmid encoding the respective tGFP-tagged POI and a plasmid encoding a TRIM21–Tn3 or TRIM21–Tn3–Tn3 fusion protein. For MALT1-tGFP degradation, all fusion proteins induced significant degradation relative to the control TRIM21–Tn3 (Fig. S7C and Table S4), and TRIM21–ME2–ME2 led to the most significant reduction in MALT1-tGFP signal (10-fold decrease). The difference in degradation was significant as compared with both TRIM21–ME2 and TRIM21–MH4 (which had average fourfold and fivefold decreases, respectively). TRIM21–MH4–ME2 also improved degradation relative to TRIM21–Tn3 fusion proteins with only a single Tn3 domain, inducing an average sixfold decrease in MALT1-tGFP signal, which was statistically significant compared with TRIM21–ME2 although not TRIM21–MH4 (Fig. S7C). As for EED degradation, all fusion proteins led to significant degradation compared with the control TRIM21–Tn3 (Fig. S7D and Table S4), and the single Tn3 domain–containing TRIM21–EG9 performed best with an average fourfold decrease in EED-tGFP signal. Although none of the Tn3–Tn3 fusion proteins led to significantly more degradation compared with TRIM21–EG9, both TRIM21–EG9–EG9 and TRIM21–EG9–EH8 induced a significant reduction in EED-tGFP signal as compared with TRIM21–EH8 (Fig. S7D). Interestingly, TRIM21–EH8–EG9 led to significantly less degradation compared with TRIM21–EG9, highlighting the importance of orientation on degradation efficiency. Taken together, these data suggest that avidity effects resulting from serially combining multiple Tn3 domains may in some instances be used as a strategy to modulate TPD in the context of TRIM21-based bioPROTACs, although the impact of avidity may vary based on the particular permutation of Tn3 domains being used as well as the POI that is being targeted.

Choice of E3 ligase impacts degradation

To compare the performance of TRIM21–Tn3 fusion proteins with that of other Tn3 domain–containing E3 ligases, we fused our lead Tn3 domains (ME2 for MALT1 and EG9 for EED) to three additional E3 ligases: VHL, CHIP, and SPOP. Tn3 domains were fused to the C terminus of full-length VHL (residues 1–213) via a Gly₄Ser linker (94). Conversely, Tn3 domains were fused to the N terminus of truncated CHIP (residues 128–303) and truncated SPOP (residues 167–374) via a Gly₄Ser linker (1) (Fig. S8A and Table S1). Degradation of tGFP-tagged POI was analyzed by flow cytometry 72 h after cotransfection of HEK 293F cells with a plasmid encoding the tGFP-tagged POI and a plasmid encoding the designated bioPROTAC. For each E3 ligase, degradation was normalized to the maximum signal of a control fusion protein comprised of a Tn3 specific to a different POI and the matched E3 ligase (Fig. S8, B and C).

For MALT1 degradation, both TRIM21–ME2 and VHL–ME2 induced significant degradation relative to their EED-targeted control fusion proteins TRIM21–EG9 and VHL–EG9. Both had an average threefold decrease in MALT1-tGFP signal relative to their controls. Both ME2–CHIP and ME2–SPOP trended toward MALT1-tGFP degradation but were not statistically significant compared with their respective control fusion proteins (Fig. S8B). For EED degradation, only TRIM21–EG9 induced a significant decrease in EED-tGFP signal relative to its control fusion protein. VHL–ME2 and EG9–CHIP did not significantly alter EED-tGFP signal compared with their respective control fusion proteins, whereas EG9–SPOP led to a significant increase in their EED-tGFP signal (Fig. S8C). Taken together, these results indicate that distinct E3 ligases can differentially impact degradation efficacy, dependent upon both the POI targeted as well as the binding domain and fusion orientation.

TRIM21–Tn3 variants induce targeted degradation through the UPS

To illustrate that TRIM21–Tn3 variant–mediated degradation was occurring through the UPS, we introduced several mutations (E12R, D21R, and R55A) into TRIM21 known to inhibit E2 recruitment and subsequent polyubiquitination (52). These mutations were installed into both of our lead TRIM21–Tn3 fusion proteins (TRIM21–ME2 and TRIM21–EG9) to create inactive variants. TRIM21–Tn3 variants were fused to an mCherry tag, separated by a T2A peptide, to track expression, and degradation of tGFP-tagged POI was interrogated. Degradation was measured 72 h after cotransfection, and degradation was measured by flow cytometry, normalizing POI signal to TRIM21–Tn3 expression for each condition. Unmodified TRIM21–ME2 induced significantly more degradation of MALT1-tGFP than did inactive TRIM21–ME2, demonstrating that degradation was, in fact, orchestrated through the UPS (Fig. 7A). We noted that cotransfection with inactive EED-targeted TRIM21–EG9 led to higher levels of MALT1-tGFP compared with the unmodified TRIM21–EG9, but the MALT1-tGFP levels were not significantly different from those observed in the presence of inactive TRIM21–ME2 (Fig. 7A and Table S4). Plate reader analysis confirmed flow cytometry observations (Fig. S6A), suggesting that these methods can be used interchangeably. In the case of EED degradation, TRIM21–EG9 induced significant degradation of EED-tGFP relative to inactive TRIM21–EG9 (Fig. 7B), confirming that EED degradation invokes the UPS. As expected, no significant differences in EED levels were observed between the unmodified MALT1-targeted control TRIM21–ME2 and the inactive TRIM21–ME2 (Fig. 7B and Table S4). Results were confirmed by plate reader analysis (Fig. S6B). Overall, TRIM21 inactivation experiments confirmed that both TRIM21–ME2 and TRIM21–EG9 mediate target degradation through the UPS. These data could be further corroborated by comparing the effects of proteasome inhibitors versus lysosome and autophagy inhibitors.

Figure 7.

TRIM21–Tn3 fusion proteins induce targeted degradation through the UPS.A and B, flow cytometry analysis 72 h after cotransfection of HEK 293F cells with a plasmid encoding MALT1-tGFP (A) or EED-tGFP (B) and a plasmid encoding an mCherry tag fused to the indicated TRIM21–Tn3 variant, separated by a T2A peptide. Inactive TRIM21–Tn3 variants contain mutations in the TRIM21ΔPRYSPRY region that prevent E2 recruitment and subsequent ubiquitination. For each condition, the background-subtracted tGFP MFI was normalized by the background-subtracted mCherry MFI and then divided by the average mCherry-normalized tGFP MFI of the appropriate control sample. Data represent mean ± SD of biological triplicates, and statistical significance was determined by one-way ANOVA with a Tukey post hoc test. Statistical data are shown in Table S4. ∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, and ∗∗∗∗p ≤ 0.0001. EED, embryonic ectoderm development protein; HEK, human embryonic kidney cell line; MALT1, mucosa-associated lymphoid tissue lymphoma translocation protein 1; MFI, mean fluorescence intensity; tGFP, TurboGFP; Tn3, third fibronectin type III domain of human tenascin-C; TRIM21, tripartite motif containing-21; UPS, ubiquitin proteasome system.

TRIM21–Tn3 fusion proteins degrade overexpressed MALT1 and EED without a tGFP tag

To confirm that lysine residues on tGFP were not contributing to degradation of the POI, we sought to evaluate degradation of MALT1 and EED without a tGFP tag. HEK 293F cells were cotransfected with a plasmid encoding unfused MALT1 or EED and a plasmid encoding the appropriate lead TRIM21–Tn3 variants, and degradation was evaluated by immunoblot analysis 72 h post-transfection. MALT1 degrader TRIM21–ME2 led to an average 2.6-fold reduction in MALT1 band intensity relative to the EED-targeted control TRIM21–EG9, normalizing to the β-actin loading control (Figs. 8, A and D and S9A). EED degrader TRIM21–EG9 led to an average 1.5-fold reduction in band intensity relative to the MALT1-targeted control TRIM21–ME2, also normalizing to the β-actin loading control, which was statistically significant (Figs. 8, B and E and 9B and Table S4). These findings align with flow cytometry and immunoblot analyses of tGFP-tagged MALT1 and EED (Figs. 4 and 5), indicating that TRIM21–Tn3 variants actively degrade MALT1 and EED in the absence of a tGFP tag.

Figure 8.

TRIM21–Tn3 fusion proteins degrade untagged and endogenous POI.A and B, immunoblot analysis of cell lysate from HEK 293F cells cotransfected with a plasmid encoding MALT1 (A) or EED (B) and a plasmid encoding the indicated TRIM21–Tn3 variant. C, immunoblot analysis of cell lysate from HEK 293F cells transfected with a plasmid encoding the indicated TRIM21–Tn3 variant or an empty plasmid (“empty”). Cells were harvested and lysed 72 h after transfection. Blots were probed with an anti-MALT1 antibody (A and C) or an anti-EED antibody (B), and an anti-FLAG antibody was used to detect TRIM21–Tn3. An equivalent amount of total protein was added to each lane, and blots were probed with an anti-β-actin antibody to confirm equivalent loading. Immunoblots are representative of two replicate experiments. Immunoblot images for biological replicates are shown in Fig. S9. D–F, relative quantitation of total MALT1 (D and F) or EED (D) levels across biological replicates by densitometry using ImageJ software. MALT1 or EED band intensity was normalized to the band intensity of β-actin and then divided by the normalized band intensity of the control sample for each replicate. Statistical significance was determined by two-tailed paired Student’s t test. Statistical data are shown in Table S4. ∗p ≤ 0.05 and ∗∗p ≤ 0.01. EED, embryonic ectoderm development protein; HEK, human embryonic kidney cell line; MALT1, mucosa-associated lymphoid tissue lymphoma translocation protein 1; POI, protein of interest; Tn3, third fibronectin type III domain of human tenascin-C; TRIM21, tripartite motif containing-21.

TRIM21–Tn3 fusion proteins degrade native MALT1

To demonstrate the potential application of TRIM21 bioPROTACs in the study and treatment of human diseases, we sought to determine whether TRIM21–Tn3 fusion proteins could degrade endogenous POI. To that end, we interrogated degradation of endogenously expressed MALT1 that was observed in HEK 293F cells (Fig. 5A). with either TRIM21–ME2, TRIM21–EG9, or an empty plasmid, and degradation was evaluated 72 h post-transfection. The MALT1-specific TRIM21–ME2 led to a statistically significant 1.6-fold average reduction in normalized MALT1 band intensity relative to the EED-targeted TRIM21–EG9 control (p = 0.0045) and an average 1.4-fold reduction in normalized MALT1 band intensity relative to cells transfected with an empty plasmid (Figs. 8, C and F and S9, C and D and Table S4). This result confirms that TRIM21–Tn3 fusion proteins are capable of degrading their native POI targets at endogenous levels.

Discussion

TPD represents a compelling tool for proteome editing that has broad applications in basic research, drug discovery, and therapeutic development (1, 2, 3, 4, 16). In contrast with neutralization or other functional inhibition approaches, TPD offers several advantages, including substantial and durable knockout of target protein activity. Small molecule–based approaches such as PROTACs and molecular glues have been prominent areas of focus within the TPD drug development landscape, with several molecules currently in clinical development (21, 22, 23, 28, 29, 30). In addition, protein-based degraders (bioPROTACs) have emerged as a complementary approach and have demonstrated success in both cellular and animal models (1, 2, 16).

Here, we developed bioPROTACs based on the E3 ligase TRIM21 targeted against two different cancer targets commonly associated with B-cell lymphomas, MALT1 and EED. Specifically, we replaced the native substrate-binding domain of TRIM21 (the PRYSPRY domain), which naturally engages antibody Fc domains, with an engineered binding scaffold based on the fibronectin domain Tn3. Tn3 domain discovery campaigns suggested a preference for evolution of specific binders from the loop and flat libraries, although not the concave library, which mutagenized residues within a concave beta-sheet surface (Fig. S3B). We found that when constructed as TRIM21–Tn3 fusion proteins, clones that were isolated from the flat library performed better in terms of POI binding and degradation (Figs. 3 and 4). This could be due to the orientation of the Tn3 domain relative to the fused TRIM21ΔPRYSPRY domain, as the three engineered loops of the loop libraries are predicted to be in close proximity to the N terminus of the Tn3 domain where it is fused to TRIM21ΔPRYSPRY, based on structural modeling (Fig. S4A). It is possible that those loops may no longer be accessible to bind the POI when the Tn3 domain is fused to TRIM21ΔPRYSPRY, and reversing the topology of the TRIM21–Tn3 fusion protein could alleviate this steric interference. This is important to keep in mind since during the discovery of these Tn3 domains, the Tn3 domain was tethered at the C terminus to Aga2p, thus leaving the N terminus free (Fig. S3A).

Interestingly, for the lead Tn3 domains isolated from the flat libraries (ME2 and EG9), fusion to other E3 ligases in the Tn3-CHIP and Tn3-SPOP formats (both of which position the Tn3 domain at the N terminus), no significant degradation was observed (Fig. S8). In contrast, for VHL–ME2 (Tn3 domain at the C terminus), significant degradation of MALT1-tGFP was observed relative to the control nontarget-specific TRIM21–Tn3 fusion protein (Fig. S8B), although no significant degradation of EED-tGFP was induced by the VHL–EG9 fusion protein (Fig. S8C). Thus, differences in degradation efficiency can result from fusion orientation, E3 ligase choice, or a combination of both, and also are dependent on the POI. More extensive engineering of TRIM21–Tn3 constructs and other Tn3-fused E3 ligase proteins will further elucidate the role of construct geometry, E3 ligase use, and target structure on TPD efficacy. The workflow that we followed in this study provides an infrastructure for further investigation of these important molecular properties as well as several high-throughput strategies employing flow cytometry and fluorescence plate readers to screen for degradation. Our study also highlights the modularity of bioPROTAC formats compared with small-molecule approaches; POI-binding domains and E3 ligases can easily be interchanged, enabling rapid screening of a diverse panel of bioPROTACs.

In addition to evaluating the choice of E3 ligase, this study also evaluated the impact of avidity effects resulting from incorporation of multiple Tn3 domains in TRIM21 fusion proteins (Fig. S7A). In the case of MALT1 degradation, we observed a significant benefit when multiple Tn3 domains were employed, with TRIM21–ME2–ME2 inducing the largest decrease in MALT1-tGFP signal (Fig. S7C). However, in the case of EED degradation, incorporation of additional Tn3 domains did not improve upon the degradation activity of TRIM21–EG9 (Fig. S7D). Thus, although we observe that avidity effects can in some cases enhance degradation efficacy, the effects will vary depending on the POI and the properties of the particular Tn3 domains that are used, most importantly their binding epitopes, affinities, and cooperativity. Indeed, significant disparities in degradation were observed upon altering Tn3 domain orientation or valency in both MALT1- and EED-targeted degraders. Thus, Tn3 domain layout represents another key design consideration when engineering bioPROTACs, particularly for the case of POI-binding designs with weak affinity toward the target POI.

We demonstrated that our most active MALT1-specific TRIM21–Tn3 fusion protein significantly degrades endogenous MALT1 in HEK 293F cells, both in the presence and absence of MALT1 overexpression (Figs. S5C and 8, C and F). Lower levels of MALT1 expression in this system could lead to reduced avidity effects and impair degradation activity, especially since TRIM21 clustering may be important for inducing degradation (49). This issue could potentially be overcome by enhancing Tn3 domain affinities or exploring alternative delivery strategies to increase the expression of TRIM21–Tn3 fusion proteins. Moreover, to interrogate the therapeutic potential of engineered TRIM21–Tn3 fusion proteins, it will be essential to demonstrate degradation of native POIs in the context of cancer cell lines. In addition, it will be important to address target specificity of our TRIM21–Tn3 fusion proteins through proteomic analyses to advance future clinical development of these molecules.

Overall, this study provides a new method for designing TRIM21-based bioPROTACs, offers mechanistic insights into the structural and molecular correlates of TRIM21 bioPROTAC-induced degradation, and presents a robust workflow for evaluating degradation activity of bioPROTACs. Future work will expand upon these efforts to evaluate whether engineered TRIM21–Tn3 fusion proteins can degrade target proteins in B-cell lymphomas and other cancer types that natively express either MALT1 or EED. It will also be critical to characterize the effects of native POI degradation on the dynamics and kinetics of downstream signaling in these systems in order to demonstrate their functional utility and determine their potential as pharmacological interventions. Moreover, extensive in vivo testing will be required to assess and realize the therapeutic potential for our TRIM21 bioPROTAC approach. We also note that affinity maturation of the Tn3 domains may be required, as most of our degradation studies were conducted in systems where either the TRIM21–Tn3 fusion protein and/or the target protein was overexpressed. In relevant biological systems, there may be low expression levels of the target protein, and it may be difficult to achieve robust expression of bioPROTACs in these cells, thus necessitating target affinity enhancement or improved delivery strategies. Ultimately, the success of our approach will be largely dependent on the ability to deliver TRIM21–Tn3 fusion proteins to cells of interest, which remains one of the main challenges for bioPROTACs and other biologics targeting intracellular proteins (2). Recent approaches have primarily focused on gene delivery of bioPROTACs through viral transduction; however, protein delivery through nanoparticle-based strategies remains a complementary approach. In addition, bioPROTACs could be engineered with a cell-penetrating peptide to facilitate cell entry (2). Another approach could be to develop bispecific bioPROTACs that target an internalizing cell surface receptor as well as an intracellular POI. This would create a targeted approach that can enter the cell through receptor-mediated endocytosis, and then using an endosomal escape peptide, it can escape into the cytosol to allow subsequent targeting and degradation of an intracellular POI. Indeed, the development of innovative delivery strategies will be vital to the successful translation of bioPROTACs into the clinic.

Recently, biologics in the TPD field have also expanded toward targeting extracellular or cell surface proteins through approaches such as lysosome-targeting chimeras and antibody-based PROTACs (2, 5, 6, 7, 16). This demonstrates the growing interest in specifically degrading a wide range of disease-relevant proteins using biologics-based approaches. Our study presents the design of TRIM21 bioPROTACs that can be used as new tools for redirecting the native E3 ligase activity of TRIM21 to degrade other POIs. Recent technologies, including TRIM-Away (56, 57), TRIMbody-Away, optoTRIM-Away, and miniTRIM-Away (49, 59), have demonstrated that the E3 ligase TRIM21 can be successfully incorporated into biologics-based degraders. To build upon that work, we sought to develop TRIM21 bioPROTACs that directly engage a target POI, without the need for an antibody intermediary. We identified a panel of Tn3-binding domains via directed evolution of in-house developed yeast-displayed Tn3 libraries against two different cancer-associated intracellular targets for incorporation into the TRIM21 scaffold. We demonstrated that these TRIM21–Tn3 fusion proteins induced degradation of their target when the POI was overexpressed as well as when the POI was present at endogenous levels. Furthermore, we developed a high-throughput fluorescence-based workflow for the characterization and optimization of TRIM21–Tn3 fusion proteins that directly engage and degrade their target POI. Overall, the workflow and design insights from this study promise to inform and inspire further development of bioPROTACs for fundamental research as well as for therapeutic discovery and design applications.

Experimental procedures

Plasmids

All plasmids used in this study are provided in Table S5. One Shot TOP10 Chemically Competent Escherichia coli were used for the construction and propagation of all plasmids. All genes and PCR products were cloned into plasmids by Gibson Assembly (NEB).

To create pEBNA3-HER4, the extracellular domain (ECD) of HER4 (residues 26–649) was cloned into the pEBNA3 vector with an N-terminal CD33 signal peptide and a C-terminal 6× His tag (95).

pOE-HR4–Fc was constructed by cloning the HR4 Tn3 sequence fused to the N terminus of the Fc domain of human IgG4 into an in-house expression vector (pOE). An N-terminal CD33 signal peptide was included for expression.

To construct pOE-TRIM21, a gene fragment (Twist Bioscience) encoding full-length human TRIM21 (residues 1–475) was cloned into the pOE vector with an N-terminal 6× His and FLAG tag and no signal peptide.

pOE plasmids encoding TRIM21–Tn3 variants were created by gene assembly of TRIM21ΔPRYSPRY (residues 1–286) and the appropriate Tn3 domain using overlap extension PCR with primers that introduced a flexible 15-amino acid (Gly₄Ser)₃ linker between TRIM21ΔPRYSPRY and the Tn3 domain. The PCR product was then cloned into the in-house pOE vector.

pOE plasmids encoding TRIM21–Tn3 variants with a “self-cleaving” N-terminal mCherry tag were created by gene assembly of mCherry and the appropriate TRIM21–Tn3 fusion using overlap extension PCR with primers that introduced a T2A ribosomal skip peptide (GSGEGRGSLLTCGDVEENPGP) between mCherry and the TRIM21–Tn3 fusion (93). The PCR product was then cloned into a pOE vector that did not include an N-terminal 6× His or FLAG tag (pOE-mCherry-T2A-TRIM21-Tn3).

To create inactive TRIM21 mutants that do not recruit E2 enzymes, the following TRIM21 mutations were introduced into our TRIM21–Tn3 constructs: E12R, D21R, and R55A (52). These mutations were introduced into the pOE-mCherry-T2A-TRIM21-Tn3 constructs.

pOE plasmids encoding TRIM21–Tn3–Tn3 fusion proteins were created by fusing two Tn3 domains to the C terminus of TRIM21ΔPRYSPRY using a flexible (Gly₄Ser)₃ linker. A (Gly₄Ser)₃ linker was also introduced between the two Tn3 domains. The control TRIM21–Tn3 used for evaluating degradation with TRIM21–Tn3–Tn3 constructs refers to TRIM21ΔPRYSPRY with a C-terminal Tn3 domain that was isolated as a binder against the extracellular target glycoprotein A33 (Fig. S7).

To create plasmids encoding VHL–Tn3, Tn3–CHIP, and Tn3–SPOP variants, gene fragments (Integrated DNA Technologies) were cloned into the pOE vector. For VHL–Tn3 fusion proteins, Tn3 domains were fused to the C terminus of full-length VHL (residues 1–213) via a Gly₄Ser linker. An N-terminal 6× His and FLAG tag were also included. For Tn3–CHIP fusion proteins, the Tn3 domains were fused at the N terminus of truncated CHIP (residues 128–303). For Tn3–SPOP fusion proteins, the Tn3 domains were fused at the N terminus of truncated SPOP (residues 167–374). Both Tn3–CHIP and Tn3–SPOP constructs had a C-terminal 6× His and FLAG tag.

pCMV6-AC-tGFP-MALT1 and pCMV6-AC-tGFP-EED plasmids were purchased from OriGene. These plasmids encode respectively for human MALT1 and EED with a C-terminal tGFP tag. To construct pCMV6-AC-MALT1 and pCMV6-AC-EED (plasmids encoding MALT1 and EED, respectively, without a tGFP tag), MALT1 and EED were PCR amplified and cloned into pCMV6-AC using BamHI and PmeI.

Cell lines

HEK 293F cells (ThermoFisher Scientific) were cultivated in Freestyle 293 Expression Medium (ThermoFisher Scientific) supplemented with 2 U/ml penicillin–streptomycin (Gibco). To generate the HER4⁺ AD-293 cell line, AD-293 cells were transduced with human HER4 lentivirus and cultured in high-glucose Dulbecco's modified Eagle's medium (Gibco) supplemented with 10% fetal bovine serum (FBS) and 10 μg/ml of blasticidin for selection of HER4⁺ cells. HBL-1 cells were cultured in RPMI1640 with GlutaMAX Supplement (Gibco) and 10% FBS. Karpas 422 cells were cultured in RPMI1640 with GlutaMAX Supplement and 20% FBS. All cell lines were maintained at 37 ºC in a humidified atmosphere with 5% CO₂.

Protein purification and expression

Full-length human TRIM21 (residues 1–475) was expressed recombinantly in HEK 293F cells via transient transfection using the pOE-TRIM21 plasmid. HEK 293F cells were grown to 1.2 × 10⁶ cells/ml and diluted to 1.0 × 10⁶ cells/ml. Midiprepped DNA (0.66 mg total plasmid per liter of cells) and 1 ml per liter of cells of 293fectin (ThermoFisher Scientific) were independently diluted in 33.3 ml of Opti-MEM I (ThermoFisher Scientific), respectively, and incubated at room temperature for 5 min. The DNA mixture was then added to the 293fectin mixture and incubated at room temperature for an additional 25 min. Subsequently, the diluted HEK 293F cells and the DNA/293fectin mixture were added to a shaking flask and incubated at 37 °C and 5% CO₂ with rotation at 125 rpm for 5 days. After 5 days, the cells were pelleted (800g for 20 min), and the supernatant was discarded. The cells were then lysed with 10 ml/g of cell pellet of M-PER Mammalian Protein Extraction Reagent (ThermoFisher Scientific) with EDTA-free Pierce Protease Inhibitor Tablets (ThermoFisher Scientific; one tablet/50 ml of lysis buffer). Cells were lysed for 20 min at room temperature, and following the incubation, the mixture was centrifuged at 10,000g for 20 min. The supernatant was then diluted to a final volume of 1 l in Tris-buffered saline (TBS, 50 mM Tris–HCl [pH 8], 150 mM NaCl). TRIM21 was then purified from cell lysate via Ni–NTA affinity chromatography followed by SEC using a Superdex 200 Increase 10/300 GL column (GE Healthcare) on an FPLC instrument, equilibrated in TBS (50 mM Tris–HCl [pH 8] and 150 mM NaCl) containing 1 mM DTT. Purity was verified by SDS-PAGE analysis.

TRIM21–Tn3 constructs were expressed and purified in the same manner as described for TRIM21 except that cells were transfected with the pOE plasmid encoding the appropriate TRIM21–Tn3 variant.

The ECD of HER4 (residues 26–649) was expressed recombinantly in HEK 293F cells following the same transfection protocol as aforementioned except that cells were transfected with the in-house pEBNA3 plasmid encoding the ECD of HER4. Transfected HEK 293F cells were incubated at 37 °C and 5% CO₂ with rotation at 125 rpm for 5 days, and then the ECD of HER4 was purified from the supernatant via Ni–NTA chromatography. The protein was then biotinylated using EZ-Link Sulfo-NHS-Biotin (ThermoFisher Scientific) and buffer exchanged by dialysis into PBS to remove excess biotin.

The Fc–HR4 fusion protein was expressed recombinantly in HEK 293F cells following the same transfection protocol as aforementioned except that cells were transfected with the in-house pOE plasmid encoding the HR4 Tn3 domain linked to the Fc domain of human IgG4. Transfected HEK 293F cells were incubated at 37 °C and 5% CO₂ with rotation at 125 rpm for 5 days, and then, Fc–HR4 was purified from the supernatant via protein A chromatography.

BLI binding measurements

Binding studies were performed using BLI on an OctetRED96 BLI instrument (Sartorius). For proof-of-concept studies, biotinylated HER4 was immobilized to streptavidin-coated biosensors (Sartorius) in 0.45 μm filtered PBSA (PBS [pH 7.2] containing 0.1% BSA). HER4 was immobilized at a concentration of 50 nM for 120 s. Once baseline measurements were collected in PBSA, binding kinetics were measured by submerging the biosensors in wells containing serial dilutions of the appropriate analyte for 300 s (association) followed by submerging the biosensor in wells containing only PBSA for 450 s (dissociation). Tips were regenerated in 0.1 M glycine (pH 2.7).

For evaluating binding to MALT1 or EED, TRIM21–Tn3 variants were immobilized to Anti-Penta-His-coated biosensors (Sartorius) in 0.45 μm filtered PBSA (PBS [pH 7.2] containing 0.1% BSA). TRIM21–Tn3 variants were immobilized at a concentration of 150 nM for 120 s. Once baseline measurements were collected in PBSA, binding kinetics were measured by submerging the biosensors in wells containing serial dilutions of recombinant MALT1 (BPS Bioscience) or EED (BPS Bioscience) for 300 s (association) followed by submerging the biosensor in wells containing only PBSA for 450 s (dissociation).

Data were visualized and processed using the Octet Data Analysis software, version 7.1 (Sartorius). For kinetic fit values, analysis and kinetic curve fitting (assuming a 1:1 binding model) were conducted using Octet Data Analysis software, version 7.1. Equilibrium titration curve fitting and equilibrium K_D value determination was implemented using GraphPad Prism (GraphPad Software, Inc), assuming all binding interactions to be of first order. Experiments were reproduced two times with similar results.

Cell binding

For binding studies performed on HER4⁺ AD-293, 2 × 10⁵ cells per well were transferred to a 96-well plate and incubated in PBSA containing serial dilutions of recombinant TRIM21, TRIM21-HR4, or Fc–HR4. Cells were then washed and stained with either antihuman IgG Fc APC (HP6017; BioLegend; catalog no.: 409306; 1:50 dilution) or anti-Penta-His AlexaFluor 647 (Qiagen; catalog no.: 35370; 1:50 dilution). After a final wash, cells were analyzed for binding using a CytoFLEX flow cytometer (Beckman Coulter). Background-subtracted and normalized binding curves were fitted to a first-order binding model, and K_D values were determined using GraphPad Prism. Studies were performed two times with similar results.

In vitro ubiquitination

Ubiquitination reactions for TRIM21 versus TRIM21–HR4 were performed with 0.2 mM ubiquitin (R&D Systems), 2 mM ATP (ThermoFisher Scientific), 0.5 μM Ube1 (R&D Systems), 1.5 μM Ube2W (R&D Systems), 0.5 μM Ube2N/2V2 (R&D Systems), and 1.5 μM of recombinant TRIM21 or TRIM21–Tn3 in 50 mM Tris (pH 8), 2.5 mM MgCl₂, and 0.5 mM DTT. Ubiquitination reactions for TRIM21 versus TRIM21–ME2 and TRIM21 versus TRIM21–EG9 were performed with 0.2 mM ubiquitin, 2 mM ATP, 0.5 μM Ube1, 1.5 μM Ube2W (Sino Biological), 0.5 μM Ubc13(Ube2N)/2V2 (Ubiquitin-Proteasome Biotechnologies), and 1.5 μM of recombinant TRIM21 or TRIM21–Tn3 in 50 mM Tris (pH 8), 2.5 mM MgCl₂, and 0.5 mM DTT. Reactions were also performed in the presence or the absence of pathway components as indicated. The reaction was performed at 37 °C for an hour and was stopped by the addition of SDS sample buffer. The samples were boiled at 90 °C for 5 min and resolved by SDS-PAGE. Immunoblotting was also performed using a horseradish peroxidase (HRP)-conjugated antiubiquitin antibody (BioLegend; catalog no.: 646304, 1:1000 dilution).

Tn3 library construction

Tn3 was previously engineered for enhanced thermal stability creating the CM4 Tn3 mutant (63). The amino acid sequence for this Tn3 mutant is as follows:

RLDAPSQIEVKDVTDTTALITWFKPLAEIDGFELTYGIKDVPGDRTTIKLTEDENQYSIGNLKPDTEYEVSLISRRGDMSSNPAKITFKTGL (92)

This sequence was cloned into an in-house yeast display vector with a C-terminal V5 tag (GKPIPNPLLGLDSTRT) followed by the Aga2 protein.

Several yeast-displayed Tn3 libraries were constructed in a similar manner as to what was described for the phage-displayed Tn3 libraries (61). In summary, four loop libraries were constructed to randomize codons in the BC (residues 23–31), DE (residues 51–56), and FG (residues 75–80) loop coding regions. Sequence diversity was also introduced at positions 33, 49, and 73. In addition to sequence diversity, length diversity was also introduced, such that BC loops of 9, 11, and 12 amino acids in length were created and FG loops of six and nine amino acids. Two concave libraries were designed in which codons were randomized in the CD (residues 39–45) and FG (residues 75–80) loop coding regions. Sequence diversity was also introduced at positions 30, 31, 33, 37, 47, 49, and 73. Length diversity was introduced such that CD loops of five and seven amino acids in length were created. A flat library was also designed in which sequence diversity was introduced at positions 7, 9, 11, 12, 14 to 17, 19, 21, 23, 24, 54 to 56, 58, 60, and 61.

Amino acid diversity in the Tn3 libraries was designed taking into consideration the structural context of each position as well as conservation across Tn3 homologs and the FN3 family. For loop libraries, BC, DE, and FG loop positions as defined previously were diversified using a “loop mixture” of 15% each of Tyr, Gly, and Ser, 10% each of Asp and Arg, and 5% each of Ala, Leu, Thr, Val, His, Trp, and Pro. However, position 25 in the BC loop was diversified using 50% Pro and 50% loop mixture as defined previously. Position 33 was diversified using 80% Glu, 10% Tyr, and 10% Arg. Position 49 was diversified using a “strand mixture” of 15% each Tyr, Thr, and Ser, 10% each of Arg and Val, 7.5% each of Ala and Leu, and 5% each of Asn, Asp, Glu, and Trp. Finally, position 73 was diversified using a mixture of 80% Ile, 10% Tyr, and 10% Arg.

For concave libraries, FG and CD loop residues were diversified using the loop mixture of amino acids (positions and mixture defined previously). Positions 30 and 31 were also diversified using the loop mixture. Positions 33, 37, 47, and 49 were diversified using the strand mixture of amino acids as defined previously, and position 73 was diversified using 80% Ile, 10% Tyr, and 10% Arg.

For the flat library, positions 7, 9, 11, 12, 14, 16, 17, 21, and 58 were diversified using the strand mixture of amino acids. Positions 15, 23, 24, 54 to 56, 60, and 61 were diversified using the loop mixture of amino acids, and position 19 was diversified using 25% Val, 25% Ile, 25% Thr, and 25% Leu.

Tn3 library selections

Recombinant human MALT1 (BPS Bioscience; catalog no.: 100360) and recombinant human EED (Abcam; catalog no.: ab156723) were buffer exchanged into PBS via dialysis and then biotinylated using EZ-Link Sulfo-NHS-Biotin. Proteins were then buffer exchanged by dialysis into TBS (40 mM Tris–HCl [pH 8] and 150 mM NaCl) to remove excess biotin.

For each round, the initial number of yeast used was chosen to ensure 10-fold coverage of the library. Positively selected clones from each round were grown at 30 °C in SDCAA liquid media (pH 4.5) for 2 days, followed by induction in SGCAA liquid media (pH 4.5) for 2 days at 20 °C. Sorting concentrations were determined by library titrations against the target.

All seven of the previously described libraries were separately screened against MALT1 and EED. The protocol described later is the same for selections against both targets, and the concentration of target used for each round is indicated.

For the first round of selection, libraries were debulked via MACS. First, a negative selection for each of the libraries was performed in which 10¹⁰ induced yeast were incubated with 100 μl of streptavidin microbeads (Miltenyi Biotec) and 9.9 ml PBE (PBS with 0.1% BSA and 1 mM EDTA) for 1 h at 4 °C. Yeasts were then washed and applied to LS MACS separation columns (Miltenyi Biotec), and clones nonspecifically binding to the microbeads were discarded. Yeast that flowed through the column (i.e., did not bind to streptavidin microbeads) were then pelleted and resuspended in 5 ml of PBE. Then, 100 nM of biotinylated target (for both MALT1 and EED) was incubated with 100 μl streptavidin microbeads for 30 min at 4 °C in PBE. About 5 ml of yeast was then added to the streptavidin microbeads coated with target and incubated for 2 h at 4 °C. Yeasts were then washed and applied to an LS MACS separation column, and yeasts eluted from the column were collected, grown, and induced for the next round of selection.

For the second round of selection, prior to inducing the yeast, the libraries were combined into two groups: group 1 (four loop libraries) and group 2 (two concave libraries and the flat library). Equal numbers of cells were taken from each library within the group to achieve an absorbance of 0.7 in 100 ml of induction media. For each library group, the selection strategy was the exact same as described for the first round of selections, except that 10⁹ induced yeast were used from each group.

For the third round of selection, library groups were screened against either monomer or tetramer and sorted via FACS using a Sony cell sorter. For MALT1 selections, group 1 was screened against 50 nM MALT1 and group 2 was screened against 50 nM MALT1 tetramer. For EED selections, both groups 1 and 2 were screened against 50 nM EED tetramer. Tetramers were formulated by incubating a 4.5:1 ratio of biotinylated MALT1 or EED to AlexaFluor 647-conjugated streptavidin (ThermoFisher Scientific) for 15 min on ice. The monomer or tetramer was then added to 10⁷ induced yeast for each library group and incubated for 2 h at 4 °C. For tetramer selections, a 1:1000 dilution of DyLight 488-conjugated anti-V5 (E10/V4RR; ThermoFisher Scientific; catalog no.: MA5-15253-D488) was also added to the mixture. For monomer selections, after the 2 h incubation, yeasts were washed and incubated with a 1:1000 dilution of DyLight 488-conjugated anti-V5 and 1:200 dilution of AlexaFluor 647-conjugated streptavidin for 20 min on ice. Yeasts were then washed and sorted by FACS in which the double positive population was collected.

For the fourth round, library groups were screened against monomer and sorted via FACS using a BD FACSAria II. For MALT1 selections, group 1 was screened against 3.5 nM MALT1 and group 2 was screened against 75 nM MALT1. For EED selections, group 1 was screened against 25 nM EED and group 2 was screened against 8 nM EED. A negative selection for each of the libraries was performed in which 3 × 10⁷ induced yeast were incubated with a 1:200 dilution of AlexaFluor 647-conjugated streptavidin for 2 h at 4 °C. Yeasts were washed and then incubated with 1:20 anti-Cy5/anti-Alexa Fluor 647 microbeads (Miltenyi Biotec) for 20 min at 4 °C. Yeasts were washed and then applied to an LS MACS separation column, and yeasts that flowed through the column (i.e., did not bind to AlexaFluor 647-conjugated streptavidin) were collected and prepared for FACS. Target at the desired concentration was then added to the yeast and incubated for 2 h at 4 °C. After the 2 h incubation, yeasts were washed and incubated with a 1:1000 dilution of DyLight 488-conjugated anti-V5 and a 1:200 dilution of AlexaFluor 647-conjugated streptavidin for 20 min on ice. Yeasts were then washed and sorted by FACS in which the double positive population was collected. Yeasts selected after the fourth round were plated, and individual clones were picked for screening.

Individual clones were screened for binding to the target and for binding to streptavidin. Any nonspecific clones that bound streptavidin were discarded. DNA was isolated from the specific binding clones using the Zymoprep Yeast Plasmid Miniprep II kit and then sequenced.

Evaluation of Tn3 library enrichment

For evaluation of Tn3 library enrichment, both library groups from every round were incubated with 50 nM tetramer. Tetramers were formulated by incubating a 4.5:1 ratio of biotinylated MALT1 or EED to AlexaFluor 647-conjugated streptavidin for 15 min on ice. The tetramer was then added to induced yeast for each library group and incubated for 2 h at 4 °C. A 1:1000 dilution of DyLight 488-conjugated anti-V5 was also added to the mixture. For evaluating library enrichment, naïve libraries (round 0) and libraries from round 1 selections were pooled into the same groups as described for round 2 selections. Data were collected on CytoFLEX flow cytometer and analyzed using FlowJo software (FlowJo, LLC).

Yeast surface binding studies

For binding studies on yeast, binding clones isolated from the Tn3 libraries were grown at 30 °C in SDCAA liquid media (pH 4.5) for 2 days, followed by induction in SGCAA liquid media (pH 4.5) for 2 days at 20 °C. After induction, 1 × 10⁵ cells of Tn3-displaying yeast per well were transferred to a 96-well plate and incubated in PBE containing serial dilutions of recombinant biotinylated MALT1 or EED for 2 h at room temperature. Cells were then washed and stained with a 1:200 dilution of AlexaFluor 647-conjugated streptavidin and a 1:1000 dilution of DyLight 488-conjugated anti-V5 diluted in PBE for 15 min at 4 °C. Expression of the V5 tag was measured to confirm full-length expression of the Tn3 domain. After a final wash, data were collected on a CytoFLEX flow cytometer and analyzed using FlowJo software. Background-subtracted and normalized binding curves were fitted to a first-order binding model, and K_D values were determined using GraphPad Prism. Normalized data are presented as mean fluorescence intensity (MFI) for biotinylated target binding (AlexaFluor 647-conjugated streptavidin) divided by MFI for anti-V5 antibody binding (as detected DyLight 488-conjugated anti-V5 antibody). Studies were performed three times with similar results.

Binding of TRIM21–Tn3 fusion proteins to native MALT1 and EED

Native MALT-1-expressing HBL-1 cells and native EED-expressing Karpas 422 cells were lysed using Pierce IP Lysis Buffer (ThermoFisher Scientific) with the addition of Roche cOmplete Mini EDTA-free Protease Inhibitor Cocktail (MilliporeSigma) and Roche PhosSTOP tablets (MilliporeSigma) according to the manufacturer’s protocol. The lysate was centrifuged at 13,000g for 10 min, and the supernatant was transferred to a fresh tube. Protein concentration was measured using the Pierce BCA Protein Assay Kit following the manufacturer’s protocol, and absorbance was measured using a Biotek Synergy2 plate reader (BioTek Instruments, Inc).

Anti-FLAG M2 Magnetic Beads (MilliporeSigma) were washed with TBS (50 mM Tris–HCl [pH 8] and 150 mM NaCl) and then incubated with 15 μg of TRIM21–Tn3 variant for 2 h at 4 °C. The beads were then washed with Pierce IP Lysis Buffer. About 1 mg of either HBL-1 or Karpas 422 cell lysate was then added to the beads and diluted to a final concentration of 1 mg/ml in Pierce IP Lysis Buffer and incubated overnight at 4 °C. The following day, the beads were washed with Pierce IP Lysis Buffer adjusted to a final salt concentration of 300 mM. To elute bound protein, the beads were incubated for 20 min at room temperature with 60 μl of 3× FLAG peptide at a concentration of 100 μg/ml.

About 20 μl of eluted protein was then used to detect the presence of native antigen by immunoblotting. To detect the presence of TRIM21–Tn3, 5 μl of eluted protein was used for immunoblotting. The experiment was repeated two times with similar results.

Degradation of tGFP-tagged MALT1 and EED

HEK 293F cells were grown to 1.2 × 10⁶ cells/ml and diluted to 1.0 × 10⁶ cells/ml. About 2 ml of cells were then transferred into each well of a 6-well plate. Cells were transiently transfected with 1.33 μg total DNA (1:1 mass ratio of either pCMV6-AC-tGFP-MALT1 or pCMV6-AC-tGFP-EED and pOE encoding the appropriate TRIM21–Tn3 variant). The DNA and 2 μl of 293fectin (ThermoFisher Scientific) were independently diluted in 66.67 μl of Opti-MEM I, respectively, and incubated at room temperature for 5 min. The DNA mixture was then added to the 293fectin mixture and incubated at room temperature for an additional 25 min. Subsequently, the DNA/293fectin mixture was added to the 6-well plate containing HEK 293F cells and incubated at 37 °C and 5% CO₂ with rotation at 125 rpm for 72 h.

After 72 h, 150 μl from each well was transferred into each well of a 96-well plate, and the cells were pelleted (500g for 5 min) and resuspended in 20 μl of eBioscience Fixable Viability Dye eFluor780 (1:1000 dilution). Cells were then washed with PBSA, resuspended in 100 μl of PBSA, and data were collected on a CytoFLEX flow cytometer and analyzed using FlowJo software. Data represent biological triplicates (three separately transfected wells) of the background-subtracted tGFP MFI normalized to the average tGFP MFI of the control sample. tGFP MFI was calculated based on the full population of live cells (including both transfected and nontransfected cells). Statistical significance was determined by one-way ANOVA (Dunnett’s multiple-comparison test compared with the control sample). The experiment was repeated twice with similar results.

The remaining transfected cells were pelleted (500g for 5 min) and lysed for 20 min at 4 °C with 50 μl Pierce IP Lysis Buffer with the addition of Roche cOmplete Mini EDTA-free Protease Inhibitor Cocktail and Roche PhosSTOP tablets. The lysate was centrifuged at 13,000g for 10 min, and the supernatant was transferred to a fresh tube. Protein concentration was measured using the Pierce BCA Protein Assay Kit following the manufacturer’s protocol, and absorbance was measured using a Biotek Synergy2 plate reader. Lysate was saved for analysis by immunoblotting (20 μg of MALT1-tGFP transfected cells or 10 μg of EED-tGFP-expressing cells were used for immunoblotting). A representative biological replicate from each experiment was analyzed by immunoblotting.

Fluorescence microscopy and plate reader analysis

HEK 293F cells were transfected with 293fectin as previously described, except cells were transfected with a 1:1 mass ratio of either pCMV6-AC-tGFP-MALT1 or pCMV6-AC-tGFP-EED and pOE-mCherry encoding the appropriate TRIM21–Tn3 variant (1.33 μg total DNA).

After 72 h, 20× images were acquired with a Zeiss Axio Observer with AxioVision 5 software (Zeiss United States), and composite images were generated with ImageJ (96). Mean tGFP (488 ± 9 nm excitation; 510 ± 9 nm emission) and mCherry fluorescence (580 ± 20 nm excitation; 630 ± 20 nm emission) was measured using a BioTek SynergyMX plate reader (BioTek Instruments, Inc) that read a 3 × 3 grid distributed across the well surface area. Background-subtracted tGFP fluorescence of each well was normalized by the mCherry fluorescence and divided by the mean mCherry-normalized tGFP fluorescence of the appropriate control sample. Data represent biological triplicates (three separately transfected wells), and statistical significance was determined by a two-tailed unpaired Student’s t test, and the experiment was repeated twice with similar results.

Degradation assay with TRIM21–Tn3–Tn3 fusion proteins

HEK 293F cells were transfected with 293fectin as previously described, except cells were transfected with a 1:1 mass ratio of either pCMV6-AC-tGFP-MALT1 or pCMV6-AC-tGFP-EED and pOE encoding the appropriate TRIM21–Tn3 or TRIM21–Tn3–Tn3 fusion (1.33 μg total DNA).

After 72 h, 150 μl from each well was transferred into each well of a 96-well plate, and the cells were pelleted (500g for 5 min) and resuspended in 20 μl of eBioscience Fixable Viability Dye eFluor780 (1:1000 dilution). Cells were then washed with PBSA, resuspended in 100 μl of PBSA, and data were collected on a CytoFLEX flow cytometer and analyzed using FlowJo software. Data represent biological triplicates (three separately transfected wells) of the background-subtracted tGFP MFI normalized to the average tGFP MFI of the control sample. Statistical significance was determined by one-way ANOVA with a Tukey post hoc test. The experiment was repeated twice with similar results.

Degradation assay with other E3 ligases

HEK 293F cells were transfected with 293fectin as previously described, except cells were transfected with a 1:1 mass ratio of either pCMV6-AC-tGFP-MALT1 or pCMV6-AC-tGFP-EED and pOE encoding TRIM21–Tn3, VHL–Tn3, Tn3–CHIP, or Tn3–SPOP fusion proteins (1.33 μg total DNA).

After 72 h, 150 μl from each well was transferred into each well of a 96-well plate, and the cells were pelleted (500g for 5 min) and resuspended in 20 μl of eBioscience Fixable Viability Dye eFluor780 (1:1000 dilution). Cells were then washed with PBSA, resuspended in 100 μl of PBSA, and data were collected on a CytoFLEX flow cytometer and analyzed using FlowJo software. For each E3 ligase, the background-subtracted tGFP MFI was normalized to the average background-subtracted tGFP MFI of a control nontarget-specific Tn3 domain containing the same E3 ligase. BioPROTACs with the EG9 Tn3 domain were used as controls for MALT1 degradation, and bioPROTACs with the ME2 Tn3 domain were used as controls for EED degradation. Data represent biological triplicates (three separately transfected wells). Statistical significance for each fusion protein (target specific versus control Tn3 domain) was determined by a two-tailed unpaired Student’s t test. The experiment was repeated twice with similar results.

Degradation assay with inactive TRIM21–Tn3 variants

HEK 293F cells were transfected with 293fectin as previously described, except cells were transfected with a 1:1 mass ratio of either pCMV6-AC-tGFP-MALT1 or pCMV6-AC-tGFP-EED and pOE-mCherry-T2A-TRIM21-Tn3-ME2 or pOE-mCherry-T2A-TRIM21-Tn3-EG9 (1.33 μg total DNA).

After 72 h, 150 μl from each well was transferred into each well of a 96-well plate, and the cells were pelleted (500g for 5 min) and resuspended in 20 μl of eBioscience Fixable Viability Dye eFluor780 (1:1000 dilution). Cells were then washed with PBSA, resuspended in 100 μl of PBSA, and data were collected on an Attune NxT (ThermoFisher Scientific) and analyzed using FlowJo software. Data represent biological triplicates (three separately transfected wells), and the background-subtracted tGFP MFI was divided by the background-subtracted mCherry MFI for each condition. The data were further normalized to the average mCherry-normalized tGFP MFI of the control sample. Statistical significance was determined by one-way ANOVA with a Tukey post hoc test.

Plate reader analysis was performed in the same manner as previously described. For each condition, the background-subtracted tGFP MFI was normalized by the background-subtracted mCherry MFI and then divided by the average mCherry-normalized tGFP MFI of the appropriate control sample. Data represent biological triplicates (three separately transfected wells), and statistical significance was determined by one-way ANOVA with a Tukey post hoc test. The experiment was repeated twice with similar results.

Degradation of MALT1 and EED without a tGFP tag

HEK 293F cells were transfected with 293fectin as previously described, except cells were transfected with a 1:1 mass ratio of either pCMV6-AC-MALT1 or pCMV6-AC-EED and pOE encoding the appropriate TRIM21–Tn3 variant (1.33 μg total DNA).

After 72 h, the transfected cells were pelleted (500g for 5 min) and lysed for 20 min at 4 °C with 100 μl Pierce IP Lysis Buffer with the addition of Roche cOmplete Mini EDTA-free Protease Inhibitor Cocktail and Roche PhosSTOP tablets. The lysate was centrifuged at 13,000g for 10 min, and the supernatant was transferred to a fresh tube. Protein concentration was measured using the Pierce BCA Protein Assay Kit following the manufacturer’s protocol, and absorbance was measured using a Biotek Synergy2 plate reader. Lysate was saved for analysis by immunoblotting (20 μg of MALT1-transfected cells or 10 μg of EED-expressing cells were used for immunoblotting). The experiment was repeated twice with similar results.

Degradation of native MALT1 in HEK 293F cells

HEK 293F cells were transfected with 293fectin as previously described, except cells were transfected with only pOE encoding the appropriate TRIM21–Tn3 variant (1.33 μg total DNA). Cells were also transfected with an empty pOE plasmid as a control.

After 72 h, the transfected cells were pelleted (500g for 5 min) and lysed for 20 min at 4 °C with 100 μl Pierce IP Lysis Buffer with the addition of Roche cOmplete Mini EDTA-free Protease Inhibitor Cocktail and Roche PhosSTOP tablets. The lysate was centrifuged at 13,000g for 10 min, and the supernatant was transferred to a fresh tube. Protein concentration was measured using the Pierce BCA Protein Assay Kit following the manufacturer’s protocol, and absorbance was measured using a Biotek Synergy2 plate reader. Lysate was saved for analysis by immunoblotting (75 μg of lysate was used for immunoblotting). The experiment was repeated twice with similar results.

Immunoblotting

Cell lysates were prepared as described in the previous sections, and the amount of lysate used for immunoblotting is described in each appropriate section of the methods. For all blots, an equivalent amount of total protein was added to each lane, and β-actin was used to confirm equivalent loading.

Proteins were transferred onto poly(vinylidene fluoride) membranes using the iBlot 2 Dry Blotting System (ThermoFisher Scientific) and blocked for 1 h at room temperature with 5% Blotting Grade Blocker Non-Fat Dry Milk (Bio-Rad) in PBST (1× phosphate-buffered saline with 0.01% Tween-20). Blots were probed with the appropriate primary antibody for 1 h at room temperature in 5% Blotting Grade Blocker Non Fat Dry Milk (Bio-Rad) in PBST. Blots were then washed with PBST and probed with the appropriate HRP-conjugated secondary antibody for 1 h at room temperature in 5% Blotting Grade Blocker Nonfat Dry Milk in PBST. Primary antibodies used were rabbit antihuman MALT1 (EP603Y, Abcam, catalog no.: ab33921, 1:10,000 dilution), sheep antihuman/mouse EED (R&D Systems, catalog no: AF5827, 1:400 dilution), rabbit anti-β-actin (13E5, Cell Signaling Technology, catalog no.: 4970, 1:2000 dilution), HRP-conjugated antiubiquitin (BioLegend, catalog no.: 646304, 1:1000 dilution), and HRP-conjugated anti-FLAG (M2, MilliporeSigma, catalog no.: A8592, 1:1000 dilution) to detect TRIM21–Tn3. Secondary antibodies used were HRP-conjugated anti-rabbit IgG (Abcam, catalog no.: ab97051, 1:2000 dilution) and HRP-conjugated antisheep IgG (R&D Systems, catalog no.: HAF016, 1:1000 dilution). SuperSignal West Pico PLUS Chemiluminescent Substrate (ThermoFisher Scientific) was used for chemiluminescence detection, and images were taken with an ImageQuant LAS 4000 Mini.

Densitometry analysis of immunoblot bands was performed using ImageJ’s gel analyzer. For each treatment, the MALT1 or EED band intensity was normalized to the band intensity of β-actin. Data were then divided by the normalized band intensity of the control sample for each experiment. Immunoblot analyses were performed at least two times with similar results.

If necessary, blots were stripped with Restore Western Blot Stripping Buffer (ThermoFisher Scientific) and reblocked before reprobing.

Data availability

All data are contained within the article.

Supporting information

This article contains supporting information (97, 98, 99).

Conflict of interest

J.H. and R.G. are employees and hold shares in AstraZeneca. All the authors declare that they have no conflicts of interest with the contents of this article.

Reviewed by members of the JBC Editorial Board. Edited by George DeMartino

Supporting information

Supplemental Fig. S1

Structural model of TRIM21-Tn3 fusion proteins. Structural model of monomeric TRIM21-Tn3 fusion generated in PyMOL from a composite of wildtype Tn3 (PDB ID 1TEN) (60) and the AlphaFold predicted structure of TRIM21ΔPRYSPRY (97, 98).

Supplemental Fig. S2

TRIM21-Tn3 fusion proteins retain ubiquitination activity. Ubiquitination of recombinant TRIM21 or TRIM21-Tn3 variants was assessed in the presence of E1 (Ube1), E2 (Ube2W and Ube2N/2V2), ubiquitin (Ub), and ATP. Reactions were carried out at 37 °C for 1 h in the presence (+) or absence (−) of pathway components, as indicated. SDS-PAGE analysis of ubiquitination reactions (A, C, and E) and immunoblots of ubiquitination reactions probed with an anti-ubiquitin antibody (B, D, and F) are shown. An equivalent amount of total protein was added to each lane. Data are representative of two replicate experiments.

Supplemental Fig. S3

Tn3 library design and evolution against MALT1 and EED.A, schematic depicting the yeast surface display library format, consisting of the third FN3 domain of human tenascin-C (Tn3, PDB ID 1TEN) (60) with a C-terminal V5 tag followed by the yeast a-agglutinin subunit Aga2p. B, design of yeast-displayed Tn3 libraries. Crystal structures of the wildtype Tn3 are shown (PDB ID 1TEN) (60), although an engineered Tn3 with enhanced thermal stability was used as the basis for library designs (63). Amino acid positions that were selected for mutagenesis are colored and the corresponding side chains are shown. Libraries generated included 4 loop (left), 2 concave (middle), and 1 flat (right) library. Clones that were isolated from each class of libraries are listed below the respective schematic. Clones that bound and degraded the target POI in TRIM21-Tn3 format are indicated in bold font. C and D, yeast surface binding titrations of recombinant MALT1 (C) or EED (D) against individual clones isolated from Tn3 library selections, as measured by flow cytometry analysis. Signal was normalized to V5 expression for each Tn3 variant.

Supplemental Fig. S4

Structural model of TRIM21-Tn3 fusion proteins overlaid with Tn3 library designs. Structural model of monomeric TRIM21-Tn3 fusion proteins generated in PyMOL from a composite of wildtype Tn3 (PDB ID 1TEN) (60) and the AlphaFold predicted structure of TRIM21ΔPRYSPRY (97, 98), as shown in Fig. S1. The Tn3 domains show the loop library (A), concave library (B), and flat library designs (C), as illustrated in Fig. S3B. Amino acid positions that were selected for mutagenesis are colored and the corresponding side chains are shown.

Supplemental Fig. S5

Immunoblot analysis and quantification of MALT1 and EED degradation.A and B, immunoblot analysis of lysates from HEK 293F cells co-transfected with a plasmid encoding MALT1-tGFP (A) or EED-tGFP (B) as well as a plasmid encoding the indicated TRIM21-Tn3 variant. Cells were harvested and lysed 72 h after transfection. Blots were probed with an anti-MALT1 antibody (A) or an anti-EED antibody (B) as well as an anti-Flag antibody to detect TRIM21-Tn3. An equivalent amount of total protein was added to each lane, and blots were also probed with an anti-β-actin antibody to confirm equivalent loading. Bands for MALT1-tGFP and MALT1 are present in panel A because HEK 293F cells natively express MALT1 (99, proteinatlas.org). Data represent biological replicates from the sample sets shown in Figs. 4, B and C and 5. C, relative quantitation of total endogenous MALT1 levels by densitometry using ImageJ software across biological replicates (Figs. 5A and S5A). Band intensity was normalized to the band intensity of β-actin and then divided by the normalized band intensity of the control sample for each experiment. Statistical significance was determined by two-tailed paired Student’s t-test. Statistical data are shown in Table S4. ∗p ≤ 0.05.

Supplemental Fig. S6

Plate reader analysis of MALT1 and EED degradation. HEK 293F cells were co-transfected with a plasmid encoding MALT1-tGFP (A) or EED-tGFP (B) as well as a plasmid encoding an mCherry-tag fused to the indicated TRIM21-Tn3 variant, separated by a T2A peptide. Inactive TRIM21-Tn3 variants contain several mutations in the TRIM21ΔPRYSPRY region that prevent E2 recruitment and subsequent ubiquitination. After 72 h, mean tGFP (488 ± 9 nm excitation; 510 ± 9 nm emission) and mCherry fluorescence (580 ± 20 nm excitation; 630 ± 20 nm emission) was measured using a BioTek SynergyMX plate reader (BioTek Instruments, Inc) that analyzed a 3 × 3 grid distributed across the well surface area. Background-subtracted tGFP fluorescence of each well was normalized by the mCherry fluorescence and divided by the mean mCherry normalized tGFP fluorescence of the appropriate control sample. Data represent mean ± SD of biological triplicates. Data in panels A and B correspond to the samples analyzed by flow cytometry in Fig. 7, A and B. Statistical significance was determined by one-way ANOVA with a Tukey post hoc test. Statistical data are shown in Table S4. ∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, ∗∗∗∗p ≤ 0.0001.

Supplemental Fig. S7

Engineering TRIM21-Tn3 fusion proteins with tandem Tn3 domains.A, linear representation of TRIM21-Tn3-Tn3 fusion proteins. Tn3 domains were fused to the C-terminus of TRIM21ΔPRYSPRY (residues 1–286) via a flexible 15-amino acid (Gly₄Ser)₃ linker and Tn3 domains were also separated by a (Gly₄Ser)₃ linker. B, schematic of dimeric TRIM21-Tn3-Tn3 fusion proteins. C and D, flow cytometry analysis 72 h after co-transfection of HEK 293F cells with a plasmid encoding MALT1-tGFP (C) or EED-tGFP (D) as well as a plasmid encoding and the indicated TRIM21-Tn3 or TRIM21-Tn3-Tn3 variant. The background-subtracted tGFP MFI for each sample was normalized to the average background-subtracted tGFP MFI of the appropriate control sample. Data represent mean ± SD of biological triplicates and statistical significance was determined by one-way ANOVA with a Tukey post hoc test. For panel C, statistical significance is shown compared to TRIM21-ME2 (purple) and TRIM21-MH4 (light blue). For panel D, statistical significance is shown compared to TRIM21-EG9 (blue) and TRIM21-EH8 (red). Statistical data are shown in Table S4. ∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, ∗∗∗∗p ≤ 0.0001.

Supplemental Fig. S8

Comparison of bioPROTACs that utilize different E3 ligases.A, linear representations of VHL-Tn3, Tn3-CHIP, and Tn3-SPOP fusion proteins. Tn3 domains were fused to the C-terminus of VHL (residues 1–213) via a Gly₄Ser linker, whereas Tn3 domains were fused to the N-terminus of truncated CHIP (residues 128–303) and truncated SPOP (residues 167–374) via a Gly₄Ser linker. B and C, flow cytometry analysis 72 h after co-transfection of HEK 293F cells with a plasmid encoding MALT1-tGFP (B) or EED-tGFP (C) as well as a plasmid encoding the indicated bioPROTAC. For each E3 ligase, the background-subtracted tGFP MFI was normalized to the average background-subtracted tGFP MFI of a control nontarget specific Tn3 domain containing the same E3 ligase. Data represent mean ± SD of biological triplicates. Statistical significance for each fusion protein (target specific versus control Tn3 domain) was determined by a two-tailed unpaired Student’s t test. Statistical data are shown in Table S4. ∗p ≤ 0.05, ∗∗p ≤ 0.01, ∗∗∗p ≤ 0.001, ∗∗∗∗p ≤ 0.0001.

Supplemental Fig. S9

Immunoblot analysis of untagged and endogenous POI.A and B, immunoblot analysis of lysates from HEK 293F cells co-transfected with a plasmid encoding either MALT1 (A) or EED (B) and a plasmid encoding the indicated TRIM21-Tn3 variant. C and D, replicate immunoblot analyses of lysates from HEK 293F cells transfected with a plasmid encoding the indicated TRIM21-Tn3 variant or an empty plasmid (“Empty”). Cells were harvested and lysed 72 h after transfection. Blots were probed with an anti-MALT1 antibody (A, C and D) or an anti-EED antibody (B), and an anti-Flag antibody was used to detect TRIM21-Tn3. An equivalent amount of total protein was added to each lane, and blots were probed with an anti-b-actin antibody to confirm equivalent loading. Data represent biological replicates from the sample sets shown in Fig. 8.