Crystal structure of the mucosa-associated lymphoid tissue lymphoma translocation 1 (MALT1) paracaspase region

Abstract

The mucosa-associated lymphoid tissue lymphoma translocation 1 (MALT1) paracaspase, a key component of the Carma1/Bcl10/MALT1 signalosome, is critical for NF-κB signaling in multiple contexts. MALT1 is thought to function as a scaffold and protease to promote signaling; however, the biochemical and structural basis of paracaspase action remains largely unknown. Here we report the 1.75-Å resolution crystal structure of the MALT1 paracaspase region, which contains the paracaspase domain and an ensuing Ig-like domain. The paracaspase and the Ig domains appear as a single folding unit and interact with each other through extensive van der Waals contacts and hydrogen bonds. The paracaspase domain adopts a fold that is nearly identical to that of classic caspases and homodimerizes similarly to form an active protease. Unlike caspases, the active and mature form of the paracaspase domain remains a single uncleaved polypeptide and specifically recognizes the bound peptide inhibitor Val-Arg-Pro-Arg. In particular, the carboxyl-terminal amino acid Arg of the inhibitor is coordinated by three highly conserved acidic residues. This structure serves as an important framework for deciphering the function and mechanism of paracaspases exemplified by MALT1.

The transcription factor NF-κB is a key constituent of all cell types and is activated by various receptors to regulate survival, proliferation, migration, and differentiation (1). In particular, NF-κB functions early in the development and maintenance of innate and adaptive immune systems and execution of the immune response. Although caspases, cysteine proteases that cleave substrate proteins after aspartate residues, are widely known as executioners of programmed cell death or apoptosis (2), a subset and related members also activate NF-κB to promote lymphocyte proliferation and inflammation. One such caspase-like family member, the mucosa-associated lymphoid tissue lymphoma translocation 1 (MALT1) paracaspase, was identified through weak sequence homology to caspases (3) and was subsequently found to play an important role in lymphocyte activation (4) and disease progression in MALT lymphomas (5).

Upon antigen receptor stimulation, the MALT1 paracaspase and Bcl10 assemble into the Carma1/Bcl10/MALT1 (CBM) signalosome to activate NF-κB in the adaptive immune system. Specifically during T-cell receptor signaling, the CBM signalosome is thought to oligomerize MALT1 and its associated ubiquitin ligase tumor necrosis factor receptor-associated factor 6 (TRAF6) or TRAF2 (6, 7), which in turn facilitates K63-linked polyubiquitylation of multiple proteins including the regulatory γ-subunit of the IκB kinase (IKK) complex (6, 7), TRAF6 itself (7), Bcl10 (8), and MALT1 (9). Poly ubiquitylation of these proteins ultimately leads to the recruitment of transforming growth factor β-activated kinase 1 (TAK1), TAK1 binding protein (TAB), and the IKK complex to lipid rafts where the IKKβ-subunit is phosphorylated and activated. In the canonical NF-κB pathway, the activated IKK complex phosphorylates IκB, enabling proteasome-mediated degradation of IκB and subsequent translocation of NF-κB into the nucleus where it acts as a transcriptional activator.

Within the CBM signalosome, MALT1 is an essential component and appears to transmit signals predominantly by recruiting and oligomerizing TRAF6 or other E3 ubiquitin ligases. In addition to its scaffolding function, MALT1 also possesses arginine-specific proteolytic activity that modulates NF-κB and possibly other signaling pathways. For example, cleavage of the negative regulator A20 by MALT1 appears to disrupt the deubiquitylating enzyme (DUB) activity of A20, leading to enhancement of NF-κB signaling (10). Cleavage of Bcl10 by MALT1, on the other hand, was reported to exert integrin-mediated cell adhesive properties independent of NF-κB (11). More recently, MALT1-mediated proteolysis of the DUB cylindromatosis (CYLD) has been implicated in JNK activation during T-cell receptor signaling (12). Despite these tantalizing insights, the overall contribution of MALT1 protease activity to NF-κB and related pathways remains largely enigmatic. Nevertheless, targeted inhibition of MALT1 proteolytic activity may contribute to the therapy of various diseases where NF-κB and related pathways play a major role. In support of this view, MALT1 protease inhibition has been shown to dramatically inhibit growth and promote apoptosis of the highly aggressive activated B cell-like form of diffuse large B cell lymphomas (13, 14).

Although MALT1 was initially linked to immune system function, it has emerged as a critical component of NF-κB signaling in multiple contexts. For example, MALT1 and Bcl10 have been shown to couple with different Carma1 homologues to transduce NF-κB signaling downstream of multiple receptor types (15). It is unclear, however, how MALT1 integrates its scaffolding and protease activities through its C-terminal region to modulate proper cellular responses in these different contexts. Sequence analysis of MALT1 initially identified a region with loose homology to the caspase fold (3) followed by an Ig-like domain critical for NF-κB activation (16). Subsequent investigations have failed to identify significant similarity between MALT1 and classic caspases in terms of substrate recognition, activity and specificity, and autoproteolytic maturation. Although the Ig domain had been implicated in modulating protease activity and recruiting components required for proper ubiquitin ligase activity (7, 16), the underlying mechanism remains unknown.

To gain mechanistic insight into MALT1 function, we determined the crystal structure of an inhibitor-bound MALT1 fragment containing the paracaspase domain and a critical Ig domain. Our study reveals a number of findings that may serve as a basis for understanding the mechanism of MALT1 function in the CBM complex.

Results

Crystallization and Structure Determination.

Full-length human MALT1 (residues 1–824) contains a death domain (DD) and two Ig domains at the N-terminal region, and a predicted caspase-like domain (paracaspase domain) and Ig fold at the C-terminal half (Fig. 1A). The N-terminal region is thought to interact with Bcl10 and assemble into the CBM signalosome, whereas the C-terminal half is believed to harbor scaffolding and protease activities (4). We attempted to bacterially overexpress various MALT1 fragments encompassing the C-terminal half. The paracaspase domain in isolation (residues 339–561) was found to be extremely prone to aggregation and precipitation, suggesting that other domains might be required for stabilization of the paracaspase domain. Using a combination of sequence analysis and limited proteolysis of longer MALT1 fragments, we identified a 45-kDa structural core encompassing the paracaspase domain and the Ig fold (residues 339–719). This MALT1 fragment (hereafter referred to as MALT1-C) was expressed at a high level, purified to homogeneity, and exhibited excellent solution properties. Despite extensive effort, we failed to generate crystals of this protein fragment. Taking lessons learned from working with caspases, we incubated MALT1-C with the peptide inhibitor benzoxycarbonyl-Val-Arg-Pro-Arg-fluoromethylketone (z-VRPR-fmk) (11, 17). The treated protein assembled into a stable homodimer that was readily resolved from the untreated MALT1-C by gel filtration (Fig. S1). The inhibitor-bound MALT1 fragment was crystallized and the structure was determined at 1.75-Å resolution using multiwavelength anomalous dispersion (MAD) (Table S1).

Fig. 1.

Overall structure of the MALT1 paracaspase region bound to a peptide inhibitor. (A) Domain organization of full-length MALT1. The crystallized MALT1 fragment (MALT1-C) described in this study is indicated by a bracket and labeled “MALT1 paracaspase region.” (B) Ribbon representation of the MALT1-C homodimer. Each protomer consists of a paracaspase domain (blue) and an adjacent Ig domain (orange). The covalently bound peptide inhibitor VRPR is colored red and the intersubunit linker in the caspase-like fold, which is normally cleaved in mature caspases, is colored yellow. The intervening peptide sequences between the paracaspase domain and the Ig fold are disordered and are represented here as gray dotted lines. The active site loops L1–L4 are labeled.

Overall Structure of the MALT1 Paracaspase Region.

There are two molecules of MALT1-C in each asymmetric unit of the crystal lattice. As predicted from our biochemical analysis, these two MALT1 molecules form a symmetric homodimer, with each molecule bound to a peptide inhibitor (Fig. 1B). Each MALT1-C protein adopts a cone-shaped structure, with the Ig fold as the sharp tip and the paracaspase domain as the blunt bottom of the cone. Together, the MALT1-C homodimer resembles a prolate spheroid, measuring 160 Å in length and 60 Å in diameter (Fig. 1B and Fig. S2). Confirming previous conjecture, the paracaspase domain indeed adopts a caspase-like fold that closely resembles that of canonical caspases. The Ig fold packs against the β2 strand and the α1 and α2 helices of the caspase-like fold, resulting in the burial of approximately 1,348 Å² surface area. The paracaspase-Ig assembly appears to be a single folding unit glued together by a combination of electrostatic and hydrophobic interactions (see below). The intervening loop between the paracaspase domain and Ig domain, which extends from strand β6 of the caspase-like fold to helix α1 of the Ig fold, is likely flexible in solution as it is partially disordered in the crystal structure (Fig. 1B).

Despite a relatively low sequence identity of 17% between the MALT1 paracaspase domain and a representative caspase such as caspase 9, the paracaspase fold is virtually identical to that of all known caspases (Fig. 2A). Compared to a single inhibitor-bound protomer of the caspase-9 dimer (18), this fold can be superimposed with an rmsd of 1.9 Å over 194 C^α atoms (Fig. S3). Briefly, the paracaspase domain contains a central β-sheet of six β-strands, which is surrounded by two α-helices (α2 and α3) on one side and three α-helices (α1, α4, and α5) on the other side (Fig. 1B and Fig. S2). Two inhibitor-bound MALT1-C molecules form an antiparallel homodimer through their respective β6 strands of the paracaspase domains. This overall arrangement is identical to that of all known inhibitor-bound and activated caspase homodimers (Fig. 2A). This observation, along with the finding that purified MALT1-C forms a stable homodimer in the presence of an irreversible inhibitor (Fig. S1), suggest that MALT1 exists as a monomer in solution and, similar to the initiator caspases, requires dimerization for proteolytic activity.

Fig. 2.

Structural comparison of the MALT1 paracaspase domain with caspase 9. (A) Ribbon comparison of the MALT1 paracaspase domain dimer (blue) with the caspase-9 dimer (gray/magenta) (18). The peptide inhibitors bound to MALT1and caspase 9 are shown as red and brown sticks, respectively. The active site loops L1–L4 are indicated and the unprocessed intersubunit linker of MALT1 containing L2 and L2′ is highlighted in yellow. Loop-bundle interactions present in the caspase-9 dimer are indicated by the box. Note the absence of loop-bundle interactions in the MALT1 paracaspase domain dimer. (B) A close-up view of the active site loops in MALT1 (blue) and caspase 9 (gray). The catalytic residue Cys464 of MALT1 is highlighted in magenta.

Beyond these structural similarities, the MALT1 paracaspase region exhibits a number of unique structural features that may be important for its function. The most prominent feature is the presence of the Ig domain, which contains a β-sandwich of seven β-strands (Fig. 1B). An exhaustive search of the protein data bank using DALI (19) identified a number of structures that are similar to the MALT1-C Ig domain, exemplified by the γ-adaptin ear domains of the adaptor protein 1 (AP1) (20) and Golgi-localizing, γ-adaptin ear domain homology, Arf-binding proteins (GGA) (21) complexes (Fig. S4). The AP1 and GGA complexes participate in vesicular trafficking between the Golgi and endosomal compartment, and intriguingly the Ig ear domains in each case appear to function by recruiting accessory proteins. Similar to the AP1 and GGA complexes, the Ig domain of MALT1-C might also serve to recruit additional proteins. Consistent with this notion, the Ig fold of MALT1-C has been proposed to interact with the E2 ubiquitin-conjugating enzyme unit Ubc13/Mms2 (6, 16) and the E3 ubiquitin ligase TRAF6 (7).

Sequence Conservation.

The primary amino acid sequences of the MALT1 paracaspase region are highly conserved from fish to human (Fig. S5). To gain insights into the combined paracaspase-Ig structural assembly, we mapped the invariant and conserved amino acids of MALT1-C across several species onto the surface of the structure (Fig. S6). This analysis reveals a number of highly conserved, surface-exposed patches, which might be critical for MALT1 function. One such surface patch is the dimerization interface surrounding the β6 strand of the paracaspase domain; this region is likely essential for dimer formation and subsequent acquisition of protease activity. Another highly conserved surface patch is the substrate-binding groove, suggesting conservation of substrate recognition across multiple species. Other conserved patches include a potential TRAF6 binding site (7) and an ubiquitylation site (9) which may be involved in the recruitment of the IKK complex. Notably, invariant surface patches are present on the opposite side of the substrate recognition pocket. Because this region contains surface grooves or potential binding pockets, it is tempting to speculate that this region might function as a scaffold to recruit auxiliary proteins.

Active Site Conformation.

As described above, part of the MALT1 paracaspase-Ig structure shares the same overall fold as caspases and this unit appears to dimerize similarly as classic caspases to form functional active sites (Figs. 1B and 2A). Given these shared structural features, it is unsurprising that MALT1 uses the same active site loop arrangement to engage substrate (Fig. 2B). Within the substrate-binding groove, the L3 loop serves as the base that sits right below the substrate, whereas the L1 and L4 loops form two sides of the groove to sandwich the substrate. The L2 loop contains the catalytic residue Cys464 and requires proper positioning for efficient catalytic activity.

For all caspases studied to date, the intersubunit linker that contains the L2 and L2′ loops is autocatalytically cleaved upon maturation. Cleavage of this linker sequence releases the L2 and L2′ loops, allowing L2′ of one caspase protomer to engage the L2 and L4 loops of the neighboring protomer (22) (Fig. 2A). These interactions stabilize the conformation of the neighboring active site, which gives rise to normal catalytic activity. Strikingly, the intersubunit linker in the MALT1-C structure remains intact (Figs. 1B and 2A). Consequently the L2 and L2′ loops are locked toward the center of the homodimer, rather than toward L4, and are unable to assemble the loop-bundle interactions seen in all other caspases (Fig. 2A). Consistent with this observation, previous studies have shown that MALT1, under a variety of conditions, does not undergo autocatalytic processing in mammalian cells (3). This unique feature indicates that MALT1 functions quite differently from classic caspases and suggests that it might not acquire high catalytic activity associated with the loop-bundle interactions.

Substrate Recognition.

Another prominent feature of MALT1 that differs from classic caspases is substrate specificity. MALT1 has been shown to cleave A20, Bcl10, and CYLD at specific sites that contain Arg, rather than Asp, at the P1 position (10–12). In particular, proteolytic cleavage of Bcl-10 after Arg228 of the sequence Leu225-Arg226-Ser227-Arg228 is required for T-cell receptor-induced adhesion to fibronection (11). Recombinant WT MALT1, but not the catalytic mutant C464A, efficiently cleaved the substrate Leu-Arg-Ser-Arg-7 amino-4-methylcoumarin (LRSR-AMC) (11). This cleavage sequence of Bcl-10, LRSR, closely resembles the peptide inhibitor VRPR used for crystallization.

Analysis of the VRPR-bound MALT1-C structure provides a satisfying explanation to the reported substrate specificity in Bcl-10. Three of the four amino acids in VRPR are specifically recognized by residues in the substrate-binding groove of MALT1. The positively charged P1 residue Arg in the inhibitor VRPR makes a total of 10 hydrogen bonds, six of which are directed to three negatively charged residues in the S1 pocket: Asp365 from L1, Asp462 from L2, and Glu500 from L3 (Fig. 3A). Each of the three acidic residues uses its carboxylate side chain to accept a pair of charge-stabilized hydrogen bonds from the guanidinium group of P1-Arg. These polar contacts to the P1-Arg side chain are buttressed by four additional main-chain hydrogen bonds mediated by Pro362 in L1, Cys464 in L2, Ala498 in L3, and Gly416 in β3a. The P3-Arg side chain from VRPR donates three hydrogen bonds, one to the side-chain carboxylate of Glu500 and two to the side chain of Gln502 through an intermediary water molecule (Fig. 3A). The main-chain groups of Glu500 also hydrogen bond with the amide nitrogen atoms of P3-Arg and P4-Val as well as the main-chain carbonyl oxygen of P3-Arg (Fig. 3A). The nonpolar side chain of the P4 residue Val is nicely accommodated by the hydrophobic S4 pocket formed by residues in L1 (Phe499 and Ile501), L4 (Leu541, Thr542, and the aliphatic portion of Lys545 side chain), and the α4 helix (Ile510) (Fig. 3B).

Fig. 3.

Specific recognition of the peptide inhibitor VRPR by MALT1. (A) A stereoview of specific VRPR P1 and P3 recognition by MALT1 (blue). The inhibitor VRPR-fmk (in red) is covalently bound to Cys464 in the substrate-binding groove of MALT1. Recognition of P1-Arg and P3-Arg is mediated mainly by specific hydrogen bonds. (B) A stereoview of P4 recognition by MALT1 (blue). P4-Val is accommodated by the hydrophobic residues in the S4 pocket. (C) A stereoview of specific benzoxycarbonyl-Glu-Val-Asp-dichlorobenzylmethylketone (orange) P1 and P3 recognition by caspase 9 (gray) (18). The catalytic residue Cys285 is indicated and hydrogen bonds are represented as dashed magenta lines. (D) A stereoview of the hydrophobic residues lining the S4 pocket in caspase 9.

Importantly, all residues involved in these interactions are highly conserved in MALT1 across several species (Fig. S5). The three acidic amino acids that recognize the P1-Arg side chain are invariant in MALT1 orthologs, whereas the hydrophobic residues that form the S4 pocket are highly conserved. This analysis suggests a degree of conserved substrate specificity in diverse organisms, with Arg invariant at the P1 position and preference for a hydrophobic residue at P4. Supporting this analysis, the P1 and P4 residues in CYLD are Arg and Phe, respectively (12). Although the P3-Arg residue is nicely accommodated in the S3 pocket of the MALT1-C structure, specific recognition is not as stringent as observed for the P1 and P4 residues, suggesting a greater degree of variation at this position. This observation is consistent with the cleavage specificity for A20 and CYLD, with Ala and Met as the P3 residue, respectively (10, 12). The only residue that does not have a well-defined mode of recognition is the P2 residue Pro, which is located in proximity to Glu497 and Phe499. Glu497 is invariant and Phe499 is preserved in all but two organisms where it is replaced by another aromatic residue Tyr. It is possible that a polar residue such as Ser, Thr, Asn, Gln, or His at the P2 position may allow additional hydrogen bonds to be made with Glu497.

Comparison of Substrate Recognition with Caspases.

In contrast to MALT1, all classic caspases favor Asp as the P1 residue. Such substrate specificity is similarly safeguarded by corresponding sequence and structural conservation in the active sites of classic caspases. For example, whereas MALT1 uses acidic residues Asp365 from L1 and Glu500 from L3 to select for P1-Arg, caspase 9 employs basic residues at the same locations (Arg179 from L1 and Arg341 from L3) to selectively engage the side chain of P1-Asp (18) (Fig. 3C). Therefore, the caspase fold utilizes different residues at the same active site locations to modulate cleavage specificity for MALT1. On the other hand, both MALT1 and caspase-9 favor a hydrophobic residue at the P4 position. Accordingly, accommodation of P4-Val by MALT1 closely resembles the recognition of a hydrophobic P4 residue by the S4 pocket of caspase 9 (Fig. 3D).

The Dimerization Interface.

The dimerization interface of MALT1-C is mediated by two structural elements, the β6 strand and the α5 helix (Figs. 1B and 4A). The interactions mediated by the antiparallel β6 strands, involving four main-chain hydrogen bonds, are similar to what has been observed for all other caspases (Fig. 4B). At this interface, the main-chain carbonyl oxygen and amide nitrogen atoms of Ile550 in strand β6 of one protomer make a pair of hydrogen bonds to the main-chain amide nitrogen and carbonyl oxygen atoms of Ser552 in strand β6 of the adjacent protomer (Fig. 4B). These highly conserved contacts assemble a contiguous β-sheet that extends from one MALT1 protomer into the other. In a notable departure from all other caspases, the interactions mediated by the α5 helices, involving eight hydrogen bonds, are unique to the MALT1 paracaspase domain (Fig. 4C). These hydrogen bonds are solely mediated by side-chain interactions. The carboxylate side chains of Asp530 and Glu534 in one protomer each accept two hydrogen bonds, from Lys524/Thr526 and from Lys524/Lys557, respectively, of the adjacent protomer (Fig. 4C). These hydrogen bonds are stabilized by the complementary charges of Asp530/Glu534 in one protomer and Lys524/Lys557 in the adjacent protomer. Thus, at the α5 helix interface, the side chains of five amino acids mediate specific, intermolecular hydrogen bonds, which presumably contribute significantly to dimerization specificity. Among these five residues, Lys524, Thr526, Asp530, and Glu534 are invariant among MALT1 orthologs (Fig. S5). Lys557, which resides outside the α5 helix, is maintained in all but two orthologs where it is replaced by a conserved Arg residue. Given the preservation of key residues in both the β6 strand and the α5 helix elements, the observed homodimeric interface is likely conserved in all MALT1 orthologs.

Fig. 4.

Features of the homodimerization interface in MALT1. (A) The homodimerization interface consists of two structural elements from each protomer: the β6 strand and the α5 helix. The two MALT1 protomers are colored blue and green. Hydrogen bonds are highlighted as dashed red lines. (B) A stereoview of the homodimerization interface mediated by the β6 strands. The α5 helices are omitted for clarity. (C) A stereoview of the homodimerization interface mediated by the α5 helices.

Paracaspase–Ig interface.

The interface between the paracaspase domain and Ig fold of MALT1-C is rigidly held in place by a combination of hydrogen bonds and hydrophobic interactions (Fig. 5). Highly conserved hydrogen bonds are primarily mediated between the α2 helix and β2 strand in the paracaspase domain and the β5/β6 and β2/β3 loops within the Ig domain (Fig. 5A). At one end of the polar interface, Glu390 from α2 uses its carboxylate side chain to accept three hydrogen bonds from the side chains of Gln676 and Lys677 on the β5/β6 loop (Fig. 5A). At the center, Leu384 and Ser382 from β2 use a combination of main- and side-chain groups to make a total of six hydrogen bonds with Ser609 and Asn610 on the β2/β3 loop. At the other end of the polar interface, Glu368 on α1 accepts a hydrogen bond from the side chain of Tyr657. Most of the interface residues are highly conserved, suggesting functional importance. For example, Glu390, Ser382, Leu384, Ser609, Asn610, and Gln676 are all invariant in MALT1 across several species (Fig. S5).

Fig. 5.

Features of the interface between the paracaspase domain and the Ig fold in MALT1. (A) A stereoview of the interdomain hydrogen bonds between the paracaspase domain (blue) and the Ig fold (orange) in MALT1. These hydrogen bonds are stabilized by opposing charges between donor and acceptor. Hydrogen bonds are highlighted as red dashed lines. (B) A stereoview of the interdomain van der Waals interactions among hydrophobic residues from the paracaspase (blue) and Ig (orange) domains in MALT1.

Compared to the hydrogen bond networks described above, the hydrophobic contacts are made in the same vicinity but centered mostly on the β2 strand and α1 helix in the paracaspase domain and the β2/β3 loop, β4/β5 loop, and the C terminus of the Ig domain (Fig. 5B). Leu384 and Val381 from the paracaspase β2 strand play a central role and form van der Waals contacts with Phe608 (β2/β3 loop) and Ile712 (C terminus) of the Ig domain. In addition, Leu363 from α1 of the paracaspase domain also forms hydrophobic interactions with Phe608 (β2/β3 loop) and Tyr657 (β4/β5 loop) of the Ig fold. It is also possible that Val364 and Leu506 contribute to the hydrophobic interface by packing against Tyr657. Overall, the highly conserved van der Waals contacts from the paracaspase domain impinge on Phe608 of the Ig fold. Tyr657 and Ile712 of the Ig domain are not absolutely conserved, but virtually all species examined have hydrophobic residues in similar or neighboring positions which may contribute to the stability of the buried interface.

Discussion

MALT1 is a critical effector of the CBM signalosome that mediates NF-κB activation downstream of multiple receptor types. Oligomerization of the MALT1 C terminus, by assembly into the CBM complex or as a consequence of the MALT lymphoma translocation, is thought to drive activation of associated TRAF6 or another TRAF member. The ubiquitin ligase TRAF6, along with other conjugating factors such as Ubc13, promotes Lys63-linked polyubiquitylation of a range of proteins to ultimately recruit and activate the IKK complex. Supporting this notion, artificial oligomerization of the MALT1 C terminus alone is sufficient to drive NF-κB activation in cell culture (6, 7). Moreover, the same region of MALT1 has been shown to possess oligomerization-dependent proteolytic activity which is also proposed to regulate NF-κB signaling (10, 11). Precisely how these critical functions are coordinated and executed by MALT1 have been enigmatic, in part due to lack of structural information.

In this manuscript, we present the crystal structure of the C-terminal fragment of MALT1 and report a number of key findings that are critical for paracaspase activity. A previous study indicated the presence of a distantly related caspase-like family in MALT1 (3). Despite the low sequence homology, our structure reveals a bona fide caspase fold similar to that of caspase 9 or any other caspase. Unexpectedly, this caspase-like domain (paracaspase) is packed closely against an adjacent Ig domain to form a single, conformationally restrained unit. The residues involved in maintaining this interface are highly conserved across species, suggesting that the combined unit is a common feature of MALT1 in different organisms.

Structural mimicry between MALT1 and caspases is also reflected by the way homodimers are formed to induce catalytic activity. In the case of the initiator caspases, transient dimerization is thought to elicit a basal level protease activity which in turn promotes autoproteolysis to yield a fully active and stable homodimer. Similarly, the covalent peptide inhibitor appears to artificially lock the transient active form of MALT1-C in precisely the same dimeric arrangement as initiator caspase zymogens (23). However, in contrast to classic caspases, MALT1 does not undergo autoproteolysis to yield a stable and highly active homodimer (3). Consequently, the L2 and L2′ loops of MALT1 remain tethered together, away from the active site, and are unable to form loop-bundle interactions, which stabilize the active site conformation. Based on these observations, we speculate that MALT1 dimerization is likely to elicit only a low level of protease activity, which is more consistent with a regulatory role by MALT1. Because no contacts are made to the active site upon dimerization, we further predict that conformational changes resulting from β6 strand dimerization are propagated through the caspase-like fold to form functional active sites as postulated for other caspases (23, 24).

MALT1 contains a C-terminal Ig domain which may recruit key factors to promote NF-κB activation, and this is consistent with our DALI analysis showing this region is similar to modules in the AP1 (20) and GGA (21) complex that recruit auxiliary proteins for membrane transport. In fact, this Ig domain has been reported to interact with the E2 ubiquitin-conjugating enzyme unit Ubc13/Mms2 (6, 16) and the E3 ubiquitin ligase TRAF6 (7). Previous studies demonstrated that TRAF6 directly interacts with the E2 Ubc13 (25) to facilitate ubiquitin transfer to appropriate substrate. Hence, TRAF6 association with Ubc13 and MALT1 is required to bring substrates within the CBM into proximity of Ubc13 to allow Lys63-linked ubiquitylation to occur. So far, TRAF6 binding sites have been mapped to the C-terminal Ig domain as well as to the N terminus and the extreme C terminus (804–809) of MALT1 (7, 26). Additional effort is needed to determine the precise mechanism by which TRAF6 or possibly TRAF2 couples with MALT1 to promote NF-κB activation.

The Ig domain in our structure is also proposed to undergo Lys63-linked ubiquitylation via TRAF6 in potentially nine different lysines to recruit the IKK complex (9). Of the possible coupling sites outlined, only one lysine appears to be invariant across species and our structure indicates that it is freely accessible in a solvent exposed loop (Fig. S6). Nonetheless, any neighboring lysine may still accept ubiquitin chains because RING E3 ligases such as TRAF6 do not appear to exhibit stringent specificity for lysines but rather act generally on proximal lysine residues (27). As more details in this aspect of MALT1 regulation are uncovered and unique substrates and binding partners for MALT1 are revealed, our structure will serve as an important framework for understanding how the various elements in the CBM complex integrate with the scaffolding and protease features of MALT1.

Materials and Methods

Protein Preparation.

Identification of the MALT1-C construct boundaries through limited proteolysis experiments and sequence analysis is described in the SI Materials and Methods and Fig. S7. Human MALT1 (residues 339–719) was cloned into pET21b (Novagen) using standard PCR-based cloning procedures. This construct was expressed in the Escherichia coli strain BL21(DE3) and purified by Ni²⁺-nitrilotriacetate resin (Qiagen). MALT1 (339–719) was further fractionated by ion exchange (Source-15Q; GE Healthcare) and gel filtration chromatography (Superdex-200; GE Healthcare). Purified MALT1-C was then incubated with an excess amount of the inhibitor z-VRPR-fmk (Enzo Life Sciences). This mixture was further fractionated by gel filtration chromatography and the fractions corresponding to dimeric protein were pooled and concentrated for protein crystallization. For production of selenomethionine protein, MALT1-C was expressed in the E. coli strain B834 and purified in the inhibitor-bound dimeric form as described above.

Crystallization, Data Collection, and Structure Determination.

Crystals were grown via the hanging-drop vapor-diffusion method by mixing protein (8–10 mg/mL) with an equal volume of reservoir solution. Inhibitor-bound MALT1-C was initially crystallized in well buffer containing 0.1 M MES (pH 6.0), 0.2 M calcium acetate, 2% (wt/vol) benzamidine, and 2-propanol (6–9% vol/vol for native protein and 2–5% for selenomethione-labeled protein). These crystals were then used as microseeds to generate diffraction quality crystals in a well buffer containing 0.1 M MES (pH 6.0), 0.2 M calcium acetate, and 2% benzamidine. Native and selenomethione-labeled MALT1 protein crystals were subsequently equilibrated in reservoir buffer containing 25% ethylene glycol (vol/vol) and were flash frozen in liquid nitrogen. Native data and MAD sets for both were collected at National Synchrotron Light Source beamline X25 and processed with Denzo and Scalepack software (28). Structure determination is described in the SI Materials and Methods.