Ubiquitin Regulates Caspase Recruitment Domain-mediated Signaling by Nucleotide-binding Oligomerization Domain-containing Proteins NOD1 and NOD2*

Background: NOD1 and NOD2 are innate immune receptors implicated in inflammatory diseases. Results: The CARDs of NOD1 and NOD2 bind ubiquitin; mutations that block this interaction increase inflammatory signaling. Conclusion: NOD1 and NOD2 signaling is regulated in part by interaction of their CARDs with ubiquitin. Significance: Understanding the regulatory mechanisms of NOD1 and NOD2 is crucial to defining their role in inflammation. NOD1 and NOD2 (nucleotide-binding oligomerization domain-containing proteins) are intracellular pattern recognition receptors that activate inflammation and autophagy. These pathways rely on the caspase recruitment domains (CARDs) within the receptors, which serve as protein interaction platforms that coordinately regulate immune signaling. We show that NOD1 CARD binds ubiquitin (Ub), in addition to directly binding its downstream targets receptor-interacting protein kinase 2 (RIP2) and autophagy-related protein 16-1 (ATG16L1). NMR spectroscopy and structure-guided mutagenesis identified a small hydrophobic surface of NOD1 CARD that binds Ub. In vitro, Ub competes with RIP2 for association with NOD1 CARD. In vivo, we found that the ligand-stimulated activity of NOD1 with a mutant CARD lacking Ub binding but retaining ATG16L1 and RIP2 binding is increased relative to wild-type NOD1. Likewise, point mutations in the tandem NOD2 CARDs at positions analogous to the surface residues defining the Ub interface on NOD1 resulted in loss of Ub binding and increased ligand-stimulated NOD2 signaling. These data suggest that Ub binding provides a negative feedback loop upon NOD-dependent activation of RIP2.

NOD1 and NOD2 (nucleotide-binding oligomerization domain-containing proteins) are intracellular pattern recognition receptors that activate inflammation and autophagy. These pathways rely on the caspase recruitment domains (CARDs) within the receptors, which serve as protein interaction platforms that coordinately regulate immune signaling. We show that NOD1 CARD binds ubiquitin (Ub), in addition to directly binding its downstream targets receptor-interacting protein kinase 2 (RIP2) and autophagy-related protein 16-1 (ATG16L1). NMR spectroscopy and structure-guided mutagenesis identified a small hydrophobic surface of NOD1 CARD that binds Ub. In vitro, Ub competes with RIP2 for association with NOD1 CARD. In vivo, we found that the ligand-stimulated activity of NOD1 with a mutant CARD lacking Ub binding but retaining ATG16L1 and RIP2 binding is increased relative to wild-type NOD1. Likewise, point mutations in the tandem NOD2 CARDs at positions analogous to the surface residues defining the Ub interface on NOD1 resulted in loss of Ub binding and increased ligand-stimulated NOD2 signaling. These data suggest that Ub binding provides a negative feedback loop upon NOD-dependent activation of RIP2.
The innate immune system relies on pattern recognition receptors (PRRs) 4 to detect pathogens by recognizing small molecular motifs called pathogen-associated molecular patterns (1)(2)(3). Nucleotide-binding oligomerization domain (NOD)like receptors (NLRs) are a family of cytosolic PRRs that mediate immune responses to a wide variety of intracellular pathogens and cellular damage (1)(2)(3)(4)(5)(6)(7). The NLRs NOD1 and NOD2 detect invading bacteria by recognizing minimal fragments of cell wall peptidoglycan called ␥-D-glutamyl-meso-diaminopimelic acid (iE-DAP) (8,9) and muramyl dipeptide (MDP) (10), respectively. Recognition of their cognate pathogen-associated molecular patterns causes NOD1 and NOD2 to translocate to sites of bacterial entry into the cell, where they form a signaling scaffold (11,12). NOD1 and NOD2 are structurally and functionally related to the apoptotic signaling NLR APAF-1. APAF-1 resides in an autoinhibited state until its ligand-binding domain is engaged by cytochrome c, releasing it to assemble into a large signaling complex, termed the apoptosome, that recruits and activates procaspase-9 (13). Similarly, NOD1 and NOD2 are thought to assemble into a multimeric NODosome scaffold in a ligand-dependent manner and recruit the downstream effector RIP2 (receptor-interacting protein kinase 2) and/or ATG16L1 (autophagy protein 16-like) (11,12,14). Recruitment of the autophagy protein ATG16L1 to sites of bacterial entry helps mediate the eventual lysosomal degradation of invading bacteria (15). Activation of the effector kinase RIP2 induces the NF-B and MAPK pathways, resulting in activation of inflammatory genes (16,17). The NOD-induced inflammatory response is mediated in part by the formation of Lys-63-linked polyubiquitin chains via ubiquitin (Ub) ligases associated with the NODosome (18,19). These in turn activate the IKK complex through IKK␥/NEMO, resulting in the eventual phosphorylation and degradation of IB and liberation of the NF-B inflammatory transcription factor (20 -24). Similarly, polyubiquitin chains have been shown to nucleate signaling complexes containing the MAPKKK Tak1, leading to activation of the IKK complex and induction of p38 MAPK inflammatory pathways (22,25). NOD1 and NOD2 signaling is mediated by their N-terminal caspase recruitment domains (CARDs). CARDs are part of a larger family of protein-protein interaction domains, known as the death domain superfamily, that share a common six-helical bundle structural fold. Collectively, the death domain superfamily members are involved in at least some aspect of nearly all immune and apoptotic signaling pathways (26). Like other death domain subfamilies, CARDs were thought to engage almost exclusively in homotypic interactions with other CARDs. Indeed, the interaction between NOD1 and RIP2 is mediated by their respective CARDs. Recently, however, several heterotypic interactions for the tandem CARDs of RIG-1, a PRR that recognizes viral RNA, were identified (27,28), with one compelling observation being that they function as Ubbinding domains (29). This provided a potential mechanistic link for how Ub might regulate or instigate signal transduction by innate immune receptors at the beginning of their respective pathways. In general, polyubiquitin chains serve as scaffolds, allowing downstream target proteins harboring Ub-binding domains to associate and undergo activation (22,23). However, the role of Ub in immunity and autophagy continues to expand in complexity and scope (20,30).
The finding that a related CARD has Ub binding activity presented the possibility that Ub might trigger, amplify, or attenuate the early steps of immune signaling pathways by other intracellular CARD-containing PRRs, such as NOD1.
Here we demonstrate that NOD1 CARD binds Ub. Mutations in NOD1 CARD that ablate Ub binding but retain the ability to bind to downstream effectors RIP2 and ATG16L1 display increased ligand-dependent inflammatory signaling, supporting a model whereby Ub regulates proximal signaling of NOD1. In addition, mutations that inactivate Ub binding of the NOD2 tandem CARDs also increase ligand-dependent RIP2 signaling, suggesting a common mode of Ub-dependent regulation.

EXPERIMENTAL PROCEDURES
Plasmids and Reagents-Plasmids used are summarized (supplemental Table S1). Antibodies to V5 epitope (R96025, Invitrogen), HA epitope (MMS-101R, Covance, Princeton, NJ) and lamin B1 (ab16048, Abcam, Cambridge, MA) were purchased. The NOD1 ligand iE-DAP and the NOD2 ligand MDP were purchased from Invivogen (San Diego, CA). Bacterial codon-optimized hATG16L1 WD40 CS was synthesized by Genscript (Piscataway, NJ). All eight cysteines present in the wild-type protein were mutated to alanines in the synthesized gene to improve soluble expression. This gene contained 5Ј EcoRI and 3Ј XhoI cut sites, which were used to generate a fragment for ligation between the same sites in the vector pGEX-6P-1.
Protein Purification-Ub and 15 N-labeled Ub were purified as described previously (31).
His 6 -NOD1 CARD was affinity-purified over Talon Co 2ϩagarose as described previously (32). Imidazole-eluted protein was then purified by gel filtration over Superdex 75 equilibrated with GF buffer (25 mM sodium phosphate buffer, pH 7.0, with 25 mM NaCl and, depending on the experiment, 5 mM DTT). Fractions corresponding to the monomeric and/or dimeric form of the CARD were pooled and concentrated. For alkylated NOD1 CARD, pooled and concentrated monomeric fractions from gel filtration were incubated in iodoacetamide (Pierce) added to a final concentration of 50 mM. Alkylated NOD1 CARD was then repurified over Superdex 75 in GF buffer lacking DTT. Production of 15 N-labeled NOD1 CARD was accomplished by growing cells to an OD of 0.8 in LB and resuspending them in 15 N-spectra 9 minimal medium supplemented (10%) with 15 N-containing Celltone (Spectra Stable Isotopes, Andover, MA) prior to induction with 0.5 mM isopropyl ␤-D-1-thiogalactopyranoside. Production of 13 C, 15 N -labeled NOD1 CARD for backbone assignments was done similarly to the 15 Nlabeled protein, except cells were resuspended in 13 C, 15 N-containing Celltone complete medium (100%). Production and purification of GST fusion proteins and MBP fusion proteins was performed according to standard protocols as described previously (31). GST-ATG16L1 WD40 was produced in MC1061 bacteria co-transformed with plasmid expressing GroES and GroEL as described previously (33). Supplemental Fig. S4A shows the GST and MBP fusion proteins used in this study.
NMR Spectroscopy-For backbone assignments of NOD1 CARD, symmetrical NMR microtubes (Shigemi, Allison Park, PA) were loaded with 0.3 ml of 1 mM 13 C, 15 N-labeled, alkylated NOD1 CARD in 25 mM sodium phosphate buffer, pH 7.0, with 25 mM NaCl, 0.01% NaN 3 , and 10% (v/v) D 2 O. Spectra were collected at 25°C on a four-channel Varian UnityInova NMR spectrometer operating at 600 MHz and equipped with a triple resonance PFG probe. Assignments (supplemental Fig. S2A) were made using the following experimental data sets: 1 H-15 N HSQC, HNCACB, CBCA(CO)NH, HNCO, and HN(CA)CO. All spectra were processed using NMRPipe (34), and peak picking was done using NMRViewJ (35). Initial peak lists were submitted to the PINE server for automated assignment (36), and results were verified manually in NMRViewJ.
For NMR titration experiments, HSQC spectra were collected at 25°C using a Bruker Avance II 800-MHz spectrometer equipped with a cryoprobe. Spectra were processed with NMRPipe (34) and analyzed using SPARKY (37). Chemical shift differences were calculated using the formula (0.07*⌬N 2 ϩ ⌬H 2 ) 1 ⁄2 . To identify the most significant chemical shift changes, values exceeding one S.D. above the mean chemical shift change for all observable residues were used for Ub, whereas values exceeding two S.D. were used for NOD1 CARD. To ensure sensitivity and specificity in measuring chemical shift differences, all comparisons used spectra of a matched sample of 15 N-labeled binding partner alone that was collected sequentially. Control samples were made using the same buffer in which the binding partner protein was dialyzed. Spectra of an experimental run were processed using identical parameters. Using this method, the intraexperimental variability in measuring 15 N-labeled Ub alone in three separate samples showed an average chemical shift difference of 8.8e Ϫ4 ϩ 5.9e Ϫ4 S.D. A base line for measuring a significant difference was set to 10 S.D. above the highest variance.
Titration data were fit to the equation, where ␦ obs is the observed chemical shift of the labeled protein upon titration; ␦ P is the chemical shift of the labeled protein alone; ␦ PL is the chemical of the labeled protein-ligand complex; P T is the total concentration of labeled protein; L T is the total concentration of ligand; and K D is the dissociation constant for the protein-ligand complex.
Pull-down Binding Experiments-Each GST protein (100 g), immobilized on 50 l of glutathione-Sepharose (GE Healthcare), was used per binding reaction. Beads and input proteins, either purified, in vitro translated, or as cell lysates, were incubated in a total volume of 800 l of reaction buffer (PBS containing 0.1 mg/ml BSA, 0.05% Triton X-100, and 5% v/v glycerol) with gentle agitation for 1 h at 25°C. Beads were washed three times in reaction buffer (4°C), and bound complexes were eluted with 50 l of PBS containing 50 mM reduced glutathione. Samples were diluted with Laemmli sample buffer, heated, and subjected to SDS-PAGE and immunoblotting. Pulldown experiments using MBP fusion proteins were performed similarly except that MBP fusions were immobilized onto 100 l of amylose resin (New England Biolabs, Ipswich, MA) and eluted with 50 mM maltose in PBS. Pull-down experiments using NOD2 CARD-GFP fusion proteins were done using extracts from HEK293T cells transiently transfected with plasmids expressing WT or mutant NOD2 CARDs-GFP-V5 fusion protein. Cells were lysed in 1 M NaCl, 20 mM NaPO 4 , pH 9.0, diluted to 100 mM NaCl, 20 mM NaPO 4 , pH 7.5. The resulting extract was diluted (1:10) in PBS, 0.1 mg/ml BSA, 0.05% Triton X-100, and 5% (v/v) glycerol and used for GST pull-down assays.
Radiolabeled in vitro translated proteins were produced using rabbit reticulocyte lysate as per the manufacturer's procedures (TNT T7 Coupled Reticulocyte Lysate System, Promega, Madison, WI) in the presence of 1 mM amino acid mixture minus Met/Cys and supplemented with 1 mM 35 S-labeled Met/Cys (EasyTag TM EXPRESS 35 S Protein Labeling Mix, PerkinElmer Life Sciences). Proteins were synthesized at 30°C for 90 min.
Chain binding assays were performed with tetra-Ub chains purchased from Boston Biochem. Affi-Gel beads (Bio-Rad) were covalently linked to polyclonal affinity-purified anti-V5 antibodies, which were subsequently bound with saturating concentrations of bacterially produced V5-tagged NOD1 CARDs. Beads were divided and resuspended in 100 l of PBS with 0.1% casein and 5 g/ml tetra-Ub (Ub 4 ) chains. Beads were washed in cold PBS with 0.01% Triton X-100 and immunoblotted to detect bound Ub 4 and V5-CARD domains.
Transfections and IL-8 Secretion Assays-HEK293T cells (ATCC) were maintained in high glucose DMEM supplemented with 10% FBS and penicillin/streptomycin (Invitrogen). Transfections were done using Polyfect (Qiagen, Valencia, CA). IL-8 ELISAs were done using the OptEIA TM IL-8 kit (BD Biosciences). For IL-8 assays, confluent flasks of HEK293T cells were split into 48-well plates and grown to ϳ70% conflu-ence before transfection. Cells were transfected with the same amount of plasmid DNA, and titration of NOD1-expressing plasmid was afforded by a compensating addition of vectoronly DNA (pcDNA). Twenty-four hours after transfection, medium was replaced with either new medium alone or medium supplemented with the indicated concentrations of stimulatory ligand. After 12 h, supernatants were collected for IL-8-specific ELISA analysis, and cells were lysed in Laemmli buffer and subjected to SDS-PAGE and immunoblotting.
Structural Analysis and Modeling-Structures were analyzed and displayed with PyMOL (38).

Direct Interactions of NOD1 CARD with RIP2, ATG16L1, and
Ubiquitin-NOD1 shares a common architecture with other NLRs, including a central NOD related to the AAA ATPase superfamily of proteins followed by leucine-rich repeats thought to comprise a ligand-binding domain (39). In the case of NOD1, its N-terminal CARD is responsible for specific downstream signaling activity (Fig. 1A). We were interested in how protein interactions mediated by the effector domains of NLRs control their signaling output. Our first step was to characterize the key interactions between NOD1 CARD and its anticipated downstream target proteins as well as their roles in immune regulation.
One downstream target of NOD1 is the kinase RIP2, which activates inflammation through NF-B and p38 MAPK. Yeast two-hybrid assays, as well as immunoprecipitation assays, have indicated that the interaction between NOD1 and RIP2 is mediated by their respective CARDs (39 -41). This has supported a model in which the NOD1-RIP2 association is mediated by direct CARD-CARD interactions; however, this has not been directly tested in vitro. Fig. 1B demonstrates that epitopetagged NOD1 CARD binds directly to an MBP fusion of RIP2 CARD. This interaction was specific in that binding to MBP alone was not observed. We observed no binding of NOD1 CARD to either procaspase 1 or procaspase-9 CARDs, although previous immunoprecipitation experiments have indicated that NOD1 associates with both procaspases in a CARD-dependent manner (39,42). Although we cannot exclude the possibility that our MBP-procaspase CARD proteins were misfolded, both CARDs have been produced successfully in other studies (43,44). Together, our results confirm previous conclusions using in vivo approaches that NOD1 and RIP2 CARDs directly interact to mediate a signal transduction pathway leading to NF-B activation, yet our results also indicate that the in vivo association of NOD1 with procaspase-1 and procaspase-9 might not be mediated by direct CARD-CARD interactions.
Another downstream target of NOD1 is a complex of autophagy proteins composed of ATG5, ATG12, and ATG16L1, which together act as a ligase for conjugating LC3 to phosphatidylethanolamine (45,46). The N-terminal domain of ATG16L1 is required for its association with the ATG5-ATG12 subcomplex. Immunoprecipitation experiments from cells expressing epitope-tagged proteins clearly showed association of NOD1 with ATG16L1 and ATG16L1 lacking its N-terminal 85 residues (⌬N85-ATG16L1) (12), indicating that association of NOD1 with ATG16L1 does not require ATG5 or ATG12.
However, exactly how the ATG16L1-NOD1 association is mediated and whether this is direct has not been determined. Fig. 1C confirms that NOD1 CARD binds directly to the C-terminal WD40 region of human ATG16L1, predicted to form a seven-bladed ␤-propeller. In vitro translated full-length [ 35 S]Met-labeled ⌬N85-ATG16L1 as well as a fragment encompassing the C-terminal ␤-propeller specifically bound GST fused to NOD1 CARD. In contrast, neither the coiled-coil region nor the intermediate domain of ATG16L1 bound NOD1 CARD. In addition, the WD40 ␤-propeller alone or a fragment containing both the intermediate domain and the ␤-propeller also bound the tandem CARDs of NOD2, in agreement with previous experiments showing association of NOD2 with ATG16L1 in vivo (12,15). To confirm a direct interaction, we produced the human ATG16L1 WD40 ␤-propeller as a GST fusion protein and found that it bound recombinant NOD1 CARD (Fig. 1D). Previous experiments have shown that a wide variety of WD40 ␤-propellers are capable of binding Ub (31). We suspected the ␤-propeller of ATG16L1 might also have this activity and included this parameter in our experiments as a potential control for the validity of our binding experiments. Here we found that in vitro translated ⌬N85-ATG16L1 as well as the ␤-propeller alone bound to GST-Ub (Fig. 1C). In addition, we found that the GST-␤-propeller fusion protein bound linear Ub 5 , a polyubiquitin chain composed of a concatemer of five in-frame Ub moieties (Fig. 1F). Together, these data provide the first demonstration of a direct interaction of NOD1 CARD with either RIP2 CARD or the ATG16L1 WD40 ␤-propeller, and these interactions explain the biochemical basis for NOD1 signaling through these pathways.
We also found that NOD1 CARD directly binds Ub (Fig. 1E). The tandem CARDs of another innate immune receptor, , and a C-terminal WD40 ␤-propeller (␤Prp). B, binding of bacterially produced V5 epitopetagged NOD1 CARD to MBP alone (ø) or MBP fused to procaspase-9 CARD (P9 CARD), procaspase-1 CARD (P1 CARD), or RIP2 CARD. C, binding of 35 S-labeled in vitro-translated fragments of ATG16L1 to GST alone (ø) or GST fused to Ub, NOD1 CARD, or tandem CARDs of NOD2. D, binding of bacterially produced V5 epitope-tagged NOD1 CARD to GST alone or as a fusion with the ␤-propeller of human ATG16L1. E, binding of NOD1 CARD to GST or GST fused to Ub or mutant Ub (ub*; L8A, I44A, R42E, R72D, and R74D). F, binding of V5-tagged penta-Ub linear chains (Ub 5 ) to GST or GST ATG16 WD40 ␤-propeller. G, binding of bacterially produced HA epitope-tagged RIP2 CARD to GST or GST fused to Ub, ub*, or NOD1 CARD.
RIG-1, have been shown to bind directly to Ub (29). This interaction is thought to be important for activation of its downstream target, mitochondrial antiviral signaling protein (MAVS), because deletion of the RIG-1 CARDs or the addition of a point mutation in the first CARD, rendering it incapable of binding MAVS and Ub, blocks RIG-1-dependent signaling. Fig.  1E shows that NOD1 CARD also binds Ub in pull-down experiments using GST-Ub. NOD1 CARD did not bind GST alone or GST fused to Ub that contained mutations around the surface Ile-44-containing hydrophobic patch on Ub, which mediates the vast majority of Ub interactions (47). In contrast, the CARD of RIP2, made as a recombinant HA-tagged protein, did not bind GST-Ub but still bound to NOD1 CARD (Fig. 1G). Thus, our data indicate that the ability to bind Ub is not unique to the CARDs of RIG-1, but it is also not universal to all CARDs.
To confirm the interaction of NOD1 CARD with a single Ub (mono-Ub), we performed HSQC NMR experiments with 15 Nlabeled mono-Ub in the presence of increasing concentrations of unlabeled NOD1 CARD. NOD1 CARD caused specific chemical shift perturbations (CSPs) in the 15 N/ 1 H HSQC spectrum for a subset of backbone amides in Ub (Fig. 2A). The CSPs generated were dose-dependent and yielded a calculated K d of binding of Ͼ1 mM (Fig. 2B). Although this affinity is low, it was measured utilizing proteins taken out of their usual context, where they are multimerized as either part of an oligomeric NODosome or in poly-Ub chains. Within that context, it is plausible that this binding activity plays a role in vivo. The observed CSPs were adequate to define the structural basis for this binding. Mapping the residues that underwent the largest CSPs (Fig. 2C) onto the known structure of mono-Ub revealed a surface surrounding the hydrophobic patch on the surface of Ub (Fig. 2D). Together, these data demonstrate that NOD1 CARD binds multiple partners via direct protein-protein interactions and qualify NOD1 CARD within the subset of CARDs that can directly interact with mono-Ub.
Structural Basis for the NOD1 CARD-Ub Interaction-We next sought to define a functional role for the ability of NOD1 CARD to bind Ub. This required identifying NOD1 mutations that would selectively block Ub binding without diminishing binding to the downstream targets RIP2 and ATG16L1. We used NMR to follow chemical shift perturbations in 15 N-labeled NOD1 CARD to identify residues that engage Ub. We first obtained backbone assignments for NOD1 CARD; however, we found that this method required initial optimization. Our previously determined crystal structure of NOD1 CARD showed that it formed helix-swapped homodimers, juxtaposing two cysteines (Cys-39) near each other (supplemental Fig. S1, A andD) (32). Indeed, NOD1 CARD in solution lacking added reducing agents readily forms dimers linked by a disulfide bond, and this dimeric form has been observed in a different crystal structure (48). Mutation of the other cysteine residues, Cys-59 and Cys-61, to alanine does not prevent dimer formation, but it does eliminate higher order aggregation. We found that a proportion of wild-type NOD1 CARD as well as the C59A/C61A CARD could still form dimers even in the presence of DTT. Therefore, we generated an obligate monomeric form by alkylating the wild-type NOD1 CARD or the C59A/C61A mutant CARD with iodoacetamide. This treatment, predicted to disrupt the dimer interface and also block oxidation of Cys-39, produced a purely monomeric form by gel filtration in the absence of reducing agent (supplemental Fig. S1B). We also confirmed that this form still bound Ub using NMR HSQC experiments with 15 N-labeled mono-Ub. In addition, we prepared a homogeneous sample of dimeric NOD1 CARD, made by oxidizing C59A/C61A mutant NOD1 CARD and purifying it over gel filtration, and found that NOD1 CARD dimers also bound Ub (supplemental Fig. S1C). Although the magnitude of chemical shift perturbations was somewhat lower for dimeric NOD1 CARD, this binding activity indicated that the interface occluded in the dimer was not critical for binding Ub.
Backbone assignments from 15 N/ 13 C-labeled alkylated NOD1 CARD (supplemented Fig. S2A) allowed us to identify residues in NMR HSQC experiments that were perturbed upon binding a fusion protein of linear Ub 5 (Fig. 2E). The oligomerized Ub 5 chain, which is larger and tumbles slower, was used to avoid detecting HSQC peaks attributable to mono-Ub that are observed at high concentrations. The most significant chemical shift perturbations (Fig. 2F) mapped to a region of the NOD1 CARD composed of helices 5 and 6 as well as the loop connecting them (Fig. 2G). Chemical shift changes were not observed for residues that are occluded upon dimer formation. This is consistent with our observation that the dimer still retains the ability to bind Ub and verifies the specificity of the NMR analysis (supplemental Fig. S1D).
NMR experiments measuring chemical shift perturbations upon binding linear Ub 5 indicated two key residues on the CARD surface that were mediating Ub-binding. The hydrophobic Tyr-88 residue, which is also solvent-exposed in the crystal structure, could potentially mediate interaction with the hydrophobic surface patch on Ub identified in Fig. 2D. The nearby acidic Glu-84 residue could potentially mediate electrostatic interactions near Arg-72 of Ub. To test this, we generated the E84A/Y88R double mutant CARD and assessed its ability to directly bind its effectors. Unlike the wild-type CARD, the E84A/Y88R mutant no longer bound GST-Ub (Fig. 3A). However, the E84A/Y88R mutant still bound to RIP2 CARD (Fig.  3A) and the WD40 ␤-propeller of ATG16L1 (Fig. 3B). Together, these data show that the Ub-binding interface of NOD1 CARD can be uncoupled from binding to RIP2 kinase and the ATG16L1 autophagy complex, thus allowing a functional dissection of how various CARD interactions contribute to NOD1 function.
We also examined whether NOD1 CARD had specificity for a particular Ub chain linkage. Fig. 3C shows that V5-tagged NOD1 immobilized on Sepharose beads linked to polyclonal anti-V5 antibodies bound both Lys-63-linked Ub 4 chains and Met-1-linked linear Ub 4 chains. Both of these chains show an open conformation, where the binding surface on Ub used by NOD1, as defined by our NMR experiments, is solvent-exposed and accessible for binding. In contrast, Lys-48-linked Ub, which can adopt a closed conformation that occludes the hydrophobic binding surface of Ub, did not readily bind NOD1 CARD. These experiments also showed that the double E84A/Y88R NOD1 mutant as well as the single Y88R NOD1 mutant CARD were defective in binding Lys-63 and linear Ub chains.
The selective defect of the E84A/Y88R NOD1 mutant characterized here differs from mutants used in other studies that were used to disrupt CARD function. One such mutant is R69E, which has been shown to block the ability of NOD1 to associate with RIP2 in co-immunoprecipitation experiments and to eliminate NOD1-dependent NF-B activation (49) . Fig. 3, A and B, shows that the R69E mutation blocks not only direct binding of NOD1 CARD with RIP2 CARD but also binding to both Ub and the WD40 ␤-propeller of ATG16L1. Thus, mutations such as R69E are nonspecific and cannot be used to dissect the functional role of individual CARD interactions. Nonetheless, the inability of R69E mutant CARD to bind to Ub, RIP2, or ATG16L1 verifies the specificity of the binding reactions followed here. Other surface mutations on NOD1 CARD have also been found to block association of full-length NOD1 with RIP2 in cells. These mutations include E53K, D54K, and E56K (49); however, none of these mutant CARD domains could be recovered as soluble proteins upon expression in Escherichia coli, indicating that they were even more structurally compromised than the R69E mutant. Hence, these NOD1 mutants serve as a caveat for inferring functional significance from mutations that may or may not specifically block pertinent pathways. Interestingly, Arg-69 on NOD1 CARD structurally aligns closely to Thr-55 on the first CARD of RIG-1 (supplemental Fig. S3). Our studies demonstrating the overall dysfunction of the NOD1 MARCH 8, 2013 • VOLUME 288 • NUMBER 10 JOURNAL OF BIOLOGICAL CHEMISTRY 6895 CARD R69E mutation may help explain why the RIG-1 mutant T55I nonspecifically blocks general RIG-1 function, including interactions with TRIM25, MAVS, and Ub (29,50,51).

Ub Regulates Caspase Recruitment Domain-mediated Signaling
Ubiquitin Competes with RIP2 CARD for Binding to NOD1 CARD-We considered two distinct possibilities for how Ubbinding might affect NOD signaling through RIP2. Ub could potentiate interaction between NOD1 CARD and RIP2, thus stimulating signaling. This scenario is plausible because the NOD1-RIP2 complex associates with Ub ligases, such as cIAP1, cIAP2, and XIAP (18,19), and also because RIP2 becomes ubiquitinated upon activation of the NOD1 and NOD2 pathways (52,53). By analogy, binding of the RIG-1 CARDs to Ub was FIGURE 3. Ubiquitin competes with RIP2 CARD for binding to NOD1 CARD. A, binding of either WT or mutant forms of V5 epitope-tagged NOD1 CARD to GST alone or fused to Ub or RIP2 CARD. B, binding of V5-tagged CARD domains to GST fused to the ATG16L1-␤-propeller. C, Sepharose beads covalently linked to polyclonal anti-V5 used alone (ø) or bound to mutant and WT V5-tagged NOD1 CARD were incubated with tetra-Ub chains (5 g/ml) linked via Met-1 (linear), Lys-48, or Lys-63. Beads were washed and immunoblotted with anti-Ub or anti-V5 monoclonal antibodies. D, binding of WT and Ub binding-deficient mutant (E84A/Y88R) V5-NOD1 CARD to MBP alone (ø) or MBP-RIP2 CARD. Binding to MBP-RIP2 CARD was performed with increasing levels (2-fold) of Ub 5 chains to a final concentration of 2.5 mg/ml. Quantification of NOD1 CARD in each lane of the WT pull-down was performed, and the percentage bound was referenced to the lane corresponding to the absence of competing Ub 5 (100%). These values were plotted versus Ub 5 concentration. Single-exponential regression yielded the equation, y ϭ 101.89e Ϫ0.979 , with an R 2 of 0.86. Inhibition by 50% of binding was found at ϳ1 mM Ub 5 , in agreement with our NMR titrations.
proposed to stimulate downstream activation through MAVS, which can be accomplished in vitro with the addition of Ub chains (29). Alternatively, Ub binding might compete for bind-ing to RIP2, which would provide a way to dampen signaling through RIP2 as a negative feedback mechanism that is sensitive to the formation of polyubiquitin chains.
Using pull-down assays as a biochemical test, we monitored the binding of NOD1 CARD to RIP2 CARD in the presence of increasing concentrations of linear Ub 5 (Fig.  3D). Here we found that linear Ub 5 interfered with the ability of NOD1 CARD to associate with RIP2 CARD. Com-petition was not observed using the E84A/Y88R NOD1 CARD that lacked its Ub binding activity, demonstrating the specificity of the competition reaction. Thus, at the biochemical level, Ub competes with RIP2 CARD for binding to NOD1 CARD.
Conserved Properties of Ub Binding by NOD2 CARDs-We next used a computational approach to find residues in NOD2 CARDs that correspond with those in the NOD1 CARD required for Ub binding. NOD2 contains two N-terminal CARDs in contrast to NOD1, which contains a single CARD. Models generated for each of the two NOD2 CARDs predicted hydrophobic residues Ile-104 and Leu-200, respectively, to be at positions analogous to Tyr-88 on the NOD1 CARD (Fig. 4). We then tested the ability of wild-type and I104R/L200R mutant tandem NOD2 CARDS to bind Ub. Both were expressed as a GFP fusion protein with the CARDs positioned at the N terminus, as they are oriented within full-length NOD2. Fig. 4D shows that wild-type NOD2 CARDs bound Ub in GST-Ub pull-down experiments, but the mutant NOD2 CARDs did not bind. This demonstrates that the residues identified in both NOD1 and NOD2 define a conserved Ub interaction surface.
Disrupting Ub Binding by NOD1 and NOD2 CARDs Enhances Inflammatory Signaling-We assessed the functional role of Ub binding by NOD1 and NOD2 CARDs by comparing full-length wild-type NOD1 and NOD2 with the corresponding NOD1 (E84A/Y88R) and NOD2 (I104R/L200R) mutants for their ability to activate RIP2. We transiently transfected HEK293T cells with NOD expression vectors and measured ligand-induced production of the proinflammatory cytokine IL-8, which is a well characterized target of the RIP2 pathway due to its response to NF-B and MAPK activation (54 -56). Cells expressing NOD1 or NOD1 mutants were stimulated in the absence and presence of its dipeptide-based ligand iE-DAP. Similarly, cells expressing NOD2 or NOD2 mutants were stimulated in the absence or presence of MDP. As described in previous studies, we found that high level transient expression of NOD1 or NOD2 alone activated RIP2 and stimulated IL-8 secretion, even in the absence of ligand stimulation (39,40,49). We circumvented this effect, using a strategy similar to previous studies (57), by titrating the expression level of NOD1 or NOD2 into the range where ligand-dependent production of IL-8 could be detected. The full-length NOD1 and NOD2 proteins were tagged with an N-terminal HA epitope to allow us to monitor their levels. We found that the mutant NOD1 and NOD2 proteins were expressed to the same levels as their comparable wild-type counterpart (Fig. 5A). As expected, NOD1 bearing the R69E mutation, which rendered its CARD domain unable to interact with any of its known effectors, did not stimulate IL-8 secretion in the presence or absence of ligand (Fig.  5A). Using sequence alignments and predicted NOD2 CARD models (Fig. 4), we hypothesized that Arg-86 and Arg-182 of NOD2, corresponding to Arg-69 of NOD1 in each of the tandem CARDs of NOD2, respectively, would render NOD2 unable to activate RIP2. Indeed, similar to the R69E NOD1 mutant, R86E/R182E NOD2 failed to induce IL-8 production at all levels of transfection in the absence or presence of ligand (Fig. 5A). Expression of wild-type NOD1 and NOD2 induced IL-8 secretion at all levels; however, clear ligand-dependent IL-8 secretion was only observed at lower levels of expression (Fig. 5A). In contrast, E84A/Y88R NOD1 and I104R/L200R NOD2 induced a far larger ligand-dependent IL-8 secretion response when compared with their wild-type counterparts (Fig. 5A). This can be better appreciated when the extent of ligand-dependent IL-8 secretion is plotted over the range of NOD1 and NOD2 expression levels (Fig. 5B). We also found that a single Y88R mutation in NOD1 blocked Ub binding (Fig.  3C) and showed a larger ligand-stimulated response than WT NOD1 (supplemental Fig. S4). Together, our results demonstrate that Ub binding by the CARD of NOD1 can modulate signaling through the NF-B pathway, and they indicate that Ub binding may work to attenuate signaling because loss of Ub binding potentiates ligand-dependent NF-B activation and IL-8 secretion.

DISCUSSION
Our current understanding of the signaling mechanism for NOD1 and its close relative NOD2 is based on an induced proximity model, whereby a multitude of effectors are assembled onto a NOD-based scaffold. The CARDs of NOD1 or NOD2 directly bind core components, which in turn recruit a host of other effectors to the NODosome to provide an appropriately tuned response to various stimuli. One of those target components is RIP2 kinase, verified in our studies to associate via a direct homotypic CARD-CARD interaction with NOD1. RIP2 mediates the indirect association of several Ub ligases with NOD1 and NOD2 (18,19,52). This results in formation of Lys-63-linked Ub chains on RIP2, leading to activation of IKK␥/ NEMO and recruitment of the TAK1-TAB1/2 complex (52,53,58), which leads to induction of NF-B and p38 MAPK signaling pathways. Our finding that the CARDs of NOD1 and NOD2 can form a heterotypic interaction with Ub provides an additional way in which the Ub system might control these signal transduction pathways. The fact that Ub competes with RIP2 CARD for NOD1 CARD binding in vitro and that loss of Ub binding activity in NOD1 or NOD2 enhances ligand-stimulated activation of RIP2 favors a model whereby ubiquitination of associated components disrupts the architecture of the NOD1/ 2-RIP2 NODosome. This would comprise a negative feedback regulatory loop that is engaged when poly-Ub chain formation within the NODosome reaches a critical threshold (Fig. 5C). One key question posed by our findings is the source of poly-Ub that binds NOD1 CARD. There is a strong potential for Ub FIGURE 5. Disrupting Ub binding of NOD1 and NOD2 CARDs enhances inflammatory signaling. A, HEK293T cells were transfected with pcDNA vector alone (V) or pcDNA carrying full-length WT or mutants of NOD1 or NOD2. The levels of NOD expression plasmids were 1000, 100, and 10 ng/ml. At 24 h post-transfection, cells were incubated in the absence (white bars) or presence (black bars) of 10 g/ml iE-DAP or MDP for an additional 12 h. IL-8 levels in the medium were measured by ELISA (bottom). Cells were solubilized in SDS, and lysates were immunoblotted for NOD1-HA or NOD2-HA expression with anti-HA antibodies or immunoblotted with anti-lamin B1 as a loading control (top). B, plot of ligand-stimulated IL-8 secretion from cells expressing wild-type (white bars) NOD1 or NOD2 and mutant (black bars) NOD1 (E84A/Y88R) or NOD2 (I104R/L200R). Results are shown over the range of NOD expression levels achieved by the indicated concentrations of plasmid transfected. Statistical analysis of the -fold change with ligand for NOD1 and NOD2 at the 10 ng/ml plasmid transfection level revealed that the differences were significant, with two-tailed p values of less than 0.001 for each. C, a model for the role of Ub binding by NOD1 and NOD2 CARDs on RIP2-mediated inflammatory signaling. RIP2 kinase is recruited to the activated NODosome through CARD-CARD interactions. This complex undergoes ubiquitination, allowing polyubiquitin chains to bind NOD1 or NOD2 CARDs and interfere with sustained RIP2 association. Error bars, S.D. chains to be in close proximity to the NOD CARDs upon their association with RIP2, given that RIP2 and NEMO form a complex and both are targets of polyubiquitination during activation of the NF-B pathway. This would make recognition of Ub chains by NOD1 and NOD2 within the complex akin to an intramolecular interaction, providing a mechanism for how the limited affinity of NOD1 CARD for Ub could exert a biological effect. Interestingly, although Lys-63-linked polyubiquitination of Lys-209 on RIP2 stimulates signaling (53), other ubiquitination events, such as those catalyzed by the E3 ubiquitin ligase ITCH, appear to antagonize RIP2 signaling (52). ITCH attaches Lys-63-linked poly-Ub chains to RIP2 on residue(s) distinct from Lys-209, suggesting that it is an ITCH-specific ubiquitination event that conveys an inhibitory effect through the Ub binding activity of the NOD CARDs. The association of Ub ligases with the NODosome also provides a potential mechanism for ubiquitination of NOD1 or NOD2 themselves. Although this has not been observed to our knowledge, one might expect only a minimal level of NOD ubiquitination to be achieved if this in turn disrupted proximity to RIP2-associated Ub ligases.
Our results may also add to current understanding of signaling by the viral RNA immune receptor RIG-1. Recently, the crystal structure of Anas platyrhynchos RIG-1, including its tandem CARDs, was determined (59). Based on this structure, we generated a model of human RIG-1 CARDs. This model contains surface-exposed hydrophobic residues in each CARD of RIG-1 that correspond to the same position of the Tyr-88 residue in NOD1, which we have shown is important for Ub binding (supplemental Fig. S3). We took a similar approach to correctly predict which residues of NOD2 were required for Ub binding by NOD2 CARDs, which also have conserved surfaceexposed hydrophobic residues. Thus, if the structural basis of Ub binding by NOD CARDs is conserved, these residues may constitute a Ub-binding surface in the CARDs of RIG-1 and define mutants that can clarify the exact role of Ub binding in these additional pathways. Our study also identifies a novel heterotypic interaction between NOD1 CARD and the WD40 ␤-propeller of ATG16L1. ATG16L1 binds ATG5-ATG12 to form a complex with ligase activity that conjugates LC3 to phosphatidylethanolamine during the course of autophagosome formation (46,60,61). NOD1 and NOD2 are important for directing autophagy of intracellular bacterial pathogens, and previous studies found that NOD1 and NOD2 associate with ATG16L1 by co-immunoprecipitation and help direct ATG16L1 to the site of bacterial entry (12,15). The ATG16L1 WD40 ␤-propeller is a feature of animal ATG16 homologs but is lacking in budding yeast and plants. As a common protein interaction domain, the ␤-propeller probably extends the ability of ATG16L1 to link with a variety of spatial adaptors, including NOD1, to specify where autophagosome formation is initiated and elongated. Previous studies have shown that a variety of WD40 ␤-propellers harbor the ability to directly bind Ub (31); accordingly, we find that the ATG16L1 ␤-propeller also binds Ub, suggesting that Ub may act as another spatial adaptor. A number of recent studies have shown that Ub plays a role in selective autophagy by acting as a sorting tag for incorporation of autophagy substrates (reviewed in Refs. 62-64). A major mechanism in recognizing these substrates involves a family of autophagy receptors, including p62/SQSTM1 and NBR1, that bind both Ub and LC3 and are thought to tether cargo to the nascent autophagosome (65). Perhaps the capacity of the WD40 ␤-propeller of ATG16L1 to associate with Ub serves as an additional regulatory device in selective autophagy, whereby ATG16L1 directly recognizes ubiquitinated autophagy substrates. In summary, our study demonstrates that Ub binds to the CARDs of NOD1 and NOD2, modulates their signaling, and may influence the balance of inflammatory signaling with that of other pathways.