RNAi screening identifies mediators of NOD2 signaling: Implications for spatial specificity of MDP recognition

Identification and Functional Network Analysis of NOD2-Signaling Modulators.

To identify modulators of the NOD2-signaling pathway, we targeted 7,783 genes using a commercially available “druggable” genome siRNA library. The screen was performed in HEK293 cells, which were transfected with siRNA, plasmids encoding NOD2, and dual–NF-κB luciferase reporter constructs. Luciferase activity was measured 12 h after MDP stimulation (Fig. S1A). Results showed a reproducible six- to eightfold induction of NF-κB activation in cells transfected with control siRNA. Independent knockdown of RIPK2 significantly impaired NF-κB activation, and knockdown of firefly luciferase served as internal control (Fig. S1B). Each transcript was targeted using three different siRNAs, resulting in a total number of 23,349 assays for NF-κB activation. Genes with normalized fold-induction values higher than 1.5 or lower than −1.5 were considered as candidate genes and were further filtered according to the scores of the individual siRNA (Fig. S1C and SI Materials and Methods). Bioinformatics analysis demonstrated significantly enriched gene ontology (GO) terms such as signal transduction, NF-κB signaling, proteolysis, and inhibition of apoptosis among 261 candidate genes from the first screen (Fig. 1A). The 261 candidate genes and a network analysis of their connection to NOD2 are illustrated in Fig. S1D. To narrow down the number of candidate genes, the 261 genes were rescreened using the same setup. The remaining 69 candidate genes were validated in a third screen changing the application of individual siRNAs to a pool of four siRNAs per transcript, which differed in their respective target sequences. A network score for the identified hits versus randomly selected genes was calculated based on information from a protein-interaction database (STRING 8.3) (21). Candidate genes were scored (9, 4, 1, or 0 points) according to their distance to NOD2 in the precalculated matrix (Fig. S1E, Upper), and the ratio of aggregated scores of candidate genes to aggregated scores of randomly picked genes and a false discovery rate (FDR) from 50,000 random simulations (out of the druggable genome pool) were calculated.

This procedure finally generated a set of 20 positive NF-κB regulators (Fig. 1B). We found a continuous increase in ratio from the primary to the tertiary screen, implying a stepwise accumulation of hits that map closer to known NOD2 interactors and a decrease in FDR, indicating relevance of the remaining 20 candidate genes for the NOD2 pathway (Fig. S1E, Lower). Because initiation of NOD2 signaling activates the canonical NF-κB pathway, we sought to distinguish between general NF-κB modulators and those specific for NOD2 signaling. To this end, we rescreened the set of 20 candidate genes using TNF-α stimulation (“counterscreen”; Fig. S1F). Ten out of 20 candidate genes showed no effect upon TNF-α stimulation (blue font) and were thus considered to be specific modulators within the NOD2-signaling pathway. Normalized NF-κB activation of the final 20 candidate genes from all screens is summarized in Table S1.

FRMPD2 Represents a Positive Regulator of NOD2-Mediated NF-κB Activation.

To identify genes relevant for the intestinal epithelium, we rescreened the 20 candidate genes using Caco-2 cells and found 14 of them beyond the cutoff level (bold in Fig. 2A). We next mapped functional relationships of candidate proteins including known NOD2-signaling components based on interaction data from STRING 8.3 (highest confidence; P > 0.9) and BioGRID (http://thebiogrid.org/) (Fig. 2B). We found that 16 out of 20 candidate genes are either directly or by a short distance connected to NOD2, whereas a subset (including the gene FRMPD2) has not yet been associated with immunological functions or NF-κB signal transduction. Next, we selected genes for further follow-up studies. We included genes from the TNF-α counterscreen that were MDP-responsive only (blue in Fig. 2B), and from these we selected genes confirmed as regulators from the Caco-2 screen (bold in Fig. 2B). This assortment method resulted in six genes: RIPK2, EP400, BIRC4, PSMB4, FRMPD2, and EEF1E1. Because RIPK2 and BIRC4 are known modulators of NOD2 signaling (3, 8), only RIPK2 was included as positive control. Moreover, karyopherin subunit β 1 (KPNB1) was integrated in the follow-up as a “canonical” control. Expression of the selected candidate genes was confirmed in different cell lines of epithelial and myeloid origin (Fig. S2A). Next, we measured IL-8 chemokine release of selected candidate genes, which is induced upon MDP-mediated NF-κB activation (22, 23). Knockdown of all six genes significantly reduced IL-8 release in response to MDP stimulation in NOD2-transfected HEK293 cells (Fig. 2C) and in the monocytic cell line THP-1, which endogenously expresses NOD2 (Fig. S2B). Because NOD2 signaling is linked to antimicrobial mechanisms (10, 22, 24), we also analyzed bacterial cytoinvasion using fluorescence-labeled Listeria monocytogenes. Knockdown of all six genes significantly altered bacterial entry and (except for KPNB1) led to an increased intracellular bacterial load (Fig. S2C). For further functional studies, we chose FRMPD2 because of its function related to basolateral membrane targeting in polarized epithelial cells (20). We found that silencing of Frmpd2 in murine IECs significantly decreased keratinocyte-derived cytokine (KC) release from WT cells but not in Nod2 KO cells (Fig. 2D). Viability of cells was assured using the MTS viability assay (Fig. S3). Although other mechanisms could also account for the observed change in intracellular bacteria, we could show using real-time PCR that induction of mRNA levels of antimicrobial peptides/proteins [e.g., human β-defensin 1 (BD-1), phospholipase A2 (PLA2), and human α-defensin 6 (HD6)] was significantly decreased after infection with L. monocytogenes upon FRMPD2 silencing (Fig. S2D).

Next, we determined the influence of FRMPD2 on the NOD2-signaling pathway. Knockdown of FRMPD2 abolished MDP-induced IκB-α degradation, supporting a function upstream of IκB-α (Fig. 2E). Further, the MDP-induced increase in NF-κB activation upon overexpression of FRMPD2 was abrogated when RIPK2 was silenced, indicating that FRMPD2 acts upstream of RIPK2 activation (Fig. 2F). Silencing of RIPK2 and FRMPD2 was confirmed at 72 h after siRNA transfection using Western blot analysis (Fig. S4). To further investigate the specificity of the FRMPD2 effect for the NOD2 pathway, we found that FRMPD2 silencing does not alter NF-κB activation induced by various Toll-like-receptor (TLR) ligands (Fig. S5A), whereas NOD1-induced NF-κB activation was significantly decreased, too (Fig. S5B).

The functional network analysis suggested that FRMPD2 and NOD2 are functionally connected through ERBB2IP (ERBIN) (Fig. 2B), a member of the LRR and PDZ domain family. ERBB2IP has been shown to interact with NOD2 at the basolateral plasma membrane (19, 25). We therefore asked whether FRMPD2 is directly linked to NOD2. Coexpression of FRMPD2 and ERBB2IP as well as FRMPD2 and NOD2 showed a colocalization at the plasma membrane of Caco-2 cells (Fig. 2G).

To assess a potential physical interaction between NOD2 and FRMPD2, we performed coimmunoprecipitation experiments. We demonstrate that FRMPD2 (both overexpressed and endogenous protein) coprecipitates with NOD2 (Fig. 2 H and I) without addition of ERBB2IP. Simultaneous overexpression of ERBB2IP clearly facilitated interaction of NOD2 and FRMPD2 (Fig. 2H). Subsequent NOD2-domain mapping showed that the C-terminal LRR domains and to a lesser extent the central nucleotide-binding domain (NBD) are required for the interaction with FRMPD2 (Fig. S6A). In a reciprocal setup, all FRMPD2 domain constructs containing PDZ domains fused to GFP were capable of precipitating with full-length NOD2, whereas interaction was observed neither with the FRMPD2-FERM domain nor with the FRMPD2-KIND domain alone (Fig. S6B). In addition, we found that FRMPD2 and ERBB2IP interact via the FERM domain of FRMPD2 independent of the presence of NOD2 (Fig. S6C). From these results, we concluded that NOD2, FRMPD2, and ERBB2IP form a tripartite interaction complex that modulates NOD2 signaling.

FRMPD2 Controls NOD2 Signaling from the Basolateral Membrane in Intestinal Epithelial Cells.

To study FRMPD2 function in the context of NOD2 signaling in polarized IECs, we stained for endogenous FRMPD2 and NOD2 in human colonic biopsy sections. We demonstrate colocalization of endogenous FRMPD2 and NOD2 at the lateral membrane and in proximity to the basal membrane of the IEC layer (Fig. 3A). To confirm this finding, we used NOD2- and FRMPD2-transfected polarized Caco-2 monolayers and found them to be colocalized with the basolateral marker Na⁺/K⁺-ATPase (Fig. 3B). It has been previously shown that membrane recruitment of NOD2 is essential for NF-κB activation (18) and that NOD2 is localized at the basolateral plasma membrane (19). We therefore asked (i) whether MDP induces NOD2 signaling in a polarized manner, and (ii) whether FRMPD2 is involved in compartmentalized NOD2 signaling. To this aim, we used small intestinal organ cultures from WT and Nod2 KO mice and disrupted epithelial junctions by calcium depletion to facilitate substance diffusion to the basolateral compartment (26, 27) or left tissue in normal media. MDP-mediated KC release was only induced in calcium-free cultured intestinal segments from WT mice, not in Nod2-deficient mice (Fig. 3C). Next, we stimulated polarized Caco-2 monolayers with MDP at either the apical or basal side. We found that only basal MDP stimulation induced BD-2 mRNA expression (Fig. 3D) and led to IL-8 cytokine secretion (Fig. S7A). Notably, when FRMPD2 was silenced, basal BD-2 and IL-8 induction was completely abrogated. These results suggest that MDP-mediated NF-κB activation and subsequent cytokine release are specifically triggered from the basolateral membrane. Because we found an impact of FRMPD2 on NOD1-mediated NF-κB activation as well, we analyzed the response of polarized Caco-2 monolayers to D-gamma-Glu-mDAP (iE-DAP) stimulation. Comparable to NOD2, we found that only basal iE-DAP stimulation induced IL-8 secretion (Fig. S7B), indicating that FRMPD2 could function as a general modulator of NLR signaling. To test whether FRMPD2 is able to restrict NOD2 to the membrane compartment, we silenced FRMPD2 in Caco-2 cells and analyzed NOD2 distribution in the cytosolic and membrane compartments after MDP stimulation. We found a reduction of NOD2 in the membrane fraction in FRMPD2-silenced cells both at the endogenous level (Fig. 3E) and when NOD2 was transfected (Fig. S7C). In addition, using fluorescence microscopy, we observed diminished NOD2 levels at the plasma membrane when cells were treated with FRMPD2 siRNA (Fig. S7D). To examine the function of FRMPD2 as a scaffold for NOD2 signaling, we checked whether other components of the NOD2 signalosome are recruited to the membrane in an FRMPD2-dependent manner. Notably, we found RIPK2 recruited to the plasma membrane when FRMPD2 and NOD2 are coexpressed in RIPK2-transfected polarized Caco-2 cells (Fig. 3F). As demonstrated previously (28), endogenous RIPK2 was recruited to the membrane in Caco-2 cells in an MDP-dependent fashion (Fig. S8A) in the presence of NOD2. Silencing of FRMPD2 resulted in diminished recruitment of RIPK2 after MDP stimulation (Fig. 3E). Interestingly, we were not able to detect a strong direct interaction of FRMPD2 and RIPK2 (Fig. S8B).

FERM-PDZ2 Domain of FRMPD2 Serves as a Membrane Scaffold and Binds to the LRR of NOD2.

To further dissect the molecular basis of the FRMPD2-scaffolding complex, we studied the localization and function of FRMPD2 and NOD2 domains. First, we analyzed whether a specific FERM-PDZ domain of FRMPD2 is sufficient to restore NF-κB activation after FRMPD2 silencing. As shown in Fig. 4A, overexpression of FERM-PDZ2 increased MDP-induced NF-κB activation significantly, whereas FERM-PDZ1 and FERM-PDZ3 had no or little effect. We next analyzed whether the FERM-PDZ2 domain influences localization of CARD, NBD, and LRR domains of NOD2 and found the LRR to be redistributed from the cytosolic to the membrane fraction when coexpressed with the FERM-PDZ2 domain (Fig. 4B). In addition, we demonstrated that coexpression of the FERM-PDZ2 domain and NOD2 increases RIPK2 membrane localization (Fig. 4C). We observed that the LRR of NOD2 is recruited to the membrane already at low levels of overexpressed PDZ2 domain (Fig. S8C).

Interaction with FRMPD2 Is Abrogated in the Crohn Disease-Associated NOD2 Variant L1007fsinsC.

We next asked (i) whether the spatially defined interaction of NOD2 and FRMPD2 is altered in the CD-associated NOD2 variant L1007fsinsC, which affects the LRR domain of NOD2, and (ii) whether this could in part explain the molecular dysfunction of the variant protein. The NOD2 L1007 frameshift mutation encodes a C-terminal truncated NOD2 protein, thereby abrogating its MDP-mediated NF-κB activation ability (4) and, interestingly, its membrane recruitment as well (18, 28). Notably, wild-type NOD2 immunoprecipitated with FRMPD2, reaching the highest interaction level at 90 min after MDP stimulation, whereas NOD2 L1007fsinsC had a very weak interaction consistently throughout the time-course stimulation study (Fig. 4D).

To further analyze whether FRMPD2 expression is differentially regulated under proinflammatory conditions, we stimulated Caco-2 cells for 2 and 8 h with TNF-α or IFN-γ and found that FRMPD2 mRNA levels were diminished compared with unstimulated cells (Fig. 4E). Next, we determined the FRMPD2 expression profile in normal and inflamed human intestinal mucosa using sigmoid colon samples from healthy individuals, patients with CD, and diseased controls (sigma diverticulitis patients). We found that in comparison with healthy individuals, FRMPD2 was significantly down-regulated in the inflamed intestinal mucosa of diseased individuals and in the noninflamed tissues as well as in the inflamed intestinal mucosa from patients with CD (Fig. 4F). Accordingly, we found that in a mouse model of intestinal inflammation [dextran sodium sulfate (DSS)-induced colitis], Frmpd2 expression was significantly decreased (Fig. 4G). This result remained consistent when relative expression was calculated using Villin as reference gene (Fig. S9) to correct for a potential loss of epithelial cells.