Animal NLRs provide structural insights into plant NLR function

Abstract

Background The plant immune system employs intracellular NLRs (nucleotide binding [NB], leucine-rich repeat [LRR]/nucleotide-binding oligomerization domain [NOD]-like receptors) to detect effector proteins secreted into the plant cell by potential pathogens. Activated plant NLRs trigger a range of immune responses, collectively known as the hypersensitive response (HR), which culminates in death of the infected cell. Plant NLRs show structural and functional resemblance to animal NLRs involved in inflammatory and innate immune responses. Therefore, knowledge of the activation and regulation of animal NLRs can help us understand the mechanism of action of plant NLRs, and vice versa.

Scope This review provides an overview of the innate immune pathways in plants and animals, focusing on the available structural and biochemical information available for both plant and animal NLRs. We highlight the gap in knowledge between the animal and plant systems, in particular the lack of structural information for plant NLRs, with crystal structures only available for the N-terminal domains of plant NLRs and an integrated decoy domain, in contrast to the more complete structures available for animal NLRs. We assess the similarities and differences between plant and animal NLRs, and use the structural information on the animal NLR pair NAIP/NLRC4 to derive a plausible model for plant NLR activation.

Conclusions Signalling by cooperative assembly formation (SCAF) appears to operate in most innate immunity pathways, including plant and animal NLRs. Our proposed model of plant NLR activation includes three key steps: (1) initially, the NLR exists in an inactive auto-inhibited state; (2) a combination of binding by activating elicitor and ATP leads to a structural rearrangement of the NLR; and (3) signalling occurs through cooperative assembly of the resistosome. Further studies, structural and biochemical in particular, will be required to provide additional evidence for the different features of this model and shed light on the many existing variations, e.g. helper NLRs and NLRs containing integrated decoys.

INTRODUCTION

Innate immune responses are used in organisms across phyla as a first line of defence against pathogens. The corresponding pathways rely on receptors, located both on the cell surface and in the cytosol, called pattern-recognition receptors (PRRs), which sense conserved pathogen-associated molecular patterns (PAMPs) from different organisms, including bacteria, fungi and viruses, as well as intrinsic danger-associated molecular patterns (DAMPs) that result from cellular injury. In humans, the concept of PRRs has a relatively short history (Janeway, 1989), with the first human innate immune receptor only identified in 1997, based on the similarities between the mammalian and the Drosophila Toll proteins (and therefore termed Toll-like receptor, TLR) (Medzhitov et al., 1997). A number of other PRR families have now been identified in mammals, including nucleotide-binding and oligomerization domain (NOD)-like receptors (NLRs), absent in melanoma 2 (AIM2)-like receptors (ALRs), RIG-I-like receptors (RLRs) and the cGAS/STING family (Brubaker et al., 2015).

In contrast to the animal innate immunity pathways, the recognition of the pathogen-secreted effector proteins (termed avirulence [Avr] proteins) within a plant cell by the plant surveillance proteins (termed resistance [R] proteins) was characterized genetically some considerable time ago as the gene-for-gene model of plant resistance (Flor, 1971). However, biochemical characterization of plant immune receptors lags behind the animal systems. Plant immunity is now known to consist of two layers, the perception of pathogens in the extracellular matrix by trans-membrane PRRs (commonly referred to as PAMP-triggered immunity [PTI]) (Dangl and Jones, 2001), and the intracellular recognition by R proteins (now typically referred to as plant NLRs) of effector proteins secreted by pathogens (referred to as effector-triggered immunity [ETI]) (Ellis et al., 2000).

Animal and plant innate immune pathways employ proteins with analogous domains. Despite this, it has been proposed that the similarities in these pathways between plants and animals are a result of convergent evolution following independent origins (Yue et al., 2012). Nevertheless, the similarities allow us to draw parallels between animal and plant systems and infer from the better-characterized system (Kagan, 2014). It is these parallels that we will focus on in this review, in particular attempting to use the more advanced structural knowledge on animal NLR proteins to discuss the mechanism of action of plant NLRs.

INNATE IMMUNITY IN PLANTS AND PARALLELS TO ANIMAL PATHWAYS

The plant innate immune response involves an autonomous, multi-faceted system, whereby individual cells are capable of recognizing specific and non-specific pathogen-derived molecules, to detect pathogen invasion and activate defence (Jones and Dangl, 2006; Dodds and Rathjen, 2010). The PRRs involved in PTI are mechanistically analogous to the trans-membrane TLRs of the animal innate immune system. Plant PRRs and TLRs encode membrane-spanning proteins, the extracellular domains of which are able to detect conserved PAMPs, such as flagellin, lipopolysaccharide (LPS), elongation factor Tu (EF-Tu) or chitin (Felix et al., 1999; Meyers et al., 1999; Peck et al., 2001; Kunze et al., 2004). Ligand perception by plant PRRs and TLRs results in a signalling cascade, culminating in changes in host transcription and upregulation of defence genes (Akira and Takeda, 2004; Chisholm et al., 2006).

Plant pathogens secrete virulence-related proteins, known as effectors, into the host cell during invasion (please note the different meaning of the term ‘effector’ in different contexts). Some of these effectors have been shown to inhibit PTI-related defence signalling pathways, preventing defence responses and allowing colonization of the host (Kjemtrup et al., 2000). In a co-evolutionary response to the use of effectors by pathogens, plants have evolved R proteins and ETI to detect secreted effector proteins and trigger further defence signalling. Most R proteins have been found to contain nucleotide-binding (NB) and leucine-rich repeat (LRR) domains, and are therefore mechanistically analogous to the animal NLRs. Both plant and animal receptors are therefore now collectively referred to as NLRs. However, animal NLRs function as cytosolic immune receptors to detect PAMPs and host-derived DAMPs, rather than the strain-specific pathogen effector proteins that are detected by plant NLRs (Franchi et al., 2009).

NLR DOMAIN ARCHITECTURE AND DOMAIN FUNCTIONS

NLRs typically have a modular tri-domain structure, with distinct roles for each of the three domains (Fig. 1). The family is defined by the central NOD region belonging to the AAA+ ATPase family; this domain usually possesses regulatory and oligomerization functions (Inohara and Nunez, 2001; Dyrka et al., 2014). The C-terminal domain in both plant and animal NLRs typically contains LRRs, but can instead consist of other superstructure-forming repeats (Kobe and Kajava, 2000), e.g. WD/WD40, HEAT, ankyrin or TPR (tetratricopeptide) motifs; this domain often has sensing and auto-regulatory functions (Yuen et al., 2014). The N-terminal domain usually has a signalling function (for clarity, we will refer to this domain as the signalling domain, although in the literature it is often called the effector domain). In plants, it is most often a TIR (Toll-like/interleukin-1 receptor/resistance) domain (also found in animal TLRs) or a coiled-coil (CC) domain (Dangl and Jones, 2001). In animals, it instead usually corresponds to a death-fold structure (death-fold structures include the caspase recruitment domain [CARD], pyrin domain [PYD], death domain [DD] and death effector domain [DED]) or a baculovirus inhibitor of apoptosis protein repeat (BIR) domain (Inohara and Nunez, 2003).

Fig. 1.

Domain architecture of representative plant and animal NLR and STAND proteins. A tripartite ‘signalling domain-NOD-sensor domain’ architecture is conserved throughout most proteins of the NLR family. Variability is often observed for the N-terminal signalling domains, with a suite of different protein-interaction domains occupying this region. In the animal NLRs, we typically see two classes, NLRC (NLR family, containing caspase recruitment domain [CARD]) and NLRP (NLR family containing pyrin domain [PYD]), and additionally the NLR-family apoptosis-inhibitory proteins (NAIPs), in which the N-terminal domain consists of one or more tandem baculovirus inhibitor-of-apoptosis repeat (BIR) domains (Inohara and Nunez, 2003). Plant NLRs are divided into two classes based on their N-terminal signalling domains: coiled-coiled (CC) domains (CNL proteins) or Toll/interleukin-1 receptor/resistance (TIR) domains (TNL proteins). Some variability exists in the NACHT/NB-ARC NOD modules between species; the second helical domain (HD2) found in animal NLRs, which connects the winged-helix (WHD) domain and C-terminal sensor domain, has not yet been identified in plant NLRs. Notably, the apoptotic response proteins APAF-1 (apoptotic protease-activating factor 1) and CED-4 differ from NLRs in the sensor domain; APAF-1 contains WD/WD40 repeats, whereas CED-4 uses separate sensor proteins to detect stimuli (Fairlie et al., 2006). Some NLRs incorporate additional domains to assist with function, e.g. the HMA domains of the rice NLRs RGA5 and Pik (Okuyama et al., 2011; Maqbool et al., 2015), the WRKY domain in the Arabidopsis NLR RRS1 (Deslandes et al., 2003), the CARD in NOD2 (Bertin et al., 1999) and PYD in DEFCAP (Chu et al., 2001; Hlaing et al., 2001).

By containing the defining NODs, NLRs are members of an even broader family of STAND (signal-transducing ATPase with numerous domains) proteins (Leipe et al., 2004; Danot et al., 2009). STAND proteins can bind to diphosphate (NDP) and triphosphate (NTP) nucleotides and cycle between the respective inactive and active states (Tameling et al., 2006; Franchi et al., 2009). There are two main classes of NODs: the NB-ARC (nucleotide-binding adaptor shared by APAF-1 [apoptotic protease-activating factor 1], R proteins and CED-4) domain (van der Biezen and Jones, 1998) and the NACHT (named after the NAIP, CIITA, HET-E and TP-1 proteins) domain (Koonin and Aravind, 2000). Usually, plant and animal NLRs contain NB-ARC and NACHT domains, respectively. The NODs can be divided into several sub-domains (Fig. 1). In a NACHT domain, the N-terminally positioned NB sub-domain is followed by a helical domain (HD1), a winged-helical domain (WHD) and another helical domain (HD2) (Koonin and Aravind, 2000). In NB-ARC domains, a four-helix bundle called ARC1 and a WHD called ARC2 succeed the NB sub-domain. While the ARC1 and ARC2 sub-domains of the NB-ARC domain are structural equivalents of the HD1 and WHD of the animal NACHT domain, plant NLRs appear to lack the HD2 sub-domain based on sequence analyses (Maekawa et al., 2011b). However, sequences at the C-terminus of plant NLRs are poorly conserved, and it remains to be seen whether conserved structural features exist in this region and whether they play a conserved role in plant NLR activation and regulation.

The importance of NODs is exemplified by the many loss-of-function and gain-of-function mutations within this domain. Mutations in the NOD regions of NLRs lead to hereditary inflammatory diseases in animals and auto-necrosis in plants, whereas loss-of-function mutations result in increased disease susceptibilities in both cases (Eisenbarth and Flavell, 2009; Maekawa et al., 2011b). The NODs of animal NLRs are involved in oligomerization essential to receptor function (Inohara and Nunez, 2001). However, while co-immunoprecipitation experiments on the NB-ARC domain of the tobacco NLR protein N show that it may be able to self-associate, there is no structural or biophysical evidence to date demonstrating that plant NLR oligomerization is mediated through this domain (Mestre and Baulcombe, 2006).

The LRR domain in plant NLRs has been implicated in pathogen effector recognition and in some cases physical interactions with the effector protein (Jia et al., 2000; Dodds et al., 2006). Domain-swap experiments involving the removal or substitution of the LRR domain can result in loss of sensitivity to the cognate pathogen effector (Rairdan and Moffett, 2006; van Ooijen et al., 2008). However, it has also been observed that the removal of this domain can lead to auto-necrosis in some NLRs, suggesting that the LRR domain also plays a role in auto-inhibition of plant NLRs (Ade et al., 2007). This auto-regulatory function of the LRR domain has been suggested to be its main function in animal NLRs, as only a few sub-groups of animal NLRs (including NAIPs [NLR family of apoptosis inhibitory proteins]/NLRBs [NLR family, BIR domain-containing and NOD1/NOD2 proteins]) have been shown to interact directly with elicitors (Inohara and Nunez, 2001; Fritz et al., 2006; Faustin et al., 2007; Zhao et al., 2011).

The N-terminal domains of NLRs in both plants and animals are variable and are necessary for signal transduction (Bernoux et al., 2011a; Bai et al., 2012; Proell et al., 2013; Vajjhala et al., 2015). In plant NLRs there are two predominant types: the TIR and CC domains. How these domains relay signals is not fully understood, with no direct signalling partners identified to date. However, self-association has been demonstrated to be necessary for signalling (Bernoux et al., 2011b; Maekawa et al., 2011a), suggesting that this creates a protein scaffold to which downstream signalling partners can bind. In the more extensively studied animal NLRs, this mechanism has been demonstrated to be true. For example, in NLR-containing inflammasomes, NLRC4 (NLR family, CARD containing 4) recruits caspase-1 directly through CARD:CARD interactions (Hu et al., 2015; Zhang et al., 2015; Lu et al., 2016), while NLRP3 (NLR family, PYD containing 3) recruits the bridging adaptor ASC (apoptosis-associated speck-like protein containing a CARD) through PYD:PYD interactions, and in turn ASC recruits caspase-1 through CARD:CARD interactions (Hofmann and Bucher, 1997; Bertin and DiStefano, 2000; Schroder and Tschopp, 2010; Lu et al., 2014). Open-ended higher-order signalosome assemblies of death-fold-domain-containing proteins represent an emerging theme in innate immunity pathways, and extend beyond NLR- and ALR-containing inflammasomes to RLR and CARMA-Bcl10-MALT1 pathways (Qiao et al., 2013; Wu et al., 2014). In general, a higher-order assembly pathway involves a sensor component (e.g. NLRP3), an adaptor component (e.g. ASC) and an ‘enzyme effector’ component (e.g. caspase-1; the reader should again note that the terms ‘sensor’ and ‘effector’ have different meaning in different contexts discussed in this paper). Activation requires the conversion of the sensor component from an auto-inhibited state to an active oligomerized state, and the formation of a large assembly upon nucleation, which eventually activates the effector enzyme(s) through proximity-induced activation (Wu, 2013). Signalling through structured yet open-ended assemblies appears extremely suitable for innate immunity pathways, resulting in a high-magnitude signal once a threshold is reached.

Recent biochemical and functional studies of plant NLRs have provided us with new insights into their modes of activation. However, we lack the structural and biophysical data required to decipher the molecular mechanisms driving plant NLR activation and signalling, and we also lack the components involved in the process. Recent structural studies on the animal NLR proteins NAIP and NLRC4 have been particularly insightful for understanding the activation and regulation of this important class of defence receptors (Hu et al., 2013, 2015; Diebolder et al., 2015; Zhang et al., 2015). In the remainder of this review, we will compare the activation models for plant and animal NLRs (1) to assess to what extent they can inform each other; and (2) to address the feasibility of reconciling the unknowns of plant NLR activation with the NAIP/NLRC4-derived animal NLR activation model in the light of our current biochemical and genetic knowledge of plant NLRs.

MECHANISTIC UNDERSTANDING OF ANIMAL NLR FUNCTION

Although animal NLRs function through the recognition of conserved microbial signatures rather than the polymorphic pathogen effector proteins detected by plant NLRs, there are clearly many parallels in how they operate. There are over 20 NLR members in mammals (Ting et al., 2008; Chavarria-Smith and Vance, 2015). Different NLR members either signal through MAP kinases and activate the transcription factor NF-κB, or oligomerize to form inflammasomes and activate caspase-1, which can lead to pyroptosis or pro-inflammatory cytokine IL-1β maturation (Fig. 2) (Inohara and Nunez, 2003; Bergsbaken et al., 2009; Franchi et al., 2009).

Fig. 2.

Simplified signal transduction pathway diagrams for representative NLRs and STAND proteins. Signalling of NLR proteins is complex and occurs in a family-dependent manner. One of the earliest-characterized NLRs, NOD1 (nucleotide oligomerization domain protein 1), directly interacts with MAMPs such as LPS and activates the transcription factor NF-κB, resulting in the of release pro-inflammatory cytokines (Inohara et al., 1999; Ogura et al., 2001). The NLRC (NLR family, CARD containing) and NLRP (NLR family, PYD containing) family members have also been demonstrated to release pro-inflammatory cytokines through the activation of caspase-1, mediated by CARD/PYD domain interactions directly with caspase-1 or through the ASC (apoptosis-associated speck-like protein containing a CARD) adaptor protein (Dowds et al., 2003; Masumoto et al., 2003). However, there is a large body of evidence suggesting there is no direct interaction between these proteins and their stimuli. Members of the NAIP family of NLRs function as sensors in cooperation with NLRC proteins, which act as adaptors (Lightfield et al., 2008; Halff et al., 2012; Lu et al., 2014). The cooperation of NAIP proteins with NLRP proteins is yet to be determined, but has been speculated (Yin et al., 2015). NLRP and NLRC proteins have been implicated in the activation of pyroptosis through ASC-dependent and independent caapase-1 activation (Fink et al., 2008; Bergsbaken et al., 2009; Schroder and Tschopp, 2010). Activation of NLRP and NLRC proteins results in the formation of large oligomeric assemblies termed inflammasomes (Gross et al., 2009; Halff et al., 2012; Lu et al., 2014; Hu et al., 2015; Zhang et al., 2015). The NOD-containing apoptotic protease-activating factor 1 (APAF-1) has been demonstrated to activate caspase-9-mediated apoptosis through direct interaction with cytosolic cytochrome c, released from a disrupted mitochondrial matrix (Zou et al., 1997; Benedict et al., 2000). Like NLRC proteins, activated APAF-1 forms a large oligomeric assembly, the apoptosome, but in a 1:1 stoichiometry with cytochrome c (Zou et al., 1999; Reubold et al., 2011). Little is known about the signalling pathways of plant NLRs in comparison with their animal counterparts. What is clear is that stimulus detection can occur in a direct or indirect manner, and that self-association of N-terminal signalling domains is vital for NLR signalling (Bernoux et al., 2011b; Maekawa et al., 2011a; Williams et al., 2014,). However, upstream and downstream signalling partners for most plant NLRs remain elusive.

The NLR system that is characterized best involves the NAIP/NLRC4 pair (Zhao et al., 2011; Hu et al., 2013, 2015; Diebolder et al., 2015; Zhang et al., 2015). NAIP proteins act as sensors for bacterial protein ligands such as flagellin (NAIP5) and components of the type-III secretion system (T3SS; NAIP1 and NAIP2) (Lightfield et al., 2008; Kofoed and Vance, 2011; Zhang et al., 2015). Upon detection of the cognate ligands, they co-oligomerize with the adaptor protein NLRC4 (another member of the NLR family). NLRC4 is then able to directly recruit caspase-1 to induce pyroptosis (a type of rapid inflammation-induced cell death) or interact with ASC, which then mediates an interaction with caspase-1, to induce the activation of pro-inflammatory interleukins (ILs) 1β and 18 (Fink et al., 2008; Bergsbaken et al., 2009; Brodsky and Monack, 2009; Schroder and Tschopp, 2010; Lu et al., 2014).

Binding of ATP to animal NLRs has been demonstrated to be necessary for activation and oligomerization. Biochemical studies of NLRP1 demonstrated that the presence of both ATP and the elicitor muramyl dipeptide (MDP) was required for caspase-1 activation in vitro; the absence of either resulted in a lack of activation of caspase-1 (Faustin et al., 2007). While ATP-mediated activation of caspase-1 by NLRP1 has been observed to be concentration-dependent, ATP binding alone failed to promote activation. Furthermore, non-hydrolysable analogues of ATP, AMP-PNP (adenosine 5′-(β-γ-imido) triphosphate tetralithium) and ATP-γ-S failed to initiate NLRP1-dependent caspase-1 activation (Faustin et al., 2007). Further experiments uncoupling the LRR domain from the NACHT domain demonstrated caspase-1 activation independent of MDP, and in vitro experiments with the NACHT domain of NLRP1 in the presence of ATP showed oligomer formation (Faustin et al., 2007).

The available data suggest that the binding and hydrolysis of ATP is mechanistically linked to the activation and regulation of NLRP1, and that the interaction of the NOD and LRR domains contributes to negative regulation of the protein by maintaining an inactive state (Faustin et al., 2007). This notion is supported by crystal structure and biochemical studies of NLRC4, in which deletions of the HD2 and LRR domains lead to constitutively active protein (Hu et al., 2013). Furthermore, disruption of ADP binding through mutations of interacting residues in the WHD also resulted in auto-activation, suggesting that the binding of ATP and ADP by NLRs is important to stabilize the active and inactive conformations, respectively (Hu et al., 2013, 2015; Zhang et al., 2015).

Our structural understanding of the NLR mechanism of action comes from recent crystallography, cryo-electron microscopy and biochemical studies of the NAIP/NLRC4 pair (Hu et al., 2013, 2015; Diebolder et al., 2015; Zhang et al., 2015). Based on these structural studies, the signalling by this pair can be divided into three steps: (1) interaction with the activating elicitor; (2) structural remodelling; and (3) oligomerization (Fig. 3). Structural studies of the auto-inhibited mouse NLRC4 demonstrated that the LRR domain acts as an auto-regulatory domain that maintains the protein in an inactive/ADP-bound state, to prevent premature signalling (Hu et al., 2013). Binding of the cognate ligand (T3SS component PrgJ) to the NOD of NAIP2 causes a conformational change in the protein that creates an oligomerization surface, labelled the nucleating interface. The nucleating interface of NAIP2 is able to interact with a pre-formed interface on the surface of the NB domain of NLRC4, called the receptor interface. NAIP2 binding to the receptor interface of NLRC4 results in a conformational change in this protein, with the NACHT sub-domains WHD and HD2 and the LRR domain rotating ∼90° as a rigid body (Zhang et al., 2015). Conformational changes in NLRC4 lead to the formation of a secondary oligomerization interface, consisting of elements from the NB, HD1 and WHD domains. This interface, referred to as the catalytic interface, is capable of interacting with the receptor interface of another inactive NLRC4 molecule, causing the same conformational changes as observed with the NAIP2 interaction. What results is a unidirectional oligomerization of NLRC4 molecules, nucleated by a single initial NAIP2-activated NLRC4 molecule, and formation of an NLRC4 inflammasome ring that contains 10–12 protomers, including NAIP2 (Hu et al., 2015; Zhang et al., 2015).

Fig. 3.

Structural studies of N-terminal signalling domains of plant NLRs. Structures are displayed in cartoon representation with transparent surface. All protein structure figures were prepared using Pymol (DeLano Scientific LLC). (A) Crystal structure of flax L6 TIR domain (green; Protein Data Bank [PDB] ID 3OZI) (Bernoux et al., 2011b). (B) Crystal structure of the complex of TIR domains from Arabidopsis RPS4 and RRS1 in blue and magenta, respectively (PDB ID 4C6T) (Williams et al., 2014). (C) Crystal structure of the CC domain of barley MLA10 (the two molecules in the dimer are shown in yellow and lime; PDB ID 3QFL) (Maekawa et al., 2011a). (D) Structure of the CC domain of potato Rx (orange) in complex with the WPP domain of RanGAP2 in grey (PDB ID 4M70) (Hao et al., 2013).

The formation of the NLRC4 inflammasome allows the oligomerization of the N-terminal CARDs, and leads to the activation of caspase-1 and ultimately the processing of pro-IL-1β and release of mature IL-1β, and induction of cell death via pyroptosis (Bergsbaken et al., 2009; Schroder and Tschopp, 2010). The prion-like self-activating property of NLRC4 results in an extremely sensitive receptor pathway, where a single MAMP or DAMP can lead to inflammasome formation and defence signalling.

PROGRESS TOWARDS A MECHANISTIC UNDERSTANDING OF PLANT NLR FUNCTION

Plant genomes can encode hundreds of NLR proteins, as a result of gene expansion in higher plants (Yue et al., 2012). Within these, there exist several classes of NLRs, with the major two classes separated by the presence of either CC or TIR domains at their N-termini. In addition, modes of pathogen effector recognition are diverse. Some effector proteins interact directly with the LRR domains of plant NLRs, although only limited examples are known (Jia et al., 2000; Dodds et al., 2006; Wang et al., 2007; Catanzariti et al., 2010; Krasileva et al., 2010; Ve et al., 2013). Other NLRs monitor host proteins that are targeted by pathogen effectors (Dangl and Jones, 2001; van der Hoorn and Kamoun, 2008), whereas certain NLRs function in pairs, where each protein has a specific role in pathogen effector perception and defence signal activation (Cesari et al., 2014b; Williams et al., 2014). In paired plant NLRs, the sensor NLR often includes an additional domain where effectors bind (Cesari et al., 2013; Maqbool et al., 2015; Le Roux et al., 2015; Sarris et al., 2015). These domains have been suggested to resemble the primary effector targets in the host and are subsequently thought of as decoys or sensors that have been integrated into an NLR (Cesari et al., 2014a). Collectively, such diverse mechanisms of action suggest that there will exist more than a single mode of NLR activation and signalling. Here, we attempt to describe the current understanding of the common features of how plant NLRs control defence signalling, based largely on biochemical studies.

The NB-ARC domain has been shown to possess nucleotide binding and ATP hydrolysis activities, similar to animal NLR NACHT domains (Maekawa et al., 2011b; Tameling et al., 2002, 2006; Williams et al., 2011; Bernoux et al., 2016). Several conserved motifs have been defined within NB-ARC and most of these are believed to play roles in binding nucleotides (Meyers et al., 1999, 2003). Like the NACHT domains of animal NLRs, NB-ARC domains have also been shown to interact with other domains of plant NLRs (TIR/CC and LRR domains) (Moffett et al., 2002; Leister et al., 2005; Rairdan and Moffett, 2006; Ade et al., 2007; Fenyk et al., 2012; Slootweg et al., 2013; Wang et al., 2015). Much of what we understand about the function of the plant NB-ARC domain is derived from biochemical studies, functional in planta experiments and structural information inferred from the crystal and cryo-electron microscope structures of related proteins, including Apaf-1 (Riedl et al., 2005; Reubold et al., 2011), CED-4 (Yan et al., 2005; Qi et al., 2010) and NLRC4 (Hu et al., 2013, 2015; Zhang et al., 2015). These studies have shown that the nature of the bound nucleotide determines the conformation of this domain, with ATP-bound CED-4 having a conformation different from that of ADP-bound Apaf-1 (Yan et al., 2005; Qi et al., 2010).

A critical motif within the NB-ARC domain is the ‘P-loop motif’. It corresponds to a flexible glycine-rich loop containing a highly conserved lysine. The crystal structure of Apaf-1 shows that electrostatic interactions of this lysine with the β-phosphate of ADP are essential for nucleotide binding (Riedl et al., 2005). In the plant NB-ARC domains, it is thought that the conserved lysine of the P-loop forms hydrogen bonds important for the interaction with β- and γ-phosphates of a bound nucleotide. Mutation of this conserved lysine residue results in the abolition of function in numerous plant NLRs (Dinesh-Kumar et al., 2000; Bendahmane et al., 2002; Tameling et al., 2002; Bernoux et al., 2011b, 2016). P-loop mutations also perturb the auto-activity observed in gain-of-function mutants of many different plant NLRs (Bendahmane et al., 2002; Moffett et al., 2002; Howles et al., 2005; Tameling et al., 2006; Gabriels et al., 2007; Sueldo et al., 2015; Wang et al., 2015).

Downstream of the P-loop motif is the ‘RNBS-B/sensor 1 motif’, thought to be involved in the differentiation of bound nucleotides, with a proposed role in interacting with the γ-phosphate of ATP bound to the NB domain (Ogura and Wilkinson, 2001; Takken et al., 2006). Both gain-of-function and loss-of-function mutations have been described for this motif, dependent on the nature of the substitution (Salmeron et al., 1996; Dinesh-Kumar et al., 2000; Tornero et al., 2002; Sueldo et al., 2015). It has been proposed, based on structural modelling of the NB-ARC domain of the tomato NLR NRC1 (using the structure of CED-4 as the template), that the arginine in the RNBS-B motif interacts with γ-phosphate of the ATP molecule (Sueldo et al., 2015).

Another important motif in the NB domain is the ‘Walker-B motif’. This motif is thought to be involved in the coordination of a Mg²⁺ ion, a crucial catalytic base for the ATPase activity of the NLR (Iyer et al., 2004; Tameling, et al., 2006; MacDonald et al., 2013). Impairment of ATPase activity in the tomato CC-NLR protein I-2 resulted in a constitutively active protein. The equivalent mutation in the Arabidopsis CC-NLR RPS5 also resulted in auto-activity, but the effect on hydrolysis was not assessed (Ade et al., 2007). The flax TIR domain-containing NLR M did not become auto-active upon the replacement of a conserved aspartate in the Walker-B motif with a glutamate (Williams et al., 2011). Recombinant M protein carrying this mutation was completely occupied with ADP, suggesting a slower dissociation of ADP compared with wild-type M. These different observations may be a characteristic of the differences in the amino acid residues that occupy the Walker-B motif in CC- or TIR-NLRs; an additional conserved aspartate is only present in the Walker-B motif of the TIR-NLRs (Meyers et al., 2003).

The ‘GxP/GLPL motif’ in the ARC1 sub-domain has been proposed to act as a hinge, where the nucleotide-dependent changes converting the closed/inactive to open/active conformation can be executed (Iyer et al., 2004). Structural modelling of plant NB-ARC domains based on the crystal structures of Apaf-1 and CED-4 indicate that GLPL residues are involved in stabilizing the adenosine and ribose backbone of the nucleotide through van der Waals interactions (McHale et al., 2006; Sueldo et al., 2015). Various gain-of-function and loss-of-function mutants have been described in this motif as well, further suggesting that structural changes caused by nucleotide exchange are crucial for NLR auto-inhibition and activation (Dodds et al., 2001; Bendahmane et al., 2002; Sueldo et al., 2015).

The ‘MHD motif’ is one of the most important and best characterized motifs in plant NLRs. Located in the ARC2 sub-domain, the methionine–histidine–aspartate signature is highly conserved in both the NB-ARC and the NACHT domain. In plants, mutation of the conserved histidine or aspartate generally results in a gain-of-function phenotype in planta (Bendahmane et al., 2002; van Bentem et al., 2005; Howles et al., 2005; Tameling et al., 2006; Williams et al., 2011; Bernoux et al., 2016). The histidine in the analogous LHD motif binds the β-phosphate of ADP in Apaf-1 and brings the ARC2 domain into contact with the NB domain (Riedl et al., 2005), stabilizing the closed inactive conformation of the NLR. It has therefore been postulated that MHD mutations release important auto-inhibitory interactions between the NB and ARC2 sub-domains through reduced affinities with ADP, and in doing so facilitate a preference for ATP and the active form of the protein.

The direct measurement of nucleotide binding has only been demonstrated in a few plant NLRs. This is most likely due to difficulties associated with obtaining recombinant plant NLR proteins for biochemical analysis (Schmidt et al., 2007). The first report of ADP/ATP binding was performed on the NB-ARC domain of the tomato NLRs I-2 and Mi-1 (Tameling et al., 2002, 2006). The full-length recombinant barley CC-NLR MLA27 was shown to be ADP-bound when purified from insect cells (Maekawa et al., 2011a), while ADP and ATP binding have been measured in the yeast-produced flax-rust TIR-NLRs M, L6 and L7 (Williams et al., 2011; Bernoux et al., 2016). Various mutants of these flax NLRs provide some evidence that ADP, ATP and no-nucleotide forms correspond to inactive, active and non-functional states of the protein. For example, an MHD-to-MHV mutant of M was found to be constitutively active, and has a preference for binding ATP (Williams et al., 2011). Mutation of the conserved lysine in M, L6 and L7 resulted in non-functional proteins that do not bind nucleotides, while L7, a weaker phenotype variant of L6, was demonstrated to bind more ADP than L6, suggesting a slower rate of ADP dissociation than L6 (Bernoux et al., 2016). It is anticipated that this higher affinity for ADP may prevent the protein from binding ATP and forming a signalling-competent state. Interestingly, it appears that in the case of paired plant NLRs, the sensor NLR does not require a functional NB-ARC domain (Cesari et al., 2014b; Williams et al., 2014), as mutations that would normally render an NLR inactive do not influence the function of the protein.

While there has been some progress elucidating the roles of nucleotide-binding and hydrolysis activity in plant NLR signalling and regulation, further biochemical studies, guided by new structural knowledge, are required to progress towards a mechanistic understanding of plant NLR function. It remains to be seen whether the DNA-binding properties of certain NLRs prove to play a role in plant NLR signalling (Fenyk et al., 2016, 2015).

STRUCTURES OF PLANT NLR SIGNALLING DOMAINS

At present, no three-dimensional structure of a full-length plant NLR protein is available; the only published structural studies of plant NLRs have been performed on the N-terminal TIR (Bernoux et al., 2011b; Williams et al., 2014) and CC domains (Maekawa et al., 2011a; Hao et al., 2013) (Fig. 4), and on the integrated-decoy heavy-metal associated (HMA) domain of the NLR Pik (Maqbool et al., 2015). In a number of cases, the CC and TIR domains have been shown to be capable of independently activating cell-death pathways (Swiderski et al., 2009; Bernoux et al., 2011b; Collier et al., 2011; Maekawa et al., 2011a; Bai et al., 2012; Cesari et al., 2014b; Williams et al., 2014). These domains are therefore thought to have a signalling function.

Fig. 4.

Structure-based model of NAIP/NLRC4 activation. (A) NLRC4 is maintained in an auto-inhibited state through ADP binding and NACHT:LRR domain interactions. Interactions between NLRC4 and NAIP2 bound to PrgJ leads to disassociation of ADP and conformational changes in the protein, leading to the release of the NACHT domain from the LRR and the binding of ATP. (B) Activated NLRC4 molecules are able to trigger activation of inactive NLRC4 molecules, leading to a disc-like assembly. The structures are shown in cartoon and surface representations, based on the structures of inactive NLRC4 (PDB ID 4KXF) (Hu et al., 2013) and activated NLRC4 without the N-terminal CARD (PDB ID 3JBL) (Hu et al., 2015). In each protein chain, the colour changes continuously from the N-terminus (NOD, blue) to the C-terminus (LRR domain, red).

The first plant TIR domain structure was reported for an Arabidopsis TIR domain-only protein (AtTIR) with unknown function (Chan et al., 2010). Since then, four structures of TIR domains from plant NLRs have been published (Bernoux et al., 2011b; Williams et al., 2014). The work on the flax NLR protein L6 revealed that self-association of the TIR domain was required for cell-death signalling. Subsequent studies of the paired NLRs RPS4 and RRS1 from Arabidopsis revealed that a TIR:TIR domain heterodimer interaction was required for the immune function of these NLRs (Williams et al., 2014), despite the involvement of the NB-ARC and LRR domains in maintaining the RPS4:RRS1 pre- and post-activation complexes. It was hypothesized that the TIR-domain hetero-association maintains the RPS4:RRS1 complex in a primed but inactive complex, which would facilitate signalling through the RPS4 TIR domain self-association after effector recognition (Williams et al., 2014).

While the structures of plant TIR domains appear to be structurally similar (Ve et al., 2015), the available structures of CC domains are structurally distinct. The CC domain from the barley powdery mildew NLR MLA10 has a helix-turn-helix fold and crystallizes as a dimer (Maekawa et al., 2011a). On the other hand, in the potato NLR Rx, which confers resistance to potato virus X, the CC domain forms a four-helix bundle. The structure of Rx was determined to be bound to the WPP domain from the Rx co-factor Ran GTPase-activating protein 2 (RanGAP2) (Hao et al., 2013), which is required for Rx function (Tameling and Baulcombe, 2007).

Based on these studies, it appears plausible that the N-terminal signalling domain is maintained in a repressed/protected state through intra-molecular interactions (Bernoux et al., 2016), or hetero-molecular associations in the case of RRS1:RPS4 (Williams et al., 2014). Upon perception of a pathogen effector and nucleotide exchange, conformational changes are expected to release the N-terminal domains from the repressed state and facilitate self-association and protein oligomerization (Bernoux et al., 2016).

RECONCILIATION OF PLANT AND ANIMAL NLR ACTIVATION

The absence of structural information on plant NLRs means we are uncertain of the molecular interactions that take place to allow these proteins to effectively perform their functions. As demonstrated above, there is a considerable knowledge gap between what we know about animal and plant NLRs. However, due to their inherent similarities, it seems possible to infer some aspects of plant NLR function based on the animal NLR structures and structure–function relationships. This section assesses the feasibility of an animal NLR-derived plant NLR activation model, through a reconciliation of functional and biochemical properties of plant NLRs with the structurally characterized activation model of animal NLRs.

Mutations of NODs that impair nucleotide binding demonstrate the critical role of nucleotide binding in the structure and function of NLRs (Tameling et al., 2006; Faustin et al., 2007; Williams et al., 2011; Bernoux et al., 2016). Biochemical studies of animal NLRs have shown the ability of NACHT domains to hydrolyse ATP and have linked nucleotide binding to the transition between active and inactive states (Faustin et al., 2007; Hu et al., 2013, 2015). Studies that incorporated non-hydrolysable analogues of ATP in the NOD show that the protein becomes constitutively active (Faustin et al., 2007). ATP hydrolysis studies of plant NLRs have also demonstrated that the NB-ARC domain is capable of hydrolysing ATP and that, like animal NLRs, impairment of the ability to hydrolyse ATP leads to constitutive activation (Bernoux et al., 2011b, 2016). These observations support the idea that nucleotide binding and ATP hydrolysis in both plant and animal NLRs play similar roles in the stabilization of different activation states of the protein.

Studies have shown that the LRR domains of several plant NLRs are essential for the recognition of pathogen effector molecules. Disruption or removal of the LRR domain leads to a loss of recognition of the effector by the plant NLRs (Dodds et al., 2001, 2006; Rairdan and Moffett, 2006; van Ooijen et al., 2008). By contrast, interactions between the LRR domains and activating elicitors have not been observed in canonical animal NLRs. Instead, elicitor recognition in NAIPs is mediated through interactions with the NACHT, rather than the LRR domain (Lightfield et al., 2008; Kofoed and Vance, 2011). Conversely, the pro-apoptotic NLR-like caspase-activating STAND proteins APAF-1 and DARK have been shown to interact with their elicitors directly, leading to the formation of heteromeric apoptosomes with a 1:1 receptor:elicitor stoichiometry (Purring et al., 1999; Rodriguez et al., 1999; Zou et al., 1999). These proteins have a CARD-NACHT-WD40 architecture, and are implicated in the recognition of cytosolic cytochrome c, with activation leading to oligomerization and caspase-9-mediated apoptosis (Zou et al., 1999). However, while similar to NLRs, these proteins have no observed involvement in innate immunity or defence signalling. MAMP interaction has been demonstrated for the LRR domains of the trans-membrane TLRs of the animal innate immune system. Activating ligand recognition is proposed to bring together the cytoplasmic C-terminal TIR domains in TLRs to form homotypic interactions with TIR domain-containing TLR adaptor proteins and initiate signalling (Akira and Takeda, 2004).

There are clear similarities and differences between the proteins involved in the innate immunity pathways in plants and animals. With respect to plant NLR function, it would appear that the amalgamation of knowledge on the mechanisms of action of domains from animal NLRs, STAND proteins and TLRs could explain the function of full-length plant NLRs. In the absence of both structural information on plant NLRs and knowledge of downstream signalling partners, the development of a cross-inferred model requires great care.

Oligomerization of the N-terminal domain of plant NLRs has been shown to be critical for signal transduction in several cases, and is also observed in the N-terminal CARDs and PYDs of animal NLRs and STAND proteins (Inohara et al., 1999; Bertin and DiStefano, 2000; Ogura and Wilkinson, 2001; Bernoux et al., 2011b; Maekawa et al., 2011a; Williams et al., 2014). The self-association of TIR domains has been demonstrated to be weak, and in many cases it has been proposed that other factors must drive the initial self-association process (Krasileva et al., 2010; Bernoux et al., 2011b; Williams et al., 2014). In the case of the TIR domain of the Arabidopsis NLR RPP1, the auto-active signalling directly correlated with a YFP tag that facilitated dimerization; disruption of YFP self-association prevented RPP1 TIR domain auto-activity (Krasileva et al., 2010). With respect to CC domains, it is difficult to determine how these proteins are involved in activation, with the two available structures differing dramatically (Maekawa et al., 2011a; Hao et al., 2013). In plant NLRs, it is likely that the NB-ARC and LRR domains facilitate oligomerization of the receptor and supply the necessary driving force for N-terminal domain interactions, but this mode of action is yet to be confirmed experimentally.

The studies of the flax NLRs have provided evidence of signalling-domain regulation through binding to the NB-ARC domain (Bernoux et al., 2011b, 2016). Expression of the L6 TIR domain in planta yielded an auto-active phenotype. Addition of the NB domain dampened this auto-activity, and the additions of the ARC1 and ARC2 sub-domains further inhibited the auto-active phenotype (Bernoux et al., 2011b). Changes in the NB-ARC domain of a weaker variant of L6, L7, were able to reconcile elicitor-dependent activity and changed the nucleotide-binding capabilities of the protein, suggesting an inhibitory interaction between NB-ARC and TIR domains (Bernoux et al., 2016). Furthermore, there is a strong body of evidence that implicates the LRR domain in the regulation of the activity of plant NLRs. Constitutive activation of the protein could be achieved by the removal of the LRR domains from NLRs Rx (Moffett et al., 2002; Rairdan and Moffett, 2006) and RPS5 (Qi et al., 2012), consistent with the observations for NLRC4 and NLRP1 (Faustin et al., 2007; Hu et al., 2013). These observations are consistent with the animal NLR model, and suggest that activation of plant NLRs through NB-ARC domain association is conceivable. However, there are differences between animal and plant NLR activation that suggest the two may not act in exactly the same fashion. In the auto-active maize NLR allele Rp1-D21, for example, the LRR domain suppressed the interaction between the NB-ARC and CC domains and facilitated auto-activity, suggesting additional roles for the LRR domains, at least in some plant NLRs (Wang et al., 2015).

A hybrid model reconciling the plant NLR activation model (Bernoux et al., 2016) and the NLRC4 activation model (Hu et al., 2015; Zhang et al., 2015) could effectively explain the activation of plant NLRs (Fig. 5). The common features of both the animal and plant NLR models include (1) an initial inactive auto-inhibited state of the NLR; (2) a combination of the binding of activating elicitor and ATP leading to a structural rearrangement of the NLR; and (3) signalling occurring through a cooperative process we term here SCAF (signalling by cooperative assembly formation). In the proposed model, NLRs exist in a conformational equilibrium between inactive and active states. The inactive state corresponds to a closed conformation stabilized by ADP and with the LRR and signalling domains folded over the NOD and preventing activation. The dissociation of ADP and binding of ATP, and the binding of the activating elicitor, would shift the equilibrium to a more open state compatible with activation. In this model it is not important where the elicitor binds, as long as it favours the active state over the inactive state; it could be at the C-terminal LRR domain, it could be at the NOD region in an area where the C-terminal or signalling domains bind, or even at an ‘integrated decoy’ domain incorporated in some NLRs. The pathogen effector-binding integrated decoy HMA domains of rice proteins RGA5 and Pik best exemplify this, with the HMA domain of RGA5 located at the C-terminal end of the LRR, but instead inserted between the CC and NB domains in Pik (Cesari et al., 2014a; Maqbool et al., 2015; Kroj et al., 2016; Sarris et al., 2016). The ultimate consequence of binding must correspond to a shift of the inactive/active state equilibrium towards the active structure. This active state leads to the cooperative formation of a ‘resistosome’ (Nishimura and Dangl, 2010), which could be a ring-like structure, as in the case of NAIP2/NLRC4, or an open-ended structure, as observed in a number of death-fold assemblies. Downstream, the consequence of this assembly formation must be the activation of an enzyme analogous to caspase-1 in the case of inflammasome signalling, through a proximity-induced activation. Such activation mechanisms are clearly feasible for enzymes such as proteases or protein kinases. There are two key aspects of this mechanism: (1) a stable auto-inhibited state and the prevention of activation under non-pathological conditions, accomplished by the requirement for more than one activating cue (both activating elicitor and ATP binding); and (2) the cooperative, prion-like nature of the activation, which can result in a signal of large magnitude, leading to a fast all-or-nothing response, necessary under pathogen-attack or cell-damage conditions.

Fig. 5.

Model for signalling by cooperative assembly formation (SCAF) of plant NLRs. When unchallenged by pathogen effectors, plant NLRs exist in equilibrium between a closed inactive conformation (stabilized by ADP binding) and an open activated conformation, with the equilibrium strongly skewed towards the former. Both ATP and effector (or effector-induced elicitor in the case of indirect effector recognition) binding stabilize the active conformation, but only when both ATP and effector are bound, the equilibrium shifts sufficiently towards the active conformation to cause downstream events to take place. The active conformer presents new interfaces supporting oligomerization, and the NLRs are able to oligomerize. In analogy with NAIP/NLRC4, a small proportion of active NLRs can seed the conformational transition of further inactive NLRs to the active conformation and allow them to participate in the oligomerization, leading to a cooperative assembly of the resistosome. The downstream adaptors and ‘effector enzymes’ have not been identified in plant systems at this stage.

CONCLUSIONS, OUTLOOK AND FUTURE DIRECTIONS

Structural work on the mammalian NLR pair NAIP/NLRC4 has provided some key insights that may apply to the mechanisms of action of NLRs from different families, including plant NLRs. It is clear that many variations exist of how NLRs can be inhibited and activated; it appears that anything goes when it comes to maintaining and disrupting the auto-inhibited state. Subsequently, this needs to be taken into account when thinking about plant NLR activation. For example, in the case of a plant NLR responding to a direct or indirect effector interaction, one can relatively easily envisage a chain of events that resembles closely the NAIP/NLRC4 model, whereby plant NLRs play both sensor and activator roles. We propose that the ability to activate further inactive NLRs is likely a key component of signal proliferation; relying on multiple rounds of pathogen effector recognition to proliferate HR may not be an efficient approach to address the threat of pathogen colonization. A key question that one might ponder is whether an activated plant NLR can interact with a related non-identical inactive NLR. So-called ‘helper’ NLRs that act downstream of sensor NLRs (Peart et al., 2005; Gabriels et al., 2007; Bonardi et al., 2011, 2012; Maqbool et al., 2015; Wu et al., 2016; Sarris et al., 2016) are compatible with such a mechanism. Intriguingly, at least in the case of the Arabidopsis helper NLR ADR1, the integrity of the P-loop is not required for HR signalling (Bonardi et al., 2011), in contrast to the ATP-dependent NAIP/NLRC4 system. The plant NLR pairs that have been described to contain ‘integrated decoy’ or ‘sensor’ domains (Cesari et al., 2014a; Maqbool et al., 2015; Le Roux et al., 2015; Wu et al., 2015; Kroj et al., 2016; Sarris et al., 2016) may in fact be more difficult to reconcile with the NAIP/NLRC4 model. For instance, RPS4 and RRS1 appear to interact in both pre- and post-activated states, with the protein partners playing key roles in regulation as well as activation. By contrast, NLRC4 is auto-regulated in the NAIP/NLRC4 pair. While a 1:1 stoichiometry was observed for the interaction of the TIR domains of RPS4 and RRS1, it is unclear whether this is the case in the context of the full-length proteins in planta and whether this changes after effector perception. It also appears that in the cases of RPS4/RRS1 and RGA4/RGA5 the pathway to activation requires self-association, which could in theory resemble an apoptosome-like structure. Finally, a range of additional proteins can be involved in the regulation of signalling by plant NLRs, e.g. RIN4 in the case of the NLRs RPM1 and RPS2 (Mackey et al., 2002; Axtell and Staskawicz, 2003).

We believe the mechanism we termed ‘signalling by cooperative assembly formation (SCAF)’ can be generalized to most innate immunity pathways, including the mechanisms previously termed ‘signalling through higher-order assembly formation’ or ‘supramolecular organizing centres’ (Wu, 2013; Kagan et al., 2014) and the nucleated polymerization observed in the case of NAIP/NLRC4 (Hu et al., 2015; Zhang et al., 2015). And while we are still scratching the surface with regard to our structural understanding of plant NLR function, there is little if any evidence that would suggest that plant NLRs cannot operate using such a mechanism.

Structural biology methodologies are rapidly advancing, allowing higher resolution, better models and more informative outcomes. The cryo-electron microscopy approach has advanced to atomic resolution with the implementation of new direct detectors and improved algorithms for particle analysis, and is constantly progressing towards smaller particles (Elmlund and Elmlund, 2015). Great examples of the current power of cryo-electron microscopy include the 2·2-Å resolution structure of β-galactosidase (Bartesaghi et al., 2015) and 1.8-Å resolution structure of glutamate dehydrogenase (Merk et al., 2016). Cryo-electron microscopy also comes with the added advantages of requiring less protein than X-ray crystallography and avoiding the need for crystallization. With the reports of the ability to produce full-length plant NLRs in different expression systems (Schmidt et al., 2007; Maekawa et al., 2011a), cryo-electron microscopy appears like the obvious approach.

To conclude, the high-resolution full-length structure of a plant NLR would provide an answer to many of the functional unknowns and speculations in the field. For example, we would be able to validate the homology models of the NB-ARC domains based on the structures of Apaf-1, NLRC4 and CED-4; better characterize the key functional motifs; and maybe most importantly, provide an atomic blueprint of the inter-domain interactions demonstrated biochemically. The structure of an auto-inhibited form of a plant NLR will provide insights into their regulation, and the structure of a plant NLR:effector complex will define the consequences of NLR activation, including the domain rearrangements and interactions necessary for activation, and the nature of the resistosome complex. Understanding the regulation and activation of plant NLRs will take us a step closer to designing bespoke NLRs with new specificities, e.g. by helping us decide where to insert new integrated decoy domains (Maqbool et al., 2015; Ellis, 2016).