The Arabidopsis TIRome informs the design of artificial TIR (Toll/interleukin-1 receptor) domain proteins

Significance

TIR (Toll/interleukin-1 receptors) domain proteins perform immune signaling across the Tree of Life. TIR domains are enzymes that can process NAD⁺ to generate diverse small molecule signals. In order to better understand TIR-based signaling, we leveraged Arabidopsis natural variation across ~150 TIR proteins to inform the design of artificial TIRs, define features that control output, and tune the output of a natural TIR immune receptor. The engineering of plant immune pathways will enable the optimization of disease resistance to safeguard yields.

Abstract

The TIR (Toll/interleukin-1 receptor) domain is an ancient protein module that functions in immune and cell death responses across the Tree of Life. TIR domains encoded by plants and prokaryotes function as enzymes to produce diverse small molecule immune signals. Plant genomes can encode hundreds of TIR-domain containing proteins—many of which confer important agricultural disease resistance as TIR-NLR (nucleotide-binding, leucine-rich repeat) immune receptors. Despite their importance, how natural variation influences TIR enzymatic output and immunity-associated cell death is largely unexplored. We assayed a complete collection of the TIR domains of Arabidopsis thaliana Col-0 (the “AtTIRome”) to explore variation in TIR metabolite production and cell death signaling. Roughly half of the AtTIRome triggered cell death in transient assays. Artificial TIR proteins designed based on consensus sequences of the AtTIRome’s cell death phenotypic classes revealed polymorphisms controlling variation in TIR cell death elicitation and metabolite production. Structure–function analyses of artificial TIRs revealed that natural variation in the “BB-loop”, a flexible region overlying the catalytic pocket, determines differences in function across Arabidopsis TIR-containing proteins. We further demonstrate that artificial TIRs are functional on an NLR chassis and that BB-loop variation can tune the activity of a natural TIR-NLR protein. These findings shed light on the diversity of TIR outputs and reveal methods to design and engineer TIR-based immune receptors.

Humanity is projected to require 40% more food by 2050, yet around 15-20% of crop yields are lost each year to pests and disease (1, 2). Enhancing plant immune system function can help limit losses, and recent advances in genome editing and protein engineering offer to complement traditional plant breeding efforts to safeguard food production (3–5). Many plant disease resistance traits are conferred by TIR-NLR (Toll/interleukin-1 receptor, nucleotide-binding site leucine-rich repeat) intracellular receptor proteins (6, 7). TIR domains are evolutionarily ancient and often function in prokaryotic and eukaryotic immune systems (7–11). In plants and bacteria, TIR domains act as enzymes to produce diverse small molecules that can activate immunity and cell death. A better mechanistic understanding of TIR functions should enable the rational engineering of plant immune systems and allow for predictable tuning of signal outputs to boost resistance or reduce the agronomic costs of autoactivity.

The plant immune system can sense and respond to extracellular microbial patterns (pattern-triggered immunity), as well as to intracellular virulence factors (effector proteins) injected into plant cells by pathogens (effector-triggered immunity, or ETI) (6, 12). Often, ETI activation results in a localized host cell death known as the hypersensitive response (HR) that limits pathogen spread. While NLR-based ETI disease resistance is a valuable agronomic trait, in some cases the costs of mis-regulated immune activation can be high (13, 14). Mis-regulation of innate immune receptors is particularly likely to occur when immune receptors are moved between genomes, or when engineered to achieve new specificities (15). Thus, rational engineering of NLR innate immunity could benefit from generalizable solutions to reregulate immune outputs.

Plant NLRs function intracellularly as multidomain switches to activate innate immunity (16). The canonical NLR has an N-terminal signaling domain, a central NBS (nucleotide binding site) domain, and a C-terminal LRR (leucine-rich repeat) domain. NLRs are typically activated when the C-terminal LRR domain recognizes pathogen effectors and/or their activities (12). Once activated, conformational changes in the central NBS domain allow the NLR to oligomerize into a “resistosome,” which promotes immune signaling by induced proximity of the N-terminal TIR or CC (coiled coil) domains (17–20). Similar to mammalian inflammasomes, plant resistosomes are wheel-like structures composed of four to six protomer NLR subunits (21–23). While CC domains can directly transduce signals by oligomerizing into ion channels, TIRs signal by oligomerizing to engage their intrinsic enzymatic activities and generate small molecule signals (24–31).

Enzymatic plant TIR domains convert nucleotide-containing substrates like NAD⁺ (nicotinamide adenine dinucleotide), NTPs (nucleoside triphosphates), and DNA or RNA either directly or indirectly into a variety of metabolites, including pRib-AMP/ADP (phosphoribose-adenosine monophosphate/diphosphate), ADPr-ATP (ADP-ribosylated ATP), ADPr-ADPR (di-ADPR), 2’cADPR (cyclic ADP-ribose), 2’,3’-cNMP, (cyclic nucleotide monophosphate), and RFA (ribofuranosyladenosine) (17, 20, 25, 30–34). The enzymology underlying the production of these diverse molecules by TIR domains is poorly understood, but requires a conserved putative catalytic glutamate and oligomerization-dependent coordination of a nearby “BB-loop” motif. (7, 35). The best characterized TIR-generated signals, pRib-AMP/ADP and ADPr-ATP/di-ADPR, are relayed by EDS1 (Enhanced disease susceptibility 1) family proteins into ETI outputs (30, 31). After binding TIR metabolites, EDS1 family protein complexes activate the downstream helper NLR (hNLR) proteins ADR1 (Activated disease resistance 1) or NRG1 (N requirement gene 1), which oligomerize into ion channels to promote transcriptional defenses and localized host cell death (26, 27, 32, 36–39). pRib-AMP/ADP and ADPr-ATP/di-ADPR are apparently low abundance and/or unstable molecules whose detection is presently limited to cryo-EM or LCMS analysis after capture and protection by an in vitro-purified EDS1 complex (30, 31). Other TIR-produced metabolites (2’cADPR, 3’cADPR, 2’,3’cNMP, and RFA) are more easily detectable, but their specific roles as plant signaling molecules remain to be firmly established (7, 25, 32, 40, 41). RFA and 2’cADPR are structurally similar to the EDS1 immune signal pRib-AMP, and as they are readily detected in planta have been used as biomarkers for TIR activity (34, 40, 42). 2’cADPR can be hydrolyzed to pRib-AMP by plant extracts and is a plausible precursor or storage form of this EDS1-activating signal (33, 39). Certain plant pathogens induce RFA accumulation (42, 43), but only recently was RFA characterized as a biomarker of enzymatic TIR activities (34). Any role for RFA as an immune regulator remains to be demonstrated. In the prokaryotic TIR-based Thoeris innate immune system, 3’cADPR functions as a signal to activate antiphage defense, and both 2’ and 3’cADPR are produced by bacterial plant pathogen virulence proteins inside host cells (29, 40, 44–46). How these diverse TIR-produced small molecules regulate plant–pathogen interactions remains a major unanswered question.

Plant TIRs require a conserved catalytic glutamate for catalysis, as well as conserved oligomerization interfaces (the “AE” and “BE” interfaces), yet TIR domains within a genome can share less than 40% identity and occur in various domain architectures aside from canonical TIR-NLRs (7, 25, 47–49). Because dicot plants frequently encode hundreds of different TIR domain-containing proteins, an understanding of how TIR diversity influences their enzymatic profile (product types, abundance), and in turn, immune outputs, will be critical to successfully engineer crop TIR signaling and design customized TIR immune receptors (4, 50–53).

Plant pathogens continually evolve virulence effectors that evade or overcome recognition by host LRR domains of NLR receptors (54–57). Similarly, LRR domains of NLRs have been engineered to restore or expand effector detection (4, 5, 58–61). Recently, engineering of the central NBS domain has been shown as a viable strategy to resurrect defeated NLRs (5). However, the engineering of NLR signaling domains (TIR or CC)—to heighten or dampen immune outputs—is largely unexplored.

Relatively few TIR proteins have been studied, and the vast majority of plant TIR diversity remains unexamined (62). Here, we survey TIR natural variation within the Arabidopsis thaliana Col-0 genome and characterize the metabolite and EDS1-dependent cell death output of 148 TIR domains (the “AtTIRome”). We then leverage the AtTIRome to design artificial TIR proteins with distinct profiles of metabolite production and cell death-triggering function. Finally, we transfer a motif revealed by artificial TIR functions back into the full-length TIR-NLR, RPP1 (Resistance to Peronospora parasitica 1), to demonstrate the tuning of effector-activated TIR-NLRs outputs. These studies suggest that tuning of TIR domains at the BB-loop may be a generalizable strategy to reregulate immune receptors.

Results

Identification of Active TIR domains in the TIRome of A. thaliana (Col-0).

To understand how natural variation influences TIR signaling, we screened each A. thaliana (Columbia; Col-0) encoded TIR (the AtTIRome, 148 TIRs) for cell death induction and metabolite production in Nicotiana benthamiana (Fig. 1 and SI Appendix Fig. S1). The AtTIRome includes TIR proteins with diverse architectures, including TIR-NBS-LRR, TIR-NBS, TIR-only, TIR–TIR, and xTNx/TNP (TIR-NBS/ARC-tetratricopeptide-like repeat). Plant TIR-NLRs are typically activated by effector-triggered oligomerization, but the specific effector triggers for most Arabidopsis TIRs are not known. Given these constraints, we activated the Arabidopsis TIR domains via constitutive oligomerization, enabling us to focus on the intrinsic properties of the TIR domains themselves. Plant TIR domain oligomerization and activation can be artificially driven by fusion to the SAM (sterile alpha motif) domain of the human TIR protein, SARM1 (sterile alpha and TIR motif containing 1) (24, 25, 35, 40, 63). SAM-TIR proteins retain a functional requirement for the native TIR self-association interfaces, the catalytic glutamate (E) residue, and still signal through EDS1, indicating that SAM-TIRs at least partially mimic TIR-NLR activation (24, 25). We fused HA-tagged SAM domains on each Arabidopsis TIR to assess their activity and protein accumulation (via immunoblot) in Nicotiana. In most cases, the initiating methionine defined the N terminus of the TIR ORF (open reading frame) clones. TIR domains of multidomain proteins (e.g., TIR-NLR and TIR-NBS) were typically C-terminally truncated immediately upstream of the easily identifiable conserved Walker A motif (GxxxxGK[S/T]) in the NBS domain. This C-terminal truncation site was chosen based on the observation that the TIR-NBS “TIR+80” linker is required for full autoactivity of the RPS4 TIR (25, 64).

Fig. 1.

Approach to screen natural diversity within the Arabidopsis thaliana (Col-0) TIRome and generate artificial TIR proteins. Each A. thaliana (Col-0) TIR domain was fused to a HA_SAM domain from SARM1 (sterile alpha and TIR-motif containing 1) to promote TIR activation, and screened in Nb.

The core immunity mediator EDS1 relays TIR metabolite signals into outputs such as localized cell death (7, 17, 25, 53). Accordingly, we scored each AtTIR for the ability to trigger EDS1-dependent cell death. Agrobacterium tumefaciens delivering each SAM-TIR construct was infiltrated into leaves of wild-type (WT) Nicotiana benthamiana (Nb) and scored for visible cell death (Fig. 1 and SI Appendix, Figs. S1 and S2). Approximately 47% (70/148) of the AtTIRome consistently activated EDS1-dependent cell death, as defined by necrosis or chlorosis. As expected, none of the 15 AtTIRs that lacked a catalytic glutamate (E) residue triggered autoactive cell death. Roughly 43% of AtTIRs had a catalytic E but did not trigger autoactive cell death (SI Appendix, Figs. S1 and S2). We also examined the accumulation of each SAM-TIR protein using Nb eds1 plants, as EDS1-dependent cell death could increase TIR turnover and confound interpretation (See SI Appendix, Fig. S3 for immunoblots). A strict correlation between protein abundance and cell death induction was not apparent—importantly, some strongly accumulating TIRs did not signal cell death and vice versa (SI Appendix, Figs. S1–S3). Many AtTIRs from atypical architectures like TIR-PP2 (phloem protein domain) or TIR–TIR (“dumbbell” TIRs) also signaled cell death, revealing that SAM fusions can activate diverse TIR domains from non-TNL architectures.

Most tested plant TIRs are dependent on EDS1 to trigger resistance and autoactive cell death (65). A monocot TIR protein of the xTNx (or TNP) class was recently proposed to function independently of EDS1; however, in that study no dicot xTNx proteins were reported to cause cell death (65, 66). Consistent with this, we did not observe cell death driven by the TIR domains of two Arabidopsis xTNx encoding genes (At5G56220 and At4G23440) on either WT or eds1 Nb plants (SI Appendix, Fig. S2).

Nearly half of the 148 AtTIRs signaled EDS1-dependent cell death, despite encompassing a wide range of sequence diversity (See SI Appendix, Fig. S4 for a phylogenetic tree). To understand differences between TIRs that did, and did not, activate cell death, we split the AtTIRs into two classes based on their cell death signaling phenotype: “Death” or “Non-death” (hereafter “Non”). We next assessed TIR domain metrics for the classes such as TIR–TIR interface conservation, BB-loop length/electrostatic charge, TIR domain length, and lengths outside the core TIR domain (SI Appendix, Fig. S5). Among the “Death” group, the BB-loop was typically 17-19 residues, and previously described conserved residues were often present at the AE- and BE- interfaces (e.g., SH/SF and G residues, respectively), while the “Non” group was more variable for either metric (25, 67). However, we did not detect substantial differences in predicted core TIR domain length, BB-loop electrostatic charges, or lengths outside the core TIR domain (SI Appendix, Fig. S5). The above TIR regions were delineated according to multiple sequence alignment (MSA) and mapping onto the RPP1 (Resistance to Peronospora parasitica 1) TIR-NLR structure, a subset are shown as examples in (SI Appendix, Fig. S5) (17).

As noted above, plant TIRs produce diverse metabolites including 2’cADPR, 2’,3’-cNMP, RFA, pRib-AMP/ADP, and ADPr-ATP/di-ADPR (7, 30–32). pRib-AMP/ADP and ADPr-ATP/di-ADPR activate EDS1-signaling; however, no methodologies to detect these molecules directly within plant extracts have been reported (30, 31). As 2’cADPR can be metabolized to pRib-AMP, it is a plausible proxy for this EDS1-activating signal (39). Therefore, we assessed AtTIR enzymatic activity via qualitative LC–MS detection of 2’cADPR and ribofuranosyladenosine (RFA) within Nb eds1 leaf extracts (SI Appendix, Fig. S1; chromatographs shown in SI Appendix, Fig. S6). The analytic method used is unable to discriminate between 2’RFA and 3’RFA, so our usage of “RFA” is agnostic as to the ribose–ribose linkage (34). Three different metabolite profiles were observed among the 70 AtTIRs that consistently signaled EDS1-cell death: RFA-only, RFA and 2’cADPR, or none detected. 53% (37/70) elevated RFA, however, only 20% (14/70) generated 2’cADPR above background levels. Notably, any AtTIR that produced 2’cADPR also generated RFA. The other 47% (33/70) of death signaling TIRs did not elevate RFA (or 2’cADPR) above background. This suggests that while these 33 TIRs exhibit reduced capacity to produce RFA/2’cADPR, they are still sufficient to stimulate EDS1 (potentially via catalysis skewed toward ADPr-ATP/di-ADPR). Although all LC–MS profiling included transitions for pRib-AMP/ADP and ADPr-ATP/di-ADPR, these were not detected, indicating that they were absent or below the limit of detection. 3’cADPR was not detected in any of the transient assays. The metabolite 2’,3’-cAMP was consistently not elevated among the initial batches of AtTIRome samples (which included both cell death positive and negative classes), and so was not measured in the remaining samples (SI Appendix, Fig. S7). 2’,3’-cNMP production by TIR proteins has only been reported in the context of TIR-only proteins or isolated TIR domains; therefore, it is unclear if our addition of an octameric SAM domain might limit the formation of the indeterminate-length TIR filament structures associated with 2’,3’-cNMP generation (32).

The AtTIRome Can Inform the Design of Functional Artificial TIR Proteins.

Given that the AtTIRs could be divided into phenotypic classes, we next attempted to leverage TIR diversity to better understand structural motifs associated with TIR function across the AtTIRome. Particularly, we asked whether our AtTIRome phenotypic classes could inform the design of artificial TIR proteins with predictable phenotypes (Fig. 2A). We determined the consensus sequence of the AtTIRome groups, “Death” and “Non,” as well as the consensus of all 146 AtTIRs (hereafter “All”). Any sequences outside of the predicted core TIR domain were trimmed, and the two xTNx TIRs were excluded, as they are proposed to be mechanistically distinct (65). The resulting artificial consensus TIRs for “Death,” “All,” and “Non” were 160 to 162 residues in length each. AlphaFold3 models indicate that the artificial consensus TIRs had catalytic E residues, canonical TIR–TIR interfaces, and had similar residues within the putative catalytic pocket proposed by Manik et al. (29) (SI Appendix, Fig. S8). Given this conservation of TIR features, the “Non” consensus was outwardly similar to enzymatically active TIRs and lacked predictable loss of function polymorphisms. Fig. 2B provides AlphaFold2 predictions for each artificial consensus TIR, as compared to the RPP1 TIR domain structure (confidence scores in SI Appendix, Fig. S8) (17). The majority of variation among the artificial consensus TIRs was found at the BB-loop or the αC and αD helices, which are located away from TIR–TIR interaction interfaces (see SI Appendix, Fig. S9 for sequence alignment).

Fig. 2.

Artificial TIRs designed from AtTIR consensus groups are enzymatically active and signal cell death. (A) Artificial TIR design using AtTIR consensus groups. (B) (Left) Structure of activated TIR domain of TIR-NLR RPP1 (17). BB-loops colored green, catalytic glutamates shown orange, AE/BE interfaces shown purple. (Center-Right) AlphaFold2 predictions of artificial consensus TIRs, “All,” “Death” or “Non.” N: N terminus, C: C-terminus. (C and D) Nb WT or eds1 expressing artificial TIRs as HA_SAM fusions. Control RBA1, SAM_SARM1-TIR (residues 478 to 724), or empty vector (EV; 35S:YFP). WT leaves shown ~5 dpi; eds1 ~6 dpi. Framed numbers denote cell death positive/total replicates per set. (E) Anti-HA immunoblot in Nb eds1 at 40 hpi. (F) NAD⁺-detection assay in Nb eds1 at 40 hpi. RFU: relative fluorescence units. (E/A): TIR catalytic glutamate to alanine substitutions. (G) LC–MS traces of 2’cADPR or RFA in Nb eds1 at 40 hpi. Similar experiments performed at least three times. Separate letter class indicates P < 0.05 or better by One-way ANOVA and Tukey’s HSD.

Next, we synthesized ORFs encoding the “Death,” “Non” or “All” artificial TIRs, and cloned them into SAM constructs, as used to screen the AtTIRome. In agreement with their class phenotypes, both “Death” and “All” triggered EDS1-dependent and catalytic E-dependent cell death in Nb, while “Non” did not (Fig. 2 C and D). The artificial consensus TIR proteins all accumulated protein by immunoblot (Fig. 2E). Curiously, “Death” caused a minor chlorosis in Nb eds1 plants, which required the catalytic glutamate (SI Appendix, Fig. S10). Because cellular NAD⁺-depleting TIRs like SARM1 or AbTir cause EDS1-independent cytotoxicity, we examined whether “Death” might also diminish NAD⁺ levels in Nb eds1 plants (Fig. 2F) (40). However, unlike SARM1-TIR, “Death” did not deplete NAD⁺, indicating that the apparent chlorosis is NAD⁺ depletion-independent, or may be triggered by NAD⁺ reduction below the limits of detection (Fig. 2D). Despite being divergent at only 15 residues, the artificial consensus TIRs recapitulated the cell death phenotype of their class constituents and provide testable hypotheses on how natural variation might influence TIR signaling.

Because plant TIRs activate EDS1 via production of small molecule signals, the “All” and “Death” artificial TIRs were likely active enzymes. Accordingly, we used LC–MS to qualitatively profile their measurable metabolic outputs (2’cADPR, RFA) (Fig. 2G). Similar to many AtTIRs in SI Appendix, Fig. S1, “All” elevated RFA but not 2’cADPR, while “Death” elevated both 2’cADPR and RFA, as well as an unknown peak (m/z 542) at retention time ~8.10 min (Fig. 2G). Metabolite elevation by either “All” or “Death” was dependent upon the catalytic glutamate (Fig. 2G). We also assayed TIR-only versions of “All” and “Death” that lacked SAM domains, and both generated metabolites and signaled EDS1-dependent cell death, indicating that their function does not necessarily require SAM-enforced oligomerization (SI Appendix, Fig. S10). “Non” did not activate EDS1-dependent cell death, and as expected, it did not elevate metabolites relative to EV (empty vector) controls (Fig. 2G).

In addition to designing artificial TIRs by consensus, as a control, we tested whether artificial TIRs generated by randomly picking residues at each position (from a pool of 146 AtTIRs) might also be functional (SI Appendix, Fig. S11). We synthesized 10 artificial “random TIRs” ranging from ~48 to 58% identity to the closest natural occurring TIR (SI Appendix, Fig. S11). Each random TIR had a catalytic E. When expressed as a SAM fusion in WT Nb, none caused apparent cell death or chlorosis (SI Appendix, Fig. S11). Together, these findings suggest that a consensus-based approach can reliably design functional artificial TIR proteins.

Variation in the BB-Loop of Artificial TIRs Controls Cell Death Signaling and Enzymatic Activity.

Several TIR regions, such as the BB-loop, the TIR–TIR interfaces, and catalytic pocket, can impact self-association, enzymatic activity, and signaling (24, 25, 49, 67, 68). The “All” and “Non” artificial TIRs are ~90% identical, possess catalytic E residues, and have identical TIR–TIR interfaces (Fig. 2). Yet “All” produces RFA and activates cell death, while “Non” does not. An amino acid sequence alignment revealed variation at the BB-loop (Region I) and two alpha-helices (helices αC and αD; referred to as Regions II and III, respectively) (Fig. 3A and SI Appendix, Fig. S9). SI Appendix, Fig. S12 shows these variant regions in context of a TIR tetramer (RPP1 model). To understand which polymorphisms were sufficient to restore activity, we swapped each of the three “All” regions, and combination of regions, into “Non” (Fig. 3 B, C, E, and F). BB-loop (Region I) transfer from “All” to “Non” was sufficient to restore signaling and elevate RFA production (Fig. 3 B, C, and F); however, adding either helix αC or αD variation (Region II or III) did not restore function, nor enhance signaling or metabolite production, if combined with BB-loop transfer (SI Appendix, Fig. S12). We then asked which variant residues within the “All” BB-loop were necessary for functionality in the context of the Non TIR domain. Removing the Q or E residues from the BB-loop abolished cell death triggered by the “Non” TIR domain containing the “All” loop (Fig. 3D). Similarly, adding back the “Non”-specific G residue also abolished cell death triggered in the same context (Fig. 3D). The T/V substitution had no impact on cell death in the same context (Fig. 3D). Together, this indicates that multiple BB-loop polymorphisms are necessary to restore signaling to “Non.”

Fig. 3.

BB-loop variation determines cell death and metabolite production by the artificial consensus TIRs. (A) AlphaFold2 predictions of the artificial consensus TIRs, “All” or “Non,” highlighting 11 of 15 variant residues. BB-loop colored green, catalytic glutamate shown orange, and AE/BE interfaces shown purple. Variant residues colored red and indicated by arrows. N: N terminus, C: C-terminus. (B–D) Nb WT expressing “All” or “Non” artificial TIRs, and BB-loop or regional swaps. I, II, or III refers to a variant region in “All” vs. “Non.” Nb WT shown ~5 dpi; eds1 at ~6 dpi. Framed numbers denote cell death positive/total replicates per set. (E) Anti-HA immunoblot from Nb eds1 at ~40 hpi. (F) LC–MS traces of 2’cADPR or RFA from Nb eds1 at ~40 hpi. (G and G’) Nb WT or eds1 expressing artificial TIRs, including “All” with the BB-loop of “Death.” (H) LC–MS traces of 2’cADPR or RFA from Nb eds1 at ~40 hpi.

We next examined whether the BB-loop from the “Death” artificial TIR modulated enzymatic outputs. As such, we replaced the loop of “All” with that from “Death” and detected a new peak at ~8.10 min, as well as a gain of EDS1-independent chlorosis similar to that observed with the “Death” TIR protein (Figs. 3 H, G, G’ and 2D). Collectively, these results suggest that the AtTIRs comprising the “Non” group are enriched for BB-loops that are suboptimal for triggering cell death, and that BB-loop variation can influence TIR metabolic outputs.

In Vitro Characterization of Artificial TIRs.

As noted above, methodologies to directly detect pRib-AMP/ADP or ADPr-ATP/di-ADPR in planta are lacking, so we also examined TIR outputs in vitro, using recombinant proteins. We first performed kinetic NADase assays and observed that “Non” was inactive compared to the “All” or “Death” TIRs, and that, consistent with in planta assays, only “Death” produced 2’cADPR (Fig. 4A and SI Appendix, Fig. S13 A–C). Consistent with in planta assays in Fig. 3, swapping the “All” BB-loop into “Non” promoted NAD⁺-hydrolysis, while placing the “Death” BB-loop into “All” enhanced the rate of NAD⁺ hydrolysis over 10-fold (Fig. 4A). Each in vitro NADase timepoint was assessed using ¹H NMR spectroscopy and compared with available standards (Nam, 2’cADPR, and NAD⁺) (see SI Appendix, Fig. S13A for spectra). Interestingly, the 2’cADPR peak for “Death” increased exponentially until 4 h, and then increased linearly despite no more NAD⁺ being present, suggesting this TIR might also bind and cyclize ADPR (SI Appendix, Fig. S13B).

Fig. 4.

In vitro NADase assays of artificial and natural TIRs. (A) Kinetic NADase assay of artificial TIRs. (B) Structure of ADPr-ATP metabolite; 1,048.06 Daltons. (C) MALDI-TOF MS detection of ADPr-ATP from in vitro end-point reactions. Blue (1,049.06 m/z) corresponds to ADPr-ATP, while 1,048.97 m/z (black) is in background controls NAD⁺ + ATP and “Non.” Y-axis as 10⁴ (black) or 10⁵ (green).

After establishing robust in vitro TIR assays, we examined ADPr-ATP production from endpoint reactions via MALDI-TOF (matrix assisted laser de/ionization—time of flight)-MS and ¹H NMR (Fig. 4 B and C and SI Appendix, Fig. S13 D–I). As controls, we included the RPP1 and RPS4 TIR domains—which are known to produce ADPr-ATP—as well as several TIR-only proteins, TN7, TN10, and TN11 (30, 31, 47). SI Appendix, Fig. S13D shows NAD⁺ consumption for each examined TIR. Notably, “Non” and TN10 did not accumulate Nam, consistent with a lack of NADase activity and being unable to trigger cell death in planta (SI Appendix, Fig. S13D). All enzymatically active TIRs (except TN11) generated MS peaks matching predictions for ADPr-ATP (+H⁺) (1,049.1 m/z) (Fig. 4 B and C). Curiously, all TIRs that produced ADPr-ATP peaks - as detected by MALDI-TOF-MS—also produced two unique peaks (U1, U2) detectable by ¹H NMR (SI Appendix, Fig. S13 E and F). Although “Death” was the only artificial TIR that could produce 2’cADPR, all surveyed Arabidopsis TIRs with enzymatic activity generated 2’cADPR (SI Appendix, Fig. S13F). Together, the MALDI-TOF-MS and NMR spectra show that both natural and artificial TIRs can generate diverse metabolites in vitro, including EDS1-activating signals like ADPr-ATP.

Several active TIRs produced products with m/z 462.1 and 638.1 (SI Appendix, Fig. S13 G–K) in addition to 1,049.1 (SI Appendix, Fig. S13 G–J). Interestingly, 638.1 m/z was observed within 1,049.1 ms/ms spectra, while 462.1 and 638.1 share similar fragment ions to ADPr-ATP. Hence, 638.1 and 462.1 species could represent di-ADPR, pRib-ADP, or ADPr-ATP degradation. Standards of di-ADPR, ADPr-ATP, and pRib-ADP may help to explore these mystery species. Their biological relevance and varied abundance between tested TIRs remains to be understood.

Artificial TIR Variation Can Inform TIR-NLR Engineering.

Because plant NLRs couple effector detection to signal generation, we examined whether the artificial TIRs were functional if attached to a biologically relevant NBS-LRR chassis (6, 12). We chose RPP1 as a model system, as it is well understood, with a cryo-EM structure and genetic information on both recognized and unrecognized alleles of the pathogen effector ATR1 (A. Thaliana recognized1) available (17, 69). We replaced the native TIR of RPP1-WsB (residues 85-248) with the “Non,” “Death” or “All” TIR, and coexpressed each in Nb with the cognate ATR1-Emoy effector, or a nonactivating allele, ATR1-Emwa (69) (Fig. 5 A and B and SI Appendix, Fig. S14). Relative to RPP1 WT, the artificial TIR-NLRs showed reduced accumulation, although “Death”-NLR was more stable than “All”-NLR or “Non”-NLR. LC–MS analysis revealed that both “Death”-NLR and “All”-NLR’ elevated RFA. Curiously, “Death”-NLR did not elevate 2’cADPR, unlike expression as a SAM fusion or TIR-only protein (Figs. 2G and 5C and SI Appendix, Fig. S10). We also compared the “All”-NLR and “Death”-NLR for cell death signaling and found that the “Death”-NLR triggered effector-dependent cell death in 100% of Nb leaves, in contrast to 12% for the “All”-NLR (Fig. 5B). These results indicate that artificial TIRs can be functional on an NBS-LRR chassis and have outputs largely consistent with those observed in the context of an orthologous SAM fusion.

Fig. 5.

Incorporating BB-loop variation from artificial TIRs can modulate TIR-NLR signaling; artificial TIRs exhibit effector-mediated activation on NLR chassis. (A) Strategy to replace native RPP1 TIR domains with artificial TIRs. TIRs shown yellow, NBS-LRRs gray, and ATR1 effectors pink. (B) Nb WT expressing artificial TIR-NLRs at OD₆₀₀ = 0.01. Cognate ATR1 (Emoy) allele triggers RPP1 WsB while noncognate Emwa does not. (C) LC–MS traces of 2’cADPR or RFA from Nb eds1 expressing artificial TIR-NLR fusions and ATR1 (Emoy) at ~40 hpi. (D) Monomer from RPP1 TNL resistosome showing TIR (yellow), NBS-LRR (gray), and ATR1 (pink). Top: Artificial TIR BB-loop variations. Bottom: RPP1 BB-loop mimics of artificial TIR variations. (E) Nb WT expressing RPP1-WT or RPP1-loop variants at OD₆₀₀ 0.005; ATR1-Emoy at OD₆₀₀ 0.10. RPP1 constructs contain C-ter YFP. Leaves imaged ~5 dpi and framed numbers denote cell death positive/total replicates per set. (F) Anti-YFP immunoblot in Nb eds1 at ~40 hpi. (G) LC–MS traces of 2’cADPR or RFA from Nb eds1 leaves at ~40 hpi.

We next asked whether variation within the artificial TIRs could inform TIR-NLR engineering. For example, might incorporating variation within the “Non” BB-loop into a natural TIR-NLR dampen signaling? To do so, we replaced two BB-loop residues (N119, N120) of the TIR-NLR, RPP1, with the corresponding loop residues from “All,” “Non,” or “Death” (Fig. 5 D, E, and G). These two residues were selected as N119 is conserved between “All,” “Non,” and RPP1, while a glutamate (aligning to RPP1 N120) was required for cell death signaling by the “Non” TIR containing the BB-loop from the “All” TIR (Figs. 3D and 5D). When cognate ATR1-Emoy was coexpressed, we observed that RPP1_{“Non”-Loop} triggered cell death in less than 10% of leaves, as compared to RPP1_{“Death”-Loop}, RPP1_{“All”-Loop} or RPP1_WT, (Fig. 5E). Similarly, RPP1_{“Non”-Loop} did not elevate RFA, as compared to RPP1_WT, RPP1_{“Death”-Loop}, or RPP1_{“All”-Loop} (Fig. 5G). The RPP1 BB-loop variants accumulated similarly, as revealed by anti-YFP immunoblot, consistent with the “Non”-type loop polymorphisms affecting enzymatic activity (Fig. 5F). Thus, a BB-loop with reduced output discovered in the context of artificial TIR overexpression could inform tuning of native TIR-NLR activity in response to its cognate pathogen trigger.

Discussion

TIR domains are conserved components of innate immune systems across the Tree of Life. In both prokaryotes and plants they function by generating diverse small molecule signals that activate downstream immune pathways (7, 48, 52, 53). To better understand the diversity of TIR domain function, we characterized each A. thaliana (Col-0) TIR domain and found that differences in metabolite production (type, abundance) and cell death signaling are common. Further, we demonstrate that the Arabidopsis TIRome has sufficient information to enable the design of functional artificial TIRs, and that variation between the “Death” and “Non” classes can guide TIR engineering to tune natural TIR-NLR outputs in a predictable manner. Assessing TIR domain outputs using alternate oligomerization chassis (e.g., NBS-LRR), as well as generating artificial consensus TIRs based on other phenotypes of interest (e.g., product type, abundance, domain architecture, etc.) could provide further insights into TIR function and artificial TIR design. Our results suggest that the BB-loop may be a useful generic target for controlling TIR-NLR activity and mitigating dysregulation resulting from NLR transfer between species or varieties.

Based on our current understanding of TIR pathways in plants, any cell death-signaling TIR should generate EDS1-activating metabolites like pRib-AMP/ADP or ADPr-ATP/di-ADPR (30, 31). Because current methodologies cannot directly detect these metabolites in planta, we measured 2’cADPR and RFA as biomarkers to assess TIR enzymatic activation (25, 34, 42, 43). Published literature contains conflicting results for 2’cADPR. Based on the structure, 2’cADPR was proposed as a plausible precursor of pRib-AMP (7, 29). However, early experiments expressing AbTir (a prokaryotic TIR that produces 2’cADPR) found that the resulting cell death was EDS1-independent, and thus a link to 2’cADPR was not supported (40, 41). More recently, Yu et al. found that AbTir can induce oligomerization of the Arabidopsis EDS1/ADR1 complex via pRib-AMP production and that 2’cADPR can indeed be hydrolyzed by plant extracts to pRib-AMP (39). Although Yu et al reports that 2’cADPR activates the Arabidopsis ADR1 immune module, any 2’cADPR-dependent signaling intrinsic to N. benthamiana remains to be demonstrated. While all active plant TIR domains produced 2’cADPR in our in vitro assays, we did not detect its cleavage into pRib-AMP. As Yu et al. found that 2’cADPR conversion into pRib-AMP in vitro required plant extract, presumably a non-TIR enzyme is required for linearization. A full understanding of 2’cADPR’s function as a precursor or storage form of pRib-AMP remains to be established. While RFA is structurally similar to 2’cADPR and pRib-AMP (SI Appendix, Fig. S1), there has been no demonstration to date that RFA has an immune function. Because RFA elevation in planta requires the TIR domain’s catalytic glutamate residue, RFA is at least reflective of TIR enzymatic activity as a biomarker (34). Accumulation of detectable 2’cADPR and RFA were not 100% correlated in planta: while all 2’cADPR-producing TIRs produced RFA, several TIRs accumulated RFA in the absence of detectable 2’cADPR. How diverse TIR products are related and how their catalysis is determined remains open questions. RFA accumulates in Arabidopsis following Pseudomonas syringae DC3000 inoculation (42, 43). However, in these studies, RFA increases could reflect the enzymatic activity of host TIRs as well as that of the bacterial TIR effector HopAM1 (29, 40, 45). The TIR effector HopAM1 requires enzymatic activity for virulence and produces 3’cADPR, yet how 3’cADPR and/or manipulating host NAD⁺ mediates pathogenesis is not understood (29, 44, 45). 3’cADPR does not stimulate EDS1-mediated cell death, although 3’cADPR activates the Thoeris prokaryotic TIR immune system (29, 40, 41, 46). Our finding that RFA was detected more widely from active TIRs in planta than 2’cADPR reveals it as a useful biomarker, yet RFA was not elevated by ~40% of the cell death signaling AtTIRs. It is plausible that some of the cell death-inducing TIRs that do not make detectable 2’cADPR or RFA are producing relatively more ADPr-ATP/di-ADPR to activate the SAG101/NRG1 branch of the EDS1 pathway. This will be important to understand as it might allow retuning of TIRs to signal preferentially for resistance via EDS1/PAD4/ADR or for cell death via the EDS1/SAG101/NRG1 branch (70). This possibly occurs for some natural TIRs such as TN11 (AT1G72940), which in our SAM-TIR system did cause cell death, and in our in vitro assays, exhibited NADase activity, produced 2’cADPR but not ADPr-ATPI Hence, the development of methodologies that can directly measure EDS1-activating signals in planta is essential to enable plant TIR research.

Almost half of the AtTIRome did not trigger cell death in Nicotiana. Does this represent the underlying biology (e.g., ADR vs NRG signaling as discussed above), or are there reasons that our screen might artifactually underrepresent AtTIRome signaling? There are several potential artifacts in our screen. For instance, we used autoactive SAM domain fusions to induce homo-oligomerization of the TIR domains, which might not effectively activate every TIR, particularly if hetero-oligomerization is required (71). How TIR oligomerization promoted by various domain architectures (including the SAM domain used here) influences TIR catalysis and product specificity is not well understood. Additionally, we noted substantial variation in SAM-TIR protein accumulation that might not reflect native TIR protein stability and/or accumulation. Given that certain nonaccumulating SAM-TIRs could still trigger cell death, it is unclear to what extent we should expect protein accumulation to impact the phenotypes measured. For instance, while the TN7 (At1g72900) TIR produced ADPr-ATP in vitro, the SAM_TN7 TIR accumulated poorly in planta and did not trigger cell death. Regardless, nonenzymatic TIR-containing proteins could still fulfill valuable roles as sensors, similar to RRS1, and/or regulate the activities of tetrameric TIR complexes via hetero-oligomerization (17, 20, 71). Despite potentially underrepresenting TIR activity, our analysis of artificial TIRs in a transient expression system was able to pinpoint two residues in the BB-loop that predictably decreased activity of the natural TIR-NLR RPP1 after effector activation. This suggests that our approach, albeit artificial, was sufficiently robust to provide useful information about natural TIR-NLR receptor function.

Our findings add to a growing list of reports illustrating BB-loop influences on enzymatic TIR signaling (24, 25, 40, 68). The artificial consensus TIRs “All” and “Non” have identical TIR–TIR interface and catalytic residues, yet BB-loop differences prevent signaling by “Non.” The in vitro enzymatic activity of the RUN1 TIR domain can be enhanced by adding positive arginine residues into the BB-loop (24), yet the “Death” TIR is enriched for negative BB-loop charges and transferring the “Death” TIR BB-loop into “All” increased enzymatic activity. BB-loop alterations can have different impacts on plant and prokaryotic immune TIRs. For instance, shortening and replacing several BB-loop residues within the prokaryotic ThsB TIR enables autoactive signaling, whereas similar replacements within autoactive plant TIRs abolish activity (40). A recent study by Song et al indicates that the BB-loop and TIR–TIR association interfaces can contribute to the formation of TIR condensates (68). Intriguingly, the BB-loop modifications introduced by Song et al. strongly impaired TIR-only function, but only weakly impacted the NADase activity of the TIR-NLR RPP1, presumably reflecting contributions of the NBS-LRR chassis to TIR oligomerization and the distinct structures for TIR-only and TIR-NLR proteins (32). Even within the TIR-NLR class BB-loop functions appear complex, for instance mutation of the BB-loop in the TIR-NLR CHS3 revealed differential requirements between Arabidopsis accessions (72).

The artificial consensus TIRs “All,” “Non,” and “Death” had distinct metabolite and signaling profiles. “All” only elevated RFA, while “Death” produced RFA, 2’cADPR, and an unknown peak at ~8.1 min. Curiously, when ‘Death’ was fused to the RPP1-NLR, it no longer produced 2’cADPR or the 8.1 min peak, unlike TIR-only or SAM fusion contexts. Such context dependency requires further investigation. Isolated TIR domains are reported to generate 2’,3’-cNMP, but how domain architecture might influence metabolite production is not understood (32). The activation of artificial TIR-NLRs by cognate effectors suggests that other “customized” TIRs could likely perform regulated signaling upon an NLR chassis. Given that the RPS4 TIR can signal as a NLRC4 chimera suggests that artificial TIRs could likely function on atypical recognition domains, including customized LRR-nanobody detection domains (“pikobodies”) (41, 58).

While we designed artificial TIRs using a consensus approach on a mesoscale set of ~150 TIR domains, artificial proteins of various families can also be generated using large language models (LLMs) trained on millions of protein sequences (3, 73, 74). For instance, Madani et al. produced functional artificial lysozyme proteins with identities as low as 31% to WT HEWL (hen egg white lysozyme), although artificial lysozyme proteins with identities closer to 70% of WT displayed higher activity (3). Resources like the OpenPlantNLR initiative, which houses ~60,000 NLR sequences—including thousands of TIRs from various species—could help design new artificial TIRs (https://zenodo.org/communities/openplantnlr/). Similarly, a large set of species-specific TIRs as in the Arabidopsis pangenome could provide considerable design power (62). The compact size of TIR domains (~160 residues) allows economic large-scale screening of natural TIR domains and artificial TIR designs. While thousands or millions of sequences are available for many protein families, phenotyping proteins remains a bottleneck. Our consensus-based approach indicates that smaller scale studies, employing low/medium-throughput in vivo phenotyping, can be informative. In this study, our consensus approach binned the TIRome into only two subclasses based on cell death elicitation. This may be masking TIR structure–function properties that could be revealed with more refined subclasses. Designing artificial consensus TIRs based on other metrics like kinetics, product type, or domain architecture may provide further insights into TIR function.

Although this study demonstrates the production of artificial immune signaling domains, key questions remain about plant TIR engineering. For instance, which TIR variations dictate substrate usage, or the production of specific metabolites? How do different TIR signals translate to actual crop success, and is yield benefited by TIRs with broad (2’-cADPR, RFA, 2’,3’-cNMP, pRib-AMP/ADP, ADPr-ATP/di-ADPR) or more narrow metabolite outputs? And can similar engineering principles be applied to TIRs that reportedly function outside of EDS1-signaling pathways (34, 65)? If successful, TIR engineering may provide new modalities to enhance resistance or mitigate autoimmunity by decreasing cell death that could complement engineering of recognition or activation conferred by NBS-LRR domains (SI Appendix, Fig. S15).

Enzymatic TIRs that make unusual signaling molecules continue to be discovered (75–77). Prokaryotes encode diverse assortments of enzymatic TIRs, and a mechanistic study of these should offer further insights into how particular TIR regions and their overall context in a protein complex impart product and/or substrate specificities (8, 9, 11, 29, 44, 45, 78). An in-depth investigation of sequence divergent TIRs possessing similar kinetic and metabolite profiles will likely reveal new TIR regions/motifs that influence catalysis beyond those uncovered by our consensus-based approach. Artificial protein-based approaches may also facilitate the design of immune proteins belonging to other families, such as those containing CC domains. As our mechanistic understanding of plant immune receptors increases, protein engineering offers to complement traditional crop improvement techniques with rationally engineered disease resistance.

Methods

Detailed information regarding each procedure is provided in SI Appendix, Supporting Methods.

Gene Cloning and DNA Vector Construction.

A. thaliana accession Col-0 TIR ORFs (AtTIRs) were cloned from cDNA (or synthesized), BP-cloned into shuttle vectors, and LR-cloned into previously described binary vectors as in refs. 25, 39. Mutagenesis was performed using polymerase incomplete primer extension (PIPE) and Q5 High-Fidelity polymerase (New England Biolabs, Ipswich MA). The HA_SAM oligomerization domain was described previously (25). See SI Appendix for full methods and TIR Amino acid sequences.

Phylogenetic Tree Construction/Multiple Sequence Alignment.

A. thaliana TIR domain protein sequences were aligned via Muscle with MEGA7 (79). Sequences were C-terminally truncated directly upstream the Walker A motif (GxxxxGK[S/T]) (the reported TIR+80 region or RPS4 activity (64). Maximum likelihood trees were constructed via MEGA7.

Transient Expression in Nicotiana benthamiana.

Agrobacterium tumefaciens strain GV3101 was infiltrated into leaves of ∼4-5 wk old Nicotiana benthamiana (Nb) at OD₆₀₀ = 0.80, except for titration analyses as noted in figure panels. For EDS1-dependent cell death assays, Nb leaves were covered with foil for 2 d and scored at 5 dpi. GV3101 was transformed and induced as in ref. 39. N. benthamiana were grown in Conviron or Percival plant growth chambers at 25 °C, 70% relative humidity, and 15 h photoperiod with a light intensity of ~80 μE m⁻² s⁻¹.

In Planta LC–MS Analysis.

LC–MS/MS analysis of leaf extracts was carried out on a Waters TQ-XS triple quad mass spectrometer with a Waters Acquity UPLC-C18 column similar to ref. 39, and run alongside standards of 2’cADPR or 2’RFA.

Immunoblots.

Three 6 mm discs from three different Nb leaves were harvested, extracted, and immunobloted [anti-HA (3F10, Roche) or anti-GFP (Cat# 1181446001, Roche)] as in ref. 39.

In Planta NADase Assays.

NADase assays on leaf discs were performed as in ref. 39.

Artificial TIR Design and Consensus Determination.

Arabidopsis TIR group (“Death,” “All,” or “Non”) sequences were aligned using Muscle within MEGA7 (79). Consensus sequences were calculated with Gene-Calc (https://gene-calc.pl/sequences-analysis-tools/consensus-sequence). “Random” artificial TIR sequences were generated by randomly selecting an amino acid from the AtTIRome for each position of the core TIR domain. Artificial TIR ORFs were codon-optimized and synthesized by Twist Biosciences.

In Vitro Expression Constructs and Protein Purification.

RPS4, RPP1, and SNC1 TIR domain constructs were generated as previously described (24, 29). AtTN7 (At1g72900), AtTN10 (At1g72930), and AtTN11 (At1g72940) were cloned into pET-28a(+). Artificial consensus TIRs ORFs were synthesized by Genscript and cloned into pET-30a(+). TIR domain proteins were produced using BL21 E. coli similar to refs. 24 and 29, concentrated to at least 5 mg/mL, and stored at -80 °C.

In Vitro NADase Kinetic Assays.

Recombinant TIRs were incubated with 2 mM NAD⁺, 2 mM ATP, and 15 µL of Ni-NTA beads at 25 °C. Time points were taken at 0, 0.5, 1, 1.5, 4, 22, and 45 h. Spectra were recorded using a Bruker Neo and NAD⁺ cleavage was validated by quantifying Nam % relative to 0 h NAD⁺ intensity.

End Point Product Assays.

TIR kinetic NADase assays were quenched and analyzed on MALDI-TOF-MS alongside standards of cADPR, 2’cADPR, 3’cADPR, pRib-AMP, ADPR, ATP, NAD⁺, or Nam in reaction buffer.

MALDI-TOF MS.

In vitro TIR NADase reactions were analyzed on a Bruker TIMS TOF Flex, as in (24, 29).

Protein Structure Modeling.

Protein structure models for artificial TIR proteins, “All,” “Death” and “Non” were generated using AlphaFold3 (80) and analyzed with PyMOL.

Statistical Analyses.

Multiple comparisons were analyzed via one-way ANOVA with post hoc Tukey HSD. Significance indicated by compact letter display (CLD); separate letter classes indicate P < 0.05 or better.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Code S01 (R)

Data, Materials, and Software Availability

All study data are included in the article and/or supporting information.