TIR-like NADases act in bacterial immunity and the RNA vault

Summary

Across all domains of life, organisms exploit NAD⁺ metabolism as a central line of defense against invading pathogens. Here, we show that domain of unknown function 4062 (DUF4062) is a widespread family of TIR-like NADases that hydrolyze NAD⁺ to ADP-ribose and nicotinamide. In bacteria, DUF4062 homologs form a previously unrecognized antiphage defense system, which we name Swarożyc, that assembles with the phage portal into a supramolecular NADase complex to induce abortive infection. In eukaryotes, DUF4062 is found in TEP1, which we demonstrate functions as an active NADase within the RNA vault, an enigmatic organelle-like structure. Single-particle cryo-electron microscopy reveals ADP-ribose bound within the shoulder of both reconstituted and human brain vaults, while cryo-electron tomography positions TEP1 along the central axis at the shoulder. Thus, TEP1, like bacterial Swarożyc, functions by depleting NAD⁺, providing new insight into the long-standing mystery of vault function.

Introduction

Bioinformatic analyses of the vast protein sequence space have revealed potential functions for previously uncharacterized proteins. By integrating evolutionary relationships, conserved motifs, and sequence and structural similarities to known enzymes, these approaches uncover distant homologies that can be used to predict biochemical activities for experimental validation. The AlphaFold revolution^1,2, which enables accurate structural modeling for most proteins, has further expanded this capability by revealing deep evolutionary relationships between enzyme families even when their amino acid sequences have diverged beyond the detection limits of the most advanced computational methods.

We have previously used bioinformatic methods to uncover new members of the kinase superfamily^3–6. We reasoned that a similar strategy could be applied to the Toll/Interleukin Receptor (TIR) domain superfamily to uncover hidden functional diversity, catalytic versatility, or new biology. TIR domains, once thought to function solely as protein-protein interaction modules in innate immune signaling, are now recognized as a family that includes active enzymes⁷. Enzymatically active TIR domains mediate immunity by either depleting NAD⁺ to trigger altruistic cell death or generating small-molecule signals that activate downstream defense pathways^7,8. TIR domains catalyze diverse reactions that utilize NAD⁺ as a substrate, including NAD⁺ hydrolysis^9–18, cyclic ADP-ribose formation^19,20, and production of ADP-ribose derivatives that act as second messengers in bacterial^21–24 and plant^25–28 immunity. TIR activation requires oligomerization, which aligns multiple TIR domains in the correct orientation to form the NAD⁺ binding site^{10,11,16,19,29,30}.

Here, we identify DUF4062 as a conserved family of TIR-like NADases. In bacteria, DUF4062 functions in antiphage defense, acting as the executioner by depleting NAD⁺ to block phage propagation. In humans, DUF4062 is found in TEP1, which we show functions as an active NADase within the RNA vault, a mysterious eukaryotic cellular structure³¹.

Results

DUF4062 is a new family of TIR-like domains

To uncover new members of the TIR domain family, we combined sequence and structural similarity searches³² and identified DUF4062, which shows distant sequence similarity to known TIR and nucleoside 2-deoxyribosyltransferase (NDT) domains (Figures 1A and S1A). Nevertheless, it is predicted to adopt a three-dimensional structure that closely resembles canonical TIR domains, such as SARM1 (Figure 1B). Sequence alignments identified a conserved Glu as a likely catalytic residue, corresponding to the known catalytic site of TIR NADases (Figure S1B, 1C). DUF4062 often co-occurs with auxiliary domains (Figures S1D, 1E), consistent with the modular architecture of TIR-containing proteins, which often rely on accessory domains for signal sensing, oligomerization, and TIR activation⁷.

Figure 1. DUF4062 is a new family of TIR-like domains found in eukaryotes and bacterial antiphage defense islands.

(A) CLANS sequence similarity network of representative TIR-like and NDT superfamily proteins (Pfam clans CL0173/STIR and CL0498/Nribosyltransf; BLAST E ≤ 0.01). Each point represents a single protein sequence, and distances reflect relative sequence similarity.

(B) Structures of the human SARM1 TIR domain (PDB: 7NAG; left) and K. pneumonia DUF4062 domain (KpDUF4062 UniParc: UPI001927123B; AlphaFold; right). SARM1 ligand 1AD is shown as sticks.

(C) Phylogenetic tree of 345 representative DUF4062 domains. Branches highlight DUF4062 loci that neighbor defense genes (light beige), reside within antiphage defense islands (light teal), or occur in defense operons (orange). Colors denote taxonomic origin: bacteria (green), eukaryotes (blue), and archaea (red).

(D) Prevalence of DUF4062 domains across analyzed genomes from selected taxa. Archaea (red), Bacteria (green), and Eukaryotes (teal).

See also Figure S1

DUF4062 is found in bacterial antiphage defense islands

Prokaryotic DUF4062 homologs display a broad but patchy distribution and are found across 26 bacterial phyla where they occur in ~1–15% of analyzed reference genomes (Figures 1C, 1D, and S1F). In bacteria, ~20% of DUF4062 homologs reside within predicted antiphage defense islands, and over half neighbor at least one homolog of a known defense gene (Figure 1C). Some are also found in operons together with homologs of characterized defense genes (Figure S1G). Notably, identical DUF4062 protein sequences can appear in diverse genera, often neighboring different phage defense systems. These observations suggest that DUF4062-containing proteins spread by horizontal gene transfer and may function in bacterial immunity.

Bacterial DUF4062 mediates antiphage defense

We selected five DUF4062-containing proteins, that have similar C-terminal auxiliary domains (Figure 2A) and lie within bacterial defense islands (Figure 2B), expressed them in Escherichia coli BW25113 cells and challenged them with a subset of the BASEL phage library³³. We observed strong and broad defense against the tested phages, most notably when DUF4062 proteins from Klebsiella pneumoniae and Enterobacteriaceae were expressed (Figures 2C, 2D and S2A). Phage defense required the active site E97 of K. pneumoniae DUF4062 and its auxiliary domain (Figures 2E and S2B). We named this newly discovered defense system Swarożyc (Swz), after a Slavic god of fire.

Figure 2. DUF4062 mediates antiphage defense by depleting NAD⁺ and blocking phage propagation.

(A) Domain organization of a DUF4062-containing protein from K. pneumoniae with DUF4062 in green and auxiliary domain (AUX) in grey.

(B) Genomic neighbourhood of 5 selected DUF4062 proteins highlighting known or predicted defense genes. DUF4062 is in green and each of the other colors denotes a different characterized defense system (all protein sequence accessions are located in the Supplemental Materials).

(C) Heatmap depicting the efficiency of plaquing (EOP) of DUF4062-containing proteins expressed in E. coli BW25113 against a subset of the BASEL library ³³.

(D) Phage plaques from ten-fold serial dilutions of Bas14, Bas26 and Bas47 on lawns of BW25113 cells expressing RFP or the indicated KpDUF4062^16-C and EcDUF4062 (Enterobacteriaceae/E. coli DUF4062) constructs. Images are representative of 3 biological replicates.

(E) Phage plaques from ten-fold serial dilutions of Bas14 or Bas26 on lawns of BW25113 cells expressing RFP or the indicated KpDUF4062^16-C constructs. Images are representative of 3 biological replicates.

(F, G) Growth of E. coli BW25113 expressing RFP or the indicated KpDUF4062 constructs (renamed Klebsiella pneumoniae Swarożyc: KpSwz) after infection with Bas14 at MOI 10⁻³ (F) or 10 (G), alongside uninfected controls. Curves are representative of 3 biological replicates.

(H) Graph showing NAD⁺ levels in uninfected or Bas14-infected E. coli expressing RFP control, KpSwz^16-C or the inactive E97A mutant.

See also Figure S2

TIR domain–containing defense systems often act through abortive infection, in which phage-infected cells undergo altruistic cell death to protect the bacterial population^34,35. To determine whether K. pneumoniae Swarożyc (KpSwz) induces an abortive defense, we infected liquid cultures expressing KpSwz with phage Bas14 at high (10) and low (10⁻³) multiplicities of infection (MOI). KpSwz rescued growth at low MOI (Figure 2F), whereas high MOI caused culture collapse (Figure 2G) and reduced phage titers (Figures S2C, S2D). Moreover, NAD⁺ levels were reduced upon phage infection in cells expressing WT KpSwz, but not the E97A mutant (Figure 2H). Thus, KpSwz protects the bacterial population by depleting NAD⁺ to trigger abortive infection, thereby blocking phage propagation.

Structural insights into the KpSwz TIR-like domain

We determined a cryo-EM structure of KpSwz^15-C,E97A, revealing a symmetric TIR-like core dimer resembling a pair of flowers with intertwined stems (Figures 3A, S3 and Table S1). These dimers assemble into tetrameric and hexameric bouquets forming a short antiparallel fiber-like structure with a helical turn of ~53° and a rise of ~17 Å (Figures 3B and S3). The TIR-like domain consists of a five-stranded, parallel β-sheet flanked by five α-helices in a β-α-β-α repeating topology (Figure S3C). The core dimer interface is formed between the αA and αB helices (AB interface) (Figure 3C), which is distinct from the AE interface seen in other TIR domains, such as SARM1 (Figure 3D)³⁶. Interactions at the AB interface push the regulatory BB loops, which link βB to αB, into the TIR-like active sites, where they occlude the catalytic pocket and lock the enzyme into an inactive conformation (Figure 3C). Beneath the TIR-like domains, strands from both αE helices form a small antiparallel β-sheet that intertwines with their partner (Figure 3E) before extending into a pair of helical stalks.

Figure 3. The KpSwz TIR-like domain adopts a distinct, bouquet-like higher-order assembly.

(A) Model of the KpSwz^{15-C, E97A} core dimer (modeled residues 15–205, with gaps) shown as a cartoon. Shades of blue represent the two monomers. The TIR-like domains, shaped like flowers, intertwine through a small β-sheet into two helical stalks that resemble flower stems.

(B) Side (left) and top (right) orthographic projections of the hexameric KpSwz mini “fiber”. Each laterally assembled core dimer is shown in two different shades of violet, blue, or green. The side view is perpendicular to the “fiber” axis, whereas the top view represents an arbitrary rotation.

(C) Top view depicting the AB interface of the KpSwz core dimer. α-helical elements are labeled, and interface residues are shown as sticks.

(D) Cartoon depiction of the active TIR domain of SARM1 bound to the inhibitor 1AD (PDB: 7NAK; ⁶⁹). The symmetric dimerization interface (AE) and the asymmetric oligomerization interface (BE) are indicated. Molecules of 1AD are shown as sticks.

(E) Zoomed-in view of the core dimer showing the two chains intertwining beneath the TIR-like domains, stabilizing the dimer and exchanging the αF helices of the stalk region. The interaction is asymmetric, with R18 participating either in salt bridges with D71/D16 or in a cation–π interaction with W176.

(F) Top view depicting the AEB oligomerization interface of the KpSwz tetramer. Interacting helices are highlighted, and interface residues are shown as sticks.

See also Figure S3

The KpSwz core dimers further assemble into tetrameric and hexameric bouquets (dimer/trimer of dimers) (Figure 3B). The αA and αE helices from two core dimers come together (αAEB and αAEC; subscripts indicate protomers), while αBA/D interact with αEC/B to form the AEB oligomerization interface (Figure 3F). The helical stalks also mediate inter-dimer contacts, with stem helices (αF) from protomers A and D forming the core of this interaction (Figure 3B). In the tetramer, each chain contacts all others, generating a distinct higher-order assembly in which the AB and AEB interfaces enforce a head-to-head, autoinhibited architecture.

KpSwz is a portal-activated NADase

Bacterial immune systems are typically maintained in an inactive state until phage infection is detected. In response, phages frequently evolve resistance by mutating the proteins recognized by these systems, thereby evading detection and preventing immune activation³⁷. When challenged with KpSwz, Bas14 phage consistently formed plaques that evaded immune defense. To identify the phage-encoded triggers of KpSwz, we isolated these immune escapers and sequenced their genomes (Figures 4A and S4A). All escapers carried mutations in bas14_0003, which encodes the portal protein (portal^Bas14). Phage portal proteins mediate capsid assembly and genome transfer in tailed dsDNA bacteriophages³⁸ and are known triggers of bacterial immune defense systems^39–41. Therefore, we monitored the growth of E. coli BW25113 cultures expressing portal^Bas14 and KpSwz. Expression of WT portal^Bas14, but not 3 of the 4 escaper mutants led to KpSwz-mediated cell death (Figures 4B, 4C and S4B).

Figure 4. KpSwz is a portal activated NADase.

(A) Phage plaques from ten-fold serial dilutions of Bas14 (parental) or escaper mutants (Clones A-D) on lawns of BW25113 cells expressing RFP, KpSwz^16-C (KpSwz) or the inactive E97A mutant. Images are representative of 3 biological replicates. The mutations are shown on the right.

(B, C) Growth assays of ten-fold serial dilutions of E. coli BW25113 expressing RFP, KpSwz^16-C, or the inactive E97A mutant (KpSwz^EA) in combination with empty vector (EV) or Bas14 portal (portal¹⁴) (B), or with Bas14 portal escaper mutants (C). KpSwz expression was induced with IPTG, and portal expression was induced with arabinose (Ara).

(D) HPLC-UV260 traces depicting nicotinamide (NAM), NAD⁺ and ADP-ribose standards (top), and reaction products generated after incubation of NAD⁺ with KpSwz^15-C (KpSwz, 0.05 mg/mL) alone (middle) or with KpSwz in the presence of excess portal^Bas14 (bottom).

(E) Top (upper) and side (lower) orthographic projections of the symmetry-relaxed KpSwz–portal^Bas14 complex. Coulombic potential maps are overlayed with the atomic model shown as a cartoon, visible in regions of poor density. The 12 portal^Bas14 monomers are uniquely colored, and the 13 KpSwz^AUX domains in the crescent are shown in a grey gradient. Residual density for the unresolved crown region of the portal is visible as grey “dust” at the bottom of the panel.

(F) Cartoon depiction of a single portal (light purple)–KpSwz^AUX (grey) heterodimer. The portal wing, stem, and clip regions are indicated.

(G) Cartoon depiction of the heterogeneous Portal^Bas14–KpSwz^AUX interface. Each KpSwz^AUX domain interacts differently with the portal. The precession of KpSwz^AUX around the portal ring (violet surface) is illustrated by overlaying KpSwz^AUX monomers from all 13 portal–KpSwz^AUX pairs.

(H) Cartoon representation of three neighboring KpSwz^AUX domains, highlighting examples of their interactions with portal^Bas14 within the hairpin anchor (panels a, b) and the α-loop-β module (panels c, d). Each chain is shown as a gray gradient, darkening toward the C-terminus, all chains are uniquely colored in the example panels. The side-by-side assembly is mediated by the rigid HTH/winged helical core, with the C-terminus tucked beneath the N-terminal portion of the domain. All 13 KpSwz^AUX monomers were overlaid on the middle monomer.

See also Figures S4–S6

To determine whether portal^Bas14 directly activates KpSwz, we incubated KpSwz with portal^Bas14 and NAD⁺ and analyzed reaction products by HPLC-UV. NAD₊ was converted to ADP-ribose and nicotinamide, and this activity required the active-site residue E97 and portal^Bas14 (Figures 4D and S4C). Thus, portal^Bas14 is sufficient to trigger KpSwz-induced cell death by activating KpSwz NADase activity.

KpSwz and portal^Bas14 form a supramolecular NADase complex

Portal^Bas14 formed a stable complex with KpSwz on size-exclusion chromatography (Figures S4D). We determined the cryo-EM structure of KpSwz^15-C,E97A bound to portal^Bas14 (Figures 4E, S5 and Table S1). Although the TIR-like domain was not resolved, we observed strong density for the KpSwz auxiliary domain (KpSwz^AUX). The portal assembles into a dodecameric ring, with KpSwz^AUX engaging the stalk that projects from the portal stem (Figures 4E and 4F). In asymmetric reconstructions of the complex, obtained by relaxing the C12 symmetry, 13 KpSwz^AUX subunits assemble into an incomplete ring, forming a crescent around the portal stem and clip (Figures 4E and 4F). Each KpSwz^AUX monomer bound portal^Bas14 in a position slightly offset from its neighbors, creating a heterogeneous interface around the ring (Figure 4G). KpSwz^AUX monomers adopt a helix-turn-helix (HTH)/winged fold and associate side by side through contacts in the α-helical core and interactions between their N- and C-termini (Figure 4H, upper).

KpSwz^AUX monomers engage the portal through flexible portal-recognition modules that include a hairpin anchor and an α-loop-β motif (Figure 4H, lower). The conformations of both modules adjust to the portal surface and change from one unit to the next. A subset of monomers function as anchors, with W238 and Y239 at the hairpin tip buried in a pocket between the wing and stalk domains. Depending on their position within the crescent, W238 and Y239 insert between portal residues R14 and R30 (Figure 4H, panel a), extend toward M21 and K282 (Figure 4H, panel b), or are disordered. Similarly, the KpSwz^AUX α-loop-β motif forms heterogeneous interactions with the helical stalk of the portal surrounding C35 (Figure 4H, panels c, d). Notably, the C35R Bas14 escaper mutation (Figure 4A) fails to activate KpSwz when co-expressed in E. coli (Figure 4C).

Sequencing of the portal gene from 17 Bas14 escaper plaques revealed that all identified escaper mutations localize at or near the KpSwz^AUX-binding interface (Figure S6A). While their exact effects are unclear, they likely alter the position or conformation of the portal^Bas14 stalk to evade detection by KpSwz. However, the E199K portal^Bas14 mutation is not at the KpSwz^AUX-binding interface and likely reflects a distinct escape mechanism, as it remains capable of triggering bacterial cell death when co-expressed with KpSwz (Figure 4C).

Our structural analysis suggests that the flexibility of the KpSwz^AUX-portal^Bas14 interaction allows up to 13 KpSwz^AUX protomers to bind a dodecameric portal^Bas14, forming a supramolecular complex and activating KpSwz NADase activity. Although density for the TIR domain in the KpSwz-portal^Bas14 assemblies was not observed, we propose an activation mechanism whereby portal^Bas14 engagement with KpSwz^AUX pulls on the helical stalks, aligns them, and ultimately repositions the TIR-like domains into an active conformation ⁴². AlphaFold modeling supports this proposed activation mechanism (Figures S6B and S6C).

Together, our results suggest that bacterial DUF4062-containing proteins function in antiphage defense by sensing phage infection, thereby triggering structural rearrangements, activating DUF4062 NADase activity, and inducing abortive infection.

Human TEP1 is an active NADase

Unlike in bacteria, DUF4062 is widely conserved across diverse eukaryotic lineages, including Metazoa (Figures 1C and 1D). Building on our findings in bacteria, we sought to determine whether DUF4062 NADase activity is conserved in eukaryotes.

In human NWD1, NWD2, TEP1, and NPHP3, the DUF4062 domain lies adjacent to a NACHT ATPase (Figure 5A), a domain that facilitates oligomerization and is common to innate immune proteins across all domains of life¹⁵. These proteins contain an apoptotic protease-activating factor 1 (APAF-1)–like helical domain (α), and several also have multiple WD40 repeats. A similar domain organization is present in APAF-1 (Figure 5A), which senses cytochrome c release from mitochondria to induce apoptosis⁴³.

Figure 5. Human TEP1 is an active NADase.

(A) Domain architecture of human DUF4062-containing proteins TEP1, NWD1, NWD2, and NPHP3. APAF1 is shown for comparison. (Telomerase, Ro and vault, TROVE; von Willebrand A, vWA; CARD, C; alpha helical, α).

(B) SDS-PAGE and Coomassie-stained gel showing purified NWD1, NWD2, NPHP3, and TEP1. (C) HPLC traces depicting the reaction products generated after incubation of NAD⁺ with TEP1, NWD1, NWD2 and NPHP3. Protein preparations were used at 0.5 mg/mL.

(D) Cartoon depiction of the TEP1^TIR domain, highlighting unique structural features, including the extended BB loop with an embedded α-helix (green), a loop–helix–loop motif extending from αC (loop–α–loop; blue), and a β-hairpin preceding αE (purple).

(E) Zoomed-in view of the TEP1^TIR active site, highlighting the BB loop (green) and three Arg residues pointing toward a putative sulphate ion. The catalytic E1008 is shown, and sulphate-interacting residues are depicted as sticks. 2Fo-Fc map is presented as a surface.

(F) Graph depicting NADase activity of TEP1^TIR and select active site mutants (1 μM, 15 min, ambient temperature). Reaction products were analyzed as in Figure S4C.

(G) HPLC traces depicting the reaction products generated after incubation of NAD⁺ with the RNA vault containing either TEP1 (red) or the E1008A mutant (green). Vault preparations were used at 0.75 mg/mL.

See also Figures S7 and S8

We purified full-length human NWD1, NWD2, NPHP3, and TEP1 (Figure 5B), incubated them with NAD⁺ and NADP⁺, and analyzed reaction products by HPLC. Only TEP1 hydrolyzed NAD⁺ and NADP⁺, producing ADP-ribose or ADP-ribose phosphate (ADPRP) and nicotinamide (NAM) with activity dependent on the active site E1008 (Figures 5C and S7A). The isolated DUF4062 domain (TEP1^886–1113, hereafter TEP1^TIR) was sufficient to hydrolyze NAD⁺ and NADP⁺ (Figures S7B and S7C). Unlike other TIR-domain proteins such as SARM1, AbTIR, or HopAM1, TEP1^TIR did not generate additional ADP-ribose derivatives (Figure S7D). When using 1 ∝M TEP^TIR, the reaction was slow (kcat = 173.4 min⁻¹), with an apparent Km of ~6.9 mM for NADP⁺ and undeterminable for NAD⁺ (Figures S7E and S7F). However, the specific activity of TEP1^TIR increased proportionally with its concentration (Figure S7G), suggesting activation via concentration-dependent oligomerization. This is consistent with oligomerization-driven activation observed for other TIR NADases¹¹. After SARM1⁹, TEP1 represents the second human TIR domain known to have catalytic activity.

We determined the crystal structure of human TEP1^TIR, which adopted a canonical TIR-like fold (Figure 5D and Table S2). Electron density quality varied throughout the asymmetric unit, with refined B-factors ranging from ~20-150 Ų and progressively worsening from αA/αE toward αB (Figure S8). The BB loop folds into the active site, suggesting an inactive conformation (Figure 5D), and no relevant oligomerization was observed in the crystal lattice. Compared to the KpSwz TIR-like domain, TEP1^TIR contains an extended BB loop with an embedded, highly charged α-helix, a flexible loop-helix-loop motif extending from αC, and a β-hairpin preceding αE (Figure 5D). Within the BB loop, R954 projects into the active site and, along with R908 and R976, coordinates electron density consistent with a sulfate ion present in the crystallization buffer (Figure 5E). Alanine substitutions of R908 and R976 reduce TEP1^TIR activity (Figure 5F). These unique structural inserts distinguish TEP1^TIR from the KpSwz TIR-like domain.

TEP1 is active in the RNA vault

TEP1 was originally identified as a telomerase-interacting protein, suggesting a role in telomerase activity⁴⁴; however, genetic studies show that TEP1 is dispensable for telomerase function and telomere length maintenance in vivo⁴⁵. Instead, TEP1 is thought to function within the RNA vault, a ~65 nm capsid-like ribonucleoprotein particle first identified in rat liver microsomal fractions^46,47. The vault consists of two dome-shaped halves, each assembled from 39 copies of the major vault protein (MVP)⁴⁸, and commonly encloses TEP1, the ADP-ribosyltransferase PARP4 and small non-coding vault RNAs ^49,50. Despite extensive study, its function remains unknown³¹.

To test whether TEP1 is catalytically active within the vault, we reconstituted human vaults with MVP, and TEP1 in HEK293F cells. Purified vaults containing TEP1, but not the E1008A mutant, displayed NADase and NADPase activities (Figures 5G and S7H). Thus, NAD(P)ase activity within the vault is conferred by the TEP1 TIR-like domain.

Cryo-EM analysis of reconstituted and native RNA vaults reveals a bound ADP-ribose

To investigate how TEP1 catalyzes NAD(P)⁺ hydrolysis in the vault, we performed single-particle cryo-EM analysis (SPA) of reconstituted human vaults containing TEP1. We added the NAD⁺ analog benzamide adenine dinucleotide (BAD) to occupy the TEP1 TIR-like domain, and AMP-PNP to occupy the TEP1 NACHT domain. Although this approach yielded high resolution RNA vault maps (Figure S9 and Table S3), the TEP1 density could not be resolved.

Similarly, SPA of reconstituted vaults containing TEP1 and PARP4 (Figures S10 and S11 and Table S4) in the presence of NADP⁺ failed to resolve the internal proteins. However, sub-particle extraction yielded a ~2.9 Å reconstruction of the PARP4 MVP-interacting (MINT) domain bound to the vault wall, complementing a recent report⁵¹ (Figures S10 and S11C). Exhaustive 3D classification indicated an average of ~20 PARP4 molecules per reconstituted vault (Figure S10B). Further sub-particle analysis resolved the symmetry-mismatched cap to ~1.9 Å (C13 symmetry) (Figure S10 and S11D), in agreement with recent reports^52,53.

To assess whether the reconstituted vaults had captured the added nucleotides, we performed focused refinement of the shoulder region, recently reported to bind adenine nucleotides: NAD⁺, ADP and ADP-ribose⁵¹. In vaults containing TEP1, BAD and AMP-PNP, BAD occupied the shoulder nucleotide-binding pocket (Figure 6A, panel a). Interestingly, in vaults containing TEP1 and PARP4 supplemented with NADP⁺, we observed density corresponding to ADP-ribose rather than the expected NADP⁺ or ADP-ribose phosphate (Figure 6A, panel b). The source of this ADP-ribose is unclear; it may result from TEP1-mediated hydrolysis of contaminating NAD⁺ in the sample or co-purified from the cells. In any event, its presence despite excess NADP⁺ suggests a preference for ADP-ribose binding to the shoulder of the vault.

Figure 6. Structural analysis reveals ADP-ribose binding proximal to TEP1 within the vault.

(A) Cartoon representation of a single MVP monomer, with repeats R1–R9 and the shoulder region labeled. Panels a-c shows the contents of the nucleotide-binding pocket in the three analyzed datasets (a, MVP/TEP1/BAD/AMP-PNP; b, MVP/TEP1/PARP4/NADP⁺; c, human brain EMPIAR-10766): BAD and ADP-ribose nucleotides are shown as sticks, along with the corresponding Coulombic densities.

(B) Isosurface renderings of four reconstituted vault particles from a single tomogram. Segmentation was performed using Dragonfly with individual models trained for the vault exterior (grey) and interior (magenta).

(C) 5nm-thick central slices of reconstituted, MVP-only vault particles extracted from high contrast tomograms and aligned using PEET.

(D-E) Central slice (5 nm, D) and isosurface rendering (E) of the MVP subtomogram average, split along the central axis. Note that the material that appears in some of the particles in (C) is averaged out during averaging.

(F-H) 5-nm thick central slices of MVP-TEP1 vault particles (F) and the subsequent subtomogram average (G) with the isosurface rendering depicted in (H). The MVP shell is depicted in grey and the interior density corresponding to TEP1 is colored blue.

(I-K) 5-nm thick central slices of MVP-PARP4 vault particles (I) and the subsequent subtomogram average (J) with the isosurface rendering depicted in (K). The density corresponding to PARP4 is colored yellow.

(L) Composite image with isosurface renderings of the MVP-TEP1 subtomogram average overlaid onto the MVP-PARP4 average. Note the minimal overlap between the densities corresponding to TEP1 and PARP4 in the overlay. Scale bars represent 10 nm.

See also Figures S9–S13

A recent cryo-EM study determined structures of the RNA vault identified as a contaminant in the sarkosyl-insoluble fraction of frontal cortex from a human brain with Type I globular glial tauopathy^52,54 (Dataset EMPIAR-10766). We reprocessed this dataset focusing on the shoulder region and observed an ADP-ribose molecule, despite, to the best of our knowledge, no nucleotides having been added (Figures 6A, panel c, S12, and Table S4). These results indicate that NAD⁺ hydrolysis products are natively present in human RNA vaults.

Cryo-electron tomography reveals the internal protein architecture of reconstituted vaults

Given that vaults containing TEP1 exhibit NADase activity and that we observe its product, ADP-ribose, bound at the vault shoulder, we sought a direct structural readout of TEP1 localization within the vault and its proximity to the nucleotide-binding pocket. Early studies proposed that TEP1 localizes to the cap⁵⁵, but the supporting evidence is limited and the assignment remains uncertain. Therefore, we used cryo-electron tomography (cryo-ET) to visualize reconstituted human vaults. While tomograms of vaults containing MVP, TEP1, and PARP4 readily resolved the external shell, internal features showed limited contrast until Volta phase plate (VPP) imaging revealed reproducible internal densities distinct from the MVP shell (Figures S13A and S13B). Segmentation of individual particles identified recurring ultrastructural motifs within the vault body, consistent with a highly ordered internal organization rather than heterogeneous or freely diffusing cargo (Figures 6B and S13C).

To assess whether these internal features reflect component-specific localization, we performed subtomogram averaging on vaults reconstituted with MVP alone, MVP and TEP1, or MVP and PARP4. Vaults containing only MVP showed minimal internal density (Figures 6C–E and S13D). In contrast, TEP1-containing vaults displayed a distinct density along the central axis in the shoulder region (Figure 6F–H and S13E), while PARP4-containing vaults showed a broader peripheral density near the vault wall extending into the shoulder (Figure 6I–K and S13F), consistent with a recent report⁵¹. Comparison of these independent reconstructions showed that the TEP1-associated density occupies a cavity within the PARP4-associated region (Figure 6L). Because TEP1- and PARP4-containing vaults were reconstituted and analyzed separately, their complementary localization suggests that TEP1 and PARP4 occupy self-defined positions within the vault.

Previous cryo-ET analyses of intracellular vaults concluded that lumenal densities are not ordered and average out during subtomogram averaging⁵⁶. Those studies analyzed native vaults containing heterogeneous cargo and endogenous cellular components. In contrast, our reconstituted systems isolate defined vault compositions and minimize combinatorial heterogeneity. Under these conditions, and using high-contrast cryo-ET, we observe reproducible, component-specific internal densities. These findings suggest that lumenal organization is context- and occupancy-dependent rather than intrinsically random.

Together, these data support a model in which MVP enforces a highly ordered internal architecture, positioning TEP1 near the central axis at the shoulder while restricting PARP4 to belts close to the shoulder and above the waist. The shoulder region of the vault concentrates the NADase activity of TEP1 and the ADP-ribosyltransferase activity of PARP4 near the MVP nucleotide-binding pocket. This spatial arrangement provides a structural framework to guide future studies aimed at elucidating the function of the vault.

Discussion

In this work, we identify DUF4062 as a widespread family of TIR-like NADases present in both prokaryotes, where it functions as the executioner in an antiphage defense system, and in eukaryotes, where it hydrolyzes NAD⁺ to ADP-ribose and nicotinamide in the RNA vault.

Immune mechanisms are frequently shared between bacteria and humans; for example, homologs of eukaryotic immunity- and cell death-related genes have been identified as key mediators of bacterial antiphage defense. These include components of the cGAS–STING pathway^16,57, caspases²⁴, gasdermins^58,59, viperins⁶⁰, signal-transducing ATPases with numerous domains (STAND)¹⁵, SIR2^21,61,62 and TIR domains^21,24,61. Interestingly, vaults appear to be involved in innate immunity, as they are highly expressed in macrophages and dendritic cells⁶³ and promote host resistance to Pseudomonas aeruginosa⁶³ and influenza A virus⁶⁴. Thus, we propose that the TIR-like domain in TEP1, like bacterial Swarożyc, may function as an executioner in a cell death pathway linked to innate immunity.

NWD1, NWD2, NPHP3 and TEP1 share a domain architecture reminiscent of other STAND ATPases, such as APAF-1⁶⁵. APAF-1 contains an N-terminal caspase recruitment domain (CARD), an NB-ARC ATPase/oligomerization domain, and two WD40 propellers that fold back onto the ATPase to prevent oligomerization and subsequent activation. Cytochrome c binding to the WD40 domains relieves this inhibition, promoting nucleotide exchange and assembly of a heptameric APAF-1 ring that recruits and activates procaspase-9⁶⁶. In NWD1, NWD2, and TEP1, DUF4062 occupies the equivalent position of the APAF-1 CARD domain (Figure 5A). We speculate that the TEP1 WD40 domains respond to an unknown signal, which promotes NACHT ATPase oligomerization, leading to activation of the TIR-like domain.

While DUF4062 proteins are far more prevalent in eukaryotes, the broad yet patchy taxonomic distribution of prokaryotic DUF4062 homologs is consistent with the paradigm of defense systems functioning as a “community resource” within a broader pan-immune system, whose components are unevenly distributed across strains and propagated by horizontal gene transfer⁶⁷. Similar to the diversity seen in Thoeris antiphage defense systems that use canonical TIR domains⁶⁸, some DUF4062-containing proteins harbor additional enzymatic domains, whereas others occur in operons alongside diverse enzymes (Figure S1). Thus, the full spectrum of antiphage activities mediated by DUF4062 may involve mechanisms more complex than the NADase activity described here.

In summary, we identify DUF4062 as a previously unrecognized family of TIR-like NADases that function in bacterial antiphage defense. Its presence and activity in human TEP1 suggest that RNA vaults may employ a conserved NAD⁺-depleting mechanism in innate immunity. Further, our findings reinforce the idea that bacterial antiphage defense systems are powerful discovery systems for uncovering biochemical activities of human proteins.

Data and materials availability

All materials developed in this study will be made available upon request.