Filament formation and NAD use by unusual human FAM118 sirtuin enzymes

We encountered human FAM118B and FAM118A when searching for uncharacterized sirtuins in InterPro and Pfam databases, where they are both annotated with a SIR2-like (SIR2_2) domain. Analyzing them in 2019 with HHPred⁴⁰, which searches the Protein Data Bank (PDB) for entries with sequence and secondary-structure similarity, confirmed that both FAM118s are significantly similar to human SIRT1 (PDB 4KXQ↗, E values of 2.9 × 10⁻¹² for FAM118B and 5.3 × 10⁻¹³ for FAM118A).

Repeating the HHPred analysis in 2022, after several bacterial sirtuin structures were deposited to the PDB, predicted a stronger similarity between FAM118s and these bacterial enzymes than with canonical sirtuins. Notably, ThsA from Streptococcusequi ranked as the strongest hit for both FAM118B and FAM118A (PDB 7UXT↗, E values of 1.2 × 10⁻²⁵ for FAM118B and 2.3 × 10⁻²⁶ for FAM118A). At the time of this article's submission, an even better score was observed for Escherichiacoli ThsA (PDB 8WC0↗, E values of 9.6 × 10⁻²⁶ for FAM118B and 1.3 × 10⁻²⁶ for FAM118A).

To further explore the relationship between FAM118s and other sirtuins, we constructed an evolutionary tree of representative sirtuin sequences (Fig. 1b and Extended Data Fig. 2). Beyond the canonical sirtuins (the SIR2 subfamily)—found primarily in eukaryotes with few bacterial representatives—the tree features two further distinct subfamilies. One is SirTM, whose members, found mainly in fungi and bacteria, catalyze protein ADP-ribosylation, as we demonstrated previously^23,41. The other, a much broader group, corresponds to the SIR2_2 category used by InterPro and Pfam (the name we adopt here for this subfamily) and to the recently proposed class of SIRims (SIR and immunity)³⁹. The SIR2_2 subfamily consists of proteins from bacteria, archaea and eukaryotes, including human FAM118s alongside bacterial sirtuins ThsA^7,27,28,32, DSR1 and DSR2 (refs. ^29,37).

We next investigated how closely FAM118s resemble other SIR2_2 members. Sequence-based searches for FAM118B and FAM118A homologs revealed two sets of SIR2_2 proteins: a vertebrate-specific set with high sequence identity to human FAM118s (≥41%), termed henceforth the FAM118 clade, and a set of proteins with lower identity (<32%, and often <25%) found in a broader set of species including prokaryotes and corresponding to non-FAM118 SIR_2 proteins.

Mapping the amino acid residues conserved within the FAM118B clade onto an AF3 model of human FAM118B revealed conservation not only within the hydrophobic domain core and predicted NAD-binding site, expected for proteins sharing the same overall fold and NAD-binding capability, but also across extensive surface patches (Fig. 1c, top). The latter may imply shared homomultimerization or interaction properties. In contrast, when mapping, onto the same model, the residues conserved between FAM118s and the closest among non-FAM118 SIR2_2 proteins, we observed conservation only within the hydrophobic core and the predicted NAD-binding site (Fig. 1c, bottom).

A further phylogenetic analysis showed that, within the FAM118 clade, all proteins can be distinctly categorized into one of two clusters, corresponding to FAM118B-type or FAM118A-type proteins (Extended Data Fig. 3).

Looking specifically at the FAM118B orthologs, we observed that they are present in almost all vertebrates, from jawless fish to humans (with a few exceptions, notably in the genera Myxine, Ascaphus, Alectura and Opisthocomus), demonstrating the emergence of FAM118B before the common ancestor of all or most extant vertebrates. The FAM118A orthologs, on the other hand, are consistently missing from jawless fish, suggesting that FAM118A emerged by duplication of FAM118B after the separation of jawless and jawed fish lineages. Moreover, FAM118A proteins have apparently been lost in a number of other fishes and amphibians (with exceptions such as the catshark Scyliorhinuscanicula or the toad Pelobatesfuscus) and are more consistently retained among reptiles, birds and mammals, where an FAM118B and FAM118A pair is a norm. Rare exceptions that feature FAM118A but not FAM118B include newts, guan and tinamou birds and possibly alligators, crocodiles and gavials, which have low-quality FAM118A sequences.

We set out to biochemically and structurally characterize a representative member of the FAM118 clade, focusing on human FAM118B. We recombinantly produced and purified full-length FAM118B (referred to simply as FAM118B) and its slightly truncated variant lacking poorly conserved and likely flexible terminal regions (designated FAM118B_24–334), both with a noncleavable His₆ tag.

To elucidate FAM118B structure, we first attempted crystallography. Although both constructs crystallized, they did not diffract. In light of the insights below, this may be because of the difficulty of achieving orderly crystal packing from large multimeric units⁴².

The part of each FAM118B protomer that is well resolved in the density maps is limited precisely to residues 28–325 for both FAM118B and FAM118B_24–334, indicating that the flexibility of N and C extremities extends slightly beyond the parts truncated in the shorter construct (Fig. 2g). Noteworthily, the resolved region matches exactly the part of the AF3 model with good predicted local distance difference test (pLDDT) values. The final structures of all individual protomers of both FAM118B and FAM118B_24–334 are similar to each other and align well with the AF3 model.

As expected, the FAM118B sirtuin domain comprises the large Rossmann-fold subdomain and a small α-helical subdomain but lacks a zinc-binding module (Fig. 2h). The zinc-binding module, missing from FAM118B and other SIR2_2 proteins, is required for the deacetylation activity of canonical sirtuins^10,44, stabilizing the overall fold and/or facilitating protein substrate binding. Given that SIR2_2 proteins, including FAM118B, are likely ancestral to canonical sirtuins³⁹, the zinc-binding module may be a later development associated with the apparent shift from NADase to protein deacetylation activity and from homomultimerization, which may help stabilize the fold, to a merely monomeric or dimeric state^45,46.

To assemble into FAM118B filaments, a part of the small α-helical subdomain (the head surface) of one protomer binds to the opposite end of the large subdomain (the tail surface) on a neighboring protomer (Fig. 3a,b).

The large head-to-tail interaction interface (>1,000 Ų) is predominantly polar in nature, featuring numerous potential hydrogen bonds and salt bridges, along with several hydrophobic residues (Fig. 3c). It likely has an electrostatic component, with a negatively charged head (Fig. 3d) and a positively charged tail (Fig. 3e) surface.

The head and tail surfaces account for a substantial part of the surface conserved among FAM118-clade proteins (Fig. 3f). Moreover, at the sequence level, key head and tail residues, notably including the central hydrophobic residues F122 and L32, are well conserved among FAM118B and FAM118A proteins across vertebrates (Fig. 3g), suggesting that the ability to form filaments with this particular topology is their shared, ancestral feature. Conserved patches extend beyond the head and tail surfaces (Fig. 3f), indicating additional interactions conserved among FAM118s.

We also tested whether AF2 or AF3 could predict the filamentous arrangement to assess whether it is encoded in the evolutionary and structural features that these algorithms recognize. Importantly, our cryoEM structures were not included in the PDB training sets for either AF version and, as mentioned, no previously characterized sirtuin forms filaments with this topology. Modeling human FAM118A or FAM118B homodimers or an FAM118A:FAM118B heterodimer, we found that AF2 and AF3 reproducibly predict a head-to-tail dimer that can be extrapolated to a filament (Fig. 3h for AF3 and Extended Data Fig. 8 for AF2). The interface predicted template modeling (ipTM) scores and other confidence metrics of these predictions—falling within an intermediate range (Extended Data Fig. 9)—would have left these predictions uncertain without experimental validation; in our case, these models align with and support the experimental data. Of note, a recent large-scale AF2-based analysis of homomultimerization within the human proteome predicted human FAM118A (but not FAM118B) as potentially multimerizing with a translational symmetry⁴⁷, consistent with our predictions.

As cryoEM showed that both FAM118B and FAM118B_24–334 have similarly strong filamentation tendencies—suggesting that the N-terminal and C-terminal tails are dispensable for this property—we used the FAM118B_24–334 construct to investigate filament formation in solution using biophysical techniques.

Analyzing instead WT FAM118B_24–334 with dynamic light scattering (DLS), we observed, at 10 µM, a well-defined main peak corresponding to the hydrodynamic diameter of around 35 nm (Fig. 4b). Given that a single FAM118B_24–334 protomer is about 6 nm long and 5 nm wide, this suggests oligomers of <10 subunits on average at the analyzed concentration.

Next, to disrupt filament formation, we designed two substitutions, F122R and L32R, inserting bulky residues into the head or tail interface (Fig. 3c). Analytical SEC at the injection concentration of 20 µM showed that each mutant in isolation is largely monomeric; L32R eluted as a narrow monomeric peak, while F122R showed slight tailing, suggesting residual filament formation. Mixing the two mutants at a total injection concentration of 20 µM produced a sharp peak at the expected dimer volume, consistent with a single head-to-tail interaction (Fig. 4d).

We further analyzed the interaction between the two mutants, which mimics a single link in a WT FAM118B chain, using isothermal titration calorimetry (ITC). The sharply decaying exothermic peaks observed when injecting L32R into F122R are consistent with an approximately 1:1 stoichiometry and high-affinity binding (Fig. 4e and Extended Data Fig. 10), with a dissociation constant (K_D) likely in a low-nanomolar to high-nanomolar range, although the exact value could not be determined because of high affinity.

Overall, these results confirm that FAM118B possesses two independent surfaces capable of interacting with each other with high affinity. In the WT protein, these interactions mediate the self-assembly of individual protomers into multimers in a submicromolar or low-micromolar concentration regime.

Above a certain level of expression, the protein partly localizes to distinct nuclear foci (Fig. 5a, right), which we address later in this article. Below this threshold, it is diffusively distributed, primarily in the nucleus, with a minor cytoplasmic population (Fig. 5a, left). It is within the lower-expressing cells, which likely better reflect physiological conditions, that we probed the size and shape of FAM118B complexes using various approaches.

First, we monitored their multimerization using photon-counting histogram (PCH) analysis, which fits photon count distributions of mEGFP-tagged complexes diffusing in and out of the confocal volume to extract brightness and concentration (Fig. 5b). Assuming that the brightness of a complex scales linearly with the number of mEGFP tags, PCH allows the estimation of the number of protomers within a multimeric species independently of diffusion time. We verified that a tandem fusion of two mEGFPs is approximately twice as bright as a single mEGFP (Fig. 5c). Moving on to mEGFP–FAM118B analysis (Fig. 5d), we performed PCH measurements separately in the nucleus and the cytoplasm, revealing that, in both compartments, FAM118B forms homomultimers composed of 2–20 tagged protomers, scaling linearly with protomer concentration ranging from approximately 30 nM to 2 µM (Fig. 5e,f).

The comparable PCH results obtained for the nucleus and the cytoplasm suggest that FAM118B multimerization is not affected by association with the previously reported⁴⁸ nuclear interactors such as coilin (known to be well expressed in U2OS cells⁴⁹) or any other factors unequally distributed between the nucleus and the cytoplasm and present at sufficient levels in these cells. In fact, PCH shows a broadly similar increase in multimer size across the submicromolar to low-micromolar range to that observed for purified recombinant protein by SEC in vitro (Fig. 4a), suggesting that FAM118B multimerization in cells reflects an intrinsic property of this protein.

We next wanted to verify that the FAM118B multimers detected in cells correspond to filaments similar to those characterized in vitro. Another possibility is the aggregation of misfolded protein, although the small, well-defined size of the diffuse multimers argues against this. To further distinguish between these possibilities, we conducted measurements that provide information on particle shape. Whereas filaments are rod-like, aggregates are expected to approximate a globule, except in a specific case of amyloid formation⁵⁰. By autocorrelating the raw fluorescence fluctuations acquired for PCH analysis in the nucleus, we were able to quantify the average time spent by the multimers within the confocal volume and plot it as a function of estimated multimer size. These measurements revealed a linear correlation between the diffusion time and the protomer number (Fig. 5g), which fits better a rod-like filament—for which the hydrodynamic drag is expected to be proportional to length—than a globular aggregate⁵¹.

As a second approach, we used fluorescence correlation to calculate the ratio of diffusion speeds of mEGFP–FAM118B multimers containing two versus ten protomers. As a control, we calculated the corresponding ratio for a single mEGFP versus five mEGFP molecules linearly fused through flexible linkers. The comparison of these diffusive ratios to the theoretical predictions of the Perrin's equation indicates that neither the mEGFP fusions, as shown previously^51,52, nor the FAM118 multimers adopt a globular shape (Fig. 5h). Instead, both show an elongated conformation with FAM118B being straighter than the flexible mEGFP pentamer. These findings further support that cellular FAM118B forms filament-like structures rather than aggregates.

Overall, the cellular data support that FAM118B forms extended filaments that grow in protomer number with concentration and ultimately condense into foci at high overexpression (Fig. 5i).

Continuing with the live-cell analysis, we wanted to verify that the head-to-tail association, which is at the basis of filament formation by the WT FAM118B protein in vitro, is formed within living cells. To do so, we assessed the interaction between FAM118B L32R or F122R mutants differentially tagged with either mEGFP or mCherry using fluorescence cross-correlation spectroscopy (FCCS), which detects protein–protein association through their correlated displacement within the confocal volume (Fig. 5j,k). These mutants were exclusively diffusively distributed in cells, localizing mainly to the nuclei. When coexpressing the same mutant (either L32R or F122R) with two different tags, we did not observe any interaction, consistent with abolished multimerization (Fig. 5l). In contrast, the mEGFP–FAM118B-L32R:mCherry–FAM118B-F112R pair or the mEGFP–FAM118B-F122R:mCherry–FAM118B-F112R pair associated strongly in cells. These data align with the head-to-tail interaction configuration observed in vitro (Fig. 5m).

To test whether focus formation requires filament formation, we examined the subcellular localization of filament-deficient FAM118 mutants. Introducing the F122R substitution, which impairs filament formation in vitro (Fig. 4c), into YFP–FAM118B (Fig. 6a, bottom), or the equivalent F109R substitution into YFP–FAM118A (Fig. 6d, bottom) led to an exclusively diffusive distribution across all examined expression levels. These findings were further confirmed using mEGFP-tagged FAM118B, where foci were completely lost in the tail (L32R) or head (F122R) mutants (Fig. 6b, middle and bottom).

To determine whether the observed foci colocalize with known subnuclear domains, including Cajal bodies (reported to require FAM118B for assembly⁴⁸), nucleoli and PML bodies, we coexpressed mCherry–FAM118B WT with EGFP-tagged versions of their respective marker proteins: coilin, nucleolin and PML isoform 4 (Fig. 6c). Although all three markers formed distinct nuclear foci as expected, none overlapped with mCherry–FAM118B foci, indicating that these correspond to distinct subnuclear domains.

Lastly, to probe the dynamics of FAM118B within foci and the extent to which the foci remain separated from their environment, we used a recently described 'half-focus' fluorescence recovery after photobleaching (FRAP) assay (Fig. 6d,e)⁵³. mEGFP–FAM118B was highly mobile within the foci, arguing against misfolded aggregates, expected to show little dynamics⁵⁴. Furthermore, monitoring the fluorescence reequilibration between the bleached and unbleached area showed that, in contrast to their high internal fluidity, the FAM118B foci display little exchange with the rest of the nucleoplasm. More specifically, the mean dip observed for the unbleached area on the fluorescence redistribution curve was equal to 0.29 (Fig. 6e, arrow), significantly above the value expected if proteins freely diffused in and out the foci (P = 10⁻⁹), indicating a diffusion barrier at the foci border.

Together, these results suggest that the FAM118B and FAM118A foci observed at high overexpression are dynamic higher-order assemblies composed of filaments and separated from their environment (Fig. 6g). Given their micrometer-scale dimensions, each focus likely contains numerous filaments.

Next, we investigated whether human FAM118A and FAM118B physically interact in cells by transiently coexpressing mEGFP-tagged FAM118B and mCherry-tagged FAM118A in human U2OS cells. As for FAM118B alone, expression levels varied; cells with lower expression showed exclusively diffusive distribution of both proteins, while highly expressing cells displayed large nuclear foci alongside a diffusive subpopulation (Fig. 6h).

The two paralogs colocalized in foci at high expression but we sought to investigate whether the two proteins also show signs of interaction in focus-deficient, lower-expressing cells more closely resembling physiological conditions. Compared to coexpressed free mCherry and mEGFP used as a negative control (Fig. 6i), the two tagged FAM118s showed strong correlated movement in these cells in FCCS (Fig. 6j,k), confirming physical association.

Potential NAD-binding and/or catalytic residues include a histidine, H206, and an adjacent aspartate, D255, both from the large subunit (Fig. 7a). A histidine in a similar position is found in many sirtuins and is important for SIRT2 and ThsA ativity^11,27,55. In our model, H206 is positioned near the 2′ and 3′ hydroxyl groups of the nicotinamide-proximal ribose ring of NAD and might not only interact with but also polarize or deprotonate these groups, which in turn could stabilize the transition state of NAD decomposition where a partial positive charge develops on the ribose as the nicotinamide departs⁵⁶. While the pK_a of histidine would normally be too high to allow efficient hydroxyl deprotonation, this might change in an appropriate microenvironment. Compared to H206, D255 appears more specific to FAM118s, as it is absent from SIRT2 or ThsA; on the basis of its position within our model, it may directly interact with the ribose hydroxyls of NAD and/or affect these groups indirectly by polarizing H206 to lower its pK_a. Notably, both H206 and D255 are strictly conserved in FAM118B and FAM118A across vertebrates, in line with their potential functional role (Fig. 7b).

Another active-site residue, N163, apparently has a central role in an interaction network that holds together C-terminal and N-terminal regions constituting the NAD pocket (Fig. 7a). Additionally, N163 may stabilize the build-up of the partial positive charge on the ribose during nicotinamide cleavage, as shown for the equivalent residue in yeast Sir2 (refs. ^23,57). N163 is highly conserved in FAM118B and FAM118A (Fig. 7b), as well as other sirtuins, including SIRT2 and ThsA.

The NAD-bound FAM118B model suggests that two aromatic residues, W65 (small subdomain) and Y307 (large subdomain), may interact with the nicotinamide and adenine rings of NAD, respectively (Fig. 7a). W65, which occupies a similar position to F273 in SIRT1, may contribute to both NAD binding and its cleavage, as nicotinamide desolvation promoted by a hydrophobic environment facilitates the latter process^56,58. The similarly positioned phenylalanine of canonical sirtuins seemingly switches from interacting with nicotinamide in the NAD-bound state to itself occupying the nicotinamide site in the absence of NAD⁵⁹ and our NAD-free FAM118B cryoEM density, albeit at moderate resolution, is consistent with W65 sliding into the predicted nicotinamide pocket in the absence of NAD (Extended Data Fig. 5 and Supplementary Fig. 1), which could enhance product release or serve another role. Again, W65 and Y307 are conserved in FAM118B and FAM118A, although Y307 is sometimes replaced by a similar phenylalanine (Fig. 7b).

To experimentally test the ability of FAM118s to process NAD, we used a fluorometric assay based on an NAD analog, ε-NAD, whose fluorescence increases when the nicotinamide moiety is released through hydrolysis or similar process^60,61 (Fig. 7c). The two extra atoms on the adenosine base of ε-NAD should be accommodated within FAM118s without steric clashes (Supplementary Fig. 2).

Using this approach, we found that 5 µM purified His₆-tagged full-length FAM118B or FAM118A showed no or very low NAD-processing activity individually (Fig. 7d,e). However, mixing the two paralogs, each at 5 µM, produced a robustly detectable NAD-processing activity that was dramatically increased compared to that of individual proteins. Nonetheless, this elevated activity remained relatively weak compared to a previously characterized extracellular NADase from the fungus Fusariumoxysporum f.sp. cubense race 1 (FoNADase1)⁶², which achieved a similar effect at a much lower concentration of 0.2 µM FoNADase1. We also compared these activities to a standard positive control, commercial porcine brain NADase^63,64, a large amount (1 mg ml⁻¹) of which was used to define complete NAD hydrolysis.

The elevated NAD-processing activity observed upon mixing FAM118B and FAM118A was prevented if either FAM118A or FAM118B was mutated, targeting the mentioned histidine or asparagine residues located in their active sites (H206A or N163A in FAM118B or H191A or N148A in FAM118A) (Fig. 7f). Intriguingly, the activity was also abolished when the ability of either protein to form head-to-tail interactions was disrupted through the F122R substitution in FAM118B or the equivalent F109R substitution in FAM118A. These results suggest that, to unleash NAD processing, not only must both paralogs be present, but each must also be able to at least bind and potentially process NAD and form head-to-tail interactions. This in turn may suggest that the active form of these enzymes corresponds to a mixed FAM118B:FAM118A filament or a higher-order structure composed of homotypic FAM118B-only and FAM118A-only filaments.

Lastly, to validate the observed activity, we tested whether the FAM118B:FAM118A combination is active in depleting NAD in human 293T cells. Indeed, simultaneously overexpressing both YFP-tagged proteins in their WT versions led to severely decreased cellular NAD levels (approximately 15% of empty-vector control) as measured using a cycling colorimetric NAD/NADH assay (Fig. 7g). In contrast, co-overexpressing catalytically inactive (FAM118B-N163A and FAM118A-N148A) or filamentation-deficient (FAM118B-F122R and FAM118A-F109R) variants of both proteins resulted in only a modest NAD decrease to 70–80% of the control levels (Fig. 7g), despite a similar expression of WT and mutant YFP–FAM118 variants (Supplementary Fig. 3). These cellular data demonstrate that FAM118B and/or FAM118A are active in processing NAD in human cells, in a manner dependent on their active site(s) and the ability to form head-to-tail interactions.

For the sirtuin family tree, sequences of proteins containing different types of sirtuin domains were extracted using InterPro⁹² and National Center for Biotechnology Information (NCBI) resources^93,94. For the FAM118B clade tree, orthologs of human FAM118B and FAM118A were identified using NCBI resources. In both cases, sirtuin domains were defined using the Pfam database⁹⁵. The evolutionary history was inferred using the maximum-likelihood method and either the Whelan and Goldman model⁹⁶ (sirtuin family tree) or the Jones–Taylor–Thornton (JTT) matrix-based model⁹⁷ (FAM118-clade tree). The tree with the highest log likelihood (−30,365.33 for the sirtuin tree and −2,081.08 for the FAM118-clade tree) is presented. The initial trees for the heuristic search were obtained automatically by applying neighbor-joining (NJ) and BioNJ algorithms to a matrix of pairwise distances estimated with the JTT model and then selecting the topology with superior log-likelihood value. A discrete gamma distribution was used to model evolutionary rate differences among sites (five categories; +G, parameter = 3.2422 for sirtuin family tree or 0.9762 for FAM118-clade tree). For the sirtuin family analyses, the rate variation model allowed for some sites to be evolutionarily invariable ([+I], 0.25% sites). The final datasets contained a total of 399 (sirtuin family tree) or 152 (FAM118-clade tree) positions and the evolutionary analyses involved 86-aa or 30-aa sequences, respectively. Evolutionary analyses were conducted in MEGA X⁹⁸.

AF2 was accessed through ColabFold⁹⁹ (version 1.5.5) run with default settings, including multiple-sequence alignment generation through MMseqs2 (many-against-many sequence searching)¹⁰⁰. AF3 (ref. ¹⁰¹) was accessed through the AF server, again with default settings. All models were deposited in ModelArchive¹⁰² as described in 'Data Availability'.

To identify proteins that are most closely related to human FAM118s in NCBI resources at the sequence level to determine whether FAM118s form a distinct clade, we performed a basic local alignment search tool among proteins. The sequence of either human FAM118 paralog was used as query against the nonredundant protein sequence database with default settings except for 'max target sequences' set to 5,000. Only results with sequence coverage of at least 70% were considered.

Multiple-sequence alignments to probe the conservation of specific residues were performed using Clustal Omega with default settings¹⁰³.

To map sequence conservation onto an AF3 structural model, we first generated multiple-sequence alignments with Clustal Omega comprising either (1) a representative 15 members of the FAM118 clade (eight FAM118Bs and seven FAM118As) with sequence identity ranging from 42% to 95% or (2) four FAM118s (FAM118A and FAM118B from Homosapiens and S. canicula) and 11 related non-FAM118 SIR2_2 proteins that are part of the same higher-level clade in Extended Data Fig. 2. The conservation scores were calculated and mapped onto a structural model using ConSurf^104,105 performed using a user-provided sequence alignment.

All structural superpositions and alignments, including root-mean-square-deviation (r.m.s.d.) calculations, were performed in PyMol.

Nucleotide sequences encoding full-length human FAM118A (UniProt Q9NWS6↗) and FAM118B (UniProt Q9BPY3↗) were cloned into the pDEST-17 and pET-28a(+) plasmids, respectively, for the production of N-terminally hexahistidine-tagged (His-tagged) proteins used in structural studies and NAD cleavage assays.

Additionally, a nucleotide sequence corresponding to residues 24–334 of human FAM118B was cloned into a pET-28a(+) plasmid for the production of the C-terminally His-tagged FAM118B_24–334 protein used in structural studies and analytical SEC.

Point mutations were introduced using a standard site-directed mutagenesis PCR protocol.

An appropriate plasmid was transformed into Rosetta2 (DE3) E. coli. Bacteria were grown in either 2× yeast extract tryptone medium (for full-length FAM118A and FAM118B proteins) or Luria–Bertani Lennox medium (for FAM118B_24–334) supplemented with an appropriate antibiotic. An expression culture was inoculated with a saturated starter culture (1:100 dilution) grown overnight from a single colony from a fresh transformation plate. Large-scale expression cultures (1–2 L) were incubated at 37 °C, with shaking at 180 rpm, until an optical density at 600 nm of 0.6–0.8 was reached before cooling to room temperature. Expression was then induced with 0.25 mM or 0.5 mM IPTG for full-length proteins or FAM118B_24–334, respectively. The induced cultures were incubated at 18 °C and 180 rpm for a further ~20–24 h before harvesting by centrifugation at 4,500g for 30 min. Cell pellets were collected and stored at −20 °C until protein purification.

Recombinant full-length FAM118A and FAM118B used for structural studies and NAD cleavage assays were purified using a procedure consisting of Ni²⁺ affinity chromatography followed by SEC. Cell pellets were resuspended in lysis buffer (20 mM HEPES pH 7.5, 500 mM NaCl, 0.5 mM TCEP, 20 mM imidazole, 10 μg ml⁻¹ DNAse I, 100 μg ml⁻¹ lysozyme and 1× cOmplete EDTA-free protease inhibitor) and lysed by sonication before clarification by centrifugation at 18,000g for 30 min. The soluble fractions of cell lysates were incubated with Ni-NTA agarose (Cytiva) preequilibrated in wash buffer (20 mM HEPES pH 7.5, 500 mM NaCl, 0.5 mM TCEP and 20 mM imidazole). Bound proteins were washed using wash buffer and eluted using elution buffer (20 mM HEPES pH 7.5, 500 mM NaCl, 0.5 mM TCEP and 200 mM imidazole). Protein-containing fractions were concentrated down to a volume of 5 ml and injected onto a 16/600 HiLoad Superdex 200 prep-grade column (Cytiva) preequilibrated in SEC buffer (20 mM HEPES pH 7.5, 150 mM NaCl and 0.5 mM TCEP). The protein was eluted from the column using SEC buffer.

FAM118B_24–334 for structural studies and analytical SEC experiments was purified using a procedure comprising Ni²⁺ affinity and ion-exchange chromatography the cell pellet was resuspended in 25 mM Tris pH 8.8, 500 mM NaCl and 0.5 mM TCEP. Following sonication, the crude lysate was separated into the pellet and supernatant fractions by centrifugation (12,000g for 30 min at 10 °C). The supernatant was loaded onto a 5-ml HisTrap high-performance (HP) column (Cytiva) and the column was washed with 25 mM Tris pH 8.8, 500 mM NaCl and 0.5 mM TCEP first without and then with 50 mM imidazole, followed by protein elution with 250 mM imidazole. The eluted protein was then diluted ten times with water, loaded onto a 5-ml HiTrap Q HP column (Cytiva), equilibrated in 25 mM Tris pH 8.8 and 0.5 mM TCEP and eluted using a salt gradient from 50 to 1,000 mM NaCl in the same buffer. The final protein from the main ion-exchange peak was in an approximate buffer composition of 25 mM Tris pH 8.8, 250 mM NaCl and 0.5 mM TCEP.

All protein purification steps were performed at 4 °C, with protein purity at each stage assessed by SDS–PAGE. Final purified protein fractions were pooled, concentrated using a centrifugal concentrator, frozen in liquid nitrogen and stored in a −80 °C freezer.

A thin film of carbon on top of a copper 200-mesh grid (C101/100, TAAB) was glow-discharged for 25 s at 15 mA of current (Pelco, easiGlow) before applying 3 μl of purified human FAM118B directly onto the grid. After 1 min of incubation at ambient temperature, the sample was absorbed by touching a side of the grid with a wedge of filter paper (Whatman, 1001-090). Facing down, the grid was rapidly dropped on top of 20 μl of 2% uranyl acetate solution dispensed onto the surface of a parafilm, where it was incubated for 10 s. The excess stain was blotted away until a thin film remained and the grid was air-dried for 5 min. The images were collected at a nominal magnification of ×50,000 and a pixel size of 2.3 Å on a 200-kV EM instrument, JEM-2100Plus (Jeol) (The Dunn School of Pathology, University of Oxford).

A purified recombinant human FAM118B was diluted from thawed material to 0.1 mg ml⁻¹ (3.5 μM) in 12.5 mM Tris pH 7.5, 150 mM NaCl and 0.5 mM TCEP buffer. Then, 3.5 μl was applied to the Quantifoil 1.2/1.3 Cu 300-mesh grids pretreated in Harrick plasma in vacuum for 2 min at high power. A Leica GP2 was equilibrated to 100% humidity at 18 °C before the sample was backblotted for 1.5 s and plunge-frozen in liquid ethane. The grids were stored in liquid nitrogen until the EM imaging.

A total of 11,630 raw cryoEM images were collected in a super-resolution mode at a physical pixel size of 0.83 Å on a 300-kV Titan Krios microscope (Thermo Fisher Scientific) with an integrated K3 camera and an energy filter (20-eV slits were used). The images were acquired at a defocus range of −1.6 μm to −2.6 μm and a sum dose per video of 43.7 e⁻ per Ų. The data were collected in the EM facility COSMIC (University of Oxford).

The entire dataset was preprocessed on the fly using cryoSPARC¹⁰⁶ live (COSMIC, University of Oxford) when 40-frame videos were aligned using default settings of patch motion correction and the contrast transfer function (CTF) values were estimated using the Patch CTF. Particle picking was carried out using the blob picker with a blob diameter from 100 to 200 Å, resulting in 1,958,302 selected particles. Initially, all particles were extracted with a smaller box size of 320 pixels and a binning factor of 4 and were subjected to two-dimensional (2D) classification. The images representing the FAM188B filaments were retained, while classes most obviously presenting as bright ice contaminants were manually deselected. The ab initio models constrained to five classes were calculated from 1,540,301 particles, which were used as reference models for a round of heterogeneous refinement. A single good three-dimensional (3D) class reminiscent of a filament reconstructed from 531,673 images (34.5%) was reextracted with a box size of 600 pixels (498 Å) and subjected to a homogeneous refinement to 6.4-Å resolution. A resulting map showed discernible core of a filament made up from five repeating units; however, lateral contacts between the core and adjacent units were also found. With such a 3D map reconstructed from 2D image sampling of a linear segment of filaments, its branching points or occasional aggregates (evident from subsequent 2D classification but also in our initial negative-stain EM images), a second round of ab initio reconstruction limited to five classes was attempted. This step was followed by heterogeneous refinement forcing hard classification so that each particle was assigned to only one class at each iteration. Four meaningful 3D classes were obtained representing clearly defined filaments short (2–4-mers) and long (5–7-mer). The longest filament was the only class (111,156 images) where no lateral contacts between a linear core and adjacent subunits could be observed, which was then further subjected to a nonuniform focused refinement using a binary mask (soft padding of 50 pixels and dilation radius of 1 pixel) encompassing the length of a filament fragment comprising five protomers. To better resolve the structure of the head-to-tail interfaces and the NAD-binding pocket, all four classes of filaments were combined (368,283 particles) before carrying out a nonuniform focused refinement within a binary mask (soft padding of 60 pixels and dilation radius of 2 pixels) around a core part encompassing two full protomers and fragments of flanking protomers. The final cryoEM maps of filament fragments composed of five and two protomers reconstructed at an overall resolution (gold-standard Fourier shell correlation (FSC) = 0.143) of 5 Å and 4.1 Å, respectively, were used for structural modeling of a full-length FAM118B filament.

Individual chains of AF3-predicted structures of human FAM118B (Q9BPY3-F1) were first rigid-body fitted in the map of a full-length FAM118B filament fragment encompassing two protomers (4.1 Å) using ChimeraX¹⁰⁷. Disordered N-terminal (1–27) and C-terminal (327–351) tails, present in the AF3 model, were found outside of the cryoEM density and were removed using Coot¹⁰⁸. Each of the two molecules of FAM118B was then flexibly fitted using molecular dynamics flexible fitting implemented in ISOLDE¹⁰⁹ as a ChimeraX plugin. Torsional and distance self-restraints were applied during simulations. Occasionally, large bulky side chains were manually repositioned especially around the NAD-binding pocket and at the interfaces between the protomers. The process was repeated for fitting five individual molecules of FAM118B models into the map of the full-length FAM118B filament fragment encompassing five protomers.

A total of 4 µl of FAM118B_24–334 at 3 µM (diluted from thawed protein stock in 25 mM Tris pH 8.8, 250 mM NaCl and 0.5 mM TCEP) was applied to glow-discharged 300-mesh Formvar grids (EMS) for 60 s at room temperature. Following sample application, the grids were manually blotted to remove excess liquid. Subsequently, 20 µl of uranyl acetate was applied three times for 30 s each. The grids were then air-dried before imaging. The negative-stained grids were first screened using a Tecnai T12 microscope operated at 120 kV.

A total of 3 µl of FAM118B_24–334 at 30 µM (diluted from thawed protein stock in 25 mM Tris pH 8.8, 250 mM NaCl and 0.5 mM TCEP) was applied to negatively glow-discharged 200-mesh Lacey grids (EMS). Grids were blotted with ash-free Whatman grade 540 filter paper in a Vitrobot Mark IV (Thermo Fisher Scientific) for 4 s at 4 °C and 95–100% humidity and then vitrified in liquid ethane. The sample quality and distribution were initially assessed using a Glacios microscope (Thermo Fisher Scientific) equipped with a Falcon 4 direct electron detector (Thermo Fisher Scientific) (NanoImaging Core Facility, Institut Pasteur).

A preliminary dataset of 100 images was collected at the NanoImaging Core Facility of the Pasteur Institute using a Glacios microscope, operated at 200 kV and equipped with a BioQuantum Energy Filter (Gatan) with a 10-eV slit width. Videos were recorded using a Falcon direct electron detector in super-resolution mode at a magnification of ×105,000, corresponding to a pixel size of 0.83 Å. A dose rate of 15 e⁻ per s per physical pixel resulted in a total electron dose of 40 e⁻ per Ų, applied over 40 frames. Data collection was performed using EPU software with defocus values ranging from −1 to −3 µm.

Video stacks were processed in cryoSPARC¹⁰⁶. Super-resolution videos were frame-aligned, motion-corrected, gain-normalized, dose-weighted and binned twice using the patch motion correction module. CTF parameters were estimated with the patch CTF module in cryoSPARC. Micrographs exhibiting ice contamination, ethane contamination or poor CTF fit resolution (below 6 Å) were discarded.

Initial particle picking was performed using a circular blob picker with dimensions of 40–250 Å. Following an initial round of 2D classification, ab initio reconstruction and heterogeneous refinement, a template-based picker was applied to improve particle selection. A second round of 2D classification, ab initio reconstruction and heterogeneous refinement was conducted. Finally, 50,985 particles were selected for nonuniform refinement, resulting in the final reconstruction map.

The final dataset of 19,798 videos was collected at the European Synchrotron Radiation Facility (ESRF) Grenoble (CM01 beamline)¹¹⁰ using a Titan Krios microscope (Thermo Fisher Scientific) operated at 300 kV using a K3 camera (Gatan) in super-resolution acquisition mode and equipped with a quantum LS Energy Filter (Gatan) with a 20-eV slit width at a magnification of ×105,000, corresponding to a pixel size of 0.839 Å. A dose rate of 18.9 e⁻ per s per physical pixel resulted in a total electron dose of 40.18 e⁻ per Ų, applied over 40 frames. Data collection was performed using EPU software with defocus values ranging from −0.8 to −2.8 µm. Video stacks were processed in cryoSPARC. Super-resolution videos were frame-aligned, motion-corrected, gain-normalized, dose-weighted and binned twice using the patch motion correction module. CTF parameters were estimated with the patch CTF module in cryoSPARC. Micrographs exhibiting ice contamination, ethane contamination or poor CTF fit resolution (below 6 Å) were discarded.

Leveraging the reconstruction map obtained from the small dataset, we directly performed template picking for particle selection. A round of 2D classification, ab initio reconstruction and heterogeneous refinement was applied as described above and illustrated in Extended Data Fig. 7. The final map obtained from nonuniform refinement was derived from a filament composed of six protomers and reached a medium resolution of approximately 8.7 Å, as estimated by the FSC = 0.143 criterion, using the PDBe FSC server. To improve the resolution of individual protomers and get a better view of the interfaces between subunits, a local refinement was performed focusing on a subset of three protomers. This refinement yielded a final map with an improved resolution of around 4.8 Å, as determined by the FSC = 0.143 criterion using the PDBe FSC server. A local resolution map was subsequently calculated in PHENIX^111,112 (from half-maps), confirming that the refined region achieved resolutions ranging approximately from 3.6 to 7 Å.

The AF model of human FAM118B, residues 24–334, was fitted into the density of the focused map of the filament fragment composed of three protomers using ChimeraX¹⁰⁷. Multiple rounds of refinement in Coot¹⁰⁸ and PHENIX^111,112 were applied, resulting in a model with a cross-correlation of around 0.6 and mean Q score of 0.2.

Analytical SEC analysis was performed using a 10/300 GL Superdex 100 Increase column (Cytiva) in 50 mM Tris pH 8, 200 mM NaCl and 0.5 mM TCEP pumped at 1 ml min⁻¹. WT or mutant FAM118B_24–334 protein samples diluted to the indicated concentrations with the same buffer were prepared in a total volume of 515 µl and injected onto the column using a 500-µl loop. Then, 1-ml fractions were collected during the run and 7 µl of each from the relevant elution volume (EV) range was analyzed by SDS–PAGE.

The column was calibrated using a collection of molecules of various MW values, which eluted at indicated EVs: blue dextran (MW, 2,000 kDa; EV, 8.7 ml), ferritin (MW, 440 kDa; EV, 9.75 ml), immunoglobulin G (MW, 160 kDa; EV, 11.4 ml), BSA (MW, 66 kDa; EV, 13.3 ml), ovalbumin (MW, 44 kDa; EV, 14.6 ml), carbonic anhydrase (MW, 29 kDa; EV, 16.2 ml) and ribonuclease A (MW, 13.7 kDa; EV, 17.9 ml). A linear regression between EV and log₁₀MW was determined with Excel and applied to predict the theoretical EV values for various multimerization states of FAM118B_24–334.

ITC experiments were performed at 25 °C using a PEAQ-ITC apparatus (Malvern instruments). FAM118B_24–334-L32R, present in the syringe at 500 µM, was titrated into FAM118B_24–334-F122R, present in the cell at 50 µM. A total of 19 injections were carried out, each of 0.5 ul, with 1 s for injection and 150 s between injections. The stirring speed was set to 750 rpm. Both proteins were exchanged into 1× PBS buffer before the experiment through ten rounds of dilution and concentration in a centrifugal concentrator. The ITC data were analyzed with the MicroCal PEAQ-ITC analysis software (version 1.41) with subtraction of injection heat of buffer into buffer, L32R mutant into buffer and buffer into F122R mutant. The K_D could not be reliably estimated because of a high-affinity interaction. Attempts to perform ITC with lower concentrations of both proteins were unsuccessful because of a low signal-to-noise ratio.

DLS experiments were performed using Zetasizer Nano S (Malvern Instruments) at 20 °C. FAM118B_24–334 WT was diluted to 10 µM in PBS buffer (137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄ and 1.8 mM KH₂PO4, pH 7.4). The data were analyzed with Zetasizer software.

The full-length native DNA sequence of human FAM118B (UniProt Q9BPY3↗) was synthesized and cloned either into (1) the pEGFP-C1 plasmid, in which the gene encoding the EGFP tag was mutated to render it monomeric (A206K substitution, the final tag designated as mEGFP), or (2) the pmCherry-C1 plasmid.

The full-length native DNA sequences of human FAM118A (UniProt Q9NWS6↗) and FAM118B were also cloned into pDEST-YFP-N1 (ref. ¹¹³) or pDEST-mCherry-N1 (ref. ¹¹⁴) using Gateway recombination (Invitrogen).

Site-directed mutagenesis was performed using a standard site-directed PCR-based approach.

U2OS cells (from I. Ahel's laboratory¹¹⁵, originally ATCC-HTB-96) were maintained in DMEM supplemented with 10% FBS (Sigma) and penicillin–streptomycin (100 U per ml, Gibco) and cultured in a humidified atmosphere at 37 °C with 5% CO₂. U2OS were plated in eight-well glass-bottom chamber slides (Ibidi or ZellContact) 24 h before transfection of plasmids using TransIT-LT1 (Mirus Bio) or X-tremeGENE HP (Sigma) transfection reagents according to the manufacturer's instructions.

For Fig. 6a,b, U2OS cells transiently expressing fluorescently labeled FAM118A and FAM118B were incubated with 1.25 µg ml⁻¹ Hoescht 33342 for 30 min, before replacing the medium with imaging medium (phenol-red-free Leibovitz's L-15 medium (Life Technologies) supplemented with 20% FBS, penicillin (100 μg ml⁻¹) and streptomycin (100 U per ml)). Cells were imaged on an Olympus IX-83 inverted microscope equipped with a Yokogawa SoRa super-resolution spinning-disk head, a UPlanAop ×60 (1.5 numerical aperture (NA)) oil-immersion objective lens and a Prime BSI scientific complementary metal–oxide–semiconductor camera. The fluorescence of Hoechst and YFP was excited with 405-nm and 488-nm solid state lasers, respectively, and fluorescence detection was achieved with bandpass filters adapted to the fluorophore emission spectra. For Fig. 6c,h, U2OS cells transiently expressing mCherry-tagged or mEGFP-tagged proteins were imaged on an LSM 880 Zeiss confocal microscope equipped with a water-immersion C-Apo ×40 (1.2 NA) lens. mEGFP and mCherry were excited using 488-nm and 561-nm lasers, respectively, and fluorescence detection was collected at 500–550 nm for mEGFP and 580–650 nm for mCherry using GaAsP detectors. The pixel size was set to 70 nm.

PCH and FCCS experiments were performed on an LSM 880 Zeiss confocal microscope equipped with a water-immersion C-Apo ×40 (1.2 NA) lens. mEGFP and mCherry were excited using 488-nm and 561-nm lasers, respectively. Fluorescence fluctuations were collected at 500–550 nm for mEGFP and 580–650 nm for mCherry, both on the GaAsP detector used in photon-counting mode. Laser powers were adjusted to minimize photobleaching and acquisitions lasted 20–30 s to reduce noise on the autocorrelation and cross-correlation curves and PCHs. Cells were maintained at 37 °C with a heating chamber. PCHs were calculated from the raw fluorescence fluctuation data using time bins of 100 µs and fitted with a one-specie model assuming a 3D Gaussian detection volume using the Zen Black software¹¹⁶. The values obtained for the number of diffusing oligomers in the confocal volume and their average brightness were corrected for diffusion as previously described¹¹⁷. The concentrations of the labeled proteins were calculated assuming a confocal volume of ~1 fl (ref. ¹¹⁸). For FCCS analysis, the autocorrelation and cross-correlation curves were calculated from raw photon traces after detrending for slow fluctuations using the Fluctuation Analyzer 4G software¹¹⁹. The autocorrelation and cross-correlation curves were fitted using one-specie simple diffusion models assuming a shape factor equal to 6. To correct for unavoidable nonidealities in the acquisition setup, we used positive controls, corresponding to cells expressing mEGFP linked to mCherry, and negative controls, corresponding to cells coexpressing free mEGFP and mCherry. The bleedthrough between the mEGFP and mCherry channels was estimated from the negative controls and subtracted from the other FCCS acquisitions. Differences in autocorrelation amplitudes and incomplete cross-correlation because of partial overlap between the confocal volumes in the mEGFP and mCherry were corrected using the positive controls as previously described¹¹⁸.

NADase activity was assessed using a plate-reader-based fluorescent ε-NAD cleavage assay. Reactions (15-μl final volume) comprising purified FAM118B and/or FAM118A protein (at 5 µM each), reaction buffer (20 mM HEPES pH 7.5, 150 mM NaCl and 0.5 mM TCEP) and 0.5 mM ε-NAD were prepared in PCR tubes and transferred to 384-well black polystyrene nonbinding clear-bottom assay plates. In the positive control reaction, porcine brain NADase (Sigma-Aldrich, N9879) at the final concentration of 1 mg ml⁻¹ was used instead of FAM118 protein. Fluorescence recordings (300-nm excitation, 410-nm emission) were measured using the SpectraMax M5 plate reader (Molecular Devices) in kinetic mode every 30 s for 180 min.

Cellular NAD levels were assayed using the NAD/NADH colorimetric assay kit (Abcam, ab65348) following the manufacturer's protocol. First, 2 days before the assay, 6 × 10⁵ 293T cells (ATCC-CRL-3216) were seeded into a six-well plate (Corning Costar 3516). The next day, cells were treated with either 2 nM FK866 (NAMPT inhibitor) or DMSO or transfected with 2 µg of plasmid DNA per well of pDEST-YFP, pDEST-YFP-FAM118A WT, pDEST-YFP-FAM118A-N148A, pDEST-YFP-FAM118A-F109R, pDEST-YFP-FAM118B WT, pDEST-YFP-FAM118B-N168A or pDEST-YFP-FAM118-F122R using 20 µl of Polifect (Qiagen, 301105) following the manufacturer's instructions. In summary, Polyfect–DNA complexes were assembled in 75 µl of serum-free medium (1× DMEM GlutaMAX; Gibco, 31966-021) and incubated at room temperature for 10 min. Thereafter, the Polyfect–DNA mixture was added dropwise onto 2.5 ml of complete medium per well supplemented with 10% FBS (Sigma, 9665) and 1% penicillin–streptomycin (Gibco, 15140-122). Then, 24 h later, cells were harvested, washed with 1 ml of PBS (Sigma, D8537) and trypsinized in 0.5 ml of TrypLE Express (Gibco, 12604-013). Cells were resuspended in 2 ml of complete medium with an average count of 1 × 10⁶ cells per ml. The 6 × 10⁵ cells were sampled, briefly pelleted at 4 °C and washed twice with ice-cold PBS to remove residual trypsin. Then, 400 ul of extraction buffer II at ambient temperature was used for cell lysis by two repeated freeze–thaw cycles on dry ice. Cells were vortexed for 10 s and centrifuged at 16,873g (maximum speed) for 5 min at 4 °C to remove insoluble material (leaving behind a transparent loose pellet). The supernatant was passed through a 10-kDa spin column (Abcam, ab93349) at 16,873g for 1 h at 4 °C to remove endogenous NAD/NADH-consuming enzymes. The flowthrough from each sample was split into two to allow independent measurements of total cellular NAD and NADH (NADt) or NADH only. To measure only NADH, NAD was depleted by heating the samples to 60 °C for 30 min followed by cooling down on ice for 5 min to avoid denaturing of the cycling enzyme in the next step. Next, the NADH standard was freshly prepared from a DMSO stock by diluting 1:100 (10 pmol µl⁻¹) in extraction buffer II to a final concentration of 20–100 pmol per well. Next, 50 μl of sample in triplicates (NADt and NADH only) and NADH standard in duplicates was dispensed in a 96-well flat, clear polystyrene microplate (Corning, CLS3370). Then, 100 μl of reaction master mix (cycling buffer I and NAD cycling enzyme mix) was added to each well before mixing on a plate shaker for 5 min at ambient temperature. Lastly, 10 μl of developer solution II was added to each sample and standard before incubating on a plate shaker for 15 min at room temperature. The absorbance at 450 nm was measured on a plate reader SpectraMax 5/SoftMax Pro v7 (Molecular Devices) over the course of 4 h with 10-min intervals at 21 °C. NADt and NADH levels were calculated from absorbance values using an NADH standard curve. The cellular levels of NAD (pmol per well) were calculated by subtracting NADH-only values from NADt values. The NAD levels were normalized to the total protein content (pmol µg⁻¹ protein) of each sample determined by the Bradford Protein Assay (Bio-Rad) and the BSA (Thermo Scientific, 23209) standard curve. Statistical analyses were performed in GraphPad Prism (version 10.5).

To monitor protein levels as shown in Supplementary Figure 3, lysates were boiled in 1× NuPAGE LDS sample buffer (Invitrogen) with 50 mM DTT (Sigma-Aldrich). Thereafter, proteins were subjected to NuPAGE Novex 4–12% Bis–Tris gel (Invitrogen), followed by semidry transfer onto a 0.2-µm nitrocellulose membrane (Bio-Rad) using the Trans-Blot Turbo transfer system (Bio-Rad) and blocking in 5% nonfat dried milk in 1× PBS buffer with 0.1% Tween-20 (PBST) for 1 h at room temperature. Goat anti-GFP antibody was diluted in 5% nonfat dried milk in PBST (Abcam, ab5450; batch GR3402450-2, stock concentration 0.5 mg ml⁻¹, dilution 1:5,000) and used overnight at 4 °C followed by horseradish-peroxidase-conjugated rabbit anti-goat secondary antibody incubation (Dako, P0160; stock concentration 0.65 mg ml⁻¹, dilution 1:7,500) for 45 min at room temperature. Pierce enhanced chemiluminescence western blotting substrate (Thermo Fisher Scientific) was added onto the membrane for 5 min, followed by visualization on Hyperfilm ECL films (Cytiva).

No statistical methods were used to determine sample sizes. Data distribution was assumed to be normal but this was not formally tested.

The sample size for cryoEM data was arbitrarily chosen depending on instrument availability and the number of particles required for standard structural reconstruction, ensuring good final statistics.

Sample sizes (numbers of cells) for cellular experiment were arbitrarily chosen to be around 30 (the precise values are indicated for each experiment in figure legends) to span the variability within the data, providing a reliable mean estimate. For qualitative assessment of protein localization, we analyzed n = 15 cells at low and n = 15 cells at high expression levels to produce representative images.

The numbers of measurements in quantitative NAD-processing and NAD-level assays were arbitrarily chosen to be n = 3 or n = 4, respectively, to allow an estimate of s.d. or s.e.m. To determine the significance of differences in cellular NAD levels (Fig. 7g), a one-way analysis of variance (ANOVA) and Tukey's post hoc test for multiple comparisons or unpaired one-tailed t-test for a binary comparison was applied as indicated in figure legend.

All quantitative and qualitative assays were repeated independently at least n = 2 times with similar conclusions, as indicated in figure legends. All attempts at replication for the experiments included in the manuscript were successful.

No data were excluded except for some particles being disregarded during cryoEM analysis as per best practice in cryoEM data analysis, as explained in detail.

No randomization or blinding was performed.

Further information on research design is available in thelinked to this article. Nature Portfolio Reporting Summary