Sir2-domain associated short prokaryotic Argonautes provide defence against invading mobile genetic elements through NAD+ depletion

concentration, 5’ - 32 P radiolabelled oligonucleotides were mixed with appropriate cold 5’P oligonucleotides at a ratio of 1:4 of hot:cold, and diluted to a working concentration of 100 nM in reaction buffer (33 mM Tris-acetate, pH 7.9, supplemented with 10 mM magnesium acetate, 66 mM potassium acetate, 0.1 mg/ml BSA, 5 mM DTT). Protein dilutions were made using the same reaction buffer to 2x final reaction concentration. Protein complexes and nucleic acids were mixed in final concentrations of 50 nM NAs and 0, 50 or 500 nM protein, incubated for 1 hour at 25 °C. The reaction was stopped with the addition of 2x 95% formamide dye and incubating for 5 min at 95 °C. Reaction products were resolved by denaturing PAA gel electrophoresis (21% 29:1 acrylamide/bis-acrylamide in TBE (89 mM Tris, 89 mM boric acid, 2 mM EDTA), supplemented with 8 M urea), visualised with a phosphor imager and analysed in OptiQuant software.

Argonaute (Ago) proteins are found in all three domains of life. The so-called long Agos are composed of four major domains (N, PAZ, MID, and PIWI) and contribute to RNA silencing in eukaryotes (eAgos) or defence against invading mobile genetic elements in prokaryotes (pAgos). Intriguingly, the majority (~60%) of prokaryotic Agos (pAgos) identified bioinformatically are shorter (comprised of only MID and PIWI domains) and are typically associated with Sir2, Mrr or TIR domain-containing proteins. The cellular function and mechanism of short pAgos remain enigmatic. Here, we show that short pAgos from Geobacter sulfurreducens, Caballeronia cordobensis and Paraburkholderia graminis, together with the NAD + -bound Sir2-proteins form a stable heterodimeric Sir2/Ago complex that recognizes invading plasmid or phage DNA through the pAgos subunit and activates Sir2 subunit triggering the endogenous NAD + depletion and cell death thus preventing the propagation of invading DNA. This is the first demonstration that short Sir2-associated pAgos provide defence against phages and plasmids and underscores the diversity of mechanisms of prokaryotic Agos.
Being at the core of RNA interference eAgos are involved in the regulation of gene expression, silencing of mobile genome elements, and defence against viruses 1,2 . All eAgos use small RNA molecules as guides for target RNA recognition and are similar both structurally and mechanistically [2][3][4][5] . Monomeric eAgos are composed of four major N, PAZ, MID, and PIWI domains (Fig. 1A) and share a bilobed structure, where the N-and C-terminal lobes are formed by conserved N/PAZ and MID/PIWI domains, respectively [4][5][6][7] . The N-domain acts as a wedge that separates guide and target strands 3,7 , while the MID and PAZ domains bind, respectively, the 5′-and 3′-terminus of the guide RNA, located between the N-and C-terminal lobes 4,6 . eAgos can slice target RNA through endonucleolytic cleavage by the PIWI domain or inhibit translation through RNA binding by the catalytically inactive eAgos that may also trigger RNA decay by auxiliary cellular nucleases 2,4,5 . pAgos are quite widespread and are present in 17% of sequenced bacterial and 25% of archaeal genomes 8 . To date, more than ~1000 pAgos have been identified bioinformatically, revealing a striking diversity. Moreover, pAgos are often associated with additional putative nucleases, helicases, and DNA binding proteins that are not linked to eAgos 8 . pAgos are divided into full-length or long pAgos (~40%) sharing conserved N, PAZ, MID, and PIWI domain architecture with eAgos, and short pAgos (~60%) composed only of MID and PIWI domains ( Fig. 1A) 8,9 . Long pAgos are relatively well-characterized both structurally and functionally 10,11 and, similarly to eAgos, contain either catalytically active or inactive PIWI domain. Long pAgos with the catalytically active PIWI, as exemplified by CbAgo and TtAgo, guided by DNA cleave DNA target providing defence against invading phages or plasmids, or contributing to chromosome segregation after replication, respectively [12][13][14] . Meanwhile, long RsAgo with an inactive PIWI domain guided by small RNA is thought to mobilize an unknown cellular nuclease(s) for degradation of invading plasmids and mobile genetic elements 9,15 . In contrast to long pAgos, all short pAgos possess a catalytically inactive PIWI domain; however, they are typically associated with proteins containing an APAZ (Analog of PAZ) domain fused to predicted Sir2, Mrr nucleases or TIR domains 8,16 . Sir2-APAZ and short pAgo are sometimes fused into a single-chain protein 8 . Despite the fact that short pAgos make up the majority of all pAgos, to date, their function in the cell and in vitro remain to be established.

Results
In this study, we aimed to explore whether short pAgos can act as prokaryotic defence systems against viruses or plasmids. To this end, we selected two short pAgos, GsSir2/Ago from Geobacter sulfurreducens and CcSir2/Ago from Caballeronia cordobensis, each encoded in a putative operon together with a Sir2 domain protein, and PgSir2-Ago from Paraburkholderia graminis, encoded as a single-chain variant, comprised of Sir2 and pAgo domains (Fig. 1A). The coding regions of the Sir2 and pAgo proteins in GsSir2/Ago and CcSir2/Ago systems overlap by 11 and 8 bp, respectively, indicating that they belong to the same operon (Supplementary Text). Next, we engineered heterologous E. coli cells by cloning GsSir2/Ago and CcSir2/Ago genes, or a single gene, encoding PgSir2-Ago into pBAD expression vectors under a PBAD promoter (Supplementary Table 2) and challenged them with phages or plasmids.
Sir2/Ago systems provide defence against phages. To test whether the GsSir2/Ago system provides defence against phages, we challenged E. coli host carrying the GsSir2/Ago system with a set of six E. coli phages spanning four morphological families including Podoviridae (T7), Siphoviridae (lambda-vir, SECphi27, SECphi18), Myoviridae (T4) and Microviridae (SECphi17, a ssDNA phage) (Supplementary Table 2). We measured the efficiency of plating (EOP) of these phages with and without L-arabinose induction of the GsSir2/Ago system. The system showed protection against two out of six phageslambda-vir (~100-fold) and SECphi27 (~1000-fold) (Fig. 1C). Catalytically active pAgos contain a conserved catalytic DEDX tetrad that is mutated in the inactive pAgos. Catalytically inactive short pAgos lack the canonical PAZ (PIWI-Argonaute-Zwille) domain, however, an APAZ (analogue of PAZ) domain is identified in putative Sir2, TIR or Mrr proteins associated with short pAgos. MID indicates middle; Llinker domain; N -N-terminal domain. Short pAgos from Geobacter sulfurreducens, Caballeronia cordobensis, and Paraburkholderia graminis associated with Sir2 protein were studied in this work. B, Schematic diagram of phage restriction assays. C, Efficiency of plating (EOP) of 2 phages infecting E. coli cells with and without the GsSir2/Ago system. The x-axis represents the number of p.f.u. Shown are the means of three replicates in the absence and in the presence of the inducer L-arabinose (L-Ara), and error bars are s.d. of the mean. Grey bars represent efficiency of plating (EOP) on pAgo-lacking cells and black bars are EOP in pAgo-containing cells. Representative images of plaque assays for lambda-vir and SECphi27 are also presented. D, EOP of lambda-vir and SECphi27 phages infecting E. coli cells with the wt and mutant GsSir2/Ago systems. Data represent p.f.u./ml in the presence of the inducer L-arabinose, an average of three replicates with error bars representing s.d. GsSir2(D230A)/Ago and GsSir2/Ago-HSH are variants that contain D230A mutation in the Sir2 domain or His6-StrepII-His6-tag (HSH-tag) on the C-terminus of pAgo. E, EOP of lambda-vir and SECphi27 phages infecting pAgo-lacking and pAgo-containing cells from Geobacter sulfurreducens (GsSir2/Ago), Caballeronia cordobensis (CcSir2/Ago), and Paraburkholderia graminis (PgSir2-Ago, a single-chain protein). Data represent p.f.u./ml in the presence of the inducer L-arabinose, an average of three replicates with error bars representing s.d. F, Lambda phage infection in liquid cultures of E. coli cells containing the GsSir2/Ago system. GsSir2/Ago-lacking E. coli MG1655 (shown in grey) or GsSir2/Ago-containing cells (shown in orange) were infected at t=0 at multiplicities of infection (MOI) of 0.05, 0.5 and 5. Each curve represents one individual replicate; two replicates for each MOI are shown. The collapse of GsSir2/Ago-containing cultures at high MOI suggests that most of the infected GsSir2/Ago-containing cells undergo abortive infection, while at low MOI a minority of cells produce viable phage progeny and the culture survives.
To probe the role of individual Sir2 and Ago proteins in antiviral defence, we engineered E. coli cells, carrying mutant variants of either the Sir2 or the Ago protein of the GsSir2/Ago system and performed small drop plaque assays using the lambda-vir and SECphi27 phages. In Sir2 variants, the highlyconserved D230 residue, presumably involved in NAD + binding was replaced by Ala residue (Supplementary Fig. 1D) 16 . Phage challenge assay revealed that the D230A mutation completely abolished defence against both phages (Fig. 1D). The Ago protein is catalytically inactive due to the active site mutations in the PIWI domain, therefore, to obtain a binding-deficient Ago variant, we fused a bulky 29 aa His6-StrepII-His6-tag (HSH-tag) at the C-terminus that is important for nucleic acid binding in other Agos 3,6,8,17 . We found that Ago C-terminal modification abolished protection from both phages (Fig. 1D). Thus, both the Sir2 and the Ago proteins are required for protection against phages by the GsSir2/Ago system. Additionally, we probed the ability of homologous CcSir2/Ago and PgSir2-Ago systems to restrict lambda-vir and SECphi27 phages (Fig. 1E). After L-arabinose induction, the PgSir2-Ago system showed ~500-fold protection against lambda phage and ~400-fold against SECphi27, while CcSir2/Ago system showed ~1000-fold protection against lambda and no protection against SECphi27 phage. Thus, homologous Sir2/Ago systems from three phylogenetically distant bacteria showed protection against phage infection, albeit with different efficiency and specificity.
Next, E. coli MG1655 cells lacking or containing the GsSir2/Ago system were infected in liquid culture with lambda phage at a multiplicity of infection (MOI) 0.05, 0.5 and 5 (Fig. 1F). At high MOI where, on average, a single bacterial cell is infected by a single phage, the culture collapses, while at low MOI the culture survives. This phenotype implies that GsSir2/Ago-mediated defence triggers cell death approximately at the same time when the phage-induced lysis occurs.
Sir2/Ago systems interfere with plasmid transformation. Further, to test whether heterologous expression of the GsSir2/Ago, CcSir2/Ago and PgSir2-Ago systems in E. coli cells (strains BL21-AI and DH10B) provides a barrier for plasmid transformation, four plasmids (pCDF, pCOLA, pACYC184, and pRSF) with different ori regions and copy numbers were used in a plasmid interference assay (Supplementary Table 2, Fig. 2, and Supplementary Fig. 2). We found that GsSir2/Ago system prevented only pCDF plasmid transformation reducing its efficiency nearly ~100fold (Fig. 2B). Next, we tested pCDF transformation efficiency in E. coli cells expressing the GsSir2/Ago mutants (Fig. 2C). Both Sir2 D230A mutation and pAgo HSH-tag modification that impaired phage restriction also abolished plasmid interference making cells permissive for pCDF plasmid transformation (Fig. 2C). CcSir2/Ago system like GsSir2/Ago provided resistance only for pCDF plasmid transformation ( Supplementary Fig. 2E) while cells carrying the single-chain PgSir2-Ago system was permissive for transformation of all four plasmids ( Supplementary Fig. 2E). . D, Top -schematic representation of ori exchange between pCDF and pCOLA plasmids; bottom -comparison of plasmid transformation efficiencies. pCDF, pCDF with CloDF13 ori exchanged with ColA ori (pCDF_ColA), pCOLA and pCOLA with ColA ori exchanged with CloDF13 ori (pCOLA_CloDF13) plasmids were used for transformation of E. coli cells carrying GsSir2/Ago system. E, Cell viability in the absence of antibiotic selection. In the case of the wt GsSir2/Ago system, the cell viability decreases on the plates even in the absence of Cb and Str antibiotics, for which the resistance genes are in pBAD and pCDF plasmids, respectively. The GsSir2/Ago system causes cell death in the presence of the target (pCDF plasmid).
Interestingly, although the GsSir2/Ago system in E. coli interfered with the pCDF plasmid transformation, pCOLA plasmid that differs mainly in the ori region and antibiotic resistance gene was permissive. To test whether the ori region determines differences in the transformation efficiency between pCDF and pCOLA plasmids, we swapped the ori sequences of pCDF and pCOLA (Fig. 2D). The pCDF plasmid with ColA ori instead of CloDF13 became permissive in E. coli cells expressing the GsSir2/Ago system, whereas transformation of pCOLA plasmid bearing CloDF13 ori instead of ColA ori was prevented. These results indicate that CloDF13 ori is a key element that controls plasmid transformation efficiency in E. coli cells expressing GsSir2/Ago system. The plasmid interference by the GsSir2/Ago system could be due to either the plasmid entry exclusion, replication inhibition or plasmid degradation. To eliminate the possible role of the plasmid entry barriers on the pCDF plasmid transformation efficiency, we engineered heterologous E. coli cells carrying two plasmids: pBAD plasmid expressing GsSir2/Ago (or its mutants) under control of PBAD inducible promoter and the pCDF plasmid providing antibiotic resistance, and tested cell viability in the presence or absence of inducer. In this case, the pCDF plasmid is already in the cell and provides streptomycin resistance, however antibiotic resistance should be lost if the plasmid is restricted after GsSir2/Ago expression. In the absence of induction, cell viability of E. coli cells carrying wt GsSir2/Ago (or its mutants) and an empty pBAD vector ( Supplementary Fig. 2C) was identical. In the presence of inducer the viability of cells expressing the wt GsSir2/Ago system, but not its mutants, significantly decreased (Fig. 2E), indicating that GsSir2/Ago interferes with the pCDF plasmid already present in the cell. Notably, a decrease in cell viability is observed in E. coli BL21-AI cells (Tc-resistant) without the cell selection for Str (pCDF) and Cb (pBAD) resistance. It cannot be excluded that upon recognition of pCDF the GsSir2/Ago system becomes activated and triggers cell death. A similar cell death phenotype triggered by the GsSir2/Ago has been observed during phage infection in liquid cultures (Fig. 1F). Taken together, these data show that the GsSir2/Ago system acts as a defence system against phages and plasmids via cell death or suicidal mechanism.
Short pAgo and associated Sir2 protein form a stable heterodimeric complex. To characterize the Sir2/Ago systems biochemically, we aimed to express individual Sir2 and Ago proteins in E. coli. The GsSir2 and CcSir2 proteins (but not PgSir2-Ago) were expressed and purified by liquid chromatography (Fig. 3A and Supplementary Fig. 3). The N-terminal His6-tagged GsSir2 and CcSir2 proteins co-expressed with Ago proteins co-purified on the Ni 2+ -affinity column ( Supplementary Fig.  3), indicating that Sir2 and pAgo proteins form a stable complex. Functionally compromised GsSir2(D230A)/Ago and GsSir2/Ago-HSH variants also formed a complex indicating that while the introduced mutations abolished the activity in vivo, they did not affect the protein complex structure ( Fig. 1, Supplementary Fig. 3). Further analysis of the oligomeric state of GsSir2/Ago and CcSir2/Ago complexes in solution using SEC-(MALS), mass photometry and small-angle X-ray scattering (SAXS) demonstrated that the complexes are heterodimeric in the wide range of protein concentrations (from 20 nM to 6.5 µM) ( Fig. 3B-D, and Supplementary Fig. 4). According to the SAXS data, the heterodimeric GsSir2/Ago complex acquires a notably asymmetric shape (Fig. 3D, Supplementary Fig. 4, Supplementary Table 3) that is consistent with the structural model of the heterodimer (Supplementary Text). In summary, the results show that pAgos and associated Sir2 proteins encoded by a single operon form a stable heterodimeric complex.  Table 1). Surprisingly, neither the 5'-terminal phosphate nor the 3'-OH end are important for ssDNA binding by both Gs and Cc complexes, as both bind the circular ssDNA and linear oligonucleotides with a similar affinity ( Fig. 4, Supplementary Fig. 6). Binding affinity of the GsSir2(D230A)/Ago and GsSir2/Ago-HSH mutants that are functionally compromised in cellular assays was similar to that of the wt complex ( Supplementary Fig. 6C, D). In summary, EMSA experiments suggest that in vitro ssRNA or ssDNA are preferable GsSir2/Ago guides.

Fig. 4. Nucleic acid binding by GsSir2/Ago in vitro.
A, EMSA of GsSir2/Ago binding to various DNA and RNA oligonucleotides. The asterisk denotes the radiolabelled strand. B, Binding of different targets by binary complexes preloaded with RNA or DNA guides. The pre-formed GsSir2/Ago complex with the guide strand (binary complex) was mixed with a radiolabelled target strand (see experimental details in "Materials and Methods"). Cg* denotes a control lane, equivalent to experimental lane marked by a black triangle, but with the guide, rather than the target, bearing the radioactive label. No displacement of the radiolabelled guide by the target strand is observed.
Next, we analysed ssDNA or ssRNA target binding by the binary GsSir2/Ago complexes pre-loaded with ssRNA or ssDNA guides. GsSir2/Ago-ssRNA binary complex showed tight binding to both complementary ssDNA and ssRNA targets. Importantly, RNA-guided GsSir2/Ago-ssRNA complex bound to complementary ssDNA target ~140-fold better than apo-GsSir2/Ago bound pre-annealed gRNA/tDNA heteroduplex (Fig. 4B, Table 1). On the other hand, DNA-guided GsSir2/Ago-ssDNA complex binding to the complementary ssRNA target was only 25-fold higher than apo-GsSir2/Ago binding to pre-annealed tDNA/gRNA heteroduplex (Fig. 4B, Table 1). Given these observations, we assume that GsSir2/Ago uses ssRNA as a guide for the recognition of the ssDNA target. The GsSir2/Ago complex binds NAD + and causes its depletion. Computational analysis of Sir2 domains showed that they possess a conserved NAD + -binding pocket (Supplementary Fig. 1D and Supplementary Fig. 5). To determine whether Sir2 domains can indeed bind endogenous NAD + , purified GsSir2/Ago and CcSir2/Ago complexes were heat-treated, protein aggregates removed by centrifugation, and the supernatant analysed by MS-HPLC ( Fig. 5 and Supplementary Fig. 7). The quantitative analysis showed that both the GsSir2/Ago and CcSir2/Ago complexes co-purified with bound endogenous NAD + in approx. 1:1 molar (Sir2:NAD + ) ratio ( Fig. 5A, B, Supplementary Fig.  3). However, in the case of the functionally inactive GsSir2(D230A)/Ago mutant, only 0.6% of all complexes were NAD + -bound indicating that the mutation severely compromised NAD + binding by the Sir2 domain (Fig. 5B). NAD + binding by the GsSir2 subunit and the similarity of the GsSir2 to the N-terminal NADase of the ThsA protein from the anti-phage Thoeris system 18 prompted us to investigate the level of endogenous NAD + in the presence of the induced GsSir2/Ago system and its target pCDF plasmid. In these experiments the corresponding E. coli cells were lysed, proteins were removed, and the amount of NAD + in the supernatant was examined by MS-HPLC. When the wt GsSir2/Ago system was induced in the presence of pCDF plasmid NAD + was depleted (~120-fold decrease), whereas in the case of the functionally inactive GsSir2(D230A)/Ago and GsSir2/Ago-HSH mutants the level of endogenous NAD + was similar to that of the empty pBAD vector (Fig. 5C, Supplementary Fig. 7). It should be noted that a significant ~30-fold decrease of the NAD + level was observed when wt GsSir2/Ago system expression was induced even in the absence of pCDF suggesting that the heterologously expressed system may be toxic to the cells resulting in their slower growth (Fig. 5C, Supplementary Fig. 7). To test the hypothesis that the activated GsSir2/Ago system similarly to the Thoeris anti-phage system depletes the endogenous NAD + through hydrolysis, we attempted to identify possible hydrolysis products (ADPR, cADPR, AMP, cAMP, ADP, cADP, nicotinamide, and adenine) using MS-HPLC albeit without success. Taken together, these results show that the wt GsSir2/Ago complex causes NAD + depletion when activated by the target pCDF plasmid. After lambda phage infection or pCDF plasmid transformation, the GsSir2/Ago system becomes activated and triggers NAD + depletion and cell death. Recognition of the CloDF13 replication origin of pCDF plasmid seems to be a major factor that triggers NAD + depletion by the Sir2 domain.

Discussion
Association of Sir2-like domains with short pAgo proteins has been identified bioinformatically in the pioneering Makarova et al. paper 16 . It has been speculated that Sir2-domain proteins can act as nucleases, however, the structure and function of Sir2 proteins remained to be elucidated. Here, we show that the Sir2 and short pAgo proteins form a heterodimeric complex (Fig. 3) similar to the heterodimeric complex between the short pAgo and the Mrr nuclease domain-containing protein 19 . It has been suggested 16 that PAZ domain that is present in long pAgos is replaced by APAZ domain in short pAgo systems, however, sequence comparison and structure modelling suggest that the canonical N domain and the L1 and L2-like subdomains are part of the Sir2 proteins, while the PAZ domain is missing (Fig. 1A, Supplementary Fig. 5). In this respect, single-chain PgSir2-Ago protein is similar to long pAgos that acquired Sir2 domain but lost the PAZ domain. We further show that the Sir2 domain isolated from the heterologous E. coli cells contains bound NAD + (Fig. 5). Next, we provide the first experimental evidence that the Sir2/Agos complex functions as a defence system against invading phages and plasmids ( Fig. 1 and Fig. 2). Intriguingly, plasmid interference assay using four plasmids with different replicons (Supplementary Table 2) revealed that GsSir2/Ago system prevents transformation only of the pCDF plasmid that contains CloDF13 ori (Fig. 2), suggesting that GsSir2/Ago may recognize specific replicon elements or structures. Indeed, ori swap between pCDF and permissive pCOLA plasmid made the latter sensitive to GsSir2/Ago interference. It has been previously shown that ColE1-like origins, including CloDF13, use two sense/antisense RNAs for priming the replication process that proceeds through the R-loop intermediate [20][21][22] . RNA is also involved in the priming of lambda phage theta replication that also involves R-loop formation, suggesting that priming RNAs/replication intermediate could be a recognition determinant for Sir2/Ago systems 23 . It is possible that Ago subunit in GsSir2/Ago complex, which shows a preference for ssRNA, binds ssRNA to interact with an ssDNA target, interfering with replication. Such GsSir2/Ago interaction with an ori region during the lambda phage infection or the pCDF plasmid transformation could activate the Sir2 effector subunit that depletes NAD + leading to the cell death, thereby, restricting phage propagation (Fig. 5D, the accompanying paper). A similar anti-phage defence mechanism based on NAD + exhaustion has been shown for the Thoeris and the Pycsar systems 24,25 . In the Thoeris system of Bacillus cereus MSX-D12, the activated Sir2 domain is similar to that of the GsSir2/Ago system and performs the hydrolysis of NAD + to ADPR and nicotinamide 25 . However, in the case of the GsSir2/Ago system, we were unable to detect ADPR and nicotinamide suggesting that despite the similarity of Sir2 domains in both systems, the fate of NAD + might be different. It cannot be excluded that instead of the NAD + hydrolysis, GsSir2/Ago may catalyze the transfer or polymerization of NAD + (or its cleavage products) on other cellular components (e.g. proteins or nucleic acids). Further structural and biochemical studies are underway to establish the structure of the heterodimeric Sir2/Ago complex and mechanisms of ori recognition and Sir2 activation that triggers NAD + transformation.

Methods
Oligonucleotides used in this work. All synthetic DNA oligonucleotides used for cloning and sitespecific mutagenesis were purchased from Metabion (Germany) and are listed in Supplementary  Table 1.

Cloning and mutagenesis.
A whole operon of the GsSir2/Ago system, composed of the Sir2 (GSU1360, NP_952413.1) and Ago (GSU1361, NP_952414. To swap ori regions between pCOLA and pCDF plasmids the DNA fragments containing ColA and CloDF13 ori were amplified by PCR using the oligonucleotides MZ-1217/MZ-1218 and MZ-1230/MZ-123, respectively (Supplementary Table 1). The resulting DNA fragments were digested by NheI and XbaI (ThermoFisher cat#FD0684) and ligated into pCOLA and pCDF vectors precleaved with NheI and XbaI and dephosphorylated using FastAP.
To swap streptomycin resistance for kanamycin in the pCDF plasmidplasmids pCDF and pCOLA were cleaved with NheI and Eco81I (ThermoFisher cat#FD0374) and isolated using a runView electrophoresis system (Cleaver Scientific). The purified fragments were then ligated into pCDF to yield a pCDF_Kn plasmid.
All gene sequences were confirmed by sequencing; links to DNA and protein sequences are presented in Supplementary Table 2.
Phage restriction assay. E. coli MG1655 (ATCC 47076) cells carrying pBAD plasmids expressing wt or mutated Sir2/Ago systems were used for phage infection assays as described below. Wholeplasmid sequencing was applied to all transformed E. coli clones to verify the integrity of the system and lack of mutations, as described before 27 .

SEC-(MALS) and mass photometry.
Size-exclusion chromatography of GsSir2/Ago complexes was carried out at room temperature using Superdex 200 10/300 GL column (GE Healthcare) preequilibrated with a buffer (20 mM Tris-HCl (pH 8.0 at 25°C), 500 mM NaCl). A calibration curve was generated by measuring the elution volumes of a series of standard proteins of known molecular mass (Bio-Rad). The molecular masses of pAgos complexes were calculated by interpolating its elution volume onto the calibration curve. SEC-MALS of GsSir2/Ago and CcSir2/Ago complexes was performed at room temperature using Superdex 200 10/300 GL column (GE Healthcare) preequilibrated with a buffer (20 mM Tris-HCl (pH 8.0 at 25°C), 500 mM NaCl, 0.03% NaN3, 1 mM DTT), at 0.4 ml/min flow rate. Sample concentrations were 6 µM and 6.5 µM for GsSir2/Ago and CcSir2/Ago, respectively. The light scattering signals were monitored on a miniDawn TREOS II detector, concentrations of protein samples were measured using an Optilab T-rEX refractive index detector (Wyatt Technologies). Data were analysed in Astra software (Wyatt Technologies) using dn/dc value of 0.185 mL/g. Mass photometry of the GsSir2/Ago complex was performed using a Refeyn OneMP system (Refeyn). The protein complex was diluted to 20 nM in a buffer containing 20 mM Tris-HCl pH 8.0, 500 mM NaCl before measurement.
Small-angle X-ray scattering (SAXS) analysis. The synchrotron SAXS data were collected at beamline P12 operated by EMBL Hamburg at the PETRA III storage ring (DESY, Hamburg, Germany) 30 . GsSir2/Ago sample in the the storage buffer was transferred into the the sample buffer (20 mM Tris-HCl pH7.5, 200 mM NaCl, 5 mM MgCl2, 2 mM β-mercaptoethanol) using gel-filtration NAP column (GE Healthcare) and concentrated by ultrafiltration to 1.2, 1.3, 1.6 and 5.5 mg/ml concentrations. The data were collected at the wavelength of 0.124 nm and the distance to the detector (Pilatus 2M, Dectris) was set to 3 m. Samples in the sample changer were kept at 10°C, capillary temperature was set to 20°C. Twenty frames exposed 0.045 sec were averaged for each concentration. The s-range of collected data was from 0.0133796 to 3.7925 nm -1 . The data were analysed using programs of ATSAS 2.8.4 (r10552) suite 31 . Data were normalized to an absolute scale with water as standard. As the data collected for the sample with concentration 1.22 mg/ml were noisy at higher s ( Supplementary Fig. 4), and higher concentration data showed more aggregation at low s, we used a merged dataset produced with PRIMUS 32 . Scattering data were parameterized and indirectly Fourier transformed with GNOM5 33 . Structural parameters of this dataset are summarized in Supplementary  Table 3. Dimensionless Kratky plot in Supplementary Fig. 4   Binding experiments for the binary GsSir2/Ago complex were conducted by first pre-mixing 5'phosphorylated ssRNA or ssDNA guide with the GsSir2/Ago complex at 1:1 ratio in the same buffer as above. The binary GsSir2/Ago:NA guide complex was then diluted to 2x final reaction concentration (in respect to guide) in the same buffer and mixed with a complimentary 5'-32 P-target oligonucleotide. The final reaction contained 0.1 nM target NA and 0, 0.02, 0.05, 0.1, 0.2 0.5, and 1 nM of GsSir2/Ago:NA guide complex. In the Fig. 5 control lane (Cg*) shows that there is no guide NA displacement in the binary GsSir2/Ago complex containing 5'-32 P-labelled ssRNA or ssDNA by the non-labelled 5'-phosphorylated target NA. Three independent replicates were performed.
The binding reaction mixtures were analysed by electrophoretic mobility shift assay (EMSA) in a PAA gel (8% 29:1 acrylamide/bis-acrylamide in TAE). The electrophoresis TAE buffer was supplemented with 5 mM magnesium acetate. Radiolabelled substrates were detected and quantified using a phosphor imager. The results were analysed with OptiQuant and OriginPro software. The Kd was calculated from the following formula: where SNBunbound substrate, nM; S0initial substrate concentration, nM; E0initial protein complex concentration, nM; Kddissociation constant.

GsSir2/Ago complex resembles long PAZ-free pAgos containing an additional effector domain.
To get an insight into the 3D structure of GsSir2/Ago and the other two systems, we used AlphaFold 38 to generate corresponding structural models. The constructed models for all three systems showed close structural similarity with each other as could be expected based on their homology ( Supplementary Fig. 5). Therefore, we will focus only on the GsSir2/Ago complex. In this complex, the Sir2 and Ago subunits bind together to form a structure similar to that of a single-chain long pAgo protein such as RsAgo ( Supplementary Fig. 5). Based just on a structural similarity search against PDB, GsSir2/Ago appears to be most similar to the TtAgo structure classified as a long-A pAgo 8 . However, another close structural match, RsAgo, a member of long-B pAgos 8 , has a catalytically inactive PIWI domain. This is a shared feature with short pAgo proteins and therefore comparison of GsSir2/Ago with RsAgo might be more relevant.
The APAZ part of the GsSir2 chain is structurally similar to the combination of N, L1 and L2 domains of RsAgo (similarity to the N domain has been already proposed before 11,39 ). However, the L2 linker domain in GsSir2/Ago corresponds to the two fragments, C-terminus of Sir2 and N-terminus of Ago. Importantly, the PAZ domain, required for the 3'-end recognition of the guide strand in long pAgos, is missing from the GsSir2/Ago structure altogether. Additionally, GsSir2/Ago has a smaller N domain than long pAgos, which could also alter nucleic acid binding. PIWI and MID domains are structurally similar to corresponding RsAgo domains. Notably, GsSir2/Ago has a longer loop (residues 268-272) compared to the corresponding region in RsAgo. This loop shows some steric overlap with the copied-in RNA/DNA duplex, but presumably, its conformation may adjust to accommodate RNA/DNA heteroduplex. The loop contains two positively charged residues (R269, K270) that might be involved in the binding of the nucleic acid backbone.
In the structural model, the N-terminal Sir2 domain of the GsSir2 protein is attached to the C-terminal domain corresponding to the N-terminal region of a long pAgo through a long linker. In effect, the two domains of GsSir2 are positioned at the opposite extremes of the complex (Sir2 domain is bound to the PIWI domain whereas the C-terminal domain is bound to the MID domain). Sir2 domain of GsSir2/Ago is structurally similar to canonical Sir2 proteins suggesting it could bind NAD + in a similar way. However, the overall structurally closest homolog is the Sir2 domain of the ThsA protein from the Thoeris defence system 18 ( Supplementary Fig. 5). The ThsA residues, N112 and H152, shown to be essential for NAD + hydrolysis have their counterparts in GsSir2 (N142 and H186). Interestingly, the loop containing H186 in different GsSir2/Ago models displays some conformational heterogeneity hinting at possible flexibility, which might be relevant for the activity regulation. These two positions are conserved among both short pAgos and other Sir2 homologs ( Supplementary Fig. 1D). On the other hand, the GsSir2 D230 residue shown here to be important for NAD + hydrolysis is conserved in ThsA Sir2 domain, but not in canonical Sir2 proteins (e.g., V193 in TmSir2).

Supplementary methods
Sequences analysis. Lists of pAgo homologues and associated APAZ domains were retrieved from supplementary data of the Ryazansky et al. article 8 . Thermotoga maritima Sir2 homologues were collected from the SwissProt database 40 using BLAST 41 (1e-5 e-value cutoff). Full-length sequences were retrieved from NCBI. Bacillus cereus ThsA (PDB ID 6LHX) homologues were collected using BLAST from UniRef50 42 database (1e-10 e-value cutoff). To remove other SIR2 homologues, these ThsA homologue sequences were clustered with CLANS 43 . Cluster separated at p=1e-70 was selected as the representative ThsA group. Fragmented ThsA sequences and sequences missing one of the domains were discarded. Multiple sequence alignments were generated using MAFFT (l-INS-i mode for high accuracy) 44 . Jalview 45 was used for multiple sequence alignment analysis, cutting and visualization. Sequence motif visualization was done using WebLogo 3 server 46 . To align the multiple sequence alignments of GsSir2 homologues TmSir2 homologues and ThsA homologues, GsSir2 sequence alignments to TmSir2 and BsThsA were obtained using the HHpred server 47 .
Phylogenetic tree construction. The phylogenetic trees were constructed with FastTree 48 using the WAG model of amino acid substitution 49 and the gamma model of rate heterogeneity. Prior to phylogenetic analysis, multiple sequence alignment positions containing more than 50% gaps were removed using trimAl 50 . Phylogenetic tree visualization was done with iTOL 51 .

Genomic neighbourhood analysis.
For genomic neighbourhood analysis, the Geobacter sulfurreducens, Caballeronia cordobensis, and Paraburkholderia graminis genomes (respective GenBank accessions: GCA_000210155.1, GCA_001544575.2 and GCA_000172415.1) and all associated sequence and annotation data were obtained from NCBI (ftp://ftp.ncbi.nlm.nih.gov/genomes/Bacteria/). Genes in the neighbourhood of each pAgo (10 upstream and 10 downstream) were identified based on available genome annotations. To refine available functional annotations of these genes, searches through Pfam 52 and PDB databases were performed using the HHpred server 47 .
Expression and purification of CcSir2/Ago complex. Expression vector constructs of the CcSir2/Ago system were used to transform E. coli BL21(DE3) strain. Transformed bacteria were grown at 37°C in LB medium in the presence of 50 μg/ml ampicillin until OD600 = 0.7 was reached.
Then, the medium was cooled to 16°C temperature and proteins were expressed for 16 h by adding 0.1% w/v L-arabinose. Harvested cells were disrupted by sonication in buffer A (20 mM Tris-HCl (pH 8.0 at 25°C), 1.0 M NaCl, 2 mM phenylmethylsulfonyl fluoride, 5 mM 2-mercaptoethanol), and cell debris was removed by centrifugation. TwinStrep-CcSir2/Ago complex was purified to > 90% homogeneity by chromatography through Strep-Tactin XT Superflow (Iba), HiLoad Superdex 200, HiTrap Heparin HP columns (GE Healthcare). Purified proteins were stored at −20 °C in a buffer containing 20 mM Tris-HCl (pH 8.0 at 25°C), 500 mM NaCl, 2 mM DTT and 50% v/v glycerol. The identity of the purified proteins was confirmed by mass spectrometry. Protein concentrations were determined from OD280 measurements using the theoretical extinction coefficients calculated with the ProtParam tool available at http://web.expasy.org/protparam/. TwinStrep-CcSir2/Ago complex concentrations are expressed in terms of heterodimer.
Construction and analysis of Sir2/Ago structural models. Structural models were generated using the AlphaFold method 38 implemented as ColabFold 53 , an online Google Colaboratory notebook. For modelling, the 'Alphafold2_advanced' notebook (colab.research.google.com/github/sokrypton/ColabFold/blob/main/beta/AlphaFold2_advanced.ipy nb) was used. GsSir2/Ago and CcSir2/Ago complexes were modelled as heterodimers, whereas the Pg sequence representing a fusion of Sir2 and a short pAgo (PgSir2-Ago) was modelled as a monomer. The modelling pipeline was run with default parameters except for the multiple sequence alignment (MSA) pairing, which was set to 'paired+unpaired'. The best-of-five model in each case was selected using VoroMQA 54 . Structure similarity searches of models/domains against PDB were performed using Dali 55 . Structure analysis and visualization was performed using UCSF Chimera 56 .
Putative binding sites of Sir2/Ago complexes were investigated by simply copying the RNA/DNA duplex and NAD + from RsAgo (PDB ID: 5AWH) and TmSir2 (PDB ID: 2H4F) structures correspondingly after their superposition onto structural models. No attempts to remove possible clashes between protein models and either RNA/DNA or NAD + were made.
Nucleic acid cleavage assay. The same linear oligonucleotides used for EMSA (Supplementary  Table 1) were given as substrates for nucleic acid cleavage assay. To raise the total substrate concentration, 5'-32 P radiolabelled oligonucleotides were mixed with appropriate cold 5'Poligonucleotides at a ratio of 1:4 of hot:cold, and diluted to a working concentration of 100 nM in reaction buffer (33 mM Tris-acetate, pH 7.9, supplemented with 10 mM magnesium acetate, 66 mM potassium acetate, 0.1 mg/ml BSA, 5 mM DTT). Protein dilutions were made using the same reaction buffer to 2x final reaction concentration. Protein complexes and nucleic acids were mixed in final concentrations of 50 nM NAs and 0, 50 or 500 nM protein, incubated for 1 hour at 25 °C. The reaction was stopped with the addition of 2x 95% formamide dye and incubating for 5 min at 95 °C. Reaction products were resolved by denaturing PAA gel electrophoresis (21% 29:1 acrylamide/bis-acrylamide in TBE (89 mM Tris, 89 mM boric acid, 2 mM EDTA), supplemented with 8 M urea), visualised with a phosphor imager and analysed in OptiQuant software.
Supplementary Fig. 1. Bioinformatic analysis. A, Left -a circular phylogenetic tree was generated according to supplementary data provided with Ryazansky et al. 8 Long-A pAgo variants are coloured in green (truncated variants without the PAZ domain, light green), long-B pAgo proteins are light green (truncated variants without PAZ, green), and short pAgo proteins are orange. pAgo proteins containing the catalytic tetrad DEDX in their PIWI domain are indicated in blue on the outer circle; pAgos with inactivated PIWI domain are indicated in light grey on the outer circle. pAgo proteins of the GsSir2/Ago, CcSir2/Ago and PgSir2-Ago systems are indicated by 'Gs', 'Cc' and 'Pg', respectively. Right -Circular phylogenetic tree of APAZ domains. The circular phylogenetic tree of the five groups of APAZ domains was generated using APAZ domain alignments from Ryazansky et al. 8  MID domain biochemical assay -Kaya et al. 57 ). In Sir2/Ago models the N, L1 and L2-like domains previously identified as the APAZ domain correspond to the analogous domains of long pAgos. B, Gs, Cc and Pg Sir2 structural models (cut from full-length models) compared to canonical Sir2 deacetylase TmSir2 and the Sir2 domain of Thoeris defence system protein ThsA. Structures are coloured based on secondary structure. Positions corresponding to ThsA Sir2 N112 and H152 are indicated in green. These residues have been shown to be critical for NAD + hydrolysis in ThsA 18 . GsSir2 D230 and corresponding positions in other structures are indicated in cyan. NAD + was also superimposed on the ThsA Sir2 structure from TmSir2.