Bacterial retrons encode tripartite toxin/antitoxin systems

Retrons are genetic retroelements, commonly found in bacterial genomes and recently repurposed as genome editing tools. Their encoded reverse transcriptase (RT) produces a multi-copy single-stranded DNA (msDNA). Despite our understanding of their complex biosynthesis, the function of msDNAs and therefore, the physiological role of retrons has remained elusive. We establish that the retron-Sen2 in Salmonella Typhimurium encodes a toxin, which we have renamed as RcaT (Retron cold-anaerobic Toxin). RcaT is activated when msDNA biosynthesis is perturbed and its toxicity is higher at ambient temperatures or during anaerobiosis. The RT and msDNA form together the antitoxin unit, with the RT binding RcaT, and the msDNA enabling the antitoxin activity. Using another E. coli retron, we establish that this toxin/antitoxin function is conserved, and that RT-toxin interactions are cognate. Altogether, retrons constitute a novel family of tripartite toxin/antitoxin systems.

To test whether the retron-antitoxin is counteracting RcaT by down-regulating its expression, 147 we Flag-tagged rcaT in STm WT and antitoxin deletion strains, and quantified RcaT levels at 148 37°C and 20°C. RcaT levels remained similar in all mutants, and if anything, only slightly 149 decreased in the ΔrrtT background (ED Fig. 5A-B). This is presumably due to mild polar effects 150 of removing rrtT, which is located directly downstream of rcaT. In all cases, RcaT-3xFlag was 151 fully functional as toxin (ED Fig. 5C). Therefore, RcaT expression is not inhibited by the retron-152 antitoxin, which would have led to higher RcaT levels in the mutants. 153

154
Having excluded that RcaT is counteracted by the retron-antitoxin at an expression level, we 155 tested if this occurs by direct protein-protein interactions, as in type II TA systems. To do so, 156 we affinity purified chromosomally encoded RT-3xFlag or RcaT-3xFlag, in wildtype and retron 157 antitoxin-deletion STm strains (ΔxseA, and Δmsrmsd/Δmsd), at 37°C and 20°C, and quantified 158 interacting proteins by quantitative mass-spectrometry (AP-qMS). Indeed, RT and RcaT 159 strongly and reciprocally pulled down each other ( Fig. 3 & ED Fig. 6). Notably, the RT-RcaT 160 interaction occurred independently of the presence/maturation of msDNA, or of the 161 temperature the toxin is active in. Tagged RT-3xFlag was fully functional in inhibiting RcaT 162 (ED Fig. 7), and did not alter RT or RcaT protein levels in the input samples, compared to WT 163 (ED Fig. 8A). In contrast, tagging RcaT led to lower RT levels in the input samples (ED Fig.  164 8B), but RcaT-3xFlag remained functional as toxin (ED Fig. 5C). This likely explains the lower 165 levels of RT enrichment in the RcaT-3xFlag pull-downs (Fig. 3C). Thus, RcaT and RT stably 166 interact with each other independently of the presence or maturation of msDNA, and of 167 temperature. 168 169 Although RcaT and RT interact in the absence of msDNA, mature msDNA production is 170 essential for antitoxin activity. We thus wondered whether msDNA-protein interactions are 171 involved in the antitoxin activity. Retron-RTs from different species have been previously 172 shown to co-purify with their mature msDNA products 15,16 . In order to assess whether  Sen2 also interacts with msDNA-Sen2, we first purified an RT-Sen2-6xHis protein fusion upon 174 concomitant msrmsd-Sen2 expression in E. coli (ED Fig. 9A). The RT-Sen2-6xHis version was 175 functional, as it could counteract the RcaT toxicity (ED Fig. 9B). At a second stage, we isolated 176 total DNA from the purified RT-Sen2-6xHis protein sample, which yielded both mature and 177 unprocessed msDNA-Sen2 (ED Fig. 9C). Therefore, the RT-Sen2 and msDNA-Sen2 interact 178 with each other and are required together for the antitoxin activity. 179 6   The RT confers specificity to the antitoxin & the msDNA modulates the antitoxin activity  180   Antitoxins of TA systems are specific against their cognate toxins. Since the retron-antitoxin is  181   composed by both the RT and msDNA, we wondered which part provides the antitoxin  182   specificity. To address this, we reasoned we could use a different retron-TA, with evolutionary  183 diverged retron-components, and swap the individual components between retrons to make 184 retron chimeras. For this, we used a novel retron from E. coli NILS-16, a clinical E. coli isolate 185 21,24 , that we named retron-Eco9. Retron-Eco9 has an RT and an accessory gene, which are 186 49% and 43% identical to RT-Sen2 and RcaT-Sen2 at the protein level, respectively. msr-187 Eco9 and msd-Eco9 are 85% and 58% identical to their Sen2 counterparts at the nucleotide 188 level (ED Fig. 10A), and the msDNA-Eco9 retains a similar overall structure to msDNA-Sen2 189 (ED Fig. 10B). As for retron-Sen2, expressing rcaT-Eco9 inhibited the growth of E. coli, while 190 expressing the entire retron-Eco9 did not (ED Fig. 10C). In addition, the msDNA- Eco9 required 191 RNase H/ExoVII for its production and function (ED Fig. 10D-E). Thus, the retron-Eco9 192 encodes a TA system similar to retron-Sen2, and its toxin (RcaT-Eco9) is inhibited by a retron-193 encoded antitoxin (RT-msDNA-Eco9). 194

195
To assess which part of the antitoxin unit is cognate to the toxin, we constructed chimeric 196 retron constructs between Sen2 and Eco9. Retron-RTs are known to be highly specific in 197 reverse transcribing their cognate msrmsd-RNA, by binding to specific RNA-structures of the 198 msr region 25 . Cross-specificity between non-cognate RT-msrmsd pairs has only been 199 observed between highly homologous retrons 13 . Thus, we first evaluated if the Sen2 and Eco9 200 RTs could transcribe their non-cognate msrmsd. To assess this, we isolated msDNA from E. 201 coli co-expressing PBAD-RT plasmids (Sen2 [Se], or Eco9 [Ec]) with Ptac-msrmsd plasmids 202 (Se, Ec, or -). Both RT-Sen2 and RT-Eco9 could use their non-cognate msrmsd and produce 203 msDNA (Fig. 4A), albeit RT-Eco9 was slightly less efficient (lane Ec-Sesee also ED. Fig 10F  204 for repercussions of this lower efficiency). 205

206
To assess the specificity of RcaT to RT, we then made arabinose-inducible plasmids, carrying 207 binary combinations of both RcaT and RT from retrons Se, and Ec: Se-Se, Ec-Se, Se-Ec, and 208 Ec-Ec. We co-expressed these plasmids with IPTG-inducible Ptac plasmids, expressing 209 msrmsd from Sen2, Eco9, or the empty vector, to test which of the RT-msDNA systems 210 retained the antitoxin-activity. As expected, both Sen2 and Eco9 RcaT inhibited the growth of 211 E. coli in the absence of msDNA, irrespectively of the RT co-expressed (Fig. 4B). While 212 cognate or non-cognate msDNA template could activate the antitoxin activity in cognate RT-213 RcaT combinations (Se-Se, or Ec-Ec) and neutralize the RcaT toxicity, this did not work in 214 non-cognate RT-RcaT combinations (Se-Ec, or Ec-Se) (Fig. 4B, ED Fig 10F). This 215 demonstrates that although RTs can produce msDNA from non-cognate msrmsd templates, 216 7 and the formed RT-msDNA complex is an active antitoxin, this active antitoxin can only act 217 against its cognate toxin. Thus, the RT-RcaT interactions (Fig. 3) are cognate within retron-218 TAs, and essential for antitoxin-activity. 219 220 In summary, the RT-RcaT interaction provides the specificity for the TA system, as the RT can 221 interact directly with RcaT in the absence of msDNA, and it cannot be exchanged for a 222 homologous RT from another retron. In contrast, the msDNA sequence is interchangeable, at 223 least to some extent, but the RT-msDNA interaction is absolutely required to form an active 224 antitoxin unit. . Third, msDNA is required, but is not 245 sufficient for antitoxin activity. This last point is harder to establish, as msDNA is not produced 246 in cells lacking its cognate RT. We bypassed this limitation by using a second retron, to make 247 hybrid-retrons. Although non-cognate RT-msrmsd pairs produce msDNA (Fig. 4A), and these 248 form active antitoxins against cognate RT-RcaT pairs, they were not enough to inhibit the non-249 cognate RT-RcaT pairs (Fig. 4B). Therefore, RTs are needed for the antitoxin activity, not only 250 because they produce msDNA, but also because they provide the antitoxin specificity by 251 simultaneously binding the toxin (Fig. 3) and the msDNA (ED Fig. 9C). 252 253 8 There are also several open aspects of this retron-TA model. Although we provide strong 254 evidence that the RT-msDNA interaction activates the antitoxin unit, the exact mechanism 255 remains to be resolved. It is still possible that the msDNA also interacts with RcaT in the 256 tripartite RcaT-RT-msDNA complex. In addition, we do not know the cellular target of RcaT,257 or why it preferentially inhibits growth in cold and anaerobic conditions, and only upon 258 overexpression it impacts growth at 37°C in aerobic conditions. The two phenotypes may be 259 linked, due to the target of the toxin being more relevant in cold/anaerobiosis. Alternatively, 260 the toxin itself could be post-translationally activated in these conditions. A very different toxin 261 in Pseudomonas putida, GraT, also causes cold-sensitivity upon antitoxin perturbations, and 262 toxicity when over-expressed 31 . GraT seems to cause cold-sensitivity by attacking ribosome 263 biogenesis 32 . Yet, an anaerobic-sensitivity toxin phenotype has not been reported before for 264 GraT, or any toxin of any TA system as far as we know. Therefore, it is possible that RcaT has 265 a distinct toxicity mechanism. Finally, we show here that another retron in E. coli NILS-16 266 encodes a  Table S1, and raw data will be uploaded in an appropriate server prior to publication. 305 All unprocessed source images are available upon request. 306

COMPETING INTEREST DECLARATION 308
We declare no competing financial interests. 309

ADDITIONAL INFORMATION 311
Supplementary information is available for this paper. Correspondence and requests for 312 materials should be addressed to AT (typas@embl.de). 313       Single bacterial-colonies were inoculated in 2 mL LB, and incubated over-day aerobically at 517 37°C in a roller drum (6 hours, until OD595~5). Over-day cultures were stepwise serially-diluted 518 eight times (ten-fold) in LB (100μL culture + 900μL LB). Using a 96-pinner (V&P Scientific, 519 catalogue number: VP 404), ~10 μL of culture dilutions were spotted on LB-plates containing 520 appropriate antibiotics, and 0.2% arabinose if needed (arabinose was only present in plates, 521 not in cultures). Spots in growth tests shown in Figures 1D, 2D, ED 1A, ED 2A, ED 3C, and 522 ED 9B were spotted manually (10 μL) with a multichannel pipette. LB-plates were incubated 523 overnight (13-15 hours) aerobically at 37°C, in a humid incubator. For cold-sensitivity growth 524 tests, LB-plates were incubated for 36, 48, or 72 hours, at 25°C, 20°C, or 15°C, respectively. 525 526 msDNA isolation and running msDNA in TBE-Acrylamide gels 527 msDNA was isolated by alkaline lysis (reagents as described in 43 ). msDNA was over-produced 528 by over-expressing the reverse transcriptase and the msrmsd region (msrmsd-rrtT), in order 529 to be able to purify msDNA from small culture volumes (see Tables S2-S3 for strain/plasmid  530 combinations used for each msDNA isolation). Strains were inoculated at OD595=0.01 in 20 531 mL LB, supplemented with appropriate antibiotics and 0.2% arabinose (to over-express 532 msDNA). Cultures were incubated for 5-6 hours at 37°C with rigorous shaking. After this, cells 533 16 were put on ice and approximately 10 mL was centrifuged (4,000 rpm/15 min/4°C)after 534 correcting for OD 595 . Pellets were washed once with ice-cold PBS, re-suspended in alkaline 535 solution 1, transferred into 1.5 mL Eppendorf tubes, and alkaline solutions II and III were cycled 536 as described in 43 . After centrifugation (14,000 rpm/20 min/4°C), supernatants were extracted 537 twice with Phenol: Chloroform: Isoamyl-Alcohol (50:48:2, pH 8), and nucleic acids were 538 precipitated overnight at 4°C with isopropanol. Precipitated nucleic acids were centrifuged at 539 14,000 rpm/60 min/4°C. Pellets were re-suspended once with 1 mL of 70% Ethanol, and 540 centrifuged again at 14,000 rpm/60 min/4°C. Pellets (msDNA extracts) were air-dried (15 541 minutes), resuspended in 10 μL of distilled water containing RNase A (20 μg/mL), and 542 incubated at 37°C for 30 minutes. Samples were subsequently kept at -80°C until further use. cultures were transferred to 50 mL tubes, with culture volumes per strain being normalized 555 based on OD to adjust for total protein-levels across strains. Cultures were centrifuged at 5,000 556 rpm/10 min/4°C and the supernatant was discarded. Pellets were washed once with 50 mL of 557 ice-cold PBS, and cells were centrifuged again. Pellets were then frozen at -80°C. 558 Subsequently, pellets were re-suspended in 1.