Toxin release by conditional remodelling of ParDE1 from Mycobacterium tuberculosis leads to gyrase inhibition

Abstract Mycobacterium tuberculosis, the causative agent of tuberculosis, is a growing threat to global health, with recent efforts towards its eradication being reversed in the wake of the COVID-19 pandemic. Increasing resistance to gyrase-targeting second-line fluoroquinolone antibiotics indicates the necessity to develop both novel therapeutics and our understanding of M. tuberculosis growth during infection. ParDE toxin–antitoxin systems also target gyrase and are regulated in response to both host-associated and drug-induced stress during infection. Here, we present microbiological, biochemical, structural, and biophysical analyses exploring the ParDE1 and ParDE2 systems of M. tuberculosis H37Rv. The structures reveal conserved modes of toxin–antitoxin recognition, with complex-specific interactions. ParDE1 forms a novel heterohexameric ParDE complex, supported by antitoxin chains taking on two distinct folds. Curiously, ParDE1 exists in solution as a dynamic equilibrium between heterotetrameric and heterohexameric complexes. Conditional remodelling into higher order complexes can be thermally driven in vitro. Remodelling induces toxin release, tracked through concomitant inhibition and poisoning of gyrase activity. Our work aids our understanding of gyrase inhibition, allowing wider exploration of toxin–antitoxin systems as inspiration for potential therapeutic agents.


Introduction
Despite concerted efforts, tuberculosis remains a major cause of morbidity and a leading cause of mortality worldwide ( 1 ).There were approximately 10.6 million new cases of tuberculosis in 2021, and it is estimated that over a quarter of the world's population would demonstrate an immunological response to the causative agent Mycobacterium tuberculosis ( 1 ).DNA gyrase is the single type II topoisomerase encoded by M. tuberculosis and remains an important drug target (2)(3)(4).DNA gyrase is responsible for the maintenance of DNA topology ( 5 ), and alters supercoiling through cycles of generating and re-ligating double-stranded (ds) breaks in DNA ( 6 ).This essential process that produces potentially cytotoxic dsDNA breaks has made type II topoisomerases an attractive target in both antimicrobial ( 7 ) and anti-cancer ( 8 ) drug research.
The widespread and highly conserved ParDE TA systems, part of the RelE / ParE superfamily ( 25 ), target DNA gyrase ( 26 ,27 ).M. tuberculosis encodes two ParDE systems and several studies on the ParE1 and ParE2 toxins have shown their ability to inhibit gyrase enzymes from E. coli , M. smegmatis and M. tuberculosis ( 22 ,28-29 ).The M. tuberculosis ParDE systems have both been shown to be regulated in response to environmental stresses associated with infection (30)(31)(32)(33).The parE1 toxin gene was also identified through mutational studies as both important for survival in activated macrophages, and in dissemination to the spleen ( 33 ,34 ), potentially contributing to extrapulmonary tuberculosis.Expression profiling highlighted the ParE toxin genes, parE1 and parE2 , as some of the highest differentially regulated genes (second only to rv1045 encoding toxin MenT3 ( 35)) when M. tuberculosis was subjected to starvation, acidification, and first-line drug exposure in combinations for varying time-lengths ( 30 ).
Here, we present microbiological, biochemical, structural and biophysical characterisation of the ParDE1 and ParDE2 TA systems from M. tuberculosis .Our studies show conserved modes of antitoxin recognition with distinct, complex-specific, interactions.We demonstrate a unique quaternary ParDE structure, with ParDE1 forming a heterohexameric complex.Unexpectedly, we were readily able to detect and characterise higher order structures for the ParDE1 complex.Conditional remodelling of the ParDE1 complex can be thermally induced, allowing control of the dynamics of complex formation, driving ParDE1 from a heterotetrameric state into higher order structures.By combining the remodelling process with gyrase assays we demonstrate selective ParE1 release.These results expand our understanding of in vitro toxin-antitoxin activity, and suggest potential routes for gyrase inhibition.

DNA isolation and manipulation
Plasmid DNA was purified from transformed DH5 α cells using a NEB Monarch® Plasmid MiniPrep kit following the manufacturer's instructions.Larger amounts of negatively supercoiled plasmid (pSG483) DNA for assays was purified from transformed DH5 α cells using a Machery-Nagel Nucle-oBond Xtra Midi Plus EF kit following the manufacturer's instructions.Plasmids were eluted in dH 2 O for storage at −20 • C. Plasmids pTRB316 and pTRB696 were made previously ( 3 ).Plasmids pTRB568, pTRB569 and pTRB570 were generated commercially at Genscript using sequences optimised for E. coli expression.Plasmid derivatives of pJEM15 and pGMC were also generated commercially by Genscript.Plasmids are described in Supplementary Table S1 .

Preparation of nicked and linear form pSG483
For nicking, 10 μg pSG483 was incubated with 10 units of Nb.Bpu10I (ThermoFisher) in 1 x Buffer R (ThermoFisher) for 1 hr at 37 • C. The enzyme was deactivated by a further incubation step at 80 • C for 20 min.For linearisation, 10 μg pSG483 was incubated with 10 units of BamHI-HF® (NEB) in 1 x CutSmart buffer (NEB) for 1 h at 37 • C. The enzyme was deactivated after incubation by a further incubation step at 65 • C for 10 min.Conversion of supercoiled pSG483 into appropriate products was assessed by agarose gel electrophoresis.Both nicked and linear form pSG483 were subsequently stored at −20 • C.

Preparation of relaxed form pSG483
Initially, 50 μg pSG483 was nicked by incubation with 10 units Nb.Bpu10I (ThermoFisher) in 1 × Buffer R (ThermoFisher) for 4 h at 37 • C. The enzyme was deactivated by a further incubation step at 80 • C for 20 min.The reaction was allowed to cool to room temperature before being supplemented with ATP to a final concentration of 1 mM. 10 μl T4 DNA ligase was added and the reaction was left at room temperature for 16 h.After ligation, ethanol precipitation was performed to remove proteins.An equal volume of UltraPure™ phenol:chloroform:isoamyl alcohol (25:24:1, vol / vol / vol) (ThermoFisher) was added to the reaction mixture before vortexing briefly.The sample was centrifuged at 16 000 × g for 2 min and the resulting aqueous layer was removed and carried forward.An equal volume of chloroform (ThermoFisher) was added to the aqueous layer before centrifugation at 16 000 × g for 2 min.The resulting aqueous layer was carried forward and 1 / 10 volume 3 M sodium acetate pH 5.2 was added.Then, 2 volumes of 100% ethanol were added, briefly mixed by pipetting, and stored at -80 • C for 30 min.The sample was centrifuged at 16 000 × g and 4 • C for 20 min.The ethanol was removed, and the DNA pellet dried at room temperature.The DNA pellet was resuspended in room temperature dH 2 O to approximately 300 ng / μl.

Protein expression
Proteins were expressed and purified following published protocols ( 3 ), with small variations as appropriate.For the expression of the gyrase subunit proteins, GyrA and GyrB, Rosetta™ 2 pLysS cells were transformed with pTRB696 and pTRB316, respectively.Both gyrase subunits were expressed with a TEV protease cleavable N-terminal hexahistidine (6His) tag for purification.Cells were grown at 37 • C with shaking at 180 rpm to an optical density (OD 600 ) of 0.6 at which point the incubation temperature was reduced to 30 • C and IPTG was added to a final concentration of 0.8 mM to induce overexpression.Cells were subsequently grown for a further 4 hr at 30 • C with shaking at 160 rpm.
Each of the toxin-antitoxin systems in this study were expressed from Duet vectors ( Supplementary Table S1 ).Roset-ta™ 2 pLysS cells were transformed with the appropriate plasmids for the expression of the ParDE1 complex (pTRB569) or ParDE2 complex (pTRB570).Toxins ParE1 and ParE2 were expressed with a hSENP-2 cleavable N-terminal 6His-SUMO tag for purification of the complexes ( 35 ).Cells were grown at 37 • C with shaking at 180 rpm to an optical density (OD 600 ) of 0.6 at which point the incubation temperature was reduced to 18 • C and IPTG was added to a final concentration of 1 mM to induce overexpression.Cells were subsequently grown for a further 16 h at 18 • C with shaking at 160 rpm.To express the ParDE1 complex in the heterotetramer stoichiometry, the protocol above was followed with the following adjustments: IPTG was added to a final concentration of 0.5 mM and expression temperature was lowered to 16 • C.

Protein purification
Bacterial cells were pelleted from liquid culture by centrifugation at 4200 × g for 30 min at 4ºC. Cell pellets were resuspended in lysis buffer A500 [20 mM Tris base pH 8.0, 500 mM NaCl, 30 mM imidazole pH 8.0, 10% (vol / vol) glycerol], except for cultures expressing GyrA, which were resuspended in A800 [20 mM Tris base pH 8.0, 800 mM NaCl, 30 mM imidazole pH 8.0, 10% (vol / vol) glycerol], and sonicated using a Vibracell™ VCX500 ultrasonicator with medium tip (Sonics) for a total of 2 min (10 s on / 10 s off).The sonicated sample was centrifuged at 20 000 × g for 1 h at 4 • C to isolate the soluble fraction from cell debris.The protein rich isolated soluble fractions were passed through Ni-NTA His-Trap™ HP 5 mL columns (Cytiva) at 2 ml / min to maximise recombinant protein binding via N-terminal hexahistidine (6His) tags.A 10-column volume (cv) wash step was performed using lysis buffer.From this stage onward, purifications were optimised for each protein, detailed below.Fast protein liquid chromatography (FPLC) steps were carried out using an Åk-ta™ Pure protein chromatography system (Cytiva) at 4 • C.

Anion exchange chromatography
Protein samples were loaded on to a pre-equilibrated HiTrap Q HP anion exchange 5 ml column (Cytiva) in low salt buffer A100 [20 mM Tris base pH 8.0, 100 mM NaCl, 10% (vol / vol) glycerol].This column was then subjected to an increasing salt gradient using the Åkta™ system, titrating in high salt buffer C1000 [20 mM Tris base pH 8.0, 1000 mM NaCl, 10% (vol / vol) glycerol] until a final salt concentration of 600 mM NaCl was achieved.2 ml fractions were collected and analysed by SDS-PAGE.Fractions containing the protein of interest were carried forward for further purification or dialysed into an appropriate buffer for storage.

Siz e-exclusion chromatograph y (SEC)
HiPrep 16 / 60 Sephacryl S-200 and S-300 HR SEC columns (Cytiva) were selected dependent on the column fractionation range and size of the target protein.The column was preequilibrated in sizing column buffer, S500 [50 mM Tris base pH 8.0, 500 mM KCl, 10% (vol / vol) glycerol], prior to a concentrated protein sample being applied via capillary loops at a rate of 0.5 ml / min.Fractionation occurred at 0.5 ml / min and the resulting chromatographic peaks were sampled and analysed by SDS-PAGE.Fractions containing the protein of interest were carried forward for further purification if needed, dialysed into an appropriate buffer, or stored.

Purification and storage of M. tuberculosis GyrA
Once bound to the initial Ni-NTA and washed with 10 cv A800, the column was washed with a further 5 cv A100.The sample was eluted directly on to a pre-equilibrated anion exchange column with 10 cv B100 [20 mM Tris base pH 8.0, 100 mM NaCl, 250 mM imidazole pH 8.0, 10% (vol / vol) glycerol] before washing again in A100 to remove the high imidazole.The anion exchange column was run as above.Fractions were analysed for protein purity by SDS-PAGE, and appropriate fractions were pooled before the addition of 0.4 mg 6His-TEV protease to cleave the 6His-TEV site tag.The sample was rolled at 30 rpm in 4 • C overnight then passed down a second Ni-NTA column (ortho Ni-NTA) to remove the 6His-TEV protease and 6His-TEV site tag.The flowthrough was collected and concentrated in a 10 kDa cut-off centrifugal concentrator (Sartorius) to 2 ml.The 2 ml sample was injected into a 2 ml capillary loop on the Åkta™ Pure system before fractionation by SEC using the Sephacryl S-300 column, as per above.Fractions were analysed for purity by SDS-PAGE, appropriate fractions were pooled and concentrated to > 300 μM before diluting by one third volume with storage buffer [50 mM Tris base pH 8.0, 500 mM KCl, 70% (vol / vol) glycerol] for a final glycerol (cryoprotectant) concentration of 30%, and final protein concentration of > 200 μM.
Appropriate volume aliquots were made and flash cooled in liquid nitrogen before storage at −80 • C.

Purification and storage of M. tuberculosis GyrB
Once bound to the initial Ni-NTA and washed with 10 cv A500, the column was washed with a further 5 cv A100.The sample was eluted directly on to a pre-equilibrated anion exchange column with 10 cv B100 before washing again in A100 to remove the high imidazole.The anion exchange column was run as above.Fractions were analysed for protein purity by SDS-PAGE, and appropriate fractions were pooled before the addition of 0.4 mg 6His-TEV protease to cleave the 6His-TEV site tag.The sample was rolled overnight at 4 • C then passed down a second Ni-NTA column (ortho Ni-NTA) to remove the 6His-TEV protease and 6His-TEV site tag.The flow through was collected and concentrated in a 10 kDa cut-off centrifugal concentrator (Sartorius) to 2 ml.The 2 ml sample was injected into a 2 ml capillary loop on the Åkta™ Pure system before fractionation by SEC using the S-300 column, as above.Fractions were analysed for purity by SDS-PAGE, appropriate fractions were pooled and concentrated to > 300 μM before diluting by one third volume with storage buffer for a final glycerol (cryoprotectant) concentration of 30%, and final protein concentration of > 200 μM.Appropriate volume aliquots were made, and flash cooled in liquid nitrogen before storage at −80 • C.

Purification and storage of M. tuberculosis ParDE1
This process was identical to production of GyrB with the following exceptions: tag cleavage occurred using the hSENP-2 enzyme ( 35 ) to remove the 6His-SUMO tag and SEC was performed using the Sephacryl S-200 column, as per above.Protein was stored in SEC buffer only (10% glycerol) at a concentration of > 100 μM.

Purification and storage of M. tuberculosis ParDE1 as a heterotetramer
This process was identical to production of GyrB with the following exceptions: tag cleavage occurred using the hSENP-2 enzyme to remove the 6His-SUMO tag concentration and SEC via the HiPrep 16 / 60 Sephacryl columns was not performed.Protein was of sufficient purity and concentration after anion exchange for biophysical studies.

Purification and storage of M. tuberculosis ParDE2
Once bound to the initial Ni-NTA and washed with 10 cv A500, the sample was eluted in 5 cv B500 [20 mM Tris base pH 8.0, 500 mM NaCl, 250 mM imidazole pH8.0, 10% (vol / vol) glycerol] and 0.4 mg 6His-hSENP-2 was added to cleave the 6His-SUMO tag.The sample was dialysed into A100 overnight at 4 • C before being passed down a second Ni-NT A column (ortho Ni-NT A) to remove the 6His-hSENP-2 and 6His-SUMO.The flow through was passed directly on to an anion exchange column for fractionation as above.Fractions were analysed for protein purity by SDS-PAGE; routinely ParE2 eluted in an early 'shoulder' peak before the full ParDE2 complex.Appropriately pure ParE2 fractions were not subjected to SEC due to low yields, rather, the sample was pooled and concentrated in a 10 kDa cut-off centrifugal concentrator (Sartorius) to > 100 μM before flash cooling in aliquots for storage at −80 • C.

Gyrase assays
Gyrase assays were performed using published protocols ( 2 ,3 ), adapted where appropriate.The DNA gyrase holoenzyme was reconstituted by incubating equimolar amounts of GyrB and GyrA to a final heterotetramer (GyrB 2 A 2 ) concentration of 10 μM on ice for 5 min.Gyrase fusion proteins were incubated at a final dimer concentration of 10 μM on ice for 5 min.Gyrase enzymes were then serially diluted in twofold steps using gyrase dilution buffer [50 mM Tris base pH 8.0, 2 mM MgO Ac, 1 mM D TT, 500 mM K O Ac, 50 μg / ml BSA, 10% (vol / vol) glycerol], down to the appropriate concentration for assays.
Each cleavage assay using TA components to interrupt gyrase DNA relaxation contained 5 μl of 4 × gyrase reaction buffer and 1 μl of a 250 ng / μl solution of negatively supercoiled pSG483.4 μl of 0.15625 μM gyrase enzyme (obtained by sequential two-fold dilutions of 10 μM stock) was added before incubation on ice for 5 min.2 μl of protein dilution was added, or solvent / buffer where appropriate, before incubation on ice for a further 5 min.Reactions were diluted to 20 μl with 8 μl dH 2 O and incubated at 37 • C for 30 min.Protein additive (TA system components and complexes) dilutions were prepared by two-fold dilution in respective storage buffers to appropriate assay concentrations.
Following incubation, reactions were first quenched with 2 μl of stopping buffer [5% (wt / vol) SDS, 125 mM EDTA], followed by adding 1 μl of 12 mg / ml proteinase K and further incubation at 37 • C for 1 h.Reactions were stored at 4 • C until immediately before gel loading, whereupon a 6 × agarose gel loading dye was added to the samples and the samples were warmed to 37 • C for 5 min.Samples were separated by electrophoresis in 1.4% (wt / vol) TAE agarose gels (containing 0.5 μg / ml EtBr as stated (when appropriate) for 16 h at 45 V. Agarose gels were post-stained in TAE containing 0.5 μg / ml EtBr (when appropriate) and visualised by UV illumination and were imaged on a BioRad ChemiDoc™ XRS + with ImageLab™ software on the EtBr setting (BioRad).Gel images were analysed using ImageJ2 ( 38 ) with background subtracted.For DNA relaxation assays, supercoiled band intensity was measured throughout the titration and converted to percentage of the '0 gyrase lane supercoiled band.Cleavage assay measurements were taken from gels containing EtBr (when possible); supercoiled, linear, and nicked band intensities were calculated per lane.Linear band percentage was subsequently calculated per lane and normalised to the '0' lane linear percentage, per assay.Measurements for the DNA damage induced by thermal remodelling of ParDE1 were performed on gels containing EtBr.Linear and nicked product estimates were calculated as per cleavage assays.To estimate the amount of DNA loss per lane the total band intensity of supercoiled + linear + nicked per lane was compared as a percentage to the band intensity of the control supercoiled (S) lane.The difference in percentage between the experimental lane and control lane is presented as DNA loss.Mean values and standard deviation were calculated from triplicate data (unless otherwise stated in figure legends) for the band of interest.Data were plotted in GraphPad Prism (Version 9.4.1) and presented with connecting line and error bars.

Mass spectrometry
ES-TOF mass spectrometry of protein samples was kindly performed on the Xevo QtoF Premier mass spectrometer (Waters, UK) at our in-house Durham University Chemistry Department facility by Mr Peter Stokes. 100 μl protein samples were supplied at 1 mg / ml in 10 mM ammonium bicarbonate.

Circular dichroism spectroscopy and thermal denaturation
Both circular dichroism (CD) and thermal denaturation were performed in-house on a J-1500 JASCO CD spectrometer.CD was performed at 20 • C pre and post melting to analyse secondary structure of TA complexes.Thermal denaturation was performed between 20 • C and 80 • C with unfolding measured via the CD at 222 nm as a function of temperature.Proteins were analysed in A500 buffer.Spectra and melts were collected in a 1 mm pathlength cuvette with 1 nm data pitch on spectra and a thermal gradient of 1 • C / min.The protein concentrations were 100 μM.Both CD and thermal denaturation curves are plotted in GraphPad Prism (Version 9.4.1) as an XY table, with X as 'Numbers' and Y as a 'single Y value for each point'.Graphs are presented with the connecting line only.Melting temperatures were calculated using the JASCO thermal analysis software.

Analytical SEC
The Superose 6 10 / 300 GL SEC column (Cytiva, discontinued) was selected for its broad fractionation range and short run time, allowing for analysis and purification on the Åk-ta™ pure system (Cytiva).Calibration curves were generated for the Superose 6 10 / 300 GL SEC column using appropriate combinations of commercially available low and high molecular weight kit proteins (Cytiva) for best resolution.The column was equilibrated in buffer S300-A [20 mM Tris base pH 8.0, 300 mM NaCl).For analysis, protein samples were manually loaded into a 100 μl capillary loop in their respective storage buffers at appropriate concentrations to generate a clear UV signal, generally 1 mg / ml was sufficient.Samples were injected onto the column using S300-A buffer at a flowrate of 0.5 ml / min for fractionation across 1.2 cv.Column volume, aka V c in the equation below, was 24 ml.Where appropriate, samples were collected for further analysis in 250 μl fractions.Elution volumes ( V e ) were calculated using the Peaks function in Unicorn™ 7 (Cytiva).Elution volumes ( V e ) were converted into the partitioning coefficient ( K av ) for each sample using the following equation: The molecular weight calibration curve is subsequently plotted as K av versus log 10 ( M r , kDa).The Stokes radius ( R st ) calibration curve is subsequently plotted as log 10 ( R st , Å) versus K av .

Molecular weight ( M r ) and Stokes radius ( R st ) estimation
For estimates of M r and R st , linear regression was performed on the respective plots.The resulting line equations ( y = mx + c ) were used to calculate the observed M r and R st through the following rearrangements: Observed values were then compared to calculated values of M r and R st and presented as a ratio of calculated:observed. M r values were calculated using the online ProtParam tool (Expasy) ( 39 ).R st values were calculated using crystal structures and / or AlphaFold generated models using the HullRad calculator (Fluidic Analytics) ( 40 ).

ParDE1 complex remodelling
ParDE1 expressed and purified as above provides the starting material (theoretical heterotetramer) for remodelling experiments.Once samples were ready for analysis they were subjected to analytical SEC as described above.For initial incubation and buffer alteration experiments, ParDE1 concentration remained at 2.5 mg / ml ( ∼62.5 μM).Incubation at 4 • C was performed in the fridge, while 37 • C and 45 • C incubation was performed in a thermocycler.For concentration dependent studies, ParDE1 was concentrated in a 5 kDa cut-off centrifugal concentrator column (Sartorius) from 2.5 mg / ml and 100 μl sampled at the appropriate concentrations.For the 37 • C time-course, ParDE1 concentration begun at 4 mg / ml (100 μM) to allow for coupling with cleavage assays.100 μl was sampled at each time-point and subjected to analytical SEC.Incubation was controlled in a thermocycler.

Mass photometry
ParDE1 expressed and purified as above provided the starting material (theoretical heterotetramer) for mass photometry experiments.A 5 ml ParDE1 sample at 5.2 mg / ml was incubated at 37 • C with shaking at 180 rpm.At each time point, a 200 μl sample was snap frozen in liquid N 2 .Solution-phase mass determination of the ParDE1 species present in each sample was then performed using the TwoMP (Refeyn) mass photometer.Samples were diluted 1000-fold in A500, and experimental data were obtained in the form of mass photometry videos recorded for one minute using the AcquireMP v2.5 software (Refeyn) on precleaned, high sensitivity microscope slides.A mass calibration was done using bovine serum albumin, IgG, and thyroglobulin.The experimental data were then fit to this calibration, and graphs were generated using the DiscoverMP v2.5 software (Refeyn).

Protein crystallisation
Samples for crystallography were dialysed into buffer X [20 mM Tris base pH 8.0, 150 mM NaCl, 2.5 mM DTT] and concentrated to 12 mg / ml (ParDE1) or 16 mg / ml (ParDE2) for initial trials.Sitting drop crystallisation trials were set-up using an SPT Labtech Mosquito® robot and commercial screens (Molecular Dimensions).Crystal screens were left at 18 • C. ParDE1 required no optimisation, datasets of sufficient quality were collected from needle shaped crystals grown in 0.1 M Bis Tris Propane pH 7.5, 20% PEG 3350, 0.2 M NaNO 3 and harvested directly from the crystal screen.Following an increase of starting concentration to 20 mg / ml, the best ParDE2 crystals (large hexagonal planar) grew after 3 months in 0.1 M MES pH 6.2, 15% wt / vol PEG 3350.For harvesting, mother liquor from the condition and 100% glycerol were mixed in a ratio of 1:1 and an equal volume of this mixture was added to the sitting drop, prior to looping and flash cooling in liquid nitrogen followed by storage in a puck for transport.

X-ray data collection and processing
Data collection was performed at Diamond Light Source, Oxford, UK, via remote access on i04.Initial data processing was automated by Diamond Light Source iSpyB using the X-ray image integration programs Xia2 and Xia2-DIALS ( 41 ).Image integration and space group selection were carried out manually using the same programs as well as Mosflm ( 42 ).
For ParDE2, six, 360 • , datasets were collected from two native ParDE2 crystals and merged using iSpyB (Diamond Light Source).For ParDE1, single datasets were collected from three native crystals and merged.Diffraction data were processed with XDS ( 43 ,44 ), and then AIMLESS from CCP4 ( 45 ) was used to corroborate the spacegroups.The crystal structure of ParDE2 was solved by molecular replacement using PHASER ( 46 ) and the M. tuberculosis H37Rv ParE2 AlphaFold structure prediction ( 47 ,48 ) as the search model.The crystal structure of ParDE1 was solved by molecular replacement using the starting model 3KXE ( 27 ) split into individual protomers ParD1 and ParE1 and input as individual assemblies.The solved starting models were built in REF-MAC ( 49 ) and BUCCANEER ( 50 ).Initially, ParD2 could not be placed by PHASER.The ParE2 AlphaFold search model was edited to remove the C-terminal 12 amino acids to allow for subsequent manual building of the ParD2 chain in Coot ( 51 ).The models were then iteratively refined and built using PHENIX ( 52 ) and Coot, respectively.The quality of the final models was assessed using Coot and the wwPDB validation server ( 53 ).PyMol (Schrödinger) was used to perform sequence ('align' command) and structure-based ('super' command) alignments, and generate figures.

Generation of AlphaFold multimer models
Protein structure predictions for the monomers of the TA system proteins are readily available in the published AlphaFold database, accessible online ( 48 ).For multimer models, the Google Colaboratory (ColabFold) ( 54 ) was used.This allowed for the input of multiple protein sequences and the subsequent automated generation of multimer models.Protein sequences for M. tuberculosis TA system components were sourced from Mycobrowser ( 55 ).The highest scoring models from these structure predictions are presented in this study.

Differential toxicity and autoregulation of ParDE systems from Mycobacterium tuberculosis
Mycobacterium tuberculosis H37Rv encodes two parDE loci (Figures 1 A and B).We began the study by examining both the toxicity of the ParDE systems and their capacity for transcriptional autoregulation, using M. smegmatis mc 2 155 (Figure 1 ).When expressing ParE1 or ParE2 toxins, only ParE2 was toxic in M. smegmatis (Figures 1 A and B).Co-expression with antitoxin ParD2 restored growth (Figure 1 B).Having cloned 1000 bp of upstream sequence from each parDE locus into the promoterless lacZ reporter plasmid pJEM15 ( 36 ), we observed that both promoters were active in M. smegmatis (Figures 1 C  and D).We then found that co-expression of both ParD1 and ParE1 caused negative autoregulation of transcription, but not with either component alone (Figure 1 C).No autoregulation was demonstrated under these conditions for ParDE2 (Figure 1 D).

ParE toxins from Mycobacterium tuberculosis enhance gyrase-mediated DNA linearisation
As ParE toxins target DNA gyrase, we aimed to verify the activity of the M. tuberculosis toxins in gyrase assays.Having first purified independent GyrB and GyrA subunits of M. tuberculosis DNA gyrase, gyrase activity was confirmed by reconstituting the GyrB 2 A 2 holoenzyme in vitro and testing for supercoil relaxation ( Supplementary Figures S1 A and B).M. tuberculosis gyrase successfully converted > 90% of the supercoiled pSG483 substrate into multiple topoisomers at a concentration of 31.25 nM, followed by decreased activity at saturating gyrase concentrations ( Supplementary Figures S1 A  and B).The observed activity was comparable to previously demonstrated activity for M. tuberculosis gyrase ( 56 ).
Due to the toxicity of ParE1 in E. coli , genes parE1 ( rv1959c ) and parD1 ( rv1960c ) were cloned into a pET-Duet vector ( 57 ) for expression and purification of the ParDE1 protein complex ( Supplementary Figure S2 ).Though we could not separate ParE1 and ParD1 during purification, the final ParDE1 complex did, however, form two peaks during size exclusion chromatography (SEC) ( Supplementary Figure S2 B).The final ParDE1 sample had high purity, as shown by SDS-PA GE and ES + -T OF mass spectrometry ( Supplementary Figures S2 C and D).
In the absence of free ParE1 toxin for assays, the ParDE1 complex was nevertheless tested in a gyrase DNA relaxation assay (Figures 2 A and B).At the highest concentration tested (10 μM), ParDE1 caused a small amount ( ∼4%) of linearisation (Figures 2 A and B).This suggested that over the course of the assay a small amount of the ParE1 toxin had potentially been released, and could trap gyrase complexes.
Genes parD2 ( rv2142A ) and parE2 ( rv2142c ) were then sequentially cloned into a pET-Duet vector ( 57 ) allowing for co-expression and purification of the ParDE2 protein complex, or ParD2 alone ( Supplementary Figure S3 A).During purification of ParDE2 it was found to also be possible to separate and purify some ParE2 toxin away from the complex ( Supplementary Figure S3 B).During anion exchange of the cleaved ParDE2 complex, ParE2 was isolated in a distinct chromatographic peak.SDS-PAGE analysis showed the purified 12.18 kDa ParE2 toxin alongside the ParDE2 complex ( Supplementary Figure S3 C).ES + -TOF mass spectrometry of the purified ParE2 sample showed there was no detectable ParD2 in the sample ( Supplementary Figure S3 D).
The purified ParDE2 samples were then tested against DNA gyrase (Figures 2 C-F).ParE2 was tested first, using an M. tuberculosis gyrase DNA relaxation reaction (Figure 2 C).At the highest concentration tested, ParE2 caused linearisation of ∼50% of substrate pSG483 (Figures 2 C and F), as normalised against a toxin-only control ( Supplementary Figure S1 C).In contrast, purified ParD2 generated no significant increase in linear pSG483 in the gyrase DNA relaxation reaction (Figures 2 D and F).The purified ParDE2 complex was also tested in the gyrase DNA relaxation reaction (Figures 2 E and F).The ParDE2 complex, like ParDE1, caused ∼5-7% linearisation at the highest concentrations of 5 and 10 μM ParDE2, respectively (Figures 2 E and F).Collectively, these results ParDE2 was our first structural target.Antitoxin ParD2 and toxin ParE2 were co-expressed and purified as described earlier and put into crystallisation trials.After three months of growth, the resulting crystals were used to collect X-ray diffraction data.The phase problem was solved by molecular replacement using a truncated ParE2 (amino acids T2-G86) AlphaFold model ( 47) as a search model, and ParD2 (amino acids I36-G71) was then built into the remaining electron density.The final model was refined to 2.35 Å (Figure 3 A, Table 1 ).
A RelE / ParE superfamily alignment ( 25 ) allowed us to plot and visualise the conserved hydrophobic residues within the ParE2 toxin structure, which are clearly concentrated on the internally facing residues of the α helical hairpin and throughout the β sheet core of the toxin (Figure 3 B).This positioning creates two major hydrophobic grooves on the surface of the ParE2 toxin which run along the β sheet core and through to the underside of the toxin between the hairpin helices (Fig- ures 3 B and C).Representing the ParD2 antitoxin alongside a surface rendered ParE2 toxin clearly demonstrates the specific interaction at the conserved hydrophobic patches, as they align closely to the truncated region of ParD2 (Figure 3 C).This mimics the conserved mechanism of protein recognition identified in the C. crescentus ParDE structure ( 27 ).PISA analysis ( 58 ) of the ParDE2 complex highlights several polar and ionic interactions that stabilise the largely hydrophobic interfacing.Notable contacts are the ionic bonds formed between ParD2 E45 (on α1) and ParE2 K57 (found on the loop region between α2 and β2) ( Supplementary Figure S4 B) and between ParD2 R47 (on α1) and ParE2 (D14) (on α1, also) ( Supplementary Figures S4 C and D).These are highly specific side chain interactions that demonstrate the mechanism of ParDE2 interaction extends beyond a conserved hydrophobic groove.A truncated AlphaFold ParE2 model was used for molecular replacement.In a full-length AlphaFold model of ParE2 the C-terminal P92 -G101 region forms additional helix α3, which occupies the interface for ParD2 binding that is bound by ParD2 α1 in the crystal structure (Figure 3 D).Phasing the ParDE2 dataset is successful when the ParE2 C-terminal helix is included in the search model, but the linker region between β4 and the additional α3 is not resolved.In contrast, when the C-terminal helix of ParE2 is removed from the search model, unmodelled electron density is present in its place, and can be more successfully built as residues I36-N49 of ParD2.This demonstrated that ParD2 α1 displaced the predicted ParE2 α4, which likely became disordered as is seen for other ParDE system complexes.
AlphaFold multimer ( 59 ) was then used to generate a series of models of ParDE2 complexes.It was expected that, given the crystal structure data indicating the presence of the ParD2 α1 helix across the β sheet hydrophobic region, AlphaFold would generate a multimer model demonstrating preference for this interaction over the ParE2 α4 helix.This was not the case when the full ParE2 sequence was entered alongside the full ParD2 sequence; the full ParE2 model was generated with the additional C-terminal helix and the corresponding ParD2 helix displaced (Figure 3 E).In this predicted model, the ParD2-ParE2 heterodimers are seen interacting through the bundled helices of ParD2 at loop regions.Based on our structure this model does not appear biologically relevant.We therefore investigated whether truncating the ParE2 sequence as input for AlphaFold multimer would present alternative solutions.Using the sequence for ParE2 88-105 , with the unresolved C-terminal residues M88 -E105 deleted, AlphaFold generated two alternative complex models with a more biological quaternary structure (Figures 3 F and G).Both these models have the ParD2 antitoxin occupying both hydropho-  S3 A and B, Supplementary Figure S4 G) to determine the most likely model for ParDE2 between Models 2 and 3 (Figures 3 F and G).The R st for Model 2 (Figure 3 F) was calculated to be 26.10Å, and the R st for Model 3 (Figures 3 G) was calculated to be 30.50Å ( Supplementary Figure S4 G).Model 3 provided a closer value to the observed R st calculated from analytical SEC data (30.27Å), indicating that this is our best model for the ParDE2 heterotetrameric complex.
Finally, given the time taken to crystallise, and that there is no remaining space in the crystal due to packing, it was considered likely that ParDE2 underwent limited proteolysis during crystallisation.This was confirmed by dissolving ParDE2 crystals and analysing by SDS-PAGE ( Supplementary Figure S4 H).

ParDE1 forms a heterohexameric complex
Having determined the structure of the ParDE2 complex, we moved on to ParDE1.Antitoxin ParD1 and toxin ParE1 were co-expressed and purified as described earlier ( Supplementary Figure S2 ).Following trials and optimisation, ParDE1 crystals grew as needles.Single datasets were collected from 3 crystals and subsequently merged to a final resolution of 2.10 Å.The ParDE1 complex structure was determined by molecular replacement using the C. crescentus ParDE structure, PDB: 3KXE, ( 27) as a search model (Figure 4 A, Table 1 ).Interestingly, whilst phasing was successful, on viewing the unrefined density it was clear that a significant portion of the structure remained unmodeled.As a result, an additional two ParD1 chains were built into the model.
The heterohexameric structure somewhat resembles the search model, with the ParD1 antitoxins interacting via an anti-parallel N-terminal β sheet (though noting only one pair of the antiparallel β sheets was resolved), and a pair of ParE1 toxins positioned inside a cage-like structure (Figure 4 A).In the C. crescentus ParDE structure the ParD antitoxins interact through an additional coiled-coil between corresponding ParD1 α2 helices ( 27 ).In contrast, a coiled-coil structural motif is not seen between the primary ParD1 chains in the M. tuberculosis ParDE1 structure, rather, the two primary chains appear to lean against one another.A coiled-coil interaction is present, however, between helix α2 of the primary ParD1 chains and helix α2 of the auxiliary ParD1 chains, forming alongside the anti-parallel β sheet that is itself part of a ribbon-helix-helix (RHH) DNA binding motif (Figure 4 A).Dali searches ( 64 ) using the ParD1 tetramer indicated structural similarity to transcriptional regulator CopG, localised to the dimerised N-terminus and RHH region, and aligning with RMSD values of < 2.0 Å (PDB: 6IYA ( 65 ); 1EA4 ( 66 )).This indicates that ParD1 belongs to the CopG / Arc / MetJ family ( 67 ).Surface electrostatics also show a large electropositive patch created at the antitoxin complexing region (Figure 4 B).Using structures of S. agalactiae CopG ( 66 ,67 ), we can model the ParD1 tetramer interaction with DNA whereby the antiparallel beta sheets insert into the major groove of bent DNA ( Supplementary Figure S5 B).This proposed model accounts for how ParDE1 can perform negative transcriptional autoregulation (Figure 1 C).
PISA analysis ( 58 ) of the ParDE1 heterohexamer identified 4 key interfaces ( Supplementary Figure S6 ).The essential interfaces are at the antitoxin antiparallel β sheet and the primary ParD1:ParE1 interface, suggesting these might form more readily before higher order assembly ( Supplementary Figure S6 ).Again using the RelE / ParE superfamily multiple sequence alignment ( 25 ) we noted the positions of conserved residues on the ParE1 surface (Figure 4  as binding sites for a single ParD1 primary chain; one site along the groove created by the antiparallel β strands ( 1-3 ), and the other site on the underside of the toxin structure between the hairpin of α helices 1 and 2 (Figure 4 D), as also picked out by PISA ( Supplementary Figure S6 ).ParD1 antitoxin binds to these patches similarly to how ParD2 was observed binding ParE2 (Figure 3 C), though it is notable that ParE1 does not encode the proposed C-terminal helix in ParE2 that is displaced by ParD2.ConSurf analysis shows complementary localisation of highly conserved residues between ParD1 and ParE1 ( Supplementary Figure S7 ).The absence of a C-terminal helix in ParE1 is supported by the AlphaFold model, which aligned to the ParE1 crystal structure with an RMSD of 0.594 Å (Figure 4 E).We used Al-phaFold multimer ( 59 ) to assess how stoichiometry might influence the overall quaternary structure.Looking more closely at data collected during ParDE1 purification strongly indicated the presence of two complexes in solution, the determined 4:2 complex structure, and a slightly smaller complex ( Supplementary Figure S2 B).This smaller peak might correspond to a more canonical 2:2 ParDE complex.Interestingly, as per our prediction, a 2:2 heterotetrameric complex could be generated by AlphaFold, creating a quaternary structure as per the search model used in MR (PDB: 3KXE) (Figure 4 F).AlphaFold also successfully generated a 4:2 complex as per the crystal structure (Figure 4 G) with the complexed regions of the antitoxins displaced off-centre creating the tetrameric CopG RHH.Interestingly, the AlphaFold 4:2 complex model presents the C-terminal ParD1 E55 -R83 regions as loosely structured in all 4 ParD1 monomers, perhaps due to perceived direct competition between what we have identified as the 'primary' and 'auxiliary' for the ParE1 surfaces (Figure 4 G).Each of the C-terminal regions track along the ParE1 interface and form a weak helical structure resembling α3, however due to this direct competition neither is presented as fully folded.This indicates error in this AlphaFold solution, as the crystal structure shows only the primary ParD1 monomers form the interface.

ParDE1 exists in a dynamic equilibrium between complexes
The obtained ParDE1 structure, purification data, and supporting AlphaFold results (Figure 4 , Supplementary Figure S2 B) suggested ParDE1 can form multiple quaternary complexes in solution.We noted that analytical SEC of two independent ParDE1 purifications produced distinct, but overlapping, UV traces containing two peaks (Figure 5 A).Both peaks contained high purity protein of the appropriate sizes for ParD1 and ParE2.Conversion of the respective V e to M r indicated a complex of 84.22 kDa to be dominant in sample ParDE1 ( 1 ) (Figure 5 A, red trace, 'peak 1') and a complex of 43.77 kDa to be dominant in sample ParDE1 ( 2 ) (Figure 5 A, black trace, 'peak 2 ).Fractions corresponding to each of the noted peaks (Figure 5 A) were then pooled, and each peak was analysed by SEC using a Superose 6 10 / 300 GL column (Figure 5 B).Both peaks maintained different sizes and did not appear to re-distribute to multiple complexes under these conditions, and so represented purified ParDE1 complexes of distinct stoichiometries.
Separation of these two peaks allowed for thermal stability and circular dichroism analysis of the respective protein complexes (Figure 5 C-E).Thermal denaturation analysis (melting) ( 68 ) showed that both samples had high thermal stability with melting temperatures of around 65 • C (Figure 5 C).The melting curves do not show complete unfolding as the CD signal ceases to change significantly above around 70 • C.This is indicative of aggregation occurring as the samples unfold.The two samples show differences both in melting temperature, with the Tm estimated from the point of inflection in the melt curve being a few degrees lower for the peak 2 (black curve) than for the peak 1 (red curve), and with the loss of structure during aggregation being more significant for the peak 2 sample (Figure 5 E) than for the peak 1 sample (Figure 5 D).These differences further indicate that two different ParDE1 complex species are present, one from each peak.
To determine likely solution states for the complexes in each SEC peak, V e values for both peaks in the same trace (Figure 5 A) were used to estimate the R st for the dominant ParDE1 species (Figure 5 F).The observed R st for peak 1 was 34.51 Å, and the observed R st for peak 2 was 27.82 Å (Figure 5 F).Using the earlier AlphaFold model (Figure 4 F) supports peak 2 containing ParDE1 heterotetramer complexes, as the calculated R st value of 29.50 Å agrees well with the observed R st value of 27.82 Å (Figure 5 F).Having identified peak 2, we proposed that peak 1 contained the heterohexameric complex we observed in our X-ray structure.Using our structure, the calculated R st was 31.7 Å, which matched well with our observed R st for peak 1 (Figure 5 F).This supports peak 1 containing ParDE1 heterohexamer complexes.Taken together, these data strongly support a model wherein ParDE1 can exist in multiple stoichiometries in solution.

Conditional remodelling of ParDE1 complexes
Interestingly, the independent purifications of ParDE1 show different relative intensities between the two complex species present (Figure 5 A).This indicates a potential equilibrium between the two species and that one may become the other.We chose to explore the conditions that might impact this equilibrium.
Having developed a method to isolate the ParDE1 heterotetramer (see Materials and Methods), the sample was subjected to a range of conditions (Figure 6 , Supplementary Figure S8 ) to explore whether ParD1 2 ParE1 2 can be remodelled into the predicted ParD1 4 ParE1 2 , as signified by the emergence of 'peak 1 from 'peak 2 during analytical SEC.Should this occur, we hypothesised that the ParE1 toxin might become free in solution.High-yield purification of ParDE1 ( > 50 ml at 2.5 mg / ml) in the heterotetramer stoichiometry (Figure 6 A, black curve / 0 h) allowed for the exploration of conditions that may influence protein complex states, notably ParDE1 concentration, temperature, pH, and salt concentration.Following incubation at 4 • C for 48 h, the chromatographic trace does indeed change and the heterohexameric ParDE1 complex 'peak 1 grows from the original heterotetrameric 'peak 2 observed at 0 h (Figure 6 A).
The starting material, at 2.5 mg / ml (or 0 h) was then concentrated and sequentially analysed via SEC (Figure 6 B).The 48 h 'mixed' ParDE1 sample (Figure 5 A) is presented alongside the increasing concentrations to help identify the respective 'peaks'.It is clear to see that concentrating ParDE1, even as high as 45 mg / ml, does not have a large effect on the predicted stoichiometries; as, at each concentration, the dominant species remains aligned with the heterotetrameric (peak 2) starting material at 2.5 mg / ml.There is a minor shift at the higher concentrations with the appearance of a 'shoulder' that aligns with the heterohexameric 'peak 1'.Considering the appearance of peak 2 over time (Figure 6 A), it is possible that the shoulder is an artefact of the experiment as concentration of the sample and the sequential analytical SEC analysis took over 6 h.
As the heterohexamer appears to be around 5 • C more thermostable than the heterotetramer via melt analysis (Figure 5 C), it was hypothesised that complex remodelling may be thermodynamically driven.Incubating ParDE1 heterotetramer at a starting concentration of 2.5 mg / ml at 37 • C results in a more rapid evolution of the heterohexameric peak 1 (Figure 6 C, 16 h).Not only is the emergence of the peak apparently more rapid than in the 4 • C incubation over 48hours, it is also more dominant in terms of the respective   1 R st for the crystal str uct ure solution of P arD1 4 P arE1 2 ; 2 R st for the AlphaFold solution of P arD1 2 P arE1 2 .R st values were generated using HullRad (Fleming and Fleming, 2018).Chromatograms are representative of duplicate data and are normalised between 0 and 1 for presentation and comparison.Graphs are cropped to the appropriate scale (10-22.5 ml).
intensities of the two peaks, with the heterohexamer being the dominant species in 37 • C incubation.At the 48-hour timepoint, not only has the entire sample shifted from the heterotetramer to the heterohexamer, but a shoulder appeared on the left of the heterohexamer, indicating that the entire fraction has been remodelled into at least heterohexamer and perhaps even higher order complexes (Figure 6 C, 48 h).Crucially, an additional small peak was observed just after 20 ml elution, most evident on the 16-h curve, which could be formed by free ParE1 toxin (Figure 6 C, Peak 3), as previously hypothesised.A higher temperature of 45 • C was then selected as a further test condition, which sped up the complex remodelling futher ( Supplementary Figure S8 A).Again, a potential ParE1 peak was obtained ( Supplementary Figure S8 A, Peak 3).Next, we briefly explored the effects of reducing agent, acidic pH, and high salt on remodelling, as these conditions can all effect protein complex states and are environmental conditions to which TA systems may be responsive ( 17 ).None of these conditions caused any noticeable shift in the positioning of the starting peak ( Supplementary Figure S8 B).
Having established the positions of three chromatographic peaks of interest (peak 1 -heterohexamer; peak 2 -heterotetramer; peak 3 -ParE1), fractionation and SDS-PAGE analysis was used to examine potential purification of the respective species ( Supplementary Figure S8 C).Both proteins, ParE1 (11.17 kDa), and ParD1 (9.21 kDa), were present in each fraction ( Supplementary Figure S8 C).Peak 3 overlays with the tail of the heterotetramer peak, which accounts for the presence of ParD2 in this region of the chromatogram.Nevertheless, the AlphaFold-predicted ParE1 structure produced a calculated R st of 16.40 Å.The calculated R st of peak 3 was 15.82 Å, providing an observed / calculated ratio of 0.96 ( Supplementary Figure S8 D), indicating that the peak produced during the hypothesised ParDE1 remodelling process could theoretically be free monomeric ParE1.However, this did not appear to be a suitable method for purification of the desired ParE1 toxin due to the very small quantities of free ParE1 protein produced.We attempted to release and purify ParE1 on a larger scale via incubation at higher temperature.A ParDE1 sample was concentrated to 10 mg / ml and incubated at 37 • C for 16 h, prior to SEC.Unfortunately, post-incubation a high amount of precipitate was present.Following centrifugation, the supernatant was analysed by SEC and the chromatographic trace was aligned to previous ParDE1 purification, indicating the sample had re-modelled to heterohexamer but no ParE1 peak was obtained ( Supplementary Figure S8 E).The major peak fraction was analysed alongside the precipitate fraction from overnight incubation ( Supplementary Figure S8 F).This demonstrated that SEC produced purified ParDE1 complex ( Supplementary Figure S8 F, ParDE1 lane), whereas the liberated ParE1 toxin after 37 • C incubation appears to have precipitated ( Supplementary Figure S8 F, ParE1 lane).Though unfortunate, the appearance of ParE1 precipitate supports our hypothesis for ParE1 release as a result of complex remodelling.
Next, we performed mass photometry analysis of ParDE1 samples incubated at 37 • C to provide an additional biophysical demonstration of complex remodelling (Figure 6 D).Within two hours there was a substantial shift from heterotetramers to majority heterohexamers (Figure 6 D).At later time points (24 and 48 h) we observed a population of larger molecules that may represent higher order ParDE1 complex structures (Figure 6 D).These would correspond with the additional shoulder observed at 48 hr incubation by SEC (Figure 6 C).These two pieces of data are thought unlikely to represent aggregates, as aggregates would appear in the void volume of SEC experiments; and, by mass photometry, this species forms a monodisperse distribution of albeit larger, but still relatively small, molecules.Our mass photometry analysis supports the SEC, X-ray crystallographic and circular dichroism data demonstrating remodelling of ParDE1 complexes.

Thermally driven ParE1 toxin release induces DNA cleavage
Despite our efforts, attempts to demonstrate ParE1 release during remodelling had so far failed, likely due to an inability to capture the small amounts available protein that are briefly in solution before precipitation occurs.Noting that the ParDE1 complex we initially isolated showed a higher than basal level of activity against gyrase (Figures 1 A and B), and putting this in context with our biophysical data on remodelling, we concluded that the gyrase poisoning was likely a result of ParE1 toxin liberated by ParDE1 complex remodelling during the assay.To confirm and improve upon this result, we decided to repeat the biochemical analysis of ParDE1 in gyrase DNA relaxation assays, starting with a pure ParDE1 heterotetramer sample.Using the biochemical assay as a readout would allow observation of the small amounts of soluble ParE1 released by remodelling, without concern for low ParE1 solubility at higher concentrations.
A gyrase DNA cleavage assay was performed following preincubation of ParDE1 at 37 • C to promote toxin release, and remodelling was concomitantly monitored by analytical SEC (Figure 7 ).ParDE1 analytical SEC was performed on the hour at the presented time points (Figure 7 A).The starting, 0 hr, sample was a single peak positioned at the appropriate V e ( ∼17.3 ml) for the heterotetrameric ParDE1 peak 2 complex, as expected (Figure 7 A).Incubation at 37 • C caused a gradual shift in the positioning towards peak 1 until the entire peak elutes at the V e of the heterohexamer ( ∼16.1 ml) (Figure 7 A).It is worth noting that the 3 hr, and final 20 h time-points were not tested in the corresponding biochemical assay due to practical time constraints during the assay, however, the bulk of the fraction is remodelled by 12 h (Figure 7 A).Interestingly, at the 1 h time-point a clear peak was again fleetingly observed at the appropriate V e (peak 3) for the ParE1 toxin (Figure 7 A, thick black arrow).
At the indicated time-points, incubated ParDE1 was used to perform a gyrase DNA relaxation assay, to a final concentration of 10 μM ParDE1 as per the standard protocol for all presented relaxation assays (Figure 7 B).A new relaxation reaction was set-up for each time-point using the pre-diluted stock of GyrB 2 A 2 .Due to protocol, the reaction itself provided an extra 30 minutes of incubation at 37 • C, alongside a different buffer environment.The relaxation reaction is clearly inhibited at the 0-hr pre-incubation point with the notable presence of linear species DNA (Figure 7 B, 0.5 h total incubation).This result is similar to that first previously observed (Fig- ures 1 A and B), but in this experiment there was an increased level of linearisation, perhaps due to using homogeneous heterotetrameric ParDE1 as starting material.The relaxation reaction becomes almost fully inhibited over the pre-incubation time-course as linearisation also increases, whilst the level of nicking decreases (Figure 7 B).The area below the supercoiled band is presented to also demonstrate the increasing levels of non-specific DNA cleavage (evidenced by the smearing pattern within the lane) (Figure 7 B, + / -EtBr).This demonstrates a clear reduction in the total band intensity at increased incubation times (Figure 7 B).Together, we observe increasing linearisation and increasing non-specific DNA cleavage leading to reduction in distinct species over the time-course.This correlates with the movement of the chromatographic 'peak 2' from the right-hand side starting point for ParDE1 heterotetramer (Figure 7 A, black trace), toward the final 'peak 1 (Figure 7 A, red trace), representing remodelled heterohex-amer.Importantly, these effects are independent of the increasing age of the initial gyrase stock that was used throughout the assays, as this stock was tested for relaxation activity in the absence of ParDE1 at the final 12-h time-point (Figure 7 B).Gyrase remained stable and active throughout the experiment; thus, these results are solely due to the addition of ParDE1.
Based on these collected data we present a model for thermally driven in vitro conditional remodelling of ParDE1 complexes, leading to release of ParE1 toxins (Figure 8 , Supplementary Figure S9 ).Remodelling allows conversion of two ParDE1 heterotetramer complexes into a single heterohexameric complex, and for every heterohexamer produced two ParE1 toxin molecules are liberated.

Discussion
Our structures of both the ParDE2 complex (Figure 3 ) and ParDE1 complex (Figure 4 ) highlight key similarities and differences between the toxin and antitoxin protomers, alongside the quaternary complexes.The superfamily structure βααβββα ( 25 ) is largely present in both toxins.Sequencebased alignment of ParE2 and ParE1 returned an RMSD of 6.175 Å indicating low level sequence similarity, however, sequence-independent superposition returns a greatly improved RMSD value of 2.163 Å.This indicates that structure is more greatly conserved than sequence amongst ParE toxins.A notable difference between the toxin structures is evident at the C-termini (Figures 3 and 4 ).Unlike for ParE1, the C-terminal residues of ParE2 are predicted to form an αhelix (Figure 3 D) and occupy the superfamily's conserved hydrophobic surface across the β-sheet core.This surface is instead occupied in the crystal by ParD2 (and corresponds to the site occupied by ParD1 in the ParDE1 structure) (Figures 3 C and D).The importance of these C-terminal residues for toxicity of ParE2 has previously been demonstrated.Removing E95 -E105 (the C-terminal 10 amino acids), or making mutants E98A or R102A, renders the toxin ineffective ( 28 ).When considering the RelE / ParE superfamily ( 25 ), reorganisation of this helix to be positioned across the β-sheet core is also important in RelE toxins for the positioning of essential catalytic residues ( 61 ) and thus, ribonuclease activity when bound to the ribosome.ParE2 does not possess the canonical RelE catalytic core residues, therefore, the significance of the C-terminal helix requires further investigation.
The full ParDE2 complex structure is yet to be fully elucidated as the N-terminal dimerisation domain of ParD2 was cleaved during crystallisation ( Supplementary Figure S4 H).Our current model for the ParDE2 system is the ParD2 2 ParE2 2 heterotetramer presented in Figure 3 G.This model is supported by good alignment to the crystal structure (Figure 3 G) and analytical SEC data ( Supplementary Figure S4 G).Interestingly, the heterotetramer model includes a ParD2 N-terminal dimerisation domain with structural similarity to the dimerisation domain of the Lactococcus phage TP901-1 Clear 1 repressor (PDB: 6FXA ( 69 )), indicating DNA-binding capability .Interestingly , whilst we did not observe autoregulation for ParDE2 using the 1000 bp upstream, a previous report did show autoregulation, using only 363 bp of upstream sequence ( 28 ).Models for ParDE2 need to be developed further as this system is peculiar in its structure, especially in the ParD2 antitoxin and how it competes with the ParE2 C-terminal helix.Further to this, an open reading frame is identifiable upstream of the ParDE2 operon, the translated product of which shares 27% sequence similarity with ParD1 (though shorter at 51 amino acids vs ParD2 at 71 amino acids).It is plausible that this is the third member of a tripartite style ParDE system, similar to that seen for ParD2-PaaA2-ParE2 ( 70 ,71 ).As for ParDE2, the structure of ParDE1 was also determined in an unexpected stoichiometry, forming a ParD1 4 ParE1 2 heterohexamer (Figure 4 ).Analyses of the ParDE1 complex highlighted that the ParD1-ParE1 interaction is highly specific, not only interacting via the conserved superfamily interface as expected (Figure 4 ), but also via several tuned interactions in regions of lower sequence conservation ( Supplementary Figure S7 ).ParD1 forms a dimeric CopG RHH motif through its N-terminal region (Figure 4 ).Investigating the structure of the CopG domain, of which there is one fully resolved in the structure, indicates a DNAbinding role as seen in the FitAB system ( 72 ), and suggests a likely model to explain the autoregulation we observed for ParDE1 (Figure 1 C).Interestingly, CopG domains appear to permit interactions with elongated operators within promoter regions; the FitAB system employs two CopG domain that interact with operator sites ∼15 bp apart (PDB: 2BSQ ( 72)).In the V. cholerae ParDE system, three CopG domains exist back-to-back and create three DNA-binding sites for proposed enhanced DNA-binding through operator site interactions (PDB: 7R5A ( 73)).Further manual searches of the ParDE structures in the PDB indicated that ParDE complexes have increased plasticity in their stoichiometries; while 3KXE ( 27 ), 6X0A ( 74 ), and 6XRW ( 75 ) all exist as 2:2 heterotetramers, 5CEG (ParD 4 :ParE 4 stoichiometry) ( 76 ) and 7R5A (ParD 6 :ParE 2 stoichiometry) ( 73 ) exist as heterooctamers.Even more noteworthy is that the 7R5A structure resembles the ParDE1 complex with both fully and partially resolved ParD chains forming RHH CopG N-terminal dimers, however, an additional partially resolved ParD1 dimer places itself in-between the full-length ParD chains.Further to this, the 7B22 ( 73 ) structure of only the V. cholerae ParD chains forms a hetero-16mer (8 dimers) in a ring-like structure.Altogether, these results indicate that the ParD N-terminal domain permits higher-order stoichiometries to form.These related structures support our conclusions drawn from the observation of higher order complexes by SEC (Figure 6 C) and mass photometry (Figure 6 D).
The ParDE1 heterohexamer structure also supports our proposed model of in vitro ParDE1 complex remodelling for toxin release (Figure 8 ), based on the ParDE1 analytical SEC experiments, mass photometry, circular dichroism, and biochemistry (Figures 5 -7 ).AlphaFold was used to successfully generate a heterotetramer model, which we predict to be the initial complex state for ParDE1 (Figure 4 F).Interface analysis of the resulting heterohexamer complex seen in the crystal indicates that four interfaces are relevant in the formation of this higher order complex ( Supplementary Figure S6 ).Through comparison to the search model (PDB: 3KXE ( 27)) and the ParDE1 heterotetramer AlphaFold prediction, alongside the PISA analysis, we suggest the heterohexamer structure be considered as a dimer of ParD1 2 ParE1 trimeric structures that interact mainly through their ParD1 CopG domains.This is supported by the structures of V. cholerae ParDE (PDB: 7R5A ( 73 )) and V. cholerae ParD (PDB: 7B22 ( 73 )) whereby the highly similar CopG domain multimerises in the same orientation seen for ParDE1.This observation has allowed us to develop our model for ParDE1 complex re-modelling (Figure 8 ), to indicate likely swapping of ParE1 toxins ( Supplementary Figure S9 ).We propose that during ParDE1 complex remodelling, the ParD1 CopG domain remains intact as a highly stable and conserved dimerisation domain.We propose that the CopG domains from two independent complexes interact through central ParD1 chains (blue and teal) ( Supplementary Figure S9 A and B).Displacement occurs at the relatively weak polar ParE1 -ParE1 interface ( Supplementary Figure S6 , interface iv), and due to steric clashes, the toxin pairs are reorganised.Why this process is driven by increasing temperature (Figures 6 and 7 ) is yet to be fully understood.Additionally, further work needs to be performed on elucidating the higher-order species of ParDE1 that evolves in the later stages of sample incubation (Figures 6 C and 6 D).At this moment, we do not know whether the ParDE2 complex would behave in a similar manner, remodelling in response to temperature.This is unlikely as purification and analysis via SEC routinely resulted in a single peak.It is more likely that ParDE2 exists in the typical 2:2 stoichiometry seen for ParDE and RelBE superfamily TA systems.
Though we have observed in vitro conditional remodelling of ParDE1 causing release of ParE1 toxin and inhibition of gyrase, it remains to be demonstrated whether this process has physiological relevance.This could be tested, for instance, by mutating residues at vital complex interfaces ( Supplementary Figure S7 ) and performing in vivo analyses of both activation and toxicity.Current models for toxin activation rely either on antitoxin degradation by proteases, or increased transcription of the TA locus.If future work were to confirm in vivo application of conditional remodelling, this would suggest there should be room to consider alternate post-translational mechanisms of toxin release in cells.Notably, toxin release via degradation of the antitoxin has recently been disputed ( 77 ).Collectively, these findings advance our understanding of the type II TA complement of M. tuberculosis , examining systems that have been implicated in several roles contributing to the virulence and adaptation of M. tuberculosis during infection ( 30-31 ,78 ).ParE toxins remain of interest as potentially potent gyrase inhibitors worthy of further study.

Figure 1 .
Figure 1.Toxicity and transcriptional autoregulation of M. tuberculosis ParDE systems.( A ) and ( B ) Schematics of the M. tuberculosis parDE systems and co-transformants of M. smegmatis mc 2 155 containing the promoterless-lacZ pJEM15 vector-only and parDE promoter plasmids, together with the inducible pGMC vector-only or parDE expression plasmids, plated on LB plates containing X-gal (40 μg / ml) in the absence and presence of the pGMC inducer anh y drotetracy cline (A Tc, 1 00 ng / ml).ParE2 w as to xic under these conditions.( C ) and ( D ) β-galactosidase activity as determined from liquid culture of the abo v e co-transf ormed strains.ND indicates no dat a obt ained f or the induced ParE2 condition, due to to xicity.Graphs sho w mean v alues, and error bars represent the SD of triplicate data.ParDE1 negatively autoregulated transcriptional activity.

FFigur e 2 .Figur e 3 .
Figur e 2. P arE toxins induce gyrase-mediated DNA linearisation.( A ) P arDE1 induced DNA clea v age assa y. ( B ) Linearisation of pSG483 as sho wn in ( A ). ( C ) ParE2 induced clea v age assa y. ( D ) ParD2 induced clea v age assa y. ( E ) ParDE2 induced clea v age assa y. ( F ) Linearisation of pSG483 as sho wn in ( C -E ).ParDE system components were titrated against constant GyrB 2 A 2 (31.25 nM) and Supercoiled (S) plasmid DNA (12.5 nM).ParDE protein concentration per lane is presented below the agarose gels.Control lanes represent Supercoiled (S), Linear (L ) , Nicked (S) and Relaxed (multiple topoisomers) (R) plasmid DNA.Assa y s are presented on 1.4% agarose 1 × TAE gels (run with ethidium bromide (+EtBr) as stated, or post-stained).Assa y s sho wn are representative of triplicate experiments, and data points and error bars represent the mean and SD of triplicate data.

Figur e 4 .
Figur e 4. P arDE1 f orms a heterohe xameric comple x. ( A ) Cartoon representation of the heterohexameric P arD1 4 P arE1 2 comple x cry stal str uct ure.ParD1 antito xins e xist in tw o f orms within the str uct ure; t wo full-length (primary) chains are coloured dark green with N and C termini labelled, and t wo short-length (auxiliary), partially resolved chains are coloured pale green, with N and C labelled.The two ParE1 toxins are coloured light orange with N and C termini labelled.The ParD1 tetramer is formed through a beta sheet between the primary and auxiliary antitoxins and creates a cage-like str uct ure around the to xins.R otated vie w (right) sho ws the N-terminal dimerisation domain of the ParD1 antito xins f orming a ribbon-helix-helix str uct ure (RHH).( B ) Surface rendering of the ParDE1 complex crystal str uct ure coloured by electrostatic potential using the APBS plugin (PyMol).( C ) Cartoon representation of the ParE1 toxin, coloured grey, with conserved residues from the RelE / ParE superfamily highlighted in light orange.( D ) Surface rendered ParE1 with conserved residues in light orange complexed with Primary ParD1 (coloured forest green).( E ) Sequence-based str uct ural alignment of the ParE1 crystal str uct ure (light orange) and AlphaFold model (dark grey), which returned an RMSD of 0.594 Å. (F and G) AlphaFold models of P arD1 2 P arE1 2 heterotetramer ( F ) and P arD1 4 P arE1 2 heterohexamer ( G ). Rotated view shows the ParD1 complexing region.AlphaFold models shown with ParE1 in dark grey and ParD1 in light blue and lilac.

Figur e 5 .
Figur e 5. P arDE1 can f orm multiple comple x es. ( A ) Analytical SEC traces of tw o independent purifications of the ParDE1 comple x. ( B ) Analytical SEC traces of separated ParDE1 peaks from a mixed sample as presented in (A).( C ) Protein thermal denaturation curves for the separated ParDE1 samples normalised to 222 nm. ( D ) Circular dichroism spectroscopy scans for ParDE1 peak 1, before (solid red) and after (dashed red) melting.( E ) Circular dichroism spectroscopy scans for ParDE1 peak 2, before (solid black) and after (dashed black) melting; ( F )Table of ParDE1 Stokes Radius (R st ) calculations, observations, and comparisons.Comparison of observed / calculated is coloured green if within 10% of the predicted ratio, yellow if > 10 ≥ 25%, and red if > 25%.1 R st for the crystal str uct ure solution of P arD1 4 P arE1 2 ; 2 R st for the AlphaFold solution of P arD1 2 P arE1 2 .R st values were generated using HullRad(Fleming and Fleming, 2018).Chromatograms are representative of duplicate data and are normalised between 0 and 1 for presentation and comparison.Graphs are cropped to the appropriate scale (10-22.5 ml).

Figur e 6 .
Figur e 6. P arDE1 undergoes conditional comple x remodelling.Analytical SEC traces f or ParDE1 ( A ) P urified in the heterotetrameric fraction (0 hr, black).When incubated at 4 • C this single peak shifts into the mixed peak (48 hr, blue).Positions of peak 1 and peak 2 are indicated.( B ) Increasing complex concentration presented alongside the mixed sample from A (light grey).( C ) 37 • C incubation alongside the st arting , 0 h trace (solid black).Vertical arrow indicates potential free ParE1 'peak 3 .Chromatograms are representative of single repeats and are normalised between 0 and 1 for presentation and comparison.Graphs are cropped to the appropriate scale (10-22.5 ml).( D ) Mass photometry analysis demonstrates that, during incubation at 37 • C, ParDE1 heterotetramer comple x es (measured at mean 47-48 kDa) proportionally shift to heterohe xameric comple x es (measured at mean 74-81 kDa).At longer time points 24 and 48 h, larger species, potentially indicating higher order comple x es, are observed (measured at mean 11 5-1 21 kDa).Graph shows normalised counts from approximately 10 0 0-150 0 molecules measured per data collection.

Figur e 7 .Figure 8 .
Figur e 7. P arDE1 induced clea v age through thermally driv en to xin liberation.( A ) Analytical SEC traces f or ParDE1 37 • C incubation time course, starting with heterotetrameric complex (0 hr, solid black).A 100 μl sample of incubated ParDE1 (100 μM) was injected onto a Superose 6 10 / 300 GL column using an Åkta Pure system at each listed time point.Horizontal arrow indicates the general direction of peak shifting.Vertical arrow indicates the peak within the 1 hr SEC experiment that could represent free ParE1.* denotes SEC samples not used in the subsequent biochemical assay due to practicalities of the timings in v olv ed. ( B ) ParDE1 induced gyrase-dependent DNA clea v age assa y.GyrB 2 A 2 w as reconstituted, diluted, and stored on ice.ParDE1 (10 μM final concentration) from each time point was added to constant GyrB 2 A 2 (31.25 nM) and Supercoiled (S) plasmid DNA (12.5 nM).Presence (+) or omission (-) of GyrB 2 A 2 and ParDE1 is detailed between the agarose gels for each lane.A 12 h GyrB 2 A 2 only assay is included to ensure enzyme st abilit y throughout the time-course.ParDE1 tot al incubation time at 37 • C (pre-assa y incubation time point + assa y incubation) is sho wn belo w the gels (h).Control lanes represent Supercoiled (S), Linear (L), Nicked (S), and Relaxed (multiple topoisomers) (R) plasmid DNA.Assays are presented on 1.4% Agarose 1x TAE gels (run with ethidium bromide (+EtBr) as stated, or post-stained) alongside graphical analysis of percentage Linear / Nicked DNA and total percentage loss in band intensity per lane (+EtBr, obtained by densitometry) against time (hr).Incubation time for ParDE1 is quantified below the gels.Assays shown are representative of triplicate experiments, and data points and error bars represent the mean and SD of triplicate data.

Table 1 .
X-ray data collection and refinement statistics Table of ParDE1 Stokes Radius (R st ) calculations, observations, and comparisons.Comparison of observed / calculated is coloured green if within 10% of the predicted ratio, yellow if > 10 ≥ 25%, and red if > 25%.