Entropic pressure controls the oligomerization of the Vibrio cholerae ParD2 antitoxin

ParD2 forms an open oligomer in solution, the size of which is limited by the presence of an intrinsically disordered tail. In the absence of this tail, ParD2 forms a circular hexadecamer.


Introduction
Like all organisms, bacteria have to deal with various kinds of stress that threaten their survival. These include nutrient starvation, chemical stress arising from compounds, including antibiotic or redox stress, and physical stress such as heat or salt. To deal with this, they have developed different strategies, including the generation of persister cells, a specific stochastically induced dormant metabolic state that allows a subpopulation to survive antibiotics to which they are not resistant. Among the potential players in the bacterial stressresponse network are sets of small two-gene operons encoding a toxic protein and a corresponding neutralizing protein or RNA, known as toxin-antitoxin (TA) modules. They are often abundant in free-living bacteria and opportunistic pathogens (Pandey & Gerdes, 2005). For example, the well known Mycobacterium tuberculosis contains at least 88 such modules, while the closely related M. smegmatis only contains five (Shao et al., 2011).
The physiological role of toxin-antitoxin modules has been greatly debated. One of their potential functions is the stabilization of mobile genetic elements (Gerdes et al., 1986;Szekeres et al., 2007). After initially having been observed as stabilizing elements on low-copy-number plasmids, TA modules were also found on chromosomal regions that do not encode essential genes, including integrative and conjugative elements, the superintegrons of Vibrio cholerae and V. vulnificus, chromosome II, cryptic prophages and genomic and ISSN 2059-7983 pathogenicity islands (Díaz-Orejas et al., 2017;Yao et al., 2018).
The most popular proposed function for TA modules is stress response (Gerdes, 2000;Gerdes et al., 2005;Hõ rak & Tamman, 2017). Indeed, the activity of TA toxins is upregulated during episodes of stress. This was initially proposed to be a consequence of the protease-dependent degradation of antitoxins, a mechanism that has recently been challenged (LeRoux et al., 2020;Song & Wood, 2020a). Furthermore, it remains unclear whether stress-related TA activation is an integral part of the stress-response network or whether it is rather a side effect of the SOS response, which is the general bacterial response to DNA damage (Little & Mount, 1982). Equally, the supposed role of TA modules in the onset of persistence remains unclear (Ronneau & Helaine, 2019).
Another function that has been attributed to TA modules is protection against bacteriophages via abortive infection (Fineran, 2019;Lopatina et al., 2020;Song & Wood, 2020b). Last but not least, it should be considered that TA modules might be mere selfish genetic elements that have adapted and been conserved in a number of cases because they provide some additional beneficial properties such as the functions described above.
Toxin-antitoxin modules have been divided into eight classes based on the nature of the antitoxin (protein or RNA) and the mechanism by which it counteracts the toxin (Page & Peti, 2016;Song & Wood, 2020a). The most common and beststudied class of toxin-antitoxin modules are the type II modules, in which the antitoxin encodes a protein that counteracts the toxin via the formation of a tight noncovalent complex. Type II modules can be further classified into an increasing number of nonrelated families. Among these, the parDE family of TA modules, although one of the first families to be discovered, has been relatively understudied (Roberts & Helinski, 1992). For two members, parDE from Escherichia coli plasmid RK2 and parDE2 from V. cholerae, the target of the ParE toxin was identified as gyrase (Jiang et al., 2002;Yuan et al., 2010). Like the more intensively studied CcdB, ParE poisons gyrase by stabilizing the so-called cleavable complex between gyrase and DNA. This leads to double-strand breaks, activation of the SOS response and ultimately, if the ParE toxin is not counteracted by the antitoxin ParD, cell death.
Currently, the Protein Data Bank (PDB) contains an NMR structure of ParD from plasmid RK2, an NMR structure of E. coli PaaA2, which is a truncated ParD protein lacking a DNA-binding domain, and crystal structures of ParD-ParE complexes from Caulobacter crescentus and Mesorhizobium opportunistum and of the PaaA2-ParE2 complex from the parDE-like paaR2-paaA2-parE2 module of E. coli O157:H7 (Oberer et al., 2007;Sterckx et al., 2014Sterckx et al., , 2016Dalton & Crosson, 2010;Aakre et al., 2015). The target has not been identified for either of the ParE proteins, and evidence has been presented that E. coli ParE2 does not interact with the DNA gyrase A subunit (Sterckx et al., 2016). Plasmid RK2, M. opportunistum and C. crescentus ParD fold into an N-terminal ribbon-helix-helix DNA-binding domain that is followed by a domain which is unfolded in solution in plasmid RK2 ParD but folds into a helix-helix-strand conformation that wraps around the ParE toxin in C. crescentus and M. opportunistum ParD. E. coli PaaA2 lacks the N-terminal DNA-binding domain and is mostly disordered in solution, but adopts the same conformation as the C-terminal domain of C. crescentus or M. opportunistum ParD when bound to its cognate ParE2.
The genome of V. cholerae contains three parDE modules, all of which are located in the superintegron on chromosome II (Yuan et al., 2011). The parDE1 and parDE3 modules have identical open reading frames and regulatory sequences, while the ParD and ParE proteins encoded by the parDE2 module share 14% and 22% sequence identity, respectively, with their parDE1/parDE3 counterparts. ParE2 has been shown to inhibit gyrase in vitro and to bind to an epitope on the gyrase A subunit that differs from the one targeted by F-plasmid CcdB (Yuan et al., 2010). In vivo, both ParE1/ParE3 and ParE2 inhibit cell division, activate the SOS response and contribute to the degradation of chromosome I upon the loss of chromosome II.
Transcriptional regulation of TA modules is often complex and involves ratio-dependent interplay between the toxin and antitoxin (Garcia-Pino et al., 2010;Yamaguchi & Inouye, 2011;Jurė nas et al., 2019;Page & Peti, 2016;Vandervelde et al., 2017;Xue et al., 2020). Intrinsically disordered segments on the antitoxin or toxin often play a major role in these mechanisms (De Jonge et al., 2009;Garcia-Pino et al., 2016;Talavera et al., 2019;De Bruyn et al., 2021). Little information has been obtained, however, on the regulation of parDE modules. Early work on plasmid RK2 parDE suggested that ParD alone can act as a repressor of the operon (Roberts et al., 1993), but it is not known whether the ParE toxin modulates this action. Here, we describe the crystal structure of V. cholerae ParD2 (VcParD2) and show that interdimer contacts in the crystal mimic the functional arrangement of the structurally related Streptococcus agalactiae CopG bound to its DNA target (PDB entry 1b01; Gomis-Rü th et al., 1998). While a doughnut-shaped hexadecamer is observed in the crystal, in solution smaller decamers and dodecamers are present that form a partial doughnut with otherwise similar inter-subunit contacts. The partial disruption of the full hexadecameric ring may possibly be attributed to steric pressure from the intrinsically disordered C-terminal tails that are absent in the crystallized entity.

Expression and purification
The coding region of the V. cholerae parDE2 operon was cloned into pET-28a, placing a T7 promotor upstream of the parD2 gene and adding a His tag to the C-terminus of VcParE2. E. coli BL21 (DE3) cells were transformed with this vector and grown at 37 C in lysogeny broth supplemented with 50 mg ml À1 kanamycin. Expression was induced at an OD 600 of 0.6 by adding 0.5 mM isopropyl -d-1-thiogalactopyranoside. The cultures were incubated overnight at 20 C, after which the cells were harvested by centrifugation at 5000 rev min À1 and 4 C for 15 min.
The cells were resuspended in 20 mM Tris pH 8.0, 500 mM NaCl, 2 mM -mercaptoethanol and lysed with a cell cracker after adding 50 mg ml À1 DNase and 2 mM MgCl 2 . After centrifugation at 19 500 rev min À1 and 4 C for 35 min, the supernatant was filtered using a 0.45 mm filter and then loaded onto a 5 ml HisTrap HP Nickel Sepharose column (GE Healthcare) equilibrated with the same buffer. The column was washed with 20 mM Tris pH 8.0, 500 mM NaCl, 2 mM -mercaptoethanol, and the VcParD2-VcParE2 complex was subsequently eluted using a step gradient of imidazole (0, 10, 25, 50, 250 and 500 mM) in the same buffer. The resulting samples were loaded onto a Superdex 200 16/90 size-exclusion chromatography (SEC) column (GE Healthcare) pre-equilibrated with 20 mM Tris pH 8.0, 500 mM NaCl, 2 mMmercaptoethanol to obtain pure VcParD2-VcParE2 complex.

Electrophoretic mobility shift assay
A 350 bp DNA fragment comprising the first 67 bp of the parD2 ORF and extending further upstream into the control region of the parDE2 operon from V. cholerae biovar El Tor strain N16961 (NCBI NC_002506.1) was synthesized by PCRbased gene assembly (Stemmer et al., 1995) after optimization of the oligonucleotides using the DNAWorks platform (Hoover & Lubkowski, 2002). The assembled 350 bp fragment was used as the template for PCR amplification of a 151 bp fragment comprising the putative operator region upstream of the parD2 ORF with the oligonucleotides forward (Fw) 5 0 -TGAGGCGTTTGTTATGCGC and reverse (Rv) 5 0 -TTTGT ATTTGGCTTGTAATAAAGCCAT as primers, of which one was 5 0 -32 P single-end labelled with ( 32 P)-ATP (Perkin Elmer, 3000 Ci mmol À1 ) and T4 polynucleotide kinase (Thermo Fisher) as described by Nguyen Le Minh et al. (2018). Labelled PCR fragments were purified by gel electrophoresis on 6% polyacrylamide. For electrophoretic mobility shift assays (EMSAs), increasing concentrations of ParD2 were mixed with labelled DNA (10 000-15 000 cpm) in buffer consisting of 20 mM Tris pH 8, 150 mM NaCl, 1 mM TCEP in a total volume of 20 ml and incubated at 20 C for 30 min. After incubation, 3 ml loading buffer (25% Ficoll, 0.1% xylenexyanol, 0.1% bromophenol) was added to each sample. Separation was performed on 6% polyacrylamide gels run in TBE buffer at 130 V for approximately 3 h.

Crystallization, data collection and structure determination
The protein was concentrated to 20 mg ml À1 in 20 mM Tris pH 8.0, 150 mM NaCl. For screening crystallization conditions, 0.1 ml protein solution was mixed with 0.1 ml reservoir solution in a sitting drop and equilibrated against 100 ml reservoir solution using a Mosquito HTS robot (SPT Labtech). Crystals were observed after approximately three months in 0.2 M lithium sulfate, 0.1 M MES pH 6.0, 20%(w/v) PEG 4000. The crystals were transferred to precipitant solution supplemented with 25% PEG 400 for cryoprotection and were immediately flash-cooled in liquid nitrogen. Crystallization information is summarized in Table 2. Data were collected on the PROXIMA-2A beamline at the SOLEIL synchrotron facility, Gif-sur-Yvette, Paris, France. All data were indexed, integrated and scaled with XDS (Kabsch, 2010) via the XDSME interface (Legrand, 2017 , 2003). Data-collection and processing statistics are summarized in Table 3. The structure of VcParD2 was determined by molecular replacement with Phaser-MR (McCoy, 2007) using the dimer of the N-terminal domain (residues 1-50) of C. crescentus ParD2 (CcParD2; PDB entry 3kxe; 63% sequence identity; Dalton & Crosson, 2010). Four copies of the N-terminal domain dimer were placed in the asymmetric unit. Several cycles of refinement in Phenix (Liebschner et al., 2019) and manual model building in Coot (Emsley et al., 2010) improved the phases. Further iterative cycles were performed in phenix.refine using an intensity-based maximum-likelihood target function, including TLS and NCS refinement with automated group determination as implemented in phenix. refine. Strong NCS restraints were applied throughout refinement, leading to a relatively high value for R work but a comparably small difference between R work and R free . Because of the low resolution, we assumed that all eight chains were identical and chose to include the same number of amino acids. Consequently, rather poor fits for residues 3-4 of some of the chains remain. Refinement statistics are given in Table 4. Structural homologs were identified using DALI (http:// ekhidna2.biocenter.helsinki.fi/dali; Holm, 2020). Macromolecular interfaces were analyzed using the PISA server (http://www.ebi.ac.uk/pdbe/prot_int/pistart.html; Krissinel & Henrick, 2007).

Circular-dichroism spectroscopy
Protein concentrations were determined spectrophotometrically assuming extinction coefficients at 280 nm of 2980 M À1 cm À1 for VcParD2 and 15 470 M À1 cm À1 for VcParE2-His as determined from the amino-acid sequences using the ProtParam tool from the ExPASy server (Gasteiger et al., 2005). CD spectra were recorded at room temperature on a Jasco J-715 spectropolarimeter at a concentration of 0.15 mg ml À1 in 20 mM Tris pH 8.0, 150 mM NaCl, 1 mM TCEP.

Analytical SEC and SEC-MALS
Analytical SEC experiments were performed using Superdex Increase 200 10/300 and Superdex Increase 75 10/ 300 columns (GE Healthcare) equilibrated with 20 mM Tris pH 8, 500 mM NaCl, 1 mM TCEP. The Bio-Rad gel-filtration standards (bovine thyroglobulin, 670 kDa; bovine -globulin, 158 kDa; chicken ovalbumin, 44 kDa; horse myoglobin, 17 kDa; vitamin B 12 , 1.35 kDa) were used to make a standard curve of the logarithm of the molecular weights of the standards as a function of their elution volumes.
Multi-angle light-scattering experiments coupled to SEC (SEC-MALS) were performed using a HPLC system (Waters) connected inline with a miniDAWN TREOS II (Wyatt Technology) light-scattering detector (using three angles) followed by a Shodex refractive-index detector (RI-501). A Shodex K402.5-4F SEC column was connected to the SEC-MALS system and equilibrated with 2-3 column volumes of the corresponding running buffer. A dilution series of VcParD2 samples from 18 to 0.1 mg ml À1 was prepared in the same buffer. 10 ml was injected for each dilution. A BSA sample at 1 mg ml À1 was used as a standard for calibration. The data were processed, and consequently the molar mass was determined, using the ASTRA 7.1.4 software.    different pH values (8.0 and 5.6) using Biospin bufferexchange columns (Bio-Rad, Temse, Belgium). After buffer exchange, the concentration of the protein in the various buffer conditions was determined using a NanoDrop P2000 spectrophotometer (Thermo Scientific, Waltham, Massachusetts, USA). For each buffer condition, samples were introduced into the mass spectrometer at VcParD2 concentrations of 0.1, 1.0 and 2.5 mg ml À1 . Nano-electrospray ionization (ESI) was performed using 3-4 ml of sample loaded into home-made gold-coated borosilicate glass capillaries, with spray voltages applied in the range +1.5-2.0 kV. The spectra were recorded on an ion mobility-enabled time-of-flight mass spectrometer (Synapt G2 HDMS, Waters, Wilmslow, UK). To achieve gentle, native-like conditions, the instrument parameters were carefully optimized in order to avoid ion activation and to preserve the higher order structure of VcParD2 in the mass spectrometer. The optimized values for VcParD2 are sampling cone, 50 V; extraction cone, 2 V; trap collision energy, 20 V; trap DC bias, 45 V; transfer collision energy, 4 V. The pressure in the source region (backing) and in the trap cell (collision gas) were 5.5 Â 10 À2 and 3.1 Â 10 À2 mbar, respectively. Spectra were externally calibrated using a 10 mg ml À1 solution of caesium iodide. Analyses of the acquired spectra were performed using MassLynx version 4.1 (Waters, Wilmslow, UK). Native MS spectra were smoothed (to an extent depending on the size of the complexes) and additionally centred to calculate the molecular weights.
2.8. Small-angle X-ray scattering SAXS data were collected on the SWING beamline at SOLEIL in HPLC mode using an Agilent BioSEC 3-300 column at a protein concentration of 12 mg ml À1 in 20 mM Tris pH 8.0, 150 mM NaCl, 1 mM TCEP at 19 C. Protein samples were briefly spun down before loading onto the SEC column. 50 ml VcParD2 was injected onto the column at a constant flow rate of 0.2 ml min À1 . The final scattering curve (after buffer subtraction) was generated for each sample after a range of scattering curves around the peak (with equivalent R g values) had been normalized and averaged. The R g values were derived from the Guinier approximation at small q values, while the I(0) parameter was estimated by extrapolation to q = 0 using the ATSAS suite (Manalastas-Cantos et al., 2021). Molecular weights were determined by the Bayesian estimation implemented in PRIMUS from the ATSAS suite.
All simulations were performed in Xplor-NIH version 2.49 (Schwieters et al., 2018), starting from the hexadecameric X-ray structure of VcParD2 solved in this work. Residues that were not resolved in the X-ray structure, including the disordered C-terminal tail, were added in Xplor-NIH using the standard topologies for individual amino acids, followed by minimization of the energy function consisting of the standard geometric (bonds, angles, dihedrals and impropers) and steric (van der Waals) terms.
For refinement against the experimental SAXS data, the positions of the structured protein regions were kept fixed, while the disordered protein termini (including residues 1-4 and 48-81) were given full degrees of freedom. The computational protocol comprised an initial simulated-annealing step followed by side-chain energy minimization as described previously (Schwieters & Clore, 2014). In addition to the standard geometric and steric terms, the energy function included a knowledge-based dihedral angle potential and a SAXS energy term incorporating the experimental data (Schwieters & Clore, 2014). Truncated SAXS curves (q < 0.5 Å À1 ) were used as the sole experimental input.
In each refinement run, 100 structures were calculated and the ten lowest-energy solutions representing the best agreement with the experimental data were retained for subsequent analysis. The agreement between the experimental and calculated SAXS curves (obtained with the calcSAXS helper program, which is part of the Xplor-NIH package) was assessed by calculating 2 as where I(q) calc,i and I(q) exp,i are the scattering intensities at a given q for the calculated and experimental SAXS curves, I(q) exp,i is an expethemental error on the corresponding I(q) exp,i value and n is the number of data points defining the experimental SAXS curve. SAXS experimental details are given in Table 5.

Purification of VcParD2 and VcParE2
We initially attempted to express VcParD2 in E. coli from a pET-28a expression vector that adds a C-terminal His tag followed by a TEV cleavage site. These attempts led to very poor yields, possibly as a consequence of proteolytic degradation of VcParD2 in the absence of VcParE2. We therefore altered our strategy and co-expressed VcParD2 and VcParE2 to obtain a VcParD2-VcParE2 complex. In order to obtain isolated VcParD2 and VcParE2, we used an on-column unfolding-refolding protocol similar to those developed previously for Phd/Doc, MazEF and HigBA (Sterckx et al., 2015). VcParE2 is His-tagged at its C-terminus and the VcParD2-VcParE2 complex was trapped on an Ni-NTA column. Firstly, after extensive washing of the column, the VcParD2-VcParE2 complex was eluted and further cleaned using SEC on a Superdex 200 column. This step removes contaminating proteins of lower molecular weight that may otherwise co-migrate with VcParD2 or VcParE2 in subsequent steps. Subsequently, the VcParD2-VcParE2 complex was reloaded onto the Ni-NTA column. VcParD2 can then be eluted by applying a GndHCl gradient. This allowed us to prepare VcParD2 in the absence of an affinity tag or other cloning artefact, ensuring that the quaternary structure is not affected by the presence of a tag. After elution of VcParD2, the protein was refolded by dialysis. The refolded protein was then further polished by SEC on a Superdex 200 column ( Supplementary Fig. S1).

research papers
Our method has the additional advantage that VcParE2 can also be obtained as, like most other TA toxins, VcParE2 cannot be expressed in the absence of its cognate antitoxin. The VcParE that remains trapped was refolded on the Ni-NTA column in a three-step procedure that consists of washing with (i) 20 mM Tris pH 8.0, 25 mM NaCl, 5% glycerol, (ii) 20 mM Tris pH 8.0, 150 mM NaCl, 5% glycerol and (iii) 20 mM Tris pH 8.0, 250 mM NaCl, 2 mM -mercaptoethanol. VcParE2 was subsequently eluted using a step gradient of imidazole. The resulting protein was then applied onto a Superdex 75 column to obtain a sample that is pure on SDS-PAGE.

VcParD2 is a well folded DNA-binding protein
Both VcParE2 and VcParD2 show CD spectra that are compatible with folded proteins (Fig. 1a). VcParD2 is almost exclusively composed of -helices, while VcParE2 contains both -helices and -sheets. The SEC profile of VcParE2 is compatible with a globular monomer (Fig. 1b). VcParD2 nevertheless elutes at an apparent molecular weight of 170 kDa, which is much higher than the 18 kDa expected for a dimer (Fig. 1c). While the presence of an IDP region will lead to aberrant migration in SEC, this deviation seems to be too large to be solely explainable by this phenomenon and a higher oligomer is likely to be formed. Both elution profiles are nevertheless indicative of monodisperse samples and together with the CD spectra suggest correctly folded proteins.
It is also of interest to compare the molecular weight of VcParD2 with that of the co-expressed VcParD2-VcParE2 complex. The latter elutes from a Superdex Increase 200 10/ 300 column at an apparent molecular weight of 96 kDa, which is significantly smaller than the corresponding value for the VcParD2 oligomer (Fig. 1c). This indicates that VcParE2 binding at least partially disrupts the VcParD2 oligomer. The resulting complex is still larger than the corresponding complex from C. crescentus (44 kDa; Dalton & Crosson, 2010).
In analogy to ParD from plasmid RK2 (RKParD; Roberts et al., 1993) and ParD2 from M. tuberculosis (Gupta et al., 2016), we predicted VcParD2 to be a DNA-binding protein that represses the parDE2 operon. In order to investigate this hypothesis, EMSA experiments were performed using a 151 bp fragment upstream of the translation start of parD2  and a 207 bp control fragment derived from the operator sequence of the Cupriavidus metallidurans psrQ2 gene (Ali et al., 2020). Clear binding can be observed of VcParD2 to its own operator region, while no binding is observed to the psrQ2 operator sequence (Fig. 2). DNA binding therefore appears to be specific and the experiment thus further validates the functionality of our VcParD2 preparation. With a protein concentration of 6 mM or higher required for binding, this not only corresponds to the same order of magnitude of binding strength as is typical for many bacterial repressors, including M. tuberculosis ParD2 (MtParD2; Gupta et al., 2016), but also indicates that VcParD2 is active as a DNAbinding protein in its oligomeric state (see below), further emphasizing the relevance of our structure.

The VcParD2 monomer and dimer
The VcParD2 sample formed crystals that diffracted to 3.1 Å resolution only after three months. The structure was determined by molecular replacement using the dimer of the N-terminal domain of C. crescentus ParD2 (CcParD2; Dalton & Crosson, 2010), which shows 63% sequence identity to VcParD2 (54% for the complete protein chain), as a search model. The structure was refined to an R work of 0.2709 and an R free of 0.2988 (Table 4). The asymmetric unit contained four dimers of the N-terminal domain of VcParD2 and the corresponding models encompass residues Lys3-Arg51. The N-terminal two and C-terminal 32 residues are not visible in the electron-density map.
Although we did not succeed in determining the length of the polypeptide present in the crystal, it is highly unlikely that eight intact VcParD2 chains are present in the asymmetric unit. The solvent content calculated based on the visible part of the VcParD2 chains is 53%, which is a very reasonable value. In the case of intact 81-amino-acid VcParD2 chains, this would decrease to 22%. Although such a low solvent content is not entirely impossible, the estimated probability based on the distribution of V M values in known crystal structures (Kantardjieff & Rupp, 2003) is below 1%, and in such a case the crowded environment would be expected to induce structure in the IDP region, as is observed for M. tuberculosis YefM (Kumar et al., 2008) and in part for bacteriophage P1 Phd (Garcia-Pino et al., 2010). Indeed, degradation of the IDP regions of antitoxins during crystallization has been observed before . Furthermore, limited proteolysis experiments using trypsin, subtilisin and proteinase K showed the appearance of faster moving fragments (one or two closely migrating bands near the buffer front; Supplementary Fig. S2), which further strengthens the hypothesis that the species we have crystallized may have originated via slow (over months) noncontrolled proteolysis.
As expected, the VcParD2 N-terminal domain (VcParD2 N ) adopts a ribbon-helix-helix DNA-binding fold (Fig. 3a), with a C r.m.s.d. of 0.7 Å on 47 equivalent residues to

Method
Simulations in Xplor-NIH based on crystallographic coordinates of the hexademeric assembly q-range for fitting (Å À1 ) 0-0.5 2 value 2.11 (average of the ten best dodecameric models) Adjustable parameters in the model fit Side-chain energy minimization, geometric and steric terms, knowledge-based dihedral angle potential included in the energy function Domain/subunit coordinates and contacts, regions of presumed flexibility Positions of the structured protein regions were kept fixed, while the disordered protein termini (including residues 1-4 and 48-81) were given full degrees of freedom ( f ) Data and model deposition IDs.
PDB code of crystallographic structure 7b22; PDB files of dodecamer models are available upon request  (Table 6 and Supplementary Fig. S3). Plasmid RK2 ParD, with only 8% sequence identity, only shows up as a low-ranking hit in this search.   Like other members of the MetJ/CopG/Arc family, including its homologs from plasmid RK2, M. opportunistum and C. crescentus, the VcParD2 monomers associate into a conserved dimer (Figs. 3a and 3b). Dimer formation buries about 1240 Å 2 of mainly hydrophobic surface and creates the hydrophobic core of the protein. According to the P(Á i G) value of 0.37 provided by the PISA server (Krissinel & Henrick, 2007), the contact surfaces are somewhat more hydrophobic that the average surface of a soluble protein, which is in agreement with a dimer that would be stable in   Figure 3 Crystal structure of VcParD2. (a) Cartoon representation of the VcParD2 dimer (residues 3-51) in three orientations. One monomer is coloured according to secondary structure (-strand, yellow; -helices, red; loop regions, green). The second monomer is coloured cyan. (b) Superposition of the VcParD2 N dimer (orange) with the dimers of other RHH-type transcription factors: RKParD (green), CcPArD (purple), MoParD3 (yellow) and SaCopG (cyan).
solution. Hydrophobic contacts involve contributions from most nonpolar side chains: Ile7 and Leu9 at the end of the N-terminal -strand, Phe13 and Phe16 in helix 1 and Ile17, Ala29, Val32, Ile33, Ala36, Leu37, Leu39 and Leu40 in helix 2. Most notable is the close contact between the 213 helices of both chains, in which Ala36 seems to be essential, with any other residue except for glycine at this position resulting in steric crowding. Further stabilization arises via hydrogen bonds, most importantly via pairing of the N-terminal -strands.

Higher order structure in the crystal
In the crystal, the VcParD2 N dimers are found in a circular arrangement. This results in a torus consisting of eight VcParD2 2 dimers (Fig. 4). The interface formed by adjacent VcParD2 2 dimers buries roughly 600 Å 2 . This interface is highly hydrophilic [PISA P(Á i G) value of 0.91] and is dominated by hydrogen bonds and salt bridges, and as such deviates from typical stable oligomerization interfaces (Fig. 5a). The interface is formed by seven amino-acid side chains: Arg25, Tyr26, Glu31, Arg34, Arg38, Glu41 and Asn42. These side chains are mostly involved in inter-side-chain hydrogen bonds or salt bridges (between Glu41 and Arg25 and between Glu31 and Arg34 as well as Arg38) and cation-stacking between Arg38 and Tyr26. The only hydrogen bond involving a mainchain atom is between the main-chain carbonyl of Arg25 and the side chain of Arg34. Also of interest is the close contact between the side chains of Arg38 in both chains, with their NH2 atoms being only 3.1 Å apart. This, together with the dominance of side chain-side chain interactions (which are entropically less favourable than main chain-main chain interactions), would normally lead us to conclude that this interface only corresponds to a packing contact and is unlikely to exist in solution.
The amino acids contributing to this interface are retained in CcParD, which has previously been shown to be a dimer in solution (Dalton & Crosson, 2010). The only substitution among the residues involved in the inter-dimer interface is Asn42, which is Glu43 in CcParD. This substitution forces Glu42 of CcParD into a different conformation to the equivalent Glu41 in VcParD2 in order to separate the two negative charges. This substitution would also result in an interface with a net negative charge of À2. Another potentially relevant difference between VcParD2 and CcParD is the reorientation of Arg38 (Arg39 in CcParD), which now interacts with Glu43 of CcParD. This further results in an 'inwards' movement of Arg26 of CcParD (Arg25 of VcParD2) to fill the space vacated by Arg39. The resulting theoretical interface would then be destabilized by close contacts between the positively charged side chains of Arg26 and Arg39 in the adjacent polypeptide chains.
The hole at the centre of the torus has a diameter of about 16 Å . The C-termini of the 16 VcParD2 N-terminal domains point outwards from the middle of the side surface of the torus. If this arrangement were composed of intact VcParD2 chains, the C-terminal IDP regions (residues Leu52-Arg81) would be forced to adopt a fan-shaped ensemble that is limited in its conformational variability. In contrast, in RKParD as well as in CcdA the C-terminal IDP ensemble adopts a wide range of conformations, many of them sterically incompatible with the oligomerization of the VcParD2 N-terminal domains into the torus observed in the crystal (Madl et al., 2006;Oberer et al., 2007). It is therefore safe to state that the IDP region would provide an entropic penalty for VcParD2 oligomerization similar to that observed in the influence of the IDP region of Phd on operator binding .
The VcParD2 hexadecamer resembles the crystallographic assembly of S. agalactiae CopG, which forms a spiral structure with inter-dimer contact surfaces roughly corresponding to the inter-dimer interfaces seen in the VcParD2 hexadecamer ( Fig. 5b; Gomis-Rü th et al., 1998). In contrast to VcParD2, although similar in size, the oligomerization interface of  Higher order oligomeric structure of VcParD2 N . Two orientations of a cartoon representation of the hexadecamer of VcParD2 as seen in the crystal. VcParD2 dimers are alternately coloured pink and blue.
Yet in solution, in the absence of a DNA ligand, SaCopG behaves as a dimer.

DNA-binding site
In general, transcription factors of the ribbon-helix-helix family dock with their N-terminal strands into the major groove of their target DNA. In agreement with this, the N-terminal -ribbons present a positively charged electrostatic surface on the VcParD2 dimer that can complement the negative charges of the DNA backbone.
S. agalactiae CopG (SaCopG) is the closest structural relative of VcParD2 for which a structure of a DNA complex is available. SaCopG makes base-pair-specific contacts via the side chains of Arg4 and Thr6, while hydrogen bonds to the DNA backbone are provided by the side chains of Thr8, Lys28 and Ser29 (Gomis-Rü th et al., 1998; PDB entry 1b01). In VcParD2, the corresponding residues are Asn4, Ser6, Thr8, Ser29 and Ser31, indicating potential differences in DNA specificity between the two proteins ( Supplementary Fig. 3a).
Unfortunately, the target sequence of VcParD2 currently remains unknown.
SaCopG binds its operator in a chain-like fashion. Several dimers position themselves next to each other in a similar way to that seen in the crystal packing of the free SaCopG structure, with the DNA wrapping around the protein helix (Costa et al., 2001). In contrast to SaCopG, VcParD2 does not associate into a helical structure, but forms a closed circle. The dimer-dimer association of VcParD2 is nevertheless very similar to that of SaCopG bound to DNA (Fig. 6a), suggesting that crystal packing may also reflect associations occurring on the DNA here. Indeed, on the toroidal surface of the VcParD2 crystallographic hexadecamer, the N-terminal -strands stick out as a set of eight parallel and equally spaced ridges that lead to a cogwheel-like arrangement (Figs. 3 and 5). Fig. 6(b) shows the positioning of an 18 bp fragment with a twonucleotide overhang taken from an SaCopG-DNA complex and covering two SaCopG dimers on the surface of the VcParD2 oligomer (Fig. 6a). When superimposed on VcParD2, the helical axis of the DNA is tangential to the  cogwheel surface, and the subsequent -strand ridges are spaced so as to make it possible for a double-stranded DNA to wrap around the cogwheel. Thus, it is likely that the oligomerization mode of VcParD2 may reflect how it interacts with its operator.

Oligomerization in solution
Despite the fact that there are several arguments that disfavour the existence of the crystalline oligomeric assembly in solution, during the purification of VcParD2 we observed that it elutes from a Superdex Increase 200 10/300 SEC column at an appreciably higher molecular weight than is expected for a simple dimer, even if we take into account the fact that the IDP region would substantially increase its hydrodynamic radius. To evaluate the relevance of the oligomer seen in the crystal, we determined the oligomeric state of VcParD2 in solution using SEC-MALS and native mass spectrometry.
To obtain a better estimate of the true oligomeric state in solution and how it may be affected by the concentration, we turned to SEC-MALS. A concentration series of VcParD2 ranging from 18 to 0.1 mg ml À1 was injected into a Shodex KW402.5-4F column. At all concentrations, a single peak was observed at an elution volume indicating a higher order oligomer (Fig. 7, Supplementary Table S1, Supplementary Fig.  S4). The molecular weight determined for the protein eluting in this peak ranges from 100.0 AE 0.6 kDa for the highest concentration used to 78.5 AE 0.6 kDa for the lowest concentration used. No additional peaks were observed that could be attributed to a monomer or dimer. Given that the theoretical molecular weight for a VcParD2 dimer is 17 912 Da and that it can be assumed that the oligomeric state of VcParD2 is a multiple of a dimer, the SEC-MALS data indicate the presence of mainly decamers, which are likely to be in equilibrium with dodecamers and octamers.
Little dependence on concentration or ionic strength was observed, although lower concentrations tend to give somewhat lower average molecular weights. Similarly, a lower pH leads to a somewhat higher average molecular weight. Most important, however, is that even at the highest concentration   used the experimental molecular weight is significantly less than that of a hexadecamer (as suggested from the crystal structure). This deviation from the expected molecular weight is not due to degradation of the C-termini, as nESI-TOF mass spectra of the protein sample after performing the SEC-MALS experiments showed the protein to be intact, with no signs of degradation.
To further understand the true nature of the higher order complexes that are present in solution, we performed native mass spectrometry (Fig. 8, Supplementary Fig. S5). These data nicely mirror the results obtained using SEC-MALS and show that VcParD2 is predominantly present as a mixture of decamers and dodecamers in solution, with tetramers as a third species. Similar to the observations using SEC-MALS, a lower pH favours the dodecameric assembly. At the lowest concentration (0.1 mg ml À1 ), some monomers, tetramers and a very small amount of dimers can nevertheless also be seen, but decamers and dodecamers still dominate, indicating that these are indeed stable species. The stability of the decamers and dodecamers, particularly at high ionic strength, is surprising given the electrostatic nature of the inter-dimer contact interface, as also is the lack of substantial amounts of tetramers, hexamers or octamers. We do not observe the presence of tetradecamers or the hexadecamer that would be expected from the crystal structure.

Solution SAXS model of the VcParD2 oligomer
In order to obtain a more detailed picture of the oligomeric species in solution, we turned to small-angle X-ray scattering (SAXS). The protein used in this experiment was again confirmed to be intact using nESI-TOF mass spectrometry, and therefore a significant fraction of its polypeptide (27%) is expected to be present in a disordered ensemble. SAXS data collected on the SWING beamline at SOLEIL were of high quality up to a q value of 0.5 ( Fig. 9a and Table 5). The Guinier plot shows linear behaviour (R 2 = 0.97), rendering a radius of gyration of 33.64 AE 0.34 Å . However, the dimensionless Kratky plot of VcParD2 does not reveal the disordered nature of the C-terminal regions ( Supplementary Fig. S6). Its bellshaped curve with a maximum around (1.73, 1.1) agrees with the expected values for a globular particle. The P(r) function shows a nice bell shape and converges smoothly to zero, with a good fit at low q angles, indicative of a polydisperse sample (Manalastas-Cantos et al., 2021).
SAXS data for the VcParD2 antitoxin indicate a molecular mass of approximately 109 kDa from the Bayesian estimate implemented in the ATSAS suite (Manalastas-Cantos et al., 2021). The oligomeric state suggested by this molecular mass thus corresponds to a dodecamer. The sample used in these experiments was subsequently verified by nESI-TOF mass spectrometry and was shown to consist only of intact chains. In agreement with this observation, the theoretical SAXS curve calculated for the hexadecameric crystallographic assembly does not fit the experimental scattering curve ( 2 = 28.16). Equally, the full-length protein in this hexadecameric assembly, with an ensemble in which residues 5-47 are fixed and residues 1-4 and 48-81 are given full torsional degrees of freedom, still provides a poor fit ( 2 = 7.80). In this ensemble, the IDP tails tend to plug the central hole, rendering the whole particle more globular. While these solutions are physically possible (there are very few, if any van der Waals clashes, torsion angles are good and local geometry is perfect), they do not make sense biochemically. Entropic considerations make such a structure, with the tails crammed up in the middle, highly unlikely.
We then considered a number of ensembles consisting of full-length VcParD2 chains in octameric, decameric, dodecameric and tetradecameric arrangements. As seen in Fig. 9(b), the best agreement is obtained for a dodecamer that forms an open fragment of a torus with identical relative orientations of the contacts between adjacent dimers as seen in the hexadecameric arrangement in the crystal ( 2 = 2.11). This ensemble features a realistic spread of tails, with some of them visiting the central inter-domain region (the former central hole) and others flopping about on the periphery (Fig. 9c). Nevertheless, similar decameric and tetradecameric models also fit reasonably well ( 2 = 2.58 and 2.78, respectively) and therefore cannot be excluded. Hybrid models containing increasing fractions of decameric structures added to the dodecameric ensemble do not further improve the fit. In other words, the dodecameric ensemble on its own is sufficient to explain the scattering data and is most likely to be the dominant species present in solution.

Discussion
We were able to obtain correctly folded VcParD2 from the overexpressed VcParD2-VcParE2 complex via an on-column unfolding-refolding procedure and showed that the resulting protein interacts specifically with a 151 bp DNA segment upstream of the ATG start codon of the parD2 gene. Transcription regulation of parDE modules has also been investigated for parDE on E. coli plasmid RK2 and parDE2 on the Native mass spectrometry. The native mass spectrum of VcParD2 (2.5 mg ml À1 ) in 150 mM ammonium acetate pH 8.0 is shown with the major peaks labelled according to their charge states. The major species present are decamers and dodecamers, with smaller amounts of monomers, dimers and tetramers also being observed. chromosome of M. tuberculosis H37Rv. For plasmid RK2 parDE, the antitoxin RKParD is responsible for auto-repression of the operon, with two RKParD dimers binding to the operator. In contrast to most other TA families, RKParE does not seem to modulate RKParD-mediated repression in vivo (Roberts et al., 1993;Johnson et al., 1996;Oberer et al., 1999). In the case of M. tuberculosis parDE2, MtParD2 also represses the operon on its own, but the presence of MtParE2 reduces repression in vivo (Gupta et al., 2016). Previous in vivo data from the V. cholerae parDE2 module are in line with those for plasmid RK2 parDE, with VcParE2 not being involved in auto-regulation (Yuan et al., 2011). Our in vitro DNA-binding results are in line with the results for the homologous parDE modules located on plasmid RK2 and the M. tuberculosis chromosome. The affinity of VcParD2 is in the low-micromolar range and its higher order oligomeric structure suggests multiple adjacent binding sites, as is also the case for RKParD, although the latter is a dimer in solution.
The folded domains of VcParD2 form a partial doughnut structure in solution that is stabilized via strong electrostatic interactions. While not recognized as such by the PISA server (Krissinel & Henrick, 2007), this association is stable and is maintained even at low protein concentrations. For over four decades, hydrophobicity, together with complementarity, has Small-angle X-ray scattering. (a) Experimental SAXS data (black) overlaid with the theoretical scattering curve for a VcParD2 dodecamer averaged over the ten best-fitting conformations (red; 2 = 2.11) and the curve calculated for the hexadecameric crystallographic assembly (cyan). (b) 2 values for the best fits of octameric, decameric, dodecameric, tetradecameric and hexadecameric ensembles. A minimum is seen for a dodecameric ensemble. (c) Molecular model of the dodecameric ensemble that best represents the scattering data. The folded parts of VcParD2 dimers are alternately shown in blue and pink. The IDP tails are shown in grey. been believed to be the major factor stabilizing proteinprotein association (Chothia & Janin, 1975). Contact surfaces in protein oligomers differ from the rest of the subunit surface in that they are enriched in hydrophobic side chains and have a low density of inter-subunit hydrogen bonds (Janin et al., 1988;Lo Conte et al., 1999). In small dimeric proteins, the dimer interface often corresponds to the hydrophobic core and dimerization is essential to form a stable structure.
In contrast, high-affinity electrostatic interactions are known between IDPs, for example the disordered complex formed between histone H1 and its chaperone prothymosin- (Borgia et al., 2018). Similar poly-electrolytic interactions have also been postulated to drive the formation of membraneless organelles via liquid-liquid phase transitions (Brangwynne et al., 2015;Schuler et al., 2020). On the other hand, to find a stable association of globular proteins dominated by side chain-side chain hydrogen bonds and electrostatic interactions is highly unusual. To our knowledge, no other examples of such complexes are known, and the VcParD2 oligomer thus represents an example of a rare class of protein-protein interface. The stability of this oligomer is even more surprising when one considers that the amino acids involved in this interaction are highly conserved in CcParD from C. crescentus, with the only difference being the substitution of an asparagine by a glutamate in CcParD. The latter would bury two negative charges at the inter-dimer interface, creating a highly unstable situation.
In the crystal, the globular domain of VcParD2 forms a closed doughnut-shaped structure in the absence of its IDP tail, which most likely degraded in the two months that were required for crystals to appear. In the absence of any cooperativity, one would expect that with identical interactions between all dimers, a complete circle would also be formed in solution. Nevertheless, we were able to demonstrate that the doughnut is incomplete in solution and that the assembly consists of 5-6 VcParD2 dimers instead of eight. This implies the existence of negative cooperativity that prevents formation of the closed structure. It is likely that this negative cooperativity originates from the IDP tails, which in a full hexadecameric assembly would hinder each other's freedom and create an entropic barrier that prevents an oligomer larger than a dodecamer from forming. A similar entropic exclusion principle has previously been observed for the binding of Phd to its operator . Two copies of the Phd dimer need to bind at adjacent sites, but this is prevented due to entropic exclusion of the IDP tails. Only when Doc binds and folds these IDP tails can operator binding proceed with high affinity and the phd/doc operon be repressed. This mechanism is further related to the action of entropic bristles, which are IDP tails that can act as solubilizers to prevent aggregation (Santner et al., 2012), tune prion nucleation (Michiels et al., 2020) and tune the energy landscape of proteins in general with respect to protein assemblies and ligand binding and association (Keul et al., 2018;Niemeyer et al., 2020).
The similarities in higher order association between VcParD2 and SaCopG from S. agalactiae suggest a mechanism for DNA binding. VcParD2 and SaCopG make higher order contacts via the same spatial surface, although this surface is substantially more hydrophobic in SaCopG. This association in SaCopG leads to an extended DNA-binding surface where multiple SaCopG dimers dock next to each other on the operator. In many TA systems, the antitoxin often binds to two or more binding sites on the operator. In most cases, the toxin increases antitoxin-DNA affinity by bridging adjacent antitoxin dimers (Garcia-Pino et al., 2010;Vandervelde et al., 2017;Xue et al., 2020) or by otherwise stabilizing the DNA-binding assembly of the antitoxin (Bøggild et al., 2012;Qian et al., 2019;Jurė nas et al., 2019). Higher toxin:antitoxin ratios lead to de-repression via the formation of toxin-antitoxin complexes with altered stoichiometry. For antitoxins that only bind to a single site, regulation is less complex and the toxin only serves to weaken operator binding (Brown et al., 2013;Turnbull & Gerdes, 2017;Winter et al., 2018;Manav et al., 2019). For parDE modules, the mechanism of transcription regulation is not known. Early studies of the parDE system present on plasmid RK2 indicated that the antitoxin is sufficient to repress the operon and is likely to bind on two adjacent palindromes (Roberts et al., 1993). Although the operator of the V. cholerae parDE2 operon is not known, the higher order association of VcParD2 dimers suggests cooperative binding to the DNA in a similar way to that observed for SaCopG. The extent to which the corresponding toxin influences this interaction currently remains unknown.