Introduction

All microorganisms need to rapidly sense environmental changes and to respond to them in order to survive, adapt and reproduce. This response is mainly mediated by transcription factors (TFs) that regulate the expression of relevant proteins by either blocking or favoring the activity of RNA polymerases at specific gene sites1,2. Besides their cognate DNA-binding site region, TFs generally have a domain that recognizes effectors, such as other proteins or small ligands. Members of the CRP-FNR and Rrf2 families of dimeric microbial TFs can use the gas-binding affinity, redox properties and intrinsic lability of iron-sulfur clusters to modulate their DNA binding3,4,5,6,7. The well-studied Fumarate Nitrate Reductase regulator (FNR) uses a [4Fe-4S] cluster to sense O2 levels and to regulate a dimer ↔ monomer transition mediated by cluster disassembly and a metastable protein interface8. Only the dimer binds to specific DNA sites9. The Rrf2 family was first described in Desulfovibrio vulgaris Hildenborough10 and the three-dimensional structure of the cysteine metabolic regulator CymR from Bacillus subtilis (Bs), which does not bind an iron-sulfur cluster, was the first to be reported for a member of this family11. Structural and functional studies of the related Staphylococcus aureus (Sa) CymR revealed a sensing mechanism in which DNA binding is lost under oxidative stress conditions after oxidation of a surface cysteine thiol12. BsCymR binds to DNA after forming a complex with O-acetyl-thiol-lyase only when cysteine is present13.

Members of the Rrf2 family that coordinate iron-sulfur clusters use them to sense various effectors and regulate gene expression to enable adaptation, maintenance of homeostasis and cell protection. Characterized cluster-binding TFs include: IscR from Escherichia coli (Ec) and Thermincola potens (Tp) that sense cell iron-sulfur cluster levels and oxidative stress6,14; RsrR, a redox-sensitive response regulator from the soil bacterium Streptomyces venezuelae (Sv)5; Rhizobium leginosarum (Rl) RirA that uses a labile [4Fe-4S] cluster to regulate cellular iron concentrations15; and NsrR, a regulator of nitric oxide (NO)-induced stress in many bacterial species4,16. The regulation of their DNA binding capabilities depends on protein conformational changes caused by the status of the cluster or its absence. The first crystal structures of an iron-sulfur sensor of the Rrf2 family were those of the apo-form of Ec- and Tp-IscR17,18. More relevant to the work reported here, the structure of apo-EcIscR in complex with a 29 base pair (bp) fragment of the hydrogenase-1 hyA operator was also solved17. The structure of [2Fe-2S]-IscR, which recognizes different operator sequences19, has not yet been determined, either alone or in complex with DNA.

We have reported the crystal structures of [2Fe-2S]-SvRsrR alone20 and in complex with cognate DNA, the latter at medium resolution21. Using a combined approach of structural and mass spectrometric studies and molecular dynamics calculations, we were able to show how the one-electron cluster reduction of [2Fe-2S]-SvRsrR triggers protonation of a histidine residue and the burial of a tryptophan side chain, causing conformational changes that lead to DNA dissociation21. Our groups have also reported the holo- and apo- crystal structures of NsrR from the antibiotic-producing bacterium Streptomyces coelicolor (Sc)22. The principal regulatory target of this TF is, in most organisms, the hmp gene, which encodes a flavohemoglobin that converts NO to nitrate (NO3−) or nitrous oxide (N2O) under aerobic or anaerobic conditions, respectively23,24.

ScNsrR is so far the only NsrR to have been characterized in significant detail22,24,25,26,27,28. Its regulon is small, enabling a detailed investigation of the relative DNA binding properties of [4Fe-4S]-ScNsrR to each identified operator: hmpA1, hmpA2, and nsrR, showing a 23-base pair (bp) consensus recognition site. Tightest binding was observed for hmpA1, followed by nsrR and hmpA2.25 In the presence of NO, the ScNsrR [4Fe-4S] cluster is rapidly lost, causing the dissociation of the protein-DNA complex. Binding of hmpA2 was found to be the most sensitive to NO, followed by hmpA1 and nsrR24. Apo-ScNsrR (or [2Fe-2S]-ScNsrR, which could be generated and stabilized upon O2 exposure in the presence of β-mercaptoethanol)29 did not bind DNA with significant affinity25. In some bacteria, such as Ec and Salmonella enterica serovar Typhimurium, many more genes are under NsrR control, suggesting a broader regulatory role for NsrR in these microorganisms30,31.

A comparison of the three-dimensional structures of holo [4Fe-4S]-ScNsrR and apo-ScNsrR furnished a first model to understand how NO binding to the iron-sulfur cluster, and the resulting conformational changes, control the configuration of the protein DNA-binding surface22. Here, we report the 3.0 Å resolution X-ray structure of ScNsrR complexed with a 23-bp hmpA1 operator fragment and use it, alongside the structures referred to above, to establish what determines the DNA binding specificity among the three structurally-related iron-sulfur Rrf2 family proteins IscR, RsrR and NsrR. DNA recognition is generally considered to depend on either a direct readout of the nucleobase sequence, a readout of a sequence-dependent DNA phosphate backbone shape, or a combination thereof32,33. Related to the recognition process is the flexibility of the DNA operator regions and their propensity to undergo conformational changes upon protein binding. In a classical study34, a crystalline B-DNA dodecamer went from an almost straight double helix structure to a bent one as the temperature was changed from 7 °C to 20 °C. DNA flexibility has also been shown by solution NMR studies: the 31P chemical shifts of double stranded (ds) B-DNA fragments vary as a function of the sequence-dependent ratio of BI and BII backbone states. These are in a dynamic conformational equilibrium, with BI being the most commonly observed in regular B-DNA35. We have explored these aspects by comparing the remarkably different DNA binding modes of [4Fe-4S]-ScNsrR (this work), apo-IscR17 and [2Fe-2S]-RsrR21 in their respective crystal structures. In addition, we have tested four modified oligonucleotides sequences between two recognition sites of the NsrR regulator in order to explore the effect of DNA flexibility on its binding. This analysis allows us to propose a series of structural features that determine specific DNA binding within a class of closely related proteins.

Results

Structure of the [4Fe-4S]-ScNsrR-hmpA1 operator complex

Table 1 shows the statistics of the 3.0 Å resolution structure of [4Fe-4S]-ScNsrR complexed to the 23-bp hmpA1 operator fragment. High atomic temperature (B) factors indicate significant disorder, in agreement with the flexible and dynamic structure previously observed for the noncomplexed protein structure22. Both the operator fragment and the bound protein dimer are about 80 Å long (Fig. 1). The buried surface area (BSA) of the protein/DNA complex is 4222 Å2, comparable to the BSA of 4981 Å2 of the NsrR dimer but much larger than the 1996 Å2 buried between the two strands of the DNA helical fragment. The BSA at the protein/DNA interface of the EcIscR/hyA complex17 is 3780 Å2. This relatively low value is explained by the fact that some of the amino acids of EcIscR facing the central region of the hyA operator were not resolved and therefore not included in the structure. A higher value of 4490 Å2 is obtained for the SvRsrR/rsrR complex21 because the larger 39 bp rsrR operator fragment covers the protein sensor more extensively than the 23 bp hmpA1 one does.

Table 1 X-ray data collection and refinement statistics.
Full size table
Fig. 1: Top view of the ScNsrR/hmpA1 operator complex.
figure 1

NsrR subunits Cα tracings are depicted in blue and gold with ribbons indicating α-helices and arrows β-strands. DNA strands carry nucleotide numbers and are shown as cyan and burgundy ribbons, with stars marking four BII backbone conformations in the central region. Cluster atoms and histidine side chains forming hydrogen bonds to phosphate groups are depicted as spheres (Fe: red-brown, S: yellow, N: blue and C: gray). The first three helices and the labeled wing loop of each protein subunit interact with the DNA. Minimal and maximal P-to-P distances between phosphate groups across the minor groove are indicated with dotted lines and given in Å. The green dashed line follows the bp origins of the DNA double helix as determined with the program DSSR37.

Full size image

Residues 59–64 of the ScNsrR dimer wings, which are disordered in the free protein22, are partially resolved in the complex where they interact with the ends of the hmpA1 operator fragment (Fig. 1). Other DNA binding regions include the N-termini of α-helices 1 and 2 and the side of the third one, which corresponds to the regulatory helix (RH) of a helix-turn-helix (HTH) motif. Conversely, the C93-C99 turn, which we previously considered as a potential additional contact interface22, does not directly interact with the DNA. Compared to many other proteins with a winged (w) HTH motif36, the DNA binding region of ScNsrR contains an additional preceding α-helix and, consequently, can be better described as an H-wHTH motif.

Structure of the DNA hmpA1 fragment

The 23 bp hmpA1 operator fragment has a B-DNA conformation. The double helix runs almost straight through the central 11 bp region and bends at its extremes around α-helix 3 of each of the NsrR subunits (Fig. 1). The distance between phosphate P atoms across the minor groove (MiG) decreases from 12.6 Å at the center of the fragment to 9.6 Å between nucleotides 0 and + 4 of the other strand. It increases again to about 13.8 Å between nucleotides +9 and −6 before narrowing to 10.3 Å between nucleotides + 11 and −7. Minimal MiG distances are observed for the + 4 and −7 phosphates, which respectively interact with the His42 and His52 imidazole groups. An analysis of torsion angles with the program DSSR37 (Fig. S1) shows that 32 of the 44 phosphate groups in the bound 23 bp hmpA1 operator fragment (excluding the 5’-terminal ones) are in the BI conformation. Conversely, the phosphates bound to the C3’ atom of the 0 and + 2 nucleotides are in the BII conformation (see also Fig. S2), while the remaining ones are in a conformation intermediate between BI and BII (Fig. 2). The reorientation of the T+2 3’-phosphates from BI to BII conformations causes propeller twists of −19.0° for bp C+3c/G-3d (Figs. 3 and S3) and −15.9° for bp T-3c/A+3d.

Fig. 2: Cartoon of the ScNsrR-hmpA1 complex.
figure 2

Blue- and purple-labeled residues belong to different NsrR subunits. Nucleotides are numbered from 0 at the center of the figure. Contacts with distances (d) < 3.3 Å from hmpA1 strands c and d to NsrR residues depicted in bold are shown by black dashed lines for H-bonds, by dashed arrows if a main chain N atom is involved and by red dashed lines for possible salt bridges with His42 and His52. Thin lines depict remaining contacts with d < 3.8 Å, including those for disordered NsrR residues (see also Table S1). Circles depicting phosphate (P) groups are colored according to the corresponding minimal P-to-P distance across the MiG relative to its average value of 11.7 Å in regular B-DNA (ΔPPMiG). Small circles and black stars indicate phosphate torsion angles in the BI and BII backbone conformation, respectively; asterisks highlight those with an intermediate conformation.

Full size image
Fig. 3: Close up view of the ScNsrR/hmpA1 interface.
figure 3

Cα traces of the A and B subunits are depicted in gold and blue, respectively; ribbons are used to depict α-helices. DNA double helix bonds and contacting NsrR residues are represented with sticks. Hydrogen bonds and salt bridges are indicated with thick dashed lines; two van der Waals contacts are shown as thin dashed lines. The asterisks (*) label the T+4c-methyl group and the C+3c-C2’ atom. Atom color codes: Fe brown-red, S yellow, P pink, O red, N blue and C green, black or gray. For stereo versions of this figure see Fig. S3a, b.

Full size image

Protein-DNA interactions

Most of the ScNsrR contacts are made with the phosphate backbone of the hmpA1 operator (Figs. 2, 3 and Table S1). The His42-Nδ1 and His52-Nε2 atoms interact with the respective 5’-phosphate groups of nucleotides +4 and −7 in opposite strands (Figs. 1–3). These interactions could be salt bridges if the Nδ1 and Nε2 atoms of the imidazole groups are protonated. The side chains of Arg2, Lys5 and Lys45, present at the contact interface, are disordered (Fig. S4), as they were in the 1.95 Å resolution structure of non-complexed [4Fe-4S]-ScNsrR22. However, these basic residues could still interact with the negatively charged DNA phosphate backbone (Fig. S5). The positive dipole moments at the N-termini of α-helices 1 and 2 provide additional electrostatic stabilization via H-bonds from: (i) Lys5 main chain N atom to OP1 and Phe6-N to OP2 of the 5’-phosphate group of the +3 nucleotide, which is in a BII conformation (the non-complexed ScNsrR structure has a bound sulfate at this position, see Fig. S6); (ii) Thr29 main chain N atom to the 5’-phosphate of the −9 nucleotide in the opposite strand, which is also hydrogen bonded to Thr29-Oγ1. The Thr41 and Thr48 Oγ1 atoms of α-helix 3 make similar H-bonds to the respective 5’-phosphates of the +5 and −8 nucleotides in opposite strands.

There are only a few specific interactions between ScNsrR amino acid side chains and the nucleobases of the hmpA1 operator (Fig. 3 and Table S1). In the major groove (MaG), Lys45 could establish H-bonds to ring atoms of G−6 and surrounding bases. However, its weak electron density (Fig. S4) makes a definite assignment difficult. The 58–61 and 66 wing residues bind edge-on to the MiG (see also Fig. S3b), with Arg60 making the most contacts to the DNA. These include a H-bond between its guanidinium group and the O2 atom of the pyrimidine base at the −9 position and van der Waals (vdW) contacts with T+10-O2 and G+9d-N2. Gly61 makes vdW contacts with T+11-O2. Base-specific vdW contacts are observed in the MaG from Thr41A (4 atoms), His42-Cδ2 and Tyr40B-Oη to the C7 atoms of T+5d, T+4 and T-9d, respectively (Table S1). Additional resolved interactions with deoxyribose atoms involve the RH Thr41 and Thr48 and the wing Arg60.

Comparison of the protein-DNA complex with DNA-free ScNsrR structures

A cavity analysis indicates that the putative access path for NO that runs along the cluster ligand Asp8, previously observed in [4Fe-4S]-ScNsrR22, is also present in its complex with the 23 bp hmpA1 operator fragment (Fig. S7). The most important structural difference in the latter is a rearrangement of the region following α-helix 5 (Fig. 4a) caused by the electrostatic repulsion between the negatively charged Glu87 and the phosphate backbone. A similar effect could have provoked a small shift of Asp96 away from the DNA (see also Fig. S6). A much larger structural rearrangement of the 87–105 region is observed in the apo-3CA-ScNsrR variant where the cluster-coordinating Cys93, Cys99 and Cys105 were substituted by Ala (Figs. 4b and S8). In this variant22, helix 5 is kinked at Gly86 and extends to Gly95. Consequently, the C93-C99 loop, which in the [4Fe-4S]-ScNsrR-DNA complex faces the MiG of the hmpA1 central region (Fig. 4a), adopts an entirely different conformation. Because α-helices 1, 2 and 3 are rearranged in the apo structure, its complex with hmpA1 DNA is not stereochemically possible as it would involve several severe clashes with the N-terminal regions of these helices (Fig. 4b).

Fig. 4: Comparison of [4Fe-4S]-ScNsrR/hmpA1 complex with DNA-free structures.
figure 4

a Structural superposition of the complex to [4Fe-4S]-ScNsrR (green, cluster not shown). b Superposition to apo-3CA-ScNsrR (gray). The complex is colored as in Fig. 1 and selected amino acid residues are highlighted. Red arrows indicate movements observed for residues 87, 93 and 99. Regions of the apo structure that would collide with hmpA1 are indicated by red asterisks. Atom color codes are as in Fig. 3. Rotated stereo zooms of the same superpositions are given in Fig. S6 and S8.

Full size image

Comparison with other sensors of the Rrf2 family

The homologous E. coli NO sensor displays several amino acid sequence differences with ScNsrR in the DNA contacting region (Fig. 5). For example, ScNsrR Thr29, Thr41 and Thr48, which are hydrogen-bonded to phosphate groups of hmpA1, are neither conserved in EcNsrR, nor replaced by Ser, which would allow a similar interaction. In addition, the phosphate-binding His52 is substituted by Arg. These dissimilarities could explain why ScNsrR does not display significant affinity for EcNsrR-specific DNA sequences25. The above mentioned His and Thr residues are also not conserved in SvRsrR or EcIscR, the other two Rrf2 family iron-sulfur cluster binding sensors with known DNA complex structures, although Thr41 is replaced here by Ser. His42, the other ScNsrR residue that possibly makes a salt bridge with the phosphate backbone, is replaced by Tyr in those sensors. Amino acid sequence comparisons, including RlRirA and Sa and BsCymR, reveal a conserved Lx3Gx6Gx2GGx2L motif (Fig. 5) that spans the C-terminal region of helix 3 and the wing region. The comparison shows that most of the ScNsrR DNA contacting residues are not conserved in the eight Rrf2 family members. This indicates, as might be expected, that the different binding modes observed for these sensor/DNA complexes result from a combination of specific amino acid (Fig. 5) and nucleotide (Table 2) differences. A detailed comparison of the ScNsrR/hmpA1 and EcIscR/hyA complex structures reveals a small number of base-specific interactions (Table S1). Modelling different operator sequences on the structures of hmpA1 and hyA produces vdW collisions with the respective ScNsrR and EcIscR protein structures (shown in bold italics in Table 2).

Fig. 5: H-wHTH motif sequence alignment of eight selected Rrf2 family members.
figure 5

Numbered α-helices (H) and labeled wing β-strands are shown on top for ScNsrR; residues conserved in at least four sequences are indicated with capitals and given in bold when they are invariant. Percentages of sequence identity with ScNsrR and Protein Data Bank codes for the structures used here are given in the %ID and pdbc columns, respectively. Residues with H-bond or salt bridge distances ≤ 3.3 Å to phosphate groups in the ScNsrR/hmpA1, SvRsrR/rsrR and EcIscR/hyA structures are highlighted in orange and those that form H-bonds to nucleobases are depicted in turquoise; additional residues within 3.8 Å of the bound DNA fragments are highlighted in gray and those that are different from ScNsrR are shown in red. Cluster binding residues in ScNsrR and SvRsrR and the oxidation-sensitive cysteine of SaCymR are highlighted in purple. Abbreviations for different bacteria are defined in the text. Tp, Sa, and Bs are Gram-positive bacteria, all the others are Gram-negative.

Full size image
Table 2 NsrR and IscR base specificity for different operator fragments*.
Full size table

DNA operator fragments conformations

The DNA double helices of the 39-bp rsrR, 29-bp hyA and 23-bp hmpA1 operator fragments display different shapes in their respective complexes to SvRsrR, EcIscR and ScNsrR (Fig. 6). The DNA structures accommodate the different positions and orientations of the H-wHTH motifs found in the bound proteins (Fig. S9). The MiG and MaG P-to-P distances (Fig. 6b, c) and widths (Fig. S10) observed at different locations of the DNA double helix in the three complexes are remarkably different. The MiGs are especially different in the central 7-bp region (Fig. 6b, c) for which no base-specific contacts with the protein are observed (Fig. 2 and Table S1). These groove width variations condition the distance between equivalent operator regions (Fig. S11), which could guide recognition by different Tfs32,33.

Fig. 6: Shape comparison of the DNA complexes of three Rrf2 family members.
figure 6

a Base sequence alignment of hmpA1, hyA and rsrR operator fragments in their corresponding complexes with [4Fe-4S]-ScNsrR, apo-EcIscR and [2Fe-2S]-SvRsrR. b Comparison of P-to-P distances (dPPMig) in the MiGs of hmpA1, hyA and rsrR for the phosphate groups in both DNA strands. c Polypeptide and DNA folding of the three labeled complexes in a view perpendicular to that of Fig. 1. Protein subunits, DNA strands and bp tracing, phosphate positions, cluster atoms and amino acids are represented as in Fig. 1. Black dotted lines trace dPPMig distances (given in Å). Because of its limited resolution, no protein residues are shown for the SvRsrR-DNA complex.

Full size image

To investigate whether the sequence of the central 7-bp region of hmpA1 is important for DNA recognition by ScNsrR we performed EMSA binding studies of fragments with modified bases (Fig. 7), embedding the 23-bp consensus central region in 151-bp dsDNA probes (see Methods). Wild type hmpA1 (Av003) has four dinucleotides in a BII phosphate backbone conformation in the 7-bp central region: C0cA+1c, T+2cC+3c, G0dA+1d and T+2dA+3d (Figs. 1–2 and 6). Because XT dinucleotides have a low probability of being in the BII conformation in regular B-DNA38, we replaced A+1c and A+1d by T in probe Av004 and C+3c and A+3d by T in Av005 (see Table S2). The observation that some XT steps are found at the corresponding positions in the aligned hyA and rsrR operator fragments (Fig. 6a) suggested to us that they could play a discriminating role in DNA recognition between NsrR, IscR and RsrR. However, our EMSA results did not show any clear difference in binding affinity to ScNsrR between the two constructs having XT steps (Av004 & Av005) and the unmodified hmpA1 (Av003) sequence (Fig. 7a–c). We also checked the binding of variants with an AT-rich (Av006) and a GC-rich (Av007) central region (Fig. 7d–e), because AT-rich base sequences have been associated with an increased likelihood of MiG narrowing32, as observed in the central region of the 23-bp hmpA1 fragment (Figs. 2 and 6). However, again our results showed that ScNsrR binds the modified operators essentially stoichiometrically and we could detect no significant differences in binding affinities under these conditions. It should be noted that variations in the flexibility of the central region caused by A/T or G/C enrichments39,40,41 could affect DNA-protein binding kinetics. However, such an effect would not be detected by the EMSA assay reported here.

Fig. 7: Binding of NsrR to modified probes.
figure 7

DNA binding EMSAs showing probes (Av003 – Av007) unbound (U) or bound (B) by [4Fe-4S] NsrR. Probes contain: a Wild type hmpA1 operator, b hmpXT1, c hmpXT2, d hmpA/T, e hmpG/C sequences, f Protein-free control for hmpA/T. Base sequences for each probe are shown, with changes relative to the wild type hmpA1 site shown in red. The DNA concentration was ~4 nM. Concentration of [4Fe-4S] NsrR used and ratios of [4Fe-4S] NsrR to DNA are indicated. The binding buffer contained 10 mM Tris, 54 mM KCl, 0.3% (v/v) glycerol, 1.32 mM GSH, pH 7.5.

Full size image

Discussion

In spite of the large buried complex surface, most of the ScNsrR contacts with the hmpA1 operator are with the phosphate/sugar backbone; only a few direct interactions with nucleobase rings are observed (Figs. 2 and 3). What, then, determines the hmpA1 specificity for ScNsrR? Like most other DNA-binding proteins, sensors such as NsrR must locate specific targets in order to perform their biological role. Experiments performed with green fluorescent protein (GFP)-tagged lac repressor showed it to diffuse along the helical axis of the DNA probe via a sliding mechanism42. Furthermore, a similar study using an RNA polymerase demonstrated that this protein tracked the helical pitch along a DNA groove43. These interactions, made of non-specific electrostatic contacts of protein residues with the phosphate backbone, are dominated by a fully entropic potential44. Indeed, the electrostatic potential at the ScNsrR surface is consistent with such non-specific interactions playing an important role in DNA recognition (Fig. S5). Specific binding will then require a unique DNA sequence and protein-nucleobases interactions conditioned by the shape of both partners45.

The shape of B-DNA is determined by sequence-dependent variations in minor and major groove width and curvature that are at least partially coupled. Inter bp rolls, defined as rotations around the base pairing axis perpendicular to the double helix axis, are major contributors to DNA curvature46,47; zero rolls give straight B-DNA double helices48. The hyA central 11-bp region is curved, as reflected by an average bp roll of 2.3° (Fig. S1). This value is only 0.5° in hmpA1, in which the same region is almost straight. This may be related to the observed presence of four BII backbone conformations in hmpA1 (Fig. 2). Dinucleotide bps, such as TpC•GpA, with both connecting phosphates in a BI conformation (i.e., BI•BI) are mostly associated with positive rolls, whereas negative rolls are typically found for BI•BII conformations49. As a result, the rolls of a sequence of bps with alternating BI•BII and BI•BI phosphates (like in hmpA1) may partially cancel out, producing a smaller average roll than a sequence lacking BII conformations (like in hyA). However, our EMSA DNA binding studies (Fig. 7) do not suggest any role, at least in vitro, for the observed BII backbone conformations in hmpA1 binding to the protein sensor.

Different backbone torsion angle combinations are required for successive dinucleotides to bend in the same direction along the DNA double helix. Factors such as base orientation and stacking, and electrostatic interactions between phosphate groups may influence the torsion angle values. DNA bending towards the MiG, which narrows it, is energetically more costly than its bending towards the MaG50. However, the electrostatic repulsion between MiG opposing phosphate groups is greatly reduced by the proximity of positively charged amino acid sidechains. Thus, a minimal MiG width (MiGW) is observed at the + 4 phosphate of hmpA1 and at the +2 phosphate of hyA, which respectively make salt bridges with ScNsrR His42 (assuming it is positively charged) and EcIscR Arg2 (Fig. 6). Narrow MiGs in naked DNA have been associated with A/T bp-rich sequences32,46. Stella and coworkers found that the A/T-rich central region of the Fis protein high-affinity DNA binding sites displayed a small MiGW46. G/C substitutions in this region increased the MiGW and resulted in lower DNA binding affinity, an effect these authors explained by the increased conformational cost of compressing the MiG. However, B-DNA sequences are known to form dynamically heterogenous structures that explore a large conformational space51. When we performed an EMSA binding study of a variant of hmpA1 with 7 G/C bps in the central region, we did not observe a change in binding affinity to ScNsrR (Fig. 7). Neither did we see an effect when the region was enriched in A/T bps. This may imply that the flexibility of the hmpA1 DNA central region of the structure does not have a marked effect on the thermodynamics of the recognition of the operator -through either induced fit or conformational selection- by the different shapes and electrostatic potentials at the ScNsrR surface. A possibly kinetic effect would not be detected in the EMSA.

In view of these results, direct protein-to-DNA interactions, which mostly involve the RH and the wing (Figs. 2 and 6), are probably much more important for operator recognition. The DNA segments bending around the regulatory helix display a widening of the MaG. This widening is more pronounced in hyA than in hmpA1 and rsrR (Fig. S10), which could be due to the compression of the MiG between the P+2 phosphates of the neighboring central region in hyA (Fig. 6). The wing regions of ScNsrR, EcIscR and SvRsrR, which display the highest amino acid sequence homologies within Rrf2 family proteins (Fig. 5), interact with a more similar, relatively A/T-rich MiG (Figs. 2 and 6a). In hmpA1, as in hyA, the corresponding MiGWs are narrower than in rsrR. This is probably due to the close interaction of the positively charged ScNsrR Arg60, and EcIscR Arg59, with deoxyriboses and base rings at the MiG edge (Fig. 6c and Table S1). These interactions also compensate for the electrostatic repulsion between opposing phosphate groups. The broader rsrR MiG in the region that binds the protein wing could be explained by its lower A/T bp content and by the fact that in SvRsrR this Arg residue is substituted by Gln57 (Fig. 5).

To further analyze DNA binding specificity, we modeled the hyA and rsrR operator sequences (Fig. 6a) on that of hmpA1 in its complex with ScNsrR. Substituting hmpA1 C+6 by (hyA) T+6 introduces a vdW (d = 2.7 Å) collision between the added C7 methyl group and the Cγ2 methyl of (ScNsrR) Thr41 (Table 2). Because the latter is located in the N-terminal region of the RH and its Oγ1 atom makes a H-bond with the 5’-phosphate group of the same base in hmpA1 (Table S1), other interactions involving the RH are likely to be perturbed as well. The same substitution in the other DNA strand should similarly modify the RH of the corresponding ScNsrR subunit, which could be sufficient to abolish binding. Similarly, when modeled on the structure of hyA, the C7 methyl group of T+5 found in one strand of hmpA1 and in both strands of rsrR lies within 2 Å from the side chain of Ser40 of EcIscR (Table 2). In addition, the T-8d C7 methyl group of hmpA1 would collide with the carboxylate group of EcIscR Glu43, which has been assigned an important role in DNA recognition17. These collisions suggest that hmpA1 and rsrR are not recognized by EcIscR. Because of its relatively low resolution21, we refrained from doing a similar analysis using the structure of the SvRsrR/rsrR complex.

The observed interaction of Pro61 in the wing of EcIscR with the MiG of a 29 bp fragment of hyA (Table S1) suggests that the aligned Arg62 of ScNsrR could bind to the MiG of a longer than 23 bp fragment of hmpA1, as the one used in our EMSA studies which is 151 bp-long (Fig. 7). This should increase binding affinity. Substitution of (hmpA1) T+10 by (rsrR) G+10 introduces vdW collisions with (ScNsrR) Arg60 in both wing regions (Table 2). We also noted that the last three Gly residues of the conserved Lx3Gx6Gx2GGx2L wing motif directly interact with the DNA in the EcIscR/hyA complex structure (Table S1). This further underscores the importance of the wing region for DNA binding.

There are few structural changes in [4Fe-4S]-ScNsrR upon its binding to the hmpA1 operator (Fig. 4a). As mentioned in the Introduction, the nsrR gene codes for ScNsrR, whereas the two hmpA genes encode NO-detoxifying flavohemoglobins. Full binding of the [4Fe-4S]-ScNsrR dimer to 23 bp DNA stretches -included near the center of 267 bp constructs- as investigated by earlier reported EMSA studies25, requires corresponding ratios of 1:1 for hmpA1, 2.5:1 for nsrR and 4:1 for hmpA2. NO-induced structural changes resulting from iron-sulfur cluster modification and degradation will alter the ScNsrR structure and, consequently, modify its binding affinity. The [4Fe-4S]-ScNsrR TF binds to its known operators25 with much higher affinity than either non-physiological [2Fe-2S]-ScNsrR species29 or its apo-form. We have provided structural data to show that the loss of the iron-sulfur cluster in the latter introduces radical structural differences relative to the holo protein form at the DNA-interacting region22. Large conformational changes, including the possible formation of persulfide bonds27, should also be present in [2Fe-2S]-ScNsrR; indeed these bonds have been observed in partially oxidized [2Fe-2S] clusters of HydE52 and detected by resonance Raman spectroscopy in FNR53.

We have found that the DNA-free [4Fe-4S]-ScNsrR structure does not change much after binding of the 23 bp hmpA1 operator fragment and that changing the sequence of its central 7-bp region (which shows no direct base-protein contacts within 4 Å) has no effect on the binding affinity. Consequently, it seems reasonable to assume that the DNA backbone conformation of bound hmpA2 and nsrR operators is not very different from that of hmpA1. We have therefore modeled the hmpA2, nsrR and (Ec) hmpA sequences on that of hmpA1, knowing that (Ec) hmpA binds only very weakly to ScNsrR25. Analysis of the resulting predicted complex structures shows a number of bad contacts involving Arg60 in the two NsrR wing regions. Together, pointing into the minor groove at the two extremes of the hmpA1 operator fragment, Arg60A and Arg60B make close (<4 Å) contacts with eleven nucleotides (Fig. 2 and Table S1). A collision with one of these nucleotides should perturb the interactions with the other ones. Three bad (d < 2.85 Å) vdW contacts with Arg60 are obtained for the modeled ScNsrR/Ec-hmpA complex (Tables S3–S4 and Fig. S12), which could explain why no significant binding is observed for this operator. One bad contact with the Arg60 guanidinium group is obtained for nsrR, in line with its reduced binding affinity. The observed low affinity of hmpA2 is possibly explained by the substitution of T-9d by C and T+5d by G causing a loss of favorable vdW contacts with the T-C7 methyl groups.

Holo-ScNsrR-operator binding is abolished for NO-to-[4Fe-4S]-cluster ratios of about 2 for hmpA2, 4 for hmpA1 and 8 for nsrR24. As in the case of O2 oxidation, important conformational changes at the cluster region are expected when intermediate NO-bound degradation products are formed. Kinetic and spectroscopic studies have shown that these intermediates appear at ratios of about 2, 4, and 6 bound NOs per cluster24. They share spectral features with known NO-iron complexes such as [Fe(NO)2(RS)2] (dinitrosyl iron complex), [Fe2(NO)4(RS)2] (Roussin’s red ester) and [Fe4(NO)7(S)3] (Roussin’s black salt, RBS)24,26. In the case of the RBS-like intermediate, it has been suggested that Cys thiolates (or persulfide derivatives of these) replace sulfide ligands26,28; this would explain how these nitrosylated moieties bind to the protein after the loss of cluster sulfide ions. Several NO-modified cluster forms have also been characterized by mass spectrometry, including early mono- and di-nitrosyl intermediates of the [4Fe-4S] cluster28. At higher NO-to-cluster ratios, persulfide adducts accumulate, with a gradual breakdown of the [4Fe-4S] core to form [Fe2(NO)4] and [Fe2(NO)4S] adducts27. The ScNsrR-hmpA1 complex starts to dissociate at ~2 NO per [4Fe-4S] cluster, where only early intermediates of nitrosylation are observed. This suggests that relatively minor modifications of the cluster can abolish operator binding. The comparison of both non-complexed and complexed [4Fe-4S]-ScNsrR to apo-ScNsrR22 has defined the nature of the conformational changes induced by either iron-sulfur cluster modification or loss that will preclude sensor binding to hmpA1 (Fig. 4b). Because we have not yet characterized any NO-complex of ScNsrR, we have at this stage no structural data to explain the different NO sensitivities of its complexes with hmpA2, hmpA1 and nsrR.

In conclusion, the main determinant factors for the ScNsrR base sequence specificity can be described as follows: (i) the end regions of the hmpA1 operator interact extensively with the ScNsrR wings (this is also the case for hyA and rsrR, which respectively bind the apo-EcIsrR and [2Fe-2S]-SvRsrR wings); (ii) important structural differences between the bound operators of these three Rrf2 family members reside in variations of minor and major groove widths, which result in different phosphate positions; (iii) the DNA also displays specific curvatures around the RH of the sensor that properly orient its phosphate groups to form salt bridges and hydrogen bonds with equivalent side chains from both protein subunits; (iv) much of the specificity involves a relatively small number of direct interactions of protein side chains with nucleobase rings; (v) as shown in the EMSA experiments, changes in the flexibility of the hmpA1 central region do not affect the thermodynamics of NsrR binding (whether they modify the kinetics of this process was not investigated here). Solvent-mediated interactions may also play a role, but these are not resolved in our 3 Å resolution X-ray model. Our results extend the findings of a recently reported sequence similarity network analysis of the entire Rrf2 superfamily that places ScNsrR, EcNsrR, EcIscR and SvRsrR in distinct cluster configurations54. Our next challenges will be to crystallize complexes of [4Fe-4S]-ScNsrR with longer constructs of the hmpA1, hmpA2 and nsrR operators and solve their structures at the highest possible resolution. These structures should help us determine if and how the above-described determinants vary with the nature of the operator. In addition, we aim now at structurally characterizing NO-induced changes for at least one ScNsrR/DNA complex.

Methods

Crystallization

[4Fe-4S]-ScNsrR was purified under strictly anaerobic conditions as described previously22,25. A stock solution with a protein concentration of 43 mg/mL in 50 mM Tris pH 8.0, 2 M NaCl, 5% glycerol was used for all experiments. A 23 bp dsDNA fragment of the hmpA1 operator was prepared by annealing its 5’-AACACGAATATCATCTACCAATT-3’ and 5’-AATTGGTAGATGATATTCGTGTT-3’ complementary oligonucleotides in DNase-free water (Alfa Aesar), as already described previously24. [4Fe-4S]-ScNsrR was complexed with the 23 bp hmpA1 operator by incubating a solution of 17 mg/mL of the protein mixed with a 3-fold excess of the dsDNA fragment in 50 mM Tris pH 8.0, 500 mM NaCl, 5% glycerol for 60 minutes at 20 °C. Crystals were obtained by screening 864 different conditions from commercial kits in 96-well plates with a Gryphon robot (Art Robbins Instruments, CA, USA) installed inside an anaerobic glove box, using the vapor diffusion method. For each crystallization condition a drop was set by the robot by mixing 200 nL of the solution of the [4Fe-4S]-ScNsrR/hmpA1 complex with 200 nL of the commercial reservoir solution. The sitting drop was then equilibrated against 100 μL of the latter solution. Manual optimization of the best automated crystallization condition yielded usable crystals from hanging drops prepared by mixing 1 μL of the protein-DNA mix with 1 μL of the reservoir solution containing 27% PEG 550 MME, 13.5% PEG 20000, 30 mM MgCl2, 30 mM CaCl2 and 100 mM MOPS/HEPES pH 8.0 equilibrated against 1 mL of this solution. All these experiments were performed anaerobically in a glove box.

X-ray data collection

Crystals were transferred from the original drop to the crystallization solution, with or without 20% v/v glycerol added as cryoprotectant. They were then fished with a cryo-loop, flash-cooled in liquid propane inside the glove box55 and transferred outside for storage in liquid nitrogen. Initial X-ray diffraction data were collected to 4.1 Å resolution for a crystal mounted from a cryo-protected solution, while maintaining it under a N2 stream at 100 K at beamline PX-2A of the SOLEIL synchrotron (Saint-Aubain, France). Data to about 3.0 Å resolution were subsequently obtained from crystals flash-cooled without added glycerol. The best data set (Table 1) was collected at a temperature of 100 K and a wavelength of 1.00001 Å with an Eiger X 9 M (Dectris) detector at beamline X06SA of the Swiss Light Source (Villigen, Switzerland), using the DA+ software56. The diffraction data were indexed, integrated and scaled with the XDS package57 and the program AIMLESS58 was used to check for anisotropic diffraction. Although X-ray data reduction initially indicated that the crystal belonged to the hexagonal P6522 space group, further analysis showed it to be P65. The change in Laue class from 6/mmm to 6/m of the diffraction data set results from the fact that the dsDNA fragment is not a perfect palindrome; consequently, it does not display the twofold symmetry of the protein dimer. The intensity statistics of the data showed no evidence of twinning.

Structure determination

The structure of the [4Fe-4S]-ScNsrR/hmpA1 complex was solved by molecular replacement (MR) with PHASER59 using two search models: the known non-complexed protein structure22 and the 23 bp hmpA1 operator. The latter was constructed with B-DNA geometry using COOT60. The obtained MR solutions were confirmed by high peaks in Fobs-Fcalc and anomalous difference electron density maps for the ScNsrR [4Fe-4S] clusters, which were not included in the protein search model. The resulting model was subjected to rigid body, positional, TLS and B-factor refinement with the PHENIX and Refmac5 programs61,62, using COOT60 for manual corrections. Non-crystallographic symmetry and secondary structure restraints were used for positional refinement. H-bond restraints were included for DNA base pairing interactions and some protein-DNA contacts. The refinement procedure gave best results when defining the hmpA1 23 bp operator and two protein half-dimers as TLS groups. No hydrogens were included in the model. Each half-dimer consisted of NsrR residues 1–86 and 124–144 from one subunit and residues 87–123 from the other, including also the [4Fe-4S] cluster bound between them. The final model has good refinement statistics, stereochemistry (Table 1) and a low all-atom clash score of 3.8, as measured by MolProbity63. There are 0 outliers in the Ramachandran (φ-ψ) plot and 2.3% rotamer outliers. The dsDNA fragment fits well to the electron density map but a relatively poor match is observed for about 30% of the protein model due to disorder, as indicated by high B-factors. However, all ScNsrR side chains forming hydrogen bonds or salt bridges with the dsDNA have good matching electron density. We tried to explain residual peaks found at both ends of the hmpA1 operator fragment in the final Fobs-Fcalc map by adding additional ordered nucleotides to the model. However, this resulted in an increase of the Rfree. Because a threefold excess of DNA was used for crystallization, we tentatively assign these residual peaks to partially disordered non-complexed DNA. Other operator sequences were modeled with COOT60 on the available structures of hmpA1 (this study) and hyA17 without changing the structure of the DNA backbone.

Structural analyses

Buried surface areas (BSAs) between the various protein and DNA chains were determined with the program Pisa64. DNA geometry, including the measurement of torsion angles, bp origins, rolls and propeller twists, was analyzed with DSSR37, with groove widths calculated from phosphate-P positions as defined by El Hassan & Calladine65 (Fig. S10). We also used the in-house program CavEnv (incorporated in the CCP4 package66), which takes into account all atoms of the structure with an atom-type dependent vdW radius, to investigate cavities and groove widths (Fig. S7). Figures were prepared with either COOT60, Bobscript67 and Raster3D68, PyMOL (www.pymol.org, the PyMOL molecular graphics system, version 2.1 Schrödinger, LLC) or manually. Compilations of all the used programs were provided by SBGrid69. The A/T notation for bps and AT for base steps is used throughout, with a number and one letter for the strand added as an index in some cases. Nucleotides are numbered with respect to the central bp, which is assigned the value zero. Amino acid residues are shown using three-letter codes in the text and one-letter codes in the figures and tables. Structures of the EcIscR/hyA and SvRsrR/rsrR complexes were obtained from deposition codes 4HF1 and 6Y42 in the Protein Data Bank70.

EMSA (electrophoretic mobility shift assay) binding studies

DNA fragments carrying wild type or mutated hmpA1 promoter sequences (Table S2) were PCR-amplified using a 151-base-pair double-stranded DNA template, 6-carboxyfluorescein (6-FAM) modified primers (Metabion, Germany) and Phusion Hot Start II DNA Polymerase (Thermo Scientific). The PCR products were purified using NucleoSpin Gel and PCR Clean‑up kit (Macherey-Nagel) according to the manufacturer’s instructions. The concentration of the 6-FAM labeled probes was determined using a Varian Cary 50 UV-Vis spectrophotometer. The molecular weights and 260 nm absorption coefficients for 6-FAM labelled probes were calculated using OligoCalc71. EMSA reactions (20 μl) were carried out on ice in 10 mM Tris, 54 mM KCl, 0.3% (v/v) glycerol, 1.32 mM glutathione, pH 7.5, as previously described25. Briefly, 2 μl of each DNA probe was titrated with varying aliquots of [4Fe-4S] NsrR, to a ~20-fold molar excess. Loading dye (2 μl of 0.3% (w/v) bromophenol blue, 50% (w/v) glycerol) was added and the reaction mixtures were immediately separated at 30 mA for 30 min on a 5% (w/v) polyacrylamide gel in 1 × TBE (89 mM Tris, 89 mM boric acid, 2 mM EDTA). Gels were visualized on a Typhoon FLA9000 (GE Healthcare). Polyacrylamide gels were pre-run at 30 mA for 2 min prior to use. Each probe has a dimeric NsrR binding site with one [4Fe-4S] cluster per protein subunit. As apo-NsrR does not bind DNA25, concentrations are presented as [4Fe-4S] per DNA, as indicated in Fig. 7.

Statistics and reproducibility

X-ray diffraction data collection and refinement statistics for the reported NsrR/hmpA1 complex are given in Table 1. EMSAs for each tested DNA fragment were carried out at least twice, on different days, to ensure repeatable observations. The images shown in Fig. 7 are the best representative images for each EMSA probe.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.