DNA deformability in sequence-dependent capture of E. coli gyrase

Abstract

To understand how gyrase interacts with DNA and selects a site of action, we created an ad hoc shape-based recognition methodology to ascertain the DNA sequence from cryoEM density maps as a string of purines and pyrimidines, which matched to the DNA minicircle sequence in our two previous cryoEM structures of negatively supercoiled DNA bound to E. coli gyrase. For one structure, the Gate- or G-segment sequence contains base-pair steps that are among the most flexible in the minicircle, facilitating the bend. The sequence flanking this G-segment is highly inflexible, preventing wrapping the β-pinwheel of gyrase. In the other structure, a flexible DNA minicircle sequence wraps a β-pinwheel of gyrase and the G-segment contains base-pair steps of average deformability. This work highlights how DNA sequence and deformability impact gyrase. It also demonstrates the utility of both identifying DNA sequences from cryoEM structures and assessing base-pair step deformability.

Here, the authors develop a shape-based approach to identify DNA sequence from cryoEM density maps. Integrating sequence identification with quantitative analysis of base-pair step deformability, their work reveals that gyrase engages DNA regions with specific mechanical properties.

Introduction

Essential and ubiquitous, type II topoisomerases pass DNA helices through each other to modulate DNA supercoiling, unlink DNA catenanes, and untie DNA knots^1–5. Several steps are required: the enzymes bind and cleave the DNA (becoming transiently covalently attached to newly formed ends), pass an intact DNA helix through the break, and then religate the DNA backbone⁶. Not strictly site-specific, type II topoisomerases nonetheless cleave linear, relaxed, and positively or negatively supercoiled double-stranded DNA with the same reproducible sequence patterns^7–10. Typically, type II topoisomerases will act with high processivity, catalyzing several cycles of cleavage, strand passage, and religation before dissociating from the DNA^11,12.

Type II topoisomerases require divalent cations and ATP for activity. Cations are required for DNA cleavage and religation. ATP binding supports DNA strand passage; ATP hydrolysis is required for enzyme turnover. The bacterial type II topoisomerase, gyrase, however, can strand-pass without ATP, at least in vitro¹³. Recent research revealed that the removal or mutation of the ATPase domain in other type II topoisomerases may also allow for ATP-independent strand passage¹⁴. New insights into the molecular mechanisms of topoisomerases continue to emerge, yet there remains much we do not understand.

Gyrase, unique among type II topoisomerases, introduces negative supercoils into DNA. It is its negative supercoiling, and not a direct unlinking, that facilitates topoisomerase IV-mediated decatenation^15,16. Gyrase has no role in untying DNA knots, which can be mutagenic and potentially lethal¹⁷; topoisomerase IV unknots DNA independently of gyrase-mediated supercoiling¹⁸. Although studied for more than half a century, how gyrase interacts with DNA, how DNA sequence affects the enzyme, and important details of the molecular mechanism of this critically important antibiotic target¹⁹ remain to be determined.

We recently reported two near-atomic (2.9 and 3.0 Å resolution, EMDB ID: EMD-18603 and EMD-18342, and PDB IDs: 8QQS and 8QDX, respectively) structures of a catalytically inactive E. coli DNA gyrase mutant (GyrA^Y122F) bound to a 601 base-pair (bp) DNA minicircle that was determined by single-particle cryoEM (Fig. 1A, B)²⁰. The minicircle used was negatively supercoiled; specifically, it was underwound relative to the relaxed minicircle with a change in the linking number (ΔLk) of −2 (Lk describes how many times the two DNA strands intertwine). In one structure, which we refer to as the “wrapped structure” (Fig. 1A), a 118 bp segment of the negatively supercoiled minicircle looped around the β-pinwheel domain of GyrA to form a positively supercoiled crossover (against the negative supercoiling in the minicircle), simultaneously providing both the DNA segment to be cleaved (the G-segment) and the DNA segment to be transported (the transport- or “T-” segment) through the cleaved G-segment. The second reconstructed density map (Fig. 1B) showed gyrase bound only to a 30 bp G-segment of the minicircle²⁰, which we call the “not-wrapped structure.”

Fig. 1. Base identification of negatively supercoiled DNA minicircle sites bound to gyrase.

The cryoEM reconstruction of the DNA-wrapped gyrase (A) or the not-wrapped gyrase (B) is shown with the protein colored in gray and the DNA colored in pink. In both (A) and (B) a zoomed-in view of an extracted base pair from the G-segment DNA (bound to the gyrase) is shown. All nucleotides bound to gyrase at the G-segment were computationally extracted (pink). Shown in (C), individual bases were compared amongst each other and against purine and pyrimidine simulated density maps (~3 Å resolution, blue). The sorted bases were assigned as purines (R), pyrimidines (Y), or undefined (X). From the sorting results and enforcing standard base pairing rules, a R/Y/X profile was constructed for the G-segment in the wrapped and not-wrapped gyrase complexes, shown in (D) and (E), respectively. These profiles were used to search the converted R/Y profile of the 601 bp minicircle. The pattern seen was unique amongst the entire minicircle. F The blue or green underlined sequence represents the alignment of the profiles for the wrapped structure and not-wrapped structure, respectively. The bolded residues highlight the palindrome sequence common to both structures.

Several attempts were made previously to model the DNA with existing automated cryoEM modeling software, including using ModelAngelo²¹. The identity of the DNA bases, however, could not be unambiguously assigned from the density²⁰. One speculation to explain this failure was that the data resulted from an averaging of gyrase bound to multiple different sites along the minicircle²⁰. Therefore, like for other recent protein-DNA complexes imaged by cryoEM, such as condensin²², the origin recognition complex²³, MukBEF²⁴, and FtsK²⁵, the DNA in the two gyrase structures was modeled as a generic adenine-thymine repeat²⁰.

Although gyrase must be able to bind and act at many different DNA sequences to perform its essential roles, only certain base-pair steps^26,27 should be flexible enough to accommodate the sharp (~70°) bend in the G-segment and to completely wrap the β-pinwheel²⁰. To determine which sequence(s) in the 601 bp minicircle were bound to gyrase, we developed and utilized an ad hoc shape recognition and correlation methodology to classify individual DNA bases as purine (R)- or pyrimidine (Y)-like from the cryoEM density maps (Fig. 1C). The resulting R-Y profile was then easily mapped to the known DNA minicircle sequence. In both structures, the G-segment bound by gyrase mapped to a 7 bp overlapping region of the minicircle containing a 6 bp palindrome. In each structure, the DNA, relative to gyrase, was in opposite orientations.

We additionally developed a DNA sequence analysis tool to assess DNA sequence-dependent deformability^26,27, which revealed that both DNA G-segments and the sequence that wrapped the β-pinwheel domain of gyrase had above average base pairstep-dependent deformability. The sequences flanking the G-segment in the not-wrapped structure had much lower than average sequence-dependent deformability. Additionally, analysis of the DNA structures using Web 3DNA 2.0 (3DNA/DSSR)²⁸ revealed several base pairs distorted from B-form and significant conformational transitions in the gyrase-bound DNA double helix. These findings suggest that DNA sequence and deformability are key players in the molecular mechanism of DNA gyrase.

Results

Existing tools failed to reveal DNA sequence

To analyze and determine which DNA sequence(s) E. coli gyrase bound on the 601 bp negatively supercoiled minicircle, our initial efforts utilized available cryoEM modeling software, including Phenix²⁹ and CryoREAD³⁰. The map_to_model procedure³¹ in Phenix was unable to produce a reasonable model for the well-resolved portions of the DNA, which prevented successful mapping of the DNA back to the known minicircle sequence. Similarly, CryoREAD, a deep-learning-based nucleotide modeling tool for cryoEM density maps at resolutions of 10 Å or better, produced models with incorrect base pairing, incompatible chain directions, and broken DNA backbones (Supplementary Fig. 1). Because the training sets for CryoREAD only contained linear (relaxed) DNA and previous cryoEM data rarely includes supercoiled DNA, it is possible that the supercoiling and looping in our densities prevented success of existing methods.

To derive the DNA sequence bound to the gyrase G-segment, we also tried to computationally extract, sort, and annotate the density using simple fitting and scoring techniques (cross correlation, principal component analysis, and Fourier shell correlation). The resulting analyses proved to be inadequate for discriminating among bases, as the computed scores were largely influenced by the sugar-phosphate backbone, which was generally better resolved than the bases.

Shapes of DNA purines and pyrimidines densities differ

For over two decades, shape-based methods have been used to identify structural features and build models for proteins in subnanometer resolution cryoEM density maps^32–40. We reasoned that this approach should work for distinguishing DNA bases as well. Therefore, we examined individually extracted nucleotides to attempt to classify them by shape similarity. We generated reference density maps of purines and pyrimidines from DNA structures found in the Protein Data Bank (PDB) at similar resolutions and noticed consistent size and geometrical features—purines are wider and have a narrower groove than pyrimidines—that separated nucleotides into distinct classes (Fig. 1C). In comparing these base features from the references to the extracted nucleotides from the gyrase maps, it was possible to visually sort and classify the nucleotide density into three distinct classes: purine-like (R), pyrimidine-like (Y), or undetermined (X) density. Moreover, as we examined the bases independently (focusing on the bases in relation to the sugars and phosphates instead of focusing on the sugars and phosphates), we observed that the shape of each class does not change with DNA bending or distortion. As such, any bending or distortion of supercoiled and looped DNA, like that found in our minicircle DNA bound to gyrase, should not hinder the analysis.

Base Hunting

Guided by the observation that we could sort purine and pyrimidine nucleotides based on base size and shape, we instituted a visual, ad hoc approach for nucleotide identification, which we termed “Base Hunting.” We extracted individual nucleotides from the density maps, aligned them to a common template, and visually assigned each density to one of the three classes (R, Y, or X) based on our reference structures and without regard to base pairing. Once we assigned all bases, we then cross-validated that each base pair contained one purine and one pyrimidine. This approach resulted in a simple pattern for each DNA strand, which was a series of consecutive annotations—R, Y, or X—teased directly from the density maps. It is important to note that through this step, the known DNA minicircle sequence was not considered.

To test the capability of this Base Hunting method, we applied the approach to 13 different cryoEM DNA–protein complexes reconstructed to near-atomic resolutions, with reported resolutions of 2.2 to 3.1 Å, which included a separate E. coli gyrase–DNA holocomplex that utilized linear DNA⁴¹ (Table 1). The corresponding models for these cases contained fully annotated DNA, providing a direct assessment of our shape-based assignments. Across these datasets, the Base Hunting method achieved a mean classification accuracy of ~77% (combined average for both an experienced structural biologist and a novice) (Table 1). Particularly relevant to our goal, we accurately assigned 32 of 35 nucleotide pairs (>91%) in the E. coli gyrase-linear DNA⁴¹ structure. In general, Base Hunting accuracy correlated with the resolvability of the DNA locally rather than the resolution of the reported final structure.

Table 1.

ad hoc base identification as purine or pyrimidine by a novice or an expert structural biologist of protein-DNA structures^* deposited in the EMDB

		Novice		Expert
EMDB	Protein	# base pairs tested	% correct	# base pairs tested	% correct
19618	MCM2-7	15	80	19	79
36391	Nucleosome	25	72	140	90
51228	TFIIIC TauB	25	88	28	93
43114	HIV-1 RT	13	85	14	86
42887	Herpes virus 1 polymerase	12	83	14	93
44328	Mce3R	34	74	40	75
45194	MIP1	10	90	11	91
61624	ISW1	104	63	108	77
62196	Monkeypox DNA polymerase	10	70	11	73
50027	BrxX methyltransferase	12	50	12	75
17845	SduA	17	65	10	80
52646	RAD51	10	70	20	70
51222	Gyrase	31	58	35	91
		318 total	69 av	462 total	84 av

^*With a resolution between 2 and 3.5 Å.

av average.

Having quantified the accuracy of Base Hunting on these test datasets, we applied the method to our minicircle DNA-gyrase complexes²⁰. For this analysis, we identified 20 or 30 bp (novice and experienced structural biologist, respectively) where the local resolution was sufficient to resolve the individual nucleotides in the DNA for both the wrapped and not-wrapped negatively supercoiled minicircle DNA-gyrase structures. We extracted, aligned, and classified the observed bases into the three classes (Fig. 1C) and created a sequence profile directly from the density (Fig. 1D, E). We then converted the known minicircle sequence to an R/Y string and compared it to the profile we had generated using Base Hunting (Fig. 1F). Finally, we substituted the known minicircle sequence into the matched profile to reveal the sequence in the maps.

In the model of the negatively supercoiled minicircle positively wrapped around the β-pinwheel of gyrase (Fig. 2A–C), a DNA density sequence profile was generated for the well-resolved G-segment DNA corresponding to positions 92–115 in Chain E and positions 4–27 in Chain F of the previously deposited model (PDBID:8QDX; we maintained this chain/numbering nomenclature in our newly deposited structures). 22 of the 24 bases from our R/Y/X DNA density pattern matched a pattern of bases at positions 112–135 (as assigned in ref. ²⁰) in the minicircle DNA (Table 2). Density corresponding to base pairs at positions 122 and 130 were indistinguishable (X). Nonetheless, with this alignment we were able to reliably extend the DNA sequence to structure assignment to cover base pairs 109–138 of the minicircle DNA at the G-segment. These extended base pairs were not considered in the initial Base Hunting analysis as they were not as well resolved as the other bases within the G-segment.

Fig. 2. Modeling the G-segments in negatively supercoiled DNA minicircle.

From the R/Y profile alignments, a complete sequence model was constructed for the G-segment DNA in the wrapped and not-wrapped complexes shown in (A) and (D), respectively. Zoomed-in views of a single base pair fit to the density are boxed; green is guanine, red is adenine, blue is thymine, and yellow is cytosine. This color scheme is used throughout the manuscript. In (B) and (E), models for the G-segments for the wrapped and not-wrapped structures are shown. The extracted models were rotated such that the palindrome sequence is displayed in the upper left. The rotation is indicated by a tracing of a geodesic arc corresponding to a ~109.5 ° angular displacement (solid angle = ~3.82 steradians). The direction of the G-segment DNA containing the GAATTC palindrome relative to β-pinwheel (depicted as a light gray polygon) is indicated by the arrows (blue arrow denotes the wrapped and green arrow the not-wrapped G-segments). The sequence corresponding to the modeled G-segment of the wrapped and not-wrapped is shown in (C) and (F), respectively. The underlined regions indicate the palindrome. In (G), a zoomed-in view of the palindrome (indicated by a dotted blue box in (A, B, D, and E) and its position in the G-segment relative to the wrapped (left) and not-wrapped (right) gyrase complexes are shown.

Table 2.

Purine/pyrimidine assignment derived from the density map of the well-resolved region of the DNA G-segment in the gyrase map with DNA wrapped around the β-pinwheel. After alignment to the minicircle sequence, differences in the alignment are in bold; bases where the density could not resolve the assignment are in italics

Reference Position (Chain E)	Base Assignment (U/Y)	Minicircle Pattern (U/Y)		Base Assignment (U/Y)	Reference Position (Chain F)
SSS92	R	R	Y	Y	27
93	Y	Y	R	R	26
94	Y	Y	R	R	25
95	R	R	Y	Y	24
96	Y	Y	R	R	23
97	R	R	Y	Y	22
98	R	R	Y	Y	21
99	R	R	Y	Y	20
100	R	R	Y	Y	19
101	Y	Y	R	R	18
102	Y	R	Y	R	17
103	R	R	Y	Y	16
104	Y	Y	R	R	15
105	R	R	Y	Y	14
106	R	R	Y	Y	13
107	Y	Y	R	R	12
108	Y	Y	R	R	11
109	R	R	Y	Y	10
110	X	R	Y	X	9
111	Y	Y	R	R	8
112	R	R	Y	Y	7
113	R	R	Y	Y	6
114	R	R	Y	Y	5
115	Y	Y	R	R	4

For the not-wrapped structure (Fig. 2D–F), our shape modeling yielded a clear assignment for 28 of the 30 bp, corresponding to base positions 132–161 of the minicircle DNA (Table 3); positions 10 and 30 in Chain E (and positions 20 and 1 in Chain F, respectively, in PDB ID: 8QQS) could not be assigned directly (X) because the local resolution was not sufficient.

Table 3.

Purine/pyrimidine assignment derived from the density map of the well-resolved region of the DNA G-segment in the gyrase map without the β-pinwheel

Reference Position (Chain E)	Base Assignment (U/Y)	Minicircle Pattern (U/Y)		Base Assignment (U/Y)	Reference Position (Chain F)
1	R	R	Y	Y	30
2	R	R	Y	Y	29
3	R	R	Y	Y	28
4	Y	Y	R	R	27
5	Y	Y	R	R	26
6	Y	Y	R	R	25
7	R	R	Y	Y	24
8	R	R	Y	Y	23
9	R	R	Y	Y	22
10	Y	Y	R	R	21
11	Y	Y	R	R	20
12	Y	Y	R	R	19
13	R	R	Y	Y	18
14	R	R	Y	Y	17
15	Y	Y	R	R	16
16	R	R	Y	Y	15
17	Y	Y	R	R	14
18	Y	Y	R	R	13
19	R	R	Y	Y	12
20	R	R	Y	Y	11
21	R	R	Y	Y	10
22	R	R	Y	Y	9
23	R	R	Y	Y	8
24	R	R	Y	Y	7
25	R	R	Y	Y	6
26	R	Y	R	Y	5
27	R	R	Y	Y	4
28	R	R	Y	Y	3
29	Y	Y	R	R	2
30	R	Y	R	Y	1

After alignment to the minicircle sequence, differences in the alignment are in bold.

A simple binomial approximation, treating each R/Y assignment as a 50/50 independent choice, yields probabilities of correctly matching the bases by random chance on the order of 10⁻⁵ to 10⁻⁷ for the wrapped and not-wrapped structures, respectively. While this approximation is a heuristic rather than a rigorous statistical model, it illustrates that the observed correspondence between the density-derived and reference sequences is highly unlikely to occur by chance. Combined with an average base-type classification accuracy of approximately 77% across the 13 additional datasets (Table 1), the low random-match probability and consistent classification accuracy support the reliability of our sequence assignments and reinforce confidence in the overall analysis.

Gyrase binds a short palindrome from both sides of the DNA helix

We used standard model building and refinement techniques to replace the A/T DNA segments modeled previously²⁰ with our derived DNA sequences (Fig. 2, Supplementary Table 1). Despite being in opposite orientation and containing different bases, the overall structure of the final DNA models of both G-segments is nearly identical (DNA backbone RMSD < 1 Å) (Fig. 2, Supplementary Fig. 2). We see that the G-segment DNA is bent and unwound similar to what was seen in complexes with the TATA-box binding protein^42,43.

There is a 7 bp overlap between the two G-segment structures that resides in the same position relative to gyrase—the “front” of the DNA-binding domains of GyrA/B (Fig. 2G, Supplementary Fig. 2). This overlap includes the palindromic sequence 5’-GAATTC-3’ (Fig. 2B,E, Supplementary Fig. 2), although the directionality of the overall DNA is different in the two structures. As such, the palindrome exists in the same location on gyrase and with similar DNA-protein contacts regardless of the relative direction of the DNA. Gyrase has been previously reported to bind to repeat palindromic sequences⁴⁴ but never, that we can find, to a single palindrome. The palindrome may be a key determinant for gyrase binding or simply coincidental.

Mapping mutant gyrase binding sites on 600 bp negatively supercoiled minicircle

To verify the Base Hunting results, we incubated gyrase-DNA complexes with micrococcal nuclease followed by restriction endonuclease digestion to generally map where gyrase binds on minicircle DNA (Fig. 3A, Supplementary Fig. 3A). In the absence of gyrase, the DNA was completely degraded by micrococcal nuclease; DNA bound to gyrase was protected. Whereas analysis of the cryoEM structures revealed stable gyrase binding sites permitted under grid-attachment constraints, this biochemical mapping reflects all possible binding sites available to gyrase on the minicircle free in solution. Mutant gyrase (GyrA^(Y122F)₂GyrB₂; the same as in our cryoEM²⁰) was incubated with negatively supercoiled (ΔLk = −2) 600 bp minicircles. The 600 bp minicircles lack the biotinylated dT base but are otherwise identical to the 601 bp minicircles used previously²⁰. A large excess of gyrase was used to replicate the previous conditions²⁰. The gyrase-minicircle DNA complex was then treated with micrococcal nuclease which generated three major DNA fragments (~91 bp, ~127 bp, and ~151 bp) and at least one minor species (~47 bp)(Fig. 3B, C). What these DNA lengths correspond to is not discernable by this method but can be speculated to be gyrase bound to a G-segment, or a G-segment and the DNA wrapped around one or two gyrase β-pinwheels (although we did not observe two β-pinwheels²⁰).

Fig. 3. Restriction enzyme digestion confirms gyrase binding at Base Hunting–identified sites.

Gyrase-protected DNA fragments generated by micrococcal nuclease digestion were analyzed by restriction enzyme cleavage. Cleavage by the enzymes shown in (A) equal cleavage at the sequences identified by Base Hunting. (B) The Agilent TapeStation gel shows DNA cleavage products after incubation of the gyrase-protected DNA with different restriction enzymes (listed above each lane). C Electropherograms from the gel shown in (B) for micrococcal nuclease alone (left top), or further digested with Tsp45I, EcoRI, or KpnI, as labeled. Supplementary Data 1 contain all these data. Source data are provided as a Source Data file.

The short palindromic sequence 5’-GAATTC-3’ bound by gyrase in both directions is a recognition site cleaved by the restriction enzyme EcoRI (Fig. 3A, Supplementary Fig. 3B). Because EcoRI cleaved the DNA fragments remaining after micrococcal nuclease digestion, we can confirm that gyrase indeed bound at that palindrome. KpnI cleavage indicates gyrase binding the G-segment of the not-wrapped structure. The lengths of the KpnI-cleaved DNA fragments were longer than expected for gyrase binding to just the G-segment, indicating gyrase protected additional DNA from micrococcal nuclease. Tsp45I cleavage shows gyrase binding in the DNA wrapped about the β-pinwheel (Fig. 3A, Supplementary Fig. 3B). Together, these data show that mutant gyrase binds at the sequences identified by Base Hunting.

All tested restriction enzymes cleaved the micrococcal nuclease products indicating that gyrase bound to additional sites on the 600 bp negatively supercoiled minicircle (Fig. 3). No enzyme cleaved to completion suggesting that no site is completely occupied by gyrase. Two main sites protected by gyrase include the HindIII and the BsaHI site (Fig. 3). HindIII cleaves 11 bp away from the location of the biotinylated base used to affix the DNA to a streptavidin grid²⁰, which likely prevented gyrase binding. The BsaHI site is located diametrically opposite the EcoRI site palindrome.

In addition to the extra biotinylated base and its attachment to the grid, there were other differences between the cryoEM experiments and the biochemical ones. AMP-PNP was added for the cryoEM experiments but not for the mapping. Washes done to remove unbound or weakly bound gyrase for the cryoEM experiments were not done for these biochemical experiments. Therefore, multiple gyrases with relatively loose binding to the supercoiled minicircles might remain bound during the micrococcal nuclease digestion. A 600 bp minicircle is long enough to simultaneously accommodate more than one enzyme, and thus binding at one site would not preclude binding to another. Additionally, the biotinylated dT base is extrahelical (a “T-bulge”) because there is no nucleotide on the opposite strand. This T-bulge may impart a small bend in the DNA or increase localized flexibility, potentially modulating where gyrase binds.

Structural interpretation of the gyrase DNA cleavage motif

Topo-Seq^45,46, a DNA sequencing approach, provides precise single-nucleotide resolution mapping of topoisomerase cleavage sites. Analysis of the thousands of gyrase-mediated DNA cleavage sites observed by Topo-Seq revealed a 130 bp “gyrase motif.” This motif consists of three regions—an initial 47 bp region with periodic distribution of G/C content, a 36 bp central region, and another 47 bp G/C periodic region. The central region was hypothesized to form the G-segment whereas the periodic regions are thought to be more readily wrapped around the β-pinwheels. For the wrapped complex, the adjoining DNA sequence that is wrapped around the β-pinwheel has approximately 50% G/C content, which is evenly distributed. For the not-wrapped complex, the sequence flanking the G-segment that did not wrap the β-pinwheel is A/T-rich (>80%) and contains long stretches (up to 17 bp) of only A/T. If DNA wrapping by gyrase is facilitated by phased G/C content^45,47, this difference in G/C content may explain why one DNA sequence wrapped a β-pinwheel and the other one did not in our gyrase structures²⁰.

In the wrapped structure²⁰, the DNA wraps 49 bp around the β-pinwheel and an additional contiguous 35 bp DNA forms the G-segment. In a study that utilized a linear DNA molecule containing a naturally occurring strong gyrase site, Mu SGS, a gyrase-DNA structure was obtained in which both β-pinwheels were wrapped⁴¹ (Supplementary Fig. 4). Mu SGS closely matches the Topo-Seq-determined gyrase cleavage motif, and contains two periodic regions, therefore allowing wrapping about both β-pinwheels.

There were other striking structural deviations in the DNA wrapped around the β-pinwheel in our structure that may help determine sequence preference for gyrase. Stretches of localized DNA unwinding, from B- to A- and TA-forms, were identified (Figs. 4,5). GC-rich sequences have a propensity to undergo B-A transitions⁴⁸ and exhibit more A-like character, e.g., lower intrinsic twist, in high-resolution structures²⁷. Thus, a previously unrecognized function of the G/C periodicity observed in the gyrase motif⁴⁵ might be to enable these transitions. These contacts of gyrase with DNA on the β-pinwheel are much like those of the histone proteins of DNA on the nucleosome⁴⁹, forming close associations with the sugar-phosphate backbone at regularly spaced intervals rather than with specific DNA bases. The handedness of the wrapped DNA reflects the positions of the contacted amino acid atoms on the protein, yielding an overtwisted (overwound) left-handed pathway on the nucleosome as compared to the undertwisted (underwound) right-handed pathway of DNA around the β-pinwheel.

Fig. 4. DNA distortions in the wrapped gyrase complex.

In (A–C), DNA structural distortions and protein contacts are shown in the wrapped gyrase-DNA structure. The left panels show per-base interactions between the leading and complementary DNA strands, highlighting regions classified as A-form (gray), TA-form (black), or locally melted (gold). The right panels map these non-B DNA features onto the corresponding portion of the model to visualize their spatial organization along the duplex DNA. The contacts and deformations of DNA in the wrapped Mu SGS DNA-gyrase complex⁴¹ are reported in Supplementary Fig. 4. In (D), global geometrical features of the DNA, including the backbone centerline curvature (top panel), base-plane rotation between adjacent base pairs (middle panel), and cross-strand phosphate-phosphate (P…P) distances computed between antiparallel strands (bottom panel), are plotted. Color-coded boxes corresponding to the non-B form DNA, as in (A–C), are superimposed on the plots to illustrate the correspondence of these structural features. Supplementary Data 2 contain Web 3DNA 2.0 (3DNA/DSSR) output²⁸. Source data are provided as a Source Data file.

Fig. 5. DNA distortions in the not-wrapped gyrase complex.

In (A), DNA structural distortions and protein contacts are shown for the not-wrapped gyrase-DNA structure. The left panels show per-base interactions between the leading and complementary DNA strands, highlighting regions classified as A-form (gray), TA-form (black), or locally melted (gold). The right panels map these non-B DNA features onto the corresponding portion of the model to visualize their spatial organization along the duplex DNA. In (B), global geometrical features of the DNA, including the backbone centerline curvature (top panel), base-plane rotation between adjacent base pairs (middle panel), and cross-strand phosphate-phosphate (P…P) distances computed between antiparallel strands (bottom panel), are plotted. Color-coded boxes corresponding to the non-B form DNA, as in (A–C), are superimposed on the plots to illustrate the correspondence of these structural features. Supplementary Data 3 contain Web 3DNA 2.0 (3DNA/DSSR) output²⁸. Source data are provided as a Source Data file.

Comparison of the identified gyrase binding sites to the position weight matrix of the gyrase cleavage motif^45,46 gives a numerical score that describes how well a sequence fits the motif, with a positive score indicating a fit. Considering the entire 130 bp gyrase motif identified by Topo-Seq, we found no positive scores for where gyrase would cleave (within the positions located around the two phenylalanines that replaced the catalytic tyrosines—GyrA^Y122F) for either our wrapped or not-wrapped DNA-gyrase structures. When comparing only the DNA sequences bound to gyrase in our structures (one β-pinwheel and the G-segment for the wrapped structure and the G-segment for the not-wrapped structure) to the corresponding subsets of the Topo-Seq gyrase motif, however, there were several positive (though low) scores for the positions near the phenylalanines (Supplementary Data 4–10). Considering that there were very few positively scoring positions on the minicircle, we calculated the percentile rankings of the gyrase binding site motif scores compared to those of our entire DNA minicircle. This comparison revealed that whereas the wrapped structure sequence does not have a positive score, it does have a relatively high score compared to the rest of the minicircle sequence, with a motif score in the 89^th percentile (Supplementary Data 4–10). The DNA sequence in the not-wrapped structure, however, does not have a highly ranking position where the phenylalanines are located, falling in the 50^th percentile (Supplementary Data 4–10). Therefore, this motif analysis, given the paucity of data we have, fails to fully explain why gyrase binds the two sites on the 601 bp negatively supercoiled minicircle.

Importance of DNA flexibility for gyrase binding

It is well known that sequences that are more flexible and deformable are more readily cyclized^50,51. Intrinsic sequence-dependent cyclizability of DNA correlates with gyrase-DNA binding affinity⁵². DNA sequence strongly dictates which DNA segments can be deformed, bent, compressed, and moved as each base-pair step has an inherent deformability that is also influenced by the two base pairs around the base-pair step^26,27. A measure of this inherent sequence-dependent deformability has been determined through the amount of conformational space a base-pair step of DNA occupies on average within protein-DNA crystal structures, which is represented by the six rigid-body parameters describing inter-base-pair orientation (Tilt, Roll, Twist) and displacement (Shift, Slide, Rise)⁵³. The idea is that base-pair steps that occupy more space, on average, do so because they are inherently easier to deform into a wider range of shapes^26,27. These average volumes, or deformability values, have been determined for all 136 unique tetrameric sequences of DNA^26,27 and can be used to assess the tendency for sequences of DNA to accommodate protein binding. Dissimilar sequences can have similar deformability, and similar sequences can have vastly different deformability. It seems reasonable that not every DNA sequence can conform to the bending requirements (~70°) for the G-segment binding to gyrase. It seems even less likely that just any DNA sequence could accommodate the tight wrap around the β-pinwheel. Because the distance (~30 Å) between the end of the gyrase-bound G-segment and the start of the β-pinwheel wrap is so short (~9 bp), these two segments must be contiguous. Therefore, a DNA site simultaneously must have a bendable G-segment and a contiguous segment deformable enough to wrap the β-pinwheel.

To identify flexible regions within the minicircle, we employed a sliding-window analysis of DNA sequence-dependent deformability^26,27 based on base-pair step parameters. In this approach, a window of fixed length k (i.e., a k-mer) is passed across the sequence to read the deformability values of the base-pair steps contained within and calculate the average deformability score of all steps within that k-mer (see Supplementary Fig. 5A), which is somewhat analogous to classical hydropathy plots used in protein sequence analysis⁵⁴. This k-mer deformability score reflects how intrinsically deformable the base pairs within the segment are based on their local sequence composition, with higher scores indicating greater capacity for bending, unwinding, or wrapping. It is important to recognize that these deformability scores are not the actual deformability of a k-mer of DNA but instead are a description of the average deformability of the individual base-pair steps within the k-mer. It is unclear how the deformability of a DNA sequence changes as the sequence grows, but certainly this averaging dampens the extremes on both ends of the scale. Considering this dampening, and that we observed the single-site bound in the not-wrapped structure, we chose to analyze the deformability of the relevant regions independently rather than with one unified score.

Applying our k-mer averaging method to the 601 bp minicircle (Fig. 6) revealed that the deformability score of the base-pair steps within the 30 bp G-segment of the not-wrapped structure is in the 98th percentile of all possible 30-mers in the 601 bp minicircle (Fig. 6B). Additionally, the center of the G-segment of the not-wrapped structure contains the single most deformable base-pair step—TA within the tetramer GTAC. These highly flexible pyrimidine-purine steps are seen as sharply bending hinges at DNA-protein binding sites²⁶.

Fig. 6. Minicircle DNA sequence-dependent deformability.

A The sequence-dependent deformability value of each base-pair step in the first 300 bp of the 601 bp minicircle DNA, centered on the overlap of the two gyrase-bound G-segments (blue is wrapped and green is not-wrapped). The sequence of the DNA wrapped around the β-pinwheel is colored in orange and the sequence in the minicircle that did not wrap around the β-pinwheel is colored in red. The direction of the DNA relative to gyrase is indicated by the arrows (blue arrow is wrapped and green arrow is not-wrapped G-segments). The baseline of this bar plot was chosen to be 3.37 as this is the average deformability of all base-pair steps in the 601 bp minicircle sequence. B The sequence-dependent deformability score of a sliding k-mer (k = 35) along the entire 601-bp minicircle is shown. Each colored bp at the center of the 35-mer represents the deformability score of that bp and the ±17 bp surrounding it. The labels within the circular heatmap have the same color scheme as in (A). In yellow are the highest averaged k-mer deformability scores (~5.0), and the blue represents the lowest averaged k-mer deformability scores (~2.0). The purple arrows point to the location of the biotinylated thymine that was used to fix the particles to the grid during cryoEM. The lighter blue arrows point to the location of the GAATTC palindrome present (in opposite orientations) in both the wrapped and not-wrapped structures. The circular heatmap was created with the Python library pycirclize¹⁰⁰. Supplementary Data 4–10 contain these deformability data. Source data are provided as a Source Data file.

Contrasted to the high flexibility of the G-segment in the not-wrapped structure, the adjacent region of the minicircle that might have wrapped the β-pinwheel has a low 49-mer deformability score, falling in 2.6th percentile of all 49-mers in the minicircle sequence. This rigidity, in the region of the minicircle that would be expected to wrap the β-pinwheel of gyrase, provides an explanation as to why this sequence did not wrap. In the wrapped structure, the average deformability of the 35 bp that form the G-segment falls in the 53rd percentile by deformability score of all 35-mers in the minicircle whereas the sequence in the wrap is in the 87th percentile by average deformability score (Fig. 6B).

We assessed the deformability scores of known strong gyrase sites (SGS) to compare to our gyrase structures. We analyzed SGS from pSC101, pBR322, and Mu⁵⁵, the Topo-Seq gyrase motif consensus sequence⁴⁵, and a DNA fragment containing multiple REP sequences, all of which have been demonstrated to be cleaved preferentially by gyrase⁴⁴. For each SGS, we analyzed the 35-mer deformability score immediately around the cleavage position (the G-segment) and the 49-mer deformability scores of the sequences flanking the G-segment (the wrap candidates). The 35-mer centered on the pSC101 cleavage site is highly deformable, with a deformability score falling in the 82nd percentile of 35-mers on pSC101 and would be in the 71st percentile of 35-mers on the 601 bp minicircle (Supplementary Fig. 6, Supplementary Data 4,4–10). One of the two candidate DNA β-pinwheel wraps in the pSC101 SGS is of average deformability; its 49-mer score falls in the 55th percentile of 49-mers on pSC101 and would be in the 42nd percentile of 49-mers in the 601 bp minicircle. The other candidate β-pinwheel DNA wrap, however, is very highly deformable, falling in the 97th percentile of 49-mers in pSC101, and would be in the 92nd percentile of 49-mers in the 601 bp minicircle (Supplementary Fig. 7, Supplementary Data 4–10).

In addition to the two sites bound by gyrase in the cryoEM structures, two more major gyrase binding sites were identified by the micrococcal nuclease protection and restriction enzyme mapping experiments described above (Fig. 3). Our deformability analysis (Fig. 6B, Supplementary Fig. 5C) aligned to the restriction map (Supplementary Fig. 3B) shows that the DNA centered on the HindIII cleavage site is of low deformability score (~20th percentile for both 35-mer and 49-mer score). The DNA centered on the BsaHI cleavage site is of moderate deformability score, falling around the 70th percentile for both 35-mer and 49-mer score. However, these are the deformability scores of the k-mers centered on the cleavage sites, which do not necessarily represent meaningful or cohesive DNA segments with respect to gyrase binding. The BsaHI and HindIII sites are within the region protected by gyrase from micrococcal nuclease digestion, but beyond that it is impossible to determine where they fall in relation to the G-segment, β-pinwheel wrap, or T-segment of gyrase. Whereas DNA deformability clearly influences gyrase binding site preference, other factors may also be important and should be investigated further.

The 35-mer centered on the pBR322 SGS cleavage site falls in the 31st percentile of pBR322 and would be in the 35th percentile of the 601 bp minicircle (Supplementary Fig. 6, Supplementary Data 4–10). Additionally, both candidate β-pinwheel wraps for the pBR322 SGS are also of medium and low deformability, falling in the 54th and 16th percentiles of 49-mers in pBR322, and would be in the 50th and 23rd percentiles of 49-mers on the 601 bp minicircle (Supplementary Fig. 7, Supplementary Data 4–10). The 35-mer centered on the Mu SGS cleavage site falls in the 0.6th percentile of 35-mers in the entire Mu genome and is lower than all 35-mers in the 601 bp minicircle (Supplementary Fig. 6, Supplementary Data 4–10). The 49-mer deformability score for one of the two candidate wraps—the left arm of the Mu SGS⁴⁷—falls in the 1.6th percentile of the entire Mu genome and would fall in the 0.3rd percentile of the 601 bp minicircle. But the other candidate wrap—the Mu SGS right arm—is of high deformability, falling in the 80th percentile of 49-mers in the entire Mu genome, and would fall in the 70th percentile of 49-mers in the 601 bp minicircle (Supplementary Fig. 7, Supplementary Data 4–10). Notably, the removal of the right arm of the Mu SGS DNA sequence eliminates the preferential activity of gyrase on the Mu SGS whereas removal of the left arm has no effect⁴⁷. Within the Mu SGS structure of gyrase, it is the right arm that bears a T-segment of DNA in position for strand passage, not the left arm⁴¹.

The Topo-Seq gyrase motif consensus sequence^45,46 is highly deformable, and the 35-mer centered on the cleavage site would be in the 80th percentile of 35-mers on the 601 bp minicircle (Supplementary Fig. 6, Supplementary Data 4–10). One of the candidate β-pinwheel DNA wraps of the motif consensus sequence has a 49-mer deformability score higher than any on the 601 bp minicircle, and the other candidate wrap would be in the 94th percentile of the 601 bp minicircle (Supplementary Fig. 7, Supplementary Data 4–10). Unlike the other SGSs, the REP-containing fragment of DNA investigated by Yang and Ames⁴⁴ contains several unique cleavage sites that can be broadly split into four different loci, although these sites were not mapped to single nucleotide resolution. Of these loci, two cleavage sites—including the major cleavage site—are located within regions of low/moderate deformability by 35-mer deformability score but have at least one candidate wrap extrapolated to a region of very high deformability ( > 85th percentile on the 601 bp minicircle). The other two cleavage sites are located within regions of high 35-mer deformability score ( > 80th percentile on the 601 bp minicircle) but do not possess a candidate wrap of high deformability (Supplementary Figs. 6,7, Supplementary Data 4–10).

The results of the analysis of strong gyrase sites and the two sites identified through Base Hunting on the 601 bp minicircle together suggest that gyrase may exhibit two distinct DNA sequence preferences—high G-segment bendability or high flexibility of at least one candidate wrapping sequence. We suggest that the wrapped and not-wrapped structures represent these two preferences by gyrase.

Quantitative analysis of DNA deformation and protein-DNA contacts

The binding of proteins to DNA is known to result in conformational transitions of the double helix at individual base-pair steps^56,57. To relate these distortions to protein contacts, we used the DSSR-SNAP module to count residue-base interactions at each position⁵⁸. The bar plots in Fig. 4A–C, Fig. 5A (left) summarize these contacts for the T-segment, β-pinwheel wrap, and G-segment of the wrapped and the G-segment of the not-wrapped, respectively. Both G-segment deformations accompany the partial intercalation of isoleucine 174 from the two GyrA subunits at DNA sites separated by 12 bp. Furthermore, the transformation from the B (classically 10 bp/turn but experimentally measured as ~10.45 bp/turn) to the A (~11 bp/turn) form of DNA consists of eight residues in contact with both GyrA and GyrB (Figs. 4, 5), which are shown for the leading (magenta) and complementary (blue) strands. Bases exhibiting A-, TA-, or melted conformations, highlighted in gray, black, and gold, respectively, are mapped onto the DNA structures (right side in Fig. 4A–C and Fig. 5A). This visualization reveals that the GyrA/GyrB interactions tend to cluster where the DNA transitions from canonical B-form to locally A- or TA-like conformations, supporting a direct mechanical coupling between protein binding and DNA deformation.

To contextualize these effects, we measured backbone curvature, base-plane rotation, and cross-strand phosphate distances from the refined cryoEM models (Supplementary Table 3). Base-plane rotation corresponds to the angle between the normals of paired bases and is closely related to Propeller Twist and Buckle, two of the rigid-body parameters between paired bases determined by Web 3DNA (3DNA-DSSR)²⁸. Curvature captures global duplex bending whereas the base-plane rotation reflects local distortions of base pairs that accompany the introduction of A/TA helical features. The cross-strand phosphate distance score reflects an intermediate measure of DNA structural consistency. Together, these values provide a domain-level, structural snapshot of the DNA associated with gyrase.

The β-pinwheel wrap region shows mean curvature of 11.9 ± 4.1° per base pair and base-plane rotation of 13.9 ± 6.2° per base pair, approximately twice those of the adjacent G-segment (6.1 ± 3.0°, 6.8 ± 3.2°) and T-segment (5.2 ± 2.8°, 5.7 ± 2.6°). Cross-strand phosphate to phosphate distances across the G-segment average 19.6 ± 0.7 Å but display local minima coinciding with curvature peaks. These quantitative measurements confirm that gyrase engagement stabilizes reproducible bending and minor-groove compression characteristic of A/TA-like geometry rather than nonspecific DNA bending (Supplementary Table 3).

Comparative analysis of DNA deformations across the wrapped, not-wrapped, and Mu SGS (with both β-pinwheels wrapped by DNA) gyrase-DNA complexes reveal a strikingly conserved spatial pattern of local bending and non-B DNA conformations (Supplementary Fig. 4B–D). In all three structures, regions of elevated curvature and A/TA-like geometry occur at nearly identical positions along the duplex, particularly within the G-segment, where base-step roll and purine/pyrimidine periodicity align with the A-form region (Supplementary Fig. 4E, F). This correspondence indicates that the deformation signature observed in our minicircle structures is not unique to the negatively supercoiled substrate but reflects an intrinsic property of gyrase-DNA engagement. Nonetheless, the Mu SGS structures also display a slightly reduced extent of non-B DNA, consistent with linear DNA lacking the torsional strain present in the negatively supercoiled minicircles. These comparisons suggest that the underlying pattern of gyrase-induced or stabilized deformation is mechanistically conserved across linear and negatively supercoiled minicircles.

Taken together, these analyses reveal that gyrase engages DNA at a consistent set of mechanically privileged positions, independent of the specific DNA topology examined. The convergence of A/TA-like geometries, clustered GyrA/GyrB contacts, and groove modulation marks the structural sites of DNA duplex capture and wrapping. These deformations likely arise from intrinsic differences in local DNA flexibility afforded by specific sequences and their potential amplification by negative supercoiling could facilitate DNA strand passage and conformational coupling within the gyrase catalytic cycle. The recurrence of DNA deformation-specific features in both gyrase and Top6 complexes⁵⁹ suggests that DNA deformability provides a conserved physical basis for sequence-dependent recognition across type II topoisomerases.

Discussion

In this work, we developed and utilized a shape-based method to directly determine purine/pyrimidine sequences from cryoEM density maps, a task that was not possible with existing software. Matching these sequences to the known DNA minicircle sequence allowed us to precisely determine the sequence of DNA bound to gyrase and successfully construct more complete gyrase-DNA models. This work sets the stage for developing a robust analytical DNA modeling tool, aligned in capability and approach with existing protein modeling platforms.

Because our method relies solely on the shape of individual bases, it is agnostic to the structural form (B or non-B) or topology of DNA. As noted above, the central section of the G-segment in type II topoisomerase complexes is not B- but A-form^20,60. We were able to correctly determine the DNA sequence of the G-segment, except for two positions (102/122 and 110/130). As observed from structures of several G-segment DNAs bound to type II topoisomerases, these positions are located in A- and TA-form DNAs at the sharpest bend in the segment disturbing base-pairing. This approach may even be suitable for analyzing double-stranded RNA, and perhaps eventually for single-stranded RNA and DNA, although the confidence gained from base pairing rules is an important aspect of the ad hoc method.

A limitation of this procedure is that it requires relatively high resolvability of individual bases. From our analysis of 13 deposited cryoEM structures, we estimate that a local resolution of better than ~3.2 Å is generally required to distinguish purine from pyrimidine densities with confidence. Reported map resolution, however, represents a global measure and does not fully capture the local variation in map quality arising from differences in density contrast, occupancy, or conformational stability. In our gyrase complexes, most bases within the G-segment are clearly resolved and readily classified, but a small subset—primarily those buried deep in the G-segment cleft—lack sufficient density detail to unambiguously define base identity. These ambiguous features may reflect local resolution of the EM map or minor variability in neighboring side-chain positions, rather than global motion or conformational heterogeneity, because adjacent base pairs remain well defined. Similarly, bases located at the periphery of the G-segment binding pocket in the wrapped complex show lower resolvability, consistent with increased local flexibility at the edge of the pocket. Excluding these few uncertain positions represents a conservative treatment of the data and does not alter the overall R/Y pattern or its correspondence with the known minicircle sequence.

Gyrase wraps DNA in a positive wrap to impart directionality on strand passage^20,61. Localized regions of A- and TA-like DNA geometry are observed in both the G-segment and the β-pinwheel-wrapped portions of the wrapped gyrase-minicircle DNA structure. Analysis of the wrapped structure revealed 12 A- and TA-like bases near the bend of the G-segment, within the β-pinwheel, and 4 bases in the T-segment. In contrast, the linear E. coli gyrase-DNA complex of Michalczyk et al.⁴¹ reveals similar A/TA-form transitions in the corresponding positions—11 bases near the bend of the G-segment and 3 bases in the T-segment. The presence of locally underwound A-like DNA within a stretch of B-DNA likely facilitates the right-handed wrapping characteristic in much the same way that the insertion of overwound base-pair steps yields a left-handed superhelix⁵⁷ (Supplementary Fig. 8).

Some of the earliest insights into the mechanistic basis of the sequence preference of DNA gyrase resulted from the discovery of naturally occurring strong gyrase sites. The best studied of these, Mu SGS, is essential for the efficient replication of the bacteriophage Mu⁶². The ability of the Mu SGS to promote gyrase binding and activity was proposed to be related to the phased anisotropic bending sites within the sequence, facilitating DNA wrapping around the GyrA C-terminal domain (CTD)⁴⁷. It should be noted that the 601 bp minicircle sequence we used previously²⁰ did not contain any known strong gyrase sites. Thus, the requisite flexibility may be provided by the negative supercoiling in the minicircle. The subsequent structure of gyrase bound to a linear Mu SGS-containing DNA shows the same features of localized DNA unwinding⁴¹. Thus, sequence and interactions with gyrase both make contributions to DNA wrapping.

The DNA palindrome, GAATTC, common to both the wrapped and not-wrapped gyrase complexes, may be a key determinant for where gyrase binds the 601 bp minicircle. Or this palindrome may be coincidental. The palindromic sequence is recognized by the restriction enzyme EcoRI. Many restriction enzyme sites are palindromic, which allows the dimeric enzymes to bind the two half-sites symmetrically, facilitating sequence recognition⁶³. Restriction sites exhibit very high sequence discrimination with typically 10⁶-fold preference in k_cat/K_m for cleaving their given site compared to other sites differing by a little as 1 bp⁶⁴. This sequence preference is significantly higher than for gyrase, but the molecular mechanism of EcoRI may provide insight into why gyrase binds this palindrome.

EcoRI recognizes its restriction site through both direct and indirect DNA readout⁶⁵. In the crystal structure of EcoRI bound to its cognate sequence, the palindromic DNA is kinked at the center, aiding sequence recognition via indirect readout⁶⁶ and enabling “crosstalk” between the two active sites of the enzyme⁶⁷. Direct readout involves base-specific contacts in the DNA grooves. In gyrase structures, the protein interacts mainly with the sugar-phosphate backbone, which suggests a stronger role for indirect readout by gyrase.

The Drew–Dickerson dodecamer, used for the first B-DNA crystal structure, also contains the EcoRI site, which has contributed to the GAATTC sequence becoming one of the most extensively studied^68–71. Electrophoresis⁷² and NMR⁷³ data indicate that this sequence may be intrinsically curved in solution, although it is not significantly bent in the crystal structure of the DNA alone⁷⁰. It is not clear how a preexisting bend would affect gyrase binding given that the GAATTC palindrome was found at the end of the G-segment, away from where most of the G-segment bending occurs. Intrinsic curvature is, however, localized to superhelical apices⁷⁴, which may facilitate binding by gyrase.

Previous researchers who have investigated sequence specificity of gyrase have analyzed cleavage sites. The relationship among the DNA binding, cleavage, and strand passage activities of type II topoisomerases, however, is not fully understood. For one thing, it is unclear whether more cleavage leads to more strand passage. Strand passage is proposed to involve large scale conformational rearrangements of the protein to allow passage of the T-segment through it, a maneuver involving potentially genotoxic DNA breaks. It has been postulated that the coupling of strand passage with ATP hydrolysis in type II topoisomerases has a genoprotective role that allows the controlled separation of protein-protein interfaces and prevents accidental breaks⁷⁵. Indeed, loss of the ATPase domain of human topoisomerase IIβ leads to increased DNA damage¹⁴, supporting this hypothesis. Clearly, where and how topoisomerases act must be carefully regulated.

It has long been known that gyrase can cleave and religate short duplex DNAs that are not long enough either to wrap a β-pinwheel or to provide a T-segment on the same DNA molecule⁷⁶. Thus, gyrase can cleave the G-segment without DNA being wrapped around its β-pinwheel. Similarly, the catalytically-deficient gyrase mutant binds the G-segment without a DNA wrapped β-pinwheel²⁰. Conversely, the mutant gyrase can wrap a β-pinwheel without cleaving, indicating that cleavage is not prerequisite for wrapping.

Deformability is a probabilistic contributor rather than a deterministic predictor of DNA-protein interactions. Our deformability analysis provides an explanation for why gyrase may sometimes bind and not wrap. If one of the two DNA sequences adjacent to the G-segment is flexible enough to wrap, the T-segment becomes positioned to enable strand passage, as in our wrapped structure. If the adjacent region is not flexible enough, as in our not-wrapped structure, gyrase might dissociate (or slide) and continue a search for a different G-segment with adjacent DNA that would allow the wrapped DNA conformation required for both G- and T-segment binding. In other words, not every DNA sequence adjacent to the G-segment will accommodate the tight positive wrap^26,27.

Surprisingly, Sutormin and coworkers found that the gyrase consensus sequence that they discovered, which comprises two regions assumed to wrap both β-pinwheels, inhibited gyrase activity when incorporated on a plasmid⁴⁵. They hypothesized that the high affinity of gyrase for the consensus sequence may inhibit strand passage. Thus, the enzyme may become stalled at the cleavage step. The complete catalytic cycle of gyrase requires DNA cleavage, wrapping about one β-pinwheel, strand passage, and finally, unwrapping to allow for selection of a new T-segment and dissociation from the DNA. In fact, the gyrase-Mu SGS linear DNA structure⁴¹ shows one β-pinwheel fully wrapped and the other only partially wrapped with fewer contacts with the DNA (Supplementary Fig. 4G). Complete wrapping of the DNA around the second β-pinwheel would result in a steric clash with the T-segment emerging from the first β-pinwheel (see Supplementary Fig. 4D). Thus, in an active gyrase-DNA complex proficient to strand pass, only one β-pinwheel is likely wrapped at a time.

While the cryoEM density map used in this study was from previously published datasets that employed biotinylated minicircle DNA immobilized on streptavidin affinity grids, all biochemical and sequence analyses in this work were performed using unmodified, non-biotinylated DNA substrates. This distinction is important, as the immobilization strategy used for cryoEM was not compatible with the micrococcal nuclease assay. Whereas the presence of a single biotinylated dT base may alter DNA locally, the DNA minicircle sequence, topology, and negative supercoiling level were consistent across experiments. We acknowledge this limitation in experimental design but note that the DNA–gyrase contacts identified in our analysis are consistent with known structural features of the complex and with patterns observed in related topoisomerase work⁵⁹.

Although this study focused on E. coli gyrase, we note that application of the same base-identification and deformability analysis framework to archaeal type IIB topoisomerase VI (Top6) yielded highly similar mechanistic features⁵⁹. In this structure, DNA cleavage occurs at a junction between a highly deformable and a more rigid duplex region. Combined with the analysis of E. coli gyrase-DNA holocomplex with relaxed linear DNA, these results support a DNA deformability-dependent mechanism of sequence selection among distinct topoisomerase families.

Beyond gyrase, the principles of our shape-based base recognition approach should extend to other protein-DNA complexes where local sequence-dependent flexibility plays a mechanistic role. Assemblies such as nucleosomes, transcription factors, recombinases, and polymerase-DNA complexes often rely on local DNA shape and deformability for binding specificity. The Base Hunting framework, by emphasizing geometric and topological features of base identity, thus offers a broadly applicable strategy for resolving DNA sequence and deformability in near-atomic cryoEM maps of diverse nucleoprotein assemblies. In summary, this study provides a comprehensive methodology for interpreting DNA sequence in cryoEM densities, a means to evaluate DNA deformability, and insight into the mystery of how DNA gyrase choreographs its complex essential function.

Methods

Data processing

The density maps and models for the wrapped (EMD-18605, PDB ID: 8QDX) and not-wrapped (EMDB-18603, PDB ID: 8QQS) gyrase were downloaded, segmented, and split into “gyrase” or “DNA” maps using the coordinate data as a reference with the “Color by Zone” tool in ChimeraX (V1.7.1) with a radius of 10 Å. Initial evaluation of the DNA revealed a stretch of well-resolved base pairs corresponding to bases 89–118 on chain E (nomenclature from²⁰) and 1–30 on chain F in the wrapped structure; with bases 92–115 in Chain E (and the corresponding bases in Chain F) having the highest local resolution. Well-resolved DNA density was also observed in the same location in the not-wrapped DNA-gyrase map, corresponding to the entirety of chains E and F of the deposited model (bases 1–30 in both chains). This region was in the DNA-binding domain of GyrA/B and appeared to be the G-segment DNA.

Modeling with Phenix and CryoREAD

While previous modeling tools did not produce reliable DNA models²⁰, we initially attempted to build models for the well-resolved portions of the DNA using Phenix map_to_model and CryoREAD. DNA from the two deposited maps was isolated and segmented using ChimeraX⁷⁷. Density maps containing only the full segmented DNA and a truncated version consisting of only the well-resolved portion of the DNA at the DNA/gyrase interface for both DNA-gyrase reconstructions were submitted to the CryoREAD³⁰ webserver (https://em.kiharalab.org/algorithm/CryoREAD, accessed March/April, 2024). For each submitted map, the resolution parameter was set at 3 Å and CryoREAD was run with and without setting an explicit contour value, and with and without inputting the minicircle DNA sequence. Additionally, Phenix’s map_to_model (Phenix v1.19), which can generate a model without the user providing a protein or DNA sequence, was run with default options from command line on a Linux workstation. The resulting models for both the wrapped and not-wrapped DNA from both modeling tools yielded non-Watson-Crick base pairing, broken chains, directional inconsistencies, failure to fit the DNA model, and incompatibility with the known minicircle DNA sequence (Supplementary Fig. 1).

Base hunting

As CryoREAD, Phenix, and ModelAngelo did not produce meaningful base assignments, we adopted our own identification and modeling strategy that focuses on categorizing base shapes as purine or pyrimidine. Using the DNA models from the deposited structures (AT-only models) as a template, every base in the well-resolved portions of the DNA maps was segmented out using “Color by Zone” (10 Å radius) and aligned to each other with “Fit in Map” in ChimeraX (version 1.71).

Density for the individual bases were then compared visually and separated into three groups: density with an oval-like protrusion emanating directly from the sugar/phosphate backbone, bases containing a large density bending away from the backbone, and an ambiguous density not fitting the loop-like or bent profile. Comparison of the oval-like and bent density profiles to previously determined structures of bases suggested that these two groups belonged to purines and pyrimidines, respectively. These “reference” models were blurred to the approximate resolution of our cryoEM density maps (~3 Å) using e2pdb2mrc.py from EMAN2⁷⁸. Bases belonging to each of these respective classes were labeled as either purine (R), pyrimidine (Y), or undefined (X). Extending the analysis, we then compared bases within a single base pair to determine if each base pair contained both a purine and pyrimidine. In both the wrapped and not-wrapped structures, only two base pairs in each map did not contain both a purine and a pyrimidine. Based on the structural assignment for each base, a simple, linear R/Y/X profile could then be constructed for each DNA strand.

Similarly, a sequence-based R/Y profile was constructed for the 601 bp minicircle DNA, whereby each adenine and guanine were assigned a “R”, and each thymine and cytosine were assigned a “Y”. The structure-based R/Y profiles were then compared to the forward and reverse directions of the minicircle DNA sequence profile. In the wrapped state, a single possible alignment where 22 of the 24 bases in the R/Y/X DNA pattern matched the sequence profile was identified at positions 112–135 in the minicircle DNA. Even allowing for the two mismatches, the probability of finding this pattern in “random” DNA is 1 in 55,738 (0.001794%). Likewise, alignment of the 30-base profile from the not-wrapped structure revealed a unique sequence in the DNA minicircle profile, corresponding to positions 132 to 161 in the minicircle DNA, where 28 of the 30 base profiles matched (1 in 2,304,167 or 0.0000434%).

For the wrapped structure, only 24 of the 30 possible base pairs in the G-segment DNA were used for the profile. The other base pairs were initially excluded from the profile as the density features of these bases were less well resolved from lower local resolution. However, after the initial alignment of the 24 bases in the structural profile, it was possible to extend the alignment to cover all 30 base pairs, corresponding to positions 109–138 of the minicircle DNA, at the G-segment.

To generate an atomistic model of the G-segment DNA in the wrapped and not-wrapped gyrase structures, the individual bases in the previously deposited structures (bases 84–118 in chain E of PDB ID: 8QDX and bases 1–30 in chain F of PDB ID: 8QQS, respectively) were mutated to the corresponding nucleotides in Coot⁷⁹, which were extrapolated from the sequence/structure R/Y profile alignments. The resulting models were then iteratively refined nine times against the corresponding density maps with Coot and Phenix real_space_refine (nproc=4, run=minimization_global+local_grid_search+morphing+simulated_annealing+adp, resolution = 4). Computational and visual examination of the fit to density appeared to indicate this new assignment fit the density maps well. For the final models, chain E of the wrapped DNA-gyrase model corresponded to positions 104 through 138 in the minicircle DNA sequence. For the not-wrapped DNA G-segment-gyrase model, chain E corresponded to bases 132–161.

Chemicals and reagents

NheI-HF, XbaI, HindIII-HF, Tsp-45I, EcoRI-HF, KpnI-HF, NcoI-HF, BsaHI, Nt.BsmAI, T4 DNA ligase, and micrococcal nuclease were purchased from New England Biolabs (NEB)(Ipswich, MA). SYBR Safe and SYBR Gold were purchased from Thermo Fisher Scientific. Oligonucleotides were synthesized by Sigma-Aldrich (Woodlands, TX).

Preparation of 600 bp minicircle topoisomers

Plasmid pIA171x²⁰ was isolated using a Qiagen maxiprep kit following the manufacturer’s protocol. 200 μg of pIA171x was cleaved with NheI-HF and XbaI. The 600 bp linear fragment produced was isolated by preparative gel electrophoresis (5% polyacrylamide gel (acrylamide:bis-acrylamide = 29:1)). The gel was run at 125 V (~6 V/cm) for 6 h in 40 mM Tris-acetate, 1 mM EDTA (TAE) buffer. The gel was subsequently stained with SYBR Safe and visualized using an Invitrogen Safe Imager 2.0 blue-light transilluminator. The band containing the 600 bp fragment was excised and DNA electroeluted from the gel slice at 100 V for 6 h in a D-tube dialyzer (Millipore Sigma, Burlington, MA). Electroeluted DNA was filtered through a 0.45 μm syringe filter, extracted with butanol to both reduce the volume and remove any residual SYBR Safe, extracted with chloroform, then precipitated with ethanol. DNA was resuspended in 10 mM Tris pH 8.0, 0.1 mM EDTA (T(0.1)E buffer) and desalted using an Amicon 0.5 ml centrifugal filter (Millipore Sigma, Burlington, MA). DNA concentration was determined using a NanoPhotometer (Implen, Munich, Germany).

600 bp was circularized by incubating 48 μg of the linear fragment with T4 DNA ligase for 2 h at room temperature in 50 mM Tris pH 7.5, 10 mM MgCl₂, 10 mM DTT, 1 mM ATP (ligase buffer) in a total volume of 30 ml. DNA concentration was kept low in the ligation to minimize concatemerization. The ligated DNA was concentrated and desalted (to remove ATP) using an Amicon 15 ml centrifugal filter then precipitated with ethanol. Circularized DNA was resuspended in T(0.1)E buffer. Residual T4 DNA ligase was inactivated by incubating at 65 °C for 10 min and the DNA subsequently nicked with Nt.BsmAI. Nt.BsmAI was subsequently inactivated by incubating at 65 °C for 20 min. Negatively supercoiled topoisomers were prepared by religating the nicked 600 bp minicircle in the presence of ethidium bromide as described previously^80–82. Individual minicircle topoisomers were isolated by preparative gel electrophoresis (4% polyacrylamide gel). Gels were run at 125 V for 6 h in 40 mM Tris-acetate, 10 mM CaCl₂ buffer. Gels were stained with SYBR Safe, visualized with blue-light, DNA bands were excised, and DNA electroeluted from the gel slices as described above for the linear 600 bp fragment.

Micrococcal nuclease digestion and restriction enzyme mapping of mutant gyrase binding sites on negatively supercoiled minicircles

300 nM mutant gyrase (GyrA(Y122F)₂GyrB₂) was incubated with 7 nM (500 ng) supercoiled (ΔLk = −2) 600 bp minicircle in 20 mM Tris pH 7.9, 100 mM potassium acetate, 5 mM CaCl₂, 1 mM DTT, and 100 µg/ml recombinant albumin at room temperature for 15 min in 200 µl total volume. The sample was transferred to 37 °C and micrococcal nuclease (100 gel units) was added. After 1 min incubation at 37 °C, the nuclease reaction was quenched by the addition of EDTA (100 mM final concentration). SDS (1% final concentration) and proteinase K (100 µg/ml) were added, and the DNA was incubated at 55 °C for 2 h. The reaction was subsequently subjected to a nucleotide removal kit (Qiagen) and concentrated using an Amicon 0.5 ml centrifugal filter (10,000 Da molecular weight cutoff) (Millipore Sigma, Burlington, MA). An aliquot was analyzed using an Agilent TapeStation (D5000 ScreenTape) to confirm that gyrase protects fragments of the DNA from digestion by micrococcal nuclease. No protection was observed when gyrase was omitted in a control reaction (otherwise treated identically). Samples of the gyrase protected DNA were incubated with different restriction enzymes in either NEBuffer r2.1 or NEBuffer r1.1 (NEB). These buffers were used instead of the supplier recommended rCutSmart buffer (NEB), which is not compatible with the D1000 ScreenTape (Agilent). Samples were incubated at 37 °C for 1 h (except for the Tsp45I reaction, which was incubated at 65 °C for 1 h) and subsequently analyzed using an Agilent TapeStation (D1000 ScreenTape) to identify fragment sizes. The TapeStation output files for these data are included as Supplementary Data 1.

Analysis of DNA sequence-dependent deformability

Our custom Python program utilized the sequence-based deformability values of DNA base-pair steps in a tetrameric context (grouped by four base pairs) to analyze the sequence-dependent deformability²⁷ of the 601 bp minicircle. This program first applies periodic boundary conditions to the written sequence, considering the start site assigned in Vayssières et al. 2024²⁰, copying the start onto the end and the end onto the start, allowing the sequence to be read as circular. Then a 4 bp sliding window rolls through the entire DNA sequence, identifying the tetramers and matching the tetramer to its deformability value, building a list of the sequence-dependent deformability values for every base-pair step, according to published base-pair step parameters for the 136 unique tetramers^26,27.

To quantify local deformability along the minicircle, we implemented a k-mer-based sliding window analysis. For a given k-mer (a subsequence of length k), the deformability score was calculated as the average of all base-pair step deformabilities within that segment. The chosen length, k, also defined the lengths of the sequences copied to establish periodic boundary conditions. This window was then shifted one base at a time along the entire minicircle sequence, treating the circle with periodic boundary conditions to ensure continuity. In this way, we generated a deformability profile analogous to a hydrophobicity plot, allowing direct comparison of regional flexibility across the sequence (see Supplementary Fig. 5). For this study, k-mers of length 35, matching the length of the G-segment, and length 49, matching the length of the β-pinwheel wrap, were chosen for analysis. The sliding window moves through the list of deformability values of the entire circular sequence, calculating the deformability score of each possible k-mer as it moves through them. Regions with highest deformability were then determined by comparing the averaged deformability scores of all k-mers on a sequence to each other.

Analysis of gyrase-induced DNA deformation

To assess how gyrase binding alters DNA structure, we analyzed the spatial organization of paired bases and their interactions with the surrounding protein (Figs. 4,5). Base-pair geometry was characterized using the 3DNA/DSSR software^83,84, which defines the six rigid-body parameters⁵³ joining the origins of the base frames⁸⁵. DNA conformational states were classified using standard 3DNA-DSSR descriptors, including zP, base-pair rigid-body parameters, and base-pair step rigid-body parameters. Base pairs with rigid-body values outside the canonical B-form range were classified as melted based upon values presented by Olson et al.⁸⁶. Step parameters (Tilt, Roll, Twist, Shift, Slide, Rise) were further used to identify deviations from ideal B-DNA geometry. Standard values are reported in refs. ^26,27.

We then applied the zP and zP(h) descriptors to discriminate between distinct helical forms⁴⁸. These quantities describe the relative position of the backbone phosphorous atom with respect to the planes of successive base pairs and pinpoint the differences in A- versus B-DNA structure more clearly than other conformational parameters⁵⁶. Steps with zP ≥1.5 Å were assigned A-like character, whereas those with zP ≤0.5 Å retained B-like geometry, corresponding to phosphorous positions in the plane of the 5´-base pair in A-like DNA versus positions midway between the planes of the two base pairs in B-like DNA. Steps with zP(h) > 4 Å correspond to the strongly bent TA-form DNA seen in TATA-box-binding-protein complexes⁸⁷. In this case, the phosphorus locations are described relative to the local helical axis of the base-pair step. The characteristic pattern of overwound C-like DNA—higher Twist, negative Roll, and positive Slide⁵⁷—does not occur in the DNA-gyrase complexes. Together, these parameters allowed direct identification of A-form, TA-form, and locally melted regions in the gyrase-bound duplex. The 3DNA/DSSR output files for these data are included as Supplementary Data 4–10.

Comparison to the Topo-Seq identified gyrase motif

The position weight matrix (PWM) of the gyrase motif was obtained from the supplementary information of the Topo-Seq paper^45,46. Sequences of the negatively supercoiled 601 bp minicircle were then compared to the motif’s position weight matrix as described^45,46 to obtain the motif score. To compare the 601 bp minicircle sequence to only parts of the motif, the PWM was split into three smaller position weight matrices—one containing the position corresponding to the first periodic region and the G-segment core of the motif, one containing only the positions corresponding to the G segment core, and one containing the positions corresponding to the G-segment core then the second periodic region of the motif. These three PWMs were then compared to the 601 bp minicircle to obtain motif scores as described^45,46.

Reporting summary

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

Author contributions

M.L.B., M.V., J.M.F., V.L., W.K.O., and L.Z. designed the research; M.L.B., H.R.J., R.A.E., J.M.F., S.L.S., W.K.O., and L.Z. developed the methods; M.L.B., H.R.J., R.A.E., J.M.F., S.L.S., W.K.O., and L.Z. performed the analyses and wrote the paper; M.V., N.M., J.M.F., and V.L. generated the original cryoEM data and edited the paper.

Peer review

Peer review information

Nature Communications thanks Nei-Li Chan, John Nitiss and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data supporting the findings of this study are available in the article and Supplementary Information files. The following PDB ID structures were used for analysis: 8QQS and 8QDX²⁰ and 9GBV⁴¹. The following EMDB structures were used for analysis: EMD-19618⁸⁸, EMD-36391⁸⁹, EMD-51228⁹⁰, EMD-43114⁹¹, EMD-42887⁹², EMD-44328⁹³, EMD-45194⁹⁴, EMD-61624⁹⁵, EMD-62196⁹⁶, EMD-50027⁹⁷, EMD-17845 [https://www.ebi.ac.uk/pdbe/entry/emdb/EMD-17845]⁹⁸, and EMD-52646⁹⁹. Source data are provided with this paper.

Code availability

The code is available at Github (https://github.com/Z-Lab-BCM/Deformability-Plo).

Competing interests

J.M.F. and L.Z. are co-inventors on several patents covering the minicircle technology and are shareholders in Twister Biotech, Inc. (rebranded as Velvet Therapeutics, Inc.). The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-71884-0.