Exquisite selectivity of griselimycin extends to beta subunit of DNA polymerases from Gram-negative bacterial pathogens

Abstract

Griselimycin, a cyclic depsidecapeptide produced by Streptomyces griseus, is a promising lead inhibitor of the sliding clamp component of bacterial DNA polymerases (β-subunit of Escherichia coli DNA pol III). It was previously shown to inhibit the Mycobacterium tuberculosis β-clamp with remarkably high affinity and selectivity – the peptide lacks any interaction with the human sliding clamp. Here, we used a structural genomics approach to address the prospect of broader-spectrum inhibition, in particular of β-clamps from Gram-negative bacterial targets. Fifteen crystal structures of β-clamp orthologs were solved, most from Gram-negative bacteria, including eight cocrystal structures with griselimycin. The ensemble of structures samples widely diverse β-clamp architectures and reveals unique protein-ligand interactions with varying degrees of complementarity. Although griselimycin clearly co-evolved with Gram-positive β-clamps, binding affinity measurements demonstrate that the high selectivity observed previously extends to the Gram-negative orthologs, with K_D values ranging from 7 to 496 nM for the wild-type orthologs considered. The collective results should aid future structure-guided development of peptide antibiotics against β-clamp proteins of a wide variety of bacterial targets.

Structural genomics and binding studies show that a cyclic peptide produced by Streptomyces griseus, griselimycin, exhibits high binding selectivity for diverse DNA polymerase binding sites of various infectious Gram-negative bacteria.

Introduction

The emergence of multi-drug resistant bacteria is an urgent threat that is being met by identifying novel drug targets and developing new antibiotics to inhibit those targets. The bacterial DNA replication and repair apparatus, or replisome, is a rich source of many potential drug targets that are essential for bacterial cell function^1,2. Additionally, several key components of the bacterial and eukaryotic replisomes are unrelated, and those showing homology typically have distinct protein architectures or structural motifs³. The β-sliding clamp (β-clamp, or simply β), the expressed product of dnaN, is an attractive target because it interacts with a large number of essential replisome proteins using a single interaction site at which resistance is likely difficult to be gained via mutation without loss of function^1,4. In addition, the human sliding clamp, proliferating cell nuclear antigen (PCNA), utilizes a different protein-protein interaction architecture for recognition of its replisome binding partners^5–8.

Early studies of the β-clamp in association with the clamp loader complex demonstrated its role as a DNA polymerase III (Pol III) processivity factor during DNA polymerization^9–12, and subsequent studies showed its much broader function in coordinating both DNA replication and repair as a protein interaction hub^13–23. When isolated, the β-clamp forms a stable homodimer^24,25 from semicircular subunits, yielding an overall ring-shaped structure²⁶. The structure is highly conserved, as revealed by over one hundred crystal structures spanning 20 bacterial genera and nine classes^27,28. At primed sites, the dimer is opened via ATP binding to the clamp loader^29–31 and closed around dsDNA, where it presents an anchoring site in each of its subunits for the recruitment of replisome proteins³². In the Escherichia coli β-clamp, the anchoring sites contain two adjacent protein binding subpockets, referred to as subsites 1 and 2³³. Several replisome binding partners interact with these subsites via linear penta- or hexapeptide motifs having the conserved consensus sequences QL[SD]LF and QLxLx[LF], respectively^4,16,33,34, with the two conserved C-terminal hydrophobic residues of the motifs buried at subsite 1^{31–33,35–38}.

The first synthetic inhibitors of the β-clamp were peptides that contained the consensus motif or closely related variants⁴ (for a recent review of β-clamp inhibitors, see ref. ³⁹; also ref. ⁴⁰). Through compound library screening and structure-aided optimization of the consensus peptide, later studies developed small molecule and peptide inhibitors that bind one or both of the peptide binding subsites or disrupt dimerization^32,41–45. The two best linear peptide inhibitors reported thus far are N-acetylated derivatives of the consensus penta- and hexapeptides^43,44. It is noteworthy that, unlike for human PCNA^46,47, the E. coli β-clamp can be inhibited, albeit weakly, with simple di- and tripeptides containing the C-terminal residues of the consensus motif that bind subsite 1^4,48.

The natural product griselimycin, a cyclic depsidecapeptide produced by Streptomyces caelicus and Streptomyces griseus (Fig. 1), was shown in 2015 by Kling, Lukat et al. to inhibit the β-clamps of Mycobacterium tuberculosis and Mycobacterium smegmatis with picomolar affinities and exquisitely high selectivities—no binding to human PCNA was detected⁴⁹. The β-clamp was suggested to be the molecular target of griselimycin from the presence of a paralog of dnaN (griR) in the griselimycin biosynthetic gene cluster of Streptomyces sp. DSM-40835⁴⁹. This was confirmed via binding and structural studies, which showed that griselimycin tightly binds the M. tuberculosis and M. smegmatis β-clamps and occupies the polymerase binding site. The macrocyclic ring binds at subsite 1 and the linear region at subsite 2⁴⁹.

Fig. 1. Chemical structure of griselimycin.

Griselimycin is a cyclic depsidecapeptide containing six modified amino acid residues. It has the linear sequence N-acetylated N-MeVal1–4(R)-MePro2–N-MeThr3–Leu4–4(R)-MePro5–Leu6–N-MeVal7–Pro8–D-N-MeLeu9–Gly10. Macrocyclization occurs via the esterification of Gly10 and the side chain of N-MeThr3.

Griselimycin was discovered in the 1960s and identified as a potent inhibitor of M. tuberculosis growth, but its development as a therapeutic was discontinued at the time due to poor pharmacokinetic properties^50–53. However, the study by Kling, Lukat et al. demonstrated that griselimycin derivatives having improved in vivo properties can be produced via total synthesis and exhibit bactericidal activity against M. tuberculosis in vitro^49,54,55. The recently discovered mycoplanecins are closely related peptides that further demonstrate how derivatization might increase therapeutic potential⁵⁶. Griselimycin derivatives were also recently shown to be bactericidal against Mycobacterium abscessus in mice⁵⁷. M. smegmatis variants showing resistance to griselimycin have been produced; however, resistance is mediated not via mutation of dnaN but rather by dnaN amplification, and occurs at low frequency and with compromised cell fitness⁴⁹.

With interest in broad-spectrum antibiotics and griselimycin showing promise as a lead inhibitor, the SSGCID initiated a large-scale structural genomics effort focused on the structure determination of griselimycin-bound β-clamps of a diverse array of Gram-negative bacteria. A total of fifteen apo and griselimycin-bound β-clamp crystal structures were determined, which covers SSGCID pipeline work on this system up to 9 August 2022. The next section describes a subset of this collection of structures in detail, together with griselimycin affinities for these and other β-clamp proteins. Subsequent subsections describe and compare bound griselimycin conformations, binding site architectures, and nonbonded interactions. The analysis was primarily aimed toward the structure-guided development of cyclic peptide inhibitors based on the griselimycin scaffold, with derivatized amino acid residues that increase medicinal potential.

Results

Crystal structures generated from the SSGCID β-clamp pipeline

Fifteen crystal structures were determined of β-clamp orthologs from nine Gram-negative bacteria and one Gram-positive bacterium, Mycobacterium marinum, including 8 cocrystal structures with griselimycin (Table 1 and S1–S4). All of the targets are products of the respective wild-type genes except an engineered Rickettsia rickettsii M375L clamp, which tests whether a leucine at this site (see PDB entry 3T0P), can support griselimycin binding through a conformational change, similar to what is observed for the methionine^33,48. The Klebsiella pneumoniae β-clamp has an amino acid sequence identical to that of E. coli and is henceforth utilized as the reference Gram-negative protein. For five of the orthologs, structures of both griselimycin-bound and -free states were characterized. The structures were determined by molecular replacement and refined to resolutions of 1.70–3.05 Å, with most amino acid residues (>98% for most structures) modeled into electron density. Every residue of griselimycin was built into electron density for each of the structural models. The unbiased composite omit maps for the various griselimycin chains are shown in Supplementary Fig. S1.

Table 1.

Crystal structures of β-clamps and griselimycin-β-clamp complexes determined by the SSGCID

PDB entry	Organism	Gram stain	Res. (Å)
Griselimycin-free
5W7Z	Rickettsia conorii	−	1.70
6DEG	Bartonella birtlesii	−	2.45
6MAN	Rickettsia bellii	−	2.35
6D47	Mycobacterium marinum	+	2.45
6DLK	Rickettsia rickettsiia	−	2.00
6D46	Rickettsia typhi	−	2.00
7RZM	Stenotrophomonas maltophilia	−	2.15
Griselimycin-bound
6PTR	Bartonella birtlesii	−	1.75
6DJ8	Borrelia burgdorferi	−	2.05
6P81	Klebsiella pneumoniae	−	1.75
6DLY	Mycobacterium marinum	+	2.10
6PTH	Pseudomonas aeruginosa	−	3.05
6DM6	Rickettsia conorii	−	2.25
6PTV	Rickettsia rickettsiia	−	1.85
6DJK	Rickettsia typhi	−	1.85

^aM375L variant.

All of the orthologs characterized form ring-shaped homodimers having pseudo 6-fold symmetry, with an inner surface formed by twelve α-helices and an exterior formed by six β-sheets (Fig. 2A). The subunits contain three topologically related domains having two antiparallel β-sheets that form extended β-sheets with adjacent domains, with that formed by domains I and III fusing the two subunits.

Fig. 2. Crystal structures of griselimycin-bound β-clamp targets determined by the SSGCID.

A Biological homodimers of eight griselimycin-bound β-clamps listed in Table 1. The β-clamps are shown as ribbons and griselimycins as spheres. The right subunit is colored according to domain to indicate the subunit architecture. B Griselimycin binding site of the K. pneumoniae β-clamp. Griselimycin is shown in ball-and-stick representation. Hydrophobic core and peripheral sites are shaded yellow and dark blue, respectively. Subsites 1 and 2 are separated by KpMet362³³. C Twenty-two amino acid residues proximal to griselimycin, in stick representation. Griselimycin residues are labeled using the one-letter codes of the corresponding unmodified amino acid residues. D Multiple sequence alignment of β-clamp targets cocrystallized with griselimycin. The 22 residues surrounding griselimycins are colored according to domain. The arginine site reported to be important for E. coli cell viability⁵⁹ is indicated with an arrow. Conserved hydrophobic sites are marked with stars, with the KpMet362 site differentiated to indicate the engineered RrM375L variant.

Griselimycin binds at a crescent-shaped patch of eight hydrophobic residues situated between domains II and III, with its macrocyclic ring and linear segment bound at subsites 1 and 2 (Fig. 2B). A conserved methionine, KpMet362, which was previously suggested to function as a gating residue for peptide binding in studies of the E. coli β-clamp^48,58, serves as a dividing line between the two subsites. Due to the small side chain of AcMeVal1, subsite 2 is only partially occupied. The center of the binding sites is formed by a β-hairpin within the first β-sheet of domain II and six residues within the C-terminal strand of the protein (consisting of the C-terminal β-strand and the extension leading to the C-terminus). The binding sites are also formed by peripheral sites (outside the hydrophobic core) within a connective peptide between the two β-sheets of domain II, a short segment of the domain II-domain III linker, and a β-hairpin in domain III (Fig. 2B and C).

Our analysis focused on the griselimycin binding sites of the K. pneumoniae, Bartonella birtlesii, Rickettsia typhi, and Borrelia burgdorferi β-clamps. These were also compared to the binding site of the M. marinum ortholog, a close relative of the M. tuberculosis β-clamp, determined through structural genomics. The targets span a fairly wide range of the sequence space associated with the β-clamps of many potential human pathogenic bacteria (Supplementary Fig. S2). Additionally, the binding sites are diverse, having sequence identities of 82%, 59%, 50%, and 36%, respectively, with the K. pneumoniae β-clamp at twenty-two site-forming residues (Fig. 2D). Among the differences between the M. marinum β-clamp and the other SSGCID targets is a notable change in binding site location of a conserved basic residue (MmArg183, KpArg152, RtArg152, BbiArg154, and BbuLys169) essential for E. coli viability⁵⁹.

Griselimycin binding affinities

Steady-state binding affinities of 12 β-clamp proteins for griselimycin were obtained using surface plasmon resonance (SPR) spectroscopy. The wild-type β-clamps generally display sub-micromolar binding affinities, whereas the R. rickettsii clamp provides an example of a lower affinity interaction resulting from a mutation of the gating methionine residue (RrM375L) that separates subsites 1 and 2 (Table 2). An affinity value is not reported for the M. marinum β-clamp due to the sample responses not reaching equilibrium within a reasonable time frame. However, the mass transport limited (MTL) binding characteristics observed in the sensorgrams indicate that the interaction has a tight binding affinity consistent with previously reported data⁴⁹. Of the Gram-negative orthologs, the β-clamp from B. birtlesii shows the tightest interaction, with a binding affinity of 6.9 nM. This is more than an order of magnitude tighter than that observed for the other β-clamps, which we have shown here to bind with affinities ranging from 109 nM (B. burgdorferi) to 496 nM (Elizabethkingia anophelis). For orthologs that were characterized structurally, binding affinities, from strongest to weakest, are: M. marinum (MTL) > B. birtlesii (6.9 nM) > B. burgdorferi (109 nM) > K. pneumoniae (230 nM) > R. typhi (306 nM).

Table 2.

Summarized SPR steady-state affinity analysis results

Organism	Protein SSGCID ID and Batch number	KD (M)	σ (KD)	Experimental Rmax (RU)a	χ2 (RU2)
B. birtlesii	BabiA.17987.a.B1.PS38344	6.853E−09	1.611E−10	16.4	0.1802
B. burgdorferi	BobuA.17987.a.B1.PW37766	1.086E−07	3.834E−9	19.3	0.457
E. anophelis	ElanA.17987.a.EN11.PD38361	4.955E−07	4.189E−9	24.0	0.07564
H. pylori	HepyC.17987.a.EN11.PD38373	1.613E−07	5.342E−9	33.9	0.9765
K. pneumoniae	KlpnA.17987.a.EN11.PD38352	2.295E−07	4.917E−9	47.5	1.27
M. marinum	MymaA.17987.a.B1.PS38239	MTLb	MTLb	24.8b	MTLb
P. aeruginosa	PsaeA.17987.a.EN11.PD38369	2.492E−07	7.464E−9	71.5	14.69c
R. bellii	RibeA.17987.a.B1.PW38223	1.376E−07	3.886E−9	21.9	0.621
R. conorii	RicoA.17987.a.B1.PW38227	1.602E−07	4.142E−9	26.2	0.3831
R. felis	RifeA.17987.a.B1.PW38224	2.292E−07	2.449E−9	43.5	0.5532
R. rickettsii	RiriA.17987.a.B1.PW38222	1.923E−05	1.636E−7	37.8	0.1637
R. typhid	RityA.17987.a.B1.PW38354	3.057E−07d	2.354E−8	9.0d	0.2226

Reported values were determined by averaging the binding responses at each sample concentration prior to analysis. Two replicates of data were acquired for all samples.

^aR_max is the experimental binding signal measured in resonance units (RU).

^bMTL: mass transport limited.

^cLarge χ² value noted.

^dLow binding response noted.

Conformational adaptability of griselimycin

The bound griselimycins have similar overall conformations, but display backbone and side chain torsional angle variations that suggest conformational adaptability to the binding sites in the crystals (Fig. 3A–C). All of the structures adopt a macrocyclic ring conformation that resembles a β-hairpin, with most backbone torsional angles residing within the extended region of Ramachandran space (Fig. 3A and B). The chains reverse direction via an α-turn formed by residues 6–10 that contains a cis peptide bond between MeVal7 and Pro8. An additional hydrogen bond is formed between the carbonyl oxygen of Leu4 and the amide nitrogen of Leu6. The hydrophobic side chains are directed away from the interior of the macrocyclic ring, enabling recognition, in part, by shape.

Fig. 3. Structures of bound griselimycins.

A Comparison of conformations. Griselimycins bound to the R. typhi, B. burgdorferi, B. birtlesii, and M. marinum β-clamps were superimposed onto griselimycin bound to the K. pneumoniae β-clamp. Hydrogen bonds are indicated using dashed lines. B Ramachandran plot of griselimycins shown in panel A. C–F Pairwise superpositions of selected griselimycins. The RMSD values based on the Cα positions are, from left to right, 0.66, 0.27, 0.29, and 0.33 Å.

Pairwise superpositions of the griselimycins from the various asymmetric units presented here show a range of root mean square deviations (RMSDs) of 0.10–0.68 Å, based on the Cα atoms. The conformation tightly bound to the M. marinum β-clamp coincides closely with ones bound to the M. tuberculosis β-clamp⁴⁹. Relative to this state, the average RMSDs observed at the B. burgdorferi, B. birtlesii, R. typhi, and K. pneumoniae β-clamps are 0.38, 0.41, 0.55. and 0.67 Å, respectively. Similar conformations observed for particular pairs are noted in Fig. 3D–F. Also noteworthy are the distinct side chain conformations adopted by D-MeLeu9 and Leu6 in the M. marinum and R. typhi systems, respectively.

Binding site architectures in griselimycin-bound and -free states

Backbone superpositions of the griselimycin binding sites show that the central β-hairpin and C-terminal strand are largely conformationally invariant, with the former containing a similar non-standard β-turn⁶⁰ (Fig. 4A). In contrast, the three peripheral polypeptide segments show differences in secondary structure and vary in their positioning near griselimycin. The connective peptide of the B. burgdorferi β-clamp lacks a 3₁₀ helix and its disposition prevents interaction with griselimycin (see Supplementary Fig. S3 for a more detailed comparison of conformations). In the B. birtlesii and B. burgdorferi β-clamps, the linker segment resides within hydrogen bonding distance of MeVal7 of the macrocycle. The peripheral hairpin of the B. burgdorferi β-clamp has a 1-residue deletion that limits its interaction with griselimycin.

Fig. 4. Griselimycin binding site architectures.

Griselimycin-bound R. typhi, B. burgdorferi, B. birtlesii, and M. marinum β-clamps were superimposed onto that of K. pneumoniae based on Cα atoms of domains II and III. A Backbone structures, with secondary structure indicated. B Side chain constellations, colored according to source organism. C Griselimycin binding at E. coli Pol IV heptapeptide binding site. The K. pneumoniae β-clamp with griselimycin bound (colored as in Fig. 2) is overlayed with the E. coli β-clamp with the Pol IV peptide bound (black and cyan, PDB entry 1UNN^33,36). Asterisks indicate conserved hydrophobic sites.

The structural superpositions also show that the eight residues comprising the hydrophobic core in the K. pneumoniae β subunit illustrated in Fig. 2B are conserved spatially in the remaining binding sites, although there are important size variations at particular sites (Figs. 2D and 4B). Residues outside the hydrophobic core vary more significantly in their physicochemical properties and together with mainchain variations form a more heterogenous binding interface with the peptide.

Structural comparisons with the unliganded β subunits reveal conformational differences in the griselimycin-binding sites, particularly within domain II. The K. pneumoniae, M. marinum, and R. typhi proteins show only modest changes in backbone structure and configuration, and only a few residues (discussed below) undergo notable changes in side chain conformation (Supplementary Fig. S4; RMSDs between 22 Cα atom pairs of 0.70 ± 0.14, 0.44 ± 0.02, and 0.44 Å, respectively, after optimal superpositions of available structures^26,45,48,61 based on domains II and III^62,63). In contrast, in the unbound B. birtlesii protein, the central β-hairpin adopts a conformation that would sterically clash with griselimycin (Supplementary Fig. S5A). Additionally, in this structure as well as in the griselimycin-bound Pseudomonas aeruginosa ortholog (see PDB entry 6AMS⁶⁴ for the unliganded structure), the connective peptide undergoes a major conformational change to accommodate the binding of a symmetry-related chain, complicating the analysis.

Conserved and ortholog-specific hydrogen bonding

The binding site of residues 1–6 of griselimycin coincides closely with that of the Pol IV hexapeptide (residues 346–351) bound to the E. coli (K. pneumoniae) β-clamp (Fig. 4C and ref. ³⁶). In all of the griselimycin-bound complexes characterized, two conserved protein–peptide hydrogen bonds are formed, which recapitulate two of three hydrogen bonds formed with Pol IV. With the K. pneumoniae β-clamp as a reference, the hydrogen bonds occur between the amide nitrogen of Leu4 and the carbonyl oxygen of Gly174, and between the acetyl carbonyl oxygen of AcMeVal1 and the amide nitrogen of Arg365 (Fig. 5A).

Fig. 5. Hydrogen bonding interactions with griselimycin.

A Conserved hydrogen bonds (dashed lines) between Leu4 and KpGly174 site and AcMeVal1 and KpArg365 site. B, C Novel hydrogen bonds observed in complexes with particular orthologs. B B. birtlesii; backbone hydrogen bonding between MeVal7 and Asp250. C B. burgdorferi; backbone hydrogen bonding between MeVal7 and Asp259, and side chain-backbone hydrogen bonding between His193 and AcMeVal1.

The additional hydrogen bonds are less conserved and involve backbone and side chain sites in domain II and the domain II-domain III linker. As noted above, the carbonyl oxygen of MeVal7 is capable of reaching the linker segments of the B. birtlesii and B. burgdorferi proteins, forming a similar backbone–backbone hydrogen bond with the amide nitrogen atoms of BbiAsp250 and BbuAsp259 (Fig. 5B and C). However, it is useful to note that the latter complex exhibits less order in this region (Supplementary Fig. S6). A second novel hydrogen bond is formed between the side chain of BbuHis193 and the carbonyl oxygen of AcMeVal1 (Fig. 5C).

The remaining hydrogen bonds involve the aforementioned conserved basic residue in domain II (KpArg152). As described below, through its interaction with griselimycin, this residue can outcompete D-MeLeu9 for side chain burial rendering it an important determinant of shape complementarity. However, the hydrogen bonds exhibit either nonideal geometry or are not consistently formed in all chains displaying noncrystallographic symmetry. The side chain of KpArg152 forms weak hydrogen bonds with the carbonyl oxygen atoms of Pro8 and Gly10 (Supplementary Fig. S7A). RtArg152 likewise extends towards griselimycin but displays even less ideal geometry (Supplementary Fig. S8A). BbiArg154 in chain A adopts two conformations, one of which forms a hydrogen bond with the carbonyl of Gly10 (results not shown); in chain B, a water-mediated hydrogen bond is instead made with the carbonyl of D-MeLeu9 (Supplementary Fig. S8B). In the high-affinity complex involving the Gram-positive ortholog, MmArg183, which emanates from the β-hairpin instead of the connective peptide, adopts distinct conformations in chains A and B that involve direct or water-mediated hydrogen bonding with D-MeLeu9 (Supplementary Fig. S7B).

Shape complementarity

Conserved hydrophobic core

In all of the structures characterized, the side chains of MePro2 and Leu4 form a zipper-like motif with nonpolar residues in the C-terminal strand (Fig. 6A). The gating methionine residue (KpMet362) adopts the open conformation. In the R. rickettsii M375L variant, the Leu375 side chain also displays a conformation that accommodates griselimycin binding. Additional hydrophobic interactions are formed between residues 4–6 and the side chains of the remaining hydrophobic core residues. The binding sites also possess a Leu4-binding motif, which is known in the E. coli β-clamp to bind highly conserved leucine residues in replisome binding partners, assembled by backbone amide groups (Fig. 6B). These interactions collectively result in the nearly complete burial of Leu4 and Leu6 (Fig. 6C–E).

Fig. 6. Shape complementarity.

A Conserved hydrophobic core of griselimycin binding site consisting of eight hydrophobic residues. B π–C–H interactions (solid lines) within conserved Leu4 binding motif (shown for B. birtlesii β-clamp). C–E Percent burial of griselimycin residues. In panels (C) and (D), griselimycin residues are colored according to the values graphed in panel (E). In panel (E), each point corresponds to the difference in the accessible surface area of the griselimycin residue in the bound and free states (assuming identical bound and unbound conformations), expressed as a percentage of the accessible surface area in the free state.

Structure and interaction heterogeneity at the periphery

The remaining residues display partial and more variable extents of burial (Fig. 6C–E). Residues 7–10 predominantly bind peripheral sites in domain II and are largely solvent-exposed, especially at the B. burgdorferi β-clamp (Supplementary Fig. S6). The differences in protein–peptide interactions involving domain II are complex because of sequence heterogeneity and were further characterized via hydrophobic contact maps.

Hydrophobic contact maps: K. pneumoniae and M. marinum β-clamps

On the basis of such contacts, griselimycin displays a much lower degree of shape complementarity with domain II of K. pneumoniae β than with that of M. marinum β (Supplementary Fig. S7). Higher complementarity with the M. marinum linker segment occurs via additional contacts involving Leu6 and D-MeLeu9. Further contacts are established with the β-hairpin as a consequence of the KpGly174/MmArg183 and KpHis175/MmPhe184 replacements. Specifically, the disposition of MmArg183 enables D-MeLeu9 to become partially buried and to bind aliphatic sites of the side chain, and MmPhe184 forms a hydrophobic cluster with AcMeVal1 and MeThr3. A similar number of contacts are formed with the connective peptides. However, the burial of D-MeLeu9 is prevented by the side chain conformation adopted by KpArg152. Comparisons with the unliganded structures suggest that this side chain is largely preordered through its interactions with Asp150 and Tyr154, whereas that of MmArg183 undergoes a large conformational change to accommodate D-MeLeu9 burial (Supplementary Fig. S4B and D).

Hydrophobic contact maps: R. typhi, B. birtlesii, and B. burgdorferi β-clamps

The hydrophobic contact map topology calculated for R. typhi β is similar to that of K. pneumoniae β (Supplementary Figs. S8A and S7A). The KpVal247/RtPhe251 replacement in the linker segment leads to more contacts with residues 5–7 of griselimycin, but is associated with a conformational change in the side chain of Leu6 to a lower probability rotamer (Fig. 3E and Supplementary Fig. S8A). In the connective peptide, the side chains of Arg152, Asn154 (which replaces KpTyr154), and Leu155 are directed toward griselimycin. Arg152 is not well-positioned for salt bridge formation with Asp149, but similar to KpArg152 prevents the burial of the D-MeLeu9 side chain. The most significant conformational changes resulting from complexation involve the ordering of the side chains of Arg152, Asn154, and Phe251 (Supplementary Fig. S4F). The disorder exhibited by Arg152 in the unliganded protein thus contrasts with the preordering observed in the K. pneumoniae protein (Supplementary Fig. S4B).

The interface formed by griselimycin and the linker segment of B. birtlesii β displays the highest degree of complementarity among the Gram-negative ortholog complexes considered (Supplementary Fig. S8B). However, the side chains of several residues within the connective peptide (Arg154, Tyr155, and Tyr156) adopt multiple conformations leading to differences in the hydrophobic packing of D-MeLeu9. In chain B, residues 4–9 all form hydrophobic interactions with the linker segment. D-MeLeu9 is able to partially bury its side chain by displacing that of Arg154, although the electron density map suggests that the connective peptide is partially disordered. In chain A, D-MeLeu9 is unable to bury its side chain to the same extent, and no hydrophobic contacts are formed with Tyr156, Leu157, and Pro249. However, the high complementarity between residues 4-8 and the linker segment is maintained.

In the griselimycin-B. burgdorferi β complex, the majority of hydrophobic interactions between domain II and the macrocycle involve the linker segment (Supplementary Fig. S8C). Residues 7–10 of griselimycin display weaker electron density that is associated with solvent exposure (Fig. 6D and Supplementary Fig. S6) and a lack of interaction with the connective peptide (Supplementary Fig. S8C).

Discussion

The ensemble of crystal structures of griselimycin-bound and griselimycin-free β-clamps determined by the SSGCID constitutes the first large-scale structural genomics effort targeting an essential bacterial protein and its natural product cyclic peptide inhibitor. The structures demonstrate that griselimycin binds structurally diverse polymerase binding site architectures with well-defined conformations that can serve as starting points for structure-guided derivatization. The structure of the griselimycin-bound β-clamp of K. pneumoniae has potentially broad application because its griselimycin-binding site has a very high sequence identity with the β-clamps of many proteobacterial pathogens, some of which are listed as urgent or serious threats by the CDC (Supplementary Fig. S2).

The combination of structures and binding affinity measurements suggests that distinct mechanisms underpin the observed high affinities. In the case of the M. marinum system, which forms only the two conserved hydrogen bonds with AcMeVal1 and Leu4 and a potential direct hydrogen bond to D-MeLeu9, surface complementarity is likely the dominant contribution. In contrast, the solvent exposure of hydrophobic residues in the complex with the B. burgdorferi protein appears to be compensated for, at least in part, by two additional hydrogen bonds. The very high affinity observed for the B. birtlesii β-clamp reflects both high surface complementarity and additional hydrogen bonding, and likewise, the lower affinities for the K. pneumoniae and R. typhi β-clamps likely reflect the lack of novel hydrogen bonds with typical acceptor-donor distances and physicochemical complementarity with domain II. Although the griselimycin conformational analysis suggests an ability of the macrocycle to adapt to bind the different polymerase binding sites, the effect of the conformational deviations from the highest affinity poses observed in the complexes with Mycobacterium β-clamps on binding affinity remains to be determined.

Several hydrophilic residues in the griselimycin binding site of the E. coli (K. pneumoniae) β-clamp make highly specific interactions with replisome proteins, both within and outside linear peptide β-binding motifs, and thus are potential target binding sites for inhibiting β-clamps via griselimycin derivatization. EcArg152 interacts with the linear peptide motif of Pol V³⁸, the N-terminal domain of the δ subunit of the clamp-loader³¹, and the regulatory domain of MutL³⁵. Mutation of this residue to alanine does not support normal bacterial growth⁵⁹. His175 also interacts with the N-terminal domain of the δ subunit and the regulatory domain of MutL^31,35, and additionally with the linear peptide motif of Pol III³². Roles for Tyr154 in clamp loading and for Arg246 in binding a Pol III peptide have also been demonstrated^32,42. Thus, amino acid replacements at these sites in, e.g., the M. marinum β-clamp, likely do not reflect a lack of functional importance, but rather organism-dependent differences in the replisomes^65,66. However, the higher degree of complementarity between the M. marinum β-clamp and griselimycin is likely related to the coevolution of griselimycin to bind similar Gram-positive β-clamp architectures, which is evidenced by the very high sequence similarity of its connective peptide with that of the S. griseus β-clamp and paralog⁴⁹, with which griselimycin co-evolved (Supplementary Fig. S9).

Structural characterization of the binding site complementarity with griselimycin can facilitate derivatization by narrowing the large amino acid residue search space, potentially by several orders of magnitude (Fig. 7)^49,56. Based on the structures of the β-clamp orthologs considered here, the chemical search space for derivatization of Leu4 and Leu6 is expected to be somewhat limited due to the extensive interactions that these residues make with the β-clamp. In addition, the search space for MeThr3 and Gly10 would also likely be limited due to the role of these residues in macrocyclization. For the remaining residues, the low complementarity exhibited with some binding site residues can be offset via structure-aided derivatization. Residues 2 and 5 bind at sites that are partially hydrophobic, but which also display physicochemical heterogeneity. On the other hand, residues 1 and 7–9 bind at sites that display a fairly high degree of conservation of hydrophilic residues (K. pneumoniae Arg152, Tyr154, His175, Asp243, and Arg246 sites in domain II; Supplementary Fig. S2). Optimizations for binding to nucleophilic residues might include reactive side chains that can form covalent adducts for selective irreversible inhibition of the β-clamps, which was recently demonstrated for linear peptides⁶⁷.

Fig. 7. Structure-guided derivatization of griselimycin.

Chemical structure-binding site diagram indicates conserved physicochemical properties of sites surrounding likely candidate residues for derivatization.

The work on griselimycin and the development of griselimycin variants utilizing modified amino acids⁴⁹ is analogous to earlier work on tyrocidine A⁶⁸, a cyclic decapeptide natural product that operates by disrupting the structure of bacterial cell membranes⁶⁹. Although the work on tyrocidine A led to the development of synthesis platforms for generating improved peptide derivatives^70,71, no structural information is available for tyrocidine A bound in its biological context, given that it targets the lipid bilayer. In the case of griselimycin, the large set of β-clamp crystal structures generated by the SSGCID has the potential to greatly accelerate the development of griselimycin derivatives for combating drug-resistant Gram-negative bacteria.

Methods

Cloning and protein expression and purification

The various dnaN open-reading frames were cloned from genomic DNA into the BG1861 expression vector, which encodes a non-cleavable N-terminal hexahistidine affinity tag; the full tag sequence is MAHHHHHH-(ORF). DnaN sliding clamp proteins from E. anophelis strain NUHP1, Helicobacter pylori, K. pneumoniae, and P. aeruginosa failed to express soluble, stable protein using this affinity tag, and thus these were cloned into an expression vector encoding an N-terminal His₆Smt tag⁷². The proteins were expressed and purified using a general expression and purification protocol detailed previously^73,74. The proteins were expressed in BL21 (DE3) R3 Rosetta E. coli cells using autoinduction media⁷⁵ in a LEX bioreactor. The cells were harvested via centrifugation and frozen at −80 °C, and then later re-suspended in lysis buffer for sonication and clarification by centrifugation. The proteins were purified by nickel chelate chromatography followed by size exclusion chromatography in 25 mM HEPES, pH 7.0, 500 mM NaCl, 2 mM DTT, 0.025 mM sodium azide, and 5% (v/v) glycerol. Attempts to cleave off the His₆Smt tag using ULP-1 protease were unsuccessful. Fractions containing the β-clamps were collected, pooled, concentrated to ~20 mg/mL, flash frozen in liquid nitrogen, and stored at −80 °C prior to crystallization experiments. The SSGCID target identifiers are BabiA.17987.a (B. birtlesii), BobuA.17987.a (B. burgdorferi), ElanA.17987.a (E. anophelis), HepyC.17987.a (H. pylori), KlpnA.17987.a (K. pneumoniae), MymaA.17987.a (M. marinum), PsaeA.17987.a (P. aeruginosa), RibeA.17987.a (Rickettsia bellii), RicoA.17987.a (Rickettsia conorii), RifeA.17987.a (Rickettsia felis), RiriA.17987.a (R. rickettsii), RityA.17987.a (R. typhi), and StmaA.17987.a (Stenotrophomonas maltophilia). Plasmids are available at the following website: www.ssgcid.org/available-materials/.

Protein crystallization and structure determination

Crystals were grown using the vapor diffusion method typically at 14–20 °C with 0.2–0.4 µL of protein and an equal volume of precipitant against 100× (i.e., 40–80 µL) reservoir in either Compact Jr 96-well crystallization plates from Rigaku Reagents (0.4 µL drops) or MRC plates (0.2 µL drops). β-clamp crystals were grown under the following conditions. B. birtlesii: 13.7 mg/mL against JCSG+ screen⁷⁶ condition H7 (0.2 M ammonium sulfate, 0.1 M bis–tris, pH 5.5, and 25% PEG3350) supplemented with 15% ethylene glycol (EG) as cryo-protectant; griselimycin-bound, 13.7 mg/mL with 2 mM griselimycin (from 100 mM DMSO-d6 stock solution) against PACT screen condition F3 (0.2 M sodium iodide, 0.1 M bis–tris propane, pH 6.5, and 20% PEG 3350), supplemented with 20% EG as cryo-protectant. B. burgdorferi: griselimycin-bound, 20 mg/mL with 1 mM griselimycin against Morpheus screen⁷⁷ condition G4 (20 mM sodium formate, 20 mM ammonium acetate, 20 mM sodium citrate tribasic dihydrate, 20 mM sodium potassium tartrate tetrahydrate, 20 mM sodium oxamate, 0.1 M imidazole/MES, pH 6.5, 12.5% MPD, 12.5% PEG1000, and 12.5% PEG3350). K. pneumoniae: griselimycin-bound, 20.95 mg/mL with 2 mM griselimycin against MCSG-2 screen condition D4 (0.2 M calcium acetate, 0.1 M sodium acetate, pH 4.5, and 30% PEG300). M. marinum: 22.54 mg/mL against Morpheus screen G6 (0.02 M each carboxylic acid (sodium formate, ammonium acetate, trisodium citrate, sodium potassium l-tartrate, sodium oxamate), 0.1 M MOPS/HEPES, pH 7.5, 10% PEG8000, and 20% EG,); griselimycin-bound, 22.4 mg/mL against TOP96 screen condition D11 (0.1 M sodium acetate, pH 4.6, and 8% PEG4000), supplemented with 25% EG as cryo-protectant. P. aeruginosa: griselimycin-bound, 20.74 mg/mL with 0.8 mM griselimycin against MCSG1 screen condition A5 optimization screen (0.2 M sodium chloride, 0.1 M sodium acetate, pH 4.2, and 1.4 M ammonium sulfate), supplemented with 25% EG as cryo-protectant. R. bellii: 20 mg/mL against 0.25 M sodium thiocyanate and 26% PEG3350, supplemented with 20% EG as cryo-protectant. R. conorii: 20 mg/mL against Precipitant Synergy Screen 2 condition E8 (0.1 M imidazole, pH 6.5, 2.01% MPD, and 13.4% PEG8000), supplemented with 20% EG as cryo-protectant; griselimycin-bound, 20 mg/mL against Wizard Cryo Full screen condition A8 (0.1 M sodium cacodylate, pH 6.5, and 35% 2-ethoxyethanol), soaked with 0.5 mM griselimycin overnight. R. rickettsii: 20 mg/mL protein against TOP96 screen condition G6 (0.5 M ammonium sulfate, 0.1 M sodium citrate, pH 5.6, and 1 M lithium sulfate); griselimycin-bound, 20 mg/mL with 2 mM griselimycin against Morpheus screen condition G12 (0.03 M NPS (sodium nitrate, disodium hydrogen phosphate, ammonium sulfate), 0.1 M bicine/Trizma, pH 8.5, 12.5% PEG1000, 12.5% PEG3350, and 12.5% MPD). R. typhi: 24.49 mg/mL against JCSG+ screen condition C6 (0.1 M phosphate/citrate, pH 4.2, and 40% PEG300); griselimycin-bound, 20 mg/mL against TOP96 screen condition D10 (0.2 M lithium sulfate, 0.1 M bis-tris, pH 6.5, and 25% PEG3350), soaked with 2.5 mM griselimycin for 3.5 h then supplemented with griselimycin and 15% EG as cryo-protectant. S. maltophilia: 16.74 mg/mL against Morpheus screen condition F5 (20 mM d-glucose, 20 mM d-mannose, 20 mM d-galactose, 20 mM l-fucose, 20 mM d-xylose, 20 mM N-acetyl-d-glucosamine, 100 mM MOPS/HEPES-Na, pH 7.5, 10% PEG 20000, and 20% PEG MME 550).

Crystals were irradiated under a stream of nitrogen (100 K) either in house using a rotating anode Rigaku SuperBright FR-E + X-ray generator with Osmic VariMax HF optics and a Saturn 944 + CCD detector at a wavelength of 1.5418 Å (M. marinum β-clamp, griselimycin-bound) or using synchrotron radiation at the Advanced Light Source ALS-ENABLE beamline 8.2.1 at 1.0 Å wavelength (R. conorii and R. rickettsii β-clamps) or at the Life Sciences Collaborative Access Team beamline 21-ID-F at the Advanced Photon Source, Argonne National Laboratory at 0.97872 Å wavelength (β-clamps of B. birtlesii, B. burgdorferi, K. pneumoniae, M. marinum (unbound), P. aeruginosa, R. bellii, R. conorii (griselimycin-bound), R. rickettsii (griselimycin-bound), R. typhi, and S. maltophilia). X-ray diffraction data reduction was performed using XDS/XSCALE⁷⁸. Structure determination was accomplished using molecular replacement in Phaser MR⁷⁹. The final models were built after numerous iterative rounds of refinement in Phenix⁸⁰ and manual model building in Coot⁸¹. The structures were validated and improved using Molprobity⁸². Griselimycin structures were validated using 2Fo-Fc composite omit maps computed using Phenix⁸⁰. Model bias was removed via simulated annealing with default settings (Cartesian coordinate displacements, initial temperature of 5000 K, and 0.05 fraction asymmetric unit volumes omitted)^80,83,84. Other programs used for structure determination include CCP4⁸⁵, MOLREP⁸⁶, MoRDa⁸⁷, Parrot⁸⁸, and ARP/wARP⁸⁹. For the cocrystal structure of griselimycin and the R. rickettsii β-clamp (M375L variant), the griselimycin was modeled into electron density for only two of the four β-clamp chains in the asymmetric unit. For the final refined structures, 94.5–98.4% of the φ–ψ angle pairs fall within the favored regions of Ramachandran space. Outliers in six structures comprise <1% of the total pairs, having the following percentages: R. typhi, griselimycin-bound, 0.3%; B. birtlesii, griselimycin-bound, 0.1%; M. marinum, griselimycin-bound, 0.1%; B. birtlesii, apo, 0.7%; M. marinum, apo, 0.8%; and S. maltophilia, apo, 0.27%. For some analyses, particular chains were chosen/omitted given considerations of crystal packing effects. Additional details related to the X-ray diffraction and structure refinement statistics are provided in Tables S1 through S4.

Binding analysis

SPR experiments were performed using a Biacore 4000 instrument (Cytiva) utilizing NTA sensors and standard nickel-capture methods at an analysis temperature of 25 °C with the following assay buffer: 25 mM Tris, pH 8.0, 200 mM NaCl, 1 mM TCEP, 1% glycerol, 50 µM EDTA, 0.05% Tween-20, and 5% DMSO. β-clamp proteins were diluted to concentrations of 10–60 µg/mL for immobilization, resulting in capture levels ranging from 1170 to 3562 RU. Assays were performed with a flow rate of 30 µL/min and 11-point two-fold griselimycin dilution series and included sample blanks as a twelfth analyte sample. The maximum griselimycin concentrations utilized for each β-clamp were as follows: 200 µM (R. rickettsii), 100 µM (R. typhi), 50 µM (B. burgdorferi, E. anophelis, H. pylori, R. conorii, R. felis), 25 µM (K. pneumoniae, R. bellii), 12.5 µM (P. aeruginosa), 1.5625 µM (B. birtlesii), and 48.828 nM (M. marinum). Analytes were injected from low to high sample concentration, with two replicates being collected in series within the same experiment. Sensors were regenerated with 350 mM EDTA between each cycle, while DMSO solvent correction data were acquired both before and after the sample acquisition cycles. Data were processed in Biacore 4000 Evaluation Software version 1.1 and included corrections for capture adjustment and DMSO solvent correction. Data were also double-referenced using an adjacent unmodified sensor spot and sample blanks. Analyses were performed in the Evaluation Software using steady-state affinity methods and averaged response values for each sample concentration. The σ values provide a metric for the error associated with each experiment. The values correspond to the standard deviations generated by the Biacore Evaluation software’s cross-validation step but are not true standard deviations due to the number of replicate data acquired in each experiment (n = 2). Specific experimental variables for each protein are provided in Table S5. Steady-state analyses figures are shown in the Supplementary Information section as Supplementary Figs. S10–S21.

Figure preparation

Sequence alignments were performed with MUSCLE⁹⁰ and illustrated with ESPript3⁹¹. Figures displaying molecular structures, contact maps, and buried surface areas were prepared with the aid of Chimera^62,63. Secondary structures were assigned using BetaTurnLib18 and DSSP^62,92. Percent burial values for griselimycins bound to B. birtlesii, B. burgdorferi, and M. marinum β-clamp chains not displayed in Fig. 6E are graphed in Fig. S22. Electron density maps were calculated using Phenix⁸⁰. Chemical structures were drawn with ChemDraw. SPR sensorgrams and response versus concentration graphs were prepared using Biacore Evaluation Software.

Reporting summary

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

Author contributions

Michael K. Fenwick: Analysis, validation, visualization, writing—original draft preparation, writing—review and editing. Phillip G. Pierce: Investigation, formal analysis, writing—original draft preparation, writing—review and editing. Jan Abendroth: Investigation, formal analysis, validation, writing—original draft preparation. Kayleigh F. Barrett: Investigation. Lynn K. Barrett: Supervision, project administration. Kalinga Bowatte: Investigation. Ryan Choi: Investigation. Ian Chun: Investigation. Deborah G. Conrady: Investigation, formal analysis, validation. Justin K. Craig: Investigation. David M. Dranow: Investigation, formal analysis, validation. Bradley Hammerson: Investigation. Tate Higgins: Investigation, formal analysis. Donald D. Lorimer: Supervision, validation, formal analysis. Peer Lukat: Investigation. Stephen J. Mayclin: Investigation, formal analysis. Stephen Nakazawa Hewitt: Investigation. Ying Po Peng: Investigation. Ashwini Shanbhogue: Investigation. Hayden Smutney: Investigation. Matthew Z. Z. Stigliano: Investigation. Logan M. Tillery: Investigation. Hannah S. Udell: Investigation. Ellen G. Wallace: Investigation. Amy E. DeRocher: Supervision, validation. Isabelle Q. Phan: Data curation, conceptualization, software. Bart L. Staker: Supervision, visualization, writing-original draft preparation, writing—review and editing. Sandhya Subramanian: Data curation; visualization; writing—original draft preparation; writing—review and editing. Wesley C. Van Voorhis: Supervision, funding, resources, acquisition, project administration. Wulf Blankenfeldt: Supervision. Rolf Müller: Supervision, funding acquisition, and conceptualization. Thomas E. Edwards: Supervision, funding acquisition, resources, project administration. Peter J. Myler: Supervision, funding acquisition, and project administration.

Peer review

Peer review information

Communications Biology thanks Nicholas Dixon and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Wendy Mok and Tobias Goris.

Data availability

The atomic coordinates and structure factors (codes B. birtlesii, 6DEG, griselimycin-bound, 6PTR; B. burgdorferi, griselimycin-bound, 6DJ8; K. pneumoniae, griselimycin-bound, 6P81; M. marinum, 6D47, griselimycin-bound, 6DLY; P. aeruginosa, griselimycin-bound, 6PTH; R. bellii, 6MAN; R. conorii, 5W7Z, griselimycin-bound, 6DM6; R. rickettsii M375L variant, 6DLK, griselimycin-bound, 6PTV; R. typhi, 6D46, griselimycin-bound, 6DJK; S. maltophilia, 7RZM) were deposited in the RCSB Protein Data Bank, www.rcsb.org. The X-ray diffraction images have been deposited in the Integrated Resource for Reproducibility in Macromolecular crystallography (IRRMC), http://proteindiffraction.org.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-024-07175-5.