Accurate Substrate-Like Probes for Trapping Late-Stage Intermediates in Nonribosomal Peptide Synthetase Condensation Domains

Abstract

Nonribosomal peptide synthetases (NRPSs) are a family of multidomain enzymes dedicated to the production of peptide natural products. Central to NRPS function are condensation (C) domains, which catalyze peptide bond formation and a number of specialized transformations including dehydroamino acid and β-lactam synthesis. Structures of C domains in catalytically informative states are limited due to a lack of clear strategies for stabilizing C domain interactions with their substrates and client domains. Inspired by a β-lactam forming C domain, we report herein the synthesis and application of 1, which forms irreversible cross-links with engineered thiol nucleophiles in a C domain active site. Deployment of 1 demonstrates the synthetic tractability of trapping late-stage nascent peptides in C domains and provides a readily adaptable tactic for stabilizing C domain interactions in multidomain NRPS fragments.

Graphical Abstract

Nonribosomal peptide synthetases (NRPSs) are a super-family of multidomain assembly-line-like enzymes dedicated to the production of specialized peptide metabolites. Many NRPS products are high value medicines, including the immunosuppressor cyclosporin, the vancomycin family of antibiotics, and several β-lactams including penicillins, monobactams, and the nocardicins. Three domains are integral to NRPS function: (1) adenylation (A) domains, which select amino acids for activation and incorporation into the product, (2) peptidyl carrier proteins (PCP), which tether amino acyl and peptidyl intermediates during synthesis with a phospho-pantetheine (PPant) post-translational modification, and (3) condensation (C) domains, which facilitate peptide-bond formation by bringing together the thioester of their upstream PCP-bound donor for reaction with the amine of their intramodular PCP acceptor. In the past 20 years, C domains and their evolutionary siblings have been implicated in a variety of specialized functions as well, from substrate gating, l-to d-stereochemical inversion, macrolactam formation, hetero-cycle formation, dehydroamino acid synthesis, and even β-lactam synthesis.^1,2

Despite strong interest in C domains, their mechanisms of action have remained poorly understood. Each C domain consists of a conserved V-shaped pseudodimer of chloram-phenicol acetyltransferase-like lobes and can adopt a range of conformations from “open” to “closed”.³ These conformational dynamics coupled with transient bisubstrate binding and delivery on PCPs has made structural work challenging and the interpretation of mechanisms difficult. Furthermore, local coevolution between PCPs and various C domain subtypes implies that different substrate presentations are likely, further elevating the need for informative C domain structures from varied domain subtypes.⁴

Recent publications have focused on resolving C domains within large multidomain NRPS fragments. This approach led first to images of the PPant cofactor delivered to either the donor or acceptor site.^5,6 However, due to the instability of PPant thioesters, attempts to resolve native aminoacyl extensions have mostly failed. More recent advances have used stable thioester analogs, such as amides or thioethers, to visualize simple substrates in C domain active sites.^7,8 Nevertheless, the substrates observed in C domains to date are not rigidly held, implying that successful trapping of catalytic states is partially the result of chance. As a result, the creation of reliable methods for stabilizing the interface between PCP and C domains with substrates is highly desired to further parse the details of C domain catalysis. In addition, as NRPS synthesis proceeds down the assembly line, the donor substrate becomes increasingly complex. To understand the accommodation of this nascent peptide, accurate mimics of late-stage intermediates are optimally required.

In other megasynthases, such as polyketide synthases and fatty acid synthases, covalent cross-linkers have been used to trap intermediates in the complex machinery.^9,10 Some of the most successful and commonly used probes use a variation of carba(dethia) PPant or coenzyme A (CoA) as their core scaffold.¹¹ Acetonyldethio-CoA would be the simplest example from this class of molecules, in which the native thioester of acetyl-CoA is replaced with a ketone. In general, the carba(dethia) scaffold is thought to be a conservative replacement for PPant because ketones and thioesters are electronically similar. Unfortunately, however, adapting this substitution to mimic intermediates in assembly line enzymes often requires difficult multistep syntheses. As a result, the carba(dethia) scaffold has not been widely applied to NRPS systems. One notable exception is a recent development by Shi et al.¹² In their work, a carba(dethia) CoA was synthesized to mimic the PPant donor for the EntF C domain from enterobactin biosynthesis. The resulting aryl ketone was proposed to react with its downstream acceptor amine, yielding an imine or hemiaminal linkage. Consistent with this proposed mechanism of action, the probe exhibited tight binding to its target C domain. However, because imine and hemiaminal formation are reversible, rigid cross-linking was not observed.

To overcome these limitations, we set out to develop a synthetically generalizable and functionally irreversible cross-linking strategy for C domains. We were inspired by the putative mechanism for the nocardicin β-lactam synthesizing C domain (NocB-C5), which includes the generation of a highly electrophilic α,β-unsaturated thioester. Earlier work demonstrated that this intermediate was on the pathway to the nocardicins because synthesis and feeding of the intermediate to the termination module of the nocardicin synthetase (NocB-M5) produced the β-lactam product.¹³

More recently, it was demonstrated that evolutionarily distinct but related C domains involved in dehydroamino acid synthesis produce the same electrophilic intermediate.¹⁴ We envisioned replacing the nocardicin donor with one bearing a carba(dethia) substitution would form a similar α,β-unsaturated electrophile that could be applied to irreversible cross-linking with a thiol nucleophile (Figure 1A). While cysteine thiols are not expected in C domain active sites, rational engineering to introduce these residues could provide access to the desired cross-link. Increasing our confidence in this approach, engineered cysteines previously facilitated PCP cross-linking with A domains and the anchoring of substrate-like small molecules in a C domain active site.^15,16 To allow for the ready adaptation of 1 to trap diverse peptidyl intermediates in NRPS C domains, we envisioned that late-stage attachment of the peptide portion was tactically desirable.

Figure 1.

(A) Cross-linking strategy and components of synthetic design. (B) Proposed routes to cross-linker 1 including (a) NocB-C5 catalyzed dehydration of 2 and (b) direct synthesis of 1 via selenoxide elimination from an intermediate like 3.

To target NocB-C5 specifically, we sought to produce 1 attached to the C5 donor, PCP4 (Figure 1B). Since elimination is catalyzed natively by NocB-C5, we predicted that the carba(dethia) analog 2 could be converted to 1 enzymatically. However, as most C domains do not catalyze dehydration, this aim is not generalizable. Therefore, we devised a route directly to the reactive α,β-unsaturated ketone 1 through selenoxide elimination from an intermediate like 3. The advantage of this route is that phenyl selenoethers are generally stable to nonredox synthetic steps but can be easily eliminated in sensitive peptide backbones without racemization at other centers.¹⁷ This property is ideal for synthetic targets like 1, which contains two highly sensitive (p-hydroxyphenyl)glycines (Hpgs).¹⁸ Bearing these considerations in mind, we report the development and characterization of 1 and 2. Using active site cysteine engineering, we demonstrate that 1 can form specific protein cross-links in the active site of NocB-C5. Furthermore, we discuss the flexibility of the design of 1 such that it could be adapted to trap other late-stage nascent peptides in NRPS C domains.

Our synthetic efforts began with the simplest probe, 2, which started with the readily accessible l-serine derived Weinreb amide 4a bearing acetonide and carbamate protecting groups (Scheme 1).¹⁹ Ketone synthesis with Grignard 5 derived from 3-chloropropan-1-ol readily provided the expected γ-keto alcohol, which was immediately acylated to 6a.²⁰ The acetonide protection could be removed with stoichiometric TMSOTf to yield 7a.²¹ At this point, the carba(dethia) connection between serine and PPant was apparent. However, to facilitate downstream substitutions, the ketone moiety was protected as the 1,3-dioxolane. Originally, we tried a variety of protic acid-catalyzed ketalizations, but all failed.^22,23 We expect the failure of these reactions is owed to the adjacent α-amine, which when protonated would destabilize the highly charged intermediates of ketal formation. Successful 1,3-dioxolane formation was achieved producing 8 using the Noyori protocol, which relies on TMSOTf catalysis under aprotic conditions.²⁴ Subsequent masking of the serine-derived hydroxyl as the tert-butylsilyl ether 9 and deprotection of the acetyl moiety yielded 10.

Scheme 1.

Synthesis of the Carba(dethia) Connection

To install the β-selenoether of our target 3, we explored the possibility of substitution of intermediates like 8 (Scheme 1). However, attempts such as tosylation followed by nucleophilic substitution or halogenation under Mitsunobu conditions failed.^17,25 We reasoned that the neopentyl-like properties of the 1,3-dioxolane sterically encumbered substitution of the β-hydroxyl. Therefore, we pivoted to the earlier intermediate 7b, bearing an Alloc group. Substitution at the β-position proceeded rapidly under modified Appel conditions to give the β-iodo ketone 11 with minimal elimination as a byproduct.²⁵ Then, substitution with phenylselenoxide gave 12 in excellent yield.¹⁷ Protection of the ketone as the 1,3-dioxolane 13 and removal of the acetyl moiety yielded 14.

From intermediates 10 and 14, both syntheses followed nearly identical paths (Scheme 2). Substitution of the γ-hydroxyl proceeded in one step using Mitsunobu conditions with phthalimide.²⁶ Then, deprotection of the phthalimide group with hydrazine hydrate gave 15a and 15b. Initially, contaminating diazene in commercial hydrazine hydrate produced a mixture of 15b and its Alloc-reduced byproduct. However, addition of allyl alcohol as a scavenger inhibited reduction and byproduct formation.²⁷ The resulting γ-amine of 15a and 15b was then coupled with 16 to fully extend the pseudopantetheine arm.²⁸ Subsequent removal of protecting groups at R¹ and R² yielded 17a and 17b.

Scheme 2.

Construction of the Full Peptidyl-Warhead Dephospho-CoA

Finally, to mimic the nascent peptide engaged with NocB-C5, we appended 18 using the conventional EDC/HOBt strategy. Given the sensitivity of the C-terminal Hpg, this reaction occasionally led to partial racemization of Hpg. However, minor diastereomeric impurities could typically be removed in later purification steps. Global removal of the protecting groups in TFA/water yielded 19a and 19b. Notably, the 1,3-dioxolane required extended acid exposure to remove it. Incubation of 19a and 19b with the CoA biosynthetic enzymes PanK and PPAT along with ATP produced the dephospho-CoA’s 20 and 21 (after selenoxide elimination with aqueous hydrogen peroxide; Supporting Information, Figure S1).²⁹ Unfortunately, phosphorylated 19b was only partially soluble in buffer, resulting in incomplete conversion to 21. Given that 19b contains three aryl side chains, we expect that the yield would increase if the modular peptidyl portion of the molecule was more hydrophilic. The dephospho-CoA’s 20 and 21 were then primed to modify PCP4 and produce the desired targets 1 and 2 with the promiscuous phosphopantethineyl transferase, SfpR4–4.³⁰

We next evaluated the ability of 2 to enzymatically convert to 1 in the presence of NocB-C5. We loaded a Strep-tagged II version of PCP4 to generate 2 and incubated it for extended periods with NocB-M5 and its cosubstrates ATP and Hpg.³¹ Extracting PCP4 via its Strep-tag II and analysis by intact protein mass spectrometry revealed 2 unchanged (Supporting Information, Figure S2A). Similarly, incubation of the probe in PCP4–C5 and analysis by intact protein mass spectrometry over time revealed that 2 failed to dehydrate (Supporting Information, Figure S2B). Although we were disappointed 2 did not convert to 1 enzymatically, similarly unexpected outcomes were reported with the carba(dethia) scaffold in other enzymes.³²

As a result, we turned to 1 as the more promising and generalizable probe for targeted cross-linking with an engineered cysteine nucleophile. Guided by a homology model of NocB-C5, we picked three active site positions to mutate to cysteine, notably NocB-H792, T924, and E1028.³³ Our model was further guided by structural alignment to C domains that exhibited substrate density in the donor or acceptor sites (Supporting Information, Figure S3).^5,6,8 The positions chosen for mutagenesis were oriented in our model into the active site and close to the expected locus of catalysis. To increase chances of specific cross-linking, each mutant was engineered into the PCP4–C5 didomain where 1 would be presented by its PCP to the C domain at the highest effective concentration because the two components are constrained in the same polypeptide.

We were then faced with the challenge that both 1 in PCP4–C5 and its cysteine 1,4-cross-linked product have identical molecular weights (+846 Da). Furthermore, common methods for measuring peptide modifications such as trypsinolysis and mass fingerprinting are cumbersome and limited by digestion coverage and decreased detection sensitivity for low abundance, high molecular-weight cross-linked fragments.³⁴ As an alternative, we designed an experiment in which the small molecule cysteamine could be added to “chase” the probe after it was loaded on the didomain and provide a unique mass readout on the extent of cross-linking (Figure 2A). If 1 cross-linked efficiently (state A), cysteamine added after loading would be incapable of reacting with the electrophilic warhead. Therefore, intact protein mass spectrometry would yield the characteristic +846 Da mass for the loaded probe. If, however, a portion of 1 was un-cross-linked (state B), cysteamine could add into the α,β-unsaturated moiety and yield the unique mass +923 Da.

Figure 2.

(A) Rationale for how cysteamine chase provides a unique mass readout on the extent of cross-linking in the PCP4–C5 didomains loaded with 1. (B–E) Deconvoluted mass spectra of PCP4–C5 from cysteamine chase experiments for (B) WT, (C) H792C, (D) T924C, and (E) E1028C didomains. (F) SDS-PAGE analysis of WT and E1028C variants of PCP4–C5 loaded with 1 and the effect of SDS-PAGE sample preparation on cross-link-induced anomalous banding. *1 and *3 denote bands of the expected molecular weight for PCP4–C5. *2 highlights the anomalous band observed in cross-linked PCP4–C5*E1028C that also disappears when the sample is incubated at a high temperature during sample preparation. (G) Comparison of the extent of interprotein cross-linking between PCP4 loaded with 1 and WT or E1028C holo-C5–A5–PCP5.

Using SfpR4–4, we optimized conditions such that apo-PCP4–C5 could be quantitatively modified to 1 in under 10 min. Prior to the addition of cysteamine, a portion of the didomain was checked by intact protein mass spectrometry to guarantee complete loading. This step ensures that any cysteamine-derived mass is exclusively the result of on-protein addition. After a 30 min incubation with cysteamine, the didomain was again analyzed by intact protein mass spectrometry. We were pleased to see that the WT didomain underwent near complete conversion from apo- to 1-loaded-(1–) to cysteamine addition product (1+Cyst-; Figure 2B). This result indicates that in the WT didomain, 1 does not engage in detectable cross-linking. We repeated each cysteamine chase experiment a minimum of three times to demonstrate reproducibility (Supporting Information, Figure S4). This result provides an important negative control for the absence of cross-linking. To our surprise, variants H792C and T924C both reproducibly express partially modified with PPant from their host cells. Nevertheless, the apo species in both variants can still be modified and tracked with intact protein mass spectrometry. Interestingly, H792C was partially modifiable with cysteamine, perhaps suggesting that cysteine is nonoptimally aligned with the warhead of 1 in this variant (Figure 2C). T924C by contrast showed robust protection from cysteamine (Figure 2D). It should be noted that T924 is located near the donor site entrance. Therefore, cysteine at this position could react with 1 before it is fully extended into the active site (Supporting Information, Figure S3). Gratifyingly, the E1028C didomain expressed almost exclusively in the apo form like the WT protein and was fully protected from cysteamine addition in three independent replicates (Figure 2E). In all, these results, especially the contrasting addition of cysteamine in the WT and E1028C didomains, indicate that 1 can be applied strategically to form specific, functionally irreversible cross-links between PCPs and engineered cysteines in a C domain active site.

Encouraged by these results, we next analyzed PCP4–C5 WT and E1028C didomains loaded with 1 by SDS-PAGE. We were pleased to find that we could not identify any off-target cross-links like protein dimers (~160 kDa, Supporting Information, Figure S5). However, to our surprise, SDS-PAGE analysis revealed a striking migration behavior that correlates with intraprotein cross-linking. While WT PCP4–C5 loaded with 1 consistently produced bands at the expected migration based on molecular weight (Figure 2F, note *1), the cross-linked E1028C didomain exhibited two bands, one at an anomalously large apparent molecular weight (Figure 2F, note *2). We examined the relationship between the anomalous band and how the sample was prepared for SDS-PAGE analysis. To our further surprise, the anomalous band faded when the samples were incubated for extended periods at high temperature prior to SDS-PAGE analysis (Figure 2F, note *3). In SDS-PAGE analysis, protein migration is critically dependent on its shape and SDS-binding capacity.³⁵ When these parameters are inconsistent, anomalous bands are known to occur at larger or smaller apparent molecular weights. These effects are common in transmembrane proteins, for example, that do not always unfold.³⁵ Notably, thermal denaturation is unnecessary to observe the WT didomain migrating normally, even after loading with 1. This observation is consistent with anomalous banding being directly related to the formation of intraprotein cross-links.

Although the reason for cross-link-induced anomalous banding is not completely clear, we hypothesize that intraprotein cross-linking either stabilizes protein folding or alters its shape in a denatured state, resulting in anomalous migration. Heat could overcome these effects by either pushing the protein into a sufficiently unfolded state to migrate normally or by reversing the cross-linking event itself and, thus, linearizing the didomain.³⁶ The net result in either case would be a homogeneous protein state consistent with the expected properties in SDS-PAGE.

Given the evident success of intraprotein cross-linking with 1 and an engineered active-site cysteine, we briefly explored if 1 could form interprotein cross-links as well. The PCP4 monodomain modified with 1 was incubated equimolar with the WT and E1028C variants of holo-C5–A5–PCP5 (Figure 2G). Although the E1028C variant showed density for a higher molecular-weight cross-link on SDS-PAGE, the WT control exhibited the same band with slightly reduced intensity. In this interprotein cross-linking scenario, the specificity of 1 for the C domain is reliant on the probe’s intrinsic affinity and reactivity. If these parameters are insufficiently balanced, we would expect that 1 would react at alternative surface-exposed positions, perhaps explaining the modest cross-linking that is observed in the absence of an engineered active site cysteine (Supporting Information, Figure S6). By contrast, intraprotein cross-linking outcompetes off-target reactions by presenting the donor at a much higher effective molarity to its fused C domain active site.

In conclusion, we developed a targeted strategy for irreversible cross-linking of late-stage nascent peptides in NRPS C domains. This strategy was inspired by the nocardicin β-lactam synthesizing domain C5, which generates an α,β-unsaturated thioester as an on-pathway intermediate to its β-lactam product. By mimicking this intermediate with a carba(dethia) analog, we generated a nonhydrolyzable electrophile that can be applied to react specifically with cysteine nucleophiles introduced into the active site of NocB-C5. By synthetic design, 1 could be easily adapted to study other C domains by simply swapping the peptidyl portion in the penultimate synthetic steps. Although the utility of probes like 1 for cross-linking to other C domains is undetermined, we hypothesize that specificity relies critically on two things: first, that a single active site cysteine is available for cross-linking and, second, that the substrate is presented at a high effective concentration to the C domain. This circumstance is most easily achieved through intraprotein cross-linking in a PCP–C didomain. As we experienced, however, not all PCP–C didomains express in the apo form, excluding some situations in which our probe could be applied. However, the clear advantages of probes like 1 are their synthetic tractability and their accurate mimicry of late-stage nascent peptide intermediates in NRPS synthesis. These features are demonstrated in the synthesis of 1, which is identical to the native intermediate produced by C5 except for a single S → CH₂ substitution. Application of this approach broadly in conjunction with structural methods could allow the visualization of complex peptidyl intermediates in NRPS C domains and lead to accelerated understanding of C domain functions.

EXPERIMENTAL PROCEDURES

Loading of Synthetic Probes.

To convert apo-PCP4 or apo-PCP4–C5 to 1 and 2, the purified proteins were incubated at 50 μM with 1.2 equiv of 20 or 21 in buffer containing 50 mM PIPES, at pH 6.5, 10 mM MgCl₂, and 1 μM SfpR4–4. After incubation for 10 min at RT, an aliquot was rapidly diluted and analyzed by intact protein mass spectrometry with a Waters Acquity/Xevo-G2 UPLC-MS (Milford, MA) to ensure conversion of the apo- to the loaded species. m/z deconvolution was performed with the MaxEnt algorithm integrated into the MassLynx software package.

Cysteamine Chase.

To determine the extent of cross-linking in PCP4–C5, immediately after loading (described above), an aliquot of loading reaction was diluted to 10 μM PCP4–C5 in a buffer containing 50 mM PIPES, at pH 6.5, and 1 mM cysteamine hydrochloride. The protein was incubated for 30 min before diluting again and analyzing by intact protein mass spectrometry.

SDS-PAGE Analysis of Cross-Linking.

To visualize the impact of cross-linking on PCP4–C5 didomain migration, SDS-PAGE samples were compared after complete loading (described above) to an apo-PCP4–C5 control. SDS loading buffer (175 mM Tris, at pH 7.5, 20% glycerol, 4% SDS, 0.06% Bromophenol Blue, and 10% fresh β-mercaptoethanol) was added to each sample and either incubated at RT for 10 min or 98 °C for 10, 20, 30, or 40 min. Samples were then run at 180 V in precast 8% polyacrylamide gels using a commercially provided MOPS running buffer (SurePAGE, Bis-Tris; GenScript, Piscataway, NJ).

Modeling of the C5 Active Site and Selection of Mutants.

A homology model of C5 was constructed with RoseTTAFold using the Robetta public server and default parameters (robetta.bakerlab.org).³³ The resulting homology models were then aligned to C domain structures with the PPant cofactor and some aminoacyl mimics resolved in the donor or acceptor sites (PDB: 4zxi, 5ejd, 6mfw).^5,6,8 Residues for mutation to cysteine were chosen based on whether they were oriented into the active site and near the expected region for catalysis.

Reconstitution of 2-loaded PCP4 with C5.

To assess the ability of 2 to enzymatically convert to 1, PCP4 loaded with 2 was reconstituted with WT M5 similarly to methods previously described.^2,13 Briefly, 2-loaded PCP4 was incubated at RT at 25 μM with 25 μM holo-M5 in a reconstitution buffer (50 mM HEPES, pH 7.5, 25 mM NaCl, 0.2 mM TCEP) supplemented with 2.5 mM ATP and 1 mM L-Hpg. At 3 and 6 h, aliquots of the reconstitution reaction were passed over a bed of Strep-Tactin XT Sepharose (Cytiva, Uppsala, SE) to bind PCP4 via its C-terminal StrepII-tag. The resin was washed with 10 column volumes of reconstitution buffer, and PCP4 was then eluted with 100 μL of buffer containing 50 mM Biotin, at pH ~ 8–9. Eluted PCP4 was then analyzed by intact protein mass spectrometry. In addition, PCP4–C5 loaded with 2 was also analyzed by mass spectrometry after 3 and 6 h of incubation (Supporting Information, Figure S2).