Structural basis for intermodular communication in assembly-line polyketide biosynthesis

Abstract

Assembly-line polyketide synthases (PKSs) are modular multi-enzyme systems with considerable potential for genetic reprogramming. Understanding how they selectively transport biosynthetic intermediates along a defined sequence of active sites could be harnessed to rationally alter PKS product structures. To investigate functional interactions between PKS catalytic and substrate acyl carrier protein (ACP) domains, we employed a bifunctional reagent to crosslink transient domain–domain interfaces of a prototypical assembly line, the 6-deoxyerythronolide B synthase, and resolved their structures by single-particle cryogenic electron microscopy (cryo-EM). Together with statistical per-particle image analysis of cryo-EM data, we uncovered interactions between ketosynthase (KS) and ACP domains that discriminate between intra-modular and inter-modular communication while reinforcing the relevance of conformational asymmetry during the catalytic cycle. Our findings provide a foundation for the structure-based design of hybrid PKSs comprising biosynthetic modules from different naturally occurring assembly lines.

Assembly-line polyketide synthases (PKSs) are homodimeric multi-enzyme systems that synthesize many structurally complex and medicinally important natural products^1–4. Their modular architecture implies the existence of an inbuilt mechanism for ensuring that a growing polyketide chain is channeled along a uniquely defined sequence of enzymatic active sites, each of which is used only once in the overall catalytic cycle of the PKS⁵. Engineering these systems holds enormous potential for the biosynthesis of user-defined chemical structures⁶. To this end, we have studied 6-deoxyerythronolide B synthase (DEBS; Supplementary Fig. 1) as a paradigm for decoding the principles of assembly-line polyketide biosynthesis⁷. For a representative catalytic cycle of the first module in this assembly line, see Supplementary Fig. 2.

Recently, the application of single-particle cryogenic electron microscopy (cryo-EM) has not only opened the door to structural analysis of intact PKS modules at near-atomic resolution but also highlighted the conformationally dynamic nature of these modules as they proceed through their catalytic cycles^8–13 (see ref. 14 for a recent review on this topic). In our previous cryo-EM analysis of DEBS, we observed an asymmetric conformation (State 1) of homodimeric module 1 (M1)¹². (We specifically employed a version of M1, referred to as M1TE, that was N-terminally fused with a docking domain for antibody complexation and C-terminally fused with a thioesterase (TE) domain for improved expression of the recombinant protein.) In this structure, the 4′-phosphopantetheine (Ppant) arm of an acyl carrier protein (ACP) domain from one subunit was inserted into the active site of the ketosynthase (KS) domain from the other subunit. This conformation corroborated previous biochemical evidence supporting the involvement of inter-subunit KS–ACP interactions during the elongation of a growing polyketide chain by an assembly-line PKS¹⁵. The cryo-EM structure of State 1 also led us to propose that the elongation reaction is catalyzed asynchronously by the two equivalent KS–ACP pairs of a homodimeric PKS module. A related study of the lasalocid A PKS independently arrived at the same conclusion¹³.

In the present study, we combined cryo-EM with a largely overlooked covalent crosslinking method to visualize functionally critical domain–domain interfaces in key transient states of the catalytic cycles of DEBS modules 1–3. These structural insights, in turn, explain how reactive intermediates are efficiently channeled between successive modules while reaffirming the functional relevance of asymmetry during the PKS’s catalytic cycle.

Results

Site-selective crosslinking of KS/ACP thiols of PKS modules

Based on the observation that the catalytic Cys of the KS domain and the Ppant thiol of the ACP were approximately 4 Å apart in State 1 (ref. 12), we attempted to covalently crosslink the proximal KS and ACP thiols using 1,3-dibromoacetone (DBA) as a thiol-reactive bis-electrophile^16,17. Given the rapid reversibility of Ppant arm docking into the KS active site, we anticipated that a subset of KS and ACP partners in a population of homodimeric modules would each become pre-modified with one equivalent of DBA and, therefore, unable to crosslink upon KS–ACP docking (Fig. 1a). We, therefore, added DBA in slight excess (1.25 equivalents) to holo-form DEBS M1TE while also providing citrate (≥50 mM), a known promoter of polyketide chain elongation, to the reaction mixture^12,18. A large excess of β-mercaptoethanol (BME) was added after DBA addition to quench any unreacted electrophiles.

Fig. 1 |. Crosslinking a PKS module with DBA.

a, Only one crosslinking outcome corresponding to the asymmetrically crosslinked State 1 is shown. (See Supplementary Fig. 5a for other possible products.) b, Addition of DBA to DEBS M1TE in the presence of 200 mM sodium citrate (pH 7.2) resulted in the formation of two major products as judged by SDS-PAGE (bands 1 and 2). The un-crosslinked monomer is denoted as band 3. Three replicate 30-s reactions (n) were analyzed alongside the unreacted M1TE (−). Squiggly lines connected to sulfur atoms depict Ppant groups. For gel source data, see Source Data Fig. 1.

Product analysis of these crosslinking reactions by SDS-PAGE revealed two putatively crosslinked species (bands 1 and 2) with decreased electrophoretic mobilities relative to monomeric M1TE (band 3; Fig. 1b). Similar products were also observed when DBA was added to the DEBS module 3 analog of M1TE (that is, M3TE; Supplementary Fig. 3a). The formation of species 1 and 2 was rapid (reaching ~50% completion within 5 s) and citrate dependent, suggesting that DBA crosslinking captured conformations related to polyketide chain elongation (Supplementary Fig. 3b). Absence of the Ppant arm from the ACP domain or mutagenesis of the catalytic Cys residue of the KS greatly abolished crosslinking, highlighting DBA selectivity for KS and ACP thiols (Supplementary Fig. 4). The decreased electrophoretic mobilities of bands 1 and 2 implied that one of them may correspond to a crosslinked State 1, although the origin of the other band was unclear.

Fluorescent probe analysis of crosslinked PKS modules

To gain further insight into the identities of bands 1 and 2, we considered four possible KS–ACP crosslinked species (Supplementary Fig. 5) based on a previous DBA crosslinking analysis of the mammalian fatty acid synthase (mFAS), which shares the homodimeric architecture and domain organization of assembly-line PKS modules¹⁶. The putative species contain either one or two KS–ACP crosslinks that occur either within or between PKS subunits, and each was expected to migrate with reduced electrophoretic mobility relative to the un-crosslinked band 3 monomer. Given that one of the observed species probably mimicked the elongation state of a module, we hypothesized that it may also be possible for DBA to capture a state mimicking inter-modular chain translocation. If so, then a standalone form of the cognate upstream ACP might serve as a probe to crosslink with unoccupied KS domains in bands 1, 2 or 3. Based on structural data, it has been proposed that a translocation event might be permitted to occur concomitantly with elongation during polyketide biosynthesis¹³.

To test this hypothesis, ACP2(2) was expressed as a standalone protein, purified and tagged with a fluorophore. ACP2(2) is the cognate substrate donor of DEBS M3; it harbors the C-terminal docking domain of DEBS M2 (denoted by the parenthetical ‘2’), which is required for native inter-molecular interactions between DEBS M2 and M3 (refs. 19–21). Recombinant ACP2(2) was tagged with fluorescein isothiocyanate to prepare ACP2(2)-FL. To confirm that the N-linked fluorescein did not impair ACP binding to its KS partner, ACP2(2)-FL was shown to efficiently crosslink to the homodimeric KS–AT didomain fragment of M3 (KS3AT3) in the presence of DBA (Supplementary Fig. 6).

Upon addition of the ACP2(2)-FL probe or a buffer control to purified M3TE before DBA crosslinking, two major fluorescent adducts were observed: one that migrated between bands 2 and 3, consistent with ACP2(2)-FL labeling of un-crosslinked M3TE (that is, band 3 + ACP2(2)-FL), and another with a slightly decreased mobility relative to band 1, supporting the assignment of band 1 as the crosslinked State 1 (Fig. 1b and Supplementary Fig. 7). Notably, in contrast to extensive conversion of band 1 to its labeled counterpart, no fluorescent band indicative of labeling of band 2 was observed. Not only do these findings attest to the assignment of band 2 as a self-crosslinked monomer (Fig. 1b and Supplementary Figs. 5 and 7), but they also suggest that the unoccupied KS of State 1 is capable of crosslinking with its translocation ACP partner. Our proposed assignment of bands 1 and 2 is also consistent with previous crosslinking of the mFAS¹⁶. In that study, analogous crosslinked species were observed, with the intra-polypeptide crosslinked species also migrating faster than the inter-polypeptide crosslinked species. Further verification of our assignments of bands 1 and 2 was obtained by limited trypsinolysis (Supplementary Figs. 8 and 9), although, given that in-gel proteolysis was conducted on denatured DBA-treated M3TE, it was not possible to establish whether band 2 existed as a monomer or dimer before denaturation.

Structure of an intermodular polyketide translocation interface

Motivated by our observation that KS3AT3 can be efficiently crosslinked to its upstream ACP2(2) partner (Supplementary Fig. 6), we sought to optimize these crosslinking conditions to facilitate structural analysis of this prototypical module–module interface. Crosslinking of these two proteins was not as strongly dependent on citrate as that between KS3AT3 and its elongation partner ACP3, nor could it be driven to completion even in the presence of a large excess of ACP (Fig. 2 and Supplementary Figs. 10 and 11). To compare the crosslinking efficiency of the two ACP partners of KS3AT3, both ACP3 and ACP2(2) were co-incubated with homodimeric KS3AT3 under varying conditions before DBA addition (Fig. 2). Conditions were established under which both the elongation and translocation crosslinking events were generally competitive. A preparative-scale crosslinking reaction was performed with KS3AT3, ACP3 and ACP2(2), and the major protein complex was isolated by size-exclusion chromatography (SEC) for single-particle cryo-EM analysis (Supplementary Fig. 12).

Fig. 2 |. DBA crosslinking a KS–AT core with its upstream and downstream ACP partners.

SDS-PAGE analysis of DBA crosslinking between the homodimeric KS–AT core of module 3 and its holo-form elongation (ACP3) and translocation (ACP2(2)) ACP partners. Total protein per lane was quantified and expressed as an average percentage of two technical replicates (Source Data Fig. 2; s.d. ≤ 2%). Red asterisks denote a protein whose KS active site harbors a quenched DBA moiety (as explained in Supplementary Fig. 5a). For additional controls associated with both crosslinking reactions, see Supplementary Figs. 10 and 11. For gel source data, see Source Data Fig. 2.

Three-dimensional (3D) classification and refinement of cryo-EM data from 7,500 dose-fractioned movies produced a noteworthy map at 3.05-Å gold standard Fourier shell correlation (GSFSC) resolution²² from approximately 25% (77,996) of the total particles (Supplementary Figs. 13 and 14). This map, which featured unambiguous density for one copy of ACP2(2), represents a near-atomic resolution structure of a catalytically competent module–module interface from an assembly-line PKS. In this structure, ACP2(2) is bound to the KS active site cleft in an orientation that is entirely consistent with previous predictions based on mutational analysis, solution nuclear magnetic resonance (NMR) and in silico docking^23–25 (Fig. 3a). Specifically, helix 1 of ACP2(2) forms multiple electrostatic contacts with the proximal AT domain (Fig. 3b). Although the region of the DBA-derived crosslink was not visible, the Ppant co-factor could be partially traced in the KS active site (Fig. 3b and Supplementary Fig. 15). Notably, the map resolved an 11-residue α-helical fragment of the C-terminal docking domain of ACP2(2) engaged with its counterpart α-helices of the N-terminal docking domain of KS3AT3. The observed three α-helix bundle features a single α-helix of ACP2(2) bound to a hydrophobic groove of the α-helical coiled-coil docking domain of KS3AT3 (Fig. 3c), akin to the solution NMR structure of a similar docking domain complex²⁰. Unlike the NMR structure, however, which captured a symmetric four α-helix bundle through covalent fusion of the interacting polypeptides²⁶, our cryo-EM structure revealed an asymmetric complex in which only a single ACP donor interacts with its homodimeric KS acceptor (Fig. 3a,c). Although statistical image analysis (discussed below) revealed the presence of homodimeric KS3AT3 particles in which both KS–AT clefts were occupied by ACP partners, none of these higher-order complexes could be refined to yield near-atomic resolution structures.

Fig. 3 |. Cryo-EM structure of an intermodular docking interaction during polyketide translocation.

a, Cryo-EM structure at 3.05-Å GSFSC resolution represented by two 90°-related orientations of the homodimeric KS3AT3 fragment of DEBS crosslinked to its upstream ACP partner, ACP2(2) (translocation; Supplementary Figs. 13 and 14). Interactions involving ACP2(2) at the KS–AT cleft (b) and the N-terminal docking domain of KS3AT3 (c) are explicitly shown (67 residues separating the ACP and its C-terminal docking domain were unresolved; depicted as a dashed line). In b, the cryo-EM map surface is displayed within a 2-Å radius relative to ACP2(2) atoms. Prior mutational analysis established that residues on helix 1 (black arrow) were critical for polyketide translocation²⁴. This structure reveals that Arg23 and Arg26 on helix 1 are appropriately situated to form salt bridges with Glu820, Asp847 and Asp851 in the AT domain of KS3AT3. Based on proximity to the observed backbone atoms, additional interactions are likely to occur between Lys39 and Arg42 of ACP2(2) and Glu554 (from the KS–AT linker) and Asp851 (from the AT domain) of KS3AT3. Modeling of the Ppant and DBA crosslink atoms was not fully supported by the map density and is, therefore, inferred from the actual biochemical state of the protein. Likewise, modeling of side-chain atoms of the C-terminal helix of ACP2(2) in c relies, in part, on a reference NMR structure of a similar docking domain complex²⁰.

Our description of an asymmetric state for intermodular translocation of the growing polyketide chain contrasts with a previously reported symmetric state for the same reaction in a different assembly-line PKS⁸. Further studies will be required to establish whether these differences stem from technical considerations or if they reflect mechanistic divergence between related assembly lines.

Cryo-EM analysis of crosslinked DEBS modules 1 and 3

In addition to enabling structural characterization of the ACP2(2)–KS3AT3 complex, DBA crosslinking also yielded preparations of modules M1TE and M3TE bound to a monoclonal antibody fragment (F_ab) 1B2 (ref. 12) that were suitable for cryo-EM analysis (Supplementary Fig. 12). 1B2 did not affect the formation of bands 1 or 2 when included in DBA crosslinking reactions (Supplementary Fig. 16). A total of 9,523 and 10,480 dose-fractionated movies were collected for crosslinked M1TE and M3TE, respectively. After particle picking and two-dimensional (2D) classification, ab initio reconstruction without imposed symmetry yielded 3D volumes resembling the previous cryo-EM structures of DEBS M1TE and a hybrid module derived by fusing fragments of M3, M1 and the TE domain (M3/1TE)^12,27. Crosslinked M1TE contained two bound copies of 1B2 oriented symmetrically about the pseudo-C2 axis of symmetry, whereas crosslinked M3TE contained only a single 1B2—akin to the previous structure of M3/1TE bound to 1B2 (Supplementary Figs. 17–22). (We previously ascribed this difference to a 12° bending of the N-terminal coiled-coil docking domain relative to the pseudo-C2 axis of symmetry that precluded binding of the second F_ab (ref. 12).) Ab initio classes were refined to yield consensus cryo-EM maps of crosslinked M1TE-1B2 and M3TE-1B2 with GSFSC resolutions of 3.18 Å each (Supplementary Figs. 17–22). Overall, substantial conformational heterogeneity was observed in the C-terminal fragment of each module (harboring the KR, ACP and TE domains) (Supplementary Figs. 17–22).

Cryo-EM maps of crosslinked M1TE revealed ACP density in the cleft between the KS and AT domains (Supplementary Fig. 17), reminiscent of the State 1 pre-elongation mode¹². Also consistent with our earlier characterization of State 1 of this module, no more than one bound ACP was observed in any of the 3D classes. From these data, it also follows that, if a subunit harboring an intra-polypeptide crosslink (that is, band 2; Fig. 1b) exists in a dimeric state, then it must partner with an un-crosslinked subunit (that is, band 3; Fig. 1b).

To identify which cryo-EM maps of crosslinked M1TE corresponded to bands 1–3 (Fig. 1b), we inspected maps with observable ACPs to assign inter-modular or intra-molecular KS–ACP interactions. One map (3.73 Å) revealed an inter-molecular KS–ACP linkage and presumably corresponded to band 1 (Supplementary Fig. 23b). It also showed strong similarity to State 1 described earlier and was, therefore, assigned as crosslinked State 1. Another map (3.61 Å) supported an intra-molecular KS–ACP linkage whose subunit was partnered with an un-crosslinked subunit (Supplementary Fig. 23b), suggesting that band 2 and band 3 were, in fact, present in the same dimer (Fig. 1b).

Considerable disorder in the regions connecting the KS–AT core of crosslinked M3TE to its ACP domain prevented assignment of its cryo-EM maps to bands 1–3. Nonetheless, a single ACP clearly docking onto the active site of the KS could also be observed in crosslinked M3TE (Fig. 4a–c). Notably, Arg1439, Arg1440 and Glu1444 from ACP3 interacted with reciprocally charged residues Glu554, Glu518 and Arg480, respectively. The latter set of residues maps to the KS–AT linker subdomain (Fig. 4c,d). Previous mutagenesis studies established that these ACP residues play pivotal roles in controlling KS–ACP specificity during chain elongation (Fig. 4e)²⁴. (See Supplementary Table 1 for a summary of structures and their compositions from this study and our related previous study¹².)

Fig. 4 |. Structural analysis of crosslinked DEBS M3TE in complex with F_ab 1B2 (CL-M3TE-1B2).

a, Cryo-EM structure of CL-M3TE-1B2 deduced from one map (designated Cis F_ab/ACP; Supplementary Figs. 20 and 21) at 3.46-Å GSFSC resolution is shown in two 120°-related orientations. A similar cryo-EM map (designated Trans F_ab/ACP; Supplementary Figs. 20 and 22) was obtained at 3.40-Å GSFSC resolution; it contains an ACP bound to the opposite KS, indicating that F_ab 1B2 does not affect ACP binding. b, SDS-PAGE analysis of the CL-M3TE-1B2 sample containing bands 1–3 (Fig. 1b) that was used for structural analysis (Supplementary Fig. 12). c, Close-up view of the interactions between ACP3 and the KS–AT cleft of M3TE. Based on proximity to the observed backbone atoms, electrostatic bonds are likely to occur between the ACP and the KS–AT interdomain linker from the opposite subunit to which the ACP3 is covalently attached (magenta versus cyan). The cryo-EM map surface is displayed within a 3-Å radius relative to the ACP atoms, and density for the Ppant group was not observed at the set threshold value. Modeling of the Ppant and DBA crosslink atoms was not fully supported by the map density and is, therefore, tentative. d, Complementary electrostatic surfaces on the ACP (yellow) and KS–AT (magenta) are shown. e, Variants of ACP3 were previously tested for their ability to facilitate polyketide chain elongation in partnership with the KS3AT3 didomain²⁴. Almost all detectable activity was lost when the loop 1 of ACP3 (black arrow) was replaced with its counterpart from ACP6. However, when R1439 or R1440 was reintroduced into this construct individually, partial activity was restored. For gel source data, see Source Data Fig. 4.

Partial ACP occupancy during polyketide chain elongation

The repeated observation that only one KS–ACP interaction at a time is permitted in a homodimeric module led us to ask whether in trans crosslinking of the homodimeric KS–AT core of M3 (KS3AT3) with its elongation partner ACP3 would also be subject to this stoichiometric limit. As shown in Supplementary Fig. 11, DBA crosslinking of KS3AT3 and ACP3 under a range of ACP3 and citrate concentrations was unable to yield stoichiometric crosslinking of the two proteins. Cryo-EM analysis of the SEC-purified complex (Supplementary Fig. 12) produced a consensus cryo-EM map at 3.71-Å GSFSC resolution (Supplementary Figs. 24 and 25). It revealed the expected structural features of an extended KS3AT3 homodimer along with weak density corresponding to the ACP bound at each of the KS active site clefts (not visual at the threshold values used in Supplementary Figs. 24 and 25).

To quantify the relative abundance of particles containing zero, one or two ACPs, the data were subjected to a recently described statistical image analysis methodology for measuring inter-subunit variability at the level of individual particles²⁸. Specifically, each pseudo-C2-symmetric particle was segmented into two asymmetric subunits by C1-to-C2 symmetry expansion. Individual C2-segmented particles were then sorted into two 3D classes according to differences in ACP occupancy by employing a focused mask around their ACP binding sites. As expected, one 3D class revealed clear and unambiguous density for ACP3 in the ACP binding site, whereas the other was completely devoid of density in this site (Fig. 5a). Particles belonging to each class were then recombined with their symmetry mates to generate the intact C1-particles and then classified as KS3AT3 homodimers bound to zero, one or two ACP subunits. Strikingly, a near 1:2:1 particle ratio was observed between 3D classes with zero (25.0%), one (50.3%) or two (24.7%) ACPs (Fig. 5a), indicating that docking of the elongation ACP partner occurred stochastically in the dissociated (that is, KS–AT + ACP) state of a PKS module. In contrast, when an identical image analysis protocol was employed to quantify ACP occupancy of the crosslinked form of intact M1TE, only particles with one ACP were observed (Fig. 5b and Supplementary Fig. 26). Thus, covalent tethering of an ACP to the catalytic KS–AT core of a homodimeric PKS module precludes both KS active sites from catalyzing chain elongation at the same time.

Fig. 5 |. Measuring the ACP occupancy of individual particles by statistical cryo-EM image analysis.

The ACP occupancy of KS–AT homodimers was measured by a recently developed method (statistical per-particle image analysis²⁸) after crosslinking of KS3AT3 + ACP3 (yellow circle) (a) or intact module 1 (b). The boxed inset in a shows SDS-PAGE analysis of the crosslinked material isolated as a single peak by SEC for cryo-EM analysis (Supplementary Fig. 12). a,b, Consensus particles were (1) symmetry expanded about the pseudo-C2 axis of symmetry (blue dot) and (2) sorted according to differences in density at the ACP binding sites by 3D focused classification. Particles from each class were correlated to their C2 symmetry mates to categorize intact particles as bound to zero, one or two ACPs (Methods). As elaborated in Supplementary Fig. 26, slight differences in the ‘1 ACP’ classes in b resulted in their segregation during 3D focused classification. Red asterisks denote KS active sites harboring a quenched DBA moiety (Supplementary Fig. 5a). For gel source data, see Source Data Fig. 5.

Our observation that no more than one KS domain of a homodimeric module can crosslink with its elongation ACP partner lends credence to a previous proposal that both ACPs co-migrate between two equivalent reaction chambers during the catalytic cycle¹³. To explain this phenomenon, Bagde et al.¹³ pointed out that C-terminal dimeric motifs—which are frequently encountered in assembly-line PKS modules—might hold the ACPs in proximity such that only one of them can partake in elongation at a given time. It should be noted that our crosslinking analysis was performed on PKS modules bearing C-terminal TE domains that are homodimeric²⁹.

Statistical image analysis of the inter-modular interface

Compelled by successful application of statistical per-particle image analysis to cryo-EM data for the crosslinked product of KS3AT3 and ACP3, we also sought to establish whether two upstream (translocation) ACP partners or a translocation and an elongation ACP partner could simultaneously dock onto the homodimeric KS-AT core of a module. To do so, the cryo-EM data from Supplementary Fig. 13 were subjected to the same computational protocol outlined above, modified slightly to allow for the simultaneous docking of two identical or different ACPs onto the KS–AT homodimer. Overall, no homo-dimeric or hetero-dimeric states were forbidden in our analysis (Supplementary Fig. 27). Thus, in contrast to the inability of the KS–AT core of a homo-dimeric PKS module to accommodate both elongation ACP domains simultaneously, there does not appear to be a barrier to concomitant chain translocation into both KS active sites or to synchronous translocation and elongation reactions. Our results, however, do not distinguish how a PKS module interacts with its ACPs in the presence of protein-tethered intermediates during the catalytic cycle.

Discussion

Since the first report of a selective protein–protein interaction at the interface between two successive modules of an assembly-line PKS¹⁹, there has been a steady growth of evidence that protein–protein interactions play dominant roles in channeling reactive intermediates along the enzymatic assembly line³⁰. Our cryo-EM structure of ACP2(2) docked to the KS–AT core of module 3 of DEBS (Fig. 3) represents a near-atomic resolution snapshot of an inter-modular interface in an assembly-line PKS, thereby establishing a foundation for the structure-based engineering of hybrid assembly lines. Encouragingly, the observed KS–AT/ACP interfaces involving both ACP2(2), which participates in chain translocation to module 3, and ACP3, which participates in chain elongation catalyzed by module 3, are well supported by empirical data from mutagenesis and other approaches^23–25. Although these earlier studies flagged the ACP residues interacting with the KS–AT core, the corresponding residues on the KS–AT side of the interface remained unknown until now.

The distinct binding modes through which ACPs are recognized during polyketide chain translocation and elongation can be readily visualized by comparing our structures of DEBS module 3 (Supplementary Fig. 28a,b). Whereas ACP3 binds a relatively small portion of the KS–AT cleft during intra-modular elongation (buried surface area ≈ 1,700 Ų), ACP2(2) forms extensive interactions in the cleft while forming secondary contacts with the docking domain during inter-modular translocation (total buried surface area ≈ 3,600 Ų). Although actual KS–AT/ACP binding constants (estimated to be in the low-mid micromolar range, based on Supplementary Figs. 10a and 11a and a previous analysis³¹) cannot be deduced from our crosslinking analysis, these experiments confirm that ACP2(2) binds more favorably than ACP3 to the KS–AT cleft when both ACPs were co-incubated at equimolar concentrations (Fig. 2). In this case, tighter recognition of the upstream ACP across a module–module junction is probably a reflection of the increased demand to overcome the entropic burden of a bimolecular substrate transfer. In fact, a recent study found that KS domains from PKS assembly lines genetically co-migrate with their upstream, not downstream, ACP partners^32,33. Although these evolutionary insights reinforce the importance of the ACP–KS partnership during translocation—and can be harnessed to pinpoint appropriate cut-sites for PKS module exchange^18,34,35—our 3.05-Å structure of a module–module interface provides the molecular blueprints from which to rationally build and test hybrid assembly lines capable of generating designer polyketides. (For a summary of the structures of DEBS reported here and previously¹², see Supplementary Table 1 and Supplementary Fig. 28.)

Our cryo-EM analysis of DEBS also underscores the importance of structural asymmetry in a homo-dimeric PKS module as it progresses through its catalytic cycle while shining light on its mechanistic origins. Notably, the putatively symmetric module state, where both elongation ACP partners are simultaneously docked to the KS–AT core of the module, is strongly disfavored. In contrast, when the KS–AT core of a homo-dimeric module is genetically dissociated from its elongation ACP partner, the dual-occupancy state is readily observed. Similarly, simultaneous crosslinking of one KS domain to its elongation ACP partner and the other KS to its translocation ACP partner is also feasible, as is the symmetric dual-occupancy state where both translocation ACP partners simultaneously dock onto their downstream KS–AT core. It must, nonetheless, be recognized that DBA crosslinking is incapable of capturing module states whose formation depends on exergonic chain elongation. Such states, wherein both KS active sites are impervious to acyl-ACP substrates, were, in fact, previously described³⁶ and characterized by cryo-EM¹². Future studies will be required to understand how and when translocation events occur in relation to elongation-induced KS active site gating.

Our observation of an ACP docked to only one of two KS–AT clefts in the asymmetric chain elongation state of an intact PKS module may reflect the presence of a C-terminal homo-dimeric TE domain attached to this module¹³. Similar crosslinking analysis of an architecturally related mammalian FAS, which harbors a C-terminal monomeric TE, resulted in simultaneous ACP crosslinking in a C₂-symmetric fashion¹⁶. Further comparative analysis is warranted to establish whether this asymmetric feature does differentiate these homologous multi-enzyme systems.

The hallmark of an assembly-line PKS lies in the ability of each of its constituent enzymatic modules to toggle between intra-modular elongation and inter-modular translocation reactions involving the growing polyketide chain. Here we described an asymmetric state associated with the latter reaction in which only one monomer of a homo-dimeric acceptor module engages with a single upstream ACP at a time. Taken together with our recent structural characterization of a module involved in the former reaction¹², the catalytic relevance of asymmetry in KS–ACP interactions is clear. Our present report also includes two methodological advances: the use of DBA as a bifunctional crosslinker to stabilize transient PKS states for structural analysis and the use of statistical per-particle image analysis of cryo-EM data to analyze asymmetric conformations of C₂-symmetric PKS homodimers. We anticipate broad use of both methods in future studies of polyketide biosynthetic mechanisms.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41589-024-01709-y.

Methods

DBA crosslinking reactions

General DBA crosslinking.

Before use in crosslinking reactions, DBA was filtered through anhydrous aluminum oxide under argon gas and diluted to 0.5 M in anhydrous DMF for storage at −80 °C. DBA was then diluted to 0.25 mM in 50% aqueous DMF for initiating crosslinking reactions no more than 15 min prior. In three identical replicates, SEC-purified DEBS M1TE (see Supplementary Tables 4–11 for protein sequences) (10 μM) was incubated in 10 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), 100 μM tris(2-carboxyethyl) phosphine (TCEP) and 200 mM citric acid, pH 7.2 (NaOH) at 20 °C for 15 min before adding 12.5 μM DBA (reaction volume = 25 μl; reaction concentrations reflect their final concentrations after all components were added). Reactions were quenched after 30 s by the addition of an equal volume of 2× Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis (Fig. 1b and Supplementary Fig. 3a), using either 3–8% TA or 4–20% TGX polyacrylamide gels (3–8% TA gels provided better separation between bands 1 and 2 than 4–20% TGX gels; see Supplementary Figs. 3 and 4a,b for comparison).

Kinetics of DBA crosslinking and citrate dependence thereof.

SEC-purified DEBS M3TE (5 μM) was incubated in 10 mM HEPES, 100 μM TCEP and 50 mM, 200 mM or 500 mM citric acid, pH 7.2 (NaOH) at 20 °C for 15 min before adding 7.5 μM DBA (reaction volume = 25 μl; reaction concentrations reflect their final concentrations after all components were added). Reactions were quenched at three different timepoints (5 s, 1 min or 25 min) by the addition of an equal volume of 2× Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis (Supplementary Fig. 3b).

Requirement of an ACP domain for DBA crosslinking of M1TE.

DEBS M1TE prepared in its apo and holo states (see ‘Protein expression and purification’ in the Supplementary Information) was used in side-by-side DBA crosslinking reactions to measure the dependence of bands 1 and 2 formation on the presence of the Ppant co-factor of the ACP. M1TE (10 μM) was incubated in 10 mM HEPES, 100 μM TCEP and 200 mM citric acid, pH 7.2 (NaOH) at 20 °C for 15 min before adding 6.3 μM, 12.5 μM or 25 μM DBA (reaction volume = 20 μl; reaction concentrations reflect their final concentrations after all components were added). Reactions were quenched after 0.5 min, 5 min or 15 min by the addition of an equal volume of 2× Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis (Supplementary Fig. 4a).

Requirement of a KS domain for DBA crosslinking of M1TE.

DEBS M1TE (31 μM) was incubated with 3.1 mM cerulenin or DMSO (control) in 0.1 M NaH₂PO₄, pH 8 (NaOH) at 20 °C for 1 h. Residual cerulenin and DMSO were removed using 7 kDa molecular weight cut-off (MWCO) Zeba spin desalting columns (Thermo Fisher Scientific, 89882) equilibrated with 10 mM HEPES, 200 mM citric acid, pH 7.2 (NaOH). Cerulenin-treated and -untreated M1TE (10 μM) were incubated in 10 mM HEPES, 100 μM TCEP and 200 mM citric acid, pH 7.2 (NaOH) at 20 °C for 15 min before adding 6.3 μM, 12.5 μM or 25 μM DBA (reaction volume = 10 μl; reaction concentrations reflect their final concentrations after all components were added). Reactions were quenched after 1.2 min or 8 min by the addition of an equal volume of 2× Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis. To ensure that cerulenin inhibited M1, DEBS (5)M1(2) was treated with cerulenin or DMSO (control) and desalted in the same manner as for M1TE (Supplementary Fig. 4b).

Requirement of both KS and ACP domains for DBA crosslinking of KS3AT3.

Wild-type or C202A variant KS3AT3 (5 μM) and apo-form or holo-form ACP2(2) (25 μM) were incubated in 10 mM HEPES, 100 μM TCEP and 200 mM citric acid, pH 7.2 (NaOH) at 20 °C for 15 min before adding 10 μM DBA (reaction volume = 20 μl; reaction concentrations reflect their final concentrations after all components were added). Reactions were quenched after 1 min by the addition of an equal volume of 2× Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis (Supplementary Fig. 4c).

DBA crosslinking of M3TE with fluorescent ACP2(2).

See below for the preparation of the fluorescein-labeled ACP2(2) probe, ACP2(2)-FL. SEC-purified DEBS M3TE (5 μM) was incubated in 10 mM HEPES, 100 μM TCEP and 150 mM, 260 mM or 510 mM citric acid, pH 7.2 (NaOH) at 20 °C for 15 min before adding 10 μM DBA (reaction volume = 20 μl; reaction concentrations reflect their final concentrations after all components were added). Reactions were quenched after 4.5 min by the addition of an equal volume of 2× Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis. In-gel fluorescence measurements were made using a ChemiDoc MP Imaging System (Bio-Rad) set to the fluorescein excitation and emission settings (Supplementary Fig. 7).

Crosslinking before single-particle cryo-EM analysis.

To scale-up the crosslinked M1TE and M3TE products for single-particle cryo-EM analysis, we prepared 5–16 replicate DBA crosslinking reactions and pooled them after quenching to avoid potential mixing effects on crosslinking efficiency due to increased reaction volume. Each reaction was carried out as above (see ‘General DBA crosslinking’ subsection) with slight adaptations. That is, M1TE or M3TE (40 μM) were incubated with 10 mM HEPES, 200 μM TCEP and 200 mM citric acid, pH 7.2 (NaOH) for 15 min before adding 50 μM DBA (reaction volume = 60 μl × 5, reaction scale = 12 nmol). KS3AT3 (20 μM) and ACP3 (235 μM) were incubated with 10 mM HEPES, 100 μM TCEP and 450 mM citric acid, pH 7.2 (NaOH) for 15 min before adding 30 μM DBA (reaction volume = 60 μl × 16, reaction scale = 19.2 nmol). Finally, KS3AT3 (20 μM), ACP2(2) (100 μM) and ACP3 (200 μM) were incubated with 10 mM HEPES, 100 μM TCEP and 450 mM citric acid, pH 7.2 (NaOH) for 15 min before adding 30 μM DBA (reaction volume = 60 μl × 16, reaction scale = 19.2 nmol). Reactions were quenched after 30 s by the addition of an equal volume of 2× Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis. Reaction concentrations reflect their final concentrations after all components were added. Whereas crosslinked M1TE and M3TE were SEC purified in 0.1 M citric acid, 0.1 M NaCl, 10 mM HEPES, pH 7.2 (NaOH) (Supplementary Fig. 12a,b), crosslinked KS3AT3–ACP3 and KS3AT3–ACP2(2)/ACP3 were SEC purified in 0.3 M citric acid, 0.1 M NaCl, 10 mM HEPES, pH 7.2 (NaOH) (Supplementary Fig. 12c–e).

Crosslinking of KS3AT3 with holo-form and apo-form ACP3.

SEC-purified DEBS KS3AT3 (5 μM) was incubated with 5 μM, 50 μM or 250 μM holo-form or apo-form ACP3 in 10 mM HEPES, 100 μM TCEP and 50 mM, 200 mM or 530 mM citric acid, pH 7.2 (NaOH) at 20 °C for 15 min before adding 7.5 μM DBA (reaction volume = 20 μl; reaction concentrations reflect their final concentrations after all components were added). Reactions were quenched at two different timepoints (2 min and 10 min) by the addition of an equal volume of 2× Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis. A control reaction was also included that was prepared in the same manner as above; however, apo-ACP3 (0.5 mM) was first treated with 1 mM bismaleimidoethane for 15 min in 200 mM citric acid, pH 7.2 (NaOH), followed by the addition of 3 mM BME and buffer exchanged using a 7-kDa MWCO Zeba spin desalting column (Thermo Fisher Scientific, 89882) equilibrated with a similar buffer (Supplementary Fig. 11).

Crosslinking of KS3AT3 with holo-form and apo-form ACP2(2).

ACP2(2) was prepared in three different ways before crosslinking with KS3AT3: (1) apo-form ACP2(2) (500 μM) was incubated with 4 mM coenzyme A (CoA), 10 mM MgCl₂ and 50 μM Sfp in 0.1 M NaH₂PO₄ pH 7.2 (NaOH) for 1 h (reaction volume = 25 μl; reaction concentrations reflect their final concentrations after all components were added) before buffer exchanging using a 7-kDa MWCO Zeba spin desalting column (Thermo Fisher Scientific, 89882) equilibrated with 0.1 M NaH₂PO₄ pH 7.2 (NaOH) to afford the holo-form ACP2(2) ‘ACP2(2) / Sfp + CoA’; (2) apo-form ACP2(2) (500 μM) was incubated with 10 mM MgCl₂, 50 μM Sfp and without exogenous CoA in 0.1 M NaH₂PO₄ pH 7.2 (NaOH) for 1 h (reaction volume = 25 μl; reaction concentrations reflect their final concentrations after all components were added) before buffer exchanging using a 7-kDa MWCO Zeba spin desalting column (Thermo Fisher Scientific, 89882) equilibrated with 0.1 M NaH₂PO₄ pH 7.2 (NaOH) to afford the apo-form ACP2(2) ‘ACP2(2) / Sfp CoA’ (however, this sample was later determined to be not entirely apo due to co-purification of Sfp with CoA from E. coli BL21(DE3); Supplementary Fig. 10); and (3) a final form of apo-ACP2(2) was not treated with any of the above reaction components as in (1) and (2) and used directly in DBA crosslinking reactions with KS3AT3 after SEC purification. SEC-purified DEBS KS3AT3 (5 μM) was incubated with 5 μM, 50 μM or 250 μM of the above ACP2(2) forms (1–3) in 10 mM HEPES, 100 μM TCEP and 50 mM, 200 mM or 530 mM citric acid, pH 7.2 (NaOH) at 20 °C for 15 min before adding 7.5 μM DBA (reaction volume = 20 μl; reaction concentrations reflect their final concentrations after all components were added). Reactions were quenched at various timepoints (1.7 min, 9 min or 10 min) by the addition of an equal volume of 2× Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis (Supplementary Fig. 10).

Inhibition of DEBS M1 by cerulenin

To a 15.3-μl mixture of 457 mM NaH₂PO₄, 6.5 mM TCEP, 13 mM MgCl₂, 7.8 mM ATP, 3 μM PrpE, 3 μM MatB, 5 μM SCME and 5 μM LDD(4) was added 4.1 μl of 19 μM cerulenin-treated or -untreated (5)M1(2) (prepared in the same manner as above for M1TE), 15 μM (3)M2TE and 7.3 μM CoA. The reactions were initiated by the addition of 0.6 μl of a mixture containing 35.7 mM sodium propionate, 35.7 mM methylmalonic acid and 28.6 mM NADPH to arrive at a final reaction volume of 20 μl and the following final reaction concentrations: 350 mM NaH₂PO₄, 5 mM TCEP, 10 mM MgCl₂, 6 mM ATP, 2 μM PrpE, 2 μM MatB, 4 μM SCME, 4 μM LDD(4), 4 μM cerulenin-treated or -untreated (5)M1(2), 3 μM (3) M2TE, 1.5 mM CoA, 1 mM sodium propionate, 1 mM methylmalonic acid and 0.8 mM NADPH. The 20-μl reactions were transferred to 384-well clear-bottom plates (Corning, 3765) to measure depletion of NADPH absorbance at 340 nm every 10 s using a BioTek Synergy HT plate reader at 20 °C (Supplementary Fig. 4d).

Preparation of fluorescein-labeled ACP2(2) (ACP2(2)-FL)

A 50 mM stock solution of fluorescein isothiocyanate (FITC) was freshly prepared in anhydrous DMSO. To 250 μM apo-ACP2(2) in 0.3 M NaH₂PO₄ pH 8 (NaOH) was added 1, 2.5, 5 or 10 equivalents of FITC (that is, four different 50-μl reactions). The reactions were terminated by removing unreacted FITC using 7-kDa MWCO Zeba spin desalting columns (Thermo Fisher Scientific, 89882) equilibrated with 0.1 M NaH₂PO₄ pH 7.2 (NaOH). The eluent containing fluorescein-labeled apo-ACP2(2) (apo-ACP2(2)-FL) was 4′-phosphopantetheinylated by incubating 209 μM apo-ACP2(2)-FL with 20.9 μM Sfp, 3 mM CoA and 10 mM MgCl₂ in 0.1 M NaH₂PO₄ pH 7.2 (NaOH) for 1 h at 20 °C to generate holo-form ACP2(2)-FL (referred to throughout as ‘ACP2(2)-FL’). To measure the extent of fluorescein labeling, the absorbance at 500 nm, corresponding to the experimental maximum absorbance of ACP2(2)-FL, was measured using a NanoDrop 2000c (Thermo Fisher Scientific) to calculate the dye-to-protein ratios for each ACP2(2)-FL sample. A maximum dye-to-protein ratio of 0.25 was measured (that is, corresponding to conjugation reactions in which apo-ACP2(2) was incubated with 10 equivalents of FITC).

Limited trypsinolysis of DEBS M3TE

We used a previously identified³⁷ trypsin-labile site at the KR–ACP junction of DEBS M3 to characterize the crosslinked species (bands 1 and 2) by limited trypsinolysis. Trypsin was prepared at a concentration of 0.5 mg ml⁻¹ in 50 mM sodium acetate pH 5 (NaOH) for long-term storage at −80 °C and subsequently diluted to 50 μg ml⁻¹ in 0.1 M NaH₂PO₄ pH 7.2 (NaOH) before usage in trypsinolysis experiments. M3TE (5 μM) and CL-M3TE (5 μM) were trypsinized at 20 °C in the presence of 5 μg ml⁻¹ trypsin in 100 mM citric acid, 50 mM BME, 5 mM HEPES, 50 mM NaH₂PO₄, pH 7.2 (NaOH). Trypsin was inactivated after 1 min, 4 min, 10 min and 16 min by the addition of an equal volume of 2x Laemmli buffer (Bio-Rad, 1610737) supplemented with 5% (ν/ν) BME and heated at 95 °C for 2 min before SDS-PAGE analysis. Based on the panel of putative crosslinked structures, we predicted, a priori, the expected tryptic products, assuming a single cleavage event at the equivalent trypsin-labile sites in each subunit (Supplementary Fig. 8). Comparison of the tryptic products in the DBA-crosslinked versus un-crosslinked M3TE revealed transient species consistent with band 1 corresponding to an asymmetric, singly crosslinked dimer (that is, State 1) (Supplementary Fig. 9). We considered band 2 to be a self-crosslinked monomer, in accordance with its expected mobility relative to a singly crosslinked dimer (Fig. 1b and Supplementary Fig. 5). This assignment was further supported by fluorescent probe crosslinking (Supplementary Figs. 6 and 7) and cryo-EM analysis (Supplementary Figs. 17–19 and 23).

Isolation of (un)crosslinked module + F_ab 1B2 complexes

All DEBS modules (crosslinked or un-crosslinked) and F_ab 1B2 used in cryo-EM experiments were individually purified by SEC before preparation and re-purification of the module–F_ab complexes. Module–1B2 complexes were prepared by adding 1.5 equivalents of 1B2 heterodimer per equivalent of PKS monomer, in accordance with the binding stoichiometry³⁸, and incubated on ice for 30 min before isolating the complex by SEC. Protein complex samples (≤2 ml) were injected onto a 120-ml Superdex 200 pg 16/600 column (Cytiva) at a flow rate of 1.5 ml min⁻¹ and fractionated into 3 ml isocratically with 0.1 M citric acid, 0.1 M NaCl, 10 mM HEPES, pH 7.2 (NaOH) (SEC buffer) using an ÄKTA Pure protein purification FPLC system (Cytiva) (Supplementary Fig. 12a,b).

Isolation of crosslinked KSAT–ACP complexes

Protein complex samples (≤2 ml) were injected onto a 120-ml Superdex 200 pg 16/600 column (Cytiva) at a flow rate of 1.5 ml min⁻¹ and fractionated into 3 ml isocratically with SEC buffer containing 0.3 M citric acid using an ÄKTA Pure protein purification FPLC system (Cytiva) (Supplementary Fig. 12c–e).

Cryo-EM sample preparation and data collection

Crosslinked and un-crosslinked DEBS module–F_ab complexes were concentrated to 5–10 mg ml⁻¹ using Amicon Ultra Centrifugal Filters (50-kDa MWCO) before adding 0.03% nonyl phenoxypolyethoxylethanol (NP-40) and applying 3 μl onto glow-discharged 300-mesh R 2/1 Quantifoil copper grids. The grids were blotted for 4 s at 4 °C and 100% relative humidity and vitrified in liquid ethane using a Vitrobot Mark IV (Thermo Fisher Scientific). Grids of crosslinked KSAT–ACP complexes were prepared in the same way, except that they were diluted from 0.3 M to 0.1 M citric acid before vitrification. Samples were imaged at 300-kV accelerating voltage with a Titan Krios G3i cryo-electron microscope (Thermo Fisher Scientific) equipped with a K3 (Gatan) or Falcon 4 (Thermo Fisher Scientific) direct-electron detector (DED) and BioQuantum or Selectris energy filters at nominal magnifications of ×81,000 or ×130,000, corresponding to a calibrated sampling of 1.1 Å per pixel or 0.946 Å per pixel, respectively. EPU software (Thermo Fisher Scientific) was used to record dose-fractionated movies composed of 40 individual frames in Lzw non-gain-normalized .tiff format (with Gatan K3 detector) or gain-normalized .mrc format (with Falcon 4 detector) and with a total dose of 50 electrons and dose rates of 6.9, 16.1 and 21.7 e⁻ pixel⁻¹ s⁻¹ (see Supplementary Table 2 for details).

Cryo-EM image processing and 3D reconstruction

Single-particle cryo-EM image analysis was carried out in Relion³⁹ and cryoSPARC⁴⁰, and statistical per-particle image analysis was implemented in EMAN2 (ref. 41) (see Supplementary Figs. 13 and 14, 17–22 and 24–27 for details). In all cases, dose-fractionated image stacks were applied to motion correction, dose weighting and contrast transfer function (CTF) estimation before reference-free automated particle picking. In some cases, initial particles were used to generate templates by 2D classification followed by template-based particle picking. The full particle batches were applied directly to ab initio reconstruction in cryoSPARC for CL-M1TE-1B2 and CL-M3TE-1B2 datasets, whereas iterative rounds of 2D classification were implemented before ab initio reconstruction for CL-KS3AT3-ACP3 and CL-KS3AT3-ACP2(2)/ACP3 datasets. C1 symmetry was specified for all ab initio reconstructions and in subsequent homogenous/heterogeneous refinements. Particles were transported from Relion to cryoSPARC via ‘import particle stack’ in cryoSPARC, whereas particles in cryoSPARC format (.cs) were converted to .star format for transport into Relion via the csparc2star. py script in pyem⁴².

Model building and refinement

Existing structural models of DEBS M1 (Protein Data Bank (PDB) 7M7F), DEBS M2 (PDB 2JU2), DEBS M3 (PDB 6C9U) and F_ab 1B2 (PDB 6C9U) were fit as rigid bodies into their corresponding cryo-EM maps using ChimeraX⁴³ and then manually refined in Coot⁴⁴ and automatically refined in Phenix⁴⁵ using real-space refinement⁴⁶. For model components without a reference structure, template-based models were generated using SWISS-MODEL⁴⁷ or AlphaFold2 (ref. 48) and then fit into their corresponding cryo-EM maps and refined as above. The Ppant co-factor was modeled by substituting the Ser residue that becomes 4′-phosphopantetheinylated with a 4HH residue in Coot. The DBA crosslink could not be observed in any of the cryo-EM maps and was, therefore, not modeled. Sequence conflicts between the reference and PDB entries arose due to differences between the wild-type and recombinant proteins used for experimentation (for example, N-terminal and C-terminal tags for purification and artificial domain fusions explained in the Results section).

Statistical per-particle image analysis

Crosslinked KS3AT3 + ACP3 / crosslinked M1TE.

All particles were first aligned to the pseudo-C2 symmetry axis and then segmented into two symmetrically related asymmetric units. A soft spherical mask centered at the ACP binding site was used for focused 3D classification into the two classes shown in Fig. 5. C2-particles were regrouped into their original C1-particles by symmetry-mate pairing and classified as bound to zero, one or two ACPs based on the class assignments of their constituent asymmetric units (Fig. 5).

Crosslinked KS3AT3 + ACP2(2) + ACP3.

All particles were first aligned to the pseudo-C2 symmetry axis and then segmented into two symmetrically related asymmetric units. A soft spherical mask centered at the ACP binding site was used for focused 3D classification into the four classes shown in Supplementary Fig. 27a. C2-particles were regrouped into their original C1-particles by symmetry-mate pairing and classified as one of 16 possible classes based on the class assignments of their constituent asymmetric units. Due to different particle counts in each of the four classes (Supplementary Fig. 27b), we were unable to assess inter-subunit coupling in the same way as with KS3AT3 + ACP3 (Fig. 5a).

Reporting summary

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

Data availability

All atomic coordinates and cryo-EM maps have been deposited in the Protein Data Bank (PDB) under accession codes 8TJN, 8TJO, 8TPW, 8TPX, 8TJP and 8TKO and in the Electron Microscopy Data Bank under accession codes EMD-41305, EMD-41306, EMD-41495, EMD-41496, EMD-41307 and EMD-41355 (also declared in the authors’ Reporting Summary). Coordinates for model building were obtained from the PDB via accession codes 7M7F, 2JU2 and 6C9U. All materials used in this study that are not commercially available will be made available by the authors upon reasonable request. Source data are provided with this paper.