Engineering an Extremely Hybrid PKS for Adipic Acid Production Authors: Sarah H. Klass, Zilong Wang, Nathan Lanclos, Mia Wesselkamper, Chenyi Li, Justin Baerwald, Allison Pearson, Makenzi Thomas, Weixi Hu, Matthias Schmidt, Yan Chen, Jennifer W. Gin, Christopher J. Petzold, Jay D. Keasling Published: 2026 ABSTRACT Polyketide synthases (PKSs) are modular enzymes with exceptional potential as biocatalysts for producing non-native compounds. Here, we report the first PKS-based pathway for adipic acid (AA), an industrial monomer for nylon production, by engineering one of the most extensively hybridized PKS systems to date. Using a retrobiosynthetic approach, we identified EtnB, a succinyl-CoA-loading module that uniquely retains the terminal carboxyl group, enabling access to dicarboxylic polyketide products, rarely produced by canonical PKSs. EtnB was coupled to a fully reducing extension module through an engineered communication linker, which improved ACP–KS interactions, enhanced titers, and demonstrated selective succinyl-CoA loading in vivo. This construct integrates genes from five organismswith domains from seven PKS modules joined across six non-natural junctionsand functions in both Escherichia coli and Pseudomonas putida. Additional engineering that included AT domain exchanges, optimization of extender unit supply, and host strain metabolic rewiring further increased AA production. Together, this work demonstrates that highly chimeric PKSs can be rendered functional through rational design, expands the PKS toolkit with a carboxyl-retaining loading module, and establishes a versatile platform for engineering diacids and other noncanonical products through PKS pathways. Introduction Polyketide synthases (PKSs) are modular, multidomain enzymes that function as biosynthetic “assembly lines,” catalyzing the stepwise condensation of acyl-CoA precursors into structurally diverse natural products. − Their modular architecture, in which each domain contributes a discrete chemical transformation, makes PKSs a versatile platform for the creation of new-to-nature polyketide products for therapeutic and industrial applications. − Engineering efforts have increasingly focused on hybrid PKSs, created by combining domains and/or modules from different systems to access new chemical space. However, the functionality of such chimeric systems often depends on optimizing domain junctions to ensure efficient substrate transfer. − This design challenge has highlighted both the promise and complexity of developing hybrid PKSs for synthetic biology. Beyond the biosynthesis of complex natural product scaffolds, modular PKSs offer an untapped opportunity to construct industrially relevant monomers de novo. Adipic acid (AA), a key monomer in nylon production at million-ton scale, is traditionally manufactured via the nitric acid (HNO3) oxidation of a cyclohexanol–cyclohexanone mixture, a petrochemical process that emits substantial amounts of the greenhouse gas N2O. Developing a sustainable microbial route to AA therefore remains an important goal in green chemistry and industrial biotechnology. Although several enzymatic pathways for AA biosynthesis have been explored, − these routes are typically single-purpose and lack the flexibility of modular PKS systems, which could enable the programmable assembly of longer, branched, or otherwise functionalized diacids. In principle, a PKS-based approach could allow the systematic design of AA analogs and extended-chain dicarboxylic acids through rational domain and module exchanges. Previous work demonstrated that highly engineered PKS extension modules could generate adipic acid in vitro. However, a PKS-dependent pathway for AA biosynthesis has not yet been demonstrated in vivo. A key limitation is that most conventional PKS loading modules decarboxylate their acyl-CoA substrates through the action of KSQ domains , or a GNAT-like mechanism, eliminating the terminal carboxyl group needed for the biosynthesis of dicarboxylic acids. To overcome this limitation, we identified a succinyl-CoA loading module, EtnB, from the Etnangien PKS pathway in Sorangium cellulosum, which uniquely retains the carboxyl group during PKS chain initiation. We further characterized the substrate specificity of EtnB and incorporated it into a hybrid PKS system, coupling it to a fully reducing extension module using a rationally designed communication linker to improve ACP–KS compatibility. The resulting assembly, domains of seven modules joined across six engineered junctions, represents one of the most extensively engineered hybrid PKS systems to date and enabled the first in vivo demonstration of AA production by an engineered PKS. Importantly, this pathway functions in both Escherichia coli and Pseudomonas putida, organisms with tolerance to high adipic acid levels, , establishing two distinct microbial chassis for AA biosynthesis. Together, the extreme hybridization in this system arises from the combination of a trans-acting succinyl-CoA loading module, an engineered heterologous ACP–KS communication interface, and a multiorigin extension module assembled across six non-native junctions and evaluated in multiple configurations. These advances expand the PKS toolkit with a carboxyl-retaining loading module and establish generalizable design principles for constructing hybrid PKSs capable of accessing diacids and other noncanonical polyketide products. Results Identification of a Succinyl-CoA Loading Module To enable PKS-based AA biosynthesis, we sought a loading module (LM) capable of incorporating carboxylated precursors without the decarboxylation characteristic of canonical PKSs. Conventional malonyl-CoA (M-CoA) and methylmalonyl-CoA (Mm-CoA) LMs contain β-ketoacyl synthase Q (KSQ) domains that remove the terminal carboxyl group, preventing dicarboxylic acid formation. , Notable exceptions include the Borrelidin PKS, which incorporates trans-cyclopentane-1,2-dicarboxylic acid; the Aurantinin PKS, which protects the carboxyl group via methylesterification; and the Gladiolin/Etnangien , PKSs, which have been proposed and shown in vitro to load succinyl-CoA (S-CoA) (Figure a,b) through the action of trans-acting acyl transferases (ATs) GdnB and EtnB. However, no kinetic or in vivo studies have confirmed their substrate specificities in physiologically relevant conditions. Identification of EtnB as a succinyl-CoA loading module. (a) Examples of noncanonical PKSs that preserve the carboxyl group (red): Borrelidin (incorporates trans-cyclopentane-1,2-dicarboxylic acid), Aurantinin (methylester protection), and Gladiolin, which have been proposed to load S-CoA via trans-acting ATs. (b) A simplified representation of the Etnangien biosynthetic pathway highlighting the two predicted trans-acting ATs, EtnB and EtnK. The proteins encoded by etnD and etnB are colored purple, while EtnK is shown in blue. The distinct functional domains are illustrated by large circles: ketosynthase (KS), acyl carrier protein (ACP), ketoreductase (KR), enoyl reductase (ER), and thioesterase (TE). To enhance clarity, smaller gray and purple circles denote multiple elongation modules, including various KS, KR, dehydratase (DH), methyltransferase (MT), and ACP domains. (c) Phylogenetic tree of trans-acting AT domains showing that etnB from the Etnangien PKS is present in the S. cellulosum cluster (red) and groups closely with trans ATs from the B. gladioli cluster (yellow). Neighbor-joining tree constructed from ClustalW2 multiple-sequence alignment of EtnB, EtnK, and representative trans-acyltransferase (AT) domains; sequences were obtained by BLASTp search against the NCBI nr database (>30% identity). Bootstrap values were omitted for clarity. (d) Subset phylogenetic analysis of ATs derived from the Etnangien (EtnB, EtnK), Gladiolin (GbnB, GbnN), and Aurantinin (Art2, Art9) biosynthetic gene clusters along with the known M-CoA loading AT (FabD) from E. coli. The ATs fall into separate clades, one associated with M-CoA loading and the other hypothesized to be associated with S-CoA loading. Genomic analysis of the Etnangien biosynthetic gene cluster , reveals that it encodes two trans ATs, EtnB and EtnK (Figure b). We performed a BLASTp search using EtnB from the Etnangien PKS cluster as a query and constructed a neighbor-joining phylogeny from the resulting sequences. Phylogenetic analysis of these ATs alongside a comprehensive set of known trans-acting AT domains results in close clustering with genes from strains of Burkholderia gladioli, the producer of Gladiolin, which is structurally similar to Etnangien (Figure c). Analysis across these gene clusters revealed that each encoded at least two trans-acting ATs that separated into distinct phylogenetic clades: one associated with M-CoA loading, which clustered with the known M-CoA loading AT, FabD from E. coli, and another cluster presumably with S-CoA loading functionality (Figure d). Based on these findings and the prior literature, we hypothesized that EtnB could function as an S-CoA loading module suitable for AA biosynthesis. In Vitro Characterization of EtnB Substrate Specificity We next characterized EtnB substrate specificity. Intact protein mass spectrometry (IP-MS) showed that EtnB efficiently loaded multiple acyl-CoA derivatives, including S-CoA, onto its active-site serine, releasing free CoA (Figure a). Transfer of M-CoA, S-CoA, and glutaryl-CoA (G-CoA) to the cognate ACP EtnD yielded mass shifts of 86, 100, and 114 Da, respectively, confirming that all three substrates were transferred with retention of the terminal carboxylate (Figure b). In vitro characterization of EtnB substrate specificity. (a) IP-MS showing EtnB (m.w. = 35557 Da) loading of diverse acyl-CoA substrates onto its active-site Ser89 with release of free CoA. Peaks corresponding to the resulting mass shift due to loaded acyl-CoA substrates are annotated in red. Ac-CoA is acetyl-CoA; n-Prop-CoA is propionyl-CoA; n-Butyl-CoA is butyryl-CoA; iso-Butyl-CoA is isobutyryl-CoA; Mm-CoA is methylmalonyl-CoA; S-CoA is succinyl-CoA; G-CoA is glutaryl-CoA. (b) Mass shifts confirming transfer of malonyl (+86 Da), succinyl (+100 Da), and glutaryl (+114 Da) moieties to the cognate ACP from EtnD, retaining the terminal carboxyl group. (c) Docking of S-CoA, Ac-CoA, (not shown: M-CoA and G-CoA) to a Boltz-2–generated EtnB structure, highlighting key interactions: proximity of acyl-CoA to Ser89, salt-bridge formation with Arg114. (d) Predicted interaction profiles (H-bonds, hydrophobic contacts, salt bridges, and π-stacks) of the top three predicted positions for each substrate. Ac-CoA displayed the lowest average number of total interactions vs all other substrates. Binding affinity predictions generated by (e) Boltz-2 and (f) OnionNet-2 aligned with a similar rank order of highest to lowest predicted log­(IC50) and log­(K d ), where Ac-CoA > M-CoA > S-CoA ≈ G-CoA. Attempts to crystallize EtnB yielded needle cluster crystals but no diffraction-quality data, and structural analysis therefore relied on computational predictions. To investigate the molecular basis of the substrate selectivity, we performed docking simulations using a Boltz-2 predicted EtnB structure. The results revealed key molecular determinants of substrate recognition for the four most metabolically relevant acyl-CoAs: acetyl-CoA (Ac-CoA), M-CoA, S-CoA, and G-CoA. All tested CoAs displayed a conserved π-stacking between the adenosine moiety and Phe286, and orientation of the pantothenate arm toward catalytic Ser89. A salt bridge between Arg114 and the terminal carboxyl group was observed for M-CoA, S-CoA, and G-CoA, but absent for Ac-CoA, as it lacks the terminal carboxylic acid group (Figure c). Analysis of the top three predicted poses for each CoA analog highlighted differences in hydrogen bonding, hydrophobic contacts, π-stacking, and salt bridges (Figure d). Unique interaction patterns for each CoA analog are summarized in Supplementary Figure 1. OnionNet-2, a deep-learning–based protein–ligand affinity predictor, was used to estimate dissociation constants (K d ) complementary to Boltz-2 predicted binding affinities. Predicted binding affinities were calculated using both Boltz-2 predicted half maximal inhibitory concentration (IC50) and OnionNet-2-predicted K d values. Consistent with these interaction patterns, both methods ranked Ac-CoA with the highest log­(IC50) (Figure e) and log­(K d ) (Figure f) values, M-CoA with intermediate values, and S-CoA and G-CoA with the lowest, supporting their role as preferred substrates for productive transfer. EtnB substrate engagement was initially evaluated using a CoA-release (hydrolysis) assay. Succinyl-CoA exhibited substantial nonenzymatic hydrolysis under the assay conditions, resulting in high background signal that precluded quantitative assessment of EtnB-catalyzed turnover. Consistent with this, mutation of the catalytic serine (EtnB S89A) (Supplementary Figure 2) did not reduce the apparent hydrolysis rate of succinyl-CoA, indicating that the observed signal is dominated by chemical instability rather than enzyme-mediated acyl transfer. Consequently, hydrolysis assays were not used to infer EtnB specificity toward succinyl-CoA. Productive utilization of succinyl-CoA was instead confirmed by intact-protein mass spectrometry and ACP-loading assays, which demonstrated serine-dependent acylation of EtnB and subsequent transfer to the cognate ACP EtnD (Figure a,b). In contrast, hydrolysis rates for acetyl- and malonyl-CoA reflected enzyme-dependent CoA release, as activity was abolished in the S89A mutant (Supplementary Figure 2). Under these conditions, malonyl-CoA showed higher EtnB-dependent hydrolysis than acetyl-CoA, consistent with preferential engagement of carboxylated substrates during the acyl-loading (ping) step of the reaction. Together, these results support selective recognition of carboxylate-containing acyl-CoAs by EtnB, with succinyl-CoA specificity established by direct acyl-transfer measurements rather than hydrolysis kinetics. Adipic Acid Biosynthesis in Escherichia coli To build a functional PKS system, we coupled the EtnB/EtnD loading module to a hybrid extension module BorM1-BorDH2-SpnB­(KR-ER)-EryTE that had previously been shown to accept succinyl-N-acetylcysteamine and elongate the polyketide backbone by two fully reduced carbons to AA in vitro. We hypothesized that efficient substrate transfer would require a communication element to mediate productive interactions between EtnD-ACP and the downstream ketosynthase (BorM1-KS). Sequence alignment of known linker regions from the EtnD ACP and Borrelidin ACP loading modules (M0) identified conserved motifs and structural features suitable for engineering a fusion construct. In the native Borrelidin PKS, this communication element mediates the correct docking orientation between the upstream ACP and the downstream KS, preserving the geometry required for productive chain transfer. The borL motif has also been implicated in enhancing substrate channeling efficiency in the Borrelidin assembly line, suggesting that its incorporation into EtnD could similarly facilitate intermodule communication. Accordingly, we appended the native C-terminal communication domain (borL) from the BorM0-ACP to the C-terminus of EtnD, generating an EtnD-borL fusion designed to facilitate correct domain interactions and substrate channeling (Figure a). Communication linker design between EtnD and BorM1. (a) Design of the ACP–linker fusion construct, appending the Borrelidin ACP communication domain to EtnD-ACP. (b) Boltz-2 structural modeling shows the correct orientation of the ACP–linker fusion with the phosphopantetheinylated residue oriented toward the downstream KS active site. (c) Comparative interaction analysis demonstrating increased stabilizing contacts in the linker-containing model. (d) Specific polar residue interactions (<3.2 Å cutoff) between BorM1 KS N-terminal docking domain and EtnD-borL (red). (e) AA titers from E. coli K207–3-MatBC expressing different PKS module combinations: EtnLM (pacyc-etnB-etnD) with the nonfully reducing BorM1-EryTE extension module (yellow); EtnLM (pacyc-etnB-etnD) with the fully reducing BorM1-BorDH2-SpnB­(KR-ER)-EryTE extension module (orange); linker-fused EtnLM-borL (pacyc-etnB-etnD-borL) with the nonfully reducing BorM1-EryTE extension module (red); or EtnLM-borL (pacyc-etnB-etnD-borL) with the fully reducing BorM1-BorDH2-SpnB­(KR-ER)-EryTE extension module (green). Fermentation was performed in M9-YE or EZ Rich medium with or without 10 mM malonate (Mal). Linker inclusion significantly increased AA titers under all conditions, with malonate supplementation providing additional enhancement. Boltz-2 homology models supported this design: the EtnD-borL construct formed stable interactions with the downstream KS, correctly orienting the EtnD to the KS active site for translocation (Figure b). For models with and without the linker, the number of interacting residue pairs (Figure c) were extracted, and the sequence location of stabilizing residues between the linker and the KS were analyzed (Figure d). These predicted stabilizing residues form a coiled-coil conformation and coordinate with charged residues on the KS that are well documented in trans-AT docking domain specificity to facilitate intermodule transfer. Further, unconstrained structure prediction of EtnD with and without the BorL linker reliably failed to orient EtnD when the linker was not present, resulting in high model interface uncertainty (Supplementary Figure 3). We next tested the effect of this linker in vivo by expressing plasmids encoding the extension module together with either the native EtnLM (encoded on plasmid: pACYC-EtnB-EtnD) or the linker-fused variant EtnLM-BorL (encoded on plasmid: pACYC-EtnB-EtnD-BorL) in E. coli K207–3-MatBC, which increases intracellular malonyl-CoA (M-CoA) availability through malonate utilization. This generated strains sE.c.AA_03 (native EtnLM) and sE.c.AA_04 (EtnLM–BorL) respectively. Cultures were grown for 6 days in M9-YE or EZ Rich medium with or without 10 mM malonate supplementation. In all conditions, the linker significantly increased AA titers, and malonate supplementation further improved production, consistent with extender-unit demand by the hybrid extension module. Additional validation beyond the established controls was conducted to confirm the production of adipic acid. Samples were spiked with an authentic adipic acid standard, resulting in a proportional increase in the corresponding chromatographic peak. No additional peaks were detected, and the LC–MS extracted ion chromatograms and fragmentation spectra remained unchanged, exhibiting the characteristic fragmentation pattern of adipic acid (Supplementary Figure 4). High-resolution LC–MS (QTOF) analysis of adipic acid produced by the PKS platform revealed an [M–H]− ion consistent with the theoretical exact mass of adipic acid, with a mass error of ≤ 0.001 Da. Furthermore, both the retention time and isotopic distribution were indistinguishable from those of an authentic commercial standard analyzed under identical conditions (Supplementary Figure 5). No peaks corresponding to authentic standards of C5 or C7 diacids were detected, confirming selective loading of S-CoA by EtnB/EtnD in vivo (Figure e). This selectivity may arise from multiple factors, including ketosynthase (KS) gatekeeping and substrate availability; however, the precise basis for this specificity remains to be elucidated through further mechanistic studies. Acyltransferase Domain Exchanges and Bioreactor Scale Fermentation To further probe the versatility of this engineered PKS, we explored exchanging the acyltransferase (AT) domain of the extension module (Figure a). The boundaries of the AT domains were defined based on prior literature precedents and informed by our Boltz-2 structural models (Figure b) to preserve native interdomain interactions and structural integrity. Extension module constructs were expressed in both K207–3 and K207–3-MatBC strains containing ATs from BorM1 (sE.cAA_04 and 05), EpoM4­(sE.cAA_06 and 07), or AnsM8­(sE.cAA_08 and 09), with or without 10 mM malonate supplementation. The AnsM8 AT exchange produced negligible AA, consistent with its known specificity for bulky acyl-CoAs that are not naturally present in E. coli. In contrast, the EpoM4 AT variants produced AA at levels similar to the native AT BorM1, demonstrating that the extension module tolerates AT substitutions without disrupting downstream reduction or termination steps. AT domain exchanges and bioreactor performance in E. coli. (a) Schematic of the hybrid PKS extension module with AT domain exchanges: BorM1 (native), EpoM4, or AnsM8. (b) Overlay of Boltz-2–predicted structures for BorM1, EpoM4, and AnsM8 AT domains. (c) Adipic acid titers from strains sE.c.AA_04–09 grown in M9-YE medium with or without 10 mM malonate supplementation. AT exchanges are denoted as BorM1 (B1, orange), AnsM8 (A8, yellow), and EpoM4 (E4, green) while a control with just the loading module and no extension module is denoted as (-, red). AnsM8 AT strains were inactive, while EpoM4 AT strains produced titers comparable to BorM1 AT strains. In the K207–3-MatBC background, malonate supplementation increased titers ∼ 10-fold across all active constructs. (d) 1-L bioreactor fermentation of strain sE.c.AA_06 in M9-YE medium with 20 g L–1 glucose and 10 mM malonate at 18 or 30 °C. AA titers were consistently higher at 18 °C after 24 h. (e) Proteomic analysis of 120-h samples showing the relative abundance of the hybrid BorM1 extension module (hybrid BorM1), loading module components (EtnB, EtnD–borL), and insoluble protein marker IbpA. Across all AT variants, malonate supplementation increased titers by ∼10-fold, further supporting extender-unit availability as a major bottleneck in this pathway (Figure c). These results suggest that alternative acyl-CoA derivatives, particularly branched or longer-chain substrates, may be explored in future work to access diversified diacids or branched-chain analogs. The EpoM4-containing strain sE.cAA_06 was selected for bioreactor-scale evaluation in M9-YE medium supplemented with 20 g/L glucose and 10 mM malonate. Four 1.5-L vessels were operated in parallel, with two held at 18 °C and two at 30 °C. Over the 5-day fermentation, cultures maintained at 18 °C consistently exhibited higher adipic acid titers beginning at 24 h, compared to 30 °C (Figure d). Proteomic analysis at the final time point (day 5) revealed comparable expression levels of the hybrid BorM1 extension module under both conditions, but lower abundance of the EtnB and EtnD-BorL loading module proteins at 30 °C (Figure e). Although these differences were not statistically significant due to limited sample size, the trend suggests that higher temperatures may compromise loading-module stability or folding, consistent with the more favorable titers observed at 18 °C. Portability and Metabolic Optimization in Pseudomonas putida Having demonstrated that the engineered PKS pathway functioned in E. coli, we next tested whether it could be established in the saprotrophic soil bacterium Pseudomonas putida, a metabolically versatile chassis with high industrial potential. Using the SAGE integration system developed by the Guss lab, we introduced the loading-module genes (etnB and etnD–borL) and subsequently the hybrid extension module (borM1-borDH2-spnB­(KR-ER)-eryTE) into P. putida AG5577-ΔoplBA to generate strain sP.pAA_05. When cultured in glucose–acetate minimal medium, sP.pAA_05 produced adipic acid, confirming portability of the PKS pathway to this alternative host (Figure b). Establishing and optimizing adipic acid production in P. putida. (a) Proposed genetic knockouts to improve metabolic flux toward AA. (b) AA production in strains sP.pAA_05 (WT), sP.pAA_06 (::sfp), sP.pAA_07 (Δgcd), and sP.pAA_08 (Δpha), from three individual colonies grown in M9–MOPS with 20 g/L glucose and 5 g/L acetate for 6 days at 30 °C, 200 rpm. (c) AA titers, cell density, residual glucose, and residual KGA for sP.pAA_05 (WT, extension module at Bxb1) and sP.pAA_09 (+sfp-Δgcd-Δpha, extension module at distal R4) from six individual colonies grown in M9–MOPS with 20 g/L glucose and 5 g/L acetate for 6 days at 30 °C, 200 rpm. (d) WT strains showed two populationslow and normal AA producers. Proteomics indicated that Bxb1-integrated low producers had reduced extension module expression despite consistent EtnB levels, while sP.pAA_09 displayed uniform expression of both PKS modules across all colonies, consistent with WGS evidence of PKS module looping out when integrated at Bxb1. (e) Whole-proteome comparison of sP.pAA_05 and sP.pAA_09 showed significantly changed proteins (p < 0.05; fold change <−1 in red, >1 in blue; nonsignificant in gray). Protein identities are listed in Supporting Information_Proteomics_T-test.xls. (f) In 7-day bioreactor fermentation, sP.pAA_09 showed increasing and stable PKS pathway protein levels, steadily rising AA titers, and OD600 peaking at 24 h. Glucose and acetate were consumed sequentially, with acetate used first; 15 g of acetate was fed on days 2–3. To improve titers, we introduced host modifications known to affect cofactor availability or carbon flux. Addition of the promiscuous phosphopantetheinyl transferase sfp and deletion of either gcd (preventing 2-ketogluconate formation) or pha (eliminating PHA granule biosynthesis) each produced modest improvements in AA titers once the extension module was integrated (Figure b). Interestingly, sP.pAA_05 displayed a bimodal phenotype with both high- and low-producing colonies. Detecting genetic “loop-out” events in P. putida is challenging due to its polyploidy (∼20 genome copies per cell), where partial loss can still yield positive PCR bands. Nonetheless, whole-genome sequencing (WGS) of a low AA-producing culture confirmed loss of the extension module, but retention of the loading module, suggesting genomic instability due to the close proximity and sequence similarity of the integrated loci (Supplementary Figure 6). This instability may have been exacerbated by additional freeze–thaw cycles experienced by this early strain. To overcome this instability, we combined the beneficial knockout mutations with the sfp addition, as this may become more beneficial at higher titers. We then simultaneously repositioned the extension module to a distal genomic SAGE integration site. In strain sP.pAA_09 (+sfp-Δgcd-Δpha), the loading module remained at the RV site, while the extension module was integrated at the R4 site ∼ 1 Mb away, separated by multiple essential genes. Six biological replicates of sP.pAA_05 and sP.pAA_09 were analyzed for AA production, cell density, glucose consumption (20 g L–1 initial), and KGA accumulation. As expected, sP.pAA_09 produced no detectable KGA, consistent with the gcd knockout. This strain also showed lower apparent cell density (OD600), likely due to the pha deletion, as the absence of PHA polymer granules reduces light scattering that would otherwise inflate OD600 measurements. While both strains fully consumed acetate, sP.pAA_09 did not completely utilize the initial 20 g/L of glucose. Importantly, AA titers were higher in sP.pAA_09 relative to sP.pAA_05. Consistent with prior observations, sP.pAA_05 exhibited two subpopulations: low and normal AA producers (Figure c). Proteomic analysis indicated that low producers had reduced expression of the extension module despite stable EtnB (loading module) levels. In contrast, sP.pAA_09 showed uniform expression of both loading and extension modules across all six colonies (Figure d). Comparing the detectable proteomes between normal-producing sP.pAA_05 and sP.pAA_09 further revealed broad protein-level changes, with significantly altered proteins highlighted in Figure e. Full proteomic results are provided in Supplementary Table 4 and Supporting Information_Proteomics.xls. Notably, sP.pAA_09 had several proteins with relatively decreased expression relative to sP.pAA_05 including d-gluconate transporter GntT (log2 Fold change = −6.33, p = 0.00017), glucose dehydrogenase Gcd (log2 Fold change = −5.348, p = 0.0103), poly granule-associated protein Pp_5008 (log2 Fold change = −6.01, p = 0.00037), and poly­(3-hydroxyalkanoate) polymerase 1 PhaA (log2 Fold change = −5.29, p = 0.0075). Additionally, several lipoproteins and lipoprotein-associated proteins demonstrated significantly altered expression levels; for instance, Pp_4032, SlyB, Pp_2121, Pp_5661, Pp_0753, OsmE, and Pp_5319 showed decreased expression, while Pal experienced a slight increase (log2 Fold change = 1.3, p = 0.049). This pattern aligns with recent findings that suggest a connection between PHA granule biosynthesis and the role of lipoproteins in P. putida. Additionally, of the few proteins with increased expression were those involved in the uptake of glucose across the cell membrane, including the mannose/glucose ABC transporter GtsD (log2 Fold change = 1.97, p = 0.0039) and the carbohydrate-selective porin Oprb-I, (log2 Fold change = 1.81, p = 0.027). Altogether, these proteomic changes confirm the intended redirection of flux away from PHA and gluconate biosynthesis (Figure a). Because of its improved genetic stability and higher AA titers, sP.pAA_09 was selected for bioreactor evaluation. In a 7-day fermentation run, integrated PKS pathway protein levels steadily accumulated and plateaued, AA titers increased continuously, and glucose and acetate were consumed sequentially, with acetate used first (Figure f). Cell density peaked at 24 h and subsequently declined, likely due to lysis, while a supplemental 15 g acetate feed between days 2–3 further supported sustained production. Together, these results demonstrate that transfer of the PKS pathway into P. putida established a productive new chassis for AA biosynthesis. Moreover, combining targeted knockouts with distal genomic integration produced a genetically stable strain capable of robust performance in bioreactor-scale fermentation. Discussion This work establishes the first PKS-based route to adipic acid (AA) by engineering one of the most hybridized PKS systems reported to date. − Central to this success was the identification of EtnB, a novel succinyl-CoA loading module that retains the terminal carboxyl group, enabling access to dicarboxylic acids, a chemical functionality largely absent from most PKS systems. Coupling this novel loading module to a fully reducing extension module via an engineered communication domain, validated both in silico and in vivo, resulted in productive substrate transfer and enhanced AA titers. The system also tolerated acyltransferase (AT) domain exchanges, suggesting the potential to access structurally diverse branched-chain or functionalized diacids in future designs. Beyond pathway construction, we demonstrated its portability across E. coli and P. putida. In P. putida, we optimized the strain by redirecting metabolic flux toward PKS pathways and improving genomic stability through the careful selection of integration sites. The resulting strain, sP.pAA_09, supported adipic acid production at bioreactor scale, where pathway proteins accumulated over time and titers steadily increased across a 7-day fermentation. This provides a generalizable strategy for multimodule PKS pathway assembly in polyploid hosts. Although malonate supplementation led to a ∼10-fold increase in adipic acid titers, indicating that extender-unit supply is a major bottleneck under baseline conditions, additional constraints limit overall pathway flux. Succinyl-CoA, as a central TCA-cycle intermediate, is rapidly regenerated in vivo, and is therefore unlikely to be limiting under the conditions tested, even under elevated malonyl-CoA flux. , Instead, the dominant limitations appear to arise from the expression level and stability of the large hybrid PKS and from imperfect intermodule translocation across multiple non-native junctions. These cumulative kinetic and structural constraints are characteristic of highly engineered PKS systems and suggest that future improvements will require coordinated optimization of precursor supply, protein folding and stability, and intermodule communication rather than single-enzyme interventions. Together, these findings highlight three advances: (i) expanding the PKS toolkit with a carboxyl-retaining loading module, (ii) demonstrating the function of a hybrid-PKS with genes derived from five organisms, seven PKS modules with six unnatural junctions in vivo, and (iii) establishing and optimizing this pathway in both E. coli and P. putida. This platform opens new opportunities to engineer PKSs toward branched or functionalized diacids and to optimize microbial hosts for sustainable, industrial-scale production. Materials and Methods Chemical Compounds and Standards All chemical standards, including adipic acid and 2-ketogluconate, were obtained from Sigma-Aldrich unless otherwise noted. A 2.5 M stock solution of disodium malonate (Spectrum Chemicals) was prepared in Milli-Q water and adjusted to neutral pH. M9 minimal medium with yeast extract (M9-YE) was prepared as follows: 1× M9 salts, 5 g L–1 yeast extract, 2% glucose, 2 mM MgSO4, thiamine (1 μg mL–1), 0.1 mM CaCl2, and 1× trace elements (Teknova). All components were dissolved in Milli-Q water and filter-sterilized (0.22 μm) before use. MOPS minimal medium was prepared as follows: 32.5 μM CaCl2, 0.29 mM K2SO4, 1.32 mM K2HPO4, 8 μM FeCl2, 40 mM MOPS, 4 mM tricine, 0.01 mM FeSO4, 9.52 mM NH4Cl, 0.52 mM MgCl2, 50 mM NaCl, 0.03 μM (NH4)6Mo7O24, 4 μM H3BO3, 0.3 μM CoCl2, 0.1 μM CuSO4, 0.8 μM MnCl2, and 0.1 μM ZnSO4. Unless otherwise stated, glucose and acetate were added from sterile-filtered stocks immediately before use. Phylogenetic Tree Construction Candidate succinyl-CoA–loading trans-acyltransferases were identified by BLASTp searches of the NCBI nonredundant (nr) protein database using the EtnB sequence as a query. Sequences sharing >30% identity with EtnB were retrieved and aligned using ClustalW2 v2.1 on the EMBL-EBI Web server (www.ebi.ac.uk/Tools/phylogeny/). A neighbor-joining phylogenetic tree was constructed from the multiple-sequence alignment using default parameters, with the maximum sequence difference set to 0.85. The resulting tree was visualized in the ClustalW2 Phylogeny interface. Simplified phylogenetic tree of ATs with carboxylic acid loading potential was derived from sequences reported on MIBiG (https://mibig.secondarymetabolites.org/) with the following accession numbers: BGC0002873 - Gladiolin BSGC from Burkholderia gladioli (GbnB, GbnN), BGC0001520 - Aurantinin B from Bacillus subtilis (Art2, Art9), BGC0000179 - Etnangien from Sorangium cellulosum (EtnB, EtnK), and the known M-CoA AT (FabD) from E. coli. A neighbor-joining phylogenetic tree was constructed from the multiple-sequence alignment using default parameters. The resulting tree was visualized in the ClustalW2 Phylogeny interface. Protein Structure Determination and Small Molecule Docking EtnB protein structure and AT domains were predicted using Boltz-2. Docking simulations were performed using Boltz-2 with default settings to copredict protein and acyl-CoA ligand conformations. Predicted dissociation constant values were calculated by running OnionNet-2 using energy-minimized protein–ligand conformations from Boltz-2. Ligand binding and interaction data were extracted using PLIP. BorL-EtnD structures were copredicted with Boltz-2 by CCD code modification of the active site serine (4HH) to model the phosphopantetheine arm in complex with the KS following separate swissprot homology modeling of EtnD (PDB: 8TKO ) and the docking domain interface (PDB: 1PZR). Protein–protein interactions for these complexes were extracted using scripts where interactions were detected between heavy atoms using standard distance criteria (vdW ≤ 4.0 Å; H-bond ≤ 3.5 Å with donor/acceptor rules; salt bridge ≤4.0 Å between charged atoms), and each unique residue pair was counted once per interaction type to avoid atom-level overcounting. Plasmid Construction Plasmids were assembled using standard molecular biology protocols. All plasmids and strains in this study are described in Supplementary Table 1. Gene-specific primers and amplified fragments are listed in Supplementary Table 2, with corresponding sequences in Supplementary Table 3. When necessary, PKS DNA fragments were codon-optimized using BaseBuddy and synthesized by Genscript. DNA fragments were amplified using either Q5 High-Fidelity DNA Polymerase (NEB) or Phusion High-Fidelity DNA Polymerase (Thermo Fisher Scientific). Assemblies were performed using NEBuilder HiFi DNA Assembly Master Mix (NEB) and transformed into NEB 5-alpha F’Iq or XL1-Blue chemically competent E. coli. Plasmids were purified using the QIAprep Spin Miniprep Kit (Qiagen) according to the manufacturer’s instructions. All constructs were verified by whole-plasmid sequencing (Primordium Laboratories). Dual-Plasmid Transformation in E. coli For dual-plasmid expression in E. coli, the pACYCDuet-based plasmids carrying the PKS loading module constructs were transformed first. Chemically competent E. coli strains of K207–3 or K207–3-ompW-MatBC were prepared according to standard protocols and transformed with pACYCDuet-derived plasmids using heat shock. Cells were recovered in SOC medium at 37 °C for 1 h and plated on LB agar containing chloramphenicol (CMR, 30 μg/mL) for selection. Single colonies were picked and cultured overnight in LB + CMR to ensure stable plasmid maintenance. Subsequently, the pARH-based extension module plasmids were transformed into the loading module-containing E. coli host strains using the same transformation protocol. Following recovery, cells were plated on LB agar supplemented with chloramphenicol (30 μg/mL) and kanamycin (50 μg/mL) to select for comaintenance of both plasmids. Colonies were screened by colony PCR or restriction digest to confirm the presence of both plasmids. For all strains used in expression experiments, dual antibiotic selection (CMR + Kan) was maintained during preculture and induction to prevent plasmid loss. Gene Deletions in P. putida The parent strain used in this study is P. putida AG5577, which is derived from strain KT2440 and was provided by A. Guss at Oak Ridge National Laboratory and further modified to delete oplBA by Lee et al. Further gene deletions were generated by homologous recombination as described previously. Briefly, pK18_sacB deletion plasmids were electroporated into P. putida. Cells were recovered in LB at 30 °C for 2 h and plated on antibiotic-selective agar, followed by counterselection on tryptone-yeast agar with 10% sucrose to select for second-cross events resulting in clean deletions or wild-type revertants. Colonies with successful gene deletions were screened for via PCR amplification using primers listed in Supplementary Table 2, and final strains were verified by whole-genome sequencing (Primordium Laboratories). Genomic Integration in P. putidavia SAGE The loading-module genes (etnB and etnD–borL) were integrated into the RV landing pad to generate strain sP.pAA_000. The hybrid extension module (BorM1–BorDH2–SpnB­(KR–ER)–EryTE) was subsequently integrated at the Bxb1 site, yielding baseline AA-producing strain sP.pAA_005. For host engineering, sfp from Bacillus subtilis (encoding a broad-specificity PPTase) was integrated at the BL3 site to generate sP.pAA_01. Deletion of gcd was performed using a two-step allelic exchange method to generate sP.pAA_02, and deletion of pha (polyhydroxyalkanoate synthase) was similarly performed to generate sP.pAA_03. For each edited background, the extension module was integrated at the Bxb1 site to generate strains sP.pAA_06 (sfp), sP.pAA_07 (Δgcd), and sP.pAA_08 (Δpha). For genomic integration of the PKS loading and extension modules, overnight cultures (1 mL) were washed three times with 300 mM sucrose and resuspended in 0.1 mL of 300 mM sucrose. Cells were coelectroporated in a 0.2 cm cuvette at 1.8 kV with 100–500 ng of an attP-containing expression plasmid and 100–500 ng of a nonreplicating helper integrase plasmid. Following electroporation, cells were recovered in LB at 30 °C for 2 h and plated on LB–Kanamycin (50 μg mL–1). Colonies were screened for integration by PCR using primers in Supplementary Table 2. Positive colonies were cultured overnight, and the Kanamycin resistance and LacI marker genes were excised by electroporation with phiC31 integrase (100–500 ng DNA, 1.8 kV). After recovery, cells were plated on LB–Chloramphenicol (30 μg mL–1). Correct integration was confirmed by PCR of genomic target regions and verified by whole-genome sequencing (Primordium Laboratories). Protein Expression Plasmids encoding pET21a-EtnB or its catalytically inactive mutant EtnB­(S89A) were transformed into E. coli BL21­(DE3) and selected on LB agar containing 50 μg/mL kanamycin. Overnight cultures (20 mL LB + kanamycin, 37 °C, 200 rpm) were used to inoculate 500 mL Terrific Broth (TB) with kanamycin (50 μg/mL) to an initial OD600 of 0.01. Cultures were grown at 37 °C until OD600 ≈ 0.6, then cooled to 18 °C over 30 min. Expression was induced with 0.2 mM IPTG, and cultures were incubated at 18 °C, 200 rpm for 16–18 h. Cells were harvested by centrifugation (8,000 rpm, 15 min, 4 °C) and pellets were stored at – 20 °C. Protein Purification Frozen pellets containing the EtnB proteins were thawed on ice and resuspended in lysis buffer (50 mM sodium phosphate, pH 7.5, 300 mM NaCl; 5 mL g–1 wet cell weight). Cells were lysed by sonication (2 s on/4 s off, total 10 min, 40% amplitude) on ice. Lysates were clarified by centrifugation (15,000 rpm, 15 min, 4 °C), filtered (0.45 μm), and loaded onto cobalt affinity resin pre-equilibrated with lysis buffer. Columns were washed with 10 column volumes of wash buffer (50 mM sodium phosphate, pH 7.5, 300 mM NaCl) and eluted with elution buffer containing 300 mM imidazole. Fractions were analyzed by SDS–PAGE, pooled, concentrated (Amicon Ultra-15, 10 kDa cutoff), and buffer-exchanged into storage buffer (50 mM HEPES, pH 7.0, 5% glycerol). Purified proteins were flash-frozen and stored at – 80 °C. In Vitro Protein Assays Apo-EtnD-borL (25 μM) was converted to holo-EtnD-borL by incubation with Sfp (2.5 μM) in 50 mM phosphate buffer (pH 7.0) containing 10 mM MgCl2, 5 mM DTT, and 0.5 mM CoA at 30 °C for 30 min. Excess reagents were removed by three washes with phosphate buffer using a 3 kDa spin column. EtnB (20 μM) was incubated with or without holo-EtnD-borL (50 μM) with 200 μM acyl-CoA substrates for 3–6 h. Reactions were quenched 1:1 with acetonitrile, clarified by 0.2 μm spin filtration, and prepared for LC–MS. Intact proteins were analyzed by ESI–TOF–MS using an Agilent 1260 LC system coupled to an Agilent 6530 Q-TOF mass spectrometer with a ProSwift RP-4H column. Separation was performed with a gradient from 5% Milli-Q water +0.1% (v v–1) formic acid to 95% MS-grade acetonitrile +0.1% (v v–1) formic acid. Data were processed in MassHunter software (Agilent) and visualized using Chartograph (www.chartograph.com). HPLC CoA Assay Reactions contained 50 mM phosphate buffer (pH 7.2), 100 μM CoA, and 5 μM EtnB or EtnB­(S89A). No-enzyme controls were included. At designated time points, 20 μL aliquots were quenched with 20 μL methanol containing 0.1% formic acid, frozen at −20 °C, and stored until analysis. Samples were thawed, centrifuged (15,000 × g, 10 min), and supernatants analyzed by HPLC (Agilent 1200 Series) on a C18 column under isocratic elution (33 mM sodium phosphate, 20 mM sodium acetate, 96% water, 4% acetonitrile). Coenzyme A Quantification (Enzychrome Assay) CoA concentrations were determined using the Enzychrome CoA Assay Kit (BioAssay Systems) with modifications to the standard procedure. A four-point standard curve (0–1,000 μM) was prepared. Samples and standards (10 μL) were dispensed into 96-well plates; ACS reagent (40 μL assay buffer, 1 μL Enzyme A, 5 μL substrate, 50 μM EtnD, 1 μL ATP) was added and incubated 10 min at room temperature. ACOD reagent (55 μL assay buffer, 1 μL Enzyme B, 1 μL dye reagent, 5 μM EtnB) was then added, and absorbance at 570 nm was recorded every 30 s for 60 min. Concentrations were calculated relative to the standard curve. Production Experiments E. coli (24-Well Plates) sE.c.AA_00 to sE.c.AA_09. Dual plasmid strains were selected on LB agar (30 μg mL–1 chloramphenicol, 50 μg mL–1 kanamycin). Overnight LB cultures were diluted 1:100 into 2 mL M9-YE medium with antibiotics ± 10 mM malonate. Triplicate cultures were grown in 24-well plates (37 °C, 200 rpm). At OD600 ≈ 0.6, cultures were cooled to 18 °C and induced with 0.2 mM IPTG + 60 ng μL–1 anhydrotetracycline. Plates were incubated for 6 days at 18 °C. Aliquots (200 μL) were collected at multiple time points and stored at −20 °C for LC-MS adipic acid quantification. P. putida (24-Well Plates) sP.p.AA_00, sP.p.AA_05, sP.p.AA_06, sP.p.AA_07, sP.p.AA_08, sP.p.AA_09. Strains were streaked from glycerol stocks (−80 °C) onto LB agar with 30 μg/mL chloramphenicol. Single colonies were inoculated into MOPS minimal medium containing 30 μg/mL chloramphenicol, 1× trace elements (Teknova), 2% glucose, and 5 g L–1 acetate. Cultures (n = 3–6) were grown in 24-well plates at 30 °C, 200 rpm. Aliquots (200 μL) were collected at multiple time points and stored at −20 °C for LC-MS analysis. Bioreactor E. coli Four 1.5 L Eppendorf DASGIP bioreactors were inoculated at 37 °C with the sE.c.AA_06 strain at an initial OD600 of 0.1 in M9 medium containing 20 g L–1 glucose, 5 g L–1 yeast extract, 1× trace elements, and 30 μg mL–1 chloramphenicol and 50 μg mL–1 kanamycin (initial working volume 1,000 mL). Dissolved oxygen was maintained at 30% via an agitation–airflow cascade using an L-sparger, and pH was controlled at 7.0 ± 0.2 with 15% (w v–1) H2SO4 and 5 M NaOH. Upon reaching OD600 ≈ 0.6, two bioreactors were cooled to 18 °C and two cooled to 30 °C, after which 10 mM malonate, 0.2 mM IPTG, and 60 ng μL–1 anhydrotetracycline were added. Antifoam 204 (0.1 mL) diluted in 9 mL water was injected to a final concentration of 0.01% (v v–1). At 36 h postinduction, cultures were fed with 20 g/L glucose, 10 mM malonic acid, 0.2 mM IPTG, and 60 ng/μL anhydrotetracycline. Bioreactors were maintained for a total cultivation time of 5 days. Aliquots were collected at defined time points for adipic acid quantification by LC–MS (200 μL) and proteomics (1 mL). Bioreactor P. putida Four 1.5 L Eppendorf DASGIP bioreactors were inoculated at 30 °C with the sP.pAA_09 strain at an initial OD600 of 0.1 in MOPS minimal medium containing 20 g L–1 glucose, 5 g L–1 sodium acetate, 1× trace elements, and 30 μg mL–1 chloramphenicol (initial volume 600 mL). Dissolved oxygen was maintained at 30% by an agitation–airflow cascade using an L-sparger, and pH was controlled at 7.0 ± 0.2 with 15% (w v–1) H2SO4 and 5 M NaOH. At 29 h postinoculation, 5 g acetate was added, followed by continuous feeding of 100 g/L acetate with 2× trace metals at 2 mL/h for 31 h. Cultures were then grown for an additional 104 h (164 h total). Aliquots were collected at defined time points for adipic acid quantification by LC–MS (200 μL), proteomics (1 mL), and glucose/2-ketogluconate analysis (400 μL). Sample Preparation for LC-MS Fermentation samples (200 μL) were mixed with 200 μL methanol containing 0.1% formic acid and 10 μM p-coumaric acid (internal standard). Samples were incubated at 60 °C for 60 min with shaking (800 rpm), centrifuged (14,000 × g, 10 min), and supernatants transferred to LC-MS vials. Analysis was performed on an Agilent 1260 Infinity HPLC coupled to an Agilent 6520 QTOF mass spectrometer. Ten microliters of the sample were injected onto a Phenomenex Kinetex XB-C18 column (2.6 μm, 100 × 3 mm, 100 Å). Mobile phases: (A) 0.1% formic acid in water, (B) 0.1% formic acid in methanol. QTOF settings: drying gas 11 L/min, nebulizer 30 psi, gas temperature 330 °C, ESI negative mode, capillary voltage 3500 V, scan 100–250 m/z with SIM at 145 m/z. Data were acquired using MassHunter (Agilent) software. Sample Preparation for Proteomics Fermentation samples (1 mL) were centrifuged at 13,000 × g for 5 min to pellet cells. Supernatants were discarded, and pellets were washed once with 1× PBS, centrifuged again, and stored at – 80 °C until analysis. Before analysis, the protein was extracted from cell pellets, and tryptic peptides were prepared by following the established proteomic sample preparation protocol. Briefly, cell pellets were resuspended in Qiagen P2 Lysis Buffer (Qiagen) to promote cell lysis. Proteins were precipitated with the addition of 1 mM NaCl and 4× vol acetone, followed by two additional washes with 80% acetone in water. The recovered protein pellet was homogenized by pipetting and mixing with 100 mM ammonium bicarbonate in 20% methanol. Protein concentration was determined by the detergent compatible (DC) protein assay (Bio-Rad). Protein reduction was accomplished using 5 mM tris 2-(carboxyethyl)­phosphine for 30 min at room temperature, and alkylation was performed with 10 mM iodoacetamide (final concentration) for 30 min at room temperature in the dark. Overnight digestion with trypsin was performed at a 1:50 trypsin-to-total protein ratio. The resulting peptide samples were analyzed on an Agilent 1290 Ultra High Performance Liquid Chromatography (UHPLC) system coupled to a Thermo Scientific Orbitrap Exploris 480 mass spectrometer for discovery proteomics. Briefly, peptide samples were loaded onto an Ascentis ES-C18 Column (Sigma-Aldrich) and were eluted from the column by using a 10 min gradient from 98% solvent A (0.1% FA in H2O) and 2% solvent B (0.1% FA in ACN) to 65% solvent A and 35% solvent B. Eluting peptides were introduced to the mass spectrometer operating in positive-ion mode and were measured in data-independent acquisition (DIA) mode with a duty cycle of three survey scans from m/z 380 to m/z 985 and 45 MS2 scans with precursor isolation width of 13.5 m/z to cover the mass range. DIA raw data files were analyzed by an integrated software suite, DIA-NN. The database used in the DIA-NN search (library-free mode) is E. coli’s latest UniProt proteome FASTA sequences plus the protein sequences of the heterologous proteins and common proteomic contaminants. DIA-NN determines mass tolerances automatically based on first-pass analysis of the samples with automated determination of optimal mass accuracies. The retention time extraction window was determined individually for all MS runs analyzed via the automated optimization procedure implemented in DIA-NN. Protein inference was enabled, and the quantification strategy was set to robust LC = high accuracy. The output main DIA-NN reports were filtered with a global false discovery rate (FDR) = 0.01 on both the precursor level and protein group level. The Top3 method, which is the average MS signal response of the three most intense tryptic peptides of each identified protein, was used to plot the quantity of the targeted proteins in the samples. , The mass spectrometry proteomics data have been deposited in the ProteomeXchange Consortium via the PRIDE partner repository with the data set identifier PXD069956. DIA-NN is freely available for download from https://github.com/vdemichev/DiaNN. Significant differences between the proteomic profiles of the strains were determined by using a Welch’s t test and a cutoff of > ± 1-fold and a P value of <0.05. Significantly up and down proteins were then annotated when function could be identified using their COG name with the eggNOG-Mapper V2 tool (http://eggnog-mapper.embl.de/). Excel files containing all the significantly up and down proteins for each comparison are provided in Supplementary Data_Proteomics_T-test.xls. Sample Preparation for HPLC Fermentation samples (400 μL) were centrifuged at 13,000 × g for 5 min to remove cell debris. The supernatants were filtered through 0.2 μm spin filters and transferred to HPLC vials. Residual sugar concentrations were analyzed using an Agilent 1200 Series HPLC system equipped with a refractive index detector. Separation was performed on a sugar analysis column under isocratic conditions with a mobile phase of 4% H2SO4 in water at 0.5 mL/min for 26 min using an Aminex HPC-87H Organic Acid column 300 × 7.8 mm heated to 65 °C Supplementary Material