A survey of Burkholderiales secondary metabolites using library-scale stable isotope labeling reveals selective production of methyl-esterified 3-hydroxybutyrate oligomers

Abstract

Large-scale untargeted bacterial metabolomics studies are often challenging to interpret, particularly for natural product discovery. Complex raw mass spectrometry data contains media components, biotransformation products, workup contaminants, and in-source fragments, making it difficult to resolve true biosynthesized bacterial products from background features. Incorporating stable isotope labels into untargeted workflows enables the selective detection of actively biosynthesized compounds, drastically reducing candidate mass spectrometry features to highlight novel or robustly expressed metabolites. Recently, our lab introduced IsoAnalyst, a parallel stable isotope labeling platform that links secondary metabolites to biosynthetic gene clusters by determining the incorporation rates of a panel of isotopically labeled building blocks. In this study we applied the IsoAnalyst protocol to a library of 114 Burkholderiales strains with the goals of profiling the specialized metabolite chemistry of this bacterial order and connecting detected natural products to their cognate BGCs based on their labeling profiles. While our strain-matching protocol, which evaluates correlations between detected metabolites and BiG-SCAPE-defined gene cluster families, did not yield confident BGC assignments, it successfully highlighted widely distributed metabolites with distinct labeling profiles. We targeted a set of these molecules for isolation, revealing two classes of compounds derived from primary biomolecule metabolism that are implicated in cellular stress responses. The first class of molecules were identified as lysophosphatidyl ethanolamine analogues while the second was characterized via NMR spectroscopy and mass spectrometry as methyl-esterified 3-hydroxybutyrate oligomers (HB6-OMe, HB7-OMe, and HB8-OMe), representing higher-degree polymer units than previously reported.

One-sentence summary Large-scale isotope labelling of a panel of 114 Burkholderiales strains reveals the secondary metabolite diversity of this microbial order, leading to the discovery of novel butyric acid polymers.

Graphical Abstract

Graphical Abstract.

Introduction

Natural product (NP) diversity has an important role in drug discovery as it provides ‘templates’ for the design of new drugs and leads to the discovery of new chemical genetic probes for studying molecular mechanisms of disease (Walsh and Tang, 2017, Hong, 2011). NP discovery is one of the most important strategies for discovering bioactive compounds. However, extensive investigation of commonly studied orders such as Kitasatosporales (which includes the genus Streptomyces) and Pseudomonadales (which includes the genus Pseudomonas) has led to the frequent rediscovery of known scaffolds, limiting the value of NP projects for finding novel bioactive molecules.

Burkholderiales are a diverse order of β-Proteobacteria with high metabolic versatility which produce a diverse array of NPs with antibiotic, anticancer and pesticidal activities (Kenshole et al., 2021, Wang et al., 2018). In terms of genetically-encoded bacterial natural product diversity identified by genome mining, Burkholderiales rank third behind two Actinomycetota orders (Kitasatosporales and Mycobacteriales), highlighting the biosynthetic potential of this bacterial family (Gavriilidou et al., 2022). Additionally, many of the reported Burkholderiaceae gene cluster families (GCFs) and/or biosynthetic gene clusters (BGCs) remain orphan or cryptic, demonstrating that this bacterial family represents an untapped source of NPs (Kunakom and Eustáquio, 2019, Paulo et al., 2024).

Recent work from our labs and others has highlighted the potential of the order Burkholderiales for producing complex and unusual carbon skeletons (Fergusson et al., 2024, Paulo et al., 2024, Romanowski et al., 2023). Examples include the antifungal polyketide lagriamide B which contains a complex spiroketal motif and 11 stereocenters, megapolipeptins A and B which are hybrid polyketide, nonribosomal peptide and polyunsaturated fatty acid products containing an unusual 4-hydroxypimelinic acid moiety, and selethramide, an 18-amino acid surfactant comprised almost entirely of serine and leucine residues (Fergusson et al., 2024, Paulo et al., 2024, Romanowski et al., 2023). Together, these discoveries illustrate both the unusual biosynthetic capacity from this order, and the excellent opportunities for new compound discovery offered by exploring taxa beyond traditional sources of NPs.

Several of the discovery projects described above employed stable isotope labeling to detect target compounds from fermentation broths. Previously, our laboratory developed the IsoAnalyst platform, which is designed to connect natural products to their cognate BGCs using a combination of parallel stable isotope labeling and isotope-driven metabolomics. IsoAnalyst works by performing separate micro-scale cultures with different isotopically labeled biosynthetic precursors and then detecting labeled mass spectrometry (MS) features from extracts of these culture broths using ultra-performance liquid-chromatography high-resolution mass spectrometry (UPLC-HRMS). IsoAnalyst uses four ‘generalist’ stable isotope labeled (SIL) precursors ([1–¹³C]-acetate (ACE), [1–¹³C]-propionate (PROP), [methyl-¹³C]-methionine (MET) and [1–¹⁵ N]-glutamate (GLU)) for the detection of a wide range of NP classes. [1–¹³C]-Acetate can be a direct precursor for terpene biosynthesis through the mevalonate pathway, or it can be converted to malonyl-CoA for polyketide biosynthesis. This SIL feedstock can also be modified through the citric acid cycle (TCA) into amino acids which can allow the detection of non-ribosomal peptides (NRPs) or other amino acid-containing NPs. [1–¹³C]-Propionate can be converted to methyl-malonyl-CoA, another common precursor for polyketide biosynthesis. [Methyl-¹³C]-methionine was chosen to detect NPs that undergo methylation by S-adenosyl methionine (SAM)-dependent enzymes. Finally, [1–¹⁵ N]-glutamate is used to detect nitrogen-containing NPs.

In the IsoAnalyst workflow four replicates of each precursor are grown in 24-well plates. Labeled precursors are grown in parallel with their equivalent unlabeled precursors. IsoAnalyst profiles are then built based on the detected SIL precursors by comparing the isotopic envelopes of each MS feature between labeled and unlabeled LCMS analyses (Fig. 1).

Figure 1.

Project Workflow. IsoAnalyst experiments were performed on our environmental Burkholderiales (BKD) library of 114 strains. SIL experiments are performed using [1–¹³C] acetate (30 mM), [1–¹³C] propionate (30 mM), [methyl-¹³C] methionine (5 mM), and [1–¹⁵ N] glutamate (10 mM) and their corresponding unlabeled precursors in 24-well microtiter plates. UPLC-HRMS is performed on each culture condition for later comparisons between labeled and unlabeled MS datasets. Detected labeled MS feature represent microbially biosynthesized or modified metabolites, significantly reducing the number of molecular candidates to investigate. and constituting the SIL Burkholderiales chemistry. From detected compound labeling profiles, in conjunction with genome BGC predictive tools (antiSMASH, BiG-SCAPE, and MIBiG), we aim to connect NPs to their cognate BGCs.

In conjunction with bioinformatics tools such as antiSMASH and MIBiG, it is possible to connect detected NPs to their cognate BGCs based on label uptake (Blin et al., 2021, Medema et al., 2011, Terlouw et al., 2023). For example, in a recent study from our groups, we were able to link the labeling profile of the newly discovered natural product haereoacidicolin to its non-ribosomal peptide synthetase (NRPS) BGC using IsoAnalyst and antiSMASH (Krull et al., 2026). Investigation of the NRPS using antiSMASH revealed 5 adenylation domains predicted to encode respectively for L-Thr, D-Ser, D-Thr, L-Gln and a fifth residue of unknown identity, which was subsequently elucidated as para-aminobenzoate (PABA). The latter amino acids translate to a predicted labeling profile consisting of ACE and GLU. This signature enabled the discovery of haereoacidicolin as the metabolite was successfully prioritized from the extract after being identified by this labeling pattern (Krull et al., 2026, Zheng et al., 2020).

A major advantage of IsoAnalyst over other metabolomics strategies is that SIL methods can differentiate between metabolites that derive from de novo biosynthesis and signals that derive from media components, biotransformation, or exogenous contaminants from the workflow. This is because only products from de novo biosynthesis contain significant isotopic labeling from precursor incorporation. Using this approach, untargeted metabolomics datasets can be reduced from thousands of MS features to a condensed list of compounds with evidence of isotopic labeling that are predicted to derive from microbial biosynthesis or manipulation.

In this study, we applied the IsoAnalyst workflow to a library of 114 environmental Burkholderiales strains. Application of large-scale SIL fermentation was followed by comparative analysis between labeled analytes and gene cluster family (GCF) distributions to examine the NP repertoire in minimal media. Investigation of analytes with strong SIL profiles led to the identification of two classes of compounds derived from biomolecule metabolism that are important for cell survival and host-microbe interactions.

Results and discussion

Development of minimal media conditions for IsoAnalyst labeling in Burkholderiales

McCaughey et al. originally developed the IsoAnalyst protocol for Actinobacteria strains in minimal media to determine the ideal timing for SIL feedstock addition and the optimal concentration to maximize the degree of label incorporation (McCaughey et al., 2022). Following the same logic, we optimized the minimal media composition and incubation time for Burkholderiales strains, prioritizing conditions that yielded a higher number of detected labeled metabolites with high label counts (sum of all incorporated labels). Higher isotopic label counts facilitate more robust correlations between NPs and their cognate BGC(s) because the resulting labeling profile provides a more diagnostic biosynthetic signature for each detected metabolite.

For media optimization, we selected 17 representatives from our environmental Burkholderiales (BKD) strain library including 1–2 strains from each subclade (Table S1). Strains were grown in five minimal media: Difco M9 minimal salts (M9), M9 supplemented with 20% glucose (M9G2), M9 supplemented with 0.4% glucose (M9G); M9G supplemented with 0.5% Bacto Tryptone (M9TG) (Burtnick and Brett, 2013), a Pseudomonas minimal media (PMM) and a 10-fold dilution of nutrient broth media (10xNB); in 24-well microtiter plates as per McCaughey et al. (McCaughey et al., 2022) Bacterial growth was only observed in M9G, M9TG, and PMM, which were subsequently used as minimal media to assess incubation time for optimal stable isotope label incorporation. Paraburkholderia megapolitana RL18-039-BIC-B was used as a representative strain for this step and demonstrated that culturing in M9TG medium for 6 days returned a higher number of detected labeled MS features with robust labeling profiles (12 labeled features with label count higher than 5 in M9TG compared to 1 and 3, respectively, in M9 and PMM, Fig. 2B).

Figure 2.

IsoAnalyst system optimization. (A) Effect of media selection on the detected IsoAnalyst output of P. megapolitana RL18-039-BIC-B. Venn diagram is drawn to scale: 70 labeled MS features detected for media M9G, 58 labeled MS features detected for media M9TG, 54 labeled MS features detected for media PMM. 21 common labeled MS features between M9G and PMM, 11 between M9G and M9TG, and 8 between M9TG and PMM. 5 common labeled MS features across the 3 assessed media. Feature list data is available in Supplementary data file ‘SI_BKD_IsoAnalyst_Media_Incubation_Time_Results.xlsx’. (B) Effect of incubation time on the detected IsoAnalyst output of P. megapolitana RL18-039-BIC-B. Venn diagram is drawn to scale: 48 labeled MS features detected for media 3-day, 58 labeled MS features detected for media 6-day, 53 labeled MS features detected for 8-day. 21 common labeled MS features between 6- and 8-day, 15 between 3-day and 6-day, and 9 between 3-day and 8-day. 8 common labeled MS features across the 3 assessed incubation times. Feature list data is available in Supplementary data file ‘SI_BKD_IsoAnalyst_Media_Incubation_Time_Results.xlsx’. (C) Label incorporation results based on minimal media (M9, M9TG, and PMM) and incubation time (3-day, 6-day, and 8-day). Blue bars denote counts of MS features that were detected with at least one SIL incorporation. Pink bars denote counts of features with label counts higher than 5, indicating high SIL incorporation.

Based on this media optimization, the large-scale IsoAnalyst experiment with our 114 environmental BKD strains was conducted in M9TG minimal medium, extracting after a 6-day incubation period. As previously discussed, secondary metabolites are predominantly produced in response to stressors which are often related to the environment (Pan et al., 2019). Minimal media can itself act as a stressor to trigger secondary metabolite biosynthesis. The result of this is that metabolite profiles can differ significantly under minimal and ‘rich’ media conditions (e.g. SYP) (Pan et al., 2019). Incubation time also matters, as bacteria produce different metabolites depending on their growth phase. Thus, extraction of NPs during the exponential phase will provide a different chemical profile compared to extraction in the stationary phase. IsoAnalyst uses SIL precursors that will label many NP biosynthetic classes but not all. It is therefore important to note that this study provides a select picture of the chemical potential of the 114 environmental BKD strains used in this study (Figure S34). For example, when we cultured P. megapolitana RL18-039-BIC-B in three different media conditions (M9G, M9TG, and PMM), we only observed 5 common labeled MS features across all three conditions despite detecting 70, 58, and 54 labeled MS features in M9G, M9TG, and PMM respectively (Fig. 2A). Likewise, when we altered the incubation time from 3-days to 6-days, and 8-days, we also only observed 8 labeled MS features across all three conditions despite detecting 48, 58, and 53 labeled MS features respectively (Fig. 2B). Thus, choice of media condition and incubation time strongly influence the observed chemistry from BKD strains.

Large-scale IsoAnalyst profiling of 114 Burkholderiales strains

In line with the IsoAnalyst workflow, we used M9TG as the minimal media in 24-well microtiter plates. Each SIL condition was analyzed in four replicates along with four replicates of the corresponding unlabeled control, resulting in two microtiter plates per assayed strain (Supplementary note 6). This corresponds to a total of 228 microtiter plates (3,648 IsoAnalyst samples and 912 blanks).

After a 6-day incubation time, the metabolites were extracted by quenching 1 ml of bacterial culture with 1 ml of MeOH. The quenched culture was sonicated for 10 min, centrifuged for 3 min at 16,000 g, and diluted 20-fold in 1:1 MeOH/H₂O before injection into an UPLC-HRMS (Waters SYNAPT G2-Si). IsoAnalyst detects the degree of incorporation of each SIL by comparing the isotopologue ratio of the relative intensity of the first isotopologue peak (¹³C, M1) over the one of the monoisotopic mass (¹²C, M0) of the unlabeled dataset, to the sequential isotopologue peaks of the labeled data (for example M2: M1, M3: M2) (McCaughey et al., 2022). If the difference in sequential isotopologue ratio in the labeled dataset is significant (Welch’s two-tailed t-test) when compared to the M1: M0 of the unlabeled dataset, then these positions are considered enriched in the SIL condition (McCaughey et al., 2022).

It is important to note that IsoAnalyst does not define the exact number of precursors for the biosynthesis of a NP, but rather the minimum number of detectable incorporations. This is due to both the presence of unlabeled precursors in the culture medium (which reduces labeling efficiency) and the limit of detection for higher isotopologue signals for low abundance analytes (which limits the number of isotope incorporations that can be quantified) (McCaughey et al., 2022). On average, we detected ∼68 labeled features per strain with a label count of at least 1.

IsoAnalyst for each strain resulted in the acquisition of a total of 3 648 LCMS runs (4 x precursors, 4 x replicates, 2 x precursor variants [labeled and unlabeled], 114 strains). Using MZmine 4.3.0, we applied peak height and retention time cutoffs (Supplementary note 9) and retained only features that were present in three out of four replicates, resulting in a list of 19,060 MS features (Schmid et al., 2023). Following analysis using the IsoAnalyst pipeline, the data were filtered to retain only features with evidence of at least one label count. This selection step reduced the feature list from 19,060 MS features to 1,867 isotopically labeled MS features, highlighting the value of IsoAnalyst labeling for prioritizing products containing the selected biosynthetic precursors. We further decided to retain features that have correlation numbers defined by MZmine (features that have at least two adducts) and used an in-house Python script to predict the MS adduct ([M + H]⁺, [M + Na]⁺, etc.) for the MS feature with the highest intensity for each feature group. Together this yielded a list of 257 analytes for further investigation.

Comparative analysis between IsoAnalyst profiles and GCFs from BiG-SCAPE

Previously, Paulo et. al. performed a BiG-SCAPE analysis on the BGCs detected by antiSMASH 5.0 from the genomes of the 114 strains used in this study to create a network of gene cluster families (GCFs) (Paulo et al., 2024). A total of 151 GCFs were detected, including 78 GCFs containing two or more members and 73 singleton BGCs (Fig. 3A) (Paulo et al., 2024). Comparing the list of strains that produce a given isotopically labeled MS feature with the lists of strains that contain each GCF it is possible to generate hypotheses about the biosynthetic origins of isotopically labeled natural products from the IsoAnalyst experiment. We decided to name this evaluation strategy ‘strain matching’. From our select list of 257 analytes with high MS intensity and clear evidence of isotopic labeling, we calculated the strain match between the NP producing strains and the strains containing BGCs. The list of strains for each NP was then compared to the list of strains of each BGC using the equation:

Figure 3.

Strain-matching pipeline for Burkholderiales metabolomics. (A) Comparison of detected 257 labeled analytes versus 151 predicted GCFs through genome mining. Intersection shows 69 analytes that received non-zero strain matching scores for 54 GCF candidates. (B) Scatterplot for visualization of all 69 strain matched detected analytes, mapping their highest detected MS intensity against their calculated strain matching score (%). The analytes are color coded by whether they successfully scaled up or not (Grey dots: 30 analytes not detected in scale-up; Blue dots: 32 analytes failed manual curations; Orange: 6 succeeded, purple diamond: haereoacidicolin known NP match). Visualization with respective feature IDs are available in Figure S35 (C) Analyte Prioritization and Attrition Workflow. Strain matching validation funnel highlighting the impact of manual curation and scale-up failures, leading to the conclusion that BGCs remain largely silent in minimal media. (D) Contrast between a high confidence strain match (haereoacidicolin) of 33.33% as it is structurally different from the compound that would be produced by the remaining two strains due to differences in the NRPS adenylation domains; and statistically weak correlation of 19.45% (ahl GCF) across 114 strains.

Note that if any strains that produce a given NP do not exist in the list of strains of a given GCF then the strain match was automatically set to 0% to eliminate candidate matches where compounds are produced but the candidate BGC is not present in the genome of the producing organism. Of the 257 analytes, 69 were prioritized as potential products of 54 BGC candidates with strain matching scores ranging from 0.88%–100% (Fig. 3A). Of these 69 labeled analytes, 27 have strain matching scores equal or higher than 50%.

Investigation of labeled analytes with high strain matching scores

To investigate the structures of these putative NPs, we cultured and extracted 12 strains (4 l of culture medium per strain) that produce all the analytes that passed our strain matching workflow (69 analytes). Extracts were fractionated into six fractions based on polarity by stepwise methanol/water elution (10%, 20%, 40%, 60%, 80%, and 100% MeOH v/v) on a medium pressure liquid chromatography (MPLC) system using a RediSep Rf C18 cartridge. Disappointingly, UPLC-HRMS analysis of the resulting fractions revealed poor reproducibility for compound production under large-scale culture conditions. Of the initial 69 target analytes, 30 were not detected following large-scale fermentation and column fractionation (Fig. 3B, Grey dots). The remaining 39 analytes underwent manual verification. This process aimed to confirm that their strain-matching scores aligned with their selective production profiles across the 12 newly cultured strains (Fig. 3B and C). Only 7 out of 39 remaining analytes were retained after manual curations.

Beyond the expected strain matching score of 33.33% for the known secondary metabolite haereoacidicolin (m/z 764) representing one of three potential producing strains, manual corrections revealed that the next highest strain matching score is of 19.45% for the ahl GCF that is encoded in 109 strains. It is important to note that the gap between the strain matching score for haereacidicolin of 33.33% and the next highest score of 19.45% is due to the number of strains that contain the matched GCF within their genomes (Fig. 3D). The NRPS encoding haereoacidicolin is shared amongst 3 strains. Due to differences within the adenylation domains, we expected different structures to be encoded by different BGCs in the same GCF. This means that the expected strain matching score for haereoacidicolin is in fact 33.33%. However, a strain matching score of 19.45% to a GCF of 109 BGCs is not equivalent as there are still a substantial number of strains (88 strains) that do not seem to produce the detected NP (Fig. 3D).

Furthermore, GNPS molecular networking failed to find any structural analogs or subnetworks that would suggest this GCF was part of an active biosynthetic pathway. Thus, it is unlikely that the strain matching score of 19.45% is significant (Supplementary note 11). Ultimately, the combination of poor strain matching scores and the lack of reproducible novel scaffolds indicates that the minimal media environment used in this study does not provide the necessary conditions for robust production of the NPs encoded by annotated BGCs.

Examination and identification of labeled analytes with unique labeling patterns

Although many bacterial natural products are produced by the enzymes encoded by BGCs, some biosynthetic classes are missed by antiSMASH. As an alternative to strain matching scores for BGC-derived analyte selection, we turned our attention to MS features with strong and reproducible labeling patterns whose biosynthetic origins were unknown. Manual review of the labeling results revealed four related analytes with similar retention times (RT) and mass-to-charge (m/z) values that possessed both high label count and high label variety (9 ACE, 1 PROP, 1 MET, and 1 GLU; Fig. 4B). Examination of the mass defect and isotopologue distribution from the HRMS data for one member of this family (m/z 494.3261) indicated the presence of a phosphorous atom in the chemical formula. ³¹P NMR spectroscopy revealed a phosphorous signal with a chemical shift of ∼0.47 ppm, diagnostic of phospholipids and phosphoethanolamines (Fig. 4A, Figure S26). 1D and 2D NMR analyses, coupled with compound class prediction from MS/MS fragmentation using the SIRIUS software package revealed the presence of two closely related lysophosphatidylethanolamine (LPE) analogues, one possessing a double bond and the other a cyclopropyl ring in the lipid tail (Fig. 4C and D, Figures S21–26). The final positions of the double bond and cyclopropyl group in the lipid tail were not determined due to substantial signal overlap in the ¹H and ¹³C NMR spectra (Figure S21–22) and the absence of diagnostic fragments in the MS/MS data.

Figure 4.

Identification of labeled analytes with unique labeling patterns. (A) ³¹P NMR of LPE, for full size spectrum refer to Figure S26. and IsoAnalyst profile of LPE compounds. Bar chart numbering refers to individual label counts: A: [1–¹³C]-acetate, P: [1–¹³C]-propionate, M: [methyl-¹³C]-methionine, G: [1–¹⁵ N]-glutamate. (B) ¹H NMR of LPE at 500 MHz, for full size spectrum refer to Figure S21. (C) Structures of LPE 1 and 2. a + b = 12. (D) COSY and HMBC correlations of 3-hydroxybutyrate heptamer methyl ester (HB7-OMe, 1^') and 3-hydroxybutyrate octamer (HB8, 2^'). (E) Structures of 3-hydroxybutyrate octamer methyl ester (HB8-OMe, 3^'), and 3-hydroxybutyrate hexamer methyl ester (HB6-OMe, 4^'). (F) Organization of the pha gene cluster in Paraburkholderia fungorum sp. RL17-337-BIB-A. The arrows represent open reading frames (ORFs) and point in the direction of transcription. All ORFs are color-coded based on proposed function (Table S6).

LPEs are derived from the lipase-driven hydrolysis of the fatty acid chains of phosphatidylethanolamine (Cao et al., 2023). LPEs were previously noted to add to structural modification of cell membranes, thus aiding in antimicrobial resistance for the producing organism. They have also been hypothesized to provide a pool of compounds that can be converted to phosphatidylethanolamine for membrane maintenance to repair potential damage from environmental stress (Cao et al., 2023). These compounds are also produced by pathogenic bacteria as signaling molecules to activate virulence; for example, Campylobacter jejuni uses LPEs to permeabilize eukaryotic host cells (Cao et al., 2022). The presence of these NPs could be ecologically or physiologically significant under nutrient-limited conditions. Minimal media environments can impose significant metabolic stress, potentially triggering the production of NPs that can have more than one function, such as cellular repair or chemical defense.

Review of the labeling data identified a second group of compounds with consistent PROP labeling with m/z 571, 657, 729, 743, 915, and 1,001 [M + Na]⁺. These compounds showed low strain matching scores (<15%) with the terp1, terp2, pho, ahl, and ape GCFs, making it unlikely that they were encoded by a BGC identified by antiSMASH. The SIRIUS platform annotated these compounds as having polyketide-like structures. Their presence in a defined number of strains, consistent labeling, and potential noteworthy structures motivated us to explore this compound class.

Large-scale liquid cultures (28 × 1 l culture in 2.8 l Fernbach flasks) of P. sediminicola RL17-337-BIB-A were grown in M9TG media with adsorbent resin (20 g Amberlite XAD-7 HP resin per liter culture). Cultures were shaken for 72 h before vacuum filtration followed by organic extraction of the resin/cell slurry. Crude organic extracts were combined and separated into six fractions by stepwise methanol/water elution using MPLC as described above. The fraction that eluted with 60% methanol was further fractionated using column chromatography through stepwise methanol/water elution (10, 20, 40, 60, 80, and 100 v/v) using a C₁₈ Sep-Pak cartridge, followed by HPLC purification (Phenomenex Kinetex XB-C18 column; 250 × 4.6 mm, 5 µm) to afford 600 µg and 1 mg of the NPs with m/z values 657 and 729, respectively.

Assessment of the HRMS m/z value for the precursor ion 657.2736 [M + Na]⁺ indicated a molecular formula of C₂₉H₄₆O₁₅ (calc. 634.2837 Da, 1.11 ppm error). The planar structure was assembled using 1D and 2D NMR spectra, including ¹H, gCOSY, gHSQC, and gHMBC spectra (Table S5, Figure S1–5), despite limited compound solubility that reduced the signal-to-noise ratio for NMR spectra. COSY correlations revealed three separate spin systems: a β-hydroxybutyric acid tail, 6 butyric acid moieties, and a terminal methyl ester (Fig. 4E). The β- hydroxybutyric acid tail showed clear COSY correlations between the methine proton on C28 (4.07 ppm) and the methyl protons on C29 (1.14 ppm), as well as the methylene protons on C27 (2.34 ppm). The methine at C28 also possessed a COSY correlation to an exchangeable doublet signal that lacked a cross-peak in the HSQC spectrum and was indicative of a secondary alcohol signal. The carbon chemical shift of C28 (δ_C 64.1 ppm) was consistent with the presence of a hydroxyl group, completing this first spin system as β-hydroxybutyric acid (Fig. 4E).

The second spin system contained sets of overlapping signals with similar chemical shifts and high integration values, suggesting the presence of repeating butyric acid units. Similarly to the first spin system, a CH group with deshielded δ_C 67.3 ppm and integrating for 6 protons, shows COSY correlations with a CH₃ (integrating for 18 protons) and CH₂ group (integrating for 12 protons) indicating 6 units of butyric acid. These 6 butyric acid units were connected via HMBC correlations between both the CH (5.19 ppm) and CH₂ groups (2.5–2.6 ppm) to the carbonyl of an ester group at δC 169.4 ppm. The third spin system is a methoxy group on an ester with characteristic chemical shifts of δH 3.62 ppm and δC of 51.2 ppm and an HMBC correlation to a carbonyl at δC 171.6 ppm. All three spin systems were connected through HMBC correlations between methylene protons to the carbonyl esters: the CH₂ on β-hydroxybutyric acid (δH 2.34 ppm) correlated to the carbonyl at 171.1 ppm as well as a methylene proton from the second spin system at δH 2.6 ppm thus connecting spin system 1 and 2; methylene protons at δH 2.50 ppm of the second spin system correlated to the carbonyl of the methoxy group of spin system 3 at δC 171.6 ppm, thus connecting spin system 2 and 3 (Fig. 4E). Due to the presence of multiple butyric acid units, we named this compound 3-hydroxybutyrate heptamer methyl ester (HB7-OMe) following established oligomer nomenclature.

Assessment of the HRMS m/z value for the precursor ion 729.2941 [M + Na]⁺ indicated a molecular formula of C₃₂H₅₀O₁₇ (calc. 706.3048 Da, 0.11 ppm error). Similarly to HB7-OMe the planar structure of m/z 729 was assembled by analysis of 1D and 2D NMR spectra including ¹H, ¹³C, gCOSY, gHSQC, and gHMBC spectra with the addition of gTOCSY (Table S6, Figures S1, S6–11). The latter confirmed the COSY correlations of the first two spin systems as well as the established spin systems via TOCSY correlations between methyls and methylenes separated by CH groups. This compound has an additional butyric acid moiety in the polymeric chain but lacks the terminal methoxy carbon, possessing instead a free carboxylic acid (Fig. 4E). This compound was therefore named 3-hydroxybutyrate octamer (HB8).

¹H NMR spectra of compounds corresponding to m/z 571 and 743 [M + Na]⁺ were very similar to the spectra for HB7-OMe and HB8, suggesting that they are from the same compound family. From their HRMS values we determined their chemical formulae to be C₂₅H₄₀O₁₃ and C₃₃H₅₂O₁₇, respectively. We then used MS/MS analysis to confirm that these structures are from the same compound family via a diagnostic neutral loss of 86.0368 Da representing but-2-enoic acid (Figure S18–20). These methyl-esterified hexamer and octamer structures were named HB6-OMe and HB8-OMe, respectively (Fig. 4F).

Unmethylated polyhydroxybutyrate (PHB) oligomers have previously been reported to be produced through bacterial fermentation of sugars or lipids. For example, Pseudomonas putida produce polyesters such as polyhydroxyalkanoates (PHA), while industrial processes use recombinant Escherichia coli to produce these PHA from glucose or propionate, which aligns with the PROP labeling observed in our isolated compounds (Figure S29–32) (Vasile and Zaikov, 2009). PHAs are used in the food and agricultural industries as an ecologically-sensitive way to produce plastics for food packaging (Jayalath and Alwis, 2025). Poly-3-hydroxybutyrate (P3HB) is one type of PHA that is produced by bacteria under environmental stresses such as limited essential nutrients including bioavailable nitrogen and phosphorus (Sabbagh and Muhamad, 2017). Under these conditions some bacteria utilize available carbon to produce PHA as energy storage for future use. These oligomers are not essential components to bacterial cells for growth and replication as they are intracellular storage granules that are biosynthesized under stressful conditions and not during growth. Therefore, PHA can offer a fitness advantage for the microbe’s long-term survival under stressful conditions, classifying them as secondary metabolites (Sabbagh and Muhamad, 2017). Antimicrobial activity of HB8 was assessed against a panel of 19 clinically-relevant bacterial pathogens (Table S4 and Supplementary note 12) using the previously developed BioMAP antibacterial profiling platform (Wong et al., 2012). Testing was performed across a 13-point, twofold serial dilution series ranging from a maximum concentration of 256 µg/ml to a minimum concentration of 0.0625 µg/ml, corresponding to 12 dilution steps and an overall 4 096-fold concentration range. MIC values are given in Table S10. No measurable antibacterial activity was observed against any of the tested strains within this concentration range; therefore all MIC values are greater than 256 µg/ml.

To the best of our knowledge, there is only one report of methyl-esterified 3-hydroxybutyrate (ME-3HB) oligomers which were naturally biosynthesized by bacteria: a study by Koskimäki et al., (2016) that reported the discovery of ME-3HB dimers and trimers. In this report the authors found that these compounds protect bacteria from hydroxyl radicals as they are not only induced when exposed to reactive oxygen species (ROS) but also exhibit antioxidant activity almost 11-fold higher than ascorbic acid which was used as a positive control. This bioactivity is essential for plant microbial endosymbionts to successfully infect host cells and establish a functional symbiotic relationship. However, this paper reported ME-3HBs that only extend to the hexamer oligomer (m/z 571). Here, we report four new ME-3HBs containing additional polymer units including HB7-OMe, HB8-OMe, HB10-OMe and HB11-OMe (Fig. 4F). Interestingly, the nonamer m/z 829 [M + Na]⁺ was detected in our samples but did not contain detectable levels of isotope labeling using IsoAnalyst, so was excluded from further study.

Selective production of methyl-esterified 3-hydroxybutyrate oligomers (ME-3HBs)

Under M9TG minimal media conditions, IsoAnalyst detected the ME-3HB hexamer oligomer (HB6-OMe) in 16 out of 114 BKD strains, distributed across four clades of the phylogenomic tree (Figure S27). These include P. sediminicola/fungorum (pink clade I, 9 strains), P. phytofirmans (purple clade H, 5 strains), P. strydomiana A (teal clade F, 1 strain), and P. strydomiana B (blue clade G, 1 strain). The selective detection of these compounds among ∼15% of the assayed bacterial strains reveals the potential presence of differential regulation in phylogenomically similar BKD strains. Thus, we decided to investigate the biosynthesis of these compounds.

Biosynthesis of ME-3HB was previously reported to involve enzymes required for PHB polymerization which include phaA, phaB, and phaC, followed by depolymerization forming shorter oligomers by PHA depolymerase known as phaZ (Koskimäki et al., 2016). The biosynthesis of PHB starts with the enzyme 3-ketothiolase (PhaA) which couples two acetyl-CoA molecules to form acetoacetyl-CoA which is then reduced by the enzyme acetoacetyl-CoA reductase (PhaB) to form 3-hydroxybutyryl-CoA (Koskimäki et al., 2016). The acyl moieties of the latter go through polymerization using PHB synthase (PhaC) to biosynthesize the larger PHB oligomers. PhaZ depolymerase is then responsible for the degradation of the large PHBs into smaller PHB molecules, which can be methyl-esterified upon cleavage (Koskimäki et al., 2016).

Identification of the PHB biosynthetic gene cluster

The genomes of all 114 Burkholderiales strains were searched for homologs of phaA, phaB, phaC, phaR, phaZ, and mxaF using command-line BLAST+ (NCBI-BLAST^®, v2.15.0) (Camacho et al., 2009). A conserved pha cluster, consisting of phaABCR (Fig. 4G, Table S8), was identified in 110 strains in our collection. The individual genes phaA, phaB, phaC, and phaR can still be found in the remaining 5 strains (Caballeronia grimmiae sp. RL17-374-BIB-E, Caballeronia glebae sp. RL18-006-BIC-A, Caballeronia jiangsuensis sp. RL17-374-BIF-D, Herbaspirillum rhizosphaerae sp. RL21-008-BIB-B and Herbaspirillum lusitanum sp. RL21-008-BIB-A), but they are only partially clustered (Table S9). For example, in the two Herbaspirillum strains, phaBCR are co-localized, but a phaA homolog is found on a different contig. In two Caballeronia strains, phaAB co-localize, while phaC and phaR homologs are found on contig edges, so their co-localization cannot be confirmed from the fragmented genome assembly.

Regarding phaZ and mxaF, all 114 strains maintain at least one copy of phaZ, but this number can be as high as five for more divergent homologs. Similarly, only 19 strains in the collection lack any copies of mxaF, with the remaining strains maintaining between one and five homologs scattered throughout the genome. Only one copy of phaZ co-localizes with other pha genes, found in Caballeronia grimmiae sp. RL18-154-BIB-A, the only strain to contain two pha clusters. The first is identical to that shown in Fig. 4G as conserved across most of the collection, but the second contains duplicated phaABC genes with nearby phasin proteins, a copy of phaZ, and interspersed transposases (Table S9).

PHBs have been described as beneficial for bacterial-plant interactions, which is supported by the pervasiveness of the pha cluster in our collection of environmental, rhizosphere-derived Burkholderiales bacteria (Silveira Alves et al., 2019). However, it is surprising that despite this, the putative products (1’ and 2’) were observed from only 16 of the 114 strains. Production may be activated differently in the various strains. Future work should confirm the necessity of the pha cluster for compound production via gene deletion and interrogate expression levels of the pha genes across a variety of conditions.

Conclusions

In conclusion, this study highlights the value of stable isotope labeling experiments when used in large-scale experiments, as a strategy to significantly reduce candidate MS feature counts. Comparison of the labeled MS features from our 114 Burkholderiales strain library against the taxonomy and genomes of their producing strains revealed diverse and strain-specific labeling patterns among microbial metabolites. From our analysis, we discovered two classes of compounds that are important for cell survival and are derived from biomolecule metabolism. The first class of compounds is comprised of two analogues of LPEs, one that included a double bond embedded in the fatty acid tail and the other containing a cyclopropane group (Fig. 4D). Compounds of this class have previously been described as forming a pool of ‘back-up’ starting material that can be readily converted to phosphatidylethanolamine for membrane maintenance and repair from environmental stress damage (Cao et al., 2023). Analogues of these compounds have also been associated with virulence activation such as in Campylobacter jejuni where they are used to permeabilize eukaryotic host cells (Cao et al., 2022). The detection of these LPE analogues can be explained by the use of minimal culture media which can impose significant environmental stress, thus encouraging the production of metabolites that have more than one function to maximize nearby resources and bacterial energy expenditure. This minimal media metabolic stress also resulted in the detection of P3HBs from the PHA family, including HB8 which is a carbon storage molecule produced in response to nutrient limiting conditions (McAdam et al., 2020, Sabbagh and Muhamad, 2017). ME-3HBs have been previously reported as small dimers and trimers which were shown to be antioxidants essential for plant host infection (Koskimäki et al., 2016). This work expands the range of known PHB structures to heptamer (HB7-OMe, 1’) and octamer (HB8-OMe, 3’) variants, and provides preliminary evidence for the biosynthesis of these environmentally important metabolites. The detection of these compounds in only 16 out of the 114 Burkholderiales strains, despite the presence of at least one copy of the core pha biosynthetic genes in all strains, highlights potential differences in gene regulation in phylogenomically similar strains. Thus, using IsoAnalyst, we detected new LPE analogues and ME-3HBs that we characterized using MS and NMR spectroscopy, to better understand the chemistry employed by Burkholderiales under nutrient depleted conditions.

Author contributions

The manuscript was written through contributions of all authors. Conceptualization, RGL and LZ; methodology, RGL, LZ, SCH; investigation, LZ, SCH, MS; writing—original draft, LZ, RGL, writing—review & editing, RGL, LZ, ASE, SCH, MS; funding acquisition, RGL, ASE; supervision, RGL, ASE. All authors have given approval to the final version of the manuscript.

Funding

This work was supported by the Natural Sciences and Engineering Research Council of Canada (Discovery grant to RGL and CGSD award to LZ) and the National Institutes of Health (R01-GM129344 to ASE and RGL).

Conflicts of interest

The authors report no conflict of interest.

Data availability

Spectroscopic data, biological screening data and data on biosynthetic enzymes available in the supporting information file. Data on labeled metabolites available in supporting information .xlsx files.