Spatial chemistry of citrus reveals molecules bactericidal to Candidatus Liberibacter asiaticus

Abstract

Huanglongbing (HLB), associated with the psyllid-vectored phloem-limited bacterium, Candidatus Liberibacter asiaticus (CLas), is a disease threat to all citrus production worldwide. Currently, there are no sustainable curative or prophylactic treatments available. In this study, we utilized mass spectrometry (MS)-based metabolomics in combination with 3D molecular mapping to visualize complex chemistries within plant tissues to explore how these chemistries change in vivo in HLB-infected trees. We demonstrate how spatial information from molecular maps of branches and single leaves yields insight into the biology not accessible otherwise. In particular, we found evidence that flavonoid biosynthesis is disrupted in HLB-infected trees, and an increase in the polyamine, feruloylputrescine, is highly correlated with an increase in disease severity. Based on mechanistic details revealed by these molecular maps, followed by metabolic modeling, we formulated and tested the hypothesis that CLas infection either directly or indirectly converts the precursor compound, ferulic acid, to feruloylputrescine to suppress the antimicrobial effects of ferulic acid and biosynthetically downstream flavonoids. Using in vitro bioassays, we demonstrated that ferulic acid and bioflavonoids are indeed highly bactericidal to CLas, with the activity on par with a reference antibiotic, oxytetracycline, recently approved for HLB management. We propose these compounds should be evaluated as therapeutics alternatives to the antibiotics for HLB treatment. Overall, the utilized 3D metabolic mapping approach provides a promising methodological framework to identify pathogen-specific inhibitory compounds in planta for potential prophylactic or therapeutic applications.

Introduction

Huanglongbing (HLB) is a highly destructive citrus disease exhibiting complex symptomatology. As one of the most serious plant diseases worldwide, HLB poses a major threat to US citrus production, which is primarily concentrated in Florida, California, Arizona, and Texas¹. All commercial citrus varieties are HLB-susceptible to various degrees and HLB has been identified in 51 out of 140 countries that produce citrus². HLB is caused by the bacterium Candidatus Liberibacter asiaticus (CLas) and is spread by the insect vector Diaphorina citri, commonly known as the Asian Citrus Psyllid (ACP). Early HLB symptoms include leaf mottling and yellow shoots, which leads to root dieback and tree death³. Moreover, the effect on fruit development has major downstream consequences on fresh fruit and juice quality⁴. Fruit from HLB-infected trees are small, misshapen, with irregular maturation patterns that yield unpalatable flavor profiles. The disease has a long asymptomatic stage, making early detection and management particularly challenging^4,5.

The ACP, first reported in Florida in 1998, caused rapid regional HLB outbreaks by 2005, resulting in endemic establishment and billions of dollars in damages in industry^5,1. ACP has since spread east to west across the US citrus belt, and HLB is now detected in thousands of Californian urban citrus plants, threatening commercial groves⁶. This rapid spread of HLB follows a consistent pattern: once the ACP vector establishes itself in a new area, HLB typically appears soon after. Disease management has mainly focused on insect vector control with repeated sprays of synthetic insecticides⁷, and no sustainable bacterium-targeting HLB interventions yet exist. The uneven distribution of CLas within infected trees further complicates reliable detection using PCR and assessment of chemical and biological perturbations caused by the bacterium⁸. While HLB-specific microbial and molecular biomarkers⁹, and HLB-induced proteome changes have been identified^10,11, these advances have yet to translate into effective disease management strategies.

CLas is a phloem limited bacterium, that can not be axenically cultured⁸. The phloem is a high pressure vascular tissue that transports photosynthates (metabolic products) and sugars throughout the plant. These features of the pathosystem make it particularly difficult to study and manage. Currently, oxytetracycline antibiotic trunk injection is the primary approach to reduce and control CLas^12–14. However, antibiotic deployment is economically unsustainable for growers and imposes risk to human and environmental health by potentially increasing antimicrobial resistance in bacterial populations. A “precautionary tale”—widespread antibiotic administration has led to the rapid emergence of resistant Candidatus Liberibacter asiaticus strains in China¹⁵. Thus, alternative bactericidal agents need to be explored for viable and sustainable approaches to CLas control.

Metabolomics, both nuclear magnetic resonance (NMR) and mass spectrometry (MS)-based, gives valuable insights into molecular distributions, including those underlying pathogenesis, and thus may provide leads on potential therapeutic molecules. These techniques have repeatedly demonstrated their value in providing important insights into pathogenesis and the mechanistic details of underlying biology^16,17. Past investigations employing NMR and MS established associations between CLas infection, such as terpenoid/sugar disruption underlying symptomatic flavor decline¹⁸. However, uneven HLB distribution hinders generalizing systemic changes^19,20. As symptoms manifest locally, pinpointing locally perturbed metabolites is essential to understand the disease and propose targeted therapies.

Recent advances in metabolomics enable exploring spatial chemical distributions^21–24. This includes mass spectral molecular networking, which supports annotation propagation within a network of compounds clustered based on their structural similarities²⁵. This method can capture metabolomic transformations and cluster related compounds, partially addressing the challenge of sparse annotation due to limited compound libraries. We hypothesized that detailed infection mapping would expedite discovery of compounds active against CLas in planta, to provide candidates for novel HLB therapeutics. In this work, we visualized the HLB effect on the metabolome of citrus plants using 2D and 3D molecular maps to gain pathological insights, illuminating possible bacterium-mediated host metabolite exploitation. While we focused on citrus, this approach, in principle, should be generally applicable to a wide range of plant-pathogen systems.

Results

Metabolome shifts in field and greenhouse trees with HLB

We carried out an untargeted metabolomics analysis of citrus tree tissues across a spectrum of disease severity collected from seven HLB-infected Florida orchards, where the disease is endemic. Tissues throughout the citrus tree: stems, roots, and leaves were collected and analyzed. Because of the complex metabolome of citrus trees across different tissues, we used molecular networking^25,26 to map and explore the detected chemistries. The resulting network grouped molecules that fragmented in a similar fashion, thus capturing their structural similarity and aiding in understanding the chemical distributions^26,27. The network of the detected metabolome across grove trees is shown in Fig. 1a,b. The metabolome varied as HLB progressed (Fig. 1c), with molecules associated with disease severity forming distinct clusters in the network. This suggests that metabolic disruption in HLB progresses not through individual molecular changes, but shifts across entire biochemical families. Clusters of compounds—dilinolenins, fatty amides like octadecenamide, glycerophosphocholines, terpenoids, and, prominently, flavonoids—showed changes linked to symptom severity across tissues. To control for confounding factors such as length of infection, local conditions etc., samples were also collected from a controlled greenhouse environment, including infected and uninfected healthy trees (confirmed using qPCR). Metabolome differences related to HLB symptomatology were much more pronounced in greenhouse samples as compared to field samples, even though symptoms were significantly less pronounced. The most discerning diagnostic markers, identified via partial least squares discriminant analysis (PLSDA) and Random Forest, were found to be various flavonoids. This implicates the flavonoid pathway as the main metabolomic “barometer” of infection^19,28–33.

Fig. 1.

Global network analysis of metabolomic shifts in HLB-infected trees. Citrus trees (leaves, stems and roots) were sampled across multiple groves in Florida. (a) Global network of the metabolome detected in samples collected from grove trees in Florida, which included trees exhibiting symptoms across a range of disease severity (1—appears healthy to 5—severe HLB symptoms). (b) A close-up of a network cluster showing distributions of molecular abundances in trees with different disease ratings (higher values correspond to higher symptom severity); node size is related to total compound abundance. The compounds in the cluster were present in higher amounts when the disease was more severe. The selected example shows perturbation in amounts of tryptophan, indicative of altered metabolism, and a metabolite known to be of microbial origin, indole-3-lactic acid (ILA), a tryptophan metabolite that is known to play a role in microbe-host interactions; ILA is associated with increased disease severity. (c) A supervised analysis (partial least squares discriminant analysis, PLSDA) of tissues from trees in orchards across Florida showed metabolome stratification according to disease severity (Q2 0.1775). (d) Unsupervised analysis (principal component analysis, PCA) of tissues from greenhouse-reared trees indicated drastic differences in the metabolome of healthy and infected symptomatic tissues.

3D metabolome mapping

To explore the spatial patterns in distribution of metabolites, molecular maps were generated using the ‘ili tool²¹. We further elucidated these patterns for HLB-discriminating molecules. This mapping was conducted on tissues from greenhouse-reared trees of the same age and length of infection. 3D metabolome maps were rendered by collecting individual leaves along an infected branch and mapping specific metabolite concentrations, such as the flavonoid scutellarein tetramethyl ether, back to branch images (Fig. S1). This allowed for visualization of metabolite concentration along the same branch where apical leaves exhibited HLB symptoms and basal leaves are asymptomatic (Fig. S1a). Notably, the distribution of the flavonoid scutellarein tetramethyl ether may be related to leaf age rather than symptomology. Therefore, the symptom-related patterns were investigated at single-leaf resolution, where leaves are age-matched. At this scale, symptom-related patterns were noticeable to an expert, with various flavonoids sharing similar spatial propagation pattern; depleted in visibly chlorotic and mottle signatures of infection (Fig. S2). At the same time, oxidized flavonoids tended to exhibit opposite spatial patterns, suggestive of mutual interconversion (Figs. 2, S3).

Fig. 2.

3D mapping of the distribution of the flavonoids, didymin and hesperidin in infected and healthy citrus branches: (a) Didymin (m/z 595.2026 (rt 208.16)), (b) Hesperidin m/z 611.1973 (rt 131.25)) and infected plant: (c) Didymin and (d) Hesperidin. Didymin was present at higher abundance in younger leaves of healthy plants and was depleted in the infected plant, while hesperidin increased in abundance in the infected plant and didymin was depleted, indicating possible formation of hesperidin due to oxidation of didymin.

Other than flavonoids, we noted feruloylputrescine, a conjugate of ferulic acid (part of the flavonoid biosynthesis pathway) and putrescine as molecules descriptive of HLB symptoms. Feruloylputrescine is preferentially accumulated in apical symptomatic leaves, exactly opposite the non-oxidized flavonoid depletion patterns (Fig. S4a). The ferulic acid abundance decreased concomitantly with putrescine conjugation, indicative of the possible conversion into feruloylputrescine (Fig. S4b). The same trends were evident on the single-leaf scale, with a clear correspondence of spatial distributions to the chlorotic spots phenotype (Fig. 3; note that in greenhouse-grown plants the visual differences may not be stark, particularly in early or mild infection). Another related phenolic acid, hydroxycinnamic acid (p-coumaric acid), showed a similar distribution, enriched in symptomatic areas (Figs. S5, 3C).

Fig. 3.

2D mapping of polyamine derivatives within single leaves. (A) Unmodified images of leaves with different disease severity; the symptoms are subtle, with chlorotic spots appearing as slightly lighter areas. (B) The leaves shown in (A) in a color scheme that more clearly shows chlorotic spots (chlorotic spots appear as dark shade areas). (C) Mapped distributions of (top to bottom): ferulic acid—H₂O (only water loss ion was observed), hydroxycinnamic acid and feruloylputrescine. 2D maps: healthy, infected asymptomatic, infected symptomatic (from left to right). The mapped symptomatic and asymptomatic leaves were sampled from two branches of a tree that was confirmed by qPCR to be infected with CLas (C_t = 19.6). Increases in hydroxycinnamic acid and feruloylputrescine were concomitant with increased disease severity, while ferulic acid exhibited the opposite trend. (D) Plot showing Liberibacter crescens (a culturable surrogate for CLas) growth inhibition assay, when challenged with 35 mg/mL solution by ferulic acid, feruloylputrescine or putrescine as quantified by the diameter of the zone of inhibition. Ferulic acid had a pronounced inhibition effect.

Metabolic modeling of CLas uptake rates of plant metabolites

A possible rationale for the observed spatial patterns in the metabolome (in particular, the low abundance of ferulic acid in symptomatic tissue) may be the metabolism of ferulic acid by the bacterium that would protect itself from antimicrobial activity. A supporting argument for this would be the observation that phenolic acids, like ferulic acid, reduce the growth rate of CLas, while in conjugation with putrescine, this effect is reduced due to the synthesis of feruloylputrescine. Therefore, we tested this hypothesis by conducting model simulations using the previously established genome-scale metabolic network of CLas¹⁵. We simulated uptake rates of plant metabolites by CLas for compounds that are part of arginine and proline metabolism, as well as those present in the biosynthesis of phenylpropanoids. The model predicted that uptake rates of ferulic acid between 1 × 10^–10 and 1 × 10^–8 mmol/gDW/h would be sufficient to reduce the growth rate of CLas up to 90%. These predictions were experimentally confirmed and putrescine uptake rates in this range did not predict a reduction of CLas growth. Interestingly, when these ferulic acid uptake rates (1 × 10^–10 and 1 × 10^–8 mmol/gDW/h) were combined with putrescine uptake rates between 1 × 10^–2 and 10 × 10^–1 mmol/gDW/h, the model predicted the highest CLas proliferation (Fig. 4).

Fig. 4.

Genome-scale metabolic network simulations about the effect of feruloylputrescine biosynthesis on the growth of CLas. Contour plot shows the predicted growth rates of CLas, while varying putrescine and ferulate uptake rates. Individual scatter plots show the response of CLas growth to changes in the individual uptake rates of metabolites of interest (mmol/gDW/h). 2D maps of leaves show from left to right: healthy, infected symptomatic, infected asymptomatic samples of citrus plants. 2D maps of plants show from left to right: healthy and infected citrus plants.

In vitro verification of model predictions with disc assay

In order to further confirm in silico predictions, we assayed putrescine, ferulic acid, and feruloylputrescine for potential antibacterial activity using a previously developed disc diffusion assay based on Liberibacter crescens, a culturable surrogate of the unculturable CLas³⁴. Metabolic modeling of the pathogen described in the “Metabolic modeling of CLas uptake rates of plant metabolites” section above predicted putrescine and feruloylputrescine to be mildly beneficial and ferulic acid to be inhibitory to CLas growth. The model predictions were consistent with the metabolites’ spatial distributions in greenhouse citrus tree leaves. Disc assay results showed that putrescine and ferulic acid were indeed moderately and highly inhibitory to L. crescens, respectively. The conjugated metabolite, feruloylputrescine, was not inhibitory to L. crescens growth as predicted by our model (Fig. 3D). These in vitro results, in conjunction with the observed interconversion between ferulic acid, putrescine, and feruloylputrescine in HLB symptomatic trees, indicate that CLas likely directly, or through manipulation of host metabolic pathways, conjugate ferulic acid to putrescine producing the non-toxic feruloylputrescine, which increasing its survivability in the host.

Ex vivo verification of model predictions with CLas-citrus hairy root assay

To further validate the toxicity of ferulic acid and flavonoids, whose biosynthesis is disrupted due to conjugation, against the target pathogen CLas, we conducted a CLas-citrus hairy root cultures assay³⁵. As a possible low-cost, accessible therapeutic intervention, we tested over-the-counter citrus peel extracts as a widely available source of native citrus bioflavonoids. CLas-containing hairy roots were treated with bioflavonoids and ferulic acid for 72 h (Fig. 5a). Untreated and ethanol (0.2% v/v) samples were used as negative controls. A reference antibiotic, oxytetracycline (OXY), reported as an inhibitor of Clas^14,36,37, was used as a positive control. After treatment, all tissue samples were exposed to PMAxx dye (propidium monoazide, Biotium, Fremont, CA) that allows measurement of only live CLas bacterial DNA, following DNA extraction and molecular diagnostics. The relative titers of CLas were estimated using qPCR. We found that bioflavonoids significantly inhibited CLas (p ≤ 0.05) in a dose-dependent manner (125, 250, and 500 ppm), whereas ferulic acid showed inhibition at 125 and 250 ppm (p ≤ 0.05), on par with oxytetracycline when compared to untreated controls (Fig. 5b).

Fig. 5.

Bioflavonoids and ferulic acid inhibit the growth of Candidatus Liberibacter asiaticus (CLas) in CLas-citrus hairy roots. (a) The CLas-citrus hairy roots were treated for 72 h with 125, 250, 500, and 1000 ppm of bioflavonoids and ferulic acid. Oxytetracycline hydrochloride (OXY)-treated hairy roots (250 and 500 ppm) were used as a positive control, and untreated (UT), ethanol (0.2%) used to dissolve the compounds was used as a negative control. Relative titers of CLas (b) were estimated after 72 h of treatment, followed by qPCR analysis. Error bars represent the standard error of the mean (n = 6), and p values were calculated by Student t-test relative to untreated samples.

Discussion

We explored HLB disease metabolome dynamics using spatial mapping, predictive modeling, in vitro, and ex vivo assays to identify endogenous citrus compounds with plant therapeutic potential. Metabolites altered by HLB symptoms were consistent among field and greenhouse trees, with flavonoids comprising the most discriminating “biomarkers”, supporting previous observation^19,37–39. The discriminant flavonoids structures did not comport to a notable trend, and molecules differing by various backbone substitutions, e.g. addition of a glycan (which increases the compound’s solubility), were found to be predominantly declining in infected symptomatic tissues. This implies upstream flavonoid pathway suppression that potentially aids pathogen establishment, fitting the known antimicrobial and defense roles of these compounds. However, this information alone is insufficient to understand disease etiology^9,37,38,40. For example, certain flavonoids were instead found to increase with HLB symptom development.

Overlaying 3D distribution onto molecular networking provided further insights into functional interconnections in molecular distributions, not accessible otherwise. For example, the association with HLB symptoms is notably different for oxidized versions of some flavonoids. As shown in Fig. 2, hesperidin accumulation mirrored declines in the flavonoid didymin, possibly reflecting oxidative conversion of the latter into the former during infection. Such localized oxidative shifts were further evident across flavonoid families (Fig. S3), aligning with oxidative stress reports previously described^41,42. In another example, 3D maps revealed that leaf age was the predominant driver of flavonoid abundance, rather than disease severity. Younger leaves possess elevated flavonoids, necessitating age-matching to correctly determine infection biomarkers.

Exploring the HLB-discriminating ability and corresponding 3D maps for various metabolites allowed us to identify other molecules of interest in addition to flavonoids. In particular, feruloylputrescine was noted to be highly associated with symptom severity. Distribution patterns implicated the conversion of ferulic acid into feruloylputrescine, mirrored by ferulate disappearance and putrescine buildup (Fig. 3). This shift was further echoed by hydroxycinnamate rises, likely reflecting the oxidation of cinnamic acid precursors. Taken together, spatial patterns indicate metabolic channeling away from phenylpropanoid biosynthesis toward polyamine conjugation due to infection.

Exploring upstream pathways points to a branch point: while general phenylalanine supply appeared unaffected, p-coumaroyl-CoA intermediates depleting from the flavonoid pathway were instead visible as hydroxycinnamate en route to feruloylputrescine. This suggests that CLas has evolved to counteract both ferulic acid and flavonoids formation, ostensibly as a self-defense against their toxicity.

To confirm that these molecules are indeed toxic to CLas, we conducted the disc and hairy roots assays that confirmed their bactericidal effects toward CLas. Although the disc assay exploited a surrogate organism, Liberibacter crescens, the hairy roots assay is a way to directly assess the toxicity, both genetic- and chemical-based to CLas itself^35,43,44. In this approach, CLas is maintained in citrus hairy root tissues, and the antimicrobials are directly infiltrated into the roots by vacuum, thus overcoming the delivery bottleneck that exists with conventional whole tree assays^13,36,45. Additionally, because citrus hairy roots have intact vasculature, they are much closer to the in planta citrus environment where CLas resides, thus providing reliability over testing against surrogate bacteria or heterologous systems^46–49. The assay has demonstrated a clear bactericidal effect of the flavonoids on par with the anti-CLas activity of oxytetracycline. In the case of ferulic acid, the lower dosages (125 to 250 ppm) are highly effective, while increasing the dosage to 500 or 1000 ppm did not seem to be effective, possibly due to the toxicity of the compounds to the hairy roots themselves⁵⁰. Both flavonoids and ferulic acid bacterial inhibitory abilities have been well characterized in other systems, but the mode of action is not clear^51,52. These phytochemicals may disrupt bacterial cell-to-cell communication⁵³, modify biofilm production and inhibit swimming motility⁵⁴, impact bacterial membranes⁵⁵, or even create pores or rupture cell members⁵⁶. Nevertheless, these experiments confirmed the robust anti-CLas activity of the flavonoids^55,56. It is not known how these compounds may impact other members of the citrus microbiome. Indeed, both bacterial and fungi communities drastically shift as HLB symptoms advance⁹, and citrus-associated fungi can produce Liberibacter spp. inhibitor compounds³⁴. Future studies should consider the interactions of these compounds with the microbiome, and vice versa, to increase efficacy of disease management applications.

With the efficacy against CLas confirmed, the ferulic acid and bioflavonoids present an opportunity for field application as novel Huanglongbing therapies. These compounds are plant-derived, non-toxic, biodegradable, inexpensive to manufacture at scale, and thus would be a very appealing class of therapeutics. They may help to bypass negative environmental and economic externalities compared to current methods, such as antibiotic use and synthetic chemical pesticide applications for ACP control^12–15. While optimal therapy delivery systems for citrus plants remain an active topic of research and development, multiple practical options such as trunk injection already exist^{9,13,14,37,57–60}.

The impact of HLB has been devastating. Deployed alongside psyllid population suppression, the proposed “botanical” solutions harnessing natural plant defenses, when deployed at scale, may offer a viable path towards stable HLB mitigation and control. Finally, although we demonstrate the described approach for the citrus and CLas, the same approach of mapping the metabolic distributions coupled with toxicity assays may present opportunities to generalize the development of a therapy agent discovery pipeline that could be used for a wide range of pathosystems.

Materials and methods

Tissue collection and processing

Branch and single leaf assay

For the branch and single leaf assays, Valencia sweet orange (Citrus sinensis) grafted onto Swingle rootstock was propagated in a greenhouse and inoculated with CLas using infectious ACP. Five ACP adults enclosed in a nylon mesh drawstring bag were applied and confined to newly emerged leaves for 14 days. Trees were then treated with insecticides, imidacloprid, and carbaryl, to eliminate ACP, and the trees were kept free of ACP until the time of sampling. For branch analyses, all leaves for a given time point were detached via razor blades, flash-frozen, and held in liquid nitrogen during sampling. Petioles were removed and stored at − 80 °C for CLas quantitation via qPCR. For metabolome analysis, 1/8 in (3.175 mm) diameter punches were dispensed into the wells of 96 well plates containing 500 µL 50% ethanol. Tissues were lysed by repeat freeze–thaw cycles between − 80 °C and room temperature. The resulting extracts were filtered through 0.22 um filter plates via centrifugation before analysis via LC–MS. Leaf punches were collected in the same manner for the single-leaf assay.

Field collected tissues

Stems, roots, and leaves from 50 trees (n = 150) were collected from 5 different citrus Florida citrus groves in 2016⁹. In 2017, stems, roots, and leaves were collected from 80 trees located in 7 different orchards (n = 240). The stem collection included both xylem and phloem.

Each tree was divided into 4 quadrants (North, South, East, and West), and stems with attached leaves were collected from each of the quadrants and pooled. Topsoil from two sides of the tree and approximately 1.5 feet away from the base of the trunk near the irrigation line was removed, and the feeder roots near the irrigation line were sampled, shaken to remove soil, and sealed in a plastic bag. Gloves were changed, and clippers and shovels were sterilized with 30% household bleach between each tree that was sampled. All samples were immediately placed on ice for transit to the laboratory, where they were placed at 4˚C and processed within 24 h.

MS data acquisition

The tissue extracts were prepared in 100% ethanol, spiked with 1 µM sulfadimethoxine internal standard, and analyzed with UltiMate 3000 UPLC system (Thermo Scientific) using a Kinetex™ 1.7 µm C18 reversed-phase UHPLC column (50 X 2.1 mm) and Maxis Q-TOF mass spectrometer (Bruker Daltonics) equipped with ESI source. The column was equilibrated with 2% solvent B (98% acetonitrile, 0.1% formic acid in LC–MS grade water with solvent A as 0.1% formic acid in water), followed by a linear gradient from 2 B to 10% B in 0.2 min and then to 100% B at 12 min, held at 100% B for 2 min. Following each run, the column was equilibrated at 2% B for 1 min at a flow rate of 0.5 mL/min. MS spectra were acquired in positive ion mode in the range of 80–2000 m/z. A mixture of 10 µg/mL of each sulfamethazine, sulfamethizole, sulfachloropyridazine sulfadimethoxine, amitriptyline, and coumarin-314 was run at the beginning and the end of each batch (one 96-well plate). An external calibration with ESI-L Low Concentration Tuning Mix (Agilent Technologies) was performed prior to data collection, and internal calibrant Hexakis(1H,1H,3H-tertrafluoropropoxy)phosphazene was used throughout the runs. The capillary voltage of 4500 V, nebulizer gas pressure (nitrogen) of 1.4 bar, ion source temperature of 180 °C, and dry gas flow of 4 L/min, were used. For acquiring MS/MS fragmentation, the 7 most intense ions per MS¹ were selected. A stepping function was used to fragment ions at 50%, 100%, 150%, and 200% of the CID, with a timing of 25% for each step. Similarly, basic stepping of collision RF of 250 to 1500 Vpp with a timing of 25% for each step and transfer time stepping of 50, 75, 100, and 150 µs with a timing of 25% for each step was employed. MS/MS active exclusion parameter was set to 5 and released after 30 s. The mass of internal calibrant was excluded from the MS/MS list using a mass range of m/z 921.5–924.5. The data were deposited in the MassIVE online repository and are available under the IDs: MSV000082967 (2D leaf mapping); MSV000082962 (3D branch mapping); MSV000082963 and MSV000085416 (field study).

MS data analysis

The collected HPLC–MS raw data files were first converted from Bruker’s d to mzXML format and then processed with the open-source MZmine2 software [https://www.ncbi.nlm.nih.gov/pubmed/20650010?dopt=Abstract]. crop filtering with a retention time (RT) range of 0 to 14 min chromatograms. Mass detection was performed with a signal threshold of 1E3 and a 0.04-s minimum peak width. The mass tolerance was set to 20 ppm, and the maximum allowed retention time deviation was set to 5 s. For chromatographic deconvolution, the local minimum search algorithm with a 30% chromatographic threshold, minimum RT range of 0.6 s, minimum relative height of 1%, minimum absolute height of 5E2, the minimum ratio of peak top/edge 2, and peak duration range of 0.04—min was used. After isotope peak removal, the peak lists of all samples were aligned within the corresponding retention time and mass tolerances. Gap filling was performed on the aligned peak list using the peak finder module with 1% intensity, 10-ppm m/z tolerance, and 0.05-min RT tolerance, respectively. After the creation and export of a feature matrix containing the feature retention times, exact mass, and peak areas of the corresponding extracted ion chromatograms, the sample metadata was added to the feature matrix metadata of the samples.

All of the peaks that were present in any of the blanks with a signal-to-noise ratio (S/N) below 3:1 were removed from the final feature table.

Data pretreatment and statistical analysis

The data pretreatment and following statistical analysis were carried out with the MetaboAnalyst platform⁶¹. The feature tables generated with MZmine were filtered to remove features with near-constant, very small values and values with low repeatability using the interquartile range (IQR) estimate. A detailed description of the methodology is given in Ref.⁶². The samples were normalized using quantile normalization. The data were further scaled by mean centering and divided by standard deviation for each feature.

Principal component analysis (PCA) and partial least-squares discriminant analysis (PLS-DA)⁶³ were used to explore and visualize variance within data and differences among experimental categories. Random forests (RF)⁶⁴ supervised analysis was used to further verify the validity of determined discriminating features.

Molecular networking

The molecular network was created using the online workflow at GNPS platform (gnps.ucsd.edu)^25,26. The data were clustered with MS-Cluster with a parent mass tolerance of 0.1 Da and an MS/MS fragment ion tolerance of 0.1 Da to create consensus spectra. The consensus spectra that contained less than 3 spectra were discarded. A network was then created where edges were filtered to have a cosine score above 0.65 and more than 4 matched peaks. The edges between two nodes were kept in the network if and only if each of the nodes appeared in each other's respective top 10 most similar nodes. The spectra in the network were then searched against GNPS's spectral libraries. All matches kept between network spectra and library spectra were required to have a score above 0.7 and at least 5 matched peaks. The molecular networks and the parameters used are available at the links below: 2D leaf mapping: https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=105598d2d782412ca0c988bfe933c032, 3D branch mapping: https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=e2f1a1e367a7450fb940fe7cda04dc19, Field study: https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=2b0b1e554bc14d1fba321257b0ffc827, Field study, feature-based molecular network⁶⁵: https://gnps.ucsd.edu/ProteoSAFe/status.jsp?task=4ac8aff08af5433292143047c8cc5e90.

2D/3D visualization

The procedure for the creation and visualization of 3D models is described in detail previously²¹. Briefly, the 2D images and 3D model of the sampled plants were created and the coordinates for sampled spots were selected according to described protocol²¹. The abundances of detected metabolites were normalized and autoscaled⁶¹, and the coordinates for spots corresponding to each sample were inserted into the feature tables along with the spot size. The 2D or 3D models were drag-and-dropped into the ‘ili website in the browser (https://ili.embl.de/), followed by the feature table with coordinates. All figures were generated with the “Jet” color map. Spot size, opacity, and border opacity were adjusted for optimal visualization.

Synthesis of feruloyl putrescine hydrochloride

As the feruloyl putrescine was not available commercially, we have synthesized this compound for further compound identification validation and disc assay testing. N-Boc putrescine (1 equiv, 4.25 mmol, 0.91 mL) and ferulic acid (1.06 equiv, 4.5 mmol, 880 mg) were combined in anhydrous CH₂Cl₂ (32.5 mL) with stirring and the solution was cooled to 0 °C. Then, a solution of N,N′-dicyclohexylcarbodiimide (DCC, 1.7 equiv, 7.2 mmol, 1.49 g) in anhydrous CH₂Cl₂ was added dropwise. The reaction mixture was allowed to warm to room temperature and stirred at rt for 2 days. Then, the mixture was filtered to remove precipitated dicyclohexylurea and the filtrate was concentrated in vacuo. The crude residue was purified over silica gel using 2–4% MeOH in CH₂Cl₂ to afford N-Boc feruloyl putrescine as a light yellow solid in 78% yield (1.29 g). For Boc deprotection, trifluoroacetic acid (TFA, 15 mL) was added to a stirred solution of N-Boc feruloyl putrescine in CH₂Cl₂ (75 mL) under an inert atmosphere of Ar. The reaction was allowed to stir at room temperature for 40 min, then the solvent was removed in vacuo. The residue was dissolved in methanol (15 mL) with HCl (25 mL) and evaporated in vacuo to yield feruloyl putrescine hydrochloride as a light yellow solid (75% overall yield). ¹H and ¹³C NMR data were consistent with those previously reported⁶⁶. HRMS (ESI) exact mass calculated for [M + H]⁺ (C₁₄H₂₁N₂O₃) is m/z 265.1547.

Simulations

Simulations were performed using the CLas Ishi-1 M15 metabolic model¹⁵. Standard biomass constraints were maintained to predict the overall CLas growth rate. All model simulations were performed using the Gurobi Optimizer v.5.6.3 solver (Gurobi Optimization) in the COBRA toolbox⁶⁷ for MATLAB (MathWorks). We simulated the maximal growth rate of CLas using flux-balance analysis. The main metabolic compounds affecting CLas growth were identified using the metabolome data, that was anthranilate, ferulate, glutamate, N-acetyl-L-ornithine, ornithine, putrescine, tyrosine, and vanillin and sensitivity analysis, looking for metabolite-specific growth responses was performed by varying uptake rates among 1 × 10^–12, 1 × 10^–10, 1 × 10^–8, 1 × 10^–6, 1 × 10^–4, 1 × 10^–2, 1 × 10^–1, 1 × 10⁰, 1 × 10¹, 1 × 10², 1 × 10³. Additionally, we performed a sensitivity analysis, which deployed a phenotypic phase plane that facilitates the observation of effects on the CLas growth by varying a particular constraint, in this case putrescine and ferulic acid. Predicted growth rates were compared with experimental results.

Microbial culture assays

Citrus metabolites were assayed using a previously developed disc-diffusion assay for L. crescens³⁴. Briefly, L. crescens liquid cultures were incorporated to a soft agar (0.8%) overlay and applied to a 20 mL solid agar (1.5%). L. crescens strain BT-1] was maintained and grown exclusively on the previously described bBM7 + 1.0 methyl-β-cyclodextrin. To these overlaid plates were applied autoclave-sterilized 6 mm paper discs (Whatman, NJ, USA), previously loaded with 35 μL of a given metabolite solution and dried in a sterile biosafety cabinet. Once discs were applied, plates were sealed and stored upside down in a 28 ℃ incubator for 6 days to allow a clear zone of inhibition development for measurement. 35 mg/mL solutions of putrescine (Fisher Scientific, Waltham, MA, USA), and feruloylputrescine (synthesis described above) were prepared using sterile water.

The efficacy of bioflavonoids and ferulic acid in vitro CLas-hairy root assay

The in vitro CLas-hairy roots assay was performed according to the previously described protocol³⁵. CLas-citrus hairy roots were generated using CLas-infected citrus plant tissues, and the diagnosis of CLas was confirmed by quantitative PCR (qPCR). CLas-hairy roots were surface sterilized, and ~ 100 mg was transferred into multi-well plates containing Gamborg’s B-5 medium with 1% sucrose. Different concentrations of bioflavonoids (Horbaach, https://horbaach.com/products/citrus-bioflavonoids-complex-1500mg-300-vegetarian-caplets) and ferulic acid (Fisher Scientific, Pittsburgh, PA, Catalog No. ICN10168505): 125, 250, 500, and 1000 ppm/mL, were added, vacuum infiltrated and incubated on a rotator shaker at 50 rpm in the dark at 25 °C for 72 h. The experiments were carried out with six biological replicates, positive control of oxytetracycline hydrochloride (Sigma Aldrich, Burlington, MA), untreated CLas hairy roots, and an equal concentration of ethanol solvent used to dissolve the bioflavonoids and ferulic acid as negative controls. After the treatments, tissue samples were treated with PMAxx dye (propidium monoazide, Biotium, Fremont, CA) to inactivate dead CLas bacterial DNA. Further, total DNA was extracted, and viable bacterial titer was estimated by qPCR analysis using primers specific to the CLas gene encoding the Ribonucleotide reductase β-subunit (nrdB, RNR-F/RNR-R)⁶⁸ and the relative CLas titers were estimated and plotted relative to untreated using the 2 − ΔΔCt method. After normalization of target Ct with an endogenous reference gene (Ct') glyceraldehyde3-phosphate dehydrogenase 2 (GAPC2)⁶⁹ to correct for DNA template concentration differences among the samples, it was plotted relative to untreated controls.

Primers used in this study

RNR-F	CATGCTCCATGAAGCTACCC	68
RNR-R	GGAGCATTTAACCCCACGAA
GAPC2-F	GAGGAGATCCCATGGGCAAA	69
GAPC2-R	AAGAGGAGCTAGGCAGTTGG

Author contributions

CR, PR, PCD created the idea for the work. NG, AB, GM, PR collected plant tissues NG, AB extracted plant tissues for acquisition of mass spectrometry data AA acquired mass spectrometry data AA, AM analyzed mass spectrometry data GM conducted plants growth, infection by CLas, PCR diagnostics AM created 3D models AB, AM picked coordinates on plant models AA performed statistical analysis EG performed organic synthesis CZ, KZ performed and analyzed model simulations AB performed disk assays MR, KM conducted hairy roots assay AA wrote the manuscript CR, AB, PR, NG, AL edited the manuscript.

Data availability

The data were deposited in the MassIVE online repository and are available below links: https://massive.ucsd.edu/ProteoSAFe/dataset.jsp?task=bc1261c22e6c49d1b4f6490c7414845b, https://massive.ucsd.edu/ProteoSAFe/dataset.jsp?task=2008873282ac4beda2a008ada2917fa4, https://massive.ucsd.edu/ProteoSAFe/dataset.jsp?task=d4416036d13141b9b3008c7712395534, https://massive.ucsd.edu/ProteoSAFe/dataset.jsp?task=7db2bd0c5a0941f28286cabdc407a17f, https://massive.ucsd.edu/ProteoSAFe/dataset.jsp?task=25598a865760458fbc45a3f579c8479c.

Competing interests

AAA and AVM are founders of Arome Science, Inc. PCD is an advisor of and holds equity in Cybele, consulted for MSD Animal Health in 2023 and is a cofounder of, holds equity in and is scientific advisor for Ometa Labs, Arome and Enveda with prior approval by the University of California San Diego. The remaining authors declare no competing interests.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-70499-z.