Comparative metabolomics of resistant and susceptible pecan leaflets uncovers early defense responses to scab infection

Abstract

Background

Pecan scab (Venturia effusa) is a devastating fungal disease affecting commercial pecan production. Developing resistant cultivars remains the most sustainable strategy, but this has been hindered by limited mechanistic understanding of resistance and the presence of diverse pathogen pathotypes. Current resistance phenotyping methods rely on microscopic assessments, which are labor-intensive and prone to inconsistency due to sample preparation and interpretation variability.

Methods

A pathway-based metabolomics approach combined with machine learning was employed to identify early biomarkers of resistance in pecan. Metabolite profiles were obtained from the pecan cultivar 'Desirable' in response to the virulent scab isolate De-Tif-11 (susceptible reaction) and the avirulent scab isolate Pa-OK-11 (resistant reaction) from 0 to 7 days post inoculation (DPI). Logistic regression with L₁ regularization followed by linear regression identified potential marker compounds with high sensitivity and specificity for resistant and susceptible reactions.

Results

Three major defense mechanisms were identified that differentiate scab-resistant from susceptible reactions. In the resistant reaction, salicylic acid and 3-hydroxy-3-methylglutaryl-coenzyme A were rapidly upregulated at 1–2 DPI, prioritizing immune activation over energy metabolism. Once defense signaling was initiated, the resistant reaction shifted towards biochemical defense by accumulating flavonoids at 3–4 DPI, reinforcing pathogen restriction. The susceptible reaction exhibited delayed and ineffective defense responses throughout the infection period.

Conclusions

These results reveal early functional shifts in hormone signaling and secondary metabolism that differentiate resistant from susceptible reactions. The identified biomarkers may serve as robust indicators of resistance and facilitate accurate, early-stage screening in pecan breeding programs.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12870-026-09211-4.

Introduction

Scab, caused by Venturia effusa (Fusicladium effusum), is the most destructive disease affecting pecan trees (Carya illinoinensis) in the southern United States. The disease manifests dark lesions on leaves, fruit shucks, and twigs, resulting in reduced nut quality, premature nut drop, and significant yield losses. Leaves are highly susceptible during expansion, while fruits remain vulnerable throughout their development and maturation stages [1]. As the most economically valuable hickory species in the U.S. [2], pecans require effective scab management to ensure sustainable production.

Current management of pecan scab relies on the application of fungicides throughout the growing season. However, the repeated use of fungicides has led to resistance in scab pathogen populations against multiple chemical classes [3]. Furthermore, high rainfall and warm temperatures from June to August, typical of the southeastern U.S. growing season, further reduce the efficacy of fungicide applications by washing away treatments and promoting disease spread [4]. These challenges increase production costs and pose environmental risks due to intensive agrochemical use. Given these limitations, the most sustainable strategy for managing pecan scab is the development of scab-resistant cultivars. Such cultivars require minimal intervention from growers once established, reducing both economic and environmental burdens. Consequently, pecan breeding programs prioritize the identification and development of naturally scab-resistant cultivars to mitigate fungicide dependence and effectively manage the disease [2].

Scab resistance is complicated by the presence of multiple pathotypes of Venturia effusa, each capable of infecting a relatively narrow range of pecan genotypes [5, 6]. Sexual reproduction in the pathogen contributes to genetic diversity and variability in virulence [7]. As a result, resistance is isolate-specific and must be evaluated within defined host–pathogen combinations. Current resistance phenotyping relies on microscopic examination of fungal development after inoculation [6]. This approach requires extensive sample preparation, is time-intensive, and depends on evaluator interpretation. A rapid and reproducible method that captures early physiological responses associated with resistance would improve selection efficiency in breeding programs.

Plants respond to pathogen attacks by quickly producing specific metabolites that play key roles in defense signaling [8]. Genetic studies and resistance screening in breeding programs, often involving inoculations with defined scab isolates, have significantly advanced our understanding of resistance traits in pecan. Metabolites, biosynthesized within the plant host, may serve as a functional link between genetic regulation and the phenotypic expression of resistance, as they represent the end products of gene expression and cellular activity. Metabolomics has emerged as a powerful tool for elucidating plant defense mechanisms by profiling these biochemical endpoints [9, 10]. Resistance-associated metabolites not only reveal key physiological pathways involved in defense but also enable the identification of robust biomarkers that distinguish infection outcomes and resistance phenotypes.

Previous studies provide partial insight into the biochemical basis of scab resistance. An untargeted metabolomics study without pathogen inoculation identified antioxidant quercetin derivatives associated with resistance traits in mature pecan trees [11]. In apple scab caused by Venturia inaequalis, metabolomics has revealed that certain primary (amino acids, sugars, organic acids) and secondary (phenolic acids) metabolites are differentially regulated in resistant cultivars during infection [12]. Transcriptomic analyses have further expanded our understanding of pecan scab resistance. At early stage of infection, resistant responses were linked to the upregulation of genes involved in pathogen recognition, defense signaling, and transduction pathways, suggesting coordinated physiological responses that may be accompanied by shifts in metabolite profiles [13, 14]. Conversely, transcriptome analyses of mature trees in provenance collections showed that resistant trees downregulated defense-related genes, while diseased susceptible trees upregulated these genes compared to non-diseased susceptible counterparts [15]. These findings indicate that resistance responses are dynamic and may differ across developmental stages and infection outcomes. However, metabolic responses during the early stages of isolate-specific resistance reactions in pecan have not been characterized.

The objective of this study was to identify early-stage metabolic markers associated with resistance reactions in pecan during controlled scab inoculation. We hypothesized that resistant and susceptible reactions induced by distinct V. effusa isolates would produce separable metabolic profiles within 0 to 7 days post-inoculation. To test this hypothesis, the cultivar ‘Desirable’ was inoculated with two isolates differing in pathogenicity, one inducing a resistant reaction and one inducing a susceptible reaction. A pathway-based metabolomics combined with supervised machine learning was used to select metabolites that discriminate between reaction types. Selected metabolites were mapped to biological pathways to interpret functional defense processes. Finally, the collected data were integrated and compared to transcriptomics data obtained from the same plant–pathogen interaction system [13]. This biomarker-based framework aims to develop resistance-associated biomarkers to support breeding programs by providing a physiological complement to microscopic phenotyping methods, enabling rapid and reproducible assessment of resistance during the early stages of scab infection in pecan.

Materials and methods

Plant material

Pecan cultivar ‘Desirable’ was used for fungal inoculation and metabolomic analysis, due to its prevalence in commercial cultivation across the southeastern U.S. [16]. Test trees were produced by grafting seedling rootstocks sourced from a commercial nursery with scions obtained from ‘Desirable’ trees verified as true to type by University of Georgia pecan breeder Dr. Patrick J. Conner. Test trees were grown in individual 6.23-L tree pots (TP616, Stuewe & Sons, Tangent OR, USA) containing a growth medium composed primarily of pine bark and peat. ‘Desirable’ pecan trees yield high-quality nuts characterized by their substantial size, well-developed kernels, and a notably high percentage of intact kernel halves upon shelling.

Fungal inoculation and disease assay

Two isolates of V. effusa were selected for experimental inoculations. The first isolate, De-Tif-11, was originated from a single conidium extracted from a lesion produced on a ‘Desirable’ nut collected from a Tift Co., GA orchard. Previous experiments showed that this isolate sporulated well in culture and readily infected ‘Desirable’ leaves. The second isolate, Pa-OK-11, was cultured from a single conidium but obtained from a ‘Pawnee’ nut collected in Payne County, Oklahoma. ‘Desirable’ previously exhibited strong resistance to infection by Pa-OK-11 [6]. Both isolates were maintained and cultured to produce conidia following established methods [17].

The inoculation was conducted in April 2019, targeting the initial flush of foliage growth. Trees were prepared for inoculation by selecting the most susceptible leaflets, those ranging from 1/3 to 1/2 full expansion [17], and removing all other leaflets. Three biological replicates were prepared for each experimental condition: isolate De-Tif-11, isolate Pa-OK-11, and a water-sprayed control. Inoculation involved thoroughly applying a conidial suspension (3 × 10⁶ conidia/mL) of either De-Tif-11 or Pa-OK-11 isolates onto the selected leaflets until runoff occurred. Control leaflets were sprayed identically with sterilized distilled water. Following inoculation, trees were incubated in humidity rooms at 25–28 °C and 99% humidity (cooler with power off, overhead light on, and humidifier running) to maintain free moisture on leaf surfaces for 48 h. Subsequently, the trees were transferred to a greenhouse environment shaded at 50% with controlled temperatures set at 22 °C (actual range 19–27 °C), reflecting typical day-night temperature fluctuations during the pecan growing season in South Georgia. The tree size of samples precluded the use of controlled-environment growth chambers. Four leaflets from each tree were collected at 0-, 1-, 2-, 3-, 4-, 5-, and 7-days post inoculation (DPI). Collected leaflets were sealed in paper envelopes that were stapled shut, and frozen with liquid nitrogen. For the 0 DPI treatment, leaves were sprayed with conidia or water control, allowed to air dry, and then frozen. Additionally, samples for microscopic evaluations were collected at 5 DPI and prepared following previously reported staining techniques [17] to visualize and quantify the fungal infection and host resistance responses within the inoculated tissue, such as sub-cuticular hyphae (SCH) and halo (H).

Sample preparation

Leaf samples stored frozen were initially freeze-dried over 48 h, after which they were finely ground under liquid N₂. For reverse-phase (RP) chromatographic analysis, 20 mg of the powdered leaf material was combined with 0.5 mL of a chilled solvent mixture consisting of acetonitrile, isopropyl alcohol, water, and formic acid (3:3:1.9:0.1, v/v/v/v), which included internal standards (phenylalanine-¹³C₆, genistein-d₄, salicylic acid-d₄ for C30-1/genistein-d₄, hippuric acid-d₅, methionine-d₃, L-proline-2,5,5-d₃ for C30-2). For hydrophilic interaction chromatographic (HILIC) and sugar analysis, an equal sample weight (20 mg) was extracted using 0.5 mL of 20% methanol containing internal standards (N,N-dimethyl-d₆-glycine, methionine-d₃, hippuric acid-d₅, and L-proline-2,5,5-d₃). Metabolites extraction was performed by vortexing (10 min) followed by sonication in an ice-cold bath for another 10 min. Afterward, samples were centrifuged at 12,500 × g for 10 min at 4 °C. The pellets obtained were re-extracted once more with an additional 0.5 mL of the corresponding extraction solvent, following the same vortexing and sonication steps. The supernatant from both extractions was combined, and a portion (200 µL) was filtered through a 0.22-µm PVDF membrane filter (Cytiva, Marlborough, MA, USA) prior to instrumental analysis. Filtered extracts were analyzed via liquid chromatography–mass spectrometer (LC–MS) for global metabolite analysis and high-performance liquid chromatography with refractive index detector (HPLC–RID) for sugar analysis. Quality control (QC) samples were created by pooling equal volumes (5 µL) of extracts to ensure analytical reliability within each batch. These pooled QC samples were injected at systematic intervals throughout each analysis batch. Additionally, standard internal mixtures were analyzed at the start, middle, and end of each batch. All sample injections were randomized to minimize analytical batch effects.

Pathway-based metabolomics

Metabolomic analysis was performed using an Agilent 1260 Infinity II UHPLC coupled with a 6470 Triple Quadrupole (QqQ) mass spectrometer (Agilent Technologies, Santa Clara, CA, USA). Targeted metabolites were selected from KEGG pathways connected to plant–pathogen interactions (plant hormone signal transduction, phenylpropanoid/flavonoid/isoflavonoid biosynthesis, lignin biosynthesis, TCA cycle, carbon fixation, amino acid and purine metabolism), supplemented by compounds flagged in prior pecan/apple scab metabolomics studies [11, 12] (Supplementary Table S1). This pathway-based targeted approach was chosen because the biological hypothesis related to early defense metabolic shifts is grounded in well-characterized plant–pathogen interaction pathways, and targeted analysis with authentic standards provides higher confidence in metabolite identification and quantification than untargeted approaches. Chromatographic separation was conducted using RP and HILIC under three different conditions (C30-1, C30-2, and HILIC).

For RP chromatography under the C30-1 condition, an Acclaim C30 column (2.1 × 150 mm, 3 µm, Thermo Scientific, Waltham, MA, USA) was utilized to analyze flavonoids, amino acids, plant hormones, and organic acids. The mobile phase comprised water with 0.1% formic acid (A) and acetonitrile with 0.1% formic acid (B). Gradient conditions were: 2% B (0–2.5 min), 2–50% B (2.5–18 min), 50–90% B (18–23 min), and a re-equilibration period at 2% B for another 10 min. Flow rate and column temperature were maintained at 0.2 mL/min and 30 °C, respectively.

Under RP chromatography condition C30-2, phosphorylated and CoA-bound metabolites were separated using the same column (Acclaim C30 column, 2.1 × 150 mm, 3 µm, Thermo Scientific, Waltham, MA, USA), but with a modified solvent system of 10 mM ammonium acetate (pH 6.8, solvent A) and acetonitrile (solvent B) [18]. The mobile phase consisted of 10 mM ammonium acetate (pH 6.8) in water (A) and acetonitrile (B). The gradient was 0% B (0–3 min), 0–100% B (3–10 min), followed by washing with 100% B (10 min) and re-equilibration with 0% B (10 min). The flow rate was 0.2 mL/min with column temperature at 30 °C.

For the HILIC based separation, amino acids, organic acids, and sugars were analyzed using a Poroshell 120 HILIC-Z column (2.1 × 150 mm, 2.7 µm, Agilent Technologies, Santa Clara, CA, USA). The mobile phase consisted of 10 mM ammonium acetate (pH 9.0) in water (A) and in 90% acetonitrile (B) with 5 μM InfinityLab Deactivator Additive (Agilent Technologies, Santa Clara, CA, USA). The gradient elution was as follows: 90% B (0–2 min), 90–40% B (2–5 min), 40% B (5–13 min), followed by re-equilibration with 90% B (7 min). The flow rate was 0.25 mL/min, and column temperature was maintained at 30 °C.

The MS detection utilized electrospray ionization (ESI) in both positive and negative ionization modes. Dynamic multiple reaction monitoring (DMRM) was applied, with MS/MS parameters optimized via standard injections (Supplementary Table S1). ESI settings included: Gas temperature 275 °C (for C30-1) and 300 °C (for C30-2 and HILIC); gas flow 11 L/min; nebulizer pressure 45 psi; sheath gas temperature 325 °C (for C30-1) and 380 °C (for C30-2 and HILIC); sheath gas flow 11 L/min; capillary spray voltage + 3.0 kV (for positive mode) and −2.5 kV (for negative mode).

Sugars such as fructose, glucose, galactose, and trehalose were quantified using a Nexera 40 Series HPLC coupled with a RID-20A detector (Shimadzu Corp., Tokyo, Japan). Chromatographic separation employed a ZORBAX Original 70 Å Carbohydrate Analysis column (4.6 × 250 mm, 5 µm, Agilent Technologies, Santa Clara, CA, USA) with an isocratic mobile phase composed of water and acetonitrile (25:75, v/v) at 40 °C and a 1.4 mL/min flow rate. As the RID detector is less sensitive and more vulnerable to matrix effects, method validation was conducted in terms of linearity, limit of detection (LOD), limit of quantitation (LOQ), precision, and recovery to ensure analytical reliability and accuracy. Calibration curves were generated by plotting peak area versus concentration, and the coefficient of determination (R²) was determined to assess linearity. The LOD and LOQ were determined based on the standard deviation (σ) of the response and the slope (S) of the calibration curve, calculated as LOD = 3.3σ/S and LOQ = 10σ/S. Precision was assessed as both intra-day and inter-day relative standard deviations (RSD, %) by analyzing replicate samples (n = 6) on the same day and across two nonconsecutive days, respectively. Recovery rates were determined by spiking known concentrations of sugar standards into the sample matrix and calculating the percentage of each sugar recovered after sample preparation and analysis.

Data processing and machine learning/statistics approach for biomarker selection

Differences in SCH and halo formation among treatment groups were tested using Welch's two-sample t-tests (two-tailed, n = 3 biological replicates per group). Comparisons involving groups with no observed events were not statistically tested due to undefined variance.

The metabolomics data were processed using MassHunter Quantitative Analysis 10.1 software (Agilent Technologies, Santa Clara, CA, USA) for LC–MS analyses and LabSolutions Ver. 5.124 (Shimadzu Corp.) for sugar analysis via HPLC–RID. The relative abundance of metabolites (metabolic response) was calculated by comparing analyte peak areas against internal standards. As a targeted metabolomics approach was employed, each metabolite was detected using only one analytical mode. Analytical reliability was evaluated by analyzing quality control (QC) samples throughout the batches. Reliability criteria required QC sample relative standard deviations (RSD) to be ≤ 30%, therefore, only metabolite features with QC RSD ≤ 30% were included for further data interpretation and analysis [19, 20]. After QC filtering, datasets from each analytical platform were combined into a unified metabolite matrix.

Overall metabolic variation was evaluated using two complementary ordination analyses applied to individual biological replicates in R (https://www.r-project.org). Prior to ordination, metabolite abundances were log₁₀(x + 1)-transformed, mean-centered, and Pareto-scaled. To establish baseline metabolic variation before infection progressed, PCA was performed at 0 DPI (n = 9), and the treatment effect was tested by permutational multivariate analysis of variance (PERMANOVA) on the Euclidean distance matrix (999 permutations). For the post-inoculation window (1–7 DPI, n = 27), samples were grouped into three inoculation times (1 + 2, 3 + 4, and 5 + 7 DPI) matching those used in the biomarker analyses. The significance of treatment, inoculation times, and their interaction on metabolite variation was tested by PERMANOVA using a treatment × inoculation time model (999 permutations). To further visualize the treatment-specific metabolic signature independent of inoculation time, RDA was then applied with inoculation time included as a conditional variable to partial out inoculation time-related variance before the treatment effect was evaluated. The significance of the constrained treatment term was assessed by a 999-permutation test.

For downstream analyses, metabolite abundances were autoscaled using the scale() function in R. Hierarchical clustering analysis was performed using Euclidean distance as the distance metric and the Ward.D2 linkage method implemented in R as well. Figures were created using both RStudio (Posit, Boston, MA) and Biorender.

For statistical analysis and biomarker discovery, adjacent sampling time points were grouped into three intervals (1 + 2, 3 + 4, and 5 + 7 DPI). This grouping increased sample size within each interval and allowed evaluation of progressive stages of early host–pathogen interaction prior to visible symptom development. Because visible pecan scab symptoms such as lesions on pecan leaves generally appear approximately 8–16 days after infection [21], all sampling points represent pre-symptomatic responses. Biomarkers differentiating scab-resistant and scab-susceptible responses were identified using logistic regression with L1 regularization (LR-L1) in R. This method was selected for its ability to manage feature selection efficiently and control overfitting by penalizing redundant variables [22]. For each time group (1 + 2 DPI, 3 + 4 DPI, and 5 + 7 DPI), metabolic responses in resistant and susceptible treatments were compared with those of the control group, which served as the baseline condition for marker selection. Model performance was evaluated by 20 iterations of cross-validation with random 70:30 train:test splits (Monte Carlo cross-validation), with mean and standard deviation of classification metrics reported across iterations [23]. The prediction accuracy of the training set was then compared with the observed classifications of the test set, and the average performance across 20 iterations was calculated. Confusion matrix metrics included:

• True Positive (TP) – the number of resistant or susceptible reaction correctly classified.

• False Positive (FP) – the number of control samples misclassified as infected.

• True Negative (TN) – the number of control samples correctly classified.

• False Negative (FN) – the number of resistant or susceptible reaction misclassified as controls.

Accuracy, specificity, sensitivity, precision, F1-score, AUC (area under the curve), and the Matthews correlation coefficient (MCC) were calculated using the confusion matrix [23], further evaluating LR-L1 performance. A label-permutation test was performed for each model to test for overfitting in the LR-L1 classifier.

Biomarkers selected by LR-L1 were further validated using linear regression analysis in R, with metabolite abundance as the response variable and treatment group as the explanatory variable. Linear regression compared between resistant and susceptible groups and metabolites with p-value < 0.05 were considered as significant biomarkers. This step ensured that selected metabolites were associated with resistance or susceptibility responses rather than representing general infection markers.

To evaluate the magnitude and direction of metabolic changes relative to baseline, log₂(fold change) values were incorporated to prioritize key metabolites associated with resistance or susceptibility. Since the control group was used as a baseline in pecan breeding programs for evaluating scab reactions, fold changes were calculated by dividing metabolite responses by the control group values. A positive log₂(fold change) value indicates upregulation, whereas a negative value indicates downregulation relative to the control group. Because metabolites may increase in both infected groups but differ in magnitude, fold change values were used to prioritize biomarkers showing stronger directional shifts specific to either resistant or susceptible reactions. For metabolites identified as significant in both resistant and susceptible comparisons, the magnitude of the log2(fold change) values was used to assign group association.

Network correlation analysis between transcriptomics and metabolomics data

Transcriptomics data from our previously published study using the same plant-pathogen interaction system [13] were integrated with current metabolomic findings. Gene co-expression modules were first identified using WGCNA, and module eigengenes were calculated for each module. For the cross-omics correlation analysis, metabolomics samples collected at 1, 2, and 4 DPI were directly matched to the corresponding transcriptomics samples at the same DPI. Pearson correlations were then computed in R between individual-replicate metabolite levels and the matched gene-module eigengene values. Significantly correlated pairs (|r|> 0.50, p-value < 0.05) were used to construct the metabolic–transcriptional network. KEGG pathway enrichment of gene modules was conducted in R (clusterProfiler, enrichKEGG).

Results

Phenotypic profiling of scab-resistant and -susceptible reactions

The ‘Desirable’ leaf samples were inoculated with two distinct monoconidial scab isolates: one collected from the infected nuts of ‘Desirable’ (De-Tif-11) and another from ‘Pawnee’ nuts (Pa-OK-11), along with a water-inoculated control. Field inoculations have shown that isolate De-Tif-11 sporulates readily on ‘Desirable’ and that ‘Desirable’ is highly resistant to isolate Pa-OK-11 in field inoculations [6, 13]. Successful fungal infection is characterized by conidial germination, germ tube formation, penetration into the subcuticular space via an appressorium, and the subsequent development of sub-cuticular hyphae (SCH), which represents a susceptible reaction [5]. The formation of halo structures (H) surrounding penetration sites, indicating the host’s ability to halt fungal spread, was considered a resistant reaction [13]. Based on these phenotypic responses, scab resistance and susceptibility were evaluated by using a microscope (400X magnification) (Fig. 1A and B). At 5 DPI, SCH formation was observed in 54.1% of germinated conidia in samples inoculated with De-Tif-11, whereas no SCH was detected in either the samples inoculated with Pa-OK-11 or the control group (Fig. 1C). Halo structures were predominantly observed in samples inoculated with Pa-OK-11, accounting for 87.1% of the germinated conidia in these samples. In contrast, relatively weak resistance response was observed in samples inoculated with De-Tif-11, with 32.8% of the germinated conidia having halo structures (Fig. 1C). These results indicated that ‘Desirable’ cultivar exhibits strong resistance to Pa-OK-11 (a scab isolate from a different cultivar) while showing susceptibility to De-Tif-11 (a scab isolate from the same cultivar).

Fig. 1.

Phenotypic and metabolic differences between scab-resistant and scab-susceptible pecan leaflets. Microscopic images of (A) susceptible and (B) resistant reactions in pecan leaf tissue against scab fungus (V. effusa) infection. SCH: sub-cuticular hyphae (susceptible reaction), C: conidia, H: halo (resistance reaction). C Percent sub-cuticular hyphae and halo formation at 5 DPI in ‘Desirable’ pecan trees treated with water (control), De-Tif-11 (scab isolate from ‘Desirable’), and Pa-OK-11 (scab isolate from ‘Pawnee’). Bars in Panel C represent mean ± standard deviation (n = 3 biological replicates per treatment); ** indicates p < 0.01 by Welch's two-sample t-test

Metabolic variation between groups

The pathway-based metabolomics approach focuses on primary metabolic pathways including the TCA cycle, carbon fixation, amino acid, sugar and purine metabolisms, and secondary metabolic pathways such as flavonoid, plant hormones, phenylpropanoid and lignin biosynthesis, which have roles in the pathogen-plant interactions. This approach identified a total of 155 metabolites in ‘Desirable’ treated with water (control) or scab isolates over a 0 to 7 DPI period. These metabolites included 12 sugars, 34 organic acids, 17 amino acids, 9 plant hormones, 43 flavonoids, and 40 metabolites from other classes (Supplementary Table S1). Metabolic responses were collected using the peak area ratio of each analyte to the internal standard. Pooled quality control (QC) samples, injected every 8–10 samples, confirmed analytical reliability across batches, with relative standard deviations (RSD) of less than 30% (Supplementary Table S2). For sugars analyzed by the RID detector with QC results below 30% RSD, the method was further validated for linearity, LOD, LOQ, precision, and recovery. The validation results are presented in the Supplementary Information (Appendix I and Supplementary Table S3).

Metabolic variation in pecan leaflets across treatments and inoculation times was analyzed using PERMANOVA, PCA and RDA (Fig. 2 and Supplementary Figure S1). At 0 DPI (n = 9), the first two principal components explained 63.6% of the total variance (PC1: 45.9%; PC2: 17.7%, Fig. 2A), and PERMANOVA on the Euclidean distance matrix detected no significant treatment effect (p = 0.180; Fig. 2A), confirming baseline equivalence among treatment groups. For the post-inoculation samples (1–7 DPI, n = 27), PERMANOVA detected significant effects of treatment (p = 0.03) and inoculation time (p = 0.001), with no significant treatment × inoculation time interaction (p = 0.274). In the constrained RDA with inoculation time included as a conditional variable, the treatment effect remained significant (F = 2.38, p = 0.007). Because treatment is a three-level factor (control, resistant, susceptible), the constrained RDA comprises two axes that together capture all of the treatment-attributable variation by construction. Treatment centroids projected within the constrained RDA space showed close positioning of the resistant and susceptible centroids at 1–2 DPI (Fig. 2B), with separation along RDA1 emerging by 3–4 DPI (Fig. 2C) and maintained through 5–7 DPI (Fig. 2D). At 3–4 and 5–7 DPI, the Euclidean distance between the resistant centroid and the control centroid was smaller than that between the susceptible centroid and the control centroid.

Fig. 2.

Baseline PCA and treatment-conditioned RDA of pecan leaflet metabolomes. (A) PCA of 0 DPI samples was used to assess baseline metabolic variation among water control, De-Tif-11-inoculated, and Pa-OK-11-inoculated leaflets. Post-inoculation samples from (B) 1 + 2 DPI, (C) 3 + 4 DPI, and (D) 5 + 7 DPI were analyzed by RDA using treatment as the constrained variable while conditioning on sampling stage. The three RDA panels use the same coordinate space and show stage-specific treatment centroids calculated from three biological replicates. Grey points represent all post-inoculation samples, and colored points indicate samples from the highlighted inoculation time (DPI). The treatment effect remained significant after accounting for stage variation (permutation test, 999 permutations; F = 2.38, p = 0.007). Colors indicate treatment and shapes indicate sampling stage

Discovery of biomarkers for scab-resistance and susceptibility

Logistic regression with L1 regularization (LR-L1) was applied for feature selection and classification to identify biomarkers associated with scab resistance and susceptibility. The mean classification accuracy of LR-L1 for the resistant (Pa-OK-11 inoculated) and susceptible (De-Tif-11 inoculated) groups versus the control ranged from 76.25% to 93.75% and 75.62% to 87.50%, respectively (Table 1), confirming that LR-L1 can effectively distinguish these groups from the control. In a label-permutation test (100 permutations), real-data AUCs exceeded the shuffled-label null distributions in all twelve model configurations (p ≤ 0.020; Supplementary Fig. S2).

Table 1.

The average of confusion matrices and performance indicators from 20-fold cross-validation for filtering metabolites by logistic regression with L₁ regularization

	Mode	DPI	Confusion matrixa				Accuracy (%)	Sensitivity (%)	Specificity (%)	Precision (%)	F1 Score (%)	Matthews correlation coefficient (%)	AUC
			TP	FP	FN	TN
Control vs Susceptible	C30	1 + 2	2.75	0.05	1.25	3.95	83.75 ± 15.76	69.67 ± 28.74	99.00 ± 4.47	97.50 ± 11.18	78.00 ± 23.47	72.01 ± 26.97	0.95 ± 0.11
		3 + 4	3.10	0.45	0.80	3.65	84.38 ± 18.08	80.42 ± 27.21	89.00 ± 21.20	90.24 ± 17.97	81.21 ± 21.48	72.45 ± 30.19	0.97 ± 0.09
		5 + 7	3.10	0.05	0.95	3.90	87.50 ± 14.62	78.92 ± 26.42	99.17 ± 3.73	98.33 ± 7.45	84.39 ± 20.08	79.20 ± 22.72	0.96 ± 0.07
	HILIC	1 + 2	2.50	0.25	1.10	4.15	83.12 ± 15.85	69.00 ± 29.15	95.17 ± 10.79	91.25 ± 18.63	74.85 ± 23.61	68.02 ± 29.25	0.95 ± 0.13
		3 + 4	3.35	0.35	0.85	3.45	85.00 ± 11.89	80.17 ± 21.99	90.00 ± 22.56	93.81 ± 12.75	83.63 ± 13.16	73.22 ± 20.22	0.85 ± 0.18
		5 + 7	2.85	0.60	1.35	3.20	75.62 ± 14.89	70.90 ± 23.30	87.89 ± 23.46	88.61 ± 22.32	73.01 ± 18.28	59.33 ± 20.65	1.00 ± 0.00
Control vs Resistant	C30	1 + 2	2.95	0.05	1.25	3.75	83.75 ± 14.11	72.50 ± 24.92	99.17 ± 3.73	98.33 ± 7.45	80.59 ± 17.61	72.88 ± 21.58	0.91 ± 0.15
		3 + 4	3.10	0.40	1.05	3.45	81.88 ± 18.79	77.73 ± 25.37	90.83 ± 19.10	90.50 ± 20.33	80.82 ± 19.73	68.96 ± 33.59	0.97 ± 0.09
		5 + 7	3.25	0.25	0.55	3.95	90.00 ± 11.18	85.67 ± 22.51	94.25 ± 14.62	95.33 ± 11.87	87.68 ± 15.48	82.19 ± 19.65	0.97 ± 0.08
	HILIC	1 + 2	2.75	0.40	1.15	3.70	80.62 ± 17.43	73.50 ± 26.68	93.00 ± 17.20	90.67 ± 22.83	76.82 ± 21.02	67.71 ± 29.40	0.92 ± 0.17
		3 + 4	2.50	0.35	1.55	3.60	76.25 ± 18.98	66.08 ± 30.27	92.33 ± 16.76	91.25 ± 18.63	71.58 ± 22.79	60.58 ± 32.32	0.91 ± 0.23
		5 + 7	3.65	0.00	0.50	3.85	93.75 ± 12.50	87.75 ± 21.93	100.00 ± 0.00	100.00 ± 0.00	91.73 ± 15.83	89.31 ± 19.32	0.98 ± 0.05

^aTP True positive, FP False positive, FN False negative, TN True negative. Values represent averages across 20 cross-validation iterations

Among the candidate markers identified by LR-L1, statistically significant markers were selected via linear regression analysis by filtering insignificant metabolites between susceptible and resistant groups based on p-values (Supplementary Table S4). Figure 3 displays biomarkers identified through LR-L1 followed by linear regression, with p-value < 0.05, marked in red for resistant and blue for susceptible groups. The x-axis (LR-L1 coefficient) represents the importance of each variable in explaining resistance or susceptibility, while the y-axis (log₂ fold change) indicates the magnitude of change compared to the control group. Biomarkers identified in both groups were further evaluated by comparing their log₂ fold changes between resistant/control and susceptible/control comparisons to determine their specificity for each group (Supplementary Table S4). Group-specific biomarkers, represented as either ‘significant resistance biomarkers’ or ‘significant susceptibility biomarkers,’ are shown as filled diamond shapes in Fig. 3A and B.

Fig. 3.

Machine learning-based selection of biomarkers distinguishing scab-resistant (Pa-OK-11 inoculated) and scab-susceptible groups (De-Tif-11 inoculated). Candidate markers for resistant (A) and susceptible (B) reactions were identified by logistic regression with L1 regularization (LR-L1). Significant markers confirmed by linear regression (p-value < 0.05) are highlighted in red (resistance biomarkers) and blue (susceptibility biomarkers). Biomarkers identified as specific to resistant or susceptible groups based on comparative log₂ (resistant/control) and log₂ (susceptible/control) are represented as filled diamonds (♦). The x-axis (LR-L1 coefficient) represents the contribution of each biomarker to resistance or susceptibility, while the y-axis (log₂ fold change) indicates its magnitude of change relative to the control. Panels correspond to different time points: 1 + 2 DPI (left), 3 + 4 DPI (middle), and 5 + 7 DPI (right). C The number of significant resistance and susceptibility biomarkers identified by machine learning

The biomarkers identified in this study reflect metabolic responses in pecan leaf tissues during early infection stages (0–7 DPI) under controlled inoculation conditions. At 1–2 DPI (left panels of Fig. 3A and B and Supplementary Table S4), six out of twenty compounds were selected as significant biomarkers related to resistant reactions (resistance biomarkers), including salicylic acid, oxalate isomer, adenosine monophosphate (AMP), apigenin, genistein, and chrysin (Fig. 3C). For the susceptible group, five out of sixteen compounds were selected as significant biomarkers associated with susceptible reactions (susceptibility biomarkers), including alanine, chlorogenate, hypoxanthine, methionine and sedoheptulose-7P. At 3–4 DPI (middle panels of Fig. 3A and B and Supplementary Table S4), the resistant group exhibited eleven resistance biomarkers out of twenty-three: glyceraldehyde-3P, glucose-6P isomer, glutamate, phosphoenolpyruvate (PEP), apigenin, astragalin, biochanin A, genistein, luteolin, sissotrin, and tricin isomer. In contrast, the susceptible group had only six susceptibility biomarkers out of eighteen: alanine, fructose, α-ketoglutarate, trehalose, 3-hydroxy-3-methylglutaryl-coenzyme A (HMG-CoA), and salicylic acid (Fig. 3C). At 5–7 DPI (right panels of 3 A and 3B and Supplementary Table S4), the susceptible group displayed a greater number of susceptibility biomarkers compared to the resistant group. These biomarkers included ascorbate isomer, salicylic acid, trehalose, HMG-CoA, 2,5-dihydroxybenzoic acid (2,5-DHBA), malate, ascorbate, fructose, α-ketoglutarate, ornithine, catechin, gallocatechin, prunin, 4-caffeoyl shikimic acid, and 5-aminopentanoate. At this stage, only three metabolites were significantly associated with the resistant group: sinapic acid, chrysin, and sissotrin (Fig. 3C). Notably, susceptibility biomarkers exhibited larger log₂(fold change) values relative to the control group than resistance biomarkers, as reflected by the broader range of log₂(fold change) values on the y-axis. These metabolites were designated as susceptibility biomarkers because they showed significantly greater abundance in susceptible samples (De-Tif-11 inoculated samples) than in resistant samples (Pa-OK-11 inoculated samples) and exhibited larger log₂(fold change) values relative to the control group. Overall, the resistant reaction rapidly exhibited distinct metabolic responses within 3–4 DPI, while susceptibility-derived changes were delayed and mostly emerged at 5–7 DPI.

Indication of selected biomarkers and associated metabolic mechanisms across different stages of infection

Thirty-five metabolites were identified as significant biomarkers for scab resistance or susceptibility across different stages of infection. To identify the trends in marker behavior between groups, hierarchical clustering of scaled biomarker content was performed (Fig. 4A). The analysis identified two major clusters based on marker responses, with their respective patterns visualized in heatmaps for the resistant and susceptible groups (Fig. 4B and C, respectively). The first cluster (colored in orange, Fig. 4A) included malate, α-ketoglutarate, salicylic acid, trehalose, catechin, gallocatechin, and 4-caffeoyl shikimic acid, which progressively increased in the susceptible group over time (Fig. 4C). The second cluster (colored in green, Fig. 4A) comprised metabolites that peaked at either 1 + 2 DPI or 3 + 4 DPI in resistant/susceptible groups, exhibiting distinct metabolic patterns (Fig. 4B).

Fig. 4.

Patterns of significant biomarkers in resistant (Pa-OK-11 inoculated) and susceptible reactions (De-Tif-11 inoculated) in pecan leaflets during the inoculation period. Hierarchical clustering of the biomarkers based on scaled content (A). Heatmaps showing dynamic changes in biomarker levels across different DPIs in the resistant (B) and susceptible (C) groups. Colors in heatmap represent scaled content, with red indicating higher and blue indicating lower levels

Metabolite mapping on the biological pathways revealed metabolic variation in the metabolic network (Figs. 5 and 6). In resistant pecan trees, flavonoid biosynthesis was primarily upregulated at 3–4 DPI, with an increase in seven markers on this pathway, including astragalin, sissotrin, biochanin A, genistein, apigenin, luteolin, and tricin isomer (Fig. 5). On the other hand, susceptible reaction exhibited the late accumulation of some compounds at 5–7 DPI, such as 4-caffeoyl shikimic acid, gallocatechin, and catechin. These metabolites accumulated later during infection compared with those observed in resistant samples. The mapping also demonstrated that the salicylic acid pathway, involved in plant immune responses, was distinctively regulated between the groups at different DPIs. Salicylic acid was upregulated in the resistant group as a biomarker, showing an early response after infection (1–2 DPI), whereas in the susceptible group, its synthesis and metabolism (involving salicylic acid and 2,5-DHBA) were activated later, after 3–4 DPI, reflecting a delayed immune response (Fig. 5).

Fig. 5.

Changes in secondary metabolites in pecan leaves in response to different scab isolates during infection. A Metabolic responses of secondary metabolite markers in control, resistant (Pa-OK-11 inoculated), and susceptible (De-Tif-11 inoculated) groups across different DPIs. B Mapping of related metabolic pathway with the markers associated with scab-resistant and scab-susceptible reactions at different DPIs

Fig. 6.

Changes in primary metabolites in pecan leaves in response to different scab isolates during infection. A Metabolic responses of primary metabolite markers in control, resistant (Pa-OK-11 inoculated), and susceptible (De-Tif-11 inoculated) groups across different DPIs. B Mapping of related metabolic pathway with the markers associated with scab-resistant and scab-susceptible reactions at different DPIs

Mapping on primary metabolism revealed that pathways such as the tricarboxylic acid (TCA) cycle, amino acid metabolism, sugar metabolism, purine metabolism, and carbon fixation were dominantly affected by susceptibility biomarkers, including alanine, methionine, hypoxanthine, ornithine, malate, α-ketoglutarate, trehalose, fructose, sedoheptulose-7P, and ascorbate (Fig. 6). Only glycolysis/gluconeogenesis intermediates were elevated in the resistant group at 3–4 DPI. Interestingly, HMG-CoA, a precursor of the mevalonate pathway, was initially upregulated in the resistant group at 1–2 DPI, whereas, as the infection progressed, a notable increase in this metabolite was observed in the susceptible group at later stages (3–7 DPI). Overall, the susceptible group exhibited broadly elevated levels of primary-metabolism intermediates across all infection stages, while the resistant group showed elevated intermediates only at 3–4 DPI.

Linkage between metabolomics and transcriptomics data underlying scab resistance and susceptibility

In a previous study using the same plant–pathogen interaction system, 11 gene modules were identified through transcriptomics combined with weighted gene co-expression network analysis (WGCNA) [13]. To confirm the mechanisms underlying biomarkers at different biological levels, metabolomics data (this study) were correlated with transcriptomics data (gene modules; the previous study) using Pearson correlation analysis with hierarchical clustering. For the early marker mechanisms, data from within 4 DPI were used for analysis. The transcriptomics and metabolomics results were aligned with each other, as presented in a clustered correlation heatmap in Supplementary Fig. S3. Network analysis based on the correlations is shown in Fig. 7, with the biomarkers and related gene modules (denoted as M1: turquoise color; M2: blue color; M3: brown color; and M4: yellow color). Previously, M3 and M4 modules were identified as key gene modules associated with host defense responses [13]. In addition to these modules, strong correlations of two additional modules (M1 and M2) and biomarkers, with |r|> 0.50 and p-value < 0.05, were observed (Supplementary Fig. S3). Pathway enrichment analysis was subsequently conducted for these modules, as illustrated in Supplementary Fig. S4.

Fig. 7.

Network analysis based on metabolomics and transcriptomics data. The correlation network at 1 + 2 DPI (A) and 4 DPI (B) shows the relationships between metabolites and gene modules. Circle nodes represent metabolites linked to scab resistance (red) and susceptibility (blue), while square nodes indicate gene modules (turquoise: M1; blue: M2; brown: M3; and yellow: M4) clustered by WGCNA [13]. Metabolites with asterisk (*) indicate the significant biomarkers selected by machine learning with log₂(fold change). Red and blue edges denote positive and negative correlations between the nodes, respectively (|r|> 0.5, p-value < 0.05)

At the 1–2 DPI (Fig. 7A), HMG-CoA, a metabolite initially upregulated in the resistant group, exhibited a positive correlation with the M4 module, which is associated with defense-related genes including alkaloid biosynthesis (Supplementary Fig. S4). Similarly, salicylic acid, a resistance biomarker at 1–2 DPI, showed positive correlations with the M3 and M4 modules, both of which are enriched in plant-pathogen interaction pathways (Supplementary Fig. S4). Salicylic acid also exhibited a negative correlation with the M2 module, which is associated with energy metabolism. In contrast, the M2 module was positively correlated with hypoxanthine and sedoheptulose-7P, both of which were identified as susceptibility biomarkers. Serine and succinate, differentially expressed metabolites in the susceptible group during the early stage, correlated positively with the M3 module, which may reflect MAPK signaling activation in response to infection [24] (Supplementary Fig. S4).

At the 4 DPI (Fig. 7B), HMG-CoA and salicylic acid remained positively correlated with the M3 and M4 modules but were re-defined as susceptibility biomarkers due to alterations in their metabolic behavior between groups. Additionally, 2,5-DHBA, the susceptibility biomarker and oxidized form of salicylic acid, exhibited positive correlations with the M3 and M4 modules at this stage, supporting a shift of salicylic acid metabolism from the resistant to susceptible groups in the later stage of infection. Five flavonoids, including astragalin, genistein, tricin isomer, apigenin, and sissotrin, which are resistance biomarkers at this stage, showed positive correlations with the M1 module. The M1 module was enriched in flavonoid biosynthesis pathways (Supplementary Fig. S4).

Discussion

Metabolic shifts in scab-resistant and susceptible pecan trees during infection

Metabolomic analysis revealed distinct metabolite variations between scab-resistant and -susceptible pecan reactions, as well as across inoculation times. The absence of significant treatment-related variation at 0 DPI (Fig. 2A) confirmed that the Pa-OK-11–inoculated, De-Tif-11–inoculated, and water-control groups were metabolically equivalent before infection progressed, indicating that subsequent metabolic divergence was attributable to the host–pathogen interaction rather than to pre-existing differences among trees. For the post-inoculation window, both treatment and inoculation time independently contributed to metabolic variation, while their non-significant interaction indicated that the treatment effect was not driven by a single anomalous timepoint. The treatment effect remained significant in the constrained RDA after partialling out time-related variance, demonstrating a treatment-specific metabolic signature independent of infection progression. The centroid trajectory in the constrained RDA space revealed progressive metabolic divergence between the resistant and susceptible reactions (Fig. 2B–D). At the early stage of infection (1–2 DPI), resistant and susceptible group remained distinct from control group, showing metabolic activation against scab invasion (Fig. 2B). This suggests that metabolic shifts occurred in resistant and susceptible pecan leaflets through certain response patterns (discussed later). As infection progressed, the two groups diverged in opposite directions along RDA1 by 3–4 DPI and remained separated through 5–7 DPI, with the resistant group located closer to the control group (Fig. 2C and D). This could imply that resistant leaflets maintain a metabolic profile more similar to healthy leaflets, likely due to a well-controlled response to infection and minimized damage. At this stage, the susceptible group was clearly distinguished from the control group, potentially reflecting delayed defense responses to pathogen attack and the resulting pathogenic changes. Phenotypic changes at 5 DPI between groups, detected using a microscopic method, support these assumptions (Fig. 1C).

Metabolomics combined with machine learning enables earlier identification of scab resistance and susceptibility, complementing microscopic phenotyping

Conventional light microscopy-based screening for pecan scab resistance has some limitations. The staining process involves multiple steps and nearly a week post-inoculation to visualize infection regions [5, 17]. Compared to the microscopic method, metabolomics combined with machine learning strategy enabled earlier detection of scab resistance and susceptibility at the initial stages of infection (1–4 DPI) based on biomarkers (Fig. 3 and Supplementary Table S4). This detection occurred prior to the appearance of visible phenotypic symptoms, highlighting the potential of this method as a rapid and objective tool for early resistance assessment. At these early stages, trained machine learning models achieved 76.25%–83.75% accuracy in classifying resistant reactions and 83.12%–85.00% accuracy for susceptible reactions (Table 1). These early biomarkers could be integrated into pecan breeding programs to facilitate genotype selection while reducing the time and labor required for traditional microscopy.

Metabolic mechanisms underlying the biomarkers of scab resistance and susceptibility

Scab-resistant pecan leaflets exhibit early defense activation via HMG-CoA and salicylic acid

Upon pathogen attack, often reactive oxygen species (ROS) are generated in the plant host, triggering salicylic acid signaling for primary plant defense and systemic resistance [25–27]. In this study, the scab-resistant group showed upregulation of salicylic acid levels at 1–2 DPI, while the scab-susceptible group did not show such an increase in this early stage (Fig. 5). This suggests a rapid and effective activation of defense mechanisms in scab-resistant samples. A similar result was reported in apple scab, where resistant apple cell cultures upregulated salicylic acid earlier than susceptible cell cultures [12]. Interestingly, as the infection progressed, susceptible reaction exhibited a delayed but dramatic increase in salicylic acid, reaching levels 7.5 times higher than those in resistant reaction by 5–7 DPI (Fig. 5). 2,5-DHBA, the oxidized product of salicylic acid, had also accumulated in susceptible reaction in this later stage (Fig. 5). This late surge in the susceptible group indicates a host response to reactive stress and pathogenic symptoms, rather than a proactive restriction of the pathogen [26, 28]. Our findings coincide with a previous study on citrus disease, which observed a delayed but accumulated salicylic acid response in disease-susceptible citrus [29]. Salicylic acid thus appears to serve as both a scab-resistant marker in the early stage (1–2 DPI) and a scab-susceptibility marker in the later stage (≥ 3 DPI).

HMG-CoA showed a temporal pattern broadly comparable to that of salicylic acid, although the statistical signal differed between stages. HMG-CoA was upregulated at 1–2 DPI in resistant reaction and rapidly decreased in the later stages (Fig. 6). This may be explained by salicylic acid-induced expression of HMG-CoA synthase (HMGS) [30]. In mustard greens (Brassica juncea), HMGS is overexpressed in response to fungal infection via the activation of salicylic acid-dependent pathogenesis-related genes [31]. The level of HMG-CoA gradually decreased in resistant plant hosts as the infection progressed, likely due to the role of HMG-CoA reductase (HMGR), which converts HMG-CoA into mevalonate for further immune responses [32, 33]. The temporal pattern of HMG-CoA paralleled that of salicylic acid, although with a weaker early-stage signal, and might similarly reflect differences in the timing of metabolic responses between the resistant and susceptible reactions [31].

Scab-resistant pecan leaflets restrict fungal spread through flavonoid and lignin biosynthesis, while susceptible leaflets exhibit delayed defense activation

At early and intermediate stages (1–4 DPI), scab-resistant leaflets activated the biosynthesis of flavonoids and their derivatives, such as apigenin, genistein, chrysin, astragalin, biochanin A, luteolin, sissotrin, and tricin isomer, while their levels remained relatively stable in susceptible leaflets (Fig. 5). Flavonoids possess antioxidant and antifungal properties associated with plant defense mechanisms [34]. Supporting our findings, a study observed similar results, demonstrating that flavonoids contribute to scab resistance in mature pecan trees [11]. The upregulation of flavonoids, including biochanin A, has also been linked to enhanced resistance to the fungus Alternaria alternata in olive leaves [35]. Susceptible leaflets showed delayed flavonoid production at 5–7 DPI, and the types of flavonoids were different from those in resistant leaflets (Fig. 5). Increased catechin levels were observed in damaged apple leaves infected by scab [36, 37]. These late-stage flavonoids in susceptible leaflets are likely associated with late defense responses to damaged tissues rather than proactive restriction of the scab pathogen. In addition to flavonoids, elevated ascorbate levels were detected in susceptible reactions at 5–7 DPI. This increase likely reflects oxidative stress resulting from progressive pathogen invasion, consistent with the role of ascorbic acid in stress mitigation [38].

Following the early activation of flavonoid synthesis, lignin signaling was overall upregulated in resistant leaflets (Fig. 5). Sinapic acid, a marker for resistance, is a molecule triggering active lignification of tissue. Lignin serves as a physical barrier to restrict pathogen invasion and elicits antimicrobial activity for plant defense [39]. Lignin biosynthesis was significantly upregulated in cotton seedlings resistant to the fungus Verticillium dahliae [40]. A previous study with gene ontology (GO) enrichment analysis also demonstrated the upregulation of lignin biosynthesis in leaflets showing resistant reactions [13], supporting the role of lignification in pecan scab resistance.

Scab-resistant pecan leaflets exhibit altered energy metabolism intermediates during infection

Primary metabolism provides the cellular energy required for regular growth and maintenance, as well as for plant defense responses during pathogen invasion [41]. However, to effectively manage extreme stress conditions, plants often regulate primary metabolic pathways, such as photosynthesis, to reallocate resources toward more active defense activities [42, 43]. In line with this strategy, at the initial stage (1–2 DPI), scab-resistant leaflets exhibited relatively lower levels of Calvin cycle intermediates, glyceraldehyde-3P and sedoheptulose-7P, than susceptible leaflets (Fig. 6). At the later stage (5–7 DPI), susceptible leaflets showed increased levels of sugars (e.g., trehalose and fructose) and TCA cycle metabolites (e.g., malate) (Fig. 6). The accumulation of these metabolites may indicate tissue damage-related metabolic shifts caused by fungal growth [43, 44]. Transcriptomic analysis reported upregulation of trehalose biosynthesis genes in diseased leaflets of scab-susceptible pecan trees [15], reinforcing the link between scab susceptibility and altered sugar metabolism.

The results above suggest that resistant leaflets prioritize immune responses with altered energy metabolism in the early stages, while susceptible leaflets exhibit delayed defense responses, with increased metabolic activity occurring after damage at later stages.

Confirmation of scab-resistant and scab-susceptible mechanisms at different biological levels: Integration of transcriptome and metabolome data

Metabolomic-transcriptomic integration confirmed key defense pathways and their linkage with markers that are differentially regulated between scab-resistant and -susceptible leaflets (Fig. 7). The early activation of salicylic acid and HMG-CoA in resistant reaction at 1–2 DPI showed a strong positive correlation with defense-related genes (M3 and M4 modules) (Fig. 7A). Salicylic acid at this stage was also negatively linked to energy metabolism-related genes (M2 module), implying a strategic shift in the resistant group toward immune activation over energy processes [43]. At 4 DPI, salicylic acid and HMG-CoA were again positively correlated with the same gene modules (M3 and M4), but as susceptibility biomarkers (Fig. 7B), suggesting delayed responses as a characteristic of susceptible response. Meanwhile, the susceptibility markers, including sedoheptulose-7P and hypoxanthine, were positively correlated with M2, whose member genes are involved in photosynthesis, energy metabolism and cell growth (Fig. 7A). Sedoheptulose-7P is an intermediate in photosynthesis, contributing to energy production [45]. Together, these correlations imply a strategic shift in the resistant reaction toward immune activation over energy processes.

At 4 DPI, flavonoids identified as resistance biomarkers such as apigenin, astragalin, genistein, sissotrin, and tricin isomer were positively correlated with flavonoid biosynthesis-related genes (M1 module) (Fig. 7B). This supports our assumption that, after the early immune response, the resistant reaction shifts from defense signaling to biochemical protection, reinforcing pathogen restriction through secondary metabolites.

Conclusion

This study presents the first metabolomics-based investigation of early biomarker responses associated with scab-resistant and scab-susceptible reactions in pecan leaves during controlled inoculation. An illustration of the study’s methods and overall findings is presented in Fig. 8. By integrating pathway-based metabolomics with supervised machine learning and transcriptomics, three sequential defense mechanisms were identified that differentiate resistant from susceptible reactions prior to the onset of visible symptoms. In the resistant reaction, rapid upregulation of salicylic acid and HMG-CoA at 1–2 DPI indicated early immune activation, followed by a shift toward flavonoid and lignin biosynthesis at 3–4 DPI to reinforce pathogen restriction. In contrast, susceptible reactions were characterized by delayed and ineffective defense responses, with progressive activation of primary metabolic pathways reflecting tissue damage rather than proactive resistance. Metabolomic-transcriptomic network analysis confirmed that these metabolic shifts are coordinated with transcriptional reprogramming of defense-related gene modules, supporting their functional relevance in resistance outcomes. The identified biomarkers enabled earlier and more objective differentiation of resistance reactions (1–4 DPI) compared to conventional microscopic methods (5 DPI), demonstrating the potential of this approach as a complement or alternative to traditional resistance phenotyping in pecan breeding programs. As a controlled proof-of-concept within a single host genotype (‘Desirable’) challenged with isolates of contrasting pathogenicity, the present study identifies candidate biomarkers for the ‘Desirable’ incompatible and compatible interactions rather than universal resistance markers. Building on the established QTL framework for pecan scab resistance [14], follow-up work will extend the inoculation panel to additional V. effusa pathotypes, test biomarker performance across cultivars of distinct resistance pedigrees in the UGA pecan breeding collection, and increase biological replication, advancing these candidates toward robust application in metabolite-based resistance screening.

Fig. 8.

Overview of the study methodology and key findings

Abbreviations

2,5-DHBA

2,5-Dihydroxybenzoic acid

AMP

Adenosine monophosphate

AUC

Area under the curve

DMRM

Dynamic multiple reaction monitoring

DPI

Days post-inoculation

ESI

Electrospray ionization

FDR

False discovery rate

FN

False negative

FP

False positive

GO

Gene Ontology

HILIC

Hydrophilic interaction liquid chromatography

HMG-CoA

3-Hydroxy-3-methylglutaryl-coenzyme A

HMGR

HMG-CoA reductase

HMGS

HMG-CoA synthase

HPLC-RID

High-performance liquid chromatography with refractive index detector

KEGG

Kyoto Encyclopedia of Genes and Genomes

LC–MS

Liquid chromatography–mass spectrometry

LOD

Limit of detection

LOQ

Limit of quantitation

LR-L1

Logistic regression with L1 regularization

MAPK

Mitogen-activated protein kinase

MCC

Matthews correlation coefficient

MS/MS

Tandem mass spectrometry

PCA

Principal component analysis

PEP

Phosphoenolpyruvate

PVDF

Polyvinylidene difluoride

QC

Quality control

QqQ

Triple quadrupole

ROS

Reactive oxygen species

RDA

Redundancy analysis

RP

Reverse-phase

RSD

Relative standard deviation

SCH

Sub-cuticular hyphae

TCA

Tricarboxylic acid

TN

True negative

TP

True positive

UHPLC

Ultra-high performance liquid chromatography

WGCNA

Weighted gene co-expression network analysis

Authors’ contributions

Min Jeong Kang: Conceptualization, Investigation, Methodology, Formal analysis, Visualization, Data curation, Writing– original draft. Samuel O. Ogundipe: Methodology, Formal analysis, Data curation, Writing– original draft. Gaurab Bhattarai: Methodology, Formal analysis, Data curation, Writing– review & editing. Ronald B. Pegg: Conceptualization, Funding acquisition, Supervision, Writing–review & editing. William L. Kerr: Conceptualization, Supervision, Writing–review & editing. M. Lenny Wells: Conceptualization, Funding acquisition, Resources, Writing–review & editing. Patrick J. Conner: Conceptualization, Funding acquisition, Supervision, Resources, Writing–review & editing. Joon Hyuk Suh: Conceptualization, Funding acquisition, Supervision, Project administration, Visualization, Writing–original draft, Writing–review & editing.

Funding

This project was supported by the 2022 Specialty Crop Block Grant Program at the U.S. Department of Agriculture (Grant number: AM22SCBPGA1154-00).

Data availability

All relevant data in this study are provided in the article and its supplementary material. The code used in this study are publicly available at: (https://github.com/minjeong94/metabolomics-scab-early-biomarker/tree/main). The repository contains the data preprocessing steps, model training code, and instructions to reproduce the results.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.