Materials and Methods

Experimental Design and Plant Material Collection

Plant material and sampling were performed following the experimental design described in Commisso et al. (2017), with minor modifications. In particular, each orchard was considered as an independent sampling unit. Fully mature fruits and corresponding leaves of Prunus avium L. were collected from seven cultivars, including early‐ (Early Bigi, Burlat, and Sandra), mid‐ (Grace Star and Romana), and late‐ripening (Durone Rosso and Ferrovia) cultivars during the 2014 and 2015 seasons within the Marostica PGI production area (Vicenza, Italy; Figure 1). For each cultivar, material was obtained from three to four distinct orchards, based on availability (Table S1). Within each orchard, sampling was performed on up to 20 trees when possible. Meteorological data representative of the sampled production area were compiled from three weather stations located in Lusiana, Breganze, and Bassano del Grappa and are reported in Table S2. In particular, monthly rainfall, number of rainy days, solar radiation, and mean temperature were compiled for each station and year. Approximately 50 fruits and 20 leaves were harvested per tree, ensuring that samples originated from multiple canopy positions and from trees of different ages in order to capture intra‐orchard variability. After collection, plant material was placed in labeled paper bags and transferred to the laboratory. On the day of harvest, fruits and leaves collected from individual trees were randomly combined to form pooled samples comprising either 50 fruits or 20 leaves. This pooling process was carried out independently three times, generating three biological replicates for subsequent metabolomic analyses. For metabolite profiling, fruits were manually destoned using a commercial cherry pitter (Westmark) and longitudinally divided into four equal segments. Two segments per fruit were immediately frozen in liquid nitrogen and used to establish technical replicates corresponding to each biological replicate, while an additional replicate was retained as backup material. Leaves were manually detached from the petiole and immediately frozen in liquid nitrogen. Frozen samples were then finely ground under liquid nitrogen using an A11 basic analytical mill (IKA‐Werke), and the resulting powders were stored at −80°C until further analysis. Fruit quality parameters were assessed using a separate composite sample of 30 fruits. Fruit firmness was measured at four evenly spaced points on each fruit with a handheld sclerometer (Agro Technologies), and individual fruit weight was recorded. Juice pH was determined using a digital pH meter (microPH 2001, Crison), while soluble solid content was measured as °Brix using a digital refractometer (DBR 35, Ningbo Kingstic). Leaf fresh weight was determined using three independent composite samples, each consisting of 20 leaves.

FIGURE 1: Geographical distribution of sampled orchards and meteorological stations within the Marostica PGI production area.

Extraction of Phenolic Compounds From Fruit and Leaf Tissues

Frozen powdered fruit and leaf tissues were extracted using an acidified chloroform–methanol solvent system, consisting of LC–MS‐grade chloroform and methanol (2:1, v/v; Honeywell) supplemented with 1% (v/v) hydrochloric acid prepared from a 37% stock solution. Specifically, 300 mg of fruit powder and 150 mg of leaf powder were extracted with solvent volumes corresponding to 10‐fold and 30‐fold (w/v), respectively. Samples were vortex‐mixed for 1 min, after which LC–MS‐grade water (Honeywell) was added in volumes corresponding to 2‐fold (w/v) for fruit extracts and 6‐fold (w/v) for leaf extracts, relative to the initial tissue powder, to induce phase separation. The mixtures were subsequently sonicated for 15 min at 40 kHz in an ultrasonic bath (Falc Instruments) and centrifuged at 4500 × g for 10 min at 4°C. Following centrifugation, the upper hydroalcoholic phase was carefully transferred to 2 mL microcentrifuge tubes and subjected to a second centrifugation step at 21,000 × g for 10 min at 4°C to remove residual particulate material. The clarified supernatants were then collected and stored at −20°C until analysis.

Chromatographic Quantification of Major Phenolics by HPLC–DAD

Polyphenolic compounds were quantified by HPLC coupled to diode array detection following an established analytical workflow (Commisso et al. 2017), adapted for the present study. Prior to chromatographic analysis, fruit and leaf methanolic extracts were diluted tenfold with LC–MS‐grade water (Honeywell) and clarified by filtration through 0.2 μm Minisart RC4 syringe filters (Sartorius Stedim). Injection volumes were set to 40 μL for fruit extracts and 20 μL for leaf extracts. Analyses were performed using a Beckman Coulter liquid chromatography system equipped with a Gold 126 solvent delivery module, a Gold 168 diode array detector, and a System Gold 508 autosampler (Beckman Coulter). Chromatographic separation was achieved using a reversed‐phase gradient composed of solvent A (LC–MS‐grade water containing 0.5% v/v formic acid and 5% v/v acetonitrile) and solvent B (100% acetonitrile). The gradient program consisted of an initial increase from 0% to 10% B over 2 min, followed by a rise to 20% B over 10 min, then to 25% B over 2 min, and to 70% B over 7 min. This was followed by an isocratic step at 70% B for 5 min, a rapid increase to 90% B in 1 min, an isocratic hold at 90% B for 14 min, and a return to initial conditions within 1 min. Column re‐equilibration was performed for 18 min prior to the next injection. Separation was carried out on an Alltima HP C18 analytical column (150 × 2.1 mm, 3 μm particle size) preceded by a C18 guard column (7 × 2.1 mm), both supplied by Alltech Associates. The flow rate was maintained at 0.2 mL min−1 throughout the analysis. Chromatographic data acquisition and processing were conducted using 32 Karat Workstation software (version 7.0; Beckman Coulter). To support the identification of the main peaks detected by DAD, additional UPLC–HRMS analyses were performed specifically for peak assignment. These analyses were conducted using the UPLC–HRMS instrumentation described below, but applying the same stationary phase, chromatographic gradient, and flow rate adopted for the HPLC–DAD method, thereby enabling partial alignment of retention times between the two analytical platforms. These targeted UPLC–HRMS analyses were used exclusively for the identification of DAD‐detected compounds, whereas untargeted metabolomic profiling was carried out using a distinct UPLC–HRMS method, described in the corresponding section of the Materials and Methods. Quantification was based on external calibration curves constructed from serial dilutions of authentic reference compounds (R 2 ≥ 0.999), including 3‐caffeoylquinic acid (chlorogenic acid), quercetin‐3‐O‐glucoside (isoquercetin), kaempferol‐3‐O‐rutinoside (nicotiflorin), cyanidin‐3‐O‐glucoside (kuromanin), p‐coumaric acid, and (+)‐catechin (Extrasynthese). Detection wavelengths were selected according to compound class: 320 nm for hydroxycinnamic acids, 350 nm for flavonols, 280 nm for flavan‐3‐ols, and 520 nm for anthocyanins. Metabolites for which authentic standards were not available were quantified using structurally related compounds as equivalents. Accordingly, neochlorogenic acid and caffeoyl quinic acid methyl ester were expressed as 3‐caffeoylquinic acid equivalents; p‐coumaroylquinic acid as p‐coumaric acid equivalents; quercetin rutinosylated derivatives as quercetin‐3‐O‐glucoside equivalents; kaempferol glycosylated and acylated derivatives as kaempferol‐3‐O‐rutinoside equivalents; and anthocyanins as cyanidin‐3‐O‐glucoside equivalents.

UPLC–HRMS Profiling of Fruit and Leaf Metabolomes

Untargeted high‐resolution mass spectrometry (HRMS) analyses were conducted using an ultra‐performance liquid chromatography (UPLC) platform coupled to quadrupole time‐of‐flight detection, based on previously published methodologies (Fasani et al. 2025) with minor adjustments. Prior to injection, methanolic extracts obtained from fruit and leaf samples were diluted with LC–MS‐grade water (Honeywell) at ratios of 1:10 and 1:50, respectively. Chromatographic separation was achieved using an ACQUITY I‐Class UPLC system (Waters) interfaced with a Xevo G2‐XS QqTOF mass spectrometer (Waters) equipped with an electrospray ionization source operating in both positive and negative ionization modes. Instrument control, data acquisition, and processing were performed using MassLynx software (version 4.1, Waters). Samples were injected onto an ACQUITY UPLC HSS T3 reversed‐phase column (2.1 × 100 mm, 1.8 μm; Waters), which was maintained at 30°C throughout the analysis. The mobile phase consisted of solvent A (water containing 0.1% formic acid; Biosolve Chimie) and solvent B (acetonitrile), both of LC–MS grade (Honeywell). Elution was performed at a constant flow rate of 0.35 mL min−1 and using the following gradient: 1% B as initial condition, held to 1% B for 1 min, then increased to 40% B at 10 min, to 70% B at 13.5 min, and to 99% at 14 min. Subsequently, the method remained in 99% B for 2 min and was then decreased to 1% B at 16.1 min. The column was subsequently re‐equilibrated at starting conditions, yielding a total run time of 20 min per sample. Samples were maintained at 8°C in the autosampler, and injection order was randomized. To monitor analytical reproducibility and instrument performance, a pooled quality control (QC) sample was prepared by combining equal aliquots of methanolic extracts and was injected at regular intervals (every nine samples). In addition, system stability was assessed using a standard solution containing chlorogenic acid, daidzein, and naringenin (1 ng μL−1), phenylalanine (2 ng μL−1), and gallic acid, quercetin, and dehydroartemisininic acid (3 ng μL−1). One microliter of this mixture was injected immediately after each QC run. Mass spectrometric parameters were set as follows: capillary voltage 0.8 kV, sampling cone voltage 40 V, source offset voltage 80 V, source temperature 120°C, and desolvation temperature 500°C. Nitrogen was used as nebulizing and desolvation gas at flow rates of 50 and 1000 L h−1, respectively, while argon served as collision gas. Data were acquired in positive and negative ionization modality, and the chromatograms acquired in negative ionization mode were used to build the subsequent datamatrix, while positive analyses were used to support the metabolite identification process. The acquisition was performed in continuum mode using two scan functions with fixed collision energies. The first function was acquired without collision energy, whereas the second function was operated at 35 V. For a subset of samples, collision energy settings were adjusted to optimize fragmentation, increasing the energy to 45 V or decreasing it to 20 V depending on the metabolites of interest. In addition, a FAST‐DDA acquisition strategy was implemented to improve metabolite identification efficiency (Zorzi et al. 2023). Data were acquired in sensitivity mode between 1.4 and 14 min over a mass range of 50–2000 m/z, using a scan time of 0.3 s. Mass accuracy was maintained through continuous lock‐mass correction using a leucine‐enkephalin solution (100 pg μL−1; Waters) infused at 10 μL min−1, providing reference ions at m/z 556.2771 (positive mode) and 554.2615 (negative mode).

Computational Processing and Annotation of Metabolomic Data

Untargeted metabolomic data files obtained by using negative ionization mode were processed using Progenesis QI software (Waters) according to the default data processing workflow. Following feature alignment and peak detection, the resulting data matrix was normalized by converting metabolite signal intensities into relative abundance values. For each sample, the sum of all detected signals was set to 100%, and individual metabolite abundances were expressed as percentages of the total signal, thereby allowing comparison of relative metabolite composition among samples and organs. Following data matrix generation, an additional manual post‐processing was applied and m/z features detected in solvent blank samples were excluded prior to statistical analysis. Preliminary metabolite annotation was performed through automated matching against publicly available databases, including MassBank, PlantCyc, the Plant Metabolic Network, and the Human Metabolome Database, based on accurate mass measurements, isotopic patterns, and fragmentation spectra generated during data processing. Additional annotation was achieved by querying the Metlin database (https://metlin.scripps.edu) using a mass tolerance of 0.003 Da and by comparison with an in‐house library of authentic reference standards. Published literature was further consulted to support metabolite assignment. All proposed metabolite identifications were subsequently confirmed by manual inspection of the spectral data and the annotation confidence reported as described in Sumner et al. (2007).

Exploratory and Discriminant Statistical Analyses

The feature matrix generated by Progenesis QI, containing m/z values and relative metabolite abundances, was subjected to multivariate statistical analysis. Principal component analysis (PCA) was performed using SIMCA software version 13.0 (Umetrics) after mean‐centering and Pareto scaling to explore intrinsic data structure and sample clustering. Samples falling outside the Hotelling's T 2 confidence region or showing elevated DModX values were considered outliers and removed prior to multivariate analysis. Supervised multivariate analyses were subsequently conducted by assigning sample groupings as response variables (Y) and applying orthogonal partial least‐squares discriminant analysis (OPLS‐DA) or orthogonal bidirectional partial least‐squares discriminant analysis (O2PLS‐DA), depending on the specific comparison and experimental design. Partial least‐squares discriminant analysis (PLS‐DA) was additionally performed to support model validation, and PLS‐DA models were evaluated using permutation testing (400 permutations). Model validity for OPLS‐DA and O2PLS‐DA analyses was further assessed by cross‐validation followed by analysis of variance (CV‐ANOVA), using a significance threshold of p < 0.01. For OPLS‐DA and O2PLS‐DA models, metabolites contributing most strongly to class separation were identified based on pq(corr) values, which describe the correlation between metabolite variables (X matrix) and class membership (Y matrix). To further support multivariate results, univariate statistical analyses were performed on selected metabolites using Students t‐test (p < 0.01) or one‐way analysis of variance (ANOVA) followed by Tukeys post hoc test, with statistical significance defined as p < 0.05. Heat maps were used to visualize the relative abundance of metabolites selected from the O2PLS–DA loading plots as the variables most associated with class discrimination. No hierarchical clustering was applied to either samples or metabolites. Metabolites were arranged according to their class association in the corresponding loading plots, and cell colors represent normalized relative abundance values.