Integrated volatile and non-volatile metabolomics reveal aroma chemotypes in Capsicum cultivars

Abstract

The chemical complexity of Capsicum annuum flavor necessitates advanced profiling strategies beyond conventional methods. Here, SPME–GC × GC–MS and UHPLC–QTOF–MS combined with feature-based molecular networking (FBMN) were employed to comprehensively profile the flavor-related metabolomes of four representative Chinese chili cultivars (XML, CTJ, ZDT, and DLJ). A total of 452 volatile compounds were tentatively identified and 53 non-volatile metabolites annotated. Multivariate analysis demonstrated clear cultivar-specific metabolic separation. Odor threshold-based screening identified 13 potent odorants; notably, 2-methoxy-3-(2-methylpropyl)pyrazine was absent only in XML, while ultra-low-threshold trans-β-ionone was most abundant in this cultivar. These volatile signatures delineated three distinct aroma chemotypes: norisoprenoid–sesquiterpene (XML), monoterpenoid–floral (CTJ/ZDT), and C₉ aldehyde–green (DLJ). At the non-volatile level, FBMN revealed cultivar-specific enrichment patterns among capsaicinoid, lipid, and flavonoid molecular families. These integrated chemical fingerprints provide candidates for flavor-oriented breeding strategies.

Graphical abstract

Highlights

• •GC × GC–MS identified 452 volatiles from four Capsicum cultivars.
• •Three distinct aroma chemotypes were delineated by threshold screening.
• •IBMP was absent in XML, suggesting a cultivar-dependent biosynthesis
• •FBMN revealed cultivar-specific capsaicinoid and flavonoid patterns.
• •A three-branch-point metabolic framework was proposed for flavor divergence.

1. Introduction

Chili pepper (Capsicum spp.) serves as a cornerstone of the global food industry, dictating the sensory identity of countless culinary products (Siebert et al., 2022). Recently, the paradigm of quality evaluation has shifted from a pungency-centric model toward a holistic appreciation of flavor complexity (Gao et al., 2025; Zhang et al., 2025). This multidimensional trait arises from a synergistic interplay between volatile aroma compounds and a diverse array of non-volatile metabolites, which collectively define varietal identity (Wu et al., 2026).

Despite its industrial importance, comprehensive characterization of the chili flavorome remains a formidable challenge due to the inherent complexity of plant matrices. Traditional one-dimensional gas chromatography–mass spectrometry (1D GC–MS) often provides insufficient resolution, as the co-elution of structurally related compounds can obscure trace constituents that play disproportionate roles in aroma perception (Liu et al., 2024). Similarly, while non-targeted profiling via UHPLC-QTOF–MS offers high sensitivity for non-volatiles, the structural annotation of complex metabolites is frequently hindered by limited reference database. Although Feature-based Molecular Networking (FBMN) has emerged as a powerful computational strategy to organize metabolites based on fragmentation similarity (Li et al., 2023; Luo et al., 2025), a critical knowledge gap remains regarding the integrated chemical framework that links these diverse metabolite classes to specific sensory outputs.

The sensory profile of chili peppers is constructed by a sophisticated matrix of compounds with distinct functional roles. Capsaicinoids act as the primary sensory contributors by providing characteristic heat, while bioactive flavonoids and organic acids serve as essential quality modulators that influence mouthfeel and oxidative stability (Vella et al., 2025). Beyond these direct contributors, lipids function as vital precursor pools for aroma formation (Shahidi & Hossain, 2022). Specifically, unsaturated fatty acids undergo oxidative cleavage through the lipoxygenase (LOX) pathway to generate key C₆ and C₉ aldehydes, which are responsible for the green and fruity aroma notes (Li et al., 2025).

Despite growing knowledge of individual volatile and non-volatile metabolites in chili peppers, how upstream precursor allocation across interconnected biosynthetic pathways (lipoxygenase, terpenoid, and phenylpropanoid) collectively shapes cultivar-specific flavor remains poorly understood. To address this question, we integrated GC × GC–MS and UHPLC-QTOF–MS to characterize the flavor-related metabolome of four representative Chinese cultivars covering a full pungency spectrum: Chaotianjiao (CTJ) > Zidantou (ZDT) > Xiaomila (XML) > Denglongjiao (DLJ). The specific objectives were to (1) establish high-resolution volatile fingerprints and identify key aroma-active compounds using odor threshold information; (2) comprehensively profile non-volatile metabolites, including capsaicinoids, flavonoids, organic acids, and lipids, by UHPLC-QTOF-MS coupled with feature-based molecular networking (FBMN); and (3) integrate volatile and non-volatile metabolomic datasets to explore metabolic trade-offs and pathway interactions underlying cultivar-specific flavor formation.

2. Materials and methods

2.1. Chemicals and reagents

HPLC-grade methanol and acetonitrile (≥99.9% purity) were purchased from Thermo Fisher Scientific (Waltham, MA, USA) or equivalent reputable suppliers. Formic acid (LC-MS grade) and ammonium acetate (≥99.0% purity) were obtained from Sigma-Aldrich (St. Louis, MO, USA). The internal standard for volatile analysis, 2-methyl-3-heptanone (≥99% purity), and a homologous series of C7–C30 n-alkanes for retention index (RI) calculation were supplied by Sigma-Aldrich. For non-volatile profiling, L-2-chlorophenylalanine (≥98% purity) was used as an internal standard to monitor instrument stability. Ultra-pure water was prepared using a Milli-Q system (Millipore, Billerica, MA, USA). All other chemicals were of analytical grade. To prevent oxidative degradation and photo-isomerization of light-sensitive compounds (e.g., carotenoids, unsaturated fatty acids), all sample handling procedures were conducted under dim light, and samples were maintained at 4 °C or on ice throughout the preparation process.

2.2. Sample collection

Dried fruits of four representative Capsicum annuum L. cultivars, specifically Denglongjiao (DLJ), Zidantou (ZDT), Xiaomila (XML), and Chaotianjiao (CTJ), were collected from the main production areas of the Sichuan Basin in 2023 at the commercial red-ripe stage, as verified by supplier records. Following harvest, the peppers were hot-air dried at 60 ± 2 °C to a final moisture content of 8% to ensure stability. For each cultivar, approximately 100 g of dried fruits were obtained from three distinct industrial processing batches, representing three independent biological replicates (n = 3). All samples were visually inspected to confirm uniform color, size, and the absence of defects. The distinct morphological characteristics of the four cultivars are illustrated in Fig. 1.

Fig. 1.

Comparative GC × GC–MS three-dimensional contour plots showing volatile compound profiles of four chili cultivars. (A) Chaotianjiao (CTJ): 5.0–7.0 cm long, 1.0–1.5 cm in diameter; (B) Xiaomila (XML): 4.0–6.0 cm long, 0.6–1.0 cm in diameter; (C) Zidantou (ZDT): 3.0–4.5 cm long, 1.2–1.8 cm in diameter; (D) Denglongjiao (DLJ): 6.0–9.0 cm long, 3.0–5.0 cm in diameter.

2.3. Sample pretreatment

The dried chili fruits were inspected, and peduncles were manually removed. The edible pericarp and seeds were pulverized into a fine powder using a high-speed multifunction grinder. The resulting powder was passed through a 60-mesh standard sieve to ensure particle size homogeneity, thoroughly mixed, and immediately transferred into airtight amber glass vials. All processed samples were stored at −20 °C in the dark until extraction to prevent oxidative deterioration and loss of volatile constituents.

2.4. Extraction of volatile organic compounds (VOCs)

VOCs were extracted using headspace solid-phase microextraction (HS-SPME) (Li et al., 2022). Briefly, 1.0 g of powdered sample was accurately weighed into a 15 mL headspace vial. To facilitate volatilization, 1.5 mL of saturated NaCl solution was added, followed by 5 μL of 2-methyl-3-heptanone (50 μg/mL in methanol) as the internal standard for semi-quantification. Subsequently, a 65 μm PDMS/DVB fiber (Supelco, Bellefonte, PA, USA) was exposed to the sample headspace at 60 °C for 40 min to perform simultaneous equilibration and extraction. Following extraction, the fiber was desorbed in the GC injector at 250 °C for 3 min in splitless mode. All extractions were performed in six biological replicates per cultivar.

2.5. GC × GC–MS analysis

GC × GC–MS analysis was performed using a GC–MS (GC2010, Shimadzu, Japan) equipped with a solid-state modulator SSM1800. Modulation was achieved with a modulation period of 4 s. High-purity helium (99.999%) was used as the carrier gas at a constant flow rate of 1.2 mL/min. The column set consisted of a polar DB-WAX primary column (30 m × 0.25 mm × 0.25 μm, Agilent Technologies) and a mid-polar DB-17MS secondary column (1.3 m × 0.18 mm × 0.18 μm, Agilent Technologies). The oven temperature program was as follows: initial temperature 40 °C (held for 2 min), ramped at 6 °C/min to 230 °C, and held for 8 min. The mass spectrometer operated in electron ionization (EI) mode at 70 eV, scanning a mass range of m/z 40–330, with a solvent delay of 2 min. The ion source and transfer line temperatures were set to 230 °C and 240 °C, respectively.

2.6. VOCs data processing and compound identification

The acquired GC × GC–MS raw data were processed using Canvas software (version 2.5, J&X Technologies, Shanghai, China). This software automatically performed peak detection, spectral deconvolution, and calculation of linear retention indices (RI) based on the retention times of C7–C30 n-alkane standards analyzed under identical conditions. To ensure high data reliability and minimize background interference, a signal-to-noise (S/N) ratio threshold of 10 was applied, and all peaks were subjected to blank subtraction. Volatile compounds were putatively identified by comparing their mass spectra with the NIST 20 mass spectral library. Only compounds exhibiting both forward and reverse match factors greater than 800, together with RI deviations within ±10 units of literature values, were retained for further analysis. Final identifications were manually verified by experienced analysts when necessary.

2.7. Extraction of non-volatile metabolite

For non-volatile metabolite extraction, 1.0 g of the powdered sample was accurately weighed and extracted with 10 mL methanol. The extraction was performed under low-light conditions to protect light-sensitive compounds. The mixture was vortexed for 60 s, ultrasonicated for 30 min at room temperature (25 ± 2 °C), and centrifuged at 7000 ×g for 10 min at 4 °C. The supernatant was collected, filtered through a 0.45 μm membrane filter, and transferred into amber autosampler vials. To ensure data quality, procedural blanks and quality control (QC) samples (pooled extracts) were prepared in parallel. All extractions were performed in three biological replicates per cultivar.

2.8. UHPLC-Q-TOF-MS analysis of non-volatile metabolites

Chromatographic separation was conducted using a Brownlee Analytical DB C18 column (1.9 μm, 100 × 2.1 mm). The mobile phase consisted of solvent A (0.1% formic acid in water) and solvent B (methanol). The gradient elution program was set as follows: 0–2 min, 98% A; 2–13 min, 98%–10% A; 13–16 min, 10%–1% A; 16–16.1 min, 1%–98% A; and 16.1–20 min, 98% A. The flow rate was 250 μL/min, the injection volume was 3.0 μL, and the column temperature was maintained at 40 °C (Wang et al., 2025).

Mass spectrometric detection was performed using a TripleTOF instrument equipped with an electrospray ionization (ESI) source operating in positive ion mode. The ion spray voltage was 5500 V, and the source temperature was 450 °C. Nebulizer gas, auxiliary gas, and curtain gas were set to 50, 50, and 35 psi, respectively. Information-dependent acquisition (IDA) was used to collect MS/MS spectra for precursor ions within the m/z range of 50–600, applying a collision energy of 10 V.

2.9. Feature-based molecular networking (FBMN)

Raw UHPLC-QTOF-MS data were processed following previously established protocols (Mei et al., 2024). Briefly, the key parameters were set as follows: MS1 mass tolerance, 0.01 Da; MS2 mass tolerance, 0.02 Da; retention time tolerance, 0.2 min; minimum peak height, 1000 counts; and signal-to-noise (S/N) ratio threshold, 5. Following alignment, the feature quantification table (.csv) and the corresponding consensus MS/MS spectral summary (.mgf) were exported using the built-in GNPS export module in MS-DIAL.

The exported files were uploaded to the FBMN workflow on GNPS. Precursor and fragment ion mass tolerances were both set to 0.02 Da. Molecular network edges were filtered to retain connections with a cosine similarity score ≥ 0.7 and a minimum of 6 matched fragment ions. All remaining parameters were maintained at default values. Network nodes were annotated by spectral matching against public libraries available on the GNPS platform, including the GNPS community spectral library. The resulting molecular networks were exported and visualized using Cytoscape software (version 3.10.2).

3. Results and discussion

3.1. Comprehensive volatile profiling by GC × GC–MS

Representative two-dimensional total ion current (TIC) chromatograms acquired via SPME–GC × GC–MS are presented in Fig. 1. The distinct chromatographic topographies observed among the four cultivars reveal pronounced differences in peak density and signal intensity across the separation space. These cultivar-specific patterns visually confirm the inherent chemical diversity of each volatilome and demonstrate the effective spatial resolution of the analytes.

Following chromatographic deconvolution and alignment, a total of 452 volatile compounds were tentatively identified (Table S1) at Metabolomics Standards Initiative (MSI) Level 2 confidence, based on spectral matching against the NIST 20 library (match factor ≥ 800). These metabolites were classified into major chemical groups, including alcohols, aldehydes, ketones, esters, and hydrocarbons (Fig. 2A). Quantitatively, ZDT exhibited the richest volatile profile (283 compounds), followed by DLJ (248), XML (221), and CTJ (205), indicating substantial inter-cultivar variation in volatile biosynthetic capacity.

Fig. 2.

Comparative Analysis of Four Chili Pepper Cultivars. (A) Main classification and abundance of detected volatile compounds; (B) PCA score plot of volatile metabolites; (C) PLS-DA score plot of volatile metabolites; (D) PLS-DA Permutation Test Results (200 Iterations). Note: Chaotianjiao (CTJ), Xiaomila (XML), Zidantou (ZDT), Denglongjiao (DLJ).

The compound coverage achieved in the present study substantially exceeds that typically obtained via conventional one-dimensional GC–MS. For instance, Shan et al. reported 393 volatiles across 14 Capsicum accessions using 1D-GC–MS, whereas this study identified 452 compounds from only four cultivars (Shan et al., 2024). This enhanced detection is consistent with the multiplicative peak capacity and orthogonal selectivity of the GC × GC architecture, which effectively resolves co-eluting isomers and trace-level analytes often masked in ¹D separations. Such resolving power is critical for flavor chemistry, as many character-impact odorants in Capsicum possess low odor detection thresholds and may contribute disproportionately to the aroma profile despite their low abundance (Taiti et al., 2024).

3.2. Multivariate discrimination of cultivar-specific volatile profiles

Unsupervised Principal Component Analysis (PCA) was first applied to visualize the global chemosensory divergence among the four cultivars. As shown in Fig. 2B, the first two principal components explained 50.3% of the total variance, a level consistent with the high dimensionality of complex volatilomes. Biological replicates of each cultivar formed tight clusters, confirming high reproducibility of the GC × GC–MS platform and sample processing.

Supervised Partial Least Squares Discriminant Analysis (PLS-DA) was subsequently employed to maximize inter-cultivar separation and identify discriminant markers (Fig. 2C). The model exhibited strong explanatory and predictive performance (R²Y = 0.947; Q² = 0.932), clearly demarcating the four genotypes into discrete chemical spaces. A permutation test (n = 200) confirmed model validity, with the Q² intercept on the y-axis below zero, indicating no overfitting (Fig. 2D).

A total of 225 volatile compounds were identified as significant discriminant markers (VIP > 1.0, p < 0.05) distinguishing the four cultivars (Table S2). To prioritize flavor relevance, a subset of 13 potent odorants was selected based on reported low odor detection thresholds (<100 μg/kg in water) (Van Gemert, 2011). These key aroma-active compounds are listed in Table S4, including characteristic Capsicum volatiles such as 2-methoxy-3-(2-methylpropyl) pyrazine, (E)-2-nonenal, and nonanal.

3.3. Cultivar-specific volatile profiles and key aroma contributors

To illustrate cultivar-specific odorant profiles, a subset of 13 representative odor-active compounds was selected for visualization in Fig. 3. These compounds represent the most discriminant features combining high VIP scores (VIP > 1.0) and notably low reported odor detection thresholds (ODTs), and do not constitute an exhaustive list of all compounds meeting these criteria. The relative abundance profiles of these 13 discriminant odorants across the four cultivars are presented in Fig. 3. Given that relative peak area percentages rather than absolute concentrations are reported, the following discussion interprets aroma potential qualitatively by considering both abundance ranking and literature-reported ODTs. Formal odor activity values (OAVs) await determination through absolute quantification with authentic standards.

Fig. 3.

Relative abundance of discriminant volatile compounds with high VIP values and significantly low odor thresholds across four chili pepper cultivars (n = 6).

XML was characterized by the highest relative abundance of trans-β-ionone (1.55%; ODT 0.007 μg/kg in water) among the four cultivars and was the only cultivar in which 2-methylbutanoic acid accumulated to a notable level (3.42%; ODT 45 μg/kg). Its volatilome was dominated by the sesquiterpene β-longipinene (33.23%), together with substantial amounts of several esters. Notably, the bell-pepper-like methoxypyrazine IBMP was not detected. This absence, combined with the prominence of trans-β-ionone, suggests the potential for a floral–woody aroma profile that may differ from the green-pepper character typically associated with many C. chinense genotypes (Shan et al., 2024).

CTJ and ZDT exhibited monoterpenoid-dominated profiles. Both cultivars showed high relative abundances of linalool (15.21% and 14.78%, respectively; ODT 6 μg/kg) and 1,8-cineole (15.32% and 12.53%; ODT 12 μg/kg), the highest levels observed across all samples. These compounds have previously been reported as aroma-active in certain Capsicum species (Xiao et al., 2024), contributing floral and cooling notes, respectively. Low levels of IBMP were detected in both cultivars (0.10% and 0.05%), while methyl salicylate was present at approximately 1.0%. ZDT was further distinguished by higher relative contents of oxidized terpenoids, including sabinaketone (3.93%) and D-(+)-carvone (1.91%), which may confer additional aromatic nuances compared with CTJ.

In contrast, DLJ displayed a distinctly different profile enriched in lipid-derived C₉ aldehydes, most notably nonanal (3.44%; ODT 1 μg/kg) and (E)-2-nonenal (0.54%; ODT 0.1 μg/kg). The co-occurrence of (E)-2-hexenal (0.06%) and (Z)-3-hexen-1-ol (0.39%) indicates that both C₆ and C₉ branches of the lipoxygenase pathway are active (Kollmannsberger et al., 2011). DLJ also contained the highest relative abundance of methyl salicylate (3.96%) among the four cultivars. This aldehyde-rich profile is consistent with a potential green, cucumber-like, and fatty aroma character, which aligns with the mild, non-pungent sensory properties described for this cultivar.

While Li et al. (2025) recently characterized volatile composition across 20 dried C. annuum cultivars using conventional GC–MS, reporting key aldehydes (isobutyraldehyde, 3-methylbutanal) and esters as major discriminants (Li et al., 2025), the present GC × GC–MS analysis achieved substantially deeper profiling (452 compounds vs. ∼60 reported features), enabling resolution of trace co-eluting constituents and stereoisomers indistinguishable in 1D separations. Notably, odor threshold-weighted screening identified 13 potent odorants, including ultra-low-threshold trans-β-ionone (ODT 0.007 μg/kg in water), which was uniquely enriched in XML.

3.4. Global non-volatile metabolomic mapping via FBMN

Feature-based molecular networking enabled systematic organization of the non-volatile metabolome across the four cultivars. A total of 1172 consensus MS/MS features were clustered into 136 molecular families according to spectral similarity relationships in GNPS (Fig. 4). Based on spectral matching against public libraries with established cosine similarity and fragment match criteria, 53 metabolites were annotated at MSI Level 2 confidence, encompassing lipids, phenolics, amino acids, and capsaicinoids Table S3. The remaining features were grouped according to shared fragmentation patterns but lacked confident library matches, reflecting the limited coverage of plant secondary metabolites in current public MS/MS repositories.

Fig. 4.

Major Molecular Families Identified in Chili Peppers. (A) Lipid Families; (B) Amino Acid Families; (C) Capsaicinoid Families; (D) Phenolic Compound Families.

Beyond individual library hits, network topology facilitated structural contextualization of unannotated nodes through proximity to reference compounds. It is important to note that molecular networking reflects fragmentation similarity rather than confirmed biosynthetic connectivity. Therefore, structural propagation within subnetworks provides putative chemical relationships but does not by itself establish metabolic flux or pathway directionality.

3.4.1. Capsaicinoid structural propagation

The capsaicinoid molecular family was examined to illustrate the utility and limitations of network-based structural inference (Fig. 5). This subnetwork contained 64 closely connected nodes. Spectral matching confidently identified capsaicin (m/z 306.202), dihydrocapsaicin (m/z 308.244), and nonivamide (m/z 294.211) at MSI Level 2. These compounds shared conserved MS/MS fragmentation features, including benzyl cation cleavage and the diagnostic fragment ion at m/z 137 corresponding to the vanillyl moiety (Forino et al., 2016).

Fig. 5.

Manual annotation of unidentified compounds in the capsaicinoid molecular family based on characterized analogs: (A) Capsaicinoid molecular family network, (B) Capsaicin, (C) Homodihydrocapsaicin, (D) Nonivamide, (E) Norcapsaicin, (F) Dihydrocapsaicin, (G) Homocapsaicin.

Additional nodes exhibited systematic mass differences relative to these references and retained comparable fragmentation patterns. Based on consistent mass shifts and network connectivity, several features were tentatively assigned as putative homologues, such as homodihydrocapsaicin (m/z 322.201), homocapsaicin (m/z 320.233), and norcapsaicin (m/z 292.192). These annotations are classified as MSI Level 3. To avoid overinterpretation, it must be explicitly stated that these structures are proposed solely on the basis of exact mass differences and network connectivity. Because these structural assignments have not been confirmed by authentic reference standards or orthogonal analytical methods, they represent structurally characterized compound classes rather than definitive molecular identities. The observed mass differences are chemically consistent with variations in acyl chain length or saturation within the Capsicum vanilloid pathway (da Silva et al., 2018). This propagation approach extends structural interpretation beyond isolated spectral matches while remaining entirely inferential in nature.

3.4.2. Lipid molecular family

The lipid molecular family comprised 14 annotated species, including glycerides, free fatty acids, and several lysophosphatidylcholines such as LPC 16:0 and LPC 18:2. These LPCs are generated through phospholipase A2-mediated hydrolysis of phosphatidylcholine, a process that concomitantly releases free polyunsaturated fatty acids including C18:2 and C18:3 (Kaffe et al., 2024). These polyunsaturated fatty acids serve as established substrates for the lipoxygenase pathway. The detection of LPCs and related lipid species therefore supports a biochemically plausible connection between the non-volatile lipid pool and oxylipin-derived volatile aldehydes discussed in Section 3.5. However, the present metabolomic data represent steady-state abundances and do not directly quantify phospholipase activity or substrate flux toward volatile formation.

3.4.3. Phenolic molecular family

The phenolic molecular family contained 25 nodes, among which 21 compounds were annotated, including three phenolic acids, thirteen flavonoids, and five additional phenolic derivatives. Identified metabolites included naringenin, quercetin, naringenin-7-O-glucoside, luteolin-6-C-β-D-glucoside, and ferulic acid. Structural assignments were supported by characteristic fragmentation behavior, including sugar-neutral losses for O-glycosides and cross-ring cleavage patterns consistent with C-glycosylated flavonoids (Añibarro-Ortega et al., 2024).

These compounds arise from the phenylpropanoid pathway, in which phenylalanine is converted to cinnamic acid and subsequently diversified into flavonoids and hydroxycinnamic acid derivatives through sequential enzymatic transformations. The pronounced accumulation of specific flavonoid glycosides such as quercitrin and luteolin-6-C-glucoside in ZDT relative to other cultivars indicates cultivar-dependent differences in phenolic composition. Whether these differences reflect altered pathway flux, enzyme expression, or downstream glycosylation patterns is addressed in the integrative discussion presented in Section 3.5.

3.5. Pathway-specific metabolic divergence underlying cultivar flavor profiles

Multivariate analysis of the non-volatile datasets revealed clear cultivar-driven metabolic separation. PCA explained 73% of the total variance (Fig. S1), and PLS-DA modeling confirmed robust inter-cultivar discrimination (R²Y = 0.994, Q² = 0.976, p < 0.05 by permutation test). A total of 225 differential metabolites were screened (VIP > 1.0, p < 0.05). Combined with volatile profiling data (Sections 3.2–3.3), the cultivar-specific flavor profiles could be associated with differential metabolite accumulation across three major biosynthetic routes: the oxylipin, terpenoid, and phenylpropanoid pathways. All interpretations in the following subsections are based on steady-state metabolite co-occurrence patterns derived from untargeted profiling (Fig. 6). In the absence of enzymatic activity data, gene expression profiling, or stable-isotope flux analyses, these associations should be regarded as correlation-based hypotheses rather than established mechanistic conclusions. We explicitly propose them as testable frameworks to guide future functional studies.

Fig. 6.

Clustered heatmap analysis across four chili pepper cultivars.

3.5.1. Oxylipin pathway and lipid-derived aldehydes

Lipid-derived C₆ and C₉ aldehydes in pepper fruit originate from the sequential action of lipoxygenase (LOX) and hydroperoxide lyase (HPL) on C₁₈ polyunsaturated fatty acids released from membrane lipids (Matsui, 2006). Among the four cultivars, non-pungent DLJ accumulated the highest levels of 9-LOX/HPL-derived products, with nonanal (3.44%) and (E)-2-nonenal (0.54%) reaching concentrations 4–9 times higher than those in pungent cultivars XML and ZDT. This inverse relationship between LOX-derived aldehyde accumulation and pungency level is consistent with, though not sufficient to confirm, a hypothesis of competitive substrate allocation between the LOX pathway and capsaicinoid biosynthesis. We tentatively propose that in pungent cultivars, the fatty acyl-CoA pool may be preferentially directed toward capsaicinoid formation, thereby limiting substrate availability for aldehyde production. It should be emphasized, however, that this interpretation is inferential; the observed co-occurrence pattern could equally reflect independent transcriptional regulation of LOX/HPL expression rather than direct substrate competition. Definitive evaluation of this hypothesis would require fatty acid flux labeling experiments (e.g., ¹⁴C- or ²H-labeled linolenic acid tracing) or comparative BAHD acyltransferase and LOX enzyme activity assays across isogenic or near-isogenic lines differing in pungency.

3.5.2. Terpenoid pathways

CTJ and ZDT exhibited monoterpenoid-dominated volatile profiles. Linalool (15.21% and 14.78%) and 1,8-cineole (15.32% and 12.53%) were co-dominant in these two cultivars, whereas their levels were substantially lower in XML (1.46% and 0.56%) and DLJ (6.75% and 3.11%). Both compounds are end products of the plastidial methylerythritol phosphate (MEP) pathway (Tholl, 2015). Notably, CTJ and ZDT also showed low C₉ aldehyde levels. This co-occurrence pattern is consistent with, but does not demonstrate, preferential channeling of pyruvate flux toward MEP-derived terpenoid biosynthesis rather than toward fatty acid precursors shared with the oxylipin pathway (Rodríguez-Concepción & Boronat, 2015). We propose this as a testable hypothesis; direct evaluation would require ¹³C-pyruvate tracer experiments to partition flux between the MEP and fatty acid biosynthetic branches, or comparative quantification of DXS and DXR transcript abundance across cultivars.

In contrast, XML displayed a distinct terpenoid profile dominated by β-longipinene (33.23%), a sesquiterpene biosynthesized via the cytosolic mevalonate (MVA) pathway. This divergence among cultivars regarding monoterpenes synthesized via the MEP pathway (CTJ/ZDT) and sesquiterpenes originating from the MVA pathway (XML) raises the testable hypothesis that MEP and MVA pathway activities are differentially regulated across these cultivars, potentially reflecting divergence in plastidial versus cytosolic isoprenoid flux allocation. Transcriptomic profiling of key regulatory nodes, such as HMGR (MVA pathway) and DXS (MEP pathway), alongside enzyme activity measurements, would be necessary to substantiate this interpretation.

3.5.3. Phenylpropanoid pathway

The phenylpropanoid pathway channels phenylalanine into multiple downstream branches, including flavonoids, phenolic acids, and the vanillylamine moiety required for capsaicinoid biosynthesis (Aza-González et al., 2011). Among the cultivars examined, ZDT showed concurrent high flavonoid accumulation and strong pungency, a pattern consistent with coordinated upregulation of the phenylpropanoid pathway reported in recent studies (Hu et al., 2024; Yang et al., 2024). Conversely, non-pungent DLJ accumulated the highest levels of volatile phenylpropanoid derivatives, including methyl salicylate (3.96%) and 2-methoxy-4-vinylphenol (0.55%). In the absence of capsaicinoid biosynthesis, the pronounced accumulation of these specific volatiles suggests that phenylpropanoid precursors may be diverted toward volatile end products, representing a compelling target for future gene expression studies.

3.6. Integrative metabolic framework and future directions

Based on the steady-state metabolomic data, we propose a working metabolic framework in which cultivar-specific flavor divergence arises from differential metabolite partitioning at three interconnected branch points: (1) pyruvate allocation between the MEP pathway and fatty acid biosynthesis, supported by the inverse accumulation of monoterpenoids and C₉ aldehydes between CTJ/ZDT and DLJ; (2) fatty acyl-CoA allocation between capsaicinoid biosynthesis and LOX-mediated aldehyde production, supported by the contrasting aldehyde levels between pungent and non-pungent cultivars; and (3) phenylalanine allocation among capsaicinoid, flavonoid, and volatile phenylpropanoid branches, supported by the differential end-product profiles of ZDT (capsaicinoids + flavonoids) versus DLJ (volatile phenolics).

This framework is based exclusively on metabolite abundance patterns and should be considered a working hypothesis. Validation through transcriptomic profiling and enzymatic assays of key branch-point enzymes (e.g., DXS, LOX, PAL, CS) is warranted. Furthermore, although 225 discriminant volatiles were identified, only 13 with confirmed low odor detection thresholds were highlighted as candidate key odorants in this study. Future work integrating GC–olfactometry (GC–O) with aroma reconstitution and omission experiments will be essential to establish causal links between the metabolic divergence documented here and the perceived flavor attributes of each cultivar.

4. Conclusion

By integrating GC × GC–MS, UHPLC-QTOF-MS, and FBMN, this study tentatively identified 452 volatile compounds and annotated 53 non-volatile metabolites across four Capsicum cultivars. Odor threshold-guided multivariate screening prioritized 13 potent aroma-active compounds that delineated cultivar-specific chemotypes. Molecular networking further revealed cultivar-dependent distributions of capsaicinoids, flavonoids, and lipid-derived metabolites, providing complementary metabolic context for flavor divergence. Collectively, these findings propose a testable multi-pathway framework linking terpenoid biosynthesis, lipoxygenase-mediated oxylipin formation, and capsaicinoid accumulation to Capsicum flavor diversity, and propose validated aroma marker candidates to support sensory-directed breeding. Future integration of quantitative odor activity assays, sensory reconstitution–omission experiments, and pathway-level functional analyses will be essential to establish causal genotype-to-flavor relationships.

CRediT authorship contribution statement

Lintao He: Writing – original draft, Investigation. Hong Li: Methodology. Lanxi Yin: Data curation. Junyi YangYi: Data curation. Chunyou Ye: Software. Ting Xia: Data curation. Guangfeng Zhang: Resources. Hongtao Li: Resources. Weili Li: Resources. Tao Wu: Writing – review & editing, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material: Supplementary Table 1: Table S1. Comprehensive profiling of 452 volatile metabolites in four pepper varieties via GC×GC-MS. Supplementary Table 2: Comparative Analysis of Differential Metabolites Across Four Chili Pepper Cultivars.

Data availability

Data will be made available on request.