Unlocking the stability and storage conditions of plant-derived nanovesicles through metabolomic and lipidomic profiling

Abstract

Extracellular vesicles (EVs) are nanoscale lipid bilayer carriers that mediate intercellular communication through the transport of bioactive cargo. Plant-derived nanovesicles (PDNVs) have emerged as sustainable platforms for therapeutic, nutraceutical, and biotechnological applications; however, their physicochemical stability under storage and handling conditions remains insufficiently characterized. Here, we present an integrated dual-omics framework combining untargeted metabolomics and lipidomics with ontology-based analyses to systematically evaluate PDNV stability. Vesicles isolated in a single batch were subjected to controlled variations in temperature (+ 25 °C, + 4 °C, -20 °C, -80 °C), storage duration (1 week-3 months), buffer composition (acidic, neutral, basic), lyophilization, and freeze-thaw (F/T) cycling (1x, 3x, 5x), followed by GC-MS and LC-MS analyses. Multivariate (PLS-DA) and univariate analyses revealed that − 80 °C best preserved native metabolite and lipid signatures, whereas + 25 °C induced pronounced remodeling, including ceramide and sugar acid accumulation and depletion of phosphatidylcholine, hexosylceramides, and N-acylethanolamines. Storage at -20 °C maintained short-term stability (~ 1 week) but showed progressive molecular drift over time. Buffer composition exerted modest yet reproducible effects, with PBS maintaining near-native lipid profiles. Lyophilization caused immediate lipid reorganization that remained largely stable during storage, while F/T cycling emerged as the most disruptive stressor. Notably, although multiple conditions altered molecular composition, most changes remained within 10%, underscoring the relative robustness of PDNVs and supporting − 80 °C storage and short-term lyophilization as preferred preservation strategies.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-52282-4.

Introduction

Extracellular vesicles (EVs) are nanoscale, lipid-bilayer-enclosed structures secreted by nearly all cell types across prokaryotic and eukaryotic domains¹. They serve as key mediators of intercellular communication by delivering a diverse cargo, including proteins, lipids, nucleic acids, and metabolites, to recipient cells². Over the past two decades, EVs have garnered increasing interest in biomedical research due to their pivotal roles in homeostasis, immunity, tissue regeneration, and disease pathogenesis^3–5. More recently, attention has expanded to include plant-derived vesicular systems, particularly plant extracellular vesicles (pEVs) and plant-derived nanovesicles (PDNVs), which represent a unique and sustainable class of bioactive nanocarriers with potential applications in therapeutic, cosmetic, and nutraceutical formulations⁶. In general, pEVs refer to vesicles actively secreted by plant cells through defined biological processes, whereas PDNVs constitute a broader and more operational category encompassing vesicle-like nanoparticles isolated from plant materials such as juices or extracts. This latter group may include both bona fide secreted vesicles and membrane fragments generated during mechanical processing.

Preservation of vesicle integrity throughout processing, storage, and transport is critical for their functionality in both experimental and commercial contexts⁷. Unlike synthetic nanoparticles, PDNVs are complex biological entities whose composition and bioactivity can be affected by a wide range of external variables-including temperature, pH, buffer composition, and stress (e.g., thermal stress). For mammalian EVs, prior studies have shown that storage temperature, lyophilization, and freeze-thaw cycles can significantly impact particle morphology, protein content, and RNA yield^8–10. However, such parameters remain largely unexplored in the context of PDNVs, and no standardized protocols currently exist for their handling and preservation.

Storage temperature is one of the most critical factors influencing the stability of vesicles. While − 80 °C storage is generally regarded as the gold standard for preserving mammalian EVs^8,11, this approach may not be universally applicable to plant vesicles, particularly those rich in hydrophobic or redox-sensitive phytochemicals. On the other hand, lyophilization (freeze-drying) offers logistical advantages such as enhanced shelf-life and easier transport¹². While some studies report that lyophilization does not negatively affect vesicle structure and function¹³, other reports suggest that it may induce structural alterations in membrane dynamics or promote vesicle aggregation^10,14. Similarly, the composition of the storage buffer, whether acidic, neutral, or basic, may influence the surface charge of vesicles, lipid fluidity, and the pH-sensitive molecular cargo¹⁵. Therefore, empirical characterization of these parameters is necessary to optimize formulation and ensure reproducibility in downstream applications^16,17.

No comprehensive study has evaluated the combined effects of storage time, temperature, pH, buffer system, and freeze-thaw cycling on the molecular integrity of PDNVs. To address this, we conducted a multidimensional stability analysis of Vitis vinifera PDNVs using complementary gas chromatography mass spectrometry (GC-MS) and liquid chromatography high resolution mass spectrometry (LC-HRMS) metabolomic and lipidomic profiling^18,19. Grapevine was selected as a model species due to its rich phytochemical profile, including polyphenols (e.g., resveratrol, anthocyanins), flavonoids, stilbenes, and essential fatty acids^20,21. Vitis vinifera PDNVs encapsulate these molecules, protecting them from degradation, enhancing bioavailability, and modulating their delivery in biological systems²². Despite increasing interest in PDNVs-based biotechnological applications, knowledge of their physicochemical stability and storage requirements remains limited. We therefore examined the effects of storage temperature, time, buffer composition, lyophilization, and freeze-thaw cycling using untargeted metabolomics and lipidomics.

Results and discussion

This work presents a comprehensive omics-based evaluation of PDNVs stability derived from Vitis vinifera. PDNVs were isolated in a single batch using conventional serial centrifugation and ultracentrifugation (Fig. 1A) and aliquoted into multiple storage conditions varying by temperature, pH, buffer composition, storage duration, lyophilization status, and freeze-thaw cycles (Fig. 1B). Freshly isolated PDNVs (TEM image can be seen in Fig. S1) served as controls. Untargeted metabolomic and lipidomic profiling (Fig. 1C) was performed using mass spectrometry to identify molecular signatures associated with vesicle membrane and content stability. There are some efforts to investigate the physical and biological activities of PDNVs to assess their stability¹¹, however, this is the first study aiming to investigate metabolite- and lipid-enriched alterations in the biomolecular profile of PDNVs using sophisticated omics approaches. By evaluating protein content, particle characteristics, and molecular composition across conditions, the insights gained herein not only provide a basis for best practices in PDNVs storage and handling but also advance the broader field of EV/pEV biology by extending its principles to non-mammalian systems and even offer a systematic analytical and molecular perspective for mammalian systems.

Fig. 1.

Comprehensive evaluation of storage and handling conditions affecting the stability of Vitis vinifera-derived plant vesicles. (A) Schematic representation of the sequential centrifugation-based isolation workflow for PDNVs. Vitis vinifera extract is subjected to stepwise centrifugation at 300 ×g (15 min), 2000 ×g (15 min), and 16,000 ×g (30 min) to remove debris and larger particles. The supernatant is subsequently filtered through a 0.22 μm membrane and ultracentrifuged at 120,000 ×g to pellet the PDNVs. (B) Stability testing framework for PDNVs, incorporating various post-storage conditions. Four different freezing temperatures (+ 25 °C, + 4 °C, -20 °C, -80 °C) were evaluated over a 3-month (M) period. Storage at − 20 °C was monitored at multiple time points (1 week (W), 1 M, 2 M, 3 M). Additional variables include buffer type (Acidic: Tris-HCl, Neutral: PBS, Basic: Tris-Base), and lyophilization stability across time points (t0, 1 W, 1 M, 3 M). In addition, post-storage stress, including freeze-thaw (F/T) cycles (1x, 2x, 3x). (C) Dual-omics platform used to assess the biomolecular stability of PDNVs under varying storage conditions. Metabolomic and lipidomic analyses were performed using integrated GC-MS and LC-MS pipelines, followed by computational bioinformatics analysis for compound identification and stability profiling.

Total protein content and PDNVs size alter in different storage conditions

Freezing temperature significantly influenced PDNVs preservation. After three months of storage, the total protein content was lowest at + 25 °C and gradually increased with lower temperatures, peaking at -80 °C, which showed levels comparable to fresh PDNVs (p = 0.161) (Fig. 2Ai). Consistently, particle size was seen to alter (between 85 and 100 nm) across storage temperatures (Fig. 2Aii), though the relative particle number dropped drastically at -20 °C and 4 °C (Fig. 2Aiii), indicating changes in particle abundance under these conditions. Storage at − 80 °C best preserved the protein content, closely resembling fresh PDNVs. Storage duration at − 20 °C further confirmed progressive PDNVs degradation over time (Fig. 2Bi). Total protein content decreased slightly from week 1 to month 2 (p = 0.16), with a sharp decrease as of month 3 (p = 0.019), while particle size consistently declined from approximately 92 nm to 80 nm (Fig. 2Bii). In parallel, the particle number significantly decreased, falling below 50% as of 2 months of storage compared to fresh EVs (Fig. 2Biii). These observations indicate time-dependent changes in protein content and particle abundance during storage at − 20 °C. Buffer composition was another critical determinant. While all tested buffers (H₂O, PBS, Tris-HCl, Tris-Base) maintained similar total protein levels except H₂O, there were notable differences in particle size and number (Fig. 2Ci-iii). PBS and Tris-HCl supported higher particle stability, whereas Tris-Base induced smaller particle size and higher particle concentration, suggesting buffer-dependent effects on particle distribution^10,23. Lyophilization had a moderate impact on PDNVs. Total protein content remained relatively stable across multiple lyophilization cycles (M1 to M3) (Fig. 2Di), though particle size marginally fluctuated (Fig. 2Dii). Also, the particle number was observed to decrease after 1 month of storage post-lyophilization (Fig. 2Diii), indicating changes in particle abundance during storage. Repeated freeze–thaw cycles were associated with changes in PDNV characteristics. Total protein content decreased after three (p = 0.045) and five cycles (p = 0.005) (Fig. 2Ei), particle size showed variability (Fig. 2Eii), and particle number declined markedly across cycles (Fig. 2Eiii), indicating cycle-dependent alterations in particle properties.

Fig. 2.

Basic stability assessment of PDNVs under various storage conditions. PDNVs preparations were evaluated for total protein content (left column), particle size (middle column), and relative particle number (right column) across five experimental parameters: (A) Storage at different temperature (+ 25 °C, + 4 °C, − 20 °C, − 80 °C) for 3-months, (B) Storage at -20 °C for different time points (1 W, 1 M, 2 M, and 3 M). (C) Storage in different buffer compositions (acidic, neutral, basic), and (D) Lyophilization (immediately after lyophilization (t0), 1 M, 2 M, and 3 M after lyophilization). A post-storage processing parameter was also examined: (E) Freeze-thaw (F/T) cycles (1x, 3x, 5x). Since lysis is only related to downstream analyses and lysis was not performed in the total PDNVs protein quantification, this parameter was not included in the analyses of total protein, particle size, and particle number. Total protein analyses were performed triplicate while particle size measurements were conducted in singlet.

Our findings are consistent with previous reports demonstrating that storage at − 80 °C best preserves PDNVs integrity, including protein content and particle number, while suboptimal conditions (< 80 °C) lead to progressive degradation and particle loss over time^10,24. Moreover, our observation that repeated freeze-thaw cycles result in a sharp decline in PDNVs stability parallels prior evidence of vesicle lysis and aggregation under such conditions²⁵. Similarly, our data showing pH-driven destabilization in Tris-Base buffer aligns with earlier studies reporting membrane disruption at alkaline pH^10,25, and the partial loss of particles after lyophilization is also in agreement with reports emphasizing the need for optimization to preserve vesicle integrity during freeze-drying²⁶. Overall, our data indicate that PDNVs quality and stability are highly dependent on storage conditions. Optimal preservation is achieved at -80 °C with minimal freeze-thaw exposure, and PBS or Tris-HCl are preferable buffers, while the reduced particle number observed under Tris-HCl conditions should not be overlooked. Lyophilization is moderately safe with slight particle loss till 1 month of storage, while repeated freeze-thaw cycles and improper buffers significantly compromise PDNVs integrity. Despite this conclusion describing the changes under storage conditions, it is very important to note that these alterations generally did not exceed 10%. These findings are essential for standardizing PDNVs handling in both experimental and clinical settings.

Metabolomic perturbations in PDNVs under distinct storage conditions

A total of 111 metabolites were annotated across the metabolomics datasets of Vitis vinifera-derived PDNVs under the investigated conditions. The descriptive statistical data related to the metabolomics analyses (see Fig. S2 for quality control, QC) for all storage conditions are provided in the Supplementary Excel file. The metabolite classes detected in our PDNVs showed broad agreement with previous LC-MS-based studies of vitis vinifera-derived nanoparticles, in which carbohydrates, organic acids, flavonoids, alkaloids, terpenoids, sterols, amino acids, and phenylpropanoid-related compounds were reported among the prominent components²⁷. Although differences in isolation strategies, analytical workflows, and reporting depth limit direct one-to-one comparison, the general overlap between our dataset and previously published profiles supports the profiling plausibility of the molecular signatures identified here.

Effect of temperature

Storage temperature was one of the strongest determinants of PDNVs stability. Partial least squares discriminant analysis (PLS-DA) (Fig. 3A-i) and principle component analysis (Fig. S3) showed that vesicles stored at -80 °C clustered closely with freshly isolated PDNVs, whereas − 20 °C samples exhibited only limited multivariate separation. Model validity was high (R²X = 0.880, R²Y = 0.932, Q² = 0.662), confirming reliable separation and predictive power. By contrast, + 4 °C and + 25 °C samples showed marked divergence, reflecting degradation and remodeling. Variable importance in projection (VIP) analysis identified membrane-associated metabolites such as stigmasteryl glucoside and maslinic acid, as well as sugar derivatives including gluconic acid and ribonic acid-γ-lactone, as key discriminators (Fig. 3A-ii). Several metabolites differed significantly across temperature conditions (FDR-adjusted p < 0.05). Compared to fresh samples, − 20 °C storage was associated with significant increases in gluconic acid and lactulose and decreases in retinoic acid and phosphoric acid (p < 0.05). Samples stored at + 25 °C showed significant alterations in multiple metabolites relative to frozen conditions (p < 0.05). The accumulation of sugar acids such as gluconic acid and glucaric acid at elevated temperatures may reflect oxidative degradation of carbohydrates and membrane-associated metabolites during storage. These compounds are known oxidation products of carbohydrate metabolism, suggesting that higher storage temperatures promote oxidative reactions affecting vesicle-associated metabolites. In contrast, the relative preservation of metabolite profiles at − 80 °C indicates that deep-freezing effectively limits chemical degradation processes within PDNVs (Fig. 3A-iii).

Fig. 3.

Multivariate and univariate metabolomic profiling of PDNVs under different storage and processing conditions (n = 3 biological replicates). (A) Storage temperature (− 80 °C, − 20 °C, + 4 °C, + 25 °C), (B) storage time (fresh, 1 W, 1 M, 3 M), (C) buffer composition (H₂O, PBS, Tris-acid, Tris-base), and (D) lyophilization (1 M, 3 M). Left panels: PLS-DA score plots; middle panels: VIP score plots; right panels: Radar plots for ANOVA-based top 10 lipids.

Temporal changes at − 20 °C

The data across different temperatures showed that PDNVs are well-preserved at -80 °C for up to 3 months; therefore, the stability at -20 °C in a time-dependent study was examined. PLS-DA (Fig. 3B-i) and PCA (Fig. S3) revealed a clear trajectory away from fresh samples as storage time increased, with clustering compact at 1 week and 1 month but diverging at 2 and 3 months. The models showed strong validity (R²X = 0.905, R²Y = 0.975, Q² = 0.539) indicating both high explanatory and predictive performance. VIP analysis highlighted metabolites such as sclareol, ginsenoside Rk2, and dimethylglycine as drivers of time-dependent separation (Fig. 3B-ii). Statistically significant decreases (p < 0.05) in saccharic acid, ribose, and lanosterol in over time may reflect disturbances in redox balance, whereas increases in threose and 5-aminovaleric acid at later time points are indicative of enhanced oxidative stress and catabolic activity, potentially compromising membrane integrity (Fig. 3B-iii). These time-dependent metabolomic shifts suggest gradual biochemical remodeling during storage at − 20 °C. Early storage periods (up to one month) maintained metabolomic profiles relatively close to freshly isolated vesicles, whereas extended storage promoted the accumulation of degradation-related metabolites and shifts in carbohydrate-associated pathways. Such patterns are consistent with progressive chemical modification of vesicle-associated metabolites during prolonged storage.

Effect of buffer composition

Buffer chemistry significantly impacted metabolomic profiles. PLS-DA (Fig. 3C-i) and PCA (Fig. S3) showed that PBS and H₂O clustered closely, while Tris-acid and Tris-base diverged strongly. The model exhibited robust explanatory and predictive performance (R²X = 0.852, R²Y = 0.986, Q² = 0.731). VIP analysis identified ribose, galactose, fucose, and 2-keto-L-gulonic acid as discriminant metabolites (Fig. 3C-ii). Relative to other buffers, PBS samples showed metabolite levels numerically closer to fresh controls, whereas Tris-base was associated with higher relative abundance of selected carbohydrate-related metabolites and Tris-acid showed increased cytidine levels (p < 0.05) (Fig. 3C-iii). These observations indicate that buffer composition can influence metabolite preservation within PDNV preparations. Differences between PBS, water, and Tris-based buffers may arise from variations in ionic strength and pH, which can modulate vesicle membrane stability as well as oxidative or hydrolytic reactions affecting vesicle-associated metabolites.

Effect of lyophilization

Lyophilization caused immediate divergence from fresh samples, with PLS-DA (Fig. 3D-i) and PCA (Fig. S3) separating lyophilized vesicles even at t₀. The models showed high reliability (R²X = 0.861, R²Y = 0.981, Q² = 0.726). With storage, 1- and 3-month groups diverged further, indicating cumulative degradation. Although lyophilization induced an immediate (t₀) but mild alteration in the metabolomic profile, subsequent storage at − 20 °C maintained the metabolomic composition largely intact for up to one month; however, beyond this period, the metabolomic content began to undergo compositional changes. VIP analysis highlighted lanosterol, heptadecanoic acid, galactose, and N-acetyl-D-glucosamine as most affected (Fig. 3D-ii). Statistically significant time-dependent decreases in selected metabolites were observed during storage (p < 0.05) (Figs. 3D-iii). These findings suggest that lyophilization introduces measurable alterations in vesicle-associated metabolites, likely due to dehydration-induced stress and redistribution of small molecules during rehydration. Prolonged storage after lyophilization may further promote chemical modification of metabolites, indicating that post-lyophilization storage conditions also influence PDNV stability.

Effect of freeze-thaw cycles

Freeze-thaw cycles induced progressive remodeling of the metabolome. PLS-DA (Fig. 4-i) and PCA (Fig. S3) showed measurable changes after a single cycle and pronounced divergence after 3x and 5x cycles. The models were strongly valid (R²X = 0.773, R²Y = 0.938, Q² = 0.809), confirming robustness. VIP analysis identified maltose, gluconic acid, dimethylglycine, lanosterol, and oxalic acid as metabolites contributing most strongly to model discrimination (Fig. 4-ii). Statistically significant differences across freeze–thaw cycles, indicating cycle-dependent alterations depletion of structural lipid-based metabolites and carbohydrates and accumulation of degradation products (Figs. 4-iii). Repeated freeze–thaw cycles may destabilize vesicle membranes and facilitate the release or transformation of intravesicular metabolites. Such structural perturbations can lead to accumulation of degradation-related metabolites and depletion of vesicle-associated compounds, explaining the progressive metabolomic alterations observed across increasing freeze-thawcycles.

Fig. 4.

Multivariate and univariate metabolomic analyses of PDNVs under freeze-thaw cycling treatments (n = 3 biological replicates). (i) PLS-DA score plot showing separation of fresh EVs from samples subjected to 1x, 3x, and 5x F-T cycles; (ii) VIP score plot identifying the most discriminant metabolites; (iii) Radar plot of the top ten ANOVA-selected metabolites highlighting significant alterations across F-T cycle groups.

Lipidomic profiling in PDNVs under distinct storage conditions

A total of 83 lipids were annotated across the lipidomics datasets of Vitis vinifera-derived PDNVs under the investigated conditions. The descriptive statistical data related to the lipidomic analyses (see Fig. S2 for QC) for all storage conditions are provided in the Supplementary Excel file. Untargeted lipidomics captured both individual lipid species and ontology-level features (lipid class, molecular function, subcellular localization, curvature, bilayer thickness, headgroup charge, and transition temperature), allowing a multi-scale evaluation of how physicochemical parameters affect vesicle lipid integrity^28,29(Figs. 5 and 6). The detection of major glycerophospholipid classes, including PA, PC, PE, PI, and DG, as well as TG, Cer, FA, and HexCer-related species, was consistent with previous reports on vitis vinifera-derived nanovesicles^30,31.

Fig. 5.

Multivariate and univariate lipidomic analyses of PDNVs under different storage and processing conditions (n = 3 biological replicates). (A) Storage temperature (− 80 °C, − 20 °C, + 4 °C, + 25 °C), (B) storage time (fresh, 1 W, 1 M, 3 M), (C) buffer composition (H₂O, PBS, Tris-acid, Tris-base), and (D) lyophilization (1 M, 3 M). (i) PLS-DA score plots; (ii) VIP score plots highlighting discriminant lipid species; (iii) Class-level lipid stability across conditions. Lipid species were grouped by biochemical class, and the y-axis represents the class-level mean abundance normalized to the corresponding Fresh group (Fresh = 1); (iv) Radar plots for ANOVA-based top 10 lipids.

Fig. 6.

Comprehensive lipid characteristics profiling of PDNVs under different storage and processing conditions (n = 3 biological replicates). (A) Lipid class composition visualized as chord diagrams, comparing the effects of temperature (i), storage time (ii), buffer type (iii), and lyophilization (iv). (B) Distribution of intrinsic curvature (i), bilayer thickness (ii), and headgroup charge (iii-iv) across conditions, represented as stacked bar plots. (C) Comparative transition temperature (i, ii, iv) and bilayer thickness (iii) profiling using radial plots, highlighting the relative abundance of lipids with low, intermediate, and high parameter values.

Temperature effects

Temperature exerted a strong influence on the global lipidome profile. PLS-DA (Fig. 5A-i) and PCA (Fig. S4) clearly separated the fresh from the stored groups (R²X = 0.856, R²Y = 0.973, Q² = 0.629), with − 80 °C and − 20 °C samples clustering closest to the fresh samples, while the + 4 °C and + 25 °C samples showed pronounced divergence and wider intra-group variability. VIP analysis identified sphingolipids (Cer 42:1;O4, HexCer 41:1;O4), free fatty acids (FA 16:1, FA 24:0), and plant-specific sterol conjugates (ASG 29:1;O; Hex; FA 18:0) as the most temperature-sensitive lipids (Fig. 5A-ii). Lipid class-level analysis shows that fresh samples exhibit a balanced distribution of monoacylglycerol (MGs), phosphatidylcholines (PCs), phosphatidic acids (PAs), N-acyl ethanolamines (NAEs), hexosylceramides (HexCers), and fatty acyls (FAs), whereas storage induces clear class-specific shifts rather than uniform alterations. PAs remain largely stable at -80 °C but decreases at higher storage temperatures, while FAs and NAEs show a consistent reduction across all storage conditions and MGs is uniformly increased relative to fresh samples (Fig. 5A-iii). ANOVA analysis revealed statistically significant differences across storage conditions, with DG 30:0 and NAE 18:0 showing lower abundance in stored groups relative to fresh samples, whereas ceramide species increased at + 4 °C, consistent with remodeling of sphingolipid-rich compartments (Fig. 5A-iv).

Ontology analyses supported these patterns. Class-level distributions revealed FAs, MGs, and NAEs dominance in fresh vesicles, while + 25 °C samples accumulated ceramides and sterol lipids (Fig. 6A-i). Functional grouping revealed that membrane lipids and lipid storage were the largest fractions under all conditions; however, storage at + 25 °C shifted the relative proportions toward signaling and neutral storage lipids (Fig. S5). Subcellular mapping confirmed that lipid droplets and ER-derived origins were the predominant sources across groups, while storage at + 25 °C and + 4 °C increased lipid droplet–associated contributions (Fig. S6). Structural indices revealed a decrease in negative-curvature lipids and enrichment of neutral-curvature species at higher temperatures (Fig. 6B-i), while transition-temperature analysis demonstrated accumulation of very low-Tm, disorder-promoting lipids in + 25 °C samples compared with frozen groups (Fig. 6C-i). Together, these data indicate that while freezing largely preserves vesicle lipidomes, thermal stress remodels both species-level and structural lipid pools, particularly sphingolipids and fatty acyls.

Temporal changes at -20 °C

The temporal analysis at -20 °C highlighted a progressive remodeling trajectory. PLS-DA (R²X = 0.879, R²Y = 0.954, Q² = 0.527) (Fig. 5B-i) and PCA (Fig. S4) showed that fresh and 1-week samples clustered tightly, while 1-, 2-, and 3-month groups shifted away from fresh and partially overlapped with each other, suggesting that most remodeling occurs between 1 week and 1 month. VIP species included phosphatidylethanolamines (PE 34:3, PE 34:2), sphingolipids (HexCer 41:1;O4, SL 52:6;O), and phosphatidylcholines (PC 36:0) (Fig. 5B-ii). Class-level analysis showed a rapid increase in MGs with a concomitant reduction in FAs and NAEs evident at 1 week and persisting through 3 months, while PCs, PAs and HexCer display comparatively modest and gradual changes over time (Fig. 5B-iii). Statistically significant decreases were observed in FA 16:0, FA 44:5, and NAE 20:0 at early time points (p < 0.05), whereas MG 18:0/20:0 and SG 29:1;O; Hex showed significant increases within the first week (p < 0.05). Over extended storage, DG 28:0/36:3, TG 28:0, and NAE 18:0 exhibited progressive increases (p < 0.05) (Fig. 5B-iv).

Ontology integration showed a decline in class diversity after one month, accompanied by enrichment of ceramides, fatty acyls, and sterol glycosides (Fig. 6A-ii). Functional analysis indicated a reduction in signaling lipids and a fluctuation in structural fractions (Fig. S7). Subcellular mapping revealed reduced plasma-membrane markers and enrichment of ER- and lipid droplet–derived species (Fig. S8). Membrane thickness distributions shifted toward low and very high extremes (Fig. 6B-ii), while transition temperatures shifted upward, suggesting less fluid and more rigid bilayers (Fig. 6C-ii). These converging datasets highlight a stabilization of an altered lipidome after one month, enriched in neutral storage lipids and structural domains but depleted of labile signaling species.

Critically, our data show that − 20 °C storage induces a time-dependent lipid redistribution that stabilizes after ~ 1 month rather than preserving the native state. Early loss of FAs and NAEs alongside enrichment of MGs and ceramide/sterol-related lipids suggests selective depletion of labile species and retention of more structurally robust components. Consistent with the known role of sphingolipids and sterols in increasing membrane order^32,33, the observed rise in transition temperature likely reflects a shift toward a more ordered bilayer. Thus, the apparent plateau represents stabilization of a remodeled lipidome, in line with prior reports that EV storage can alter lipid composition without overt structural disruption³⁴.

Buffer effects

Buffer chemistry also influenced lipid composition. PLS-DA (R²X = 0.933, R²Y = 0.974, Q² = 0.662) (Fig. 5C-i) and PCA (Fig. S4) distinguished PBS/H₂O from Tris-based groups. VIP analysis identified FA 18:2, FA 28:0, PE 34:2, PI 34:1, and FA 24:0 as key drivers (Fig. 5C-ii). Class-level analysis demonstrated buffer-dependent lipid class alterations, with PAs showing pronounced enrichment in Tris Acid, HexCer increased in Tris Base, and relatively lower PAs levels observed in H₂O, while FAs, NAEs, PCs and MGs remain largely stable across buffers (Fig. 5C-iii). Statistically significant increases in neutral lipids, including TG 50:3, TG 26:0, and FA 20:0, were observed under Tris-Base conditions (p < 0.05), whereas PBS samples maintained a phospholipid-dominant distribution relative to other buffers (Fig. 5C-iv).

The enrichment of PAs observed under Tris-Acid conditions may be associated with acid-promoted hydrolytic processes affecting glycerophospholipids³⁵. Under acidic environments, partial headgroup cleavage or chemical conversion of phospholipids (e.g., PCs, PEs, or phosphatidylinositols (PIs)) may lead to increased relative PA abundance. This mechanism provides a plausible explanation for the elevated PA levels detected under acidic buffer conditions and highlights the influence of buffer chemistry on the chemical stability of vesicle membrane lipids during storage.

Ontology-level mapping showed FAs and MGs consistently dominating across all buffers, though Tris-Base shifted the balance toward neutral storage lipids while PBS/H₂O maintained phospholipid-rich profiles (Fig. 6A-iii). Functionally, Tris-Base increased signaling lipids slightly, whereas PBS favored structural lipid stability (Fig. S9). Subcellular mapping confirmed the lipid droplet and ER as the main lipid source across all groups (Fig. S10). Headgroup charge was stably negative across buffers (85–95%) (Fig. 6B-iii). Bilayer thickness profiles revealed more disordered membranes in H₂O, more ordered high-thickness fractions in PBS, and intermediate remodeling in Tris conditions (Fig. 6C-iii). Thus, while overall lipid origins were robust to buffer choice, buffer chemistry subtly modulated bilayer architecture and class balance.

Acidic buffer conditions can promote hydrolytic reactions affecting glycerophospholipids, which may contribute to the apparent enrichment of PAs. Under low pH, acid-catalyzed cleavage and headgroup loss of phospholipids (e.g., PCs/PEs/PIs) may increase the relative abundance of PAs-like species and related degradation products, thereby shifting the observed lipid class distribution. Consistent with this, PA signals were increased in Tris-Acid conditions, suggesting that buffer chemistry may modulate chemical stability of membrane lipids during storage. While biological enzymatic activity cannot be excluded in plant-derived preparations, the strong pH dependence supports a physicochemical contribution (hydrolysis/acid-driven conversion) to the observed PA elevation.

Lyophilization effects

Lyophilization produced distinct alterations relative to fresh vesicles. PLS-DA (R²X = 0.923, R²Y = 0.996, Q² = 0.950) (Fig. 5D-i) and PCA (Fig. S4) separated lyophilized vesicles (t₀, 1 month, 3 months) from fresh samples, with 1- and 3-month groups clustering closely, suggesting major shifts occur immediately. VIP lipids included PE 34:2, FA 16:1, HexCer 40:1;O3, Cer 40:1;O4, and ASG 29:1;O; Hex; FA 22:0 (Fig. 5D-ii). The overall lipid class distribution remained preserved over 3 months after lyophilization, with MGs/PCs/PAs largely maintained and no progressive class depletion indicative of membrane breakdown (Fig. 5D-iii). Statistically significant decreases were observed in saturated fatty acids (FA 16:0, FA 17:0, FA 18:0, FA 24:0) (p < 0.05), while TG 52:2, TG 52:3, and TG 54:2 showed statistically significant increases (p < 0.05) (Fig. 5D-iv).

Ontology perspectives showed immediate redistribution of lipid classes at t₀, with MGs enrichment and NAEs depletion, as well as alterations in PAs, PCs, and PEs (Fig. 6A-iv). Functional grouping indicated transient signaling enrichment at t₀ followed by partial normalization (Fig. S11). ER lipids remained dominant, but lipid droplet contributions decreased at t₀ compared to stored samples, while a similar pattern was observed in fresh samples (Fig. S12). Headgroup charge shifted toward neutral/zwitterionic species immediately after lyophilization and partially rebalanced with storage (Fig. 6B-iv). Transition temperature profiles shifted from balanced (fresh) to low-Tm dominated at t₀, with gradual recovery of average/high-Tm fractions over 1–3 months (Fig. 6C-iv).

Overall, lyophilization appears to trigger an acute physicochemical adjustment rather than progressive degradation^36–38. The initial drop in transition temperatures and shift in headgroup composition are consistent with dehydration-induced membrane reorganization followed by partial stabilization over time.

Freeze-thaw cycle effects

Repeated freeze–thaw cycles progressively reshaped the lipidome profile. PLS-DA (R²X = 0.817, R²Y = 0.988, Q² = 0.911) (Fig. 7A-i) and PCA (Fig. S4) showed measurable changes after one cycle and stronger divergence after 3x and 5x cycles. VIP analysis highlighted PAs, diacylglycerols (DGs), NAEs, HexCer, triacylglycerols (TGs), and FAs species as sensitive candidate markers (Fig. 7A-ii). Lipid class-level analysis showed that the first freeze-thaw cycle induces a marked reduction in FAs and NAEs with concurrent shifts in PAs and HexCer relative to fresh samples, whereas additional cycles^3–5 do not cause further systematic decreases, indicating stabilization of the lipidomic profile after the initial freeze-thaw stress (Fig. 7A-iii). ANOVA indicated depletion of DG 30:0, FA 44:5, and NAE 20:0 after a single cycle, transient enrichment of PAs species, and accumulation of TGs with multiple cycles (Fig. 7A-iv).

Fig. 7.

Multivariate and univariate lipidomic profiling of Vitis vinifera-derived nanovesicles under freeze-thaw cycling (n = 3 biological replicates). (A) Effect of freeze–thaw (F/T) cycles: (i) PLS-DA score plot showing separation of fresh PDNVs from those subjected to 1x, 3x, or 5x F/T cycles; (ii) VIP score plot ranking the most discriminant lipid species; (iii) Class-level lipid stability across F/T conditions; (iv) Radar plot of the top ten ANOVA-selected lipids indicating cycle-dependent lipid alterations. (B) Lipid class composition: (i) Chord diagram illustrating lipid class distributions. (ii) Biophysical property distributions: Headgroup charge profiles presented as stacked bar plots. (iii) Transition temperature classification: Radial plot showing the relative abundance of lipids categorized by transition temperature for freeze–thaw cycles.

Ontology data confirmed these species-level patterns. Lipid class diagrams revealed FAs depletion and MGs enrichment after one cycle, followed by progressive accumulation of TGs and DGs (Fig. 7B-i). Functional grouping revealed signaling lipids increased transiently after 1x cycle, while membrane fractions became dominant again with repeated cycles (Fig. S13). ER origins remained the major source, with subtle enrichment of lipid droplet–associated fractions at later cycles (Fig. S14). Headgroup charge displayed modest oscillations, with early increases in zwitterionic fractions stabilizing thereafter (Fig. 7B-ii). Transition temperature profiles shifted from balanced (fresh) to very-low-Tm enrichment after one cycle, then toward average-Tm dominance after repeated stress, consistent with remodeling toward less flexible but more stabilized bilayers (Fig. 7B-iii).

Our data show that the first freeze-thaw cycle drives the main lipidomic shift, with minimal additional change thereafter. Early depletion of FAs and NAEs and changes in PAs/HexCer indicate sensitivity of labile lipids, while TG/DG accumulation suggests redistribution toward neutral lipids. The transition toward low-Tm species after one cycle is consistent with membrane destabilization followed by partial reorganization. Similar observations have been reported in EV systems, where most alterations occur during the initial freeze-thaw event²⁴.

Chain length and double bond profiling

To evaluate the biophysical consequences of storage- and handling-related perturbations, we performed chain length and double bond profiling across all experimental variables (Fig. S15). This analysis revealed statistically robust, condition-specific alterations in lipid composition. Under temperature stress (from frozen sample to unfrozen samples), saturated lipid species with short to medium chain lengths (C14-30, zero double bonds) were most affected, whereas long-chain unsaturated lipids displayed selective shifts, suggesting alterations in membrane fluidity and packing under elevated temperatures.

Time-dependent storage induced coordinated changes in both rigid, short-chain saturated lipids and fluid, long-chain unsaturated species (C48-C54, ≥ 1 double bond), pointing to dynamic remodeling of structural and functional lipid pools. Buffer composition produced distinct patterns, with short-chain saturated species (C14-C20, zero double bonds) and highly unsaturated long-chain lipids (≥ 2 double bonds) showing the greatest variability, indicating solvent- and pH-driven susceptibility of diverse lipid classes.

Lyophilization strongly impacted both extremes of the lipidome: short to medium saturated chains (C14-C34, 0–1 double bonds) and polyunsaturated species (C22-C36, ≥ 4 double bonds; C44-C54, ≥ 1 double bond) were markedly altered, consistent with freeze-drying–induced membrane disruption. Similarly, repeated freeze–thaw cycles triggered broad remodeling across saturated and unsaturated pools, spanning a wide range of chain lengths and degrees of unsaturation. In contrast, lysis protocols caused only modest changes, with less impact on short saturated or highly unsaturated lipids, underscoring their comparatively lower disruptive potential. Collectively, these results demonstrate that storage-related stressors remodel the vesicle lipidome in a chain length– and unsaturation-dependent manner, with implications for membrane flexibility, structural stability, and reproducibility of lipidomic biomarkers.

Common metabolites and lipids

The Upset plot of ANOVA-derived metabolite alterations revealed both unique and shared responses across storage parameters (Fig. 8A). Time and buffer composition emerged as the most influential variables, followed by temperature and lyophilization. Most perturbations were condition-specific, yet 11 metabolites were commonly affected by both time and buffer conditions, suggesting potential synergistic interactions between temporal and solvent-driven stressors. Carbohydrate-related molecules were particularly sensitive. Galactose and ribose were altered under time, buffer, and lyophilization conditions, while glucuronic acid and lactulose responded to time, buffer, and temperature stress. These sugars and sugar derivatives, involved in glycosaminoglycan and disaccharide pathways, may serve as useful indicators of PDNVs metabolic stability.

Fig. 8.

Common altered metabolites and lipids by conditions: UpSet plots (top) and bar charts (bottom) showing the number of significantly altered metabolite (A) and lipid (B) species across conditions. This integrated analysis highlights key structural and compositional lipid changes driven by physical and chemical processing stresses.

Upset analysis of lipid profiles (Fig. 8B) similarly identified lyophilization (22 unique lipids) as the strongest pre-storage driver of remodeling. Twenty-four lipids were commonly affected by temperature, time, and lyophilization, while seven were shared by temperature and lyophilization alone, suggesting additive destabilization by solvent- and time-dependent stress. Few lipids were altered across three or more variables, underscoring the condition-specific sensitivity of the lipidome.

Post-storage analysis confirmed that freeze-thaw cycles had the greatest effect on vesicle lipid integrity, uniquely altering 44 lipids. Lysis induced a smaller but detectable set of changes, with 16 lipids shared between the two stressors. These overlapping effects point to common pathways of membrane remodeling triggered by physical and chemical disruption. Together, the combined metabolite–lipid analysis highlights that while individual parameters elicit distinct signatures, freeze-thaw cycling and lyophilization represent the strongest disruptors of PDNVs molecular integrity. The comparative summary of storage and processing conditions revealed distinct impacts on vesicles metabolomic and lipidomic integrity (Table 1).

Table 1.

Summary of Storage Condition Effects on the Stability of PDNVs.

Storage Parameter	Condition	Metabolomic Stability	Lipidomic Stability	Overall EV Stability	Recommendation
Temperature for 3-months storage	–80 °C	High - minimal drift in PLS-DA, highly overlap with fresh; low degradation	High - PC/PE and SM/Cer ratios preserved	Excellent -vesicles with low oxidation Implying high stability	Optimal
	–20 °C	Moderate - minor drift after 1 month	Moderate - ceramide increase, partial PC loss	Moderate - membrane partially destabilized	Acceptable for short-term
	+ 4 °C	Low - accumulation of sugars and oxidized acids	Low - increased HexCer, MG, oxidized FA	Compromised - membrane asymmetry may be disrupted	Not recommended
	+ 25 °C	Very Low - severe metabolite degradation (e.g., gluconic acid, triethylphosphate)	Very Low - lipid hydrolysis, storage lipid accumulation	Severely compromised - possible vesicle collapse	Avoid
Storage Duration at -20 °C	1 week	High - no significant compositional drift	High - lipid class and subcellular distribution retained	Stable	Recommended
	1 month	High - mild organic acid accumulation	High - limited lipid class shift	Stable with early drift signs	Recommended
	2 months	Low - early signs of lipid/metabolite degradation	Low - PC/PE loss, MG/FA enrichment begins	Destabilized - compositional shift begins	Not suitable
	3 months	Low - degradation markers accumulate (e.g., aldehydes, NAE 20:0)	Low - phospholipid loss, storage lipid dominance	Destabilized - potential vesicle disruption	Not suitable
Buffer Type	PBS	High - good buffering capacity and metabolic preservation	High - preserves lipid class, subcellular origin, and membrane markers	Excellent	Optimal
	Tris	Moderate - increased lipid and metabolite drift	Moderate - some membrane asymmetry loss	Moderate - potential interference over time	Use with caution
	H2O (unbuffered)	Moderate - oxidized and hydrolytic metabolites observed	Moderate - TG/FA accumulation, PC loss	Moderate - highly variable outcomes	Use with caution
Freeze-Drying	lyophilized, -20 °C, 1 week	High - acceptable metabolic variance	High - membrane-associated lipids preserved	Good for short-term dry storage	Feasible
	Lyophilized, − 20 °C, > 1 month	Low - aldehyde, cyanoalanine accumulation	Low - HexCer enrichment, phospholipid loss	Destabilized upon rehydration	Not suitable
Freeze–Thaw	1x cycle	Moderate - minor metabolite drift	Moderate -slight MG/FA increase	Generally safe if minimal cycles	Acceptable
	≥ 3x cycles	Low - core metabolite degradation	Low - MG/DG enrichment, PC depletion	Major destabilization risk	Avoid

Study limitations

This study provides a comprehensive multi-omics assessment of PDNV stability; however, several limitations should be considered. The analysis was conducted primarily at the molecular level using metabolomic and lipidomic profiling, which captures compositional remodeling but does not directly resolve ultrastructural integrity. Particle characterization relied on DLS, a technique sensitive to aggregation and unable to discriminate intact vesicles from co-isolated particles; complementary high-resolution imaging approaches such as Cryo-TEM or nanoparticle tracking analysis would strengthen structural validation. In addition, vesicles were derived from a single production batch to minimize biological variability and enable controlled stability comparisons; however, this approach does not capture potential inter-batch heterogeneity. Despite these constraints, the integrated dual-omics framework applied herein provides robust and reproducible insight into PDNV compositional stability and establishes a systematic analytical foundation for future structural and functional investigations.

Another limitation of the present study is that metabolomic and lipidomic analyses were performed using an untargeted workflow without the inclusion of labeled internal standards for absolute quantification. While this approach enables broad molecular coverage and relative comparison of compositional changes, future studies incorporating isotope-labeled standards or targeted validation of selected key metabolites and lipids would improve quantitative accuracy and facilitate cross-study comparability of PDNV stability markers.

In addition, lipid and metabolite identifications were assigned using an MS/MS library-based annotation workflow and were not validated against authentic reference standards. Therefore, these annotations should be considered putative rather than definitive structural assignments. Owing to the inherent nature of untargeted omics analyses, in which a large number of features are annotated simultaneously, such unconfirmed assignments, as lots studies in the literature, require follow-up targeted analyses using authentic reference standards to establish structural identity with greater confidence.

Furthermore, although the multi-omics workflow applied here provides a comprehensive characterization of PDNV compositional stability, not all analytical techniques may be required for routine stability assessment. Once critical stability markers are identified, a reduced analytical panel combining basic physicochemical characterization (e.g., particle size distribution, particle abundance, and protein content) with targeted monitoring of selected lipidomic or metabolomic indicators may offer a cost-effective strategy for quality control during PDNV storage and formulation development.

Although Vitis vinifera was selected as a representative and phytochemically rich model system, plant-derived nanovesicles from different species may exhibit distinct lipid architectures, sterol profiles, and metabolite compositions that could influence their physicochemical stability. Variations in membrane saturation, sphingolipid content, and intrinsic antioxidant capacity across plant taxa may confer differential sensitivity to thermal or processing-related stress.

Importantly, the multi-omics stability framework established in this study is not species-restricted but rather provides a transferable analytical strategy that can be applied to other PDNV systems and potentially to mammalian extracellular vesicles. Systematic cross-species investigations will be essential to determine the extent of generalizability and to define stability signatures specific to different biological origins.

Technically, particle size and concentration measurements were acquired as single technical reads per sample due to the large number of experimental conditions and the cost- and time-intensive nature of MADLS analysis. Although MADLS provides high reproducibility through multi-angle detection and internal averaging, the absence of technical replicates limits the assessment of measurement variability. Future studies incorporating replicate measurements would further strengthen the robustness of these findings.

The relationship between molecular stability and biological function of PDNVs was not directly evaluated, which is another limitation of this study. Although multi-omics analyses revealed that parameters such as lyophilization induce immediate lipid and metabolite reorganization, the functional consequences of these changes after rehydration such as cellular uptake efficiency or bioactivity were not assessed. This is primarily because the present work was designed as a comprehensive stability assessment from a molecular perspective rather than a functional potency study. Nevertheless, the molecular signatures identified herein provide a rational framework for future investigations aimed at establishing stability-function relationships. In this regard, follow-up studies integrating controlled storage conditions with functional assays will be essential to define the biological relevance of PDNV alterations and to determine acceptable stability thresholds for downstream applications.

Conclusion

This comprehensive multi-omics study demonstrates that the physicochemical stability of PDNVs is strongly dependent on storage parameters, including temperature, duration, buffer composition, lyophilization, and freeze–thaw handling. Metabolomic profiling revealed that extended storage, particularly at + + 25 °C or after multiple freeze–thaw cycles, induced significant alterations in metabolites such as N-acetyl-D-glucosamine, gluconic acid, and oxalic acid. Lipidomic data further showed pronounced shifts in MGs and NAEs, most evident in lyophilized samples stored beyond one month. Correlation analysis confirmed reduced intra-group similarity under destabilizing conditions, while lipid enrichment highlighted a progressive decline in saturated FAs and MGs, consistent with compromised membrane integrity. Samples stored at + 25 °C, + 4 °C, or subjected to long-term lyophilization (three months) clustered separately from freshly isolated PDNVs, reflecting lipid breakdown, oxidative damage, or enzymatic remodeling. In contrast, storage at -80 °C or lyophilization limited to one month preserved molecular integrity most effectively. Together, these findings underscore the critical importance of optimized storage for maintaining PDNVs biochemical fidelity and advocate for the adoption of standardized preservation protocols to ensure reproducibility and functionality in both research and clinical applications. Nevertheless, it should be emphasized that in this study, the stability and storage conditions of PDNVs were investigated solely at the molecular level, while the more critical aspect lies in their physical evaluation. Although our DLS-based particle size distribution analyses generally revealed less than a 10% size alteration following storage, it remains crucial to substantiate these findings with techniques such as Cryo-TEM. This aspect, however, was beyond the scope of the present work. Therefore, all findings and hypotheses proposed herein warrant further validation through advanced physical characterization methods.

Given the compositional complexity of PDNVs, stability assessment may benefit from a stepwise decision framework rather than reliance on a single readout. In practice, an initial evaluation of particle size, particle abundance, and protein content could be followed by targeted review of key metabolomic and lipidomic markers identified as storage-sensitive in the present study. Samples showing acceptable physicochemical properties but substantial molecular deviation may require application-specific reconsideration, particularly for carrier-based use. Such a tiered strategy may support more rational decisions on PDNV usability during storage studies and future formulation development.

Materials and methods

PDNVs isolation and sample preparations for stability testing

Fresh Vitis vinifera was obtained from a producer located in Manisa, Aegean region, Turkiye, and the same source was used throughout the study. The juice was first clarified by sequential centrifugation steps carried out at 300 ×g for 10 min, and at 2,000 ×g for 20 min to eliminate residual debris. A further centrifugation step at 16,000 ×g for 30 min was then applied to remove micro-debris and large vesicular structures. The cleared supernatant obtained after these steps was subsequently passed through a sterile 0.22 μm filter to ensure removal of remaining large particles and to provide sterility. The filtrate was then subjected to ultracentrifugation at 120,000 ×g (Himac 90NX, Eppendorf, Japan) for 90 min at 4 °C to pellet plant extracellular vesicles. After discarding the supernatant, the pellet was gently resuspended in cold phosphate-buffered saline (PBS, pH 7.4), washed by a second round of ultracentrifugation at 120,000 ×g for 90 min, and finally resuspended in a minimal volume of PBS for downstream stability assays and biochemical analyses.

For stability assays, freshly isolated Vitis vinifera nanovesicles were aliquoted (n = 3) into equal volumes (100 µL) normalized by initial particle number (10¹⁰ particle/vial) and stored under different experimental conditions. Temperature-dependent stability was assessed by keeping aliquots at -80 °C, -20 °C, + 4 °C, and + 25 °C and analyzing them after 3-months storage. Time-dependent stability was assesed after 1 week, 1 month, 2 months, and 3 months of storage at -20 °C. Additional aliquots were resuspended in different buffer systems, including water, PBS, Tris-HCl (pH 6.2–6.5), and Tris-Base (pH 8.5-9.0), and stored at -80 °C for 3-months. For lyophilization, PDNVs suspensions (10¹⁰ particle/vial) were flash-frozen and subjected to freeze-drying under vacuum, stored at − 20 °C, and reconstituted to their original volume at t₀ (immediately after freeze-drying), 1 month, and 3 months. To evaluate the effect of repeated freeze–thaw stress, aliquots were subjected to 1x, 3x, or 5x cycles consisting of freezing at − 80 °C for 1-day and thawing at + 25 °C in a water bath, followed by gentle mixing.

Assessment of total protein content, particle size, and particle number at changing parameters

Total protein content was quantified using the bicinchoninic acid (BCA) assay. PDNVs aliquots were thawed on ice, mixed gently, and lysed by adding 0.2% Triton X-100 to solubilize membranes. Samples were diluted in the corresponding buffer used for storage, and BSA standards were prepared in the same matrix to account for background absorbance. 25 µL of each sample or standard was mixed with 200 µL of BCA working reagent in a 96-well plate, incubated at 37 °C for 30 min, and absorbance was measured at 562 nm. Protein concentrations were calculated from the standard curve and expressed as µg/mL, and relative changes compared to freshly isolated vesicles were determined for each condition and time point.

Particle size distribution and number were determined by dynamic light scattering (DLS, Malvern Panalytics, UK). Prior to measurement, samples were thawed on ice, reconstituted if lyophilized, and diluted (1:100 v/v) in the corresponding storage buffer (PBS) to obtain an appropriate scattering intensity. Aliquots were loaded into clean, dust-free disposable cuvettes, equilibrated to 25 °C. For each sample, the Z-average hydrodynamic diameter and the scattering count rate (kcps) provided by the instrument was logged as a qualitative proxy of relative particle abundance, although it was not used to estimate absolute concentrations.

Sample preparation and GC-MS/LC-MS based metabolomic analyses

For metabolomic profiling, aliquots of Vitis vinifera nanovesicles stored under the indicated stability conditions were subjected to biphasic extraction to obtain both polar and non-polar metabolites. At each time point, samples were thawed on ice, and four volumes of ice-cold methanol were added to each aliquot, followed by vigorous vortexing to ensure rapid quenching and protein precipitation. Subsequently, four volumes of water and four volumes of chloroform were added sequentially, and the mixtures were vortexed for one minute and incubated on ice for ten minutes to promote phase separation. The samples were then centrifuged at 16,000 ×g for 10 min at 4 °C. After centrifugation, the upper methanol-water phase containing polar metabolites and the lower chloroform phase containing lipids and hydrophobic metabolites were carefully collected into separate vials, avoiding the interphase³⁹. Both fractions were dried to completeness using a vacuum concentrator at room temperature. For GC-MS analysis of polar metabolites, dried extracts were first methoximated by incubation with 20 µL of methoxyamine hydrochloride (20 mg/mL in pyridine) at 30 °C for 90 min. Following this step, samples were derivatized with 80 µL of N-methyl-N-(trimethylsilyl)trifluoroacetamide (MSTFA) at 37 °C for 30 min^40,41. The derivatized samples were injected in splitless mode into a GC-MS system equipped with a DB-5MS (30 m + 10 m duraguard x 0.25 mm i.d., 0.25 μm film thickness) capillary column (Q2010, Shiamdzu, Japan). The GC oven was programmed for a 37.5-minute run, beginning with an initial 1-minute isothermal hold at 60 °C, followed by a linear temperature ramp of 10 °C min⁻¹ to 325 °C, where it was maintained for the final 10 min. The injection volume was one µL in splitless mode, and data were acquired over an m/z range of 50–650 with an event time of 0.30 s.

For LC-qTOF-MS analysis, samples dried to completeness were reconstituted in 200 µL of acetonitrile: water (1:1, v/v) containing 0.1% formic acid for metabolomic analysis before being transferred to LC vials. Samples were analyzed by LC-qTOF-MS using gradient elution (26 min) with mobile phases consisting of water with 0.1% formic acid (A) and acetonitrile with 0.1% formic acid (B). The mobile phase composition was maintained at 90% A and 10% B from 0 to 1 min, linearly shifted to 10% A and 90% B by 14 min, held constant until 15 min, and then returned to the initial conditions (90% A, 10% B) by 20 min with re-equilibration until 25 min. Separation was performed at a flow rate of 0.3 mL/min on either a C18 column (2.1 × 100 mm, 2.7 μm; Agilent) operated at 35 °C. The injection volume was set at 10 µL. Mass spectrometric data were acquired in positive ionization modes, with scan ranges of 100–1700 m/z for MS1 and 50–1700 m/z for MS2 (MS/MS), at a rate of 3 and 2 spectra per second. The instrument was operated under the following conditions: spray voltage, 3500 V; skimmer voltage, 65 V; gas temperature, 300 °C; gas flow rate, 8 L/min; nebulizer pressure, 35 psig; and fragmentor voltage, 175 V. MS/MS spectra were acquired for quality control samples at different collision energies (10, 20, and 40 eV) to ensure reliable peak annotation. In MS-based analyses, analytical performance and method suitability were systematically monitored using pooled quality control samples injected at regular intervals.

Lipidomic analysis-sample preparation and LC-MS-based analysis

Lipidomic profiling was performed using LC–MS-based analysis. The chloroform phase of the extracts, previously dried to completeness as described in the metabolomics sample preparation section, were reconstituted in 200 µL of an isopropanol: acetonitrile: water (6:3:1, v/v/v) mixture for lipidomic analysis, before being transferred to LC vials. For lipidomics, samples were analyzed by LC-qTOF-MS (6530, Agilent, USA) using gradient elution (30 min) at a flow rate of 0.25 mL/min with mobile phases consisting of water: acetonitrile (6:4, v/v) containing 0.1% formic acid and 10 mM ammonium formate (A), and isopropanol: acetonitrile (9:1, v/v) containing 0.1% formic acid and 10 mM ammonium formate (B). Chromatographic separation was performed using the following gradient program for mobile phase B: 35% at 0–1 min, 55% at 5 min, 75% at 10 min, and 95% at 15–17 min, followed by re-equilibration to 35% at 20 min, which was maintained until 26 min. Chromatographic separation was performed on a C18 column (2.1 × 100 mm, 2.7 μm; Agilent) maintained at 55 °C, with a 10 µL injection volume. Mass spectrometric acquisition was performed in both positive and negative ionization modes, scanning 100–1700 m/z for MS and MS/MS at a rate of 3 and 2 spectra per second. Instrument parameters were kept similar to those used in the metabolomics analyses. MS/MS spectra were acquired for quality control samples at different collision energies (10, 20, and 40 eV) to ensure reliable peak annotation. The performance and method suitability were systematically monitored as in Sect.  4.3.

Bioinformatic analysis

Raw chromatographic data obtained from GC-MS and LC-qTOF-MS analyses of Vitis vinifera nanovesicles were processed using Mass Spectrometry – Data Independent Analysis Software (MS-DIAL, ver. 4.92, https://systemsomicslab.github.io/compms/msdial/main.html)⁴³. Complex chromatograms were deconvoluted, peak retention times were aligned, and data matrices were generated. Both GC-MS and LC-qTOF-MS data matrices were normalized to total ion current (TIC), calculated as the sum of all detected ion intensities per sample, with feature intensities divided by the corresponding TIC value prior to statistical analysis. Features with more than 50%, with relative standard deviation values in quality control samples greater than 30% (Fig. S2) or corresponding to background compounds originating from mobile phases were excluded from the matrices. The missing values were filled with 1/5 of minimum values.

For GC-MS datasets, the processed matrices were exported to Excel, and metabolite annotation was performed using retention index-matched Fiehn library⁴². For LC-qTOF-MS–based metabolomic and lipidomic datasets, peak annotation was performed by comparing MS/MS spectra, acquired at multiple collision energies (10, 20, and 40 eV) from quality control samples, with publicly available spectral libraries provided by RIKEN in both positive and negative ionization modes. MS/MS data that could not be reliably annotated by MS-DIAL (ver. 4.92) were exported to MS-FINDER (ver. 3.61, https://systemsomicslab.github.io/compms/msfinder/main.html)⁴³, where metabolomic features were searched against the HMDB database with a 10-ppm mass tolerance. According to the Metabolomics Standards Initiative (MSI) guidelines, metabolite and lipid identifications were considered Level 2.

After normalization, the omics data were mean scaled by dividing each metabolite or lipid intensity by its mean across all samples. Detailed metabolomic and lipidomic results are provided in the Supplementary Information as group-wise means ± SE of mean-scaled feature intensities. The resulting data matrices from GC-MS metabolomics, LC-qTOF-MS metabolomics, and lipidomics (after merging in positive and negative modes) were separately imported into Soft Independent Modelling by Class Analogy (SIMCA, ver. 14.1, https://www.sartorius.com/en/products/process-analytical-technology/data-analytics-software/mvda-software/simca). Multivariate statistical analyses, including PCA and PLS-DA, were performed to evaluate clustering of PDNVs samples across different storage conditions (temperature, buffer, lyophilization, and freeze–thaw cycles). PCA plots were used to visualize overall variance structure, identify systemic errors, and detect outliers (Fig. S3 and S4). PLS-DA plots were constructed to visualize differentiation among multiple groups and pairwise comparisons.

VIP scores derived from PLS-DA were used to identify metabolites and lipids contributing most strongly to group separation. Metabolites and lipids with VIP values > 1.2 were considered to have the greatest impact on discriminating among stability conditions. Statistical analyses were performed using the MetaboAnalyst platform⁴⁴. One-way ANOVA was applied with a false discovery rate (FDR)-adjusted significance threshold of 0.05 in MetaboAnalyst 6.0. Radar plots represent the relative abundance patterns of the top 10 ANOVA-selected metabolites/lipids across experimental groups.

A class-level fresh-normalized stability metric was used to assess relative changes in lipid classes across experimental conditions. Lipid species were first grouped according to their biochemical classes, including FAs, PCs, NAEs, HexCer, MGs, and PAs. For each class and condition, the mean abundance was calculated across all lipid species assigned to that class. These class-level mean values were then normalized to the corresponding Fresh group mean, yielding a fresh-normalized value for each lipid class and condition.

For each class and condition, the mean abundance was calculated as follows:

where represents the mean abundance of the i-th lipid species across all samples within the specified condition, and n is the number of lipid species assigned to that class.

These class-level mean values were then normalized to the corresponding Fresh group to evaluate how conditions affect the biochemical state compared to the native state (FN):

This normalization method provided baseline measurements of lipid composition and enabled straightforward comparison of relative changes in lipid-class abundance across experimental conditions, regardless of absolute signal intensities. FN values greater than 1 indicate a relative increase compared with Fresh, which may reflect greater preservation or accumulation of a given lipid class, whereas values below 1 indicate a relative decrease, which may be associated with loss or degradation under the applied conditions (freeze-thaw cycling, time, temperature, and lyophilization stresses).

Finally, in the profiling module, lipid characteristics such as lipid class, molecular function, subcellular localization, curvature, headgroup charge, and transition temperature were analyzed using the LipidSig platform¹⁹, where bilayer thickness was computationally inferred based on lipid structural features, primarily acyl chain length and degree of unsaturation, and categorized into predefined thickness groups (e.g., low, intermediate, high) to enable relative comparison across conditions, and chain length and double bond distribution were profiled under the Differential Expression module. Statistical evaluation was performed using one-way ANOVA post-hoc tests, with multiple testing correction applied using the Benjamini & Hochberg method. A significance threshold of p < 0.05 was employed. The resulting characteristic matrix was exported to Excel, and visualizations were generated via a GUI-based plotting tool.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

BD, CCE, and EN contributed to conceptualization, investigation, methodology and writing the original draft. BD contributed to resources and supervision. SY contributed to investigation. CB and MGC contributed to investigation.

Funding

This work was not financially funded.

Data availability

Upon reasonable requests, data will be provided by authors.

Declarations

Competing interests

The authors declare no competing interests.