The experiment was conducted from August 2024 to February 2025 in a light-quality controlled chamber at the School of Life Sciences, Qufu Normal University. Test materials consisted of perennial Begonia ‘Black Velvet’ seedlings purchased from the Weifang Flower Market. All seedlings originated from the same batch and supplier. Following, a one-month acclimatization period in the Qufu Normal University laboratory, individuals exhibiting consistent size and growth vigor were selected for experimental treatments.

Specifically, treatments were administered within plant growth chambers maintained at 22°C, 85-90% relative humidity, and a 6-h light/18-h dark cycle. The experimental design incorporated two environmental factors: light intensity and light quality. Light treatments were provided by LED modules (Jiada Lighting Co., Ltd., China; models: White light: W5730-12V, Red light: R5730-12V, Blue light: B5730-12V). The spectral power distribution of each LED type at the three intensity levels is provided in Supplementary File 1 Table 1. Each factor had three levels: light intensities of 5, 20, and 50 μmol·m^-2·s^-1; and light qualities of white, red, and blue light. Consequently, this resulted in nine distinct treatment combinations including 5 μmol·m^-2·s^-1 white light (W5), 20 μmol·m^-2·s^-1 white light (W20), 50 μmol·m^-2·s^-1 white light (W50), 5 μmol·m^-2·s^-1 red light (R5), 20 μmol·m^-2·s^-1 red light (R20), 50 μmol·m^-2·s^-1 red light (R50), 5 μmol·m^-2·s^-1 blue light (B5), 20 μmol·m^-2·s^-1 blue light (B20), 50 μmol·m^-2·s^-1 blue light (B50). For each treatment, nine pots of plants were utilized, with three pots constituting one replicate, thereby providing three biological replicates per treatment. All plant materials were grown under these controlled conditions for 14 days. Subsequently, newly developed leaves were harvested as experimental samples, flash-frozen in liquid nitrogen, and stored for subsequent analyses.

Data were analyzed in Office 2021 (Microsoft Corporation, Redmond, WA, USA). Heatmaps, upset plots and Venn diagrams were created using TBtools-II v2.154 (Chen et al., 2023). One-way analysis of variance was conducted using Duncan’s method in IBM SPSS Statistics 25. Histograms were generated using Origin 2021 (OriginLab Corporation, Northampton, MA, USA). Principal Component Analysis (PCA) visualization was performed using R version 3.5.1, metabolite classification circular plots, correlation analysis and K-means analysis were generated using R version 4.2.0, differential lipid metabolite classification scatter plots were created using R version 4.2.2.

In Begonia ‘Black Velvet’ subjected to different light qualities and intensities, we detected and annotated a total of 492 lipid metabolites. A heatmap illustrates their accumulation patterns across the treatments (Figure 1A). Classification analysis revealed diverse lipid types, categorized into 16 distinct classes (Figure 1B). Steroid and isoprene lipids were the most abundant classes, comprising 114 and 108 metabolites, respectively, each exceeding 20% of the total. Fatty acyls and Fatty acids followed, each accounting for approximately 15% of the total metabolites. Nine classes glycerophospholipids, fatty alcohol, glycerol glycolipids, fatty acid esters, other, Lpe, sphingolipid, glycosides and glycerol ester collectively contained metabolites representing 1% to 5% of the total. Conversely, metabolites classified into glycolipids, phospholipids, and Lpc (Lysophosphatidylcholines) were relatively scarce, each constituting less than 1% of the total.

A petal-style Venn diagram (Figure 1C) showed that 444 lipid metabolites were common to all nine treatments, with no treatment-specific metabolites detected. Furthermore, analysis of the number of lipid metabolites detected per treatment (Figure 1D) revealed relatively minor differences; the lowest count was 472 (R5 treatment), while the highest was 487 (W5 treatment). The plot of Principal Component Analysis (PCA) shows that triplicate samples for each treatment clustered tightly together, and the nine treatments formed distinct clusters (Figure 1E). PCA analysis accounted for 54.36% of the total variances, therein, PC1 explained39.65%, 14.71% for PC2 (Figure 1E).

Applying the defined differential lipid metabolites (DLMs) screening criteria, we detected and annotated 443 DLMs spanning all 16 classification classes. Moreover, the low-intensity treatments (R5, W5, B5) and the high-intensity treatments (W50, B50, R50) were separately clustered into different groups (Figure 2A).

To further characterize accumulation patterns, K-means clustering was performed on the 443 DLMs, partitioning them into nine distinct subclasses (Figure 2B). The subclasses exhibited considerable variation in size. Subclass 9 was the largest, containing 113 DLMs with high accumulation under W50, R50, and B50. In contrast, subclass 8 comprised 48 DLMs, showing the opposite pattern with low accumulation under W50, R50, and B50. Subclass 6 comprised 50 metabolites characterized by low accumulation under W5, R5, and B5, while subclass 4 contained 49 metabolites with high accumulation under these same low-intensity treatments (W5, R5, B5). Subclass 2 was the smallest, with only 21metabolites that displayed low accumulation across all white light intensities but high accumulation across all three red light intensities.

To assess the specific effects of light quality, we detected and annotated differential lipid metabolites (DLMs) between pairs of light qualities (blue vs red, blue vs white, red vs white) at each fixed intensity (5, 20, and 50 μmol·m^-2·s^-1). Bubble plots (Figure 3A) illustrate the number of DLMs for each comparison.

Under extremely low light intensity (5 μmol·m^-2·s^-1), about 290 DLMs were recorded from each three pairwise comparisons. Moreover, treatment B5 showed higher proportion of DLM accumulation (Figure 3A). Among the recorded DLMs, 181 DLMs were shared by these three different comparison groups (Figure 3B).

At 20 μmol·m^-2·s^-1, a higher proportion of DLM accumulation was also found under blue-light-treated samples compared to other comparison groups (Figure 3A). For instance, 246 DLMs for B20_vs_R20 comparison group, and 286 and 221 for B20_vs_W20 and R20_vs_W20, respectively (Figure 3A). Among these DLMs, 128 DLMs were shared by all comparison groups (Figure 3C).

Under 50 μmol·m^-2·s^-1 light intensity, treatment R50 showed high proportion of DLMs among these three light intensity-treated groups. Moreover, the DLM numbers were 167, 208 and 235 for comparison groups B50_vs_R50, B50_vs_W50 and R50_vs_W50, respectively (Figure 3A). Therein, only 92 DLMs were shared by these three comparison groups (Figure 3D). In addition, 25 candidate DLMs (Supplementary File 2 Table 2) that exhibit consistent differential accumulation among different light−quality treatments were selected.

To evaluate the effects of light intensity, we detected and annotated differential lipid metabolites (DLMs) between pairs of intensities (5, 20, and 50 μmol·m^-2·s^-1) within each light quality (white, red, and blue). Bubble plots summarize the number of DLMs for each pairwise comparison (Figure 4A).

Under white light, high-light-treated samples (W50, W20) showed higher proportion of DLMs accumulation compared to W5 treatment (Figure 4A). Moreover, treatment W50 has the highest metabolite accumulation under white light quality (Figure 4A). And 257, 304, and 316 DLMs were separately found from comparison groups W20_vs_W5, W20_vs_W50, and W5_vs_W50. Therein, 177 DLMs were shared by these three different light intensity-treated comparison groups (Figure 4B).

Under red-light-treated samples, the comparisons R20_vs_R5, R20_vs_R50, and R5_vs_R50 contained 292, 302, and 329 DLMs, respectively. Treatment R50 has higher accumulation of DLMs compared to other samples under low light intensity (Figure 4A). Moreover, a total of 198 DLMs were shared among all three red-light-treated comparisons (Figure 4C).

Under blue-light-treated samples, treatment B50 showed the higher DLMs accumulation compared to other comparisons (Figure 4A). The comparison groups B20_vs_B5, B20_vs_B50, and B5_vs_B50 separately annotated 279, 277, and 308 DLMs. Moreover, 174 common DLMs were shared by these three different comparison groups (Figure 4D). In addition, 23 candidate DLMs, showing consistent differential accumulation across intensity gradients were finally selected through integrated analysis among white, red, and blue light comparison groups.

Among these 432 DLMs, 420 DLMs were recorded under different light quality-treated samples, 432 for light intensity-treated samples (Figure 5A). Therein, 409 metabolites overlapped (Figure 5A), indicating a shared response to both factors. Moreover, certain intensity-specific DLMs (e.g., POS_t1278, POS_t961) showed differential accumulation under multiple light qualities (Figure 5B). Likewise, some quality-specific DLMs (e.g., POS_t569) responded differentially across all three intensities under particular spectral conditions (Figure 5C).

The activities of antioxidant enzymes (SOD, POD, CAT), malondialdehyde (MDA) content, and soluble protein (SP) contents varied across light treatments (Figure 6). Under 5 μmol·m^-2·s^-1, white- and red-light-treated samples showed the highest CAT activity, while SOD activity was lowest under blue light. In contrast, MDA contents were low across all qualities at this intensity. At optimal intensity (20 μmol·m^-2·s^-1), white light elicited peak synergistic activation of POD and SOD (303.33 U POD, 1788.24 U SOD, respectively). Soluble protein content reached maximal levels across all light treatment (1.61–1.73 mg/g), while blue light achieved minimal MDA accumulation (0.036 μmol/g). Under high-stress (50 μmol·m^-2·s^-1) conditions, divergent antioxidant strategies emerged. Blue light triggered an SOD surge (1211.6 U, + 75% vs 20 μmol) with concurrent MDA escalation (0.090 μmol/g, +151%). Red light exhibited POD compensation (114.11 U, doubling 20 μmol levels) despite SOD collapse (decreased 31%). Conversely, white light suffered catastrophic enzyme suppression (POD decreased 85%, SOD decreased 74%) alongside soluble protein depletion (decreased 93%). Intriguingly, blue light uniquely maintained elevated soluble protein (0.29 mg/g, +24%) and conserved CAT activity (21.0 U) contrasting with 64–65% CAT suppression under white and red light conditions.

Correlation analysis integrating physiological indices and candidate lipid metabolites revealed several significant associations (Figure 7). Two lipid metabolites, NEG_t647 and NEG_t874 exhibited significant positive correlations with POD (correlation coefficient > 0.6; p < 0.001). Four lipid metabolites, NEG_t647, NEG_t874, POS_t513, and POS_t721, showed similarly strong positive correlations with SOD (correlation coefficient > 0.6; p < 0.001). Two lipid metabolites, NEG_t840 and NEG_t864, were significantly positively correlated with MDA (p < 0.001), while NEG_t865 and POS_t961correlated positively with SP (p < 0.001). Additionally, NEG_t865, POS_t499, and POS_t513, demonstrated significant positive correlations with CAT (p < 0.001).

Overall, nine lipid metabolites, including NEG_t647, NEG_t840, NEG_t864, NEG_t865, NEG_t874, POS_t499, POS_t513, POS_t721, and POS_t961 showed significant correlations (p < 0.001) with physiological indices. Among these, three were previously identified as responsive to both light quality and intensity, four were primarily responsive to intensity, and two were primarily responsive to quality.

Total RNA from all samples was subjected to paired-end transcriptome sequencing. Raw sequencing data were processed to obtain clean data by removing adapters and low-quality reads. Quality metrics are summarized in Supplementary File 4 Table S4). Clean reads per sample ranged from 19.64 million to 30.44 million, corresponding to a total base count between 5.83 Gb and 8.93 Gb. GC content was consistent across samples (44.68% to 47.00%). Q30 scores were all above 95.82%, with most samples exceeding 96% and several reaching over 99%, indicating high base-calling accuracy and reliable sequencing quality. These metrics confirm the sequencing data are of high quality and suitable for downstream transcriptome assembly and differential expression analysis.

To elucidate the transcriptional basis of light-induced lipid remodeling, we performed RNA sequencing across all nine treatment groups (three light qualities × three intensities). We focused on two core biosynthesis pathways, including steroid and fatty acid metabolism that were prominently represented among the candidate lipid metabolites.

Steroid biosynthesis pathway. We mapped 38 DEGs onto nine enzymatic steps leading to ergosterol (Figure 8A). Seven ERG-family genes, including ERG1, ERG3, ERG5, ERG6, ERG24, and ERG25, exhibited expression levels significantly correlated (r > 0.6, p < 0.001) with the accumulation of key steroid metabolites (Figure 8A; Supplementary File 5 Table S5).

Fatty acid biosynthesis pathway. Similarly, we identified 57 DEGs across all steps of the fatty acid biosynthesis pathway from acetyl-CoA to hexadecanoic acid (Figure 8B; Supplementary File 5 Table 5.). Nine of these DEGs, including four accD, two accB, two FabF, and one FATB, exhibited strong correlations (r > 0.6, p < 0.001) with fatty-acyl metabolite levels (Figure 8B).

The coordinated expression of these pathway-specific DEGs, particularly the ERG family genes for steroid synthesis and accD/FabF for fatty acid metabolism, provides a transcriptional framework that supports the observed light-dependent lipid compositional changes.

This study systematically deciphers how light quality and intensity modulate lipid metabolism in the shade-adapted Begonia ‘Black Velvet’. We detected and annotated 492 lipid metabolites across nine light treatments, classified into 16 categories with steroid (23.17%) and isoprene lipids (21.95%) predominating. Principal component analysis (PCA) revealed that light intensity was more strongly associated with the overall differences in light quality on global lipid profiles, as treatments clustered primarily by intensity (5/20/50 μmol·m^-2·s^-1) and secondarily by spectral quality. The high degree of overlap (444 out of 492) in metabolites across all treatments indicates a stable core lipidome. The absence of treatment-unique metabolites and the variation in metabolites detected per treatment (472–487) together imply that light regimes primarily induce quantitative remodeling of a shared metabolite set, rather than a wholesale restructuring. Notably, despite this high commonality, treatments grouped more strongly by intensity in the PCA, underscoring the dominant role of light intensity in driving these quantitative changes.

Specific wavelengths modulate lipid metabolism through dedicated photoreceptors (Lim et al., 2023). In this study on Begonia, we annotated 25 differential lipid metabolites (DLMs) influenced primarily by light quality, including steroids and fatty acyls. Notably, pregnanetriol accumulation was specifically enhanced under blue light. This aligns with known blue-light responses in other species, where cryptochrome signaling upregulates fatty acid desaturase genes, increasing polyunsaturated fatty acids (PUFAs) that can contribute to membrane adaptation (Ma et al., 2016).

Concurrently, light intensity exerted a dominant quantitative effect. We annotated 23 DLMs significantly influenced by intensity, with metabolites such as (2R,3R,4S,5S,6R)-2-[(2R)-2,3-Dihydroxypropoxy]-6-(hydroxymethyl)oxane-3,4,5-triol peaking at 20 μmol·m^-2·s^-1, coinciding with maximal soluble protein levels. This suggests an optimal intensity for biosynthesis. Higher intensity (50 μmol·m^-2·s^-1) broadly increased the abundance of numerous lipids, a shift consistent with a redirection of carbon flux under high light stress (Alboresi et al., 2016; Ge et al., 2025).

Critically, we found extensive synergistic regulation: 409 lipid metabolites responded to both factors. For example, low-intensity blue light (B5) elevated putative PUFA-rich membrane lipids, while high-intensity blue light (B50) shifted accumulation toward potential antioxidant species like chabrolosteroid E. These coordinated changes underscore that lipid remodeling in Begonia is fine-tuned by the interplay of spectral signal and energy availability (photon flux).

It is important to note that while our lipidomic profiling revealed significant remodeling of lipid classes including potential PUFA-containing species, direct quantitative measures of membrane fluidity parameters (e.g., galactolipid ratios or fatty acid unsaturation indices) were not within the scope of this widely-targeted study. Therefore, the proposed link between observed lipid changes and membrane fluidity adjustment remains a plausible hypothesis based on established knowledge of lipid function and the stress-marker correlations observed here. Future targeted lipid analyses and biophysical assays will be valuable to directly test this mechanism in Begonia.

Lipids are pivotal for plant adaptation to light, serving dual roles in energy storage and membrane integrity (Burgos et al., 2011). Our data reveal that Begonia ‘Black Velvet’ modulates these functions differentially in response to light intensity and quality.

Energy storage is optimized under moderate, non-stressful light. The significant increase in glyceroglycolipids at 20 μmol·m^-2·s^-1, particularly under white light, aligns with peak soluble protein levels, indicating a metabolic state favoring biosynthesis. This light regime provides sufficient photosynthate to activate fatty acid synthesis and storage lipid assembly (Kowalczyk et al., 2021). These reserves can be mobilized via β-oxidation to sustain metabolism under subsequent stress (Zheng et al., 2016).

Membrane stabilization involves dynamic adjustments in lipid composition. Under high light (50 μmol·m^-2·s^-1), we observed an increased relative abundance of lipid species that are typically enriched in polyunsaturated fatty acids (PUFAs). In many plants, such a shift is a known adaptive mechanism that could help maintain membrane fluidity and mitigate photo-oxidative damage (Carmona-Salazar et al., 2015; Wu et al., 2023). Concurrently, specific glycerophospholipids like phosphatidylcholine (PC) underwent compensatory changes, which can enhance membrane stability and protein function (Wang et al., 2020; Morales-Martínez et al., 2022). These putative adaptive adjustments are reflected in the correlation of key lipids (e.g., NEG_t840, NEG_t864) with oxidative stress markers (MDA), and others (e.g., NEG_t865, POS_t513) with antioxidant enzymes (CAT/SOD).

The physiological profiling revealed that soluble protein content uniquely peaked at 20 μmol·m^-2·s^-1 across light qualities, a trend distinct from antioxidant enzymes and MDA. This indicates that 20 μmol·m^-2·s^-1 represents a light optimum for growth in this shade-adapted species, where resources are allocated to protein synthesis and general metabolism. In contrast, at 50 μmol·m^-2·s^-1, the activation of specific antioxidant defenses (SOD, POD) and the rise in MDA signify a shift in priority from growth to stress acclimation. The decline in soluble protein under high light likely reflects resource reallocation toward stress-responsive proteins and/or protein degradation under oxidative pressure. Thus, the soluble protein peak serves as a biomarker for favorable conditions, while antioxidant and MDA responses mark the onset of light stress, explaining their divergent patterns (Dahl et al., 2015; Song et al., 2016).

As an understory-adapted species, Begonia’s lipid plasticity, characterized by PUFA enrichment and sterol-mediated membrane reinforcement contrasts with sun-tolerant models. Our findings suggest precision horticulture strategies: blue-enriched spectra at ≤ 20 μmol·m^-2·s^-1 can enhance antioxidant capacity, while intensity gradients can be tuned to optimize energy storage (20 μmol) or stress resilience (50 μmol).

Nine key lipids, including chabrolosteroid E (NEG_t647) (regulated by both light intensity and quality) and culobophylin C (POS_t721) (quality-specific) emerged as central to photoprotection. The distinct response patterns of key lipid metabolites to light intensity versus quality reveal fundamental differences in their biosynthetic regulation. The contrasting responses of two key photoprotective lipids, POS_t721 (culobophyllin C, an isoprene lipid) and NEG_t647 (chabrolosteroid E, a steroid) highlight distinct biosynthetic triggers. POS_t721 accumulated primarily under specific light qualities (e.g., blue light) with little change across intensities, suggesting its biosynthesis is directly coupled to photoreceptor signaling (e.g., cryptochrome activation of the plastidial MEP pathway) (Vranová et al., 2013). In contrast, NEG_t647 responded synergistically to both light quality and intensity, reflecting a homeostatic adjustment to membrane stress. Steroids, as key membrane components, likely accumulate under high light to stabilize membranes against photo-oxidative damage, a need that scales with photon flux (Ferrer et al., 2017). Thus, Begonia employs a layered regulatory strategy: light quality fine-tunes specific signaling molecules (e.g., isoprene lipids), while light intensity broadly recalibrates structural components (e.g., steroids) in response to cellular stress.

Light quality regulates gene expression through photoreceptor-specific signaling pathways (Sanchez et al., 2020), ultimately affecting lipid synthesis (Poitelon et al., 2020). Our transcriptomic data show that key lipid metabolic genes (e.g., accD, FabF, and ERG family members) are regulated by light. This aligns with the established paradigm where photoreceptors (e.g., phytochromes for red light, cryptochromes for blue light) relay signals through intermediates like the COP1/SPA complex and PIF transcription factors to reshape transcription (Llorente et al., 2017; Maldini et al., 2015). The upregulation of fatty acid desaturation-related genes under blue light may be linked to the cryptochrome-PIF axis (Ma et al., 2016), while red light effects on membrane lipid genes likely involve phytochrome signaling. Although we did not directly validate upstream receptors, our data position lipid metabolic reprogramming as a key downstream response to light cues.

These transcriptional changes provide a mechanistic basis for the observed lipid remodeling. We focused on a core set of regulatory genes whose expression showed strong correlations with metabolite accumulation. Within the fatty acid pathway, accD (along with accB) controls the committed step of synthesis initiation (Madoka et al., 2002), while FabF regulates chain elongation and influences unsaturated fatty acid supply (Morgan-Kiss and Cronan, 2008); in Xanthoceras sorbifolium, FabF collaborates with other enzymes to promote specific fatty acid accumulation (Shi et al., 2024). All differentially expressed steroid-related genes belonged to the ERG gene family, central to plant steroid and lipid metabolism (Bishop and Koncz, 2002).

Future research should focus on elucidating the subfunctional divergence within the ERG gene family and employing synthetic biology approaches to optimize metabolic flux for enhancing stress resistance and high-value lipid production. It should be noted that the 14−day treatment period, while ecologically relevant for capturing integrated adaptive responses in this perennial shade−adapted species, may combine direct photoregulatory signals with ontogenetic adjustments during leaf development. Future time−course experiments could help disentangle the acute effects of light quality and intensity from longer−term developmental reprogramming of lipid metabolism.