Mature mango fruit samples were collected from Youjiang district of Baise City (23° 54’ 19.2996” N, 106° 36’ 53.5752” E), China between June and August 2024. The fruit belongs to the mango cultivars Tainong (TN), Aomang (AM), Qingmang (QM), and Yumang (YM). These cultivars were grown in distinct orchards within a 2 km radius of the Youjiang district, ensuring that the selected orchards were representative of the typical growing conditions within the region. In total, four independent trees were sampled; five fruits were collected from each tree, and the representative replicate from each tree was a composite of five fruits. The collected mango samples were transferred to plastic bags and immediately transported to the laboratory in a cool box. The samples collected from each cultivar provided representative coverage of the primary mango cultivar and sufficient fruit tissue for downstream analysis of metabolomics and microbiome sequencing.

The collected mango fruit samples were dissected for combined peel-pulp fractions using a sterilized scalpel, hereafter referred to as mango peel-pulp. Each replicate consisted of five composite fruit samples from each of the four trees for each cultivar, which were weighed equally before and ground using a ZT-1000A high-speed multifunctional grinder (Yongkang Zhanfan Industry and Trade Co., Ltd.) and further milled using a DLF-50S classified ultra-fine water-cooled grinder (Wenzhou Dingli Medical Instruments Co., Ltd.). The resulting powder was passed through a 32-mesh sieve and stored in vacuum-sealed drying bags at room temperature under anaerobic conditions for DNA extraction and metabolomic analysis.

The moisture content of the mango peel-pulp was determined using a drying method, following the national standards of the People’s Republic of China GB/T 6435-201 (National Technical Committee of Feed Industry Standardization, 2014). Crude protein content was measured using the Kjeldahl nitrogen method according to the China National Standard GB/T 6432-2018 (National Technical Committee of Feed Industry Standardization, 2018). Crude fat content was analyzed by Soxhlet extraction using a SOX606 fat analyzer (Hanon Instruments Co. Ltd., Jinan, China) according to the China National Standard GB 5009.6-2016 (National Technical Committee of Feed Industry Standardization, 2016). The crude ash content was determined according to the China National Standard GB/T 6438-2007 (National Technical Committee of Feed Industry Standardization, 2007). Neutral detergent fiber content was determined using the crucible method GB/T 20806-2022 (National Technical Committee of Feed Industry Standardization, 2022a). Acid detergent fiber and lignin content were examined according to the China National Standard NY/T 1459-2022 (National Technical Committee of Feed Industry Standardization, 2022b). Soluble sugars were quantified using the 3,5-dinitrosalicylic acid colorimetric method according to the China National Standard NY/T 2742-2015 (National Technical Committee of Feed Industry Standardization, 2015).

For untargeted LC-MS analysis to profile metabolites from mango fruit, the mango peel-pulp samples were accurately weighed into 2 mL centrifuge tubes, and then 1 mL of tissue extraction solvent (75% methanol:chloroform 9:1, 25% water) was added along with steel beads for homogenization. The samples were homogenized by grinding at 50 Hz for 60 s in a tissue grinder. After tissue homogenization, the sample tubes were centrifuged at 12,000 rpm for 10 min and the resultant supernatant was transferred to a new tube. Chromatographic separation was performed on a Vanquish UHPLC system using an ACQUITY UPLC^® HSS T3 column (2.1 × 100 mm, 1.8 µm) at 40°C, with a flow rate of 0.3 mL/min and 2 µL injection volume. The mobile phase for the positive-ion mode consisted of 0.1% formic acid in acetonitrile and 0.1% formic acid in water. For the negative ion mode, acetonitrile and 5 mM ammonium formate were used with appropriate gradient elution programs. Mass spectrometry was performed on a Thermo Orbitrap Exploris 120 equipped with an ESI source operating in both positive and negative ion modes, with spray voltages of +3.50 kV and -2.50 kV, respectively, sheath gas at 40 arb, auxiliary gas at 10 arb, and capillary temperature at 325°C. The MS1 resolution was set to 60,000, scanning m/z 100–1000, with data-dependent MS/MS fragmentation at 30% normalized collision energy, and MS2 resolution of 15,000, employing dynamic exclusion to improve data quality.

The obtained raw data files were converted to the mzXML format using ProteoWizard, and peak detection, filtering, retention time correction, and alignment were performed using XCMS with parameters optimized for accurate feature extraction. Batch effects were corrected using support vector regression based on QC samples and features with QC relative standard deviations above 30% were excluded. Multivariate statistical analyses, including Principal Component Analysis (PCA), were conducted using the ropls package in R to distinguish sample groups and identify differential metabolites with model validation using permutation tests. Significant metabolites were selected based on VIP scores of >1 and p <0.05, as shown in the volcano plot. Metabolites were identified by matching accurate mass and MS/MS fragmentation patterns against the Kyoto Encyclopedia of Genes and Genomes (KEGG) and the Human Metabolome Database (HMDB). Pathway enrichment and topology analyses were performed using MetaboAnalyst, and the results were visualized using KEGG Mapper to interpret the biological relevance of differential metabolites.

Bacterial 16S rRNA gene amplicon sequencing was performed using the Illumina HiSeq platform. DNA from peel-pulp samples of each mango fruit was extracted using the DNeasy PowerSoil Kit (QIAGEN, Hilden, Germany), and the concentration was measured using a NanoDrop2000 spectrophotometer (Thermo Fisher Scientific, Wilmington, DE, USA). The 16S rRNA gene amplicons were generated using primers V3-V4 region 335F (5’-CADACTCCTACGGGAGGC-3’) and 769R (5’-ATCCTGTTTGMTMCCCVCRC-3’) (Dorn-In et al., 2015). The USEARCH tool was used to process the generated paired-end Illumina reads, followed by joining the paired-end reads, relabeling the sequencing names, trimming the barcodes and primers, and filtering low- and high-quality reads. High-quality reads obtained were assigned to operational taxonomic units (OTUs) with 97% sequence similarity. The OTUs assigned to plant organelles were discarded. Taxonomic classification of the OTUs was performed against the SILVA database (Quast et al., 2013) using the RDP Classifier algorithm. All raw sequencing data from this project are available in the NCBI Sequence Read Archive (SRA) database under BioProject PRJNA1367298.

Mango fruit bacterial community analysis was performed using R software (v4.5.1). Any sequences annotated as plant mitochondria or chloroplasts were removed prior to downstream analysis. The OTU table was rarefied to 90,915 sequences per sample based on the lowest number of sequences contained in a sample (min= 90,915; max=287,564). Alpha diversity including Shannon diversity and Simpson diversity index analysis of bacterial communities were performed using R package ‘vegan’. The treatment mean values of different mango organs were compared using analysis of variance (ANOVA) and least significant difference (LSD) test. Beta diversity analysis based on the unweighted Unifrac distance for the bacterial community was performed using the ‘UniFrac’ function in R-Package ‘phyloseq’. Principal coordinate analysis (PCoA) plots were constructed based on unweighted UniFrac distance. Permutational multivariate ANOVA was performed by ‘adonis’ function in R package ‘vegan’ to determine the significant differences in bacterial community composition across four mango cultivars. The unique and shared bacterial OTUs across the four cultivars were examined in a Venn diagram using the R package ‘VennDiagram’. The average RAs of bacterial phyla and families were calculated using OTUs and bar graphs were plotted by the R package ‘ggplot2’. Heatmaps were constructed to visualize the relative abundance of bacterial genera using the function heatmap.2 in the gplots. ANOVA and LSD tests were used to determine significant differences between the groups. The correlation between the bacterial microbiota and specific metabolites was determined using the R package ‘ggcorrplot’.

The nutritional composition of mango peel-pulp from the four cultivars was analyzed using standardized methods (Figure 1A). The moisture content was highest in Aomang (AM), followed by Tainong (TN), Qingmang (QM), and Yumang (YM) (Figure 1B). Dry matter and ash contents were highest in YM (Figures 1C, D). Crude protein, fat, and soluble carbohydrate contents were high in the QM cultivar (Figures 1E–G). In contrast, the YM cultivar had significantly lower soluble carbohydrate and fat content than the other cultivars. Acid detergent fiber, neutral detergent fiber, and lignin contents were highest in YM, followed by AM, TN, and QM mangoes (Figures 1H–J).

A total of 2393 metabolites were detected in four distinct mango cultivars, TN, AM, QM, and YM, based on untargeted LC-MS analysis. All four cultivars showed signs of significant variation in metabolite composition within the principal component analysis (PCA) plots based on the overall clustering of mango samples (Supplementary Figure S1). We then performed pair-wise comparisons among the four cultivars to understand the variation in their metabolite composition. Mango fruits from all cultivars exhibited distinct metabolic profiles, as can be seen by sample clustering in the PCA plots (Figures 2A–F).

The QM and YM cultivars were clearly separated along PC1 (50.41%) and PC2 (24.7%), suggesting distinct metabolic profiles (Figure 2A). A clear separation between QM and TN was observed, with PC1 (65.4%) and PC2 (11.25%) explaining most of the variance (Figure 2B). AM and QM also showed distinct separation along PC1 (45.76%) and PC2 (23.53%), indicating differences in metabolic composition between cultivars (Figure 2C). The cultivars TN and AM, in comparison to YM, showed moderate but significant separation, indicating partial similarity in their metabolic composition (Figures 2D, E). Finally, the AM and TN cultivars also showed significant differences, with PC1 (55.46%) and PC2 (22.6%) explaining a significant portion of the variation. Overall, these results indicate a significant degree of metabolic distinctiveness between the different mango cultivars.

Comparative analysis of metabolite composition between Qingmang and the other cultivars revealed significantly distinct metabolite enrichment, with Qingmang enriched in 321, 453, and 430 metabolites compared to Yumang, Tainong, and Aomang, respectively, while these cultivars showed enrichment of 261, 314, and 249 metabolites relative to Qingmang (Figures 3A–C). Similarly, the Yumang cultivar was enriched in 301 and 226 metabolites compared to Tainong and Aomang, respectively, whereas these cultivars showed enrichment in 291 and 209 metabolites relative to Yumang (Figures 3D, E). The Tainong cultivar was enriched in 286 metabolites compared with Aomang, whereas this cultivar showed an enrichment of 324 metabolites relative to Tainong (Figure 3F).

Based on the HMDB database, the mango fruit-associated metabolites detected from the four distinct cultivars were annotated as carbohydrates and carbohydrate conjugates (12.37%), amino acids, peptides, and analogues (11.29%), fatty acids and conjugates (4.57%), flavonoid glycosides (4.3%), terpene glycosides (3.36%), sesquiterpenoids (3.09%), benzoic acids, and derivatives (2.96%), as the most abundant subclasses in the four mango cultivars (Supplementary Figure S2). Furthermore, the differential fruit metabolites from different mango cultivars based on KEGG mainly belonged to the class biosynthesis of secondary metabolites, microbial metabolism in diverse environments, biosynthesis of amino acids, degradation of aromatic compounds, biosynthesis of plant hormones, glucosinolate biosynthesis, ABC transporter, tyrosine metabolism, purine metabolism, glutathione metabolism, phenylpropanoid biosynthesis, phenylalanine, tyrosine and tryptophan biosynthesis, 2-oxocarboxylic acid metabolism, cremeomycin biosynthesis, and carbon metabolism (Figures 4A–F).

Specifically, when Yumang was compared to Qingmang, the top fold-change-enriched metabolites were 6-(methylsulfonyl)hexyl glucosinolate, mhppa sulfate, methyl 2,3,6-tri-O-galloyl-B-D-glucopyranoside, 1-dodecanoyl-sn-glycero-3-phosphocholine. While Qingmang was enriched with 5-(3’,4’,5’-trihydroxyphenyl)-gamma-valerolactone 4’-sulfate, tetramethylquercetin 3-rutinoside, N-gamma-glutamyl-S-trans-(1-propenyl)cysteine, kickxioside and pomiferin compared to Yumang (Supplementary Figure S3A). Similarly, the metabolites 6-(methylsulfonyl)hexyl glucosinolate, luteolin 7-O-[beta-D-glucuronosyl-(1->2)-beta-D-glucuronide], methyl 2,3,6-tri-O-galloyl-B-D-glucopyranoside and 4-methoxyglucobrassicin were enriched in Qingman compared with Tainong. While, the Tainong cultivar was enriched in the metabolites L-arginine, H-gamma-glutamyl-glutamine, ornithine and obacunone (Supplementary Figure S3B). Furthermore, the pair-wise comparison of top fold change metabolites revealed the enrichment of L-arginine, H-gamma-glutamyl-glutamine, ornithine, and 6’-O-galloyl salidroside in Qingman than in Aomang. In contrast, Aomang was enriched in icariside H1, 1-O-beta-D-glucopyranosyl-4-epiamplexine, xi-linalool 3-[rhamnosyl-(1-&Gt;6)-glucoside], and peonidin-3-O-arabinoside (Supplementary Figure S3C).

A comparison of Yumang and Tainong revealed enrichment of mhppa sulfate, quercetin-3-O-arabinoglucoside, sorbitan palmitate, lipoyllysine and dendronobiloside B. In contrast, Tainong was enriched in the metabolites tricetin, casoxin D, tetramethylquercetin 3-rutinoside, 4-methoxyglucobrassicin and 4’-methylepicatechin 5-glucuronide as compared to Yumang (Supplementary Figure S3D). The top fold-change-enriched metabolites in Aomang than Yumang were tricetin, casoxin D, phenylacetylaspartic acid and 4’-methylepicatechin 5-glucuronide, while in Yumang, they were mhppa sulfate, methyl 2,3,6-tri-O-galloyl-B-D-glucopyranoside, and (4-methylcyclohex-3-ene-1,1-diyl)dimethanol (Supplementary Figure S3E). Likewise, Tainong as compared to Aomang was also enriched in metabolites, such as methyl 2,3,6-tri-O-galloyl-B-D-glucopyranoside, 4-O-digalloyl-3,5-di-O-galloylquinic acid, 4-methoxyglucobrassicin and 4-methoxyglucobrassicin. Aomag was enriched in quercetin-3-O-arabinoglucoside, convallatoxin, crosatoside B and phenylacetylaspartic acid compared with that in Tainong (Supplementary Figure S3F).

The four mango cultivars showed a distinct number of bacterial OTUs based on rarefied sequence counts (Figure 5A). The boxplot based on the mean values of the Shannon index showed that the alpha diversity of Tainong was higher than that of Qingmang and Yumang. The Aomang cultivar showed significantly lower Shannon index values than those of the other three cultivars (Figure 5B). Similarly, the Simpson index values followed the same pattern as the Shannon diversity for all four cultivars (Supplementary Figure S4).

Analysis of the total bacterial OTUs across the four cultivars revealed significant differences in the number of OTUs. The cultivars Qingmang, Tainong, Yumang, and Aomang were inhabited by 1535, 1394, 1012 and 597 unique OTUs, respectively. A total of 631 OTUs were identified as core and shared OTUs across all cultivars (Figure 6A). Next, we conducted principal coordinate analysis (PCoA) of the unweighted UniFrac distance, a phylogenetic metric sensitive to changes in lineage presence and absence, to examine changes in the structure of fruit bacterial microbiota between different mango cultivars. We observed that the bacterial microbiota inhabiting the peel-pulp of Tainong, Aomang, Qingmang, and Yumang were moderately but significantly different from each other (PERMANOVA, p = 0.02). The PCoA1 first coordinate explained a total variation of 16.5% of the bacterial β-diversity among all cultivars. The second coordinate of PCoA 2 showed a variation of 13.2% in the bacterial β-diversity. Notably, the bacterial communities of Aomang and Yumang were clearly separated along PCoA Axis 2 (Figure 6B). These results imply that the mango genetics and biochemical properties of fruits create a niche that selectively enriches distinct sets of microbiota.

We identified a diverse number of bacteria belonging to different phyla in the fruits of four genetically distinct mango cultivars. Phyla Proteobacteria, Firmicutes, Bacteroidetes, Actinobacteria, and Myxococcota were dominant across the four cultivars. Specifically, the relative abundance of Proteobacteria was higher in Aomang and Qingmang than in Yumang and Tainong (Figure 7A). The abundance of Firmicutes was higher in Yumang and Tainong than that in Aomang and Qingmang. Bacteroidetes were more abundant in Yumang than in other cultivars. The relative abundance of Actinobacteria was significantly higher in Tainong than that in Qingmang and Yumang. The abundance of Acidobacteria was significantly higher in Qingmang and Tainong than in Yumang (Figure 7A).

At the family level, all cultivars showed a distinct enrichment of several bacterial taxa. The bacterial families Alcanivoracaceae and Catenulisporaceae were significantly enriched in Qingmang (Figure 7B). Similarly, the relative abundances of Ardenticatenaceae, Corynebacteriaceae, Geodermatophilaceae, Haliangiaceae, Weeksellaceae, and Xanthomonadaceae were higher in Aomang than in the other cultivars. The Yumang cultivar was enriched with the bacterial families Chromobacteriaceae, Marinifilaceae, Sphingomonadaceae, Sutterellaceae, and Xanthobacteraceae. The bacterial family Diplorickettsiaceae was more abundant in Tainong than in Aomang, Qingmang, or Yumang (Figure 7B).

Covering from the phylum to genus level, all cultivars showed distinct patterns of bacterial enrichment. The cultivar Aomang had the highest number of enriched bacterial genera, followed by Yumang, Tainong, and Qingmang, which is consistent with the results of cultivar separation in the PCoA. The bacterial genera enriched in Aomang were Actinocatenispora, Actinomyces, Amaricoccus, Burkholderia, Caenimonas, Candidatus Arthromitus, Chryseobacterium, Cloacibacterium, Clostridioides, Corynebacterium, Enhydrobacter, Geodermatophilus, Haemophilus, Haliangium, Janthinobacterium, Paludicola, Pseudoxanthomonas, Rothia, Rubellimicrobium, Silanimonas, and Vulcaniibacterium (Supplementary Figure S5). The fruits of the cultivar Yumang were enriched in the bacterial genera Alistipes, Aquabacterium, Aurantisolimonas, Bilophila, Colidextribacter, Duganella, Enterorhabdus, Flectobacillus, Harryflintia, Lachnoclostridium, Lachnospiraceae_NK4A136_group, Lachnospiraceae_UCG-010, Larkinella, Odoribacter, Oscillibacter, Parasutterella, Patulibacter, Rikenella, and Vogesella (Supplementary Figure S5). For the cultivar Tainong, the fruit-enriched bacterial genera were Acidiphilium, Aquicella, Curtobacterium, Diaphorobacter, Quadrisphaera, Rhodoplanes, and Rosenbergiella (Supplementary Figure S5). The relative abundance of bacterial genera Alcanivorax, Catenulispora, Jiella, Komagataeibacter, and Thiopseudomonas were greater in Qingmang than Aomang, Yumang and Tainong (Supplementary Figure S5). The abundances of the top 50 bacterial genera detected in the fruits of the four mango cultivars are shown in Supplementary Figure S6.

Several differentially abundant bacterial genera across mango cultivars were positively correlated with top fold-change metabolites. In the Qingmang cultivar, the bacterial genera Alcanivorax, Catenulispora, Jiella, Komagataeibacter, and Thiopseudomonas were positively correlated with the metabolites ornithine and L-arginine (Figure 8). Several metabolites, including neobonaspectin B, obacunone, pomiferin and asacoumarin A positively correlated with the bacterial genus Komagataeibacter. For the Tainong cultivar, Acidiphilium, Aquicella, Curtobacterium, Diaphorobacter and Quadrisphaera were positively correlated with tricetin and quinine (Figure 8). Casoxin D, glucoraphanin and kaltostat levels were positively correlated with the abundances of Aquicella, Curtobacterium and Diaphorobacter. The two metabolites mhppa sulfate and sorbitan palmitate were mainly correlated with most of the bacterial genera (Alistipes, Aurantisolimonas, Bilophila, Colidextribacter, Lachnoclostridium, Lachnospiraceae_NK4A136_group, Lachnospiraceae_UCG-010, Odoribacter, Oscillibacter, Parasutterella, Rikenella, and Vogesella) enriched in the Yumang cultivar (Figure 8). The metabolites in the fruits of Aomang (meconic acid, dendronobiloside B, aspidinol, myristicanol B, icariside H1, convallatoxin, dictamnoside I and rengyoside B) were positively correlated with Silanimonas, Pseudoxanthomonas, Caenimonas and Amaricoccus.