Background

Apple is one of the most economically important fruit crops all over the world. Due to their high nutritional value and strong ecological adaptability, apples are preferred by both growers and consumers [1, 2]. Following efforts of domestication and cultivar breeding, the global production of apples in 2020 has exceeded 86 million tons [3]. Moreover, modern growers are keen to produce cultivars tailored to the apple industry that focus on and cater to discerning consumers, with changing preferences and dietary habits, who are highly aware of the nutritional and health benefits of fruits [4].

Apple belongs to the genus Malus within the family Rosaceae. The cultivated apples belong to M. × domestica Borkh., while other Malus species are commonly known as wild apples or crab apples. Genetic and morphological evidence has suggested that cultivated apples are more closely related to M. sieversii, a wild species native to the Tianshan Mountains in central Asia, than other wild species [5,6,7]. As apples spread along the Silk Road between Asia and Europe, their genetic makeup changed due to hybridization with local wild apples. Following the primary progenitor M. sieversii, cultivated apples obtained intensive introgressions from M. sylvestris [6,7,8]. Fruit taste and size are two of the most important characteristics in apple breeding. The content and composition of organic acids, soluble tannins, and soluble sugars contribute to the overall taste of apples and other fruits [9]. Wild apple fruits are astringent and sour due to higher contents of tannins and organic acids than cultivated apples, which are not astringent and less acidic [10]. Also, wild apples are smaller than cultivated apples, suggesting an extensive selection for fruit size during apple domestication [7, 11]. While the domestication of maize and tomato was initiated from ancestral species with small fruits, the domestication of apple benefits from M. sieversii bearing large fruits [12].

Thousands of apple cultivars have been reported worldwide due to diverse hybridization and introgression processes [13]. Besides the fruit taste and size, other characteristics such as texture [14], storage life [15], and nutrition [16] vary in apples with different pedigrees. Recent studies have demonstrated that metabolism changes within a species are greater than expected [17, 18], and this genetic diversity determines the naturally occurring variation in metabolites. Integrating metabolic analysis with other omics has been widely used in identifying genes and elucidating plant pathways [19, 20]. In particular, metabolome-based genome-wide association studies (mGWAS) have been proven efficient in clarifying natural variation in metabolism and its genetic control in tomato [21], peach [22], rice [23], jujube [24], and other crops. GWAS has been recently used to map the genetic loci for important phenotypic traits in apples [25,26,27]. The diversity of apples has rendered them an “ideal model” to interpret the genetic basis underlying the accumulation of metabolites and, in turn, the regulation of the metabolome.

In the present study, we collected 292 accessions of wild and cultivated apples to compare the genomic and metabolic differences among apples of different pedigrees. A comprehensive population genomic analysis and metabolomic variation related to apple domestication and improvement were conducted. Subsequently, mGWAS was carried out to identify the key genes associated with specific modifications. The results provide insights into genetic variants associated with apple domestication and improvement, and the data are valuable for clarification of the mechanisms controlling quality traits.

Results

Genome variation of wild and cultivated apple accessions

We collected and re-sequenced 292 apple accessions worldwide (Fig. 1a), including 11 M. sieversii, seven M. sylvestris, 19 other wild species, 38 heirlooms, and 217 cultivars representing different pedigrees and consumption types, which yielded 2.67 Terabyte (Tb) high-quality cleaned sequences, with a median depth of 11.42 × and coverage of 94.06% of the assembled genome (Additional file 2: Tables S1 and S2). A total of 2,956,396 high-quality single-nucleotide polymorphisms (SNPs) were selected for genetic variation analyses (Additional file 2: Table S3). Principal component analysis (PCA) results indicated that the cultivars were separated from wild accessions and were much closer to M. sieversii and M. sylvestris, while heirlooms were distributed from wild to cultivars (Fig. 1b). The phylogenetic tree supported the clustering of the wild and cultivated clades, where the M. sylvestris and all apple cultivars situated in a large monophyletic cluster. Population structure analysis demonstrated that the domesticated apples along with M. sieversii and M. sylvestris were separated from other wild species when K ranged from three to five (Fig. 1c). In general, the distribution of heirlooms was partially intermingled with wild and apple cultivars, indicating a transition of heirlooms from wild to cultivars. Likewise, the “Fuji,” “Ralls Janet,” and “Starking Delicious” pedigrees formed three sub-clusters in a monophyletic clade, which suggests the “Starking Delicious” and “Ralls Janet” as the pollen parents of “Fuji” [28]. Rapid linkage disequilibrium (LD) decays were observed in cultivars and heirlooms, and the process was found slow in M. sieversii and M. sylvestris (Fig. 1d). Nucleotide diversities were different between groups, with other wild species exhibiting the highest degree of genetic variation while the lowest was observed in the “Fuji” pedigree due to erosion caused by breeding (Fig. 1e).

Fig. 1
figure 1

Genome-wide relationship in cultivated apple and wild species. a Geographic distribution of apple accessions. The circle represents the sample size. b Principal component analysis (PCA) of apple accessions. c The phylogenetic tree and population structure of apple accessions. “RJ” represents “Ralls Janet,” “SD” represents “Starking Delicious,” “GD” represents “Golden Delicious” and “Jona” represents “Jonathan.” The accession orders and positions on the X-axis are the same. d Linkage disequilibrium (LD) decay patterns in M. sieversii, M. sylvestris, other wild, heirloom, and cultivar groups. e Nucleotide diversity (π) between different groups. Lowercase letters indicate the significant differences based on Fisher’s least significant difference (LSD) test at p ≤ 0.01

Full size image

Variation of metabolites related to apple domestication and quality improvement

Apple flesh was collected from 270 accessions comprising nine other wild accessions, six M. sieversii, 38 heirlooms, and 217 cultivars, and we constructed an MS2 spectral tag (MS2T) library using a widely targeted metabolic profiling method [29]. A data matrix of 2575 metabolites, including 486 annotated and 2089 unannotated metabolites, was generated (Additional file 2: Table S4). Among the annotated metabolites, 275 were confirmed by standards, and 211 were putatively annotated as described previously, including flavonoids, phenolic acids, organic acids, lipids, alkaloids, tannins, amino acids, nucleotides, and others (Additional file 1: Fig. S1a). Over 65% of the metabolites exhibited a coefficient of variation (CV) greater than 50%, indicating significant variation between apple accessions (Additional file 1: Fig. S1b). Of these, flavonoids revealed the highest CV with an average of up to 349%, followed by phenolic acids and lipids with an average of 202 and 201%, respectively (Additional file 2: Table S4). PCA analysis of all metabolites divided the apple accessions into four groups, namely other wild, M. sieversii, heirlooms, and cultivars, where the other wild group was separated from the other three groups (Fig. 2a). The cultivars were further grouped into six clades, namely “Jonathan,” “Gala,” “Golden Delicious,” “Starking Delicious,” “Ralls Janet,” and “Fuji” pedigrees, respectively (Additional file 1: Fig. S2). The results obtained from metabolite differentiation were generally consistent with the phylogenetic tree relationships, indicating a strong relationship between changes in metabolite and genomic variation of apples.

Fig. 2
figure 2

Profiling of metabolites among different apple accessions. a PCA analysis of all metabolites in other wild accessions, M. sieversii, heirlooms, and cultivars. b Differentiation of metabolites between M. sieversii/ heirloom and heirloom/ cultivar comparisons. c Box plots for the contents of flavonoids (quercetin 7-O-rutinoside, quercetin-3-O-rutinoside, and quercetin-3-O-robinobioside) and phenolic acids (isochlorogenic acid A, isochlorogenic acid B, and isochlorogenic acid C) with large fold changes between M. sieversii and heirloom groups. d PCA analysis between the “Ralls Janet” and “Golden Delicious” pedigrees, and e between high-weight (≥ 300) and low-weight (≤ 90) domesticated apples. t[1] and t[2] represent the first and second principal component. R2x[1] and R2x[2] indicate the proportion of variance explained by t[1] and t[2], respectively. The colored dots indicate different apple accessions and the ellipses indicate the 95% confidence interval for each data set

Full size image

A total of 595 and 363 different metabolites were identified in the comparison between M. sieversii and heirloom and between heirloom and cultivar (Additional file 2: Tables S5 and S6), respectively, revealing the metabolome variation in apples during domestication and improvement [26]. Among these, the levels of most flavonoids, phenolic acids, organic acids, and tannins decreased when the M. sieversii transitioned into heirloom and from heirloom to cultivar (Fig. 2b). However, most flavonoids such as quercetin-3-O-robinobioside, quercetin-3-O-rutinoside (rutin), and quercetin 7-O-rutinoside and phenolic acids such as isochlorogenic acid A, isochlorogenic acid B, and isochlorogenic acid C exhibited large fold change (FC) of up to 20 and 10, respectively, between M. sieversii and heirloom. However, the value was lower in comparison between heirloom and cultivar, with an average of 3 (Fig. 2c). It is well known that tannins and organic acids contribute to an astringent property and acid taste in fruits, while flavonoids and phenolic acids have been associated with the astringent property and bitter taste in fruits [30,31,32]. Consequently, it can be deduced that acidity, astringency, and bitter tastes are the primary characteristics selected for removal during domestication.

Improved storage life is another selection criterion for fruit quality by the breeders. Fruits of the “Ralls Janet” pedigree are characterized by durable storability and firm flesh [33], which can be passed on to offspring, while fruits of the “Golden Delicious” pedigree soften during storage [34]. The comparison of all metabolites in the “Ralls Janet” and “Golden Delicious” pedigree divided them into two distinct groups (Fig. 2d). The two groups differed in 361 metabolites, of which 64 were annotated. Interestingly, seven lysolipids were accumulated to significantly higher levels in the “Ralls Janet” pedigree than in the “Golden Delicious” pedigree (Additional file 2: Table S7). Of these, lysophosphatidylethanolamine (LPE) 18:1 and LPE 18:1 (2n: isomer) were highly accumulated, with FC of up to 4.39 and 3.68, respectively. Lysolipids have been reported to improve cell membrane stability [35, 36] and mediate hormonal regulation of growth and senescence [37, 38]. Thus, the accumulation of lysolipids, particularly LPE 18:1, could potentially enhance the apple’s storage life.

Fruit weight is an important agronomical trait for many domesticated crops. Accordingly, the potential metabolites that could affect domesticated apple fruit weight were investigated. PCA analysis of metabolites divided the domesticated apples into high- and low-weight groups (Fig. 2e). The two groups differed in 805 metabolites, of which 181 were annotated (Additional file 2: Table S8). Interestingly, the phytohormones abscisic acid (ABA) and salicylic acid (SA) accumulated significantly higher levels in low-weight apples, with FC of 3.69 and 1.90, respectively. Moreover, 14 soluble tannins and eight organic acids were significantly accumulated in low-weight fruits, with an average FC of up to 4.54 and 3.75, respectively (Additional file 2: Table S8). The total tannin and titratable acidity decreased with an increase in weight (Additional file 1: Fig. S3), suggesting that tannin and organic acid content were inversely correlated with the evolution of enhanced fruit weight.

Genetic basis of metabolite content in different apple accessions

We then performed mGWAS for 270 apple accessions using a linear mixed model (LMM) approach implemented in GEMMA to reduce false positives by considering genome-wide patterns of genetic correlation. A total of 222,877 significant SNPs associated with 2205 apple metabolites were detected (Additional file 2: Table S9). Manhattan plots of significant SNPs of metabolites, including tannins, phenolic acids, organic acids, lipids, flavonoids, alkaloids, amino acids, nucleic acid derivatives, others, and unknown metabolites are illustrated (Fig. 3). Furthermore, an overlap of the co-mapped region of flavonoids, organic acids, tannins, and phenolic acids from 2.84 to 5.01 Mb on chromosome (Chr) 16 was investigated, suggesting this region is important for improving taste, quality, and nutrition in apples.

Fig. 3
figure 3

Genome distribution of significant SNPs related to all metabolites. The circles from outside to center represent amino acids and nucleic acids (a), alkaloids (b), flavonoids (c), lipids (d), organic acids (e), others (f), phenolic acids (g), soluble tannins (h), and unknown metabolites (i). The unit for chromosome length is Mb. The P-value of all significant SNPs associated with metabolites are listed in Additional file 2: Table S9

Full size image

Domestication based on fruit taste and its underlying biosynthetic network

Tannins, including procyanidin A1, procyanidin A2, procyanidin B4, procyanidin B5, and procyanidin C1, were accumulated to higher levels in the M. sieversii group than in heirlooms and cultivars (Fig. 4a). From GWAS results, a strong peak on Chr 16 co-mapped with procyanidin A2, procyanidin C1, procyanidin B4, cinnamtannin D1, and arecatannin B1 was identified (Additional file 1: Fig. S4), estimating a candidate region from 1.55 to 5.57 Mb harboring one putative gene (Additional file 2: Table S10). MD16G1048500 contains the consensus sequence defining leucoanthocyanidin reductase (LAR) [39] and is 32 kb from the lead SNP (Chr16:3,372,301) (Fig. 4b). In addition, a strong peak on Chr 15 was identified for procyanidin A2, indicating a candidate region from 1.82 to 4.81 Mb harboring two putative genes (Additional file 2: Table S10). One of these, Myb9-like (MD15G1051400), has the potential to regulate anthocyanin and procyanidin metabolisms (Fig. 4b). A phylogenetic tree of reported R2R3 MYBs revealed that MD15G1051400 was closely related to MdMYB9 and FaMYB9 (Fig. 4c; Additional file 1: Fig. S5), which are involved in procyanidin synthesis by regulating LAR and anthocyanidin reductase (ANR) [40, 41], suggesting that Myb9-like may regulate procyanidin synthesis in apple. Fruit weight quantitative trait locus (QTL), fw1, was co-located with vacuolar acidification-related PH4 (MD15G1051000) [42] as well as Myb9-like from 3.41 to 3.76 Mb on Chr 15, this region was found in the selective sweeps of M. sieversii/ heirloom and M. sylvestris/ heirloom comparisons but not in heirloom/ cultivar comparison (Additional file 1: Fig. S6), the results were consistent with a previous report [26].

Fig. 4
figure 4

Identifying metabolites and genes potentially involved in apple domestication. a The contents of tannins with significant differential accumulation in M. sieversii/ heirloom and heirloom/ cultivar comparisons. b Manhattan plots of procyanidin A2 by GWAS. c A phylogenetic tree for MD15G1051400 and other R2R3 MYB members. d,e qRT-PCR analysis of Myb9-like and LAR (leucoanthocyanidin reductase) in high- and low-procyanidin content apples. f,g Results of transient overexpression of Myb9-like and LAR in apple fruits than empty vector (pGreenII 62-SK) injections. h Nucleotide polymorphism identified in the promoter of LAR. i Box plot for procyanidins contents correlated with C and T in Chr16:3404850. j qRT-PCR analysis of LAR in Myb9-like overexpressing apple fruit. k GUS transactivation assay in apples. The GUS reporter vector contains the LAR promoter, while the effector vector contains Myb9-like. l A simplified metabolic biosynthetic pathway of procyanidins in fruit. ANR, anthocyanidin reductase; Cys-EC, 4β-(S-cysteinyl)-epicatechin. Data are expressed as mean ± sd. “*” indicate p ≤ 0.05 by t-test

Full size image

The expression levels of Myb9-like and LAR were higher in fruits with high procyanidin than those low in procyanidin (Fig. 4d, e). Transiently overexpressing Myb9-like and LAR exhibited a significantly higher level of procyanidin A2 and procyanidin A1 in apple fruits than the controls (Fig. 4f, g). A significant SNP (Chr16:3,404,850) was located at − 315 bp upstream of the LAR gene (Fig. 4h), where allelic mutations of C and T were significantly correlated with the contents of tannins such as procyanidin B4, procyanidin C1, cinnamtannin D1, and arecatannin B1 (Fig. 4i). The allelic mutation in the promoter could potentially lead to differential expression of LAR in apples with a high and low level of procyanidin. Furthermore, transiently overexpressing Myb9-like increased LAR expression in apple flesh (Fig. 4j), and the Myb9-like specifically activated GUS transcription as driven by the LAR promoter (Fig. 4k). LAR enzyme catalyzes leucocyanidin to catechin and converts 4β-(S-cysteinyl)-epicatechin (Cys-EC) to epicatechin [43, 44]. Thus, it is inferred that Myb9-like might participate in procyanidin synthesis by promoting LAR expression (Fig. 4l).

Improvement of metabolites related to the firmness of fruits

Seven lysolipids, in particular LPE 18:1 implicated in firmness and storage life of fruits [45], accumulated to higher levels in the “Ralls Janet” pedigree than in the “Golden Delicious” pedigree (Fig. 5a). Based on GWAS results, a strong peak on Chr 12 co-mapped for LPE 18:1 and LPE 18:1 (2n isomer) was identified, with estimated candidate region from 5.93 to 10.01 Mb, harboring three putative genes (Additional file 1: Fig. S7; Additional file 2: Table S10). One of these, MD12G1057600, which encodes fatty acid desaturase 2 (FAD2), is 0.72 kb from the lead SNP (Chr12:6,578,653) (Fig. 5b). A phylogenetic tree of the well-characterized Arabidopsis FAD family demonstrated that MD12G1057600 was closely related to AtFAD2 (Fig. 5c), which converts LPE precursor PE 18:1 to PE 18:2 [46], suggesting that the activity of FAD2 may reduce LPE 18:1 accumulation.

Fig. 5
figure 5

Functional interpretation of GWAS for lysophosphatidylethanolamine (LPE) 18:1 content in apple. a Lysolipid contents of apples in the “Ralls Janet” (RS) and “Golden Delicious” (GD) pedigrees. b Manhattan plots of LPE 18:1 content by GWAS. c An unrooted phylogenetic tree for MD12G1057600 and Arabidopsis FAD members. d The firmness of apples in “RS” and “GD” pedigrees. e Correlation analysis (R2) between firmness and LPE 18:1 in domesticated apples. f qRT-PCR analysis of FAD2 transcripts in domesticated fruits with high and low LPE 18:1. g Transient overexpression of FAD2 in apple fruits. h,i The firmness and respiration rate of apples treated with LPE 18:1 during storage. j Nucleotide polymorphisms identified in the promoter and coding sequence of FAD2. k Box plots of LPE 18:1 correlated with identified nucleotide polymorphisms. l Nucleotide polymorphisms in the “RS” and “GD” pedigrees. The data are expressed as mean ± sd. “*” indicates p ≤ 0.05 by t-test. The correlation was evaluated based on the Pearson correlation coefficient

Full size image

Consistent with their enhanced storage life, apples from “Ralls Janet” pedigree exhibited higher firmness than the “Golden Delicious” pedigree (Fig. 5d). A close correlation between the content of LPE 18:1 and the firmness of cultivated apple accessions was investigated (Fig. 5e). Furthermore, a significant difference was observed in transcript levels of FAD2 in apple accessions with high and low LPE 18:1 content (Fig. 5f). Transient overexpression of FAD2 in apple fruit led to a decreased level of LPE 18:1 than that of the control (Fig. 5g). To test the effects of LPE 18:1 on fruit firmness, apples were sprayed with LPE 18:1 before storage. Compared with the control, LPE 18:1 maintained apple firmness at 3 and 7 days of storage (Fig. 5h). Also, the respiration rate in LPE 18:1-treated fruit was reduced during storage (Fig. 5i). Furthermore, three SNPs were significantly correlated with LPE 18:1 content. Chr12:6582426 and Chr12:6582271 located at − 1890 and − 1745 bp upstream of FAD2, respectively, and Chr12:6578544 located at 615 bp in the coding sequence might induce a non-synonymous amino acid change in FAD2 (Fig. 5j). Allelic mutations that correlated with high and low LPE 18:1 content were mainly found in the “Ralls Janet” and “Golden Delicious” groups, respectively (Fig. 5k, l). Thus, FAD2 is a strong candidate for suppressing the accumulation of LPE 18:1, which was potentially maintaining the firmness and prolonging the storage life of apples.

Metabolome alterations related to fruit weight targeted selection

The phytohormones ABA and SA, involved in cell division and elongation of plants [47, 48], were accumulated to high levels in low-weight fruits (Fig. 6a–c). Based on GWAS results, a strong peak associated with SA contents was identified on Chr 16 (Fig. 6d), with an estimated region from 2.70 to 4.99 Mb harboring three putative genes (Additional file 2: Table S10). MD16G1069500 encoding a transcription factor, NAC-like activated by Apetala3/Pistillata (NAP), was identified in this region. Based on a phylogenetic tree of reported NACs, MD16G1069500 was closely related to tomato SlNAP1 and SlNAP2 (Fig. 6f), which were reported to regulate SA biosynthesis and fruit size in tomatoes [49, 50]. In addition, a strong peak on Chr 3 was identified, which was associated with ABA content, with an estimated candidate region from 30.57 to 32.58 Mb, harboring two putative genes (Additional file 2: Table S10). MD03G1222900 contains the consensus sequence of an ABA transporter, ATP binding cassette G25 (ABCG25), which controls the ABA level in Arabidopsis [51] (Fig. 6e).

Fig. 6
figure 6

Functional analysis of GWAS for SA and ABA content related to apple fruit weight. Fruit weight (a), SA (b), and ABA (c) contents in high- (H) and low- (L) weight domesticated fruits. d,e Manhattan plots of SA and ABA contents by GWAS. f An unrooted phylogenetic tree was constructed for MD15G1051400 and other NAC members. Fruit weight (g), SA (h), and ABA (i) contents during fruit development. j,k qRT-PCR analysis of NAP (MD16G1069500) and ABCG25 (MD03G1222900) in H and L apple groups. l,m Results of transient overexpression of NAP and ABCG25 in apple fruits. n Nucleotide polymorphisms identified in the promoter and coding sequence of NAP. o Box plot of SA and ABA correlated with identified nucleotide polymorphism. p Nucleotide polymorphisms in H and L apple groups. DAFB, days after full bloom. Data are expressed as mean ± sd. “*” indicate p ≤ 0.05 by t-test

Full size image

Consistent with its role in restricting the increase in fruit weight, SA content decreased during fruit development, while ABA content decreased from 40 to 100 days after full bloom (DAFB) and increased to 120 DAFB, possibly due to the role of ABA in fruit ripening (Fig. 6g–i) [52]. The transcript level of NAP was significantly higher in low-weight apples than in high-weight apples, which correlated with the SA contents in high- and low-weight groups (Fig. 6j). Transiently overexpressing NAP in apple fruits indicated that SA content was significantly increased, which further supports the role of NAP in SA synthesis (Fig. 6l). The transcript level of ABCG25 was significantly higher in low-weight apples, which correlated with the ABA content in high- and low-weight apples (Fig. 6k). Also, apple fruits transiently overexpressing ABCG25 contained a significantly higher ABA content than the controls (Fig. 6m). Two significant SNPs (Chr16:4903371 and Chr16:4903717) located upstream, and Chr16:4905296 located at 898 bp in NAP may lead to a non-synonymous amino acid change within this gene (Fig. 6n). Additionally, the Chr3:30773716 located at 2865 bp in ABCG25 may lead to a non-synonymous amino acid change in ABCG25. Allelic mutations that led to low and high SA and ABA contents were mainly found in the high- and low-weight groups, respectively (Fig. 6o, p). Thus, it is concluded that accumulation of SA and ABA, which appear to be affected by the genes, NAP and ABCG25, respectively, may restrict fruit enlargement in apples.

Discussion

The metabolome defines what we ingest from a plant product and determines its attractiveness and nutritional value. Plant metabolites play important roles in growth, cellular replenishment, whole-plant resource allocation, and the adaptation of plants to the environment [53, 54]. Consequently, the metabolome is a direct reflection of the physiological status and a connection between the genotype and the phenotype of a plant [55]. Given the importance of plant metabolism to plant development and for human health, many genetic studies have been conducted to identify metabolic pathways in apples [56], tomatoes [21], peach [22], rice [23], jujube [24], and other crops. With increasing genome resources, GWAS has been conducted on apples to clarify important agronomical traits that contribute to yield and quality [7, 26]. In the current study, mGWAS data analysis aimed at understanding the effects of domestication and improvement on the metabolic composition of apples was provided. A total of 222,877 significant SNPs were identified that are associated with 2205 apple metabolites. This information provides a valuable resource for further studies on biosynthetic pathways involved in determining the contents of metabolites in apple fruit and provides a reference for metabolic pathway elucidation to other crops. An interesting overlap of the co-mapped region of flavonoids, organic acids, soluble tannins, and phenolic acids from 2.84 to 5.01 Mb on Chr 16 was investigated. Based on previous reports, several important genes or QTLs related to apple fruit quality have been identified in this region [26, 57,58,59]. In our study, a NAC transcription factor NAP potentially related to an increment in fruit weight was identified at a 4.90 ~ 4.91 Mb region on Chr 16. Thus, the mGWAS data identified a region on Chr 16 as having great value for apple quality and nutrition breeding.

Apple accessions were divided into two major groups based on genomic and metabolomic criteria that distinguished between the wild and cultivated apples. Although many omics studies have been conducted on crop species and their progenitors [60, 61], very few researches have been performed at the metabolic level that allows evaluation of the impact of domestication on metabolites diversity and their contents [62]. In recent studies, the impact of domestication on the fruit metabolome in tomatoes [21], the combined nature and human selections on peach fruit metabolome [22], and the variation in primary metabolism in a diverse lettuce population were investigated [63]. A previous study suggested that cultivated apples were domesticated from M. sieversii in Central Asia [12]. In this study, the genomes and metabolomes of M. sieversii and heirloom (domestication), and between heirloom and cultivar (improvement) were evaluated [26], and the question was addressed of how domestication and improvement have altered fruit chemical composition. This study provided a direct evaluation of the chemical compositional consequences of apple domestication and improvement at such a scale.

Apple domestication resulted in the reduction of certain compounds that produce adverse effects on taste, including tannins [30], organic acids [10, 64], flavonoids, and phenolic acids [32] but not sugars, which are consistent with previous reports [26]. Also, organic acid content has been identified as the key criterion for the evolution of taste during domestication in citrus [65], lettuce [63], jujube [66], and pear [67], which showed both distinct and overlapping aspects of genetic control of metabolism between species. Flavonoids have been associated with astringency and bitter taste in fruit [31, 32], which characteristics are to be removed during domestication. Other research has also suggested that the flavonoids is excluded from selection during the domestication of citrus, tomato, rice, wheat, soybean, and barley [68, 69]. Interestingly, the tannins and organic acids were significantly accumulated at different levels based on the comparisons between the heirloom and M. sieversii groups (Fig. 2a, b) and between high- and low-weight domesticated apples (Fig. 2e), which were also found to be negatively correlated with fruit weight (Additional file 1: Fig. S3). Furthermore, the locus for the fruit weight, fw1, was located in a similar region with acidity and tannin-related genes, e.g., PH4 and Myb9-like, from 3.41 to 3.76 Mb on Chr 15. It was also shown in tomato that selection for alleles of genes associated with larger fruits altered metabolite profiles as a consequence of linkage with nearby genes [21]. However, fruit acidity and astringency are genetically controlled by other major genes such as Ma1 and LAR on Chr 16, whereas fruit weight is regulated by multiple genes [26]. Thus, fruit enlargement may partially affect the acidity and astringency in apples.

The control of firmness has been a major goal for improving fruit storage and shelf life [70, 71]. The loss of turgor pressure in fruits leads to a decrease in firmness [72], in addition to changes in the cell wall and the action of antioxidants [73]. The integrity of the fruit cell membranes can be damaged during storage, particularly at low temperatures, causing the loss of turgor pressure. We demonstrated that lysolipid LPE 18:1 is potentially maintaining the firmness of apple, with the “Ralls Janet” pedigree identified as a variety with firm flesh, prolonged storage life, and relatively high LPE 18:1 content (Fig. 5a, d). Previous reports have depicted that unsaturation of fatty acids in lipids, such as LPE and lysophosphatidylcholine (LPC), could improve cell membrane stability [35,36,37, 74] and mediate hormonal regulation of growth and senescence [37, 38], while saturated fatty acids produce the opposite effect [75, 76]. The ratio of unsaturated to total fatty acids has been reported greater in “Ralls Janet” than in “Starking Delicious,” which increases as the storage temperature decreases [77]. Other research has also found that pre- and postharvest LPE treatments promote fruit color and enhance shelf life [45]. Particularly, the LPE-treated banana had higher pulp firmness, lower ion leakage from peel tissue, a thicker peel, and better shelf life as compared to the control [78]. Thus, it can be concluded that fruits with a higher content of lysolipids, particularly LPE 18:1, have better storage and shelf life.

A comparison of the metabolites in low- and high-weight domesticated groups demonstrated that the increment of SA and ABA is associated with fruit enlargement. The growth phase of fruit involves an active period of cell division, followed by a cell expansion phase, and a combination of these determines the maximum final fruit size [79]. Recent studies have reinforced the importance of SA and ABA in regulating plant cell division or organ enlargement. For example, the dwarf phenotype was identified in Arabidopsis mutants with high SA content, such as constitutive pathogen response (5cpr5) [80], accelerated cell death (acd6-1) [81], and aberrant growth and death (agd2) [82]. ABA has been reported to be crucial in inhibiting cell division in plants and tissues [48, 83, 84]. Our results demonstrate that smaller fruits contain higher SA and ABA contents than bigger fruits. ABA was also found to promote ripening in some fruits [85], suggesting the possible dual roles of ABA and SA in regulating fruit size, development, and ripening. We also found evidence that the NAP transcription factor is associated with fruit size in apples based on the accumulation of SA. Previous research reported that tomato fruits that overexpressed SlNAP2 were smaller than the wild-type despite the increase in the fruit yield [49]. Analysis of different hormones revealed higher SA and ABA contents in the SlNAP1 overexpressing tomato plants. Moreover, SlNAP1 was observed to directly activate the expression of genes in both SA and ABA biosynthetic pathways [50], implying that NAP may affect both SA and ABA synthesis in apples. However, the mechanisms involved in interactions between SA and ABA and their role in fruit enlargement requires further research.

In addition to providing information on metabolic divergences in well-characterized pathways, the data can also provide insight into unknown pathways. Metabolomic or sequencing data alone can provide valuable information on plant metabolism, but the combination of these two data can create a synergistic effect, which is beneficial for the discovery of metabolic pathways [86]. In particular, these approaches can reveal novel metabolites and genes that are involved in plant metabolism. These data have also laid the foundation for cross-species comparative approaches that can provide insights into the evolution of metabolism in other crops [87]. However, we should also understand that although the multiomics method provides large data, most of them are difficult to utilize currently. For example, of the 2575 distinct metabolites identified, only 486 (18.87%) could be annotated (Additional file 2: Table S4). Therefore, the characterization of these chemicals through biochemical approaches is essential for the identification of the underlying candidate genes, which will be possible to construct novel key biological pathways of apples in the future.

Conclusion

Metabolite changes in apple accessions have accompanied human-guided evolutionary selection processes. We have associated significant SNPs with apple metabolites. Important flavor compounds, namely organic acids, soluble tannins, flavonoids, and phenolic acids, were co-mapped to a 2.84 ~ 5.01 Mb region on Chr 16. In addition, the fruit weight locus was co-mapped with organic acid and tannin-related genes in a 3.41 ~ 3.76 Mb region on Chr 15. LPE 18:1, which contents were reduced by the activity of FAD2 and found higher in the “Ralls Janet” pedigree than the “Golden delicious” pedigree, was also found beneficial for maintaining firmness and storability. The levels of two plant hormones, SA and ABA, with higher levels detected in smaller fruits (Fig. 6a–c), and their accumulation was affected by NAP and ABCG25, respectively. In summary, our results provide a metabolic perspective of selection for fruit quality related to domestication and improvement in apples. The information offered in this article can be used to investigate the mechanisms underlying apple metabolite content and quality attributes.

Materials and methods

Fruit material and treatment

A total of 292 accessions with a wide range of geographic origins (Additional file 2: Table S1) were used in this study, including 270 collected from the apple germplasm repository in Liaoning, China, and 22 wild accessions reported previously [7, 26]. In total, 217 cultivars, 38 heirlooms, 11 M. sieversii, seven M. sylvestris, and 19 accessions of other wild species, including three M. baccata, three M. sieboldii, seven M. spectabilis, and six M. toringoides (Additional file 2: Table S2), were investigated in this study. The fruit flesh was frozen in liquid nitrogen and stored at − 80 °C.

For developmental metabolites analysis and postharvest treatment, “Hongyu” and “Molisi” apples were collected from an orchard in Beijing, China. The “Hongyu” apples were collected at 40, 80, 100, and 120 DAFB. For LPE 18:1 treatment, “Molisi” apples at the commercial ripening stage were randomly divided into two groups and were separately dipped in water (as a control) and 100 mg L−1 of LPE 18:1 for 10 min. The physical data for “Molisi” were measured at 0, 3, 7, 14, 21, and 28 days of storage. All the fruits were stored at 20 ± 1 °C with 90% relative humidity. Each treatment involved 120 fruit samples with three replicates.

Physiological index measurements

Firmness was evaluated by a texture analyzer (TA.HD plus, Stable Micro Systems, UK) with a probe of 7.5 mm diameter, and results were expressed in Newton’s (N) force. The measurement speed and distance were set at 2 mm s−1 and 10 mm, respectively. Two apples were sealed in a 1.5-L chamber for 2 h at 20 °C; then, the CO2 level was monitored by a gas analyzer (F950, FELIXD, USA) to calculate the respiration rate presented as mL kg−1 h−1. To measure titratable acid, 1 g of flesh powder was mixed with 10 mL of water and centrifuged for 10 min at 12,000 × g at 4 °C. The supernatant was titrated to pH 8.2 with 0.01 mol L−1 of sodium hydroxide. ABA and SA contents were measured by enzyme-linked immunosorbent assay (ELISA) kits (Welab, China). Each measurement contained three biological replicates.

Metabolic profiling

Apple flesh was freeze-dried and ground for 1 min at 30 Hz using a mixer mill (MM 400, Retsch, Germany) with zirconia beads. A total of 80 mg of flesh powder was homogenized with 1 mL of 70% aqueous methanol at 4 °C overnight. After centrifuging for 10 min at 12,000 rpm, supernatants were filtered using a filter with a pore size of 0.22 μm (SCAA-104, ANPEL, China). The samples were measured by a high-performance liquid chromatography (HPLC)-electrospray ionization (ESI)-mass spectrometer (MS)/MS system (HPLC, Shim-pack UFLC Shimadzu CBM30A system, China; MS, Applied Biosystems 6500 plus Q TRAP, China). For HPLC system, Waters ACQUITY UPLC HSS T3 C18 (1.8 μm, 2.1 mm × 100 mm) column was used. Water (0.04% acetic acid) and acetonitrile (0.04% acetic acid) were used as a solvent system. The program was set as follows: 95:5 v/v at 0 min, 5:95 v/v at 11.0 min, 5:95 v/v at 12.0 min, 95:5 v/v at 12.1 min, 95:5 v/v at 15.0 min at a flow rate of 0.35 mL min−1, the temperature of 40 °C, and injection volume of 2 μL. The effluent was connected to an ESI-triple quadrupole-linear ion trap (Q TRAP) MS.

Linear ion trap (LIT) and triple quadrupole (QQQ) scans were collected on a Q TRAP MS using an API 6500 Q TRAP LC/MS/MS System equipped with an ESI Turbo Ion-Spray interface, operated in a positive ion mode and controlled by Analyst 1.6 software (AB Sciex, Germany). Operation parameters of the ESI source were as follows: turbo ion spray; source temperature of 550 °C; ion spray voltage (IS) of ( +) 5500 V and ( −) 4500 V; ion source gas I (GSI), gas II (GSI), and curtain gas (CUR) were set at 55, 60, and 25 psi, respectively; and collision gas (CAD) set at medium. Instrument tuning and mass calibration were performed using 10 and 100 μmol L−1 polypropylene glycol solutions in QQQ and LIT modes, respectively. QQQ scans were obtained from multiple reaction monitoring (MRM) experiments with the collision gas (nitrogen) set to 5 psi. By further decluttering potential (DP) and collision energy (CE) optimization, the DP and CE for individual MRM transitions were performed. A specific set of MRM transitions was monitored for each period based on the metabolites eluted during this period.

Genome sequencing

Genomic DNA was obtained using a Plant DNA Mini Kit (Aidlab Biotech Ltd., China). After constructing the DNA library, an Illumina HiSeq XTen/NovaSeq/BGI platform was used for sequencing, with 150 bp read lengths, conducted by a credible service provider (Biomarker Technologies, China). High-quality sequences were obtained by removing reads containing the sequencing adapters or that of low quality based on the following criteria: pair-end reads with > 10% “N” bases or reads with more than 50% of the bases have a Phred-like score < 20.

SNP calling

All clean reads of 292 apple accessions were mapped to the reference genome (GDDH13 v.1.1). The mapping results were sorted, and duplications were removed by rmdup in SAMTools (v.1.9) [88]. The HaplotypeCaller module in GATK v.3.8 was used for calling of SNPs within the whole accessions, which were filtered according to the parameters: QD < 2.0, MQ < 40.0, FS > 60.0, QUAL < 30.0, MQRankSum < -12.5, ReadPosRankSum < -8.0, -clusterSize 2 -clusterWindowSize 5. The high-quality SNPs were further selected with the following criteria: minor allele frequency (MAF) ≥ 5% and missing rate per site ≤ 10%. Based on the reference genome, SNP annotation was conducted by snpEff software [89], and SNPs were classified as intergenic, upstream or downstream regions, and exons or introns. SNPs identified in coding exons were further categorized into synonymous or non-synonymous SNPs.

Population genetics analysis and selective sweeps identification

The phylogenetic tree was constructed from the distance matrix in PHYLIP software v.3.69 with 1000 bootstrap replicates and was visualized by iTOL [90]. PCA of whole-genome SNPs was performed with EIGENSOFT 6.0 software smartpca program, and the first three eigenvectors were plotted in two/ three dimensions [91]. Population structure was modeled by ADMIXTURE v.1.3.0 [92], and TASSEL v.3.0 [93] was used to calculate kinship value. LD coefficients based on r2 values between pair-wised high-quality SNPs were calculated by PopLDdecay software with default parameters [94]. The potential selective signatures were estimated by the genetic differentiation index (FST) for M. sieversii/ heirloom, M. sylvestris/ heirloom, and heirloom/ cultivar comparisons. The region of sweeps associated with domestication and improvement were selected within the top 5% of windows of FST.

Metabolites-based GWAS analysis

A total of 2,956,396 SNPs were used to carry out GWAS. mGWAS was conducted by the linear mixed model implemented in GEMMA. The genome-wide significance thresholds were determined using the Bonferroni test threshold (p = 1.71E−7).

RNA extraction and quantitative real-time PCR (qRT -PCR)

Total RNA was extracted from the apple flesh using the cetyltrimethylammonium bromide (CTAB) method [95]. cDNA was synthesized, and qRT-PCR was performed on a Fast Real-Time PCR System (Applied Biosystems 7500, USA). The primers are listed in Additional file 2: Table S11.

Transient overexpression

The coding sequences were cloned into NotI/ SmaI sites of a pGreenII 0029 62-SK vector, which were subsequently transformed into Agrobacterium tumefaciens strain GV3101. Agrobacterium cells containing the overexpressed constructs were separately infiltrated into the apples. The opposite sides of each apple were infiltrated with cells containing the empty vector as controls. Tissues around the infiltration sites were collected after 3 days in the chamber.

GUS transactivation assay

The coding sequence of Myb9-like was cloned into the binary pRI101 vector to generate an effector construct. The reporter construct was cloned, according to the LAR promoter sequence, upstream of the GUS reporter gene in the binary vector pBI101. The reporter vector and the effector vector were transformed into A. tumefaciens GV3101. The opposite side of each apple was infiltrated with cells containing the pRI101 vector and reporter construct as control. Tissues around the infiltration sites were collected after 3 days in the chamber.

Statistical analyses

Coefficients of variations (CV) of each metabolite were independently calculated as follows: s/m, where “s” and “m” represent the standard deviation and mean of each metabolite in all accessions, respectively. Differential metabolites between different groups were determined by partial least squares to discriminate analysis (PLS-DA) with the variable importance for projection (VIP) values ≥ 1, followed by both ANOVA (p ≤ 0.05), using SIMCA software package (v.14.1). Significant differences were determined by Fisher’s least significant difference (LSD) test at the level of p ≤ 0.01.