Insights into Carotenoid Biosynthesis Mechanisms in Three Fresh-Consumption Sweetpotato (Ipomoea batatas (L.) Lam.) Cultivars with Distinct Flesh Colors via Integrated Targeted Metabolomic and Transcriptomic Analyses

Abstract

The sweetpotato (Ipomoea batatas [L.] Lam) is a globally significant crop, valued for its nutritional and economic importance. The tuberous roots of the sweetpotato are rich in carotenoids, which contribute to their vibrant colors and health benefits. This study focuses on three elite fresh-consumption sweetpotato cultivars: “Kokei No. 14,” “Xinxiang,” and “Zheshu81” with distinct flesh colors. To elucidate the metabolic pathways and genetic mechanisms underlying carotenoid biosynthesis in the sweetpotato, 20 types of carotenoids were quantified using targeted metabolomic analyses, and the key genes involved in carotenoid synthesis were identified with transcriptomic analyses. The results revealed significant differences in carotenoid content and composition among the cultivars, with “Zheshu81” exhibiting the highest carotenoid levels. Weighted gene co-expression network analysis further highlighted key regulatory genes and transcription factors influencing carotenoid accumulation. This study identifies key transcriptional regulators associated with carotenoid accumulation, sheds light on sweetpotato carotenoid biosynthesis mechanisms, and lays a foundation for breeding to improve its nutritional quality and flesh color.

1. Introduction

The sweetpotato (Ipomoea batatas [L.] Lam.), a tuberous-rooted plant of Convolvulaceae, is one of the most important food and economical crops around the world [1]. The tuberous roots have abundant nutritional compositions, including minerals, flavonoids, anthocyanins, and carotenoids [2,3]. Based on the nutrients and applications, sweetpotato cultivars are always divided into fresh consumption, food processing, and starch production types [1]. Among these, sweetpotato cultivars intended for fresh consumption are distinguished by their excellent eating qualities, attractive appearance, and high yield [4]. The attractive appearance of cultivars for fresh consumption are mainly reflected in the yellow or orange flesh based on the carotenoid content [5]. Orange-fleshed sweetpotato cultivars contain significantly higher levels of carotenoids compared to cultivars with other flesh colors [6]. Carotenoids have become increasingly popular among nutritionists and consumers due to their health benefits, as they have proved effective in alleviating obesity, treating stomach ulcers, and enhancing antioxidant activity [7,8,9]. Some xanthophyll carotenoids concentrate in the center of the retina selectively and exhibit a protective function for the eyes, such as lutein and zeaxanthin, neither of which can be synthesized in the human body [10]. Violaxanthin has demonstrated antiproliferative activity against a human mammary cancer cell line [11]. β-carotene, accounting for most of the total carotenoids in orange-fleshed cultivars, has been reported as the most efficient precursor of vitamin A [12]. Thus, the sweetpotato cultivars for fresh consumption with yellow or orange flesh are usually preferred for treating vitamin A deficiency [13].

The biosynthesis mechanisms of carotenoids, which involve metabolic pathways, genes, plastids, and signals like light, have attracted growing interest among researchers [14,15,16,17]. For the sweetpotato, the carotenoids in the tuberous roots are synthesized in the chromoplasts originated from pre-existing chloroplasts or other non-photosynthetic plastids, including proplastids, leucoplasts, or amyloplasts [18]. Light is closely associated with carotenoid biosynthesis and storage in plants, and photon flux density exerts a significant influence on the total carotenoid content (TCC) in sweetpotato seedlings [19,20]. Carotenoids in plants are produced through the 2-C-methyl-D-erythritol 4-phosphate/1-deoxy-D-xylulose 5-phosphate (MEP/DOXP) pathway, which is part of the isoprenoid biosynthesis process [21]. Transcriptome analyses and genome-wide association studies have been popular methods adopted to excavate novel genes involved in the carotenoid biosynthesis and degradation pathways [22,23]. Metabolic enzymes, including geranylgeranyl pyrophosphate synthase (GGDPS), phytoene synthase (PSY), phytoene desaturase (PDS), ε-carotene isomerase (Z-ISO), ξ-carotene desaturase (ZDS), lycopene ε-cyclase (LCYE), lycopene β-cyclase (LCYB), hydroxylation by β-hydroxylase (CHYB), zeaxanthin epoxidase (ZEP), and neoxanthin synthase (NXS), are involved in the synthesis process of carotenoids. Carotenoid cleavage dioxygenase (CCD) and 9-cis-epoxycarotenoids dioxygenase (NCED) are two key enzymes for carotenoid degradation [24]. Most enzymes in the carotenoid biosynthetic pathways, along with the orange protein, have been characterized in the sweetpotato; modification of their corresponding genes alters the content and composition of carotenoids [25,26,27,28,29]. However, based on the research progress made above, the biosynthetic pathway of carotenoids in sweetpotato tuberous roots has been insufficiently elucidated.

In the present study, to further demonstrate the carotenoid metabolic pathways of the sweetpotato and mine novel genes involved, three elite sweetpotato cultivars for fresh consumption were selected, and the content of 20 types of carotenoids was revealed using widely targeted metabolomic analyses. The underlying metabolic mechanisms of carotenoids in the sweetpotato were further illustrated through the use of transcriptomic analyses. This study will provide valuable insights into the coloration of sweetpotato root flesh, and the results will assist in further efforts in elucidating and developing carotenoids in the sweetpotato.

2. Materials and Methods

2.1. Plant Materials

Three sweetpotato cultivars with different flesh colors, namely “Kokei No. 14,” “Xinxiang,” and “Zheshu81,” were used in this study. These cultivars were cultivated at the Xuzhou Institute of Agricultural Sciences under conventional field conditions and cultivation methods. The seedlings were planted on June 24 and harvested on October 22 after a growth period of 120 days. For each cultivar, tuberous roots with a flawless surface and uniform size (weighing 100–250 g) were selected as experimental materials. The skin and flesh of the selected tuberous roots were photographed after being washed in the same environment (Figure 1A).

2.2. Determination of General Nutrients in the Tuberous Roots

To determine the variation in dry matter content (DMC) of the tuberous roots, three flawless roots were washed, peeled, and sliced into strips. Fifty grams of the fresh root flesh was weighed accurately and transferred into a drying oven. The temperature was first set at 105 °C and maintained for 5 min, then the samples were fully dried at 48 °C for 48 h. The DMC was calculated as the ratio of dry weight to fresh weight. The total carotenoid content in the tuberous roots was determined according to our previous method [30]. After being lyophilized and ground, 0.1 g of the lyophilized powder was blended thoroughly with 25 mL of acetone and kept in the dark for 1 h. Then, the supernatant was used for content evaluation using a spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). The standard equation used in this study was established with standard β-Carotene (CAS 7235-40-7): y = 135x − 0.0008, where y is the absorbance value at 454 nm and x is the carotenoid concentration in the supernatant (mg/mL).

2.3. Quantification of the Saponification Carotenoids in the Tuberous Roots

In this study, the quantification of the saponification carotenoids using liquid chromatography–high-resolution mass spectrometry (LC-MS) was conducted based on the platform by Metware Biotechnology Co., Ltd. (Wuhan, China). Twenty types of saponified carotenoids and the corresponding detailed information are listed in the Supplemental Materials (Table S1 and Figure S1).

For the sample pretreatment, 50 mg of powder was weighed accurately and immersed with 0.5 mL of mixed solution of n-hexane (CNW Technologies GmbH, Dusseldorf, Germany)–acetone (Sinopharm Chemical Reagent Co., Ltd., Shanghai, China) –ethanol (Merck, Darmstadt, Germany) = 1:1:1 (v/v/v). The extract was vortexed for 20 min (at room temperature). The supernatants were collected after centrifugation for 5 min (11,591× g at 4 °C). The residue was extracted with the above procedures once again. The supernatants were homogenized with 0.5 mL of saturated sodium chloride (Rhawn, Shanghai, China) solution, and the upper layer was collected. This collection procedure was repeated twice. Then, the supernatant was evaporated to total dryness and dissolved in 0.5 mL of MTBE completely, with 0.5 mL of 10% KOH (Husheng Reagent, Shanghai, China)-MeOH (Merck) added afterwards. The mixture reacted overnight at room temperature. After the reaction, the mixture was further mixed with 1 mL of saturated sodium chloride (Rhawn) solution and 0.5 mL of MTBE (CNW Technologies GmbH), and vortexed. The upper layer was collected for another two repeats of the procedure. At last, the supernatant was evaporated to dryness and resolved in 100 μL of a mixed solution of MeOH (Merck) –MTBE (CNW Technologies GmbH) = 1:1 (v/v). The solution was then filtered through a 0.22 μm organic filter for the following LC-MS/MS analysis.

LC-MS conditions: (1) Liquid chromatography conditions: ultra-high-performance liquid chromatography (UPLC) system, ExionLC^TM AD (AB Sciex, Foster City, CA, USA); mobile phase: methanol (Merck) –acetonitrile (Merck) = 1:3 (v/v) with 0.01% BHT (Aladdin reagent, Shanghai, China) and 0.1% formic acid (Sigma-Aldrich, St. Louis, MO, USA) (A), methyl tert-butyl ether with 0.01% BHT (B); program: 3–5 min, B volume ratio increased from 0% to 70%, 5–9 min, B volume ratio increased to 95%, 10–11 min, B volume decreased to 0%; column: YMC C30 (3 μm, 100 mm × 2.0 mm i.d) (YMC Co. Ltd, Kyoto, Tokyo, Japan); flow rate, 0.8 mL/min; temperature, 28 °C; injection volume, 2 μL. (2) MS/MS conditions: Linear ion trap (LIT) and triple quadrupole (QQQ) scans were acquired using a triple quadrupole-linear ion trap mass spectrometer (QTRAP^® 6500) + LC-MS/MS system (AB Sciex) equipped with an atmospheric-pressure chemical ionization (APCI) heated nebulizer. The parameters were set as follows: positive ion mode of APCI ion source, source temperature of 350 °C, curtain gas of 25.0 psi. Analyst 1.6.3 software (AB Sciex) was used to control the equipment and process LC-MS data. External standards: β-carotene (CAS: 7235-40-7). Carotenoids were analyzed using multiple reaction monitoring (MRM). Data acquisitions were also performed using Sciex Analyst 1.6.3 software (AB Sciex).

Principal component analysis (PCA) and heatmap visualization were conducted to analyze the carotenoid profiles among the three samples. For PCA, the relative contents of identified carotenoids were subjected to PCA using the prcomp function in R software (version 3.5.1), and the resulting principal components were plotted to assess the clustering patterns among different sweetpotato cultivars. For the heatmap, the relative contents of differentially accumulated carotenoids among the three sweetpotato cultivars were first normalized. Then, using the pheatmap package in R software (version 2.8.0), a hierarchical clustering heatmap was generated, where rows represent different carotenoid compounds, columns represent the sweetpotato cultivars, and the color scale indicates the relative abundance of each carotenoid.

2.4. RNA Sequencing Analyses and Bioinformatics Analyses

Total RNA from the root flesh was extracted with ethanol precipitation and CTAB-pBIOZOL reagents. After being extracted successfully, RNA was dissolved with another 50 μL of DEPC-treated water. Subsequently, the total RNA was quantified using a Qubit fluorometer (Qubit 4.0, Thermo Fisher Scientific, Waltham, MA, USA), and its integrity was assessed with a Qsep400 high-throughput biofragment analyzer (BiOptic Inc., New Taipei, Taiwan, China).

In the construction of mRNA libraries, mRNAs carrying poly(A) tails were firstly enriched using Oligo(dT) magnetic beads. The purified mRNAs were then cleaved into smaller fragments using a fragmentation buffer. Then, the first-strand cDNAs were generated via reverse transcription using random hexamer primers. For the synthesis of second-strand cDNAs, dUTPs replaced dTTPs to introduce dUTPs into the second-strand cDNAs. Following this, sequencing adapters were ligated to the cDNA fragments, and the resulting mixture was purified and selected for fragments ranging from 250 to 350 bp using DNA magnetic beads. The ligated products were further amplified by polymerase chain reaction (PCR) and subjected to another round of purification before being resuspended in nuclease-free water. Finally, the concentration of the constructed library and the fragment size were detected using the fluorescence quantifier and biofragment analyzer, and the library’s effective concentration was precisely quantified using real-time quantitative PCR. After quality checks, the libraries were pooled and sequenced on an Illumina platform by Metware, producing 150 bp paired-end reads. The sequencing process adopted fluorescently labeled dNTPs, DNA polymerase, and primers to simultaneously synthesize and sequence DNA.

For quality control processes, the raw data were first processed to remove adaptors and low-quality reads for clean reads. Then, the GC content and Q20 and Q30 quality scores were quantified. High-quality clean reads were then aligned and mapped to the Ipomoea batatas reference genome (pasi3.fa, accessed from https://sweetpotao.com/download_genome.html on 25 June 2025) using Bowtie and BWA software (version 2.5.4). The reads mapped to each gene were counted using the featureCounts tool from the Subread package (version 2.0.0). RSEM (version 1.2.15) was utilized to calculate the fragments per kilobase of transcript per million mapped reads (FPKM) for each gene. Differentially expressed genes (DEGs) were identified based on the fold change (FC) of FPKM values between different samples. A false discovery rate (FDR) of ≤5% was applied to set the p-value threshold. The screening criteria for DEGs included an absolute log2FC value and the p-value threshold. Finally, functional enrichment analyses of the DEGs were performed using the clusterProfiler R package, focusing on Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways.

The heatmap of the carotenoid metabolic pathway was generated using z-score standardized FPKM values referencing KEGG pathway ko00906 (https://www.kegg.jp/pathway/ko00906 accessed on 28 June 2025) and other studies on the metabolic pathways of carotenoid in plants [31,32]. This process involved normalizing the FPKM values through z-score transformation to ensure comparability across different samples, after which the standardized values were used to construct the heatmap for a visual representation of the data patterns.

For the quantitative reverse transcription polymerase chain reaction (qRT-PCR), the total RNA was first extracted from the tuberous roots of the three cultivars with the instructions of Ultrapure RNA kit (CwBio, Inc., Beijing, China), and the cDNA was obtained using UnionScript First-strand cDNA Synthesis Mix for qPCR (Genesand Biotech Co., Ltd., Beijing, China). Gene-specific primers were designed using Primer-BLAST (https://www.ncbi.nlm.nih.gov/tools/primer-blast/ accessed on 30 June 2025), a tool for locating specific primers, based on the sequencing data (Table S2). Using the sweetpotato actin gene as the internal reference gene, the qRT-PCR was performed following the procedure of GS AntiQ qPCR SYBR Green Fast Mix (Genesand Biotech Co., Ltd., Beijing, China) on QuantStudio^TM 6 Flex system (Applied Biosystems^TM, Thermo Fischer Scientific, Waltham, MA, USA). The amplification was performed under the following protocol: initial denaturation at 95 °C for 30 s; 40 cycles of 95 °C for 10 s and 60 °C for 30 s. Gene expression quantification was achieved through the 2^−ΔΔCt method with normalization to the internal reference genes.

To elucidate potential regulatory genes associated with carotenoid accumulation, DEGs exhibiting significant variations across the three cultivars were subjected to integrative analysis with corresponding carotenoid content profiles. Weighted gene co-expression network analysis (WGCNA) was implemented using the R package (version 4.1.2) WGCNA, incorporating expression data from 49,852 genes with normalized FPKM values and quantitative carotenoid measurements across multiple samples [33]. Network visualization and module characterization were subsequently performed using Cytoscape software (version 3.9.1) to facilitate topological analysis of the co-expression networks [34].

2.5. Statistical Analysis

The experimental data reported in this study are mean values calculated from three independent biological replicates and standard deviations. The correlation and difference analyses of the data were conducted using Microsoft Excel 2019 (Seattle, WA, USA) and IBM SPSS Statistics 19 (Armonk, NY, USA). Images were processed with OriginPro 9.0 (Northampton, MA, USA) and Photoshop CS6 (San Jose, CA, USA).

3. Results and Discussion

3.1. General Phenotypes of the Tuberous Roots from the Three Elite Cultivars for Fresh Consumption

Three elite sweetpotato cultivars for fresh consumption with different flesh colors were selected in this study. “Kokei No. 14,” a famous and representative Japanese cultivar for fresh consumption, has been proved to be suitable for baking with high maltose content [35]. “Xinxiang,” an early-maturing sweetpotato cultivar, is popular for fresh consumption in China [36]. “Zheshu81” is known for its high carotenoid content in the root flesh [37]. The appearance and root flesh of the tuberous roots from the three cultivars were photographed and compared (Figure 1A). Apart from the similar red skin color, cream, yellow, and orange colors were observed for the flesh of the three cultivars “Kokei No. 14,” “Xinxiang,” and “Zheshu81,” respectively.

“Kokei No. 14” exhibited a significantly higher DMC compared to “Xinxiang” and “Zheshu81,” with no significant DMC difference observed between “Xinxiang” and “Zheshu81” (Figure 1B). According to the previous report, there was minimal association between the flesh color and the DMC of sweetpotato tuberous roots [38]. The authors did, however, report that orange-fleshed cultivars typically had a DMC ranging from low to medium, whereas cream-fleshed cultivars generally had a higher DMC. This proposition aligns with our results, in which the cream-fleshed cultivar exhibited a higher DMC than the yellow- and orange-fleshed ones. The orange-fleshed cultivars were always characterized with high carotenoid contents and low DMCs [39]. Therefore, it was reasonable to notice that the orange-fleshed “Zheshu81” had the highest carotenoid content, while “Kokei No. 14” had the lowest (Figure 1C). Furthermore, a negative correlation was also observed between the DMCs and carotenoid contents, which was generally accepted, as the cream-fleshed cultivar “Kokei No. 14” exhibited both the highest DMC and the lowest carotenoid content among the three cultivars [39].

3.2. The Contents and Compositions of Carotenoids Among “Kokei No. 14,” “Xinxiang,” and “Zheshu81”

The diversity in carotenoid composition and content has been widely recognized to influence the color and nutritional value of plant tissues [40,41,42]. In our previous research, metabolomic analyses have been utilized for phytochemicals quantification in two sweetpotato cultivars for fresh consumption [30]. To further excavate the composition differences of carotenoids among the three cultivars in the present study, the saponification carotenoids using the LC-MS method were identified and quantified based on the established carotenoids bank (Table S1; Figure S1). Seven carotenes and thirteen xanthophylls were contained in the data bank, while only thirteen of them were detected in the three cultivars, including four carotenes and nine xanthophylls (Table S3). PCA results based on differentially accumulated carotenoids showed that the first principal component explained 88.63% of the total variance and separated the three cultivars completely, indicating that the data were distinguished based on the cultivar differences (Figure 2A). The heatmap also showed that the carotenoids of the three replicates for each sample showed a similar accumulation trend, and the clustered samples claimed the repeatability of the three replicates. The obvious differences among the three groups of samples further distinguished the three cultivars (Figure 2B).

Among the differentially accumulated carotenoids, three were detected in all three cultivars, including β-carotene, lutein, and violaxanthin (Figure 2B; Table S3). “Zheshu81” had the highest contents of the three carotenoids, and “Kokei No. 14” the lowest, indicating a positive correlation between the content and the flesh color. This trend was proved to be in accordance with orange cultivars [41]. Meanwhile, there were four xanthophylls (echinenone, β-cryptoxanthin, zeaxanthin, and antheraxanthin) detected in both “Xinxiang” and “Zheshu81,” and “Zheshu81” showed significantly higher content of each xanthophyll than that of “Xinxiang.” Furthermore, there were three carotenes (γ-carotene, (E/Z)-phytoene1, and phytofluene) and two xanthophylls (canthaxanthin and 8′-apo-β-carotenal) detected only in “Zheshu81.” Therefore, it seemed that the flesh colors were determined by not only the TCC but also the content of various carotenoid compositions. Apart from β-carotene, the content of various carotenoid compositions was also correlated with the flesh color [43]. However, it was noteworthy that capsorubin was detected only in “Xinxiang,” while none was identified in “Kokei No. 14” or “Zheshu81.” It was reported that capsorubin significantly counteracted UVB (ultraviolet radiation b) -induced cytotoxicity and decreased the formation of DNA strand breaks under UVB irradiation [44]. Thus, while “Xinxiang” had a lower total carotenoid content compared to “Zheshu81”, it possessed advantages in terms of carotenoid composition: capsorubin, a key carotenoid in “Xinxiang”, is applicable as a dietary supplement to improve natural photoprotection.

3.3. Transcriptome Analyses Among “Kokei No. 14,” “Xinxiang,” and “Zheshu81”

To elucidate the molecular mechanisms underlying the differential accumulation patterns of carotenoids among the three sweetpotato cultivars, RNA sequencing (RNA-seq) was performed in the present study to analyze the transcriptomic profiles of genes associated with carotenoid metabolism. The flesh samples from “Kokei No. 14,” “Xinxiang,” and “Zheshu81” were collected at the harvest stage and were used for the RNA-seq analysis. After the adapter sequences and low-quality reads had been filtered out, an average of 64.51 million clean reads from “Kokei No. 14,” 54.48 million from “Xinxiang,” and 62.12 million from “Zheshu81” were obtained as high-quality reads, respectively (Table S4). With regard to base composition analysis, the lowest guanine–cytosine (GC) content across all samples was 43.50%, with a Q30 score of 93.90%. All clean reads were aligned to the sweetpotato reference genome (available at https://sweetpotao.com/download_genome.html, last accessed on 25 June 2025), with mapping rates ranging from 77.59% to 84.56%. Moreover, the percentage of unigenes that aligned with the reference genome of all the clean reads ranged from 71.65% to 80.04%. Pearson correlation coefficients are statistical measures widely used to quantify the strength and direction of the linear relationship between each pair of variables [45]. The correlation coefficients among all the samples were calculated and compared (Figure 3A). The correlation coefficients among the replicates of the same cultivar were all 0.99, while all of the coefficients between cultivars were above 0.89. The transcriptome assembly results above demonstrated the reliability of the RNA sequencing data obtained in this study, confirming their suitability for subsequent analyses.

The DEGs among cultivars were first analyzed using transcriptomic analyses in this study. After analyzing the DEGs using DESeq2 with screening criteria of |log2FC| ≥ 1 and FDR < 0.05, a total of 15016 DEGs were screened out, and the hierarchical cluster analyses were carried out (Figure 3B; Table S5). The large number of DEGs is mainly attributed to differences in genetic background among the three cultivars. The smallest number of DEGs was found between “Xinxiang” and “Kokei No. 14,” totaling 6883. In contrast, the most substantial difference was observed between “Xinxiang” and “Zheshu81”, with a count of 10,734 DEGs (Figure 3C). Specifically speaking, compared with “Xinxiang,” 2721 down-regulated genes and 4112 up-regulated genes were identified in “Kokei No. 14,” while 4793 down-regulated genes and 5941 up-regulated genes were identified in “Zheshu81.” Out of all the DEGs identified, 1513 were shared by the three pairwise comparisons, 4834 were shared by “Zheshu81 vs. Xinxiang” and “Zheshu81 vs. Kokei No. 14,” 2613 were shared by “Zheshu81 vs. Xinxiang” and “Xinxiang vs. Kokei No. 14,” and 1855 were shared by “Xinxiang vs. Kokei No. 14” and “Zheshu81 vs. Kokei No. 14” (Figure 3D). Apart from the shared DEGs, 841, 1575, and 1774 DEGs were exclusively detected for “Xinxiang vs. Kokei No. 14,” “Zheshu81 vs. Kokei No. 14,” and “Zheshu81 vs. Xinxiang,” respectively. For the annotation of all the DEGs, 8092 (53.89%) were annotated in KEGG database, with 13,440 (89.50%) in NR, 10,011 (66.67%) in Swissprot, 12,929 (86.10%) in Tremble, 7341 (48.89%) in KOG, and 11,301 (75.26%) in GO databases.

To further verify the possible functions of the DEGs associated with the carotenoid synthetic pathway, two commonly used techniques, GO and KEGG enrichment analyses, were also used in this study [46]. The results of GO enrichment analyses of DEGs in the three pairwise comparisons and the main pathways are shown in Figure S2 and Table S6. For “Xinxiang vs. Kokei No. 14”, 14,531, 4435, and 6943 unigenes were assigned to “biological process”, “cellular component”, and “molecular function”, respectively. For “Zheshu81 vs. Kokei No. 14”, 21,424, 6515, and 9616 unigenes were assigned to “biological process”, “cellular component”, and “molecular function”, respectively. For “Zheshu81 vs. Xinxiang”, 24,175, 7485, and 10979 unigenes were assigned to “biological process”, “cellular component”, and “molecular function”, respectively. In “biological process” of the three pairwise comparisons, most genes were annotated to participate in cellular process and metabolic process, which were associated with secondary metabolite biosynthesis [47]. In “molecular function”, most genes were related to binding and catalytic activity, which might influence the function sites and catalytic activity of key enzymes involved in carotenoid synthesis [48]. Additionally, other genes were annotated with functions related to transcription regulatory activity and transporter activity, which have been shown to be associated with carotenoid synthesis [49,50]. For KEGG enrichment analyses, the DEGs among the three pairwise comparisons were matched to the KEGG database, and the significant enrichment pathways and classifications are shown (Figure S3; Table S7). Among the top 20 significantly enriched pathways, the metabolic pathway genes had the highest representation in terms of DEGs, although the enrichment significance was not pronounced (Table S7). For the three pairwise comparisons, the DEGs were matched to the pathways mainly associated with the biosynthesis of secondary metabolites, including carotenoid, flavone, and flavonol (Figure S3). In addition, apart from the metabolism-related pathways, the DEGs were also matched to the transporters, which were critical for the accumulation of secondary metabolites [51]. For example, as the main synthetic and storage organelle for the carotenoid, the transporters on the chromoplast membrane play an important role in transporting the raw materials necessary for the synthesis of carotenoids [18]. Taken together, the DEGs involved in the synthesis pathway of carotenoids were identified with the transcriptomic analyses, and the different expression levels might explain the content and composition differences among the three cultivars.

3.4. Analyses of the DEGs Involved in the Carotenoid Synthetic Pathway

The biosynthesis of carotenoids in plants primarily occurs within specialized plastids known as chromoplasts, which are responsible for producing and storing carotenoids [18]. The biosynthetic process requires the participation of numerous key enzymes, including GGDPS, PSY, PDS, ZDS, LCYE, LCYB, CHYB, ZEP, NXS, CCD, and NCED [24]. Carotenoid biosynthesis starts with the formation of geranylgeranyl diphosphate (GGPP) from isopentenyl diphosphate (IPP) and dimethylallyl diphosphate (DMAPP), produced via the plastid-localized MEP pathway. GGPP is then converted to phytoene by PSY, initiating carotenoid-specific synthesis. Phytoene undergoes desaturation by PDS and ZDS to form lycopene, which can be cyclized into α-carotene or β-carotene by LCYE and LCYB, respectively. These carotenes are hydroxylated by specific hydroxylases (CHYB) into xanthophylls such as lutein and zeaxanthin, with further epoxidation by VDE. Isomerization of prolycopene to all-trans lycopene is catalyzed by carotenoid isomerase (CRTISO). Therefore, the transcriptional expression levels of the key enzymes influence the composition and content of the carotenoids [52].

In the present study, the expression levels of the genes encoding some key enzymes were compared among the three cultivars using transcriptomic analysis, which were believed to be associated with the differences of carotenoids (Figure 4) [53]. For example, in the downstream of the synthetic pathway, NXS converts violaxanthin to neoxanthin. In our study, the expression level of NXS in “Kokei No. 14” was significantly higher than that in “Xinxiang” or “Zheshu81”, which was in accordance with the higher content of violaxanthin in “Zheshu81” compared with “Kokei No. 14” and “Xinxiang”. ZEP and VDE function as the intermediate enzymes, which provide linkages between lutein and epoxylutein, zeaxanthin and antheraxanthin, and antheraxanthin and violaxanthin. In our study, the expression levels of ZEPs and VDEs exhibited irregular patterns among the three cultivars, except that “Kokei No. 14” exhibited an extremely low level, which explained the undetected or low content of the above-mentioned carotenoids (Table S3). In the upstream of the synthetic pathway, GGDPS and PSY play an important role in activating the synthetic process of carotenoids and are acknowledged as rate-limiting enzymes. In this study, the expression levels of GGDPSs and PSYs were significantly higher in “Xinxiang” compared with “Kokei No. 14” and “Zheshu81.” However, the total carotenoid content and the content of carotenoid compositions of “Zheshu81” were higher (Table S3). This was probably due to the fact that at the sampling time, “Xinxiang” was still in great demand for carotenoid synthesis. In addition, the irregular or random expression of other key genes also affects the composition and content of carotenoids. It was noteworthy that the transcriptional level of g40864 (one of GGDPSs) in “Zheshu81” was significantly higher than that of “Kokei No. 14” or “Xinxiang,” while the transcriptional level of PSYs in “Zheshu81” was lower. This was because of the fact that GGPP, the product of GGDPS catalysis, was also the source material for synthetic pathways of other important metabolites. Meanwhile, in terms of the degradation activities, the transcriptional activities of most of CCDs and NCEDs in “Zheshu81” were much higher than those of “Kokei No. 14” or “Xinxiang,” which was in close relationship with the high content of different kinds of carotenoids in “Zheshu81.” As the two main forms of carotenoid cleavage oxygenase (CCO), NCEDs proved crucial for catalyzing the initial step of abscisic acid (ABA) biosynthesis, while CCDs generate precursors of the strigolactones hormone [54]. The differential expression patterns of these key genes (ZEPs, NCEDs, and CCDs), together with GGDPS and PSY, synergistically influences the composition and content of carotenoids in the sweetpotato.

To verify the expression levels of the DEGs associated with carotenoid metabolism in sweetpotato tuberous roots, six DEGs were randomly selected for qRT-PCR analysis (Figure S4). The selected genes, NCED-1 (g30636), CCD-7 (g30428), CCD1-1 (g21335), CCD1-3 (g57696), CCD8 (g21348), and ZEP-3 (g50580), are located in the synthesis and degradation process during the pathway representatively (Figure 4). The expression of all the selected genes was higher in “Zheshu81” than in “Kokei No. 14” or “Xinxiang,” which was in accordance with the RNA-seq results (Figure S4A). Furthermore, the cross-method validation showed high consistency (r² > 0.8577) between qRT-PCR quantification and RNA-seq profiling (Figure S4B), confirming the technical reproducibility of our transcriptomic datasets.

WGCNA characterizes gene expression correlation patterns across various samples and identifies gene clusters with significant phenotypic associations. In the present study, following the removal of 90% of genes exhibiting FPKM values < 1 in more than 50% of samples, a total of 8000 genes were retained and subsequently clustered into 10 distinct modules (Figure 5A; Table S8). Hierarchical clustering heatmaps were generated to illustrate inter-modular correlations among the identified gene modules (Figure S5). Module–carotenoid relationship analysis revealed that the “turquoise” and “brown” co-expression modules demonstrated significant associations with carotenoid metabolites (Figure 5B). Specifically, the “brown” module, comprising 1746 genes, exhibited strong negative correlations with β-carotene, lutein, and violaxanthin contents (r ≤ −0.9, p < 0.05), while the “turquoise” module, containing 2340 genes, showed pronounced positive correlations with these carotenoid components (r ≥ 0.9, p < 0.05).

To further interpret these findings, the potential biological implications of the “turquoise” and “brown” modules were studied. The robust positive correlation of the “turquoise” module with multiple carotenoids suggested it harbored genes that were core drivers of carotenoid biosynthesis. We cross-referenced genes within this module with known carotenoid pathway enzymes and found key genes such as PSY, LCYB, and ZEP. These genes are pivotal at different steps of the carotenoid biosynthetic pathway, with PSY initiating the pathway, LCYB directing the synthesis of β-branch carotenoids, and ZEP participating in the conversion towards xanthophylls. Their presence in the “turquoise” module, coupled with the module’s strong positive correlation with carotenoid levels, reinforces the module’s role in promoting carotenoid accumulation. Conversely, the “brown” module’s strong negative correlation with carotenoid contents implies it may function to repress carotenoid synthesis or divert metabolic flux away from carotenoid production. Within this module, the genes associated with processes that could compete with carotenoid biosynthesis for common precursors or energy resources were identified. For instance, some genes were related to general metabolic maintenance or other secondary metabolic pathways, which might limit the availability of substrates or energy for carotenoid production, thus leading to the observed negative correlation. Additionally, the expression patterns of representative genes from both modules across the three sweetpotato cultivars were examined. In the orange-fleshed “Zheshu81” with high carotenoid content, genes from the “turquoise” module showed significantly higher expression levels, while genes from the “brown” module had lower expression. In contrast, in the cream-fleshed “Kokei No. 14” with low carotenoid content, the expression trends were reversed. This expression pattern consistency across cultivars with differing carotenoid profiles further supports the regulatory roles of these modules in carotenoid accumulation.

3.5. Co-Expression Network Analysis for Identifying Genes Related to Carotenoid Composition and Content

To identify genes associated with carotenoid content and composition, and to explore the relationships between gene modules, we selected differentially expressed nodes with a weight over 0.2 from the “turquoise” and “brown” modules. This process identified 891 and 1344 genes in the “turquoise” and “brown” modules, respectively. We used the MCODE plugin in Cytoscape to analyze and cluster the co-expression networks. Genes from cluster 1 (positive correlation) and clusters 1, 2, and 3 (negative correlation) were extracted, with gene significance (GS) and module eigengene connectivity (kME) set to ≥0.9 or ≤−0.9. The co-expression networks of both modules were visualized, leading to the identification of 40 and 44 transcription factors and their families related to carotenoid accumulation in the “turquoise” and “brown” modules, respectively (Figure 6A,B; Table S8). In the “turquoise” network, we identified the gene NCED-1 (g30636), which is positively correlated with carotenoid synthesis. The finding was different from previous reports where the high expression of NCED explained the low β-carotene content or carotenoid degradation [55,56]. This was probably due to the emphasized initiation of NCED-1 expression by the high content of carotenoids, which acted as substrates of NCED-1. Moreover, the specific regulatory context in sweetpotato tuberous roots, such as the coordination with other carotenoid biosynthesis-related transcription factors, might also contribute to this distinct correlation pattern. The expression levels of these 40 transcription factors and their families were significantly higher in “Zheshu81 compared to “Kokei No. 14” and “Xinxiang”, with opposite expression trends observed in “Kokei No. 14” and “Xinxiang” (Figure 6A and Figure S6). In the “brown” network, we identified two genes negatively correlated with carotenoid synthesis: PICBP (g30547) and AKHSDH2 (g37772), both of which were reported for the first time. These 44 transcription factors were predominantly highly expressed in “Kokei No. 14” but showed low expression levels in “Zheshu81” (Figure 6B and Figure S6). Based on the GS, we further constructed networks of differentially expressed transcription factors (DETFs) and DEGs related to carotenoid metabolism pathways and carotenoid content in the “turquoise” and “brown” modules to investigate the role of hub genes in regulating carotenoid accumulation (Figure 6C,D).

In the “turquoise” network, NAC (g288) showed significant positive correlations with violaxanthin and lutein content, while NCED-1 (g30636) exhibited a negative correlation with β-carotene content [55,56,57]. In the “brown” network, TCP (g25256), NF-YB (g61252), and Garp-arr-b (g47091) significantly negatively correlated with violaxanthin content. Garp-arr-b (g47091), HSF (g41892), and bZIP (g47459, g56238, and g12897) significantly negatively correlated with lutein content. AKHSDH2 (g37772) and PICBP (g30547) also significantly negatively correlated with lutein content. bZIP (g47459, g56238, and g12897) negatively correlated with β-carotene content. MYB (g20760) positively correlated with both β-carotene and lutein content. AKHSDH2 (g37772) negatively correlated with both lutein and β-carotene content. These results demonstrate that genes within the “turquoise” and “brown” modules were closely associated with carotenoid content. Additionally, several key transcription factors played critical roles in regulating carotenoid biosynthesis and accumulation, thereby influencing the flesh color.

4. Conclusions

In this study, we conducted a comprehensive analysis of carotenoid biosynthesis in three elite sweetpotato cultivars with distinct flesh colors. Through metabolomic and transcriptomic approaches, significant variations in carotenoid content and composition were identified, with “Zheshu81” exhibiting the highest levels of carotenoids, particularly β-carotene, lutein, and violaxanthin. Transcriptome analysis revealed DEGs associated with carotenoid biosynthesis, including key enzymes such as PSY and LCYB. WGCNA further highlighted regulatory genes and transcription factors, such as NCED-1 and PICBP, that are closely associated with carotenoid accumulation. The expression profiles of these genes showed a correlation with the observed differences in carotenoid levels across the cultivars. These results provide a detailed understanding of the genetic and metabolic mechanisms underlying carotenoid biosynthesis in the sweetpotato, offering valuable insights for future research and breeding programs aimed at improving carotenoid content and flesh color in sweetpotato cultivars.