Plant materials

A total of 76 accessions, including 16 entries of S. melongena from Asian and European origin, 30 of S. aethiopicum belonging to the four varietal groups (Aculeatum, Gilo, Kumba and Shum) plus intermediate forms between S. aethiopicum and S. anguivi, five of S. macrocarpon, and 25 accessions of 14 wild species were used for the present study (Table 1). Among the 16 entries of brinjal eggplant, two of them are doubled haploids (S. melongena_10 and S_melongena_12) obtained by anther culture [33]. Also, four brinjal eggplant entries come from two original sources (entries S. melongena_1 and S. melongena_2 from the original source MEL1; and accessions S. melongena_6 and S. melongena_7 from the original source MEL5) (Table 1). Among the wild relatives are included the putative ancestors of brinjal eggplant (S. insanum), scarlet eggplant (S. anguivi), and gboma eggplant (S. dasyphyllum) [34–36], as well as eight other wild species from Old World origin (S. campylacanthum Hochst. ex A. Rich, S. incanum L., S. lichtensteinii Willd., S. lidii Sunding, S. linnaeanum Hepper & P.-M.L. Jaeger, S. tomentosum L., S. vespertilio Aiton, and S. violaceum Ortega), and three native to the New World (S. elaeagnifolium Cav., S. sisymbriifolium, and S. torvum) [30,37]. All these materials are conserved in the germplasm collection maintained at Universitat Politècnica de València (Valencia, Spain).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Plant materials used including taxon, accession name, accession code used in the present work, country of origin and fruit shape and predominant colour.

https://doi.org/10.1371/journal.pone.0180774.t001

Library construction and sequencing

DNA was extracted following a modified CTAB method [38] as indicated elsewhere [39]. Library construction (11/2015) was performed as proposed in Peterson et al. [40] and modified as in Acquadro et al. [41], by using a HindIII-MseI enzyme combination and adding a final biotin/streptavidin-coated beads based purification step. Quality, quantity and reproducibility of libraries were assesed on a Bioanalyzer instrument (DNA High Sensitivity chip) as well as qPCR. On the basis of the quantitation, DNA libraries were pooled and sequenced on Illumina HiSeq 2500 platform (Illumina Inc., San Diego, CA, USA), following the manufacturer protocol using 100SE chemistry.

Sequence analysis

Raw reads were analyzed with Scythe (https://github.com/vsbuffalo/scythe) for filtering out contaminant substrings and Sickle (https://github.com/najoshi/sickle), which allows to remove reads with poor quality ends (Q<30). Illumina reads were de-multiplexed on the basis of the Illumina TruSeq index. Alignment to the reference eggplant genome [42,43] was carried out using BWA aligner [44] (i.e., mem command) with default parameters and avoiding multiple-mapping reads. SNP mining was conducted by adopting a Samtools-based pipeline [45]. Homozygous/heterozygous SNP/Indel calls were considered only with phred-scaled genotype likelihood equal zero. A catalog of candidate high quality SNPs was produced. Relationships among the genotypes were computed using: i) whole genome, and ii) coding (within exons) SNP/indel datasets. The proportion of heterozygous SNPs for each genotype was estimated by the ratio of total number of heterozygous SNPs and all the detected SNPs (excluding missing SNPs) as well as the ratio of the number of heterozygous SNPs in coding regions and all the detected SNPs in coding regions.

Genetic relationships analysis and population structure

SNP data were coded according to the number of occurring polymorphisms, being assigned a 0 if they showed the homozygous reference type, a 1 if the variant occurred in one chromosome and a 2 if the variant was present in both chromosomes. Genetic similarities between pairs of entries were quantified by the Dice similarity index [46] as 2m⁺/(2m⁺ + m^-), were m⁺ is the number of matches (1–1 and 2–2) and m^- is the number of mismatches (0–1, 0–2 and 1–2). Genetic relationships were described by using both the unweighted pair-group arithmetic mean (UPGMA) method with 1,000 bootstraps, and principal coordinate analysis (PCoA) by means of Past 3.14 software [47]. A co-phenetic matrix was also produced using the hierarchical cluster system, by means of the COPH (cophenetic values) routine, and correlated with the original distance matrix, in order to test for associations between clusters and the similarity matrix.

FastSTRUCTURE [48] was used to estimate the number of sub-populations in the panel, applying the admixture model for the ancestry of individuals and correlated allele frequencies. A hierarchical FastSTRUCTURE analysis [49] was also applied on accessions which clustered in sub-group 1 and subgroup 4 following UPGMA analysis as well as on the set of all the remaining. The program was run with default setting using simple prior to obtain a reasonable range of values for the number of populations (K), FastSTRUCTURE was executed for multiple values of K (K = 1–9). The script chooseK.py [48] was then used to infer the most likely number of populations.

Sequencing and SNPs identification

A total of 225 million single reads were produced. About 94% of raw reads contained the expected restriction site overhang, along with discriminating inline barcodes. The average number of successfully de-multiplexed reads per sample was 2.7 M, with a standard deviation of 1.5 M (S1 Fig). Sequences were trimmed and quality cleaned to 210 million of useful reads (6.2% discarded). The latter were then aligned to the recently produced reference eggplant genome [42,43] and close to 100% of reads were successfully mapped to single regions (no multiple mapping was permitted). Mapped sequences showed an extensive coverage alongside the 12 chromosomes (data not shown).

In all, 75,399 polymorphic sites were identified among the 76 lines in study. Overall, all the S. melongena accessions, together with the three S. insanum accessions, showed a reduced level of polymorphism (on average 2.47 and 4.75% respectively) when aligned to the reference genome. On the other hand the frequency of polymorphic SNPs ranged from 10.62 to 24.32% in the other entries (S1 Table).

Solanum melongena is a largely autogamous species [20], thus its low level of heterozygosity (on average 1.66%) is coherent with the expectation that germplasm accessions and non-hybrid varieties should be highly homozygous (Fig 1, S1 Table). Interestingly, the two S. melongena varieties (S. melongena_10, S. melongena_12, Fig 1, S1 Table), which are the result of diploidization of haploid plants obtained through anther culture, displayed some heterozygosity (<0.5%). This might be due to somaclonal variation, which is manifested as cytological abnormalities, sequence change, and gene activation and silencing which occur through the ‘in vitro’ culture process and that provides evidence that DNA modifications occur more frequently in ‘in vitro’ cultivated than in seed-grown plants [50]. However, it might be also a consequence of SNPs mapping on paralog genes since, similarly to tomato, potato and pepper, also eggplant the genome carries signs of the “T” triplication occurred during Solanaceae evolution [42], or being the results of some mapping artifacts. This would suggest that the heterozygosity detected in the rest of S. melongena accessions would be overestimated by almost one third.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1.

SNP numbers and distribution: A) Plot of the number of SNPs (total, homozygous and heterozygous) and missing sites observed within the collection of 76 Solanum accessions; B) The distribution of missing sites along two eggplant chromosomes (i.e.: chr4 and chr8). The trend lines track missing sites in three Solanum melongena and six accessions of American origin (S. elaeagnifolium, S. sisymbriifolium and S. torvum).

https://doi.org/10.1371/journal.pone.0180774.g001

The two other cultivated eggplant species (i.e., S. aethiopicum and S. macrocarpon) showed, on average, higher heterozygosity than S. melongena, ranging from 4.52 to 9.53% (Fig 1, S1 Table). This might be a consequence of their higher allogamy and the more limited breeding efforts for stabilizing phenotypic and yield-related traits. Low heterozygosity was also observed in the wild S. insanum, S. lichtensteinii and S. linnaeanum (< 3.5%), while higher values, over 10%, were observed in the wild species S. campylacanthum, S. anguivi and S. violaceum.

Some missing data were observed in S. melongena and S. aethiopicum (ranging from 2.71% to 9.87%), some accessions showed a medium-high level of missing data (e.g., S. macrocarpon 19% on average), while others showed a surprisingly high number of missing data (up to 54.5%, 54.43 and 36.88, in Solanum torvum, S. sisymbriifolium and S. elaeagnifolium, respectively). This might be explained by the fact that these latter species are native to the New World [30,37] and in consequence have a more distant common ancestor, and greater evolutionary divergence. Missing data were distributed on the different eggplant chromosomes; however, by adopting a five million bases sliding window analysis, some hot spot regions were highlighted (Fig 1). The filtering of the whole SNP dataset for the sites present in CDS regions granted 12,859 SNPs. The latter were used for all the subsequent analyses, since the relative number of missing data was lower in the coding dataset (3% on average) than in the whole dataset (10% on average, Fig 1, S1 Table). As an example the percentage of missing data of South American accessions (S. elaeagnifolium, S. sisymbriifolium and S. torvum–S1 Table) was lowered from about 46.1% to 15.1%, thereby increasing the resolution power of our analyses.

Genetic relationships analysis and population structure

The UPGMA-based dendrogram and the output of FastSTRUCTURE [48] analysis (Fig 2) show the genetic relationships between the 76 accessions. Both, as well as the K analysis (Fig 2 box), suggest a population structure comprising four sub-groups. Each entry was fingerprinted and the co-phenetic correlation coefficient (r-value) between the Dice data matrix and the co-phenetic matrix was 0.978, highlighting a very good fit between the dendrogram clusters and the similarity matrices from which they were derived, indicating that the UPGMA method is suitable for the interpretation of our data. The fact that the sister entries S. melongena_1 and S. melongena_2, which are derivatives from the original source MEL1 cluster together in the dendrogram, and the same occurs for accessions S. melongena_6 and S. melongena_7, which derive from MEL5 provide a confirmation that the analysis is congruent.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. The genetic architecture of the full germplasm panel: Consensus UPGMA dendrogram and FastSTRUCTURE output at K = 4.

Bootstrap values (%) for the main nodes are reported in red. Each entry is represented by a vertical line representing sub-group 1 (blue) sub-group 2 (orange), sub-group 3 (yellow) and sub-group 4 (grey). The box reports the Probabilities (p) plots derived from the FastSTRUCTURE analysis of genotypic data with K values from 1 to 9.

https://doi.org/10.1371/journal.pone.0180774.g002

According to the level of membership provided by FastSTRUCTURE [48], sub-group 1 (blue) includes all the accessions of scarlet eggplant (S. aethiopicum) and S. anguivi, which on the basis of previous studies has been reported to be its wild ancestor [30,34,51]. Sub-group 2 (orange) includes members of the brinjal eggplant complex [52,53], among which the most genetically related accessions of S. melongena and its wild progenitor S. insanum, the accession of S. incanum, and the two of both S. lichtensteinii and S. linneanum. Sub-group 3 (yellow) includes the five accessions of gboma eggplant (S. macrocarpon) and the one S. dasyphyllum, which is its wild progenitor [30,35]. Sub-group 4 (grey) includes the accessions of the New World species, which form part of the tertiary genepool of brinjal eggplant [30]. Finally, the remaining accessions of S. campylacanthum, S. violaceum, S. lidii, S. vespertilio and S. tomentosum had ambiguous membership and were thus classified as admixed, as their level of membership to a single group was lower than 70% (Fig 2). With the goal to provide insight into the complex relationships of the germplasm used, and to detect additional sub-population structure, a hierarchical FastSTRUCTURE analysis was applied by running STRUCTURE on partioned data, i.e. on accessions which clustered in sub-group 1 and sub-group 4 following UPGMA analysis, as well as on the remaining materials (S2 Fig). The hierarchical FastSTRUCTURE analysis for the scarlet eggplant complex revealed that the optimal number of populations was obtained at K = 2, and that the accessions of S. aethiopicum and S. anguivi, included in the UPGMA subgroup 1, share a common genepool. For the brinjal eggplant and gboma eggplant complexes group, the hierarchical FastSTRUCTURE analysis suggests that four populations are present. In this set of accessions K = 2 separates the brinjal eggplant S. melongena and its close relatives S. insanum, S. incanum, S. lichtensteinii and S. linnaeanum [18] from the gboma eggplant S. macrocarpon and its wild ancestor S. dasyphyllum [35] together with the Canary Islands endemisms S. lidii and S. vespertilio and the related South African S. tomentosum [23,54,55], while S. campylacanthum and S. violaceum appear as an admixture (S2 Fig). At K = 3 the S. lidii, S. vespertilio, S. tomentosum and S. violaceum are separated from the gboma eggplant. Finally, at the optimal K = 4, S. campylacanthum accessions group separately, while S. incanum, S. lichtensteinii and S. linnaeanum appear as an admixture of S. melongena/S. insanum and S. campylacanthum (S2 Fig). This might be a consequence of gene flow among them or the result of the recent speciation from a common ancestor or both. These species are phylogenetically closely related but at present are distributed in different geographical areas [18]; this suggests that presumably they evolved from a common ancestor for adaptation to different niches, which might difficult gene flow. The hierarchical FastSTRUCTURE analysis of the New World species recognized at K = 2 two populations, one of which included S. elaeagnifolium while the other both S. sysimbriifolium and S. torvum. However, at the optimal K = 3, the latter was further splitted in two genetically differentiated genepools, each including one of the two species.

PCoA analyses

The whole data set was also subjected to PCoA analysis (Fig 3) which, on the whole, confirmed the grouping of genotypes based on UPGMA and FastSTRUCTURE [48] clustering. Because a limited number of samples of each of the wild relatives was included in our study, the PCoA analysis did not allow to highlight the within-species diversity as it did in the cultivated species; however, it made possible some additional inferences. The first two axes explained 71.4% of the genetic variation. The first axis, explaining 57.6% of the genetic variation, clearly separated cultivated scarlet eggplant S. aethiopicum and its wild ancestor S. anguivi from all the other accessions, with no evident separate clustering of the two species. The latter are fully inter-fertile [34,51] and the identification of intermediate forms [27,29] suggests occurrence of genetic flow between them.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The genetic architecture of the full germplasm panel of 76 Solanum accessions.

PCoA visualization of the genetic relationships within the full set of accessions.

https://doi.org/10.1371/journal.pone.0180774.g003

The second axis, explaining 13.8% of the genetic variation, clearly split the entries of S. sisymbriifolium, S. torvum and S. elaeagnifolium, which clustered in the previously described group 4, from the ones of sub-clusters 2 and 3 as well as the entries classified as admixed, i.e. brinjal and gboma eggplants, their respective progenitors S. insanum and S. dasyphyllum together with other Old World wild species, as well as the entries classified as admixed.

Both S. sisymbriifolium and S. torvum are native of South and Central America and, together with S. viarum, were classified in GP3 by Syfert et al. [30]. They have been also reported to be, within subgenus Leptostemonum (Dun.) Bitt., phylogenetically far away from the cultivated eggplants and the other Old World species [23,30,54,55]. Solanum elaeagnifolium is also a New World species [37] which was not included in the study of Syfert et al. [30], and whose origin is attributable to GP3 on the basis of the present results.

Both S. sisymbriifolium and S. torvum are of interest for eggplant breeding, as they are tolerant or resistant to many diseases [20]. Their high phylogenetic distance to cultivated eggplants is confirmed by the many ineffective attempts to hybridize them with S. melongena [29,56–58]. No sexual hybrids have ever been reported between S. melongena and S. sisymbriifolium, while interspecific hybrids obtained through embryo rescue of the cross S. melongena x S. torvum were highly sterile and no backcrosses have been reported to date [17]. Furthermore, although tetraploid somatic hybrids between either S. sisymbriifolium or S. torvum with S. melongena were obtained, they did not produce sexual offspring [59,60].

On the basis of PCoA analysis, the cultivated species which appears genetically closer to the cultivated eggplant is gboma eggplant (S. macrocarpon), clustering together with S. dasyphyllum, which has been reported by many authors to be its wild ancestor [23,35,52,54,61] (Fig 3). This seems to indicate that gboma eggplant, might be genetically closer to S. melongena than the cultivated scarlet eggplant (S. aethiopicum), which is included in section Oliganthes (Dunn.) Bit. [34,62]. However contrasting results have been reported in literature. Based on chloroplast DNA RFLPs [63], ISSRs [64], AFLPs and nrITS sequences [19] it was previously reported that S. aethiopicum is closer to S. melongena than S. macrocarpon; otherwise Sakata and Lester [65], in a study based on chloroplast DNA RFLPs, and Vorontosva et al. [23] using ITS, waxy and trnT-F regions sequences obtained opposite results. Interestingly, Furini and Wunder [66] using AFLPs as well as Levin et al. [54], Weese and Bohs [53] and Särkinen et al. [55] using several nuclear and plastid DNA sequences found that S. aethiopicum and S. macrocarpon were phylogenetically closer among them than to S. melongena. Studies based on the species inter-fertility highlighted that interspecific hybrids between S. melongena and S. aethiopicum as well as backcrosses could be easily obtained [17,67,68]; on the other hand, although hybrids between S. melongena and S. macrocarpon were obtained [56,67,69], in most cases they were high sterile and only the backcross of a tetraploid hybrid between the two species with S. melongena was successful [69]. The difficulty in obtaining the hybrids between these two species, despite being phylogenetically close [23,65], might be caused by some chromosomal rearrangement or other hybridization barriers. At last, Kouassi et al. [58] reported that the backcrosses towards S. melongena of the hybrid between S. dasyphyllum (wild ancestor of S. macrocarpon) and S. melongena was successful. A clarification is provided by our data obtained from FastSTRUCTURE analysis (Fig 2) which highlights that the three cultivated species belong to clearly separate groups, suggesting that S. macrocarpon should be excluded from section Melongena (Mill.) as proposed by Sakata et al. [63].

PCoA analysis also showed that S. campylacanthum, S. incanum, S. insanum, S. lichtensteinii and S. linnaeanum, which form part of the “brinjal eggplant” complex [52,53], cluster in proximity with eggplant (Fig 3). Among them, S. campylacanthum appears to be the most genetically differentiated from the others. This is in agreement with previous AFLP, nuclear and chloroplast DNA sequence results [23,53,61]. Indeed, interspecific hybrids were obtained between S. campylacanthum and S. melongena, but the number of seeded fruits and seeds per fruit was lower in respect to the ones obtained following crosses with other species within the “common eggplant” complex [52,58,70]. Solanum linnaeanum and the accession of S. lichtensteinii cluster together and close to S. melongena. This result confirms that the two species are genetically related [23,53] and supports the hypothesis that S. linnaeanum and S. lichtensteinii are of South African origin and share a common ancestor, although the former grows in several tropical and subtropical areas of the world [18,23].

Solanum linnaeanum and S. lichtensteinii produce hybrids with moderate or high fertility when crossed with eggplant [18], which can be also backcrossed with relative ease [17,19,23,58,28]. However our data show that they are genetically more distant from S. melongena than S. incanum or S. insanum [19,23,53,65,66]. Solanum incanum was suggested to be eggplant’s pre-domestication ancestor and is being used in eggplant breeding programs as a source of variation for phenolics content and resistance to drought [18]. Recent morphological and molecular work has shown that species-level differences exist between S. incanum and S. melongena and, on the basis of new evidence, S. insanum is considered the eggplant wild progenitor [36]. The two species are also fully inter-fertile and their hybrid produce many fruits and seeds [29]. It is also not surprising that, since frequent genetic flow occurs between both species in the indo-birmanian region [71,72], in our PCoA analysis the S. insanum accessions appear intermingled with the ones of S. melongena.

Our data show that the three species S. lidii, S. tomentosum, S. vespertilio cluster into proximity to each other and S. violaceum a little more apart (Fig 3). Solanum lidii and S. vespertilio are endemic to the Canary Islands (Spain) and are genetically similar sister species, which were found to cluster together in previous molecular studies [23,54,55,73,74]. In several molecular studies S. tomentosum was also found to cluster close to S. lidii and S. vespertilio [23,54,55,73], thus our results confirm that the three species are close relatives. Solanum violaceum clusters with these three taxa in both the FastSTRUCTURE and PCoA analyses in spite of having a native distribution in India and Southeast Asia [19].

Within-groups PCoA analyses

In order to gain a better landscape of the genetic relationships among the species in study, PCoA analyses were also separately performed on entries clustering in the sub-groups 1, 2 and 4, following FastSTRUCTURE [48] analysis (Fig 4A, 4B and 4C). The separate PCoA of entries grouped in sub-group 1 (Fig 4A) confirmed that the different S. aethiopicum varietal types are partially intermingled and show a high within varietal type genetic diversity; furthermore, the absence of an evident genetic differentiation with their wild ancestor S. anguivi was confirmed. As observed by Sunseri et al. [24] in a molecular characterization based on AFLP and SSR markers, the different cultivar groups of S. aethiopicum were intermingled in the cluster analysis. The four cultivar groups (Aculeatum, Gilo, Kumba, and Shum) are distinguished by simple morphological traits, like fruit size and shape, fruit bitterness, and the presence or absence of prickles and star leaf hairs [26,34], which allow the differentiation among cultivars based on morphological characterizations [21]. However several of these traits, like prickliness and presence/absence of star leaf hair, seem to have a simple genetic basis in scarlet eggplant [51] while, as occurs in common eggplant [75,76], other traits (fruit size and shape) are under control of a few major genes. The genetic flow occurring between different groups, as a result of spontaneous or artificial hybridization, may thus result in a lack of (or reduced) genetic differentiation. Indeed, in a previous study [26], it was reported that the Aculeatum group seems to have been derived from hybridization between S. aethiopicum group Kumba and S. anguivi. On the whole, the varietal groups that showed the highest genetic differentiation were Aculeatum, (characterized by the highest anthocyanin content and prickliness in respect to all the others) and Shum, which differed for the mean average values of 8 of 18 morphological traits analysed in a previous study [21].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

Within-groups PCoA analyses in subgroups of germplasm panel of Solanum accessions: visualization of the genetic relationships within sub-group 1 (A; scarlet eggplant complex), sub-group 2 (B; brinjal eggplant complex) and sub-group 4 (C; New World species).

https://doi.org/10.1371/journal.pone.0180774.g004

PCoA including accessions of the sub-group 2 (Fig 4B), as expected grouped separately the entries of eggplant and its wild relative S. insanum from the close relatives S. incanum, S. lichtensteinii and S. linneanum, the latter being the most genetically differentiated from all the others. The S. melongena accessions analysed included types, hailing from Sri Lanka, India and China as well as from Ivory Coast and Spain and producing fruits of different shape and colour. In a previous work [16] 191 eggplant accessions were scored for a set of 19 fruit and plant traits and the analysis of phenotypic data made it possible to classify the genotypes in three main fruit morphological groups producing: (i) elongated fruits, (mean ratio fs = fruit length/fruit maximum diameter around 5.05); (ii) semi-long fruits (fs from 1.2 to 2) and (iii) round fruits (fs around 1), which cut across the Oriental and Occidental divide. On the other hand STRUCTURE [77] analysis based on 24 microsatellite markers (22 genomic ones and two from EST), identified two major sub-groups, which to a large extent mirrored the provenance of the entries. In the present study, in spite of the wide set of polymorphisms detected, the accessions from different origin did not highlight a grouping together trend. This apparent discrepancy can be explained by either the difference in size of the two germplasm sets, but also by the number of markers applied, as the use of a limited number of selected markers might provide unrealistic estimates of genetic variability in the set of accessions in study.

PCoA including accessions of the sub-group 4 (Fig 4C) highlights that S. sisymbriifolium, S. torvum and S. elaeagnifolium are genetically far away from each other and that their grouping in the sub-group 4 is due to their common high genetic divergence from all the other entries. This is confirmed by previous molecular results that includes S. torvum and S. sisymbriifolium in different clades within subgenus Leptostemonum from the cultivated eggplants [23,30,54,55]. Furthermore, on the basis of PCoA analysis, the two accessions of S. torvum form a group 'per se' in respect to all the others.

Previous phylogenetic studies placed S. elaeagnifolium and the rest of species of the Elaeagnifolium clade closer to Old World species than either S. sisymbriifolium or S. torvum [19,23,54,55,73]. Recently, crossing data confirm that S. elaeagnifolium is closer to eggplant than either S. sisymbriifolium and S. torvum, as interspecific hybrids have been obtained which present intermediate fertility [58], and with which it is possible to obtain backcrosses with S. melongena (unpublished results).