The complete chloroplast genome of grape is 160,928 bp in length (Fig. 1) and includes a pair of inverted repeats 26,358 bp long, separated by a small and a large single copy region of 19,065 bp and 89,147 bp, respectively. The grape chloroplast genome has 113 unique genes, 18 of which are duplicated in the IR, for a total of 131 genes (Fig. 1). There are four ribosomal and 30 distinct tRNA genes; seven of the tRNA genes and all rRNA genes are duplicated within the IR. There are 17 intron-containing genes, 15 of which contain one intron, and two of which contain two introns. Overall, the gene order in the grape chloroplast genome is identical to that of tobacco. The grape genome is 37.40% GC and 62.60% AT; 57.55% of the genome corresponds to coding regions and 42.45% to non-coding regions, including introns and intergenic spacers.

Repeat analysis identified 36 repetitive elements (30 bp or longer with a sequence identity of at least 90%), 15 of which are direct repeats and 21 of which are inverted repeats (Fig. 2 and Table 1). Eight direct repeats and 12 inverted repeats were 30 – 40 bp long, and the longest direct repeats were 64 bp. The majority of the repeats were located within intergenic spacer regions, intron sequences and ycf2. Two distinct 64 bp direct repeats were found in ycf2, which is located in the IR. Additionally, a 41 bp direct repeat was located in psaA and psaB, and a shorter, 32 bp direct repeat was found in two serine transfer-RNA (trnS) genes that recognize different codons; trnS-GCU and trnS-UGA. Lastly, a 31-bp direct repeat was identified within trnG-GCC in the IR, and a 39-bp direct repeat was found three times in the grape chloroplast genome, with a single occurrence in an intergenic spacer region, and also in the ycf3 and ndhA introns.

Comparison of DNA and EST sequences for chloroplast-encoded proteins retrieved from GenBank showed that most photosynthetic machinery and ribosomal subunit genes have 100% sequence identity with their respective EST sequences. Eleven non-synonymous nucleotide substitutions, resulting in a total of nine amino acid changes, were identified for atpI, clpP, matK, petB, petD, psbA and rpl22 compared to the ESTs (Table 2). Also, there were five synonymous substitutions. In the cases of non-synonymous substitutions, all genes experienced one nucleotide substitution except clpP, which had five variable sites. Lastly, in atpI, clpP and psbA the nucleotide substitutions had an impact on the hydropathy of the amino acid, changing it from aliphatic to hydrophilic, and vice versa. These differences could be due to mRNA editing, sequencing error of either the genomic DNA or ESTs, or polymorphisms between the samples used for genomic and EST sequences (see Discussion).

We examined two datasets that differed by a single rosid taxon to assess the effect of taxon sampling on resolving relationships among rosids. The first data matrix examined for phylogenetic analyses included 61 protein-coding genes for 28 taxa (Table 3, excluding Gossypium), including 26 angiosperms and two gymnosperm outgroups (Pinus and Ginkgo), and the second data matrix included 29 taxa with the addition of Gossypium. Both data sets comprised 45,573 nucleotide positions but when the gaps were excluded there were 39,624 characters.

Maximum Parsimony (MP) analyses of the 28-taxon dataset resulted in a single, fully resolved tree with a length of 49,511, a consistency index of 0.47 (excluding uninformative characters) and a retention index of 0.62 (Fig. 3A). Bootstrap analyses indicated that 18 of the 25 nodes were supported by values ≥ 95% and all but one of these had a bootstrap value of 100%. Maximum likelihood (ML) analysis resulted in a single tree with – lnL = 289638.676. ML bootstrap values also were consistently high, with values of ≥ 95% for 21 of the 25 nodes. The ML and MP trees had very similar topologies, except for two important differences. The first concerned the position of the two basal angiosperm lineages. The MP tree placed Amborella as the most basal lineage followed by the Nymphaeales (including Nuphar and Nymphaea), whereas the ML tree placed Amborella sister to the Nymphaeales, and together this group formed the basal lineage of angiosperms. The second topological difference concerned the placement of Calycanthus, the only representative of the magnolids. The MP tree placed Calycanthus sister to the eudicots, whereas the ML tree positioned Calycanthus as sister to a large clade that included both monocots and eudicots. Support for the different placements of Calycanthus was weak in both MP and ML analyses, whereas the support for the different resolutions of basal angiosperms was stronger (Fig. 3). These two differences were also detected in a recent phylogeny of basal angiosperms based on whole chloroplast genome sequences 5. The remaining angiosperms formed two major clades, one including monocots and a second including the eudicots (highlighted with thick lines in Figs. 3 and 4). Monophyly of the monocots was strongly supported (100% bootstrap value for both MP and ML) and included members of three different orders (Acorales, Asparagales, and Poales). Ranunculales were the earliest diverging lineage of eudicots. There were two major clades of core eudicots, one including the rosids and the second including the Caryophyllales + asterids. Within the rosids, Vitis was sister to the remaining taxa, which formed two clades, one including Cucumis (Cucurbitaceae) + Myrtales, and a second with Arabidopsis (Brassicales) + Fabales. Overall, relationships within rosids were in agreement with recent phylogenies [summarized in 1] except that the eurosids I clade was paraphyletic in our analyses.

MP analysis of the second dataset of 29 taxa including Gossypium resulted in a single most parsimonious tree with a length of 51,056, a consistency index of 0.46 (excluding uninformative characters) and a retention index of 0.61 (Fig. 4A). Bootstrap analyses indicated that 24 of the 26 nodes were supported by values ≥ 95%, and all but four of these nodes had a bootstrap value of 100%. ML analysis resulted in a tree with a – lnL = 296670.545 (Fig. 4B). ML bootstrap analyses indicated that 22 of the 26 nodes were supported by values ≥ 95% and all but two of these nodes had a bootstrap value of 100%. Both MP and ML analyses provided strong support for Vitis as the earliest diverging lineage of rosids, monophyly of Myrtales, and sister relationship of Brassicales and Malvales. The ML and MP trees had three important topological differences. The first two differences concerned the position of Calycanthus and the basal angiosperms, which were identical to those described above for the analyses that excluded Gossypium. The other difference concerned relationships among rosids. The MP tree (Fig. 4A) showed strong support (100% bootstrap) for monophyly of the eurosid I clade because of the sister relationship between the Fabales and Cucurbitales. In contrast, the ML tree indicated that eurosids I are paraphyletic because Cucurbitales were sister to Mrytales rather than Fabales (Fig. 4B); bootstrap support for this relationship was also strong (92%).

The organization of the Vitis genome with two copies of an IR separating the SSC and LSC regions is identical to most sequenced angiosperm chloroplast genomes [reviewed in 41]. The size of the genome at 160,928 bp is also within the known size range for angiosperms, which generally vary from 150,519 (Lotus 42) to 162,686 bp (Amborella 3) among photosynthetic genomes from dicots that have both copies of the IR. Size of the Vitis IR at 26,358 bp is also well within the size range of other sequenced dicot genomes, which range from 23,302 (Calycanthus 43) to 27,807 bp (Oenothera 44). Gene content and order of the Vitis chloroplast genome is virtually identical to tobacco and many other unrearranged angiosperm chloroplast genomes. Several previously sequenced rosid chloroplast genomes have lost the rpl22 gene, including legumes 4546474849. The distribution of this loss on the chloroplast phylogeny (arrows in Fig. 4A) indicates that there have been at least two independent losses of rpl22 in rosids. Multiple, independent gene losses in angiosperms have been demonstrated for other genes including infA 50, rps16 48 and accD 5152. Thus, it is evident that gene losses are not always reliable indicators of phylogenetic relationships.

It is increasingly evident that chloroplast genomes contain repeated sequences other than the IR 49. Several studies have identified a higher incidence of dispersed repeats in genomes that have experienced extensive rearrangements 5354. However, dispersed repeats are also being detected in unrearranged genomes. In most cases, these repeats are more common in intergenic spacers and introns, which is also true for the Vitis genome. Repeats have been located in a number of other rosids 49 in the same regions as those identified in the Vitis genome. One of these, a 32 bp repeat in the trnS gene, is in the same location in Gossypium hirsutum 55, indicating that this repeat may be shared among rosids. Although repeats have been implicated in playing a role in chloroplast genome rearrangements 56, their effect if any in unrearranged chloroplast genomes is unknown.

Based on previous studies of Atropa 57 and tobacco 58, posttranscriptional RNA editing events, as well as deamination-facilitating attacks on nucleotides' exocyclic amino groups, yield primarily C-to-U alterations. Analyses of the Vitis chloroplast genome and the corresponding ESTs indicate that the five C-to-U changes likely represent mRNA edits. However, the remaining six differences could be either sequencing errors in the genomic DNA or EST sequences or due to the use of different cultivars and/or plants/tissues used for sequencing. Our methods eliminate the latter explanation since we only compared DNA and EST sequences from leaves of the chardonnay variety of Vitis vinifera. In view of the high depth of coverage (8X) of our genomic DNA sequences, we believe that the non C-to-U changes represent EST sequencing errors.

Evolutionary loss of RNA editing sites has been observed in earlier studies and could be attributed to a decrease in the effect of RNA-editing enzymes 59. Additionally, conversions other than C-to-U in Vitis and other plants suggest that chloroplast genomes may be accumulating a considerable number of nucleotide substitutions, and some genes might accumulate more changes than others, such as the petL and ndh genes that have a high frequency of RNA editing 60. Therefore, despite high levels sequence conservation in chloroplast genomes, variations do occur posttranscriptionally, promoting translational efficiency due to transcript-protein complex binding and/or changes in the chloroplasts microenvironment (e.g., like redox potential or light intensity 6162).

Phylogenetic analyses of 28 (Fig. 3) or 29 (Fig. 4) angiosperms based on 61 protein-coding genes identified many of the major lineages recognized in previous phylogenetic hypotheses of flowering plants [reviewed in 1]. Two groups, Amborella and Nymphaelaes (represented by Nuphar and Nymphaea) are basal, with Amborella forming the first diverging lineage in MP analyses and Amborella/Nymphaelaes together forming the most basal clade in ML trees. These results are congruent with recent 61-gene analyses by Leebens-Mack et al. 5 and support their contention that limited taxon sampling in earlier whole chloroplast genome phylogenies led some previous workers to suggest that Amborella may not be among the earliest diverging angiosperm lineages 23. Monophyly of the monocots is strongly supported, and they are sister to the remaining angiosperms. Calycanthus, the sole representative of the magnolids, is weakly supported as sister to eudicots in the MP analyses (Figs. 3A and 4A) but the genus is weakly supported as sister to a clade that includes both monocots and eudicots in ML trees (Figs. 3B and 4B). Monophyly of eudicots is strongly supported (100% bootstrap values), in agreement with phylogenies based on both pollen 6364 and other molecular data 13141819656667. Within eudicots, Ranunculales diverge first and are sister to a strongly supported eudicot clade that includes two moderately to well-supported groups comprising the rosids and asterids. The early divergence of Ranunculales among eudicots is in agreement with many recent molecular phylogenies [see chapter 5 in 1]. Although previous studies have clearly indicated that Carylophyllales belong in the core eudicot clade 1, resolution of the relationships of Caryophyllales to other major clades of eudicots remains uncertain. This order has been considered to be closely allied to rosids, asterids, or simply as an unresolved major eudicot clade sister to the Dilleniaceae 15. Although taxon sampling is limited in our 61 gene phylogeny, there is moderate to strong support for a sister relationship between the Caryophyllales and asterids (Figs. 3 and 4).

The rosid clade is very diverse, including nearly 140 families representing approximately 39% of the species of angiosperms. The most recent phylogenies of this group [summarized in chapter 8 in 1] indicate that there are seven major clades whose relationships still remain unresolved. Eight (Fig. 3) or nine (Fig. 4, includes Gossypium) representatives of four of these major clades are included in our phylogenetic analyses, including members of eurosids I, eurosids II, Myrtales, and Vitaceae. Phylogenetic analyses of both datasets using MP and ML clearly indicate that the Vitaceae is sister to the remaining rosids, and therefore represents an early diverging member of the rosid clade. Previous molecular phylogenetic comparisons that included Vitaceae could not resolve its relationship. Phylogenetic analyses of rbcL sequences alone placed the Vitaceae as sister to either the Caryophyllales or asterid clade with weak support 18. Phylogenies based on atpB provided only weak support for a sister relationship of Vitaceae to Saxifragales 14. Several phylogenies based on two to four genes suggested that the Vitaceae are sister to rest of the rosids, with relatively weak support (50–75%; 141516). However, phylogenies based on the chloroplast gene matK did not place Vitaceae sister to rosids but instead positioned the family as sister to Dilleniaceae with weak support 19. In short, the phylogenetic position of Vitaceae is equivocal, though our results strongly support the earlier findings that Vitaceae represent an early diverging clade within rosids (Figs. 3 and 4).

The two datasets we examined differed by only one taxon but the results of MP analyses differed dramatically regarding the placement of three of the four rosid clades examined (compare Figs. 3A and 4A). The 28-taxon dataset (excluding Gossypium) showed relationships that are incongruent with recent molecular phylogenies of rosids 1 by placing the eurosids II (represented by only Brassicales) sister to the Fabales in eurosids I. This made eurosids I paraphyletic because the other representative of this clade is Cucumis (Cucurbitales), which is sister to the Brassicales in molecular phylogenies of rosids 1. The addition of Gossypium in the 29-taxon dataset generates MP trees (Fig. 4A) that are congruent with previous angiosperm phylogenies. The Brassicales and Malvales are sister and there is strong support for the monophyly of eurosid II. The addition of Gossypium also makes the eurosid I clade strongly monophyletic in the MP tree by placing the Cucurbitales sister to the Fabales, both of which are members of the nitrogen-fixing clade [see chapter 8 in 1]. In contrast to the MP trees, relationships among the major rosid clades do not differ in the ML trees when Gossypium is added. In both the 28 and 29-taxon data sets the ML trees do not support the monophyly of eurosids I since Cucurbitales (eurosid I) are sister to the Myrtales and Brassicales (eurosid II) are sister to the Fabales (eurosid I). Therefore, the ML analyses are incongruent with currently accepted relationships among rosids 1, though the strongest support for the monophyly of eurosid I clade is only 77% (jackknife support) in a three-gene analysis 15. Thus, our results suggest that additional phylogenetic studies are needed to assess the monophyly of eurosids I and their relationship to other rosids.

There has been considerable debate regarding the utility of whole genome sequences for phylogeny reconstruction 578910. Some have argued that the use of more genes from whole genomes has great potential for providing much more data for resolving phylogenetic relationships 268, whereas others have suggested that problems with limited taxon sampling available for whole genomes 57810 and model misspecification 411 overshadows any potential advantages. One example that highlighted each of these concerns centered around the controversy regarding identification of basal angiosperms. Leebens-Mack et al. 5 demonstrated that inadequate taxon sampling clearly played a role in misleading some previous studies, and Goremykin et al. 4 demonstrated that ML analyses of whole chloroplast genome data sets can be sensitive to model specification. It is well known that ML methods fail when model parameters are misspecified 697071. The phylogenetic analyses in this study provide yet another example of these phenomena. Addition of the Gossypium genome to our parsimony analyses generated trees that are congruent with current understanding of relationships among the major rosid clades. However, the ML analyses are incongruent with the MP trees regarding the monophyly and relationships of the rosid clades and support for the alternative relationships was very strong in each case (compare Figs. 4A and 4B). It is possible, if not likely, that the use of a single "average" model (GTR + I + Γ) in the ML analyses is inappropriate for a data set of 61 concatenated genes [see 11 for a discussion of this issue]. Future phylogenetic analyses of complete chloroplast genome sequences should consider using methods in which different models can be applied to different partitions of the data (e.g., genes, codon positions, functional groups) 72. Development of more appropriate models of evolution of chloroplast sequences 12 may also improve the accuracy of phylogenies based on these genomes. Thus, we need more extensive sampling of whole chloroplast genomes from the major lineages of flowering plants and more rigorous phylogenetic analyses before the full potential of this approach can be realized. Ongoing projects by several labs [see 73 for a list of some of these] should greatly enhance our taxon sampling so that we can generate reliable phylogenies based on whole chloroplast genomes.

The bacterial artificial chromosome (BAC) library of grape was constructed by ligating size-fractionated partial Hind III digests of total cellular, high molecular weight DNA with the pINDIGOBAC vector. The average insert size of the grape library is 144 kb. BAC-related resources for this public library can be obtained online from the Clemson University Genomics Institute BAC/EST Resource Center 74.

BAC clones containing the chloroplast genome inserts were isolated by screening the library with a soybean chloroplast probe. The first 96 positive clones from screening were pulled from the library, arrayed in a 96-well microtitre plate, copied, and archived. Selected clones were then subjected to Hind III fingerprinting and Not I digests. End-sequences were determined and localized on the chloroplast genome of Arabidopsis thaliana to deduce the relative positions of the clones, then clones that covered the entire chloroplast genome of grape were chosen for sequencing.

The nucleotide sequences of the BAC clones were determined by the bridging shotgun method. The purified BAC DNA was subjected to hydroshearing, end repair, and then size-fractionated by agarose gel electrophoresis. Fractions of approximately 3.0–5.0 kb were eluted and ligated into the vector pBLUESCRIPT IIKS+. The libraries were plated and arrayed into 40 96-well microtitre plates for the sequencing reactions.

Sequencing was performed using the Dye-terminator cycle sequencing kit (Perkin Elmer Applied Biosystems, USA). Sequence data from the forward and reverse priming sites of the shotgun clones were accumulated. Sequence data equivalent to eight times the size of the genome was assembled using Phred/Phrap programs 75.

The Vitis genome was annotated using DOGMA (Dual Organellar GenoMe Annotator) 76, after uploading a FASTA-formatted file of the complete plastid genome to the program's server. BLASTX and BLASTN searches against a custom database of previously published plastid genomes identified Vitis' putative protein-coding genes, and tRNAs or rRNAs. For genes with low sequence identity, manual annotation was performed, after identifying the position of the start and stop codons, as well as the translated amino acid sequence, using the plastid/bacterial genetic code.

REPuter 77 was used in order to locate and count the direct (forward) and inverted (palindromic) repeats within the grape chloroplast genome. For repeat identification, the following constraints were set to REPuter: (i) minimum repeat size of 30 bp, and (ii) 90% or greater sequence identity, based on Hamming distance of 3 49. Manual verification of the identified repeats was performed in EditSeq, while performing intragenomic blast search of the identified repeat sequence.

Each of the gene sequences from the grape chloroplast genome was used to perform a BLAST search of expressed sequence tags (ESTs) from Genbank. In order to incorporate specificity into our analyses, the matching ESTs had to meet all of the following criteria: (1) belong to a Vitis vinifera cv., (2) belong to the chardonnay variety, and (3) come from leaf tissue. Due to length variations between the screened ESTs and the related gene, the retrieved EST with the highest bit score was selected for further analyses. The retrieved Vitis vinifera EST was aligned with the corresponding annotated gene using ClustalX 78, followed by screening for nucleotide and amino acid changes using Megalign and the plastid/bacterial genetic code. Because of variations in the length between an EST and the related gene, the length of the analyzed sequence was recorded.

The 61 genes included in the analyses of Goremykin et al. 23 and Leebens-Mack et al. 5 were extracted from our new chloroplast genome sequences of Vitis using the organellar genome annotation program DOGMA 76. The same set of 61 genes was extracted from chloroplast genome sequences of six other recently sequenced angiosperm chloroplast genomes, including tomato, potato, soybean, cotton, cucumber, and Eucalyptus (see Table 3 for complete list of genomes examined). In general, alignment of the DNA sequences was straightforward and simply involved adding the 61 genes for the new angiosperms to the aligned data matrix from Leebens-Mack et al. 5. In some cases, small in-frame insertions or deletions were required for correct alignment. For two genes, ccsA and matK, the DNA sequences were more divergent, requiring alignment using ClustalX 78 followed by manual adjustments. The complete nucleotide alignment is available online at 79.

Phylogenetic analyses using maximum parsimony (MP) and maximum likelihood (ML) were performed using PAUP* version 4.10 80 on two data sets, one including 28 taxa and a second including 29 taxa by the addition of Gossypium. Phylogenetic analyses excluded gap regions. All MP searches included 100 random addition replicates and TBR branch swapping with the Multrees option. Modeltest 3.7 81 was used to determine the most appropriate model of DNA sequence evolution for the combined 61-gene dataset. Hierarchical likelihood ratio tests and the Akaikle information criterion were used to assess which of the 56 models best fit the data, which was determined to be GTR + I + Γ by both criteria. For ML analyses we performed an initial parsimony search with 100 random addition sequence replicates and TBR branch swapping, which resulted in a single tree. Model parameters were optimized onto the parsimony tree. We fixed these parameters and performed a ML analysis with three random addition sequence replicates and TBR branch swapping. The resulting ML tree was used to re-optimize model parameters, which then were fixed for another ML search with three random addition sequence replicates and TBR branch swapping. This successive approximation procedure was repeated until the same tree topology and model parameters were recovered in multiple, consecutive iterations. This tree was accepted as the final ML tree (Figs. 3B, 4B). Successive approximation has been shown to perform as well as full-optimization analyses for a number of empirical and simulated datasets 82. Non-parametric bootstrap analyses 83 were performed for MP analyses with 1000 replicates with TBR branch swapping, 1 random addition replicate, and the Multrees option and for ML analyses with 100 replicates with NNI branch swapping, 1 random addition replicate, and the Multrees option.