Evolution of Oryza chloroplast genomes promoted adaptation to diverse ecological habitats

The course, tempo and mode of chloroplast genome evolution remain largely unknown, resulting in limited knowledge about how plant plastome gene and genome evolve during the process of recent plant speciation. Here, we report the complete plastomes of 22 closely related Oryza species in chronologically ordered stages and generate the first precise map of genomic structural variation, to our knowledge. The occurrence rapidity was estimated on average to be ~7 insertions and ~15 deletions per Myr. Relatively fewer deletions than insertions result in an increased repeat density that causes the observed growth of Oryza chloroplast genome sizes. Genome-wide scanning identified 14 positively selected genes that are relevant to photosynthesis system, eight of which were found independently in shade-tolerant or sun-loving rice species. psaA seemed positively selected in both shade-tolerant and sun-loving rice species. The results show that adaptive evolution of chloroplast genes makes rice species adapt to diverse ecological habitats related to sunlight preferences.

C hloroplasts were derived from endosymbiosis between independent living cyanobacteria and a nonphotosynthetic host 1 . Each has its own genome that is usually nonrecombinant and uniparentally inherited 2 . Chloroplast DNA (cpDNA) have increasingly been employed for resolving the deep phylogeny of plants because of their low rates of nucleotide substitutions and decelerated structural variation compared with nuclear genomic sequences [3][4][5] . Thus, it has been commonly recognized that the chloroplast genomes are prominently conserved in terms of genome size, genomic structure, and gene content among the majority of flowering plant species 4,6,7 . However, the course, tempo and mode in the evolution and genome architecture changes in the context of the recent speciation remain largely unknown, resulting in limited knowledge about the micro-evolution of chloroplast genes and genomes in flowering plants. The deep comparisons of closely related plant species in chronologically ordered different stages are able to dramatically improve the sensitivity of evolutionary inferences with genomic structural knowledge to thoroughly elucidate the mechanisms, rates, or directionality of chloroplast genome evolution. Such an effort could precisely generate more convincing results needed to examine patterns of genomic architectures during the speciation process than single chloroplast genome analysis can offer. With this regard, we know little about broadscale patterns of evolutionary dynamics, discrepancy, and consequences that will provide in-depth insights into microstructural genome diversification and unravel mechanisms underlying the maintenance or disruption of microstructural organization. The availability of high-quality complete chloroplast genome sequences will offer an unprecedented opportunity to address these questions. It is required to initiate comparative analyses among multiple chloroplast genomes of very closely related plant species that span the speciation continuum under a well-resolved phylogenetic framework.
The genus Oryza is one of ideal groups to understand the diversity and evolution of chloroplast genes and genomes in flowering plants. It belongs to the grass family (Poaceae) and consists of more than 20 wild species and two cultivated species 8,9 . These rice species have been assigned as 10 genome types (AA, BB, CC, EE, FF, GG, BBCC, CCDD, HHJJ, and HHKK) [10][11][12] , representing an enormous gene pool for genetic improvement of modern rice cultivars. Molecular phylogenetic and phylogenomic studies in this genus have well-resolved evolutionary relationships and reconstructed the phylogeny of different rice genome types and species 10,[12][13][14][15][16][17][18] . Besides cpDNA sequences that have been applied to determining phylogenetic relationships among Oryza species 12,[19][20][21][22][23] [24][25][26][27][28][29][30] , and for example, the fully sequenced rice chloroplast genomes 24,31 have been extensively used to phylogenetic and population genetic studies in cultivated rice and wild relatives [32][33][34] . The genus Oryza particularly becomes an unparalleled system for studying comparative and evolutionary genomics in plants owning to available nuclear genomes of the two cultivated rice subspecies [35][36][37] and the other six Oryza species having been sequenced recently 18,38 . These species with well-defined phylogenetic framework and splitting times may undoubtedly serve as an important model that deserves endeavors for obtaining an in-depth view of chloroplast gene and genome evolution.
Next-generation sequencing techniques have revolutionized DNA sequencing via high-throughput capabilities and low costs 39 , making it now fairly convenient to obtain multiple highquality plastome sequences. Here, we present the complete plastome sequences of 21 Oryza species and Leersia japonica from the tribe Oryzeae as outgroup. Our sampling representation allows us to infer the tempo and mode in Oryza plastome gene and genome evolution. To this end, we performed phylogenomic analysis based on whole-plastome sequences of almost all Oryza species to clarify evolutionary relationships among rice species towards obtaining overall insights into the rice chloroplast genome evolution. Understanding phylogenetic relationships of rice species will greatly enhance our efforts to search for favorable genes. Besides, maternally inherited nature of organelle genomes makes them potentially useful tools to identify diploid donor of the polyploidy species. Thus, another objective of this study was to elucidate the evolutionary relationships of the Oryza chloroplast genomes with emphasis on the origin and evolution of the allotetraploid rice species. In addition, all these cp genomes were pairwise compared to genome-wide identified structural variants, and such a comprehensive map of genomic variation across Oryza plastomes permits, to our knowledge, to gain the first vision about the tempos of the InDels that came into and/or removed from the chloroplast genomes during recent speciation. Specifically, we investigated the evolutionary dynamics of repeat sequences by means of assessing the pace of the occurrences and turnover as well as the magnitude of conservation since the generation along the timetabled Oryza phylogeny. We also evaluated how far-reaching structural variation has affected the evolution of chloroplast protein-coding genes; and finally, we scanned for chloroplast rice genes that evolve under Darwinian selection. These candidate genes of wild rice in particular are expected to play an important role in adaptations to their ecological niches in Asia, South America, Africa, and Australia, and become a valuable database of functional variation for the future rice improvement programs.

Results
Genome size, gene content and structure of Oryza plastomes. We sequenced and assembled plastomes of a total of 21 Oryza species and one closely related species, Leersia japonica, from the tribe Oryzeae (Supplementary Table 1). Illumina paired-end reads, after excluding non-chloroplast genome reads, were first mapped to the O. sativa ssp. japonica chloroplast genome (NC_001320) (AA-genome) as a reference, and these 21 chloroplast genomes (cp genomes) were generated with genome sizes varying from 134,074 bp (L. japonica) to 136,133 bp (O. meyeriana) (GG-genome; Supplementary Table 2). These 22 newly sequenced cp genomes in this study together with two additional O. sativa cp genomes (AA-genome; O. sativa ssp. japonica, NC_001320; O. sativa ssp. indica, NC_008155) 24,31 exhibited well-maintained quadripartite structure; they consists of two inverted repeat regions (IRa and IRb) separated by large (LSC) and small (SSC) single-copy regions, typically resembling to the majority of flowering plant cp genomes 40 (Supplementary Fig. 1). They displayed similar GC contents of~39%. In comparisons with the relatively conserved sequences in the coding or IR regions, non-coding regions of the investigated cp genomes were apparently variable ( Supplementary Fig. 2).
All Oryza cp genomes displayed a conserved gene content and commonly harbored an identical set of 111 annotated unique genes, including 77 unique protein-coding genes, 30 tRNA genes, and four rRNA genes (Supplementary Fig. 1; Supplementary  Table 2), representing typical features of land plant cp genomes 7,41,42 . All four rRNA genes were located in the IRs, 21 tRNA genes were found in the single-copy region, and the others were positioned in the IRs. Of the 15 genes that contained introns, a total of the 14 genes (trnK-UUU, rps16, atpF, trnL-UAA, trnV-UAC, petB, petD, rpl16, rpl2, ndhB, trnI-GAU, trnA-UGC, ndhA, and rps12) had one intron with the exception of ycf3 with two introns. rps12 was trans-spliced with one of its exons located in the LSC regions (5′_end) and the other existed in the IR regions (3′_end), which were separated by a long intron. A pair of the ndhH genes was annotated, which stretched across the IR/SSC junctions with the 5′ end of ndhH located in SSC.
The occurrences of gene transfer and loss have been commonly observed in plant cp genomes [43][44][45][46] . accD, ycf1, ycf2, rpl23, rpl22, infA, rps16, and ycf4 were reported partially or completely lost from the legume chloroplast genomes 47,48 , of these, some genes such as infA even have been lost for multiple times. To our knowledge, this is the first effort using such an extensively representative cp genomes to further understanding of patterns of gene loss among the closely related plants; in this case, we observed that ycf1, ycf2, and ycf15 were lost from these sequenced Oryza cp genomes. accD, coding for a subunit of acetyl-CoA carboxylase, was absent from the T. subterraneum cp genome 48 . Sequence alignments of all annotated accD revealed a remarkably different gene structure and protein-coding sequences that were unlike those in other angiosperms ( Supplementary Fig. 3), which may be suggestive of a novel gene structure. Raubeson et al. 49 proposed that ycf68 is a pseudogene in Ranunculus. In the genus Oryza, one SNV and two InDels were found within the ycf68 gene at the sites of 49 bp, 140 bp, and 289 bp, respectively (Supplementary Fig. 4), resulting in frameshift mutations and premature termination codes in the functional Ycf68 protein (Supplementary Fig. 5).
In addition to an overall conservation of gene content, the sequenced Oryza cp genomes were structurally conserved and displayed an almost uniform order of the same set of genes with slight differences at boundary junctions ( Supplementary Fig. 1). This is not exceptional as a slight IR expansion/contraction was also reported in other cp genomes 40,50,51 . In the junctions between IRs and SSCs, for example, ndhH genes marginally extended into the IR regions; the distances of rps19 from the junction of LSC/IRb varied from 40 bp in O. sativa ssp. indica to 48 bp in O. australiensis (EE-genome); a trivial difference in junction positions was also observed among the 22 Oryza and L. japonica chloroplast genomes ( Supplementary Fig. 6).
Phylogenomic analysis based upon whole plastome sequences and Oryza chloroplast genome evolution. One of advantages to have sequenced chloroplast genomes of almost all Oryza species is to explore evolutionary relationships that allow a full picture of the rice plastome genome evolution. We concatenated the chloroplast whole-genome sequences into an alignment dataset comprising of 139,206 bp in length after removing all indels across the studied species. Of them, variable and phylogenetically informative sites were 5917 bp (4.25%) and 2766 bp (1.99%), respectively (Supplementary Table 3). Using L. japonica as outgroup, maximum likelihood (ML), maximum parsimony (MP), and Bayesian inference (BI) methods yielded the fully resolved phylogenetic trees that are nearly identical to each other with almost 100% bootstrap support values or 1.00 posterior probabilities ( Fig. 1; Supplementary Figs. 7 and 8). In the resulting phylogenies, almost all Oryza species were grouped into clusters corresponding to different genome types, and the studied species formed genome type-based monophyletic clades. The genus Oryza was divided into the two main clades of which one branch comprised AA-, BB-, BBCC-, CC-, CCDD-, and EE-genome species, while the other included FF-, GG-, and HHJJ-species. The ML phylogenetic trees were further reconstructed using noncoding genomic sequences ( Supplementary Fig. 9) and a total of 77 protein-coding genes ( Supplementary Fig. 10), respectively. They were consistent to each other and largely congruent with the above-described topology based on the entire chloroplast genome sequences, strongly supporting the previous recognition of different genome types in both rice diploids and tetraploids by studying the meiotic pairing of hybrids between Oryza species 52,53 and phylogenetic analysis 12 . The topological differences were merely observed when using protein-coding gene sequences, indicating that O. brachyantha (FF-genome) independently formed a monophyletic and failed to group with the GG-and HHJJ-genome species. This robust phylogeny is generally consistent with the phylogeny of Oryza species in earlier studies 12,18 , further providing solid grounds for performing evolutionary and comparative analyses. We further reconstructed phylogenetic trees using genomic indels ( Supplementary Fig. 11) and indels presented in protein-coding genes ( Supplementary  Fig. 12). Compared to several topological differences (e.g., O. nivara and O. glumapaetula) of the phylogeny using genomic SVs ( Supplementary Fig. 11), the SV phylogeny reconstructed using indels presented in protein-coding genes is quite consistent with the above-described SNV phylogeny topology with almost 100% bootstrap support ( Supplementary Fig. 12).
Whole chloroplast genomes provide a wealth of information to determine the phylogenetic relationships of the eight AAgenome Oryza species using L. japonica as outgroup ( The genus Oryza constitutes more than one-third of rice species that are natural allotetraploids originated from two distinct diploid donors through hybridization events coupled with doubled chromosome numbers. Due to the maternally inherited nature of plant chloroplasts, obtaining the chloroplast phylogenomic tree that represents a maternal genealogy enables us to determine the maternal parents of allotetraploid species. To infer origins of allopolyploid rice species, the phylogenomic tree was reconstructed based on complete chloroplast genome sequences of the majority of Oryza species ( Fig. 1 bootstrap support values or 1.00 posterior probabilities, suggesting a significant divergence between the former two species and O. latifolia during the process of speciation and subsequent evolution. This result is seemingly in good agreement with the taxonomic treatment of Roschevicz 57 that these American tetraploids as two separate species (O. grandiglumis and O. latifolia) (CCDD-genome) but multiple characters are needed by morphological, biogeographical, cytological, and genomic evidences. In addition, our results reveal that the allotetraploid O. minuta (BBCC-genome) presented a closer relationship with the diploid O. puctata (BB-genome), suggesting that the BB genome was the maternal donor during the formation of the BBCC genome species. Nevertheless, the BBCC genome species (O. eichingeri) revealed a different origin with maternal parent being the CC genome 12 . A robust reconstruction of origins of all allotetraploid rice species requires to complete chloroplast genome sequences of remaining tetraploids including O. schlechteri and O. coarctata and put extra efforts to ensure the diploid species with the DD, HH, JJ, or KK genomes that are either undiscovered or disappeared.
A genome-wide map of genomic structural variation across Oryza plastomes. Global sequence alignments of the 23 Oryza and L. japonica chloroplast genomes identified several variants, indicative of evolutionary processes in rice. With L. japonica as outgroup, we characterized genomic variation including SNVs, insertions, and deletions (InDels) in the Oryza chloroplast genomes, revealing that they varied from one species to another (Supplementary Tables 4 and 5). On the whole-genome level, the numbers of SNVs, deletions, and insertions per kb varied from one to another with average values of 16.22, 1.15, and 1.26, respectively. In terms of these three types of genomic variation, the IR regions presented the fewest numbers per kb with average values of 3.16, 0.43, and 0.41, respectively. The SSC regions exhibited the utmost numbers of SNVs and insertions per kb with average values of 26.43 and 1.84, respectively. Meanwhile, the number of deletions per kb was the largest in the LSC region with an average value of 1.34. These results together make evident that the IR regions were more conserved than single-copy regions, and that chloroplast genes were more conserved than the intergenic regions ( Supplementary Fig. 2). Figure 2, Supplementary Figs. 2 and 13 provide an overview of genomic structural variation across Oryza plastomes, ascertaining trivial variation but overall conservation of genome architecture across these rice species. To examine the occurrences of genomic structural variants along different lineages that account for the evolution of Oryza plastomes, we mapped all genomic structural variants onto the phylogenetic tree of the 23 Oryza chloroplast genomes using L. japonica as outgroup. These InDels were inferred by aligning 24 chloroplast genome sequences using a phylogeny-aware algorithm PRANK +F 58 , which has been proven to outperform traditional methods with regard to the accuracy in InDel-rich sequences. A total of 730 insertions were identified that varied from 1 to 234 bp in length. This summed to 4908 bp of insertions, compared to 994 deletions, ranging from 1 to 608 bp in length that affect 12,592 bp (Supplementary Tables 6 and 7; Supplementary Fig. 14A; Fig. 3a). The majority of detected InDels were small in size;~99% of all InDels were <100 bp, whereas~82% were <10 bp. The size distribution of structural variations is consistent with findings in plant nuclear genomes 18,59 that longer variants were less abundant ( Supplementary Fig. 14B).
To estimate the pace of these InDels in the Oryza plastomes, we dated the occurrences of structural variants by locating them at the timetabled phylogenetic nodes. The results showed that, during the past~15.3 Mya in the middle Miocene 60 , the 24 lineages as well as their common ancestral branches, on average, had~7 insertions and~15 deletions per Myr, respectively ( Supplementary Fig. 14A). We estimated insertion and removal rates of InDels based on the previously estimated times 17,18,60 ( Fig. 3b; Supplementary Fig. 14C). Our results further revealed that insertions and deletions remarkably varied across different lineages. Nevertheless, the estimated rates of either insertions or  Fig. 2 An overview of chloroplast genome variation across the Oryza plastomes. The InDel density of 2Kb was measured for the 23 Oryza chloroplast genomes using L. japonica as outgroup. In the quadripartite structure of these chloroplast genomes, the two IR regions (IR A and IR B ) are shown with gray background, while the large and small single-copy regions (LSC and SSC) are displayed with blank background. The deletions are filled with green ridges while the insertions are occupied with yellow ridges. The studied Oryza species are indicated as below:  Fig. 3 The occurrence rapidity of chloroplast genomic variation during speciation in the genus Oryza. a Accumulation rates of insertion and deletion lengths every million years along branches of the Oryza phylogeny. Pie of top 10 branches are scaled proportionally to InDel lengths. Divergence times (Myr) of major branches are shown by following estimates by Zou et al. 60 , Zhu et al. 17 and Zhang et al. 18 ; b Total lengths of insertion and deletion for different lineages of the genus Oryza. Detailed information of these lineages is shown in Supplementary Fig. 12C deletions were mostly constant along these lineages except for, for examples, O. brachyantha (FF-genome), O. glumaepatula (AAgenome), and O. nivara (AA-genome) with elevated rates and O. barthi (AA-genome), O. alta (CCDD-genome), and O. officinalis (CC-genome) with lowered rates (Fig. 3b; Supplementary  Fig. 14C). The finding is consistent to the result that the numbers of genomic structural variants observed along different branches were largely proportional to branch lengths ( Supplementary  Fig. 14D). This is well consistent to the observation in the six AAgenome Oryza nuclear genomes that the occurrences of structural variation events are closely related to their phylogenetic positions and divergence times 18 .
The corresponding genomic positions of these identified InDels were mapped and located into the 23 Oryza plastomes, displaying that 92% of the predicted InDels were found in intronic (14%) and intergenic regions (78%) (Supplementary Table 8; Supplementary Fig. 15A). As expected, InDels were scarce in exons of protein-coding genes (4%) and RNA genes (4%). The uneven distribution of genomic structural variants throughout the plastomes suggests that they are likely to have negative impacts and thus easily eliminated by purifying selection.
Far-reaching structural variation may have affected not only genomic architectural heterogeneities but also the evolution of protein-coding genes in the Oryza chloroplast genomes. To examine whether this holds true, we characterized the length distribution of insertions and deletions within chloroplast protein-coding genes. Our results showed that, of the 71 InDels, there were merely 27 with lengths that are multiples of three ( Supplementary Fig. 15B). This finding suggests that negative selection on frameshift InDels may not really affect the chloroplast protein-coding genes, which is sharply against the observation in the six nuclear genomes of Oryza 18 and other flowering plants 38,59 . The prevalence of genomic structural variants within chloroplast genes were further characterized by mapping them into the exons of protein-coding and RNA genes of the Oryza chloroplast genomes (Supplementary Tables 9 and  10; Supplementary Fig. 15C; Supplementary Figs. [16][17][18], indicating that InDels serve as an important driver of rice chloroplast gene evolution.
Evolutionary patterns and tempo of cp genes and genomes have long been studied on a large scale, obtaining the knowledge of macro-evolution of cp genes and genomes in angiosperms 4,6,61 . With this regard, this is the first case study, to our knowledge, providing insight into the micro-evolution of cp genes and genomes in the course of recent speciation in flowering plants.
Evolutionary dynamics of plastid repetitive DNA and the growth of Oryza plastomes. Although the remarkable feature of a chloroplast genome is the conservation of its most prominent structure and genome size 62 , our results showed that there was a gradual growth of rice chloroplast genomes in size with an increase in genetic distance from cultivated rice in the genus Oryza ( Fig. 1; Supplementary Table 2). To determine the contribution of repetitive DNA sequences to genome variation and evolution, we investigated their evolutionary dynamics across these closely related chloroplast genomes. In total, we identified 112 repeats in the 23 Oryza and L. japonica cp genomes that are under the three categories, namely palindromic (Rp), dispersed (Rd), and tandem (Rt) repeats (Supplementary Table 11). Among them, the shortest 61 repeats that ranged from 15 to 29 bp in length occurred most frequently (Supplementary Fig. 19A). Although the vast majority of repeats were located within noncoding regions, the rest were found in a number of proteincoding or tRNA genes (Supplementary Fig. 19B). Comparative genomic analyses revealed that evolutionary behaviors of the three types of repeat sequences (Rp, Rd, and Rt) were contrastingly different across Oryza cp genomes. Here we define repetitive DNA sequences that locate in orthologous genomic regions with similar lengths as shared repeats. In contrast to Rd and Rp repeats that mainly generated in the common ancestor and have subsequently preserved afterwards during the past~15.3 Myr 60 , Rt repeats were fairly evolutionarily dynamic, representing the abundance of copy number variation of chloroplast repeats across Oryza lineages (Fig. 4a; Supplementary Table 12). Our visual examination of the conserved Rd and Rp repeats shared by all 24 Oryzeae cp genomes accurately documented numerous single or multiple independent SNVs and/or deletions across Oryzeae lineages, providing convincing evidence for that Rp evolve faster than Rd repeats within these rice plastomes ( Fig. 4a; Supplementary Table 12). Comparisons of genomic variation levels (SNVs and InDels) within Rt, Rp, Rd, and their flanking regions with genome background revealed that Rp and flanking regions exhibited high levels of both SNVs and InDels, Rt and flanking regions only exhibited high levels of InDels, and Rd and flanking regions showed no difference in both SNVs and InDels (Fig. 4b).
The findings together suggest that, in comparison the conserved Rd repeat, Rt and Rp repeats are highly variable that may serve as important but different forces to trigger rice chloroplast genome variation and evolution.
Although molecular mechanisms of how a novel repeated sequence de novo generates in a chloroplast genome remain largely unresolved, our results indicate that the tempos of repeat insertions are quite different among these closely related rice species while they evolve (Fig. 4a). The majority of repeat insertion events occurred at differently timetabled nodes, that is, they may be well illuminated by their phylogenetic relationships; Rt5 may represent a case that occurred independently in O. sativa ssp. indica and O. rufipogon (Fig. 4c). To look into the rapidity of repeats inserted into or removed from the Oryza plastomes, we approximately mapped the occurrences at the timetabled phylogenetic tree and estimated the speed of their insertion into and/or removal from the Oryza plastomes (Fig. 4a). The detection of 54 insertion and 10 deletion events suggests the accumulation of repeat sequences in rice plastomes during the past~15.3 Myr 60 .
In sharp contrast of recent and rapid evolution of transposable elements observed in Oryza nuclear genomes 18,38,63 , the majority of chloroplast Rp and Rd repeats possess fairly slow turnover rates since the generation in the common ancestor of these Oryza species, except for Rt repeats that evolve quickly. Of the three repeat types, Rt, accounting for 40.71% of total repeats, stood out as the most common one, while Rp and Rd were 36.28% and 23.01% of overall repeats, respectively ( Supplementary Fig. 19C). These findings reveal an increased density of chloroplast repeats (e.g., Rt in particular) in Oryza chloroplast genomes as a result of relatively fewer deletions than insertions after they came into chloroplast genomes, may partially account for the growth of chloroplast genomes in the genus Oryza ( Fig. 1; Supplementary  Table 2).
The availability of the 24 Oryzeae complete chloroplast genomes makes it also possible to study the molecular evolution of rice chloroplast SSRs (Supplementary Table 13; Supplementary  Fig. 20). Genome-wide characterization of SSRs showed that numbers of the four types of SSRs, including dinucleotide, trinucleotide, tetranucleotide, and pentanucleotide motifs, ranged from 70 to 81 with an average density of~0.55 SSRs per kb. Numbers of all SSRs varied from 5 to 13, and the average density was 0.07 SSRs per Kb across the 24 genomes. Among the four motifs of SSRs, trinucleotide SSRs were the most abundant with the occurrence frequencies of~52.7-59.21%, while pentanucleotide SSRs were the rarest (0-1.39%; Supplementary Table 14; Supplementary Fig. 20). Comparison analyses of Oryza cp genomes detected a total of 17 specific SSRs that are indicative of their incessant generation although tempos were quite different among these closely related rice species. Of the 30 SSRs that were shared by at least two chloroplast genomes, 17 SSRs were shared by more than 14 Oryzeae cp genomes, suggesting their relatively early insertions, the majority of which may be well illuminated by their phylogenetic relationships. Note that these conserved SSRs varied by repeat length polymorphisms together with SNVs and InDels (Supplementary Table 14    Genome-wide scanning for footprints of positively selected chloroplast genes and adaptive evolution of sun-loving and shade-tolerant Oryza species. Non-synonymous (dN), synonymous (dS) nucleotide substitution rates and non-synonymous/ synonymous rate ratio (ω = dN/dS) are widely used as indicators of accelerated evolution and positive selection or adaptive evolution. The elevation of ω ratios can be due to either by adaptive evolution or relaxation of selective constraints. To examine molecular evolution of chloroplast protein-coding genes in the genus Oryza, we first calculated dN, dS, and ω along different lineages for all 77 rice chloroplast protein-coding genes. We observed that ω values slightly varied among these Oryza lineages, which is in a good agreement with the phylogenetic relationships ( Fig. 1; Supplementary Figs. 7-10 Table 15).
To identify the positively selected chloroplast genes that may be beneficial in rice breeding programs, we performed CodeML and a series of LRTs to calculate the ratio of dS and dN changes at each codon on particular branches or clades of interest in the Oryza phylogeny for all 77 chloroplast genes. Applying the likelihood method and a P-value of 5% for statistical significance (FDR < 0.05), we convincingly obtained a total of 14 nonredundant positively selected genes (PSGs) in all tests (accD, matK, ndhD, ndhF, ndhH, psaA, psbB, psbD, psbH, rbcL, rpl16, rpoA, rpoC2, and ycf68; Supplementary Table 16), accounting for 17.95% of all examined chloroplast genes. Note that the 14 PSGs (accD, matK, ndhD, ndhF, ndhH, psaA, psbB, psbD, psbH, rbcL, rpl16, rpoA, rpoC2, and ycf68) detected by using branch/cladespecific LRTs covered all eight PSGs (accD, matK, ndhF, ndhH, psbH, rbcL, rpoC2, and ycf68) with site model tests for all branches (Supplementary Table 16). A total of 10, 2, and 2 PSGs were found in the LSC, SSC, and IR regions, respectively (Supplementary Table 17), suggesting that rice chloroplast genes located in the LSC region is prone to being under positive selection. These 14 PSGs were functionally assigned as three selfreplication genes, eight photosynthesis genes, one other gene, and two genes with unknown functions, indicating that the majority of positively selected genes are relevant to photosynthesis system (Supplementary Table 18). These proteins-coding genes often function chloroplast protein synthesis, gene transcription, energy transformation and regulation, and particularly photosynthesis (Supplementary Table 19).
The genus Oryza includes both sun-loving and shade-tolerant rice species, providing an ideal model to understand the role of natural selection in driving the chloroplast genome evolution and identify key chloroplast genes underlying adaptive evolution of very closely related plant species. Strikingly, the photosynthesis gene ndhD has shown signatures of positive selection in typically shade-tolerant rice species, including O. longiglumis  Table 20; Supplementary Fig. 21). Moreover, the four other photosynthesis genes, rbcL, ndhH, psbD, psbH and/or unknown functional gene accD were found to be under positive selection in either the above-mentioned shade-tolerant species or other rice species that naturally prefer to grow in shade habitats, including O. officinalis (CC-genome), O. rhizomatis (CC-genome), O. eichingeri (CC-genome), O. alta (CCDD-genome), O. grandiglumis (CCDD-genome), and O. latifolia (CCDD-genome) (Fig. 5). Remarkably, photosynthesis genes (ndhF, psbB) and selfreplication genes (rpoC2, rpl16) were under strong positive selection in sun-loving rice species, such as Asian cultivated rice O. sativa, African cultivated rice O. glaberrima, and other AAgenome species (Fig. 5; Supplementary Table 20; Supplementary  Fig. 21). Besides these lineage-specific PSGs that are relevant to ecological adaptation to sun or shade habitats, the photosynthesis gene psaA appeared positively selected in both shade-tolerant and sun-loving rice species, probably displaying adaptive evolution under differently ecological conditions. Furthermore, speciesspecific tests unambiguously detected the four PSGs (ndhD, ndhF, psbB, and rpoC2; Fig. 5; Supplementary Table 21), strongly supporting that they are under independent natural selection. Of them, ndhF and psbB were under positive selection exclusively in O. sativa ssp. japonica, ndhD was positively selected in O. longiglumis (HHJJ-genome), and rpoC2 was positively selected in O. autraliensis (EE-genome). These chloroplast functional genes may have played key roles in the adaptation of rice species to ecological habitats related to sunlight preferences. Knowledge of chloroplast genes under positive selection may provide valuable clues for understanding how these rice species adapt to diverse ecological niches in Asia, South America, Africa, and Australia. Moreover, the detected genes or sites targeted by selection will offer more opportunities for further functional and evolutionary analyses.
In conclusion, we generated complete chloroplast genomes of 22 closely related Oryza species in chronologically ordered stages. We performed deep comparisons that permit us, to our knowledge, to make the first accurate map of genomic structural variation in the course of recent plant speciation. We estimated the occurrence rate on average to be~7 insertions and~15 deletions per Myr in the genus Oryza. We surprisingly found the swelling of Oryza chloroplast genome sizes, mainly due to the relatively fewer deletions than insertions that result in an increased repeat density. Comparative genomic analysis of repeat sequences further suggests that evolutionary dynamics of palindromic, dispersed, and tandem repeats exhibited remarkably dissimilar, probably due to either the diverged turnover tempos or evolutionary rates of SNVs and/or deletions after initial insertions. We accomplished genome-wide scanning and convincingly detected a total of 14 cp genes with strong signatures of positive selection, which are mainly related to photosynthesis system. We observed that eight of them were separately selected in shade-tolerant (ndhD, rbcL, ndhH, psbD, psbH, and accD) or sun-loving (ndhF and psbB) Oryza species, while psaA was positively selected in both shade-tolerant and sun-loving rice species. Such a comparative genomic analysis of the Oryza cp genomes offers us an unprecedented opportunity to reveal footprints of adaptive evolution of chloroplast genes that makes rice species adapt to diverse ecological habitats related to sunlight preferences. Thus, this study provides new insights into the course, tempo and mode of sequence evolution and architectural variations that will greatly further our understanding about the micro-evolution of cp genes and genomes in recent plant speciation. Chloroplast DNA isolation, amplification, and sequencing. We collected 50-100 g of fresh leaves from each studied Oryza species and L. japonica to isolate the purified cpDNA with an improved high salt method as reported previously 65 . They were amplified using a total of 14 primer pairs (Supplementary Table 22), which were designed according to the chloroplast sequence of O. nivara (NC_005973) 25 , and then PCR products were mixed for sequencing. The isolated cpDNA of the studied species was sequenced using Illumina sequencing platform. We used 40 µg of cpDNA for the fragmentation by nebulization with compressed nitrogen gas, and constructed short-insert (500 bp) libraries following the manufacturer's protocol. cpDNA from the different species was indexed by tags and pooled together in one lane of Illumina's Genome Analyzer, and sequence reads of 2 × 100 bp were obtained at Germplasm Bank of Wild Species in Southwest China, Kunming Institute of Botany, Chinese Academy of Sciences, China.

Materials and methods
Genome assembly and annotation. To determine the proportion of cpDNA, we mapped sequence reads to the cp genome of O. sativa ssp. japonica (NC_001320) using Bowtie 64 with paired-end alignment and a maximum of three mismatches (-v = 3) as raw sequence reads may include non-cp DNA. Subsequently, we assembled raw sequence reads into contigs with a minimum length of 100 bp using SOAPdenovo 66 with an overlap length of 27 bp. The contigs were aligned to the reference genome using BLAST (http://blast.ncbi.nlm.nih.gov/), and aligned contigs (≥90% similarity and query coverage) were ordered according to the reference cp genome, gaps between the de novo contigs were replaced with consensus sequences of raw reads mapped to the reference genome. Also based on the reference genome, we designed primer pairs for closing gaps and low-coverage regions, respectively. PCR products were sequenced using standard Sanger protocols on ABI 3730 xl instruments. Sanger sequences and assembled genomes were aligned with ClustalW 1.83 program 67 to determine if there were any differences.
To check the accuracy of sequence assembly, we verified regions with ambiguous read mapping (conflicted reads mapped to the same genomic region) and low coverage (≤2 reads) by PCR amplifications and Sanger sequencing. Here, 1-20 primer pairs (Supplementary Table 23) were used to close the gaps and verify the assembly accuracy.
To detect and characterize genomic variation (SNVs and InDels) the Oryza chloroplast genomes were compared with L. japonica using the program MAFFT v5 70 , following by manual examination. The distribution of genomic variation among the chloroplast genomes was calculated and visualized using the Circos-0.64 71 .
Phylogenomic analyses. Phylogenomic analyses were performed based on the complete cp genome sequences, a set of 77 common protein-coding genes and noncoding DNA sequences from 22 Oryza species including two rice subspecies (O. sativa L. ssp. indica and O. sativa L. ssp. japonica) using L. japonica as outgroup. All of the cp genome sequences were aligned using the program MAFFT version 5 70 and adjusted manually where necessary. MP analyses were performed with PAUP*4.0b10 72 . Heuristic tree searches were conducted with 1000 random-taxonaddition replicates and tree bisection-reconnection (TBR) branch swapping with multrees option. Non-parametric bootstrap analysis was executed under 1000 replicates with TBR branch swapping. Maximum likelihood (ML) analyses were implemented in RAxML version 7.2.6 73 . RAxML searches relied on the general time reversible (GTR) model of nucleotide substitution with the gamma model of rate heterogeneity. Non-parametric bootstrapping as implemented in the fast bootstrap algorithm of RAxML used 1000 replicates. Bayesian analyses were completed using the program MrBayes version 3.1.2 74 . The best-fitting models were determined using the Akaike Information Criterion as implemented in the program Modeltest 3.7 75 . The Markov Chain Monte Carlo (MCMC) algorithm was run for 200,000 generations with trees sampled every 10 generations for each data partition. The first 25% of trees from all runs were discarded as burn-in, and the remaining trees were used to construct majority-rule consensus tree.  was set as outgroup and all gaps introduced by the alignment were excluded. InDels were mapped onto the phylogenetic trees determined by ML analyses of whole cp genome alignment using the ClustalW1.83 program 67 .
Genome-wide analyses of genomic structural variants. The sequences of 23 rice complete chloroplast genomes allowed a comprehensive assessment of genomewide structural variation. All datasets were aligned by a phylogeny-aware algorithm PRANK +F 58 , which has been shown to outperform traditional methods with regard to the accuracy in indel-rich sequences. In PRANK +F, we invoked the +F option to fix already inferred InDels at their places and avoid another indel to be inferred in an overlapping position during the second recursion of the algorithm. This reflects the notion that InDels introduced along one branch of the phylogeny are from an indel event another branch, even if they overlap, and thus produces more gappy but potentially more accurate alignments.
Repeat sequence analyses. The repeat sequences were divided into three categories: dispersed, tandem, and palindromic repeats. The minimal copy sizes investigated were 30 bp for dispersed, 15 bp for tandem, and 20 bp for palindromic repeats, respectively. We determined these three repeat types by first applying the program DNAMAN Version 6.0.3.99 (Lynnon Biosoft, Vaudreuil, Quebec, Canada) and then manually filtering the redundant output. The chloroplast simple sequence repeats (SSRs) were searched by the SSRHunter software (http://www. biosoft.net) 76 . The number of dinucleotide ≥8 bp, trinucleotide ≥9 bp, tetranucleotide ≥12 bp, and pentanucleotide ≥15 bp loci were counted, respectively. Accelerated evolution of protein-coding genes. We investigated the molecular evolution of all 77 chloroplast genes in the genus Oryza. All the orthologous genes were aligned by ClustalW 67 , and then used to the detection of positive selection and evolutionary rate analyses. We analyzed this set of well-aligned 1:1 orthologous gene to obtain and compare the average evolutionary rates of protein-coding genes along lineages and clades of the 23 species phylogeny. dN, dS, and ω were estimated using the codonML 77 program in the PAML software (version 4.4) 78 . The two models with (1) a single ω estimated for all branches and (2) branch-specific ω for each branch were compared in a likelihood ratio test (LRT). The experiments were replicated 10 times.
Genome-wide scan for positive selection. Positive selection was tested for all 77 chloroplast genes of the genus Oryza using the widely employed codon-based substitution models and LRTs implemented in the program PAML version 4.4 78 . In all measurements, codon frequencies were estimated from nucleotide frequencies at each codon position (model F3 × 4). Briefly speaking, our LRT for selection on any branch of the phylogenetic tree was compared site model 1a (nearly neutral) against 2a (selection) 79 , while the branch/clade-specific LRTs were based on branch-site models 80,81 , which compared the modified model A with the corresponding null model with ω 2 = 1 fixed (fix_omega = 1 and omega = 1). In the M1a-M2a comparison, the degree of freedom df = 2 was used, with the critical values to be 5.99 and 9.21 at 5 and 1% significance levels, respectively. For the lineage-specific and clade-specific LRTs, P-values were computed assuming that the null distribution was a 50:50 mixture of point mass 0 and χ 2 df=1 . Identifying sites under positive selection was achieved by using the Bayes empirical Bayes (BEB) 82 to calculate the posterior probabilities for site classes. Multiple comparisons were performed by following the method 83 of to estimate the appropriate P-value threshold for a false discovery rate (FDR) of <0.05.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request. The obtained 22 chloroplast genomes are publicly available in NCBI GenBank under accession numbers (KF359901-KF359922).