We compared gene orders of representative cpDNAs from land plants, including tobacco (Nicotiana tabacum, [GenBank:NC_001879]) 18 and liverwort (Marchantia polymorpha, [GenBank:NC_001319]) 19, a charophytic green alga (Chaetosphaeridium globosum [GenBank:NC_004115]) 20, chlorophytic green algae (Nephroselmis olivacea [GenBank:NC_000927] 21, C. vulgaris [GenBank:NC_001865] 17, C. reinhardtii [GenBank:BK000554] 16), a green flagellate alga with uncertain affinities (Mesostigma viride [GenBank:NC_002186]) 22, and the plastid of Cyanophora paradoxa [GenBank:NC_001675] 23 (Figure 1) (Additional file 1, Additional file 3).

To measure the genome structure in terms of clustering by chromosome locations and gene functions, we defined "sided blocks" as contiguous genes coded on the same strand of the plastid chromosome, and "functional clusters" as blocks of functionally related genes (see Methods). The randomness in the observed distribution of shared genes in chloroplast genomes with respect to gene function was assessed using a Kolmogorov-Smirnov test. The null hypothesis was rejected in all seven cpDNAs investigated for genes in functional categories such as ATP synthases and electron transport (p << 0.05, Table 1). While this test suggests some degree of functional clustering in all chloroplast genomes, it does not take into account the phylogenetic relationship of these organisms, so it is unclear whether functional clustering in chloroplast genomes is a legacy of genome organization in a cyanobacteria-like ancestor, or the product of selection on gene order in the face of genome rearrangements.

In order to investigate evolutionary changes of gene order, we constructed a phylogeny of seven representative cpDNAs and rooted with the sequence of C. paradoxa 23. Maximum parsimony, neighbor joining and maximum likelihood analyses of an alignment of 50 concatenated protein sequences including a total of 19,836 aligned sites (Additional file 2), all yielded identical fully resolved topologies with high bootstrap support (Figure 2A). Mesostigma was placed as a basal charophyte lineage in one previous analysis 24. The unrooted phylogeny of seven cpDNAs (Figure 2B) is congruent with the alternative placement of Mesostigma either as a basal charophyte 24 or basal to both charophyte and chlorophyte lineages 22. This tree was used as the reference phylogeny for gene order inference.

We scored the orders of 85 genes shared in the seven genomes (Gene orders are in additional file 3). Then we used modified versions of GRAPPA 1125 to compute the inversion distance between ancestral nodes and each terminal node (Figure 2B; see Methods). The branches leading to two chlorophytic green algae, C. reinhardtii and C. vulgaris, are much longer than the branches leading to the other taxa, and that many more steps were inferred on the C. reinhardtii lineage relative to the C. vulgaris lineage. Gene duplications or deletions were mapped before scoring the ancestral genomes with inversions, and were not counted as rearrangements. IRs were present in all inferred ancestral nodes, and one copy was lost in C. vulgaris. Ancestral gene orders were scored on all the phylogenies using a two-step approach (see Methods). Due to the computational time limit (the full search for ancestral gene orders may require months), we stopped scoring all possible ancestral gene orders with the data set after 25 days and took the best scored ancestral gene orders at that time (Additional file 4).

The cpDNAs of two land plants, N. tabacum and M. polymorpha, were separated by an estimated 7 inversions based on the data set. One large inversion (~30 kb) in the LSC region has long been recognized to separate the two genomes 26. Additional rearrangements are directly observable through comparison of gene order files for the two species (see additional file 5 for the sequences of gene order rearrangements). Using GRAPPA, all rearrangements were inferred as inversions, but the total number of inversion events estimated by GRAPPA may be greater than the true (but unknown) mixture of inversions and transpositions because one transposition could result in the same change in gene order as two or three inversions.

Two genomic structural characteristics were measured: the propensity of adjacent genes to be clustered on the same strand (using the sidedness index C_s) and the clustering of functionally related genes (using the functional cluster index, C_f) (see Methods). Both indices were calculated for the inferred ancestral gene orders and extant daughter lineages. Among land plants and charophytes, the inferred sidedness among ancestral genomes was similar to extant lineages, however, among the chlorophytes an opposite trend was observed, especially in the C. reinhardtii lineage (Additional file 3). The large number of rearrangements in the C. reinhardtii cpDNA lineage resulted in dramatically increased sidedness relative to the inferred most recent common ancestor of C. reinhardtii and C. vulgaris (C_s ancestor = 0.6966, C_s observed = 0.8710; Figure 3A). A small increase of C_s was found in the N. olivacea lineage and there was almost no change in the lineage leading to C. vulgaris. A large increase was also observed in the functional clustering index, C_f, for C. reinhardti (C_f ancestor = 0.01674, C_f observed = 0.03397; Figure 3B), whereas the trend was less profound in other lineages (Additional file 3). Thus, even if the ancestral genome already had a "sided" structure, sidedness increased with genome rearrangements as C. reinhardtii cpDNA evolved. The inferred increase in sidedness and functional clustering in the face of the large number of rearrangements on the lineage leading to C. reinhardtii might be adaptive, if such increases were not expected under random rearrangements.

To test the null hypothesis that the changes in C_s and C_f were consequence of random genome rearrangements rather than a consequence of directional selection (H₀: random rearrangement; H_A: constraints in rearrangements), we simulated random rearrangements starting with the inferred ancestral genome along the branch leading to C. reinhardtii. Although inversions are the most abundant type of rearrangement in cpDNAs 27, we also considered the contribution of transpositions under three inversion to transposition ratios, while the total number of rearrangements was fixed according to the branch length inferred using GRAPPA (Figure 2B). Three simulations with 10,000 replicates were conducted with inversion to transposition ratios of 1:0, 10:1 and 1:1 under the random breakage model. The mean C_s values for the three sets were 0.5929, 0.6084 and 0.5948, respectively, and the 95% confidence intervals were (0.5056,0.6742), (0.5281, 0.6854) and (0.5169, 0.6742), respectively. All datasets simulated under the random breakage model showed a significant decrease of sidedness from the ancestral level (p < 0.0001). In contrast, the C_s value calculated for C. reinhardtii increased significantly to 0.8710 (Figure 3A), greatly exceeding the sidedness that would be expected in a genome that had undergone this much evolutionary change relative to its ancestor. Simulations using inferred ancestral genomes for land plant lineages (e.g. N. tabacum) also strongly reject the null hypothesis of random rearrangements (results not shown).

Given the large number of rearrangements observed in the C. reinhardtii lineage, C_f was also predicted to decrease significantly under the random breakage model, but C_f did not decrease as observed in C. reinhardtii (Figure 3B). The simulations with three models described above (all inversions, a small fraction of transpositions, and equal inversions and transpositions) all yielded a large decrease in clustering as expected (the observed C_f in C. reinhardtii is 0.03397, and the 95% confidence intervals for C_f in simulated genomes were 0.00744–0.01401, 0.00812–0.014299 and 0.0750-0.01418, respectively). When transposition was included in simulations, decreases of C_f were on a similar scale to the inversion-only simulations. Taken together, these results indicate that the remarkable increase in sidedness and functional clustering observed in C. reinhardtii cpDNA has not been the outcome of solely chance events. Instead, the strong deviation from the range of outcomes expected under various random breakage models implies that the genome structure is the outcome of a directional selective process.

The increased level of organization in C. reinhardtii cpDNA was associated with both maintenance of ancestral clusters and growth of new clusters. There were six conserved blocks containing 19 of the 85 genes shared between the C. reinhardtii and the C. vulgaris cpDNAs. These blocks include concentrations of genes from a single functional category, such as ribosomal proteins (rpl23-rpl12-rps19, rpl16-rpl14-rps8), Photosystem II (psbL-psbF, psbB-psbT-psbN-psbH), translation apparatus (rrn16-trnI-GAU – trnA-UGC – rrn23-rrn5), and ATP synthase subunits (atpF-atpH). Moreover, a number of small clusters of functionally related genes inferred in the ancestral genome were brought together in C. reinhardtii ("rearranged clusters" in Figure 4B). These include transcription/translation genes (trnH-M-F; rpl/rps; rps3-rpoC2), electron transport genes (petA-petD), and photosynthetic genes (psbD-psaA exon 2-psbJ) (Figure 4B). The new clusters contributed to the increase of C_f in the C. reinhardtii chloroplast genome.

Co-transcription of several clusters shown in Figure 4B has been previously documented, including psbD-psaA exon 2-psbJ-atpI 28, psbF-psbL 29, petA-petD 30, and psbM-psbZ 31. Co-transcription of rpl and rps genes has been found in land plant chloroplasts 32. We documented co-transcription for an additional novel functional cluster, shown in Figure 4A. Using RNA gel blots, tricistronic transcripts of rpl36-rpl23-rpl2 and possibly dicistronic rpl2-rps19 species could be detected. Taken together, it appears that the clusters of functionally related genes observed in C. reinhardtii cpDNA may be frequently co-transcribed.

We defined a "functional cluster" as contiguous genes encoded on one strand from one of the following categories: transcription/translation, photosystem I and II, electron transport (cytochrome b6/f complex), and ATP synthase (See additional file 1).

A random cluster consists of genes from any functional category. The n = 85 genes shared in the seven chloroplast genomes shown in Figure 1 were divided into 11 equal sized blocks of r_j = 7 genes and one block of 8 genes so that the block sizes and number of blocks are equal. If m_ij genes were from the functional category i (total T_i genes) in the jth block, the observed cumulative frequency was u_i = ∑_im_ij/r_j. The Kolmogorov-Smirnov test measures the deviation of the observed u_i from the expected from the random breakage model 13. The test statistic D_n was calculated for each functional category separately.

D n = max ⁡ [ max ⁡ ( T i n − u i ) , max ⁡ ( u i − T i − 1 n ) ] MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=vr0dc8meaabaqaciaacaGaaeqabaqabeGadaaakeaacqWGebardaWgaaWcbaGaemOBa4gabeaakiabg2da9iGbc2gaTjabcggaHjabcIha4jabcUfaBjGbc2gaTjabcggaHjabcIha4jabcIcaOmaalaaabaGaemivaq1aaSbaaSqaaiabdMgaPbqabaaakeaacqWGUbGBaaGaeyOeI0IaemyDau3aaSbaaSqaaiabdMgaPbqabaGccqGGPaqkcqGGSaalcyGGTbqBcqGGHbqycqGG4baEcqGGOaakcqWG1bqDdaWgaaWcbaGaemyAaKgabeaakiabgkHiTmaalaaabaGaemivaq1aaSbaaSqaaiabdMgaPbqabaGccqGHsislcqaIXaqmaeaacqWGUbGBaaGaeiykaKIaeiyxa0faaa@55C1@

Alignments of 50 proteins shared in the 8 chloroplast genomes shown in Figure 2A were concatenated into one data matrix (Additional file 2). 1,000 bootstrap replicates were conducted on the data set using PAUP* 4.0b10 with maximum parsimony and using MEGA with neighbor-joining methods and the Poisson-corrected distance. Maximum likelihood analysis with 100 bootstrap replicates was performed using PHYLIP3.6 with JTT distance and gamma = 0.5. GRAPPA was not used to construct the reference phylogeny.

The ancestral gene order was inferred from the gene orders of extant genomes on the best-scored tree following two steps. First, the gene contents for the LSC, SSC and IR regions of ancestral genomes of IR-containing cpDNAs were inferred based on parsimony. Changes in gene copy number due to IR expansion or contraction were considered the last step of gene order changes, and thus the gene contents of ancestral genomes were determined. The ancestral gene orders on the phylogeny for five genomes (excluding C. vulgaris and C. reinhardtii) were computed using GRAPPA-IR 25, which is a modified version of GRAPPA that scores rearrangements independently within LSC, SSC or IR. Second, the chlorophyte algal gene orders (the extant chloroplast gene orders of N. olivacea, C. reinhardtii and C. vulgaris and the inferred ancestral genome of N. olivacea from step one) and the gene order of M. viride were used for the inference of the common ancestral gene order of C. vulgaris and C. reinhardtii. The data set contains duplicated trnV-UAC and trnG-GCC in C. vulgaris, trnE-UUC and psbA in C. reinhardtii and three trans-splicing psaA exons in C. reinhardtii. The IR regions contained rRNA genes in the same order and orientation in each genome except that one copy was lost in the lineage leading to C. vulgaris. To score the genomes with gene duplications and deletions, multiple data sets were created each containing genomes with equalized gene contents by the following assignment rules: one copy of each duplicate genes outside the typical IR was chosen; the IR region lost in C. vulgaris was inserted to all possible locations in that genome. Preferably, we should test all these datasets (3,936 total) with inversion medians; however, such computation on one dataset alone will take more than a month. To overcome this limitation, these datasets were computed using breakpoint medians, and the assignment yielded the shortest tree was chosen for a full evaluation by GRAPPA. Because the gene contents of LSC and SSC in C. reinhardtii were different from other chloroplast genomes in the study, we allows free rearrangements such that genes in LSC or SSC could commute across the IR.

A set of simulation experiments were conducted to evaluate the accuracy of ancestral genome reconstruction with long branches. Three genomes with 85 genes each were generated from a defined ancestral gene order, and the branch lengths (inversion distances) were 50, 20 and 20, respectively. The true gene order score was 90 (equals the tree length). The scores were computed for inferred ancestral gene orders by GRAPPA using inversion medians and the random breakage model and then compared to the true score. The experiment was repeated on 30 data sets.

Gene orders were simulated under the assumption that the rearrangements involve random breakpoints placed between genes. Initial gene orders were set based on the inferred ancestral gene orders estimated. Random rearrangement operations on the initial genomes were performed for the number of replicates according to the number of rearrangements inferred by GRAPPA. The parameters input to the model were the ratios of inversion and transposition (1:0, 10:1, 1:1) to test the sensitivity of the findings to the specific rearrangement model. The simulated genomes had identical gene content but scrambled gene orders relative to those observed in extant genomes, with the exception that inverted repeats were maintained. Test statistics (below) were calculated for each simulated replicate of 10,000 total and the frequency distributions were used to test the null hypothesis of random rearrangement.

We designed the sidedness index (C_s) to measure the degree to which neighboring genes are clustered on the same strand (side) of the chromosome. A "sided block" includes only adjacent genes on one strand, and the number of sided blocks in a genome is designated as n_SB, while the total number of genes is n. C_s is defined as

C_s = (n-n_SB)/(n-1).

When C_s reaches the maximum of 1, all genes are located on one side. If every gene resides on the strand opposite its neighbors, C_s approaches a minimum of zero.

We divided a genome of total n genes to J sided blocks (r₁, r₂,...r_J). In a block, we assigned genes to functional categories. Let the numbers of genes in the ith functional category and the jth block be m_ij, the functional cluster index C_f is

C f = 1 J ∑ j = 1 J r j n ∑ i = 1 4 ( m i j 2 ) / ( r j 2 ) . MathType@MTEF@5@5@+=feaafiart1ev1aaatCvAUfKttLearuWrP9MDH5MBPbIqV92AaeXatLxBI9gBaebbnrfifHhDYfgasaacH8akY=wiFfYdH8Gipec8Eeeu0xXdbba9frFj0=OqFfea0dXdd9vqai=hGuQ8kuc9pgc9s8qqaq=dirpe0xb9q8qiLsFr0=vr0=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@50F0@

A larger value of C_f indicates that functionally related genes are more clustered into blocks.

Wild-type CC-124 cells were grown in Tris-Acetate-Phosphate medium 60 under continuous light to mid-log phase. RNA was isolated from 10 mL of cells as previously described 61. For filter hybridization, 5 μg of total RNA was fractionated in 1.2% agarose and 6% formaldehyde gels, transferred to nylon membranes, and probed with gene-specific PCR products labeled by random priming according to Church and Gilbert 62.