Figure 1 shows that the IR-LSC junctions in 90 non-monocot angiosperms usually drift around position b (data shown in the additional file 1). In these cases, designated as type I, an intact trnH gene is always present near the J_LA but absent from the J_LB. In Chloranthus oldhami, C. spicatus, Sarcandra glabra (Chloranthales), Canella winterana (Canellales), Ranunculus japonica and R. macranthus (eudicot), a partial trnH sequence is found extending to position c in IR_B (Fig. 1A, additional file 1). The IR-LSC junctions were located upstream of position c' (i.e. upstream of trnH) in Nuphar advena (Nymphaeaceae) and Elaeagnus formosana (Elaeagnaceae, eudicot), at position a in Kadsura japonica (Schisandraceae, Austrobaileyales), and at position a' in Calycanthus fertilis and C. floridus (Calycanthaceae, Laurales, 2930) (Fig. 1A). However, Vitis vinifera (Vitaceae, eudicot) showed a complete loss of rpl2 near J_LA 31.

The Winteraceae (Canellales), exemplified by Zygogynum pauciflorum and Drimys granadensis 29, were exceptional in that the organization of the genes flanking the IR-LSC junctions resembled the one found in most monocots, rather than the organization seen in other non-monocot angiosperms. Notably, each of their IRs contained a trnH-rps19 cluster and their IR-LSC junctions were located within the 5' portion of rps19 (position d, Fig. 1).

Type II IR-LSC junctions were found in Schisandra arisanensis (Schisandraceae; Austrobaileyales) and some 41 representative eudicots (Fig. 1A; additional file 1). Unlike type I, the J_LA of type II shifted to the 5' end of the truncated rps19 in IR_A (position e and e', Fig. 1A, additional file 1).

In contrast to basal angiosperms and eudicots, most monocots (Fig. 1B) had trnH-rps19 clusters present in each of the two IRs, and the IR-LSC junctions were generally at position f (Arecales, Dasypogonaceae, Asparagus densiflorus [Liliales], Poales and Zingiberales) or g (in Asparagales and Commelinales) (Fig. 1B). This type of gene organization was classified as type III. In addition, IR-LSC junctions of some monocots were located downstream of rpl2 (position b; in Araceae, most Alismataceae, and Hydrocharitaceae), of trnH (position c' in Potamogetonaceae and Dioscoreaceae), or within rps19 (position d, Fig. 1; in Acorales, Lilium formosamum [Liliales] and Panadanales). When the IR-LSC junction was at position d, the rps19 sequence in IR_A was found to be partially truncated most of the times.

Figure 2 illustrates alignment of the sequences flanking the J_LA regions in some representatives of basal angiosperms and eudicots (A) and monocots (B). Of particular interest is the observation that the IR-LSC junctions of basal angiosperms, eudicots and monocots are commonly found at either polyA tract or A-rich regions (Fig. 2). We also found that the dicot IR sequences near the IR-LSC junctions varied little and could be aligned among orders having the same or different IR-LSC junction types, while in monocots the corresponding regions were very different and difficult to align across different orders (Fig. 2B). Moreover, within the sampled angiosperm families the sequences flanking the J_{LAregions were very similar.}

Among the chloroplast operons, the S10 ribosomal protein operon is the largest. It contains genes encoding both small (rps) and large (rpl) ribosomal protein subunits that are organized into a polycistronic transcription unit conserved in known cpDNAs 32. In angiosperms, the 5' end of the S10 operon is initiated within the IR, but only in IR_B does the operon extend into the LSC region, and the S10 operon is only partially in IR_A (viz. the S10_A operon). However, a second operon in IR_A, the psbA operon, is transcribed from LSC towards IR_A 32 and opposite to the S10_A operon.

In the Winteraceae and a majority of monocots, the trnH-rps19 cluster of IR_A is included in both the S10 and psbA operons. Therefore, this gene cluster may be regulated by two opposing promoters, the S10_A and the psbA (Fig. 3A). In monocots, if the trnH in IR_A is indeed regulated by the above-mentioned two opposing promoters, the function of the trnH gene may be repressed because antisense-trnH RNAs would be generated by both the S10_A and S10_B promoters. To verify this possibility, we conducted RT-PCR assays using specific primers for a type III representative, Asparagus densiflorus, with the IR-LSC junction located at position f (Fig. 1B).

Our results indicate that expression of the trnH gene in IR_A is regulated by both the S10_A and psbA promoters. This suggests that the duplicated trnH gene located in the IR_B region of most monocots and in some non-monocots has different fates (i.e. functional or degenerate in different lineages; see Fig. 1). Figure 3B shows that two RT-PCR products, a 250 bp and a 700 bp fragment, respectively, were generated when specific primer pairs for each were used (Fig. 3A). The former fragment was amplified from the transcripts made by the psbA promoter, and the latter by the S10 promoter. This result confirms that the trnH-rps19 cluster of IR_A is regulated by two opposing promoters (Fig. 3B), indicating that the transcription machinery in IRs of monocots may differ from that of basal angiosperms and eudicots.

As shown in Figure 1A, IR-LSC junctions of the Amborella + Nymphaeales are mainly located at position b, but junctions of monocots are further expanded to encompass LSC genes and are located at positions f or g. Since the two IRs of monocots usually include the trnH-rps19 cluster (position f or g, further downstream of rpl2; Fig. 1B), we hypothesize that at least two duplication events are required to explain the expansion of IRs in monocots during the course of IR evolution from an Amborella-like ancestor to present-day monocots. If this hypothesis is correct, it is expected that an intermediate junction type could be traceable in the cpDNAs of some early divergent monocot lineages between the two duplication events.

Narayanan et al. 33 have recently presented a model of gene amplification in eukaryotes that argues strongly for the involvement of hairpin-capped DSBs in the initiation. Based on this model and our observations, we propose two hypotheses to account for the evolution of IR expansions in monocots (Fig. 4). In hypothesis A, a DSB event (Fig. 4, red arrowhead in step 1) occurs first within the IR_B of an Amborella-like ancestor, and then the free 3' end of the broken strand is repaired against the homologous sequence in IR_A. The repaired sequence extends over the original IR-LSC junction and reaches the area downstream of trnH (Fig. 4, step 1), so that duplication of a trnH gene in the newly repaired IR_B is achieved. Similarly, a second DSB event occurs in IR_A adjacent to the IR_A-LSC junction (Fig. 4, red arrowhead at step 2) so that duplication of rps19 at IR_A can be initiated, and a trnH-rps19 cluster nearby J_LB (Fig. 4, step 2) is created. The newly formed IRs might cover the trnH-rps19 cluster and extend further into the intergenic spacer between rps19 and rpl22 (Fig. 4, step 1 to step 2). Furthermore, if one additional DSB event took place within the intergenic spacer located between rps19 and rpl22 in the LSC region, a partial rpl22 gene would be duplicated at IR_A using the rpl22 sequence of LSC as a template, and from then on the repaired IRs might have expanded towards the 5' region of the rpl22 (Fig. 4, step 2 to step 3). The exceptionally long IRs observed in the Orchidaceae and Commelinales are likely to have been generated by this process. The same outcomes could also result if the process proceeded directly from step 1 to step 3 without step 2 (Fig. 4, path indicated by green dashed arrow).

Hypothesis B, on the other hand, assumes that rps19 would be duplicated or converted prior to the duplication of trnH through a DSB event that takes place at IR_A first (Fig. 4; blue arrowhead of step 1). A second DSB event (Fig. 4; blue arrowhead of step 2) then would take place within the IR_B region through a similar repair process to the one mentioned before, so that a duplicated trnH is generated at IR_B. Finally, the IRs expand downstream of rps19. In hypothesis B subsequent extension of IRs is assumed to resemble step 3 of hypothesis A.

Duplication of a partial or complete rps19 gene was also observed in some eudicots and Schisandraceae (type II) with their respective IR-LSC junctions located at position e or e' (additional file 1; Fig. 1). However, these duplicated rps19 genes (both partial and complete) are situated between the rpl2 and trnH genes of the IR_A (refer to type II in Fig. 1A and Fig. 4 [see the light blue line at the right side leading to eudicots]) rather than downstream of trnH or upstream of psbA (refer to step (2) and (3) of hypothesis A in Figure 4). Therefore, the gene arrangement flanking the IR_A-LSC of type II deviates from that of type I, suggesting that duplication of rps19 genes in type II must have a distinct evolutionary history.

Based on comparisons of aligned rpl2-trnH and trnH-rps19 intergenic spacer sequences from representatives of major monocot orders (Figure 5A, B), it is apparent that these two spacer sequences are separately highly similar across the sampled monocot orders. These data give strong support to hypothesis A that in monocots expansion and inclusion of trnH-rps19 gene cluster in IRs might require at least two common DSBs (please refer to steps 1 to 3 of hypothesis A in Figure 4): one occurring within IR_B (refer to Fig. 4, step 1), and the within IR_A (refer to Fig, 4 step 2 or 3).

However, we did not discover any inverted repeats that might have led to the formation of hairpins in the monocot intergenic spacers of trnH and rps19. Therefore, we are inclined to conclude that the expansions of monocot IRs took the path depicted in hypothesis A.

Goulding et al. 15 proposed two models to account for two kinds of IR expansion: (1) small and random IR expansions, caused by gene conversion (viz. single strand break); and (2) large IR expansions, like those found in the Nicotiana species, rice and maize, generated via DSB events. Narayanan et al. 33 further demonstrated that DSBs can trigger gene amplification through a variety of mechanisms, and that breakage at the inverted repeats of chromosomes can cause gene amplification.

After a critical comparison of genes or sequences adjacent to the IR-LSC junctions in 33 major orders and 8 families of angiosperms (following the classification system proposed by Soltis et al. 2005 27), we hypothesize that IR expansions resulted principally from the DSB events that occurred during IR evolution from the Amborella-like ancestor to monocots. This hypothesis is founded on the following 5 observations: (1) the length of IR expansion from basal angiosperms to monocots is large (more than 100 bp); (2) trnH and rps19 are situated downstream of IR_A and IR_B, respectively, in all sampled basal angiosperms (Fig. 1A). This type of gene arrangement might represent the ancestral gene pattern in basal angiosperms; (3) IRs of several basal angiosperms (e.g. Schisandraceae, Chloranthales and Magnoliales, Winteraceae) and eudicots (Fig. 1A) have partially or completely duplicated trnH genes located at IR_B; (4) in comparison with other angiosperms, monocot IRs have expanded further to include a duplicated rps19 in IR_A, and this expansion should have occurred before the diversification of major monocot orders; and (5) the IRs of advanced monocots (from Asparagales to Poales) have expanded to encompass more LSC sequences or genes (Fig. 1B). Nevertheless, the latter expansions did not apparently result from another common DSB event but from independent ones, because among sampled monocot orders the downstream regions of rps19 genes have low sequence similarity (Fig. 2). At the infra-order level of angiosperms, gene conversion might occur frequently at meiosis and cause small IR expansion or contraction during evolution, as found in Apiaceae 14 and Nicotiana 15.

Studies on the IR-LSC junctions of Nicotiana species 15 and Apiaceous plants 14 have indicated that short repeats or "polyA tract" sequences associated with tRNAs at the IR-LSC boundaries might be likely hotspots for recombination. We also observed that polyA tract sequences are commonly present near the IR-LSC junctions in all the basal angiosperms, eudicots and monocots examined (Fig. 2), indicating that such sequences are closely linked with the dynamics of IR-LSC junctions and expansion of IRs. In this regard, we further propose that IR expansion may initiate at the DSBs and finish at the polyA tract regions, where recombination may actively occur, and that the recombination mechanism in cpDNA may resemble that reported for nuclear genomes by Narayanan et al. 33.

According to our hypothesis, DSBs within IRs must have been frequent during angiosperm evolution. However, only those which led to successful IR expansions, and have subsequently been retained in the extant taxa, are detectable. Based on our observations, it is evident that the type of IR-LSC junction appears to be informative, at least at the level of order, and is therefore useful for inferring phylogenetic relationships at this rank and above.

As shown in Figure 1B, IR-LSC junctions of basal monocots including Acorales, Pandanales and Liliales are usually located at position d. This type might represent a primitive state. In contrast, IR-LSC junctions of the derived monocots, such as Asparagales and Poales, have generally expanded to position f or g. This trend in IR expansion seems to correlate well with the divergence pattern of monocot lineages in the multigene tree of Soltis et al. 2734, which shows Acorales to be a sister group to other monocots. This correlation connotes the ancient status of the order and the continuous IR expansion experienced by the more terminal and derived lineages, viz. Asparagales, Commelinales, Zingiberales, Arecales, Dasypogonaceae and Poales.

It is worth mentioning that in some monocots (e.g. Pandanales and Liliales) the IR-LSC junctions are located at position d, with a truncated rps19 gene at IR_A. According to hypothesis A (Fig. 4), duplication of rps19 at IR_A was due to a second DSB event in IR_A (Fig. 4, red arrowhead at step 2), followed by a sequence repair supposed to have been terminated within or downstream of the rps19 gene. Duplication of the rps19 gene will lead to a shift of the IR-LSC junction to position d or f (Fig. 1B). However, in Pandanales and Liliales, the rps19 sequences of IR_A are incomplete or degraded. We considered these common degradations likely to be secondary rather than primary, since the majority of monocot orders have the trnH-rps19 clusters (Fig. 1B). Moreover, among the major monocot orders (except Alismatales) the intergenic spacer sequences within the trnH-rps19 cluster (Fig. 5B) have a high degree of similarity, suggesting that among the sampled monocots a common DSB event might have taken place adjacent to the trnH gene. Therefore, the IRs in Acorales, Pandanales and Liliales are likely to have contracted, causing a shift of the IR-LSC junctions from around position f to position d.

A comparison of the downstream non-coding or spacer sequences of the rps19 genes in monocots reveals that the sequences do not have a common origin (Fig. 2B), as they are highly variable and a reliable sequence alignment is impossible except between closely related con-ordinal taxa (e.g. Zingiberales and Asparagales). This indicates that these spacer sequences had diverse origins and are likely to have resulted from independent DSB events occurring at different points within the IRs.

In contrast, it appears that expansion of IR-LSC junctions did not parallel the evolutionary diversification of basal angiosperms and eudicot lineages (Fig. 1A). In type I (Fig. 1), IR expansion downstream of rps19 is extremely rare in eudicots, with the exception of Adzuki bean (Perry et al. 18) and a Pelargonium species (Palmer et al. 16, Chumley et al. 11). According to our hypothesis A (Fig. 4), the scenario of IR expansion in these two eudicots may have different origins from those of monocots and other eudicots (i.e. type II, Fig. 1), with IRs that have expanded downstream of rps19 genes. Similarly, significant IR contractions in the basal angiosperm Illicium oligandrum (about 1 kb), coriander (4 kb) 1314, and Cuscuta reflexa (about 700 bp to 8 kb) 35 seem to be separate events in their respective lineages.

In extant angiosperms, the relationships among the remaining 5 lineages (magnoliids, monocots, eudicots, Chloranthaceae and Ceratophyllum) are unresolved 192627. To what extent the dicot lineage is a sister group of monocots remains uncertain, probably a reflection of the rapid radiation and extinction of early angiosperms soon after they originated 3637.

Recent phylogenetic analyses based on plastid sequence data have suggested that monocots and eudicots are sister taxa (Graham et al. 38 and Cai et al. 39), but with low bootstrap support (67% and 72%, respectively). In addition, several lines of evidence have indicated that Ceratophyllaceae could be the sister group of monocots 4041424344.

Here we present an alternative view on this issue. As illustrated in Figure 1, an intact trnH is duplicated in IR_B of all monocots, one basal angiosperm (Nuphar advena, position c'), and two winteraceous magnoliid species (Zygogynum paucifolum and Drimys granadensis, position d) 29. Sequence comparison revealed that only Winteraceae and monocots have highly similar spacer sequences between the rpl2 and trnH genes (Fig. 5B), suggesting that duplication of trnH gene in IR_B of the two taxa might be common or similar (viz. convergent). On the other hand, Acorales (the most basal lineage in monocots, 27) has its IR endpoint at position d, suggesting that those lineages with IR-LSC junctions at position b and c' (most Alismatales and Dioscoreales) might have resulted from separate, independent contractions. Our alternative view on the relationships among monocots and their relatives is preliminary, as it is only based on comparison of genic organizations at IR-LSC junctions. Additional molecular and morphological data are required to improve our understanding of monocot phylogeny.

The presence of a trnH-rps19 cluster in the IRs appears to be a common feature in monocots other than some Alismatids (additional file 1, Fig. 1), in which IR-LSC junctions are located at position b and strongly resemble those of most non-monocot angiosperms. However, alignment of the intergenic spacers between rpl2 and trnH in some Alismatales (e.g. Alocasia odora) and other monocots, basal angiosperms and eudicots (Fig. 5) reveals that sequences of the Alismatids are more similar to other monocots than to non-monocot angiosperms. This implies that IR expansions in some Alismatids might share evolutionary scenarios similar to those proposed for other monocots, and that the short IRs (or IR contraction) in some other Alismatids are likely due to either an early termination of the repair-extension reaction after the first DSB in step 1 of hypothesis A (Fig. 4), or to a contraction after this step.

In monocots, each IR usually contains a trnH gene, while in most basal angiosperms and eudicots the gene is rarely present in IR_B (see Fig. 1A: type I and type II). Why is the duplicated trnH gene able to survive in IR_B of most monocots but is absent, degraded or truncated in most non-monocot angiosperms? In two studied eudicots, Lotus japonicus 18 and Spinacea oleracea 45, the transcriptional activity of S10_A dropped significantly because of either the high transcription levels of the psbA and trnH genes or the termination of S10_A proximal to J_LA 32. Therefore, in non-monocot angiosperms, trnH-encoded mRNA molecules constitute only one sense strand, transcribed solely by the psbA operon rather than by the S10_A operon. Because anti-sense RNA molecules may interfere with the normal function of the sense RNA molecules 32, in monocots the mechanism by which anti-sense trnH is regulated by two S10_A promoters is mysterious. Further study on the evolution and survival of the duplicated trnH gene in IR_B of monocots is desirable.

Species sampled in this study were listed in the additional file 1. Total cellular DNA was extracted using the method of Saghai-Maroof et al. 46. The extracted DNAs were used directly for PCR amplification.

Primer design was based on published sequence data for conserved regions flanking the IR-LSC junctions. The J_LA regions were amplified with the primer pair rpl2-psbA-F3 and rpl2-psbA-R2, which correspond to the 3' end of rpl2 and the 5' end of psbA respectively (Fig. 1). The J_LB region was amplified using two forward primers, rps3-F1 and rps3-F2, that respectively pair with a reverse primer rps3-rpl2-R2. The sequences of these primers are listed in Table 1. Amplicons were cleaned using the Gel Extraction System (Viogene, Taipei) and cloned into a pGEM T-Easy vector (Promega, Fitchsburg). Plasmid DNAs were purified using the Plasmid DNA Miniprep System (Viogene) and sequenced on an ABI 3730 automated sequencer (Applied Biosystems, Foster City). For each species two independent PCR clones were sequenced. Sequence alignments were made using GeneDoc (Ver. 2.6.02.)

To verify the transcription of trnH-rps19 that flanks the IR_A region, total RNAs were extracted and purified by RNeasy^® Plant Mini Kit (Qiagen, Hilden). The resulting RNAs were reversely transcribed to synthesize cDNA with Superscript II reverse transcriptase (Invitrogen, Indianapolis) and a specific primer (either trnH-psbA-F1 or trnH-rev), according to the manufacturer's protocol. The two synthesized cDNAs were then used with the primer pair trnH-psbA-F1 and rpl2-psbA-R2 to amplify a 674 bp fragment, and the primer pair trnH-rev and rpl2-psbA-F3 to amplify a 298 bp fragment. Each of the two reactions was conducted under the following conditions: 94°C for 5 min, followed by 30 cycles of 94°C for 30s, 55°C for 30s, and 72°C for 30s, and ending with an extension of 72°C for 10 min.