Complex Processing Patterns of mRNAs of the Large ATP Synthase Operon in Arabidopsis Chloroplasts

Chloroplasts are photosynthetic cell organelles which have evolved from endosymbiosis of the cyanobacterial ancestor. In chloroplasts, genes are still organized into transcriptional units as in bacteria but the corresponding poly-cistronic mRNAs undergo complex processing events, including inter-genic cleavage and 5′ and 3′ end-definition. The current model for processing proposes that the 3′ end of the upstream cistron transcripts and the 5′ end of the downstream cistron transcripts are defined by the same RNA-binding protein and overlap at the level of the protein-binding site. We have investigated the processing mechanisms that operate within the large ATP synthase (atp) operon, in Arabidopsis thaliana chloroplasts. This operon is transcribed by the plastid-encoded RNA polymerase starting from two promoters, which are upstream and within the operon, respectively, and harbors four potential sites for RNA-binding proteins. In order to study the functional significance of the promoters and the protein-binding sites for the maturation processes, we have performed a detailed mapping of the atp transcript ends. Our data indicate that in contrast to maize, atpI and atpH transcripts with overlapping ends are very rare in Arabidopsis. In addition, atpA mRNAs, which overlap with atpF mRNAs, are even truncated at the 3′ end, thus representing degradation products. We observe, instead, that the 5′ ends of nascent poly-cistronic atp transcripts are defined at the first protein-binding site which follows either one of the two transcription initiation sites, while the 3′ ends are defined at the subsequent protein-binding sites or at hairpin structures that are encountered by the progressing RNA polymerase. We conclude that the overlapping mechanisms of mRNA protection have only a limited role in obtaining stable processed atp mRNAs in Arabidopsis. Our findings suggest that during evolution of different plant species as maize and Arabidopsis, chloroplasts have evolved multiple strategies to produce mature transcripts suitable for translation.


Introduction
Chloroplasts are cell organelles in photosynthetic organisms where essential functions as photosynthesis, synthesis of lipids, pigments, vitamins and amino acids occur. They have evolved from endosymbiosis of photosynthetic cyanobacteria and from those they have inherited many prokaryotic-like elements. However, during the evolution process from bacteria, chloroplasts have acquired also eukaryotic features and plastid-specific characteristics [1]. The gene-expression system represents a good example for the acquirement of complexity in plastids. In chloroplasts, genes are still organized into transcriptional units or operons as in bacteria. However, differently from the bacterial counterparts, the corresponding poly-cistronic mRNAs undergo complex processing events and only upon intron splicing, editing, inter-cistronic cleavage and termini definition, the mature mRNAs are ready for translation.
An early model for inter-cistronic processing of chloroplast mRNAs predicts that endonucleases cut at specific sites and exonucleases then trim the RNA until an hairpin is encountered [2]. More recently, pentatricopeptide repeat (PPR) proteins together with other classes of RNA-binding proteins have been involved in the definition of transcript termini. Small RNAs (sRNAs) persisting in the RNA pool even upon mRNA degradation have been identified as footprints of RNA-binding proteins [3,4]. The finding of transcripts of adjacent cistrons having overlapping ends was explained by a novel mechanism in which the 39 end of the upstream cistron transcripts and the 59 end of the downstream cistron transcripts are protected from exonucleolytic degradation by the same RNA-binding protein [5]. PPR10 was shown to protect the overlapping ends of atpI and atpH mRNAs in maize chloroplasts [6].
Recently, we have analyzed the transcriptional organization of the plastid large ATP synthase (atp) operon of Arabidopsis thaliana that consists of the genes atpI/H/F/A [7]. This operon is transcribed by the plastid-encoded RNA polymerase (PEP) starting from a promoter at 2225/229 in front to the atpI coding sequence, which is dependent on the Sigma factor 2 (SIG2). In addition, we identified an operon-internal promoter just upstream of the atpH ORF, which is dependent on the Sigma factor 3 (SIG3) [8]. This atpH specific promoter has not been described in maize thus raising the question of whether the processing mechanisms of the atp operon mRNAs are comparable in maize and Arabidopsis. Furthermore, in addition to the supposed PPR10-binding site in the atpI/H inter-genic region, three other sites for RNA-binding proteins were predicted in the large atp operon, upstream of the atpI ORF and within the atpH/F and the atpF/A inter-genic regions [3,4]. In order to elucidate the mechanisms controlling the processing of the atpI/H/F/A transcripts in Arabidopsis chloroplasts as well as the functional significance of the operon-internal promoter and the four protein-binding sites for these maturation processes, we have performed a detailed mapping of the 59 and 39 ends of the complete set of mRNAs encoded by the large atp operon. Our data show that 59 ends of the nascent poly-cistronic atp transcripts are defined at the first protein-binding site following the transcription initiation sites, while the 39 ends are defined at the subsequent protein-binding sites or hairpin structures that are encountered by the progressing RNA polymerase. We observe that atp transcripts with overlapping ends are rare and in some cases they are even truncated at the 39 end, thus representing degradation products.

Mapping strategy
For mapping of transcript ends, we have chosen the circular RT-PCR (cRT-PCR) technique because this method allows determination of the 59 and 39 ends delimiting the same RNA molecule [9]. Conversely, techniques like 59 and 39 RACE, primer extension or nuclease S1 mapping allow definition of only one of the two mRNA ends and cannot distinguish among different combinations of 59 and 39 ends. We isolated total RNA from 7 day old Arabidopsis seedlings. Transcripts were self-ligated and retrotranscribed with gene-specific primers for the single cistrons in the atp operon ( Figure 1A). cDNA fragments containing the fused 59 and 39 ends were then amplified for cloning (when necessary) and sequencing ( Figure 1B). In order to characterize also primary transcripts, RNA self-ligation was preceded by treatment with Tobacco acid pyro-phosphatase (TAP) ( Figure 1C). Data obtained by cRT-PCR were then complemented by Northern blot analyses, which demonstrate the relative ratio among the different RNA populations ( Figure 2).

Mapping of the mono-cistronic atpH transcript ends
We started our analysis by mapping the ends of the atpH transcripts, which are the most abundant mRNAs encoded by the large atp operon [10]. When primers were designed inside the coding region of atpH mRNAs (h2 and h3 primers, Figure 1A), a sharp PCR product was obtained in both TAP-untreated and treated retro-transcribed RNAs ( Figure 1B, lanes 3 and 4), indicating that the transcripts were generated by a processing event at the 59 end. In parallel, amplification reactions without previous reverse transcription (lanes 1-2) or DNA template (lane 5) were carried out as controls for the specificity of the cRT-PCR reaction. The PCR product was then cloned and 8 clones were sequenced. All sequences indicated 59 ends at position 245 ( Figure 3A, bottom panel) and most of the clones (6/8) 39 ended in the inter-genic region between atpH and atpF, at position +297 from atpH stop codon (e in Figure 3A). A short RNA (sRNA) with a 59 end at position 245 was also found in Arabidopsis [3] and in maize [4] chloroplasts. This sRNA corresponds to the PPR10binding site in maize. Another sRNA with a 39 end around position +297 was also discovered in Arabidopsis chloroplasts [3]. Taken together, these data suggest that mono-cistronic atpH mRNAs starting at position 245 and ending at position +297 are generated through protection by RNA-binding proteins. However, we detected also another mono-cistronic atpH isoform with a 39 end at position +151 (d in Figure 3A), a site that was already found using the 39 RACE technique. No sRNAs were found to map in close vicinity to this site [3]. We performed a secondary structure prediction using the MFOLD 3.2 program (http://mobyle. pasteur.fr/cgi-bin/portal.py?#forms), which indicated that the 50 nucleotides upstream to the +151 site can fold into a stable stem-loop structure with a predicted DG of 214.36 kcal/mol ( Figure 3B, left panel). Similar structures were proposed to protect the ends of plastid mRNAs as an alternative to the proteinmediated mechanism [2,5]. Interestingly, we found that the atpH transcript 39 end at the +151 site included an extra sequence (AA), suggesting that this mRNA represents an intermediate product of a processing/degradation event [11]. These data indicate that the atpH transcripts are mostly present as two distinct mono-cistronic isoforms which share the same 59 processed end nearby the PPR10-binding site. However, the longer isoform (e in Figure 3A) contains a well-defined 39 end which maps close to a protein footprint, while the shorter one (d in Figure 3A) presents a 39 end which maps close to a stem loop structure. The existence of these two types of transcripts was confirmed with Northern blot analyses (d and e in Figure 2B) that were performed with single strand DNA probes hybridizing to the atpH coding region (data not shown) or to the 39UTR ( Figure 2A, probe 2-4). When probe 2 was used, which hybridizes to the atpH 39 UTR just upstream to the +151 site, three very abundant bands (d, e and f) were revealed ( Figure 2B). The bands e and f, but not the band d, were still detected when we used probe 3 which was designed to hybridize in between the hairpin structure (+151) and the protein-binding site (+297). Finally, only the band f was detected when we used probe 4 which was hybridizing directly downstream to the protein-binding site (+297). Thus, in addition to the confirmation of two different mono-cistronic atpH transcripts, the Northern experiment revealed the existence of a longer atpH transcript (band f), that corresponds to an atpH/F co-transcript (see below).

Mapping of the atpI transcript ends
Genetic and biochemical data have shown that the same barrier protein, PPR10, is involved in the stabilization and definition of the overlapping ends of the atpH and atpI mRNAs in maize [6]. However, the transcriptional organization of the large atp operon differs in Arabidopsis and in maize with respect to the internal atpH promoter [8]. Therefore, processing mechanisms might be also different in these two plant species. For that, we have verified the validity of the overlapping model at this site in Arabidopsis by mapping the ends of the atpI transcripts. Amplification of circularized atpI mRNAs using the i2 and i3 primers ( Figure 1A) revealed a diffuse band pattern with both TAP-untreated and treated RNA samples ( Figure 1B, compare lanes 6 and 7), where all atpI mRNAs result from a 59 processing event. We have sequenced 11 clones and found that 9 sequences revealed mRNA 59 ends at position 248 from atpI AUG whereas 39 ends were scattered all over the atpI/H inter-genic region (a9 in Figure 3A). One of the clones (1/11) presented an extra sequence at the 39 end, indicating it as an intermediate of a RNA processing/ degradation event. Only 2 clones (2/11) displayed 39 ends just downstream to the PPR10 sRNA mapping site (this mRNA is expected to be 1292 nucleotides long) ( Figure 3A, bottom panel). We have repeated this experiment twice, using two independent preparations of RNAs that were obtained from two independent sets of plants. All the clones analyzed (32 clones) revealed mRNA 59 ends at position 248, indicating a very stable 59 end. Only a limited proportion of the clones (9/32) revealed 39 ends inside or just downstream to the PPR10 footprint ( Figure 3A, bottom panel). In addition, 12/32 and 7/32 clones displayed 39 ends at positions +5 and +390 downstream to the atpI stop codon, respectively, indicating two additional 39 ends. These mRNAs are expected to be 803 and 1188 nucleotides long, respectively. As in the precedent experiment, few (3/32) clones presented extra sequences. Our data indicate that even though in Arabidopsis the atpH 59 end is well protected against nuclease digestion (likely by the PPR10 orthologue) as in maize, the 39 end of the atpI monocistronic transcripts is not defined with the same efficiency. This suggests that the mechanism for the atpI/H transcript processing in Arabidopsis chloroplasts is different from that in maize. On the other hand, when the reverse (i2) and forward (h3) primers were designed inside the atpI and the atpH coding sequences, respectively, in order to characterize the longer atpI/H cotranscripts (see Figure 1A), two sharp PCR products were amplified after reverse transcription of circularized mRNAs ( Figure 1B, lanes 8 and 9). 9 clones were analyzed and all mRNAs displayed 59 ends at position 248 from the atpI AUG and 39 ends either at position +151 (2 of 9 clones) or at position +295 (5 of 9 clones) downstream to the atpH stop codon (b9 in Figure 3A). These data demonstrate that stable atpI mRNAs do exist mainly as di-cistronic atpI/H mRNAs, which use the same barrier modules for 39 end definition of the atpH mono-cistronic mRNAs.
As atpI mRNA could also result from co-transcription with the preceding rps2 gene, we performed Northern blot hybridizations using single strand DNA probes complementary to either rps2 (data not shown) or atpI (Figure 2A and B, probe 1) coding regions.
Only one mRNA of about 2600 nucleotides (diamond) was revealed by both probes, suggesting that the remaining atpI bands correspond to transcripts which initiate at the atpI cistron. Two bands (a9 and a) of around 1300 and 1500 nucleotides, respectively, were detected, that from the size could correspond to atpI mono-cistronic mRNAs. In our previous analyses we have observed that only the larger of the two bands (a) disappeared in the sig2 mutant [7]. These data suggested that the larger and very faint band (a) corresponds to the SIG2 dependent primary atpI transcripts (starting at position 2229) while the smaller band (a9) corresponds to the processed atpI mRNAs (with the 59 end at  [3,4]. Upward (with right tip) and downward arrows indicate the 59 end positions of primary and processed transcripts, respectively, with negative numbers corresponding to the distance from the ATG as described [7]. Left and right-directed arrow heads correspond to primers for the cRT-PCR analysis. B. Mapping of processed atp transcript ends by cRT-PCR. Agarose gels showing the cRT-PCR data are presented, with the positions of the molecular weight markers on the right. For atpH transcripts (lanes 1-5), h1 was used as RT-primer and h2-h3 as PCR-primers. For atpI transcripts (lanes 6-7), i1 was used as RT-primer and i2-i3 as PCR-primers. For atpI/H transcripts (lanes 8-9), i1 was used as RT-primer and i2-h3 as PCR-primers. For atpF transcripts (lanes 10-11), f1 was used as RT-primer and f2-f3 as PCR-primers. For atpA transcripts (lanes 12-13), a1 was used as RT-primer and a1-a3 as PCR-primers. For atpH/F/A transcripts (lanes 14-15), h1 was used as RT-primer and h2-a2 as PCR-primers. 59 processed transcripts were distinguished from primary mRNAs because PCR products were also obtained in absence (-TAP) of a previous TAP-treatment of the RNA samples. The -RT (lanes 1 and 2) and -PCR (lane 5) controls are shown only for the mono-cistronic atpH mRNAs. C. Mapping of primary atp transcript ends by cRT-PCR. For atpI/H transcripts (lanes 1-2), i1 was used as RT-primer and i4-h3 as PCR-primers. For atpH transcripts (lanes 3-4), h1 was used as RT-primer and h4-h3 as PCR-primers. doi:10.1371/journal.pone.0078265.g001 position 248, as based on our cRT-PCR data). Two bands of the expected size for di-cistronic atpI/H transcripts (around 1800 nucleotides) were detected (bands b and b9). They were also revealed by DNA probes hybridizing to the atpH coding region (data not shown) or the 39 UTR ( Figure 2B, probe 2). Our previous analyses have shown that only the larger of these two bands (b) was SIG2 dependent, corresponding to the SIG2dependent primary atpI/H co-transcripts (starting at position 2229, see below), while the smaller band (b9) likely corresponds to the processed atpI/H mRNAs (with the 59 end at position 248 on the basis of the cRT-PCR data in this manuscript) [7].

Mapping of the atpF and atpA transcript ends
According to the overlapping model, one would expect that the 39 termini of atpH mRNAs ending nearby a protein footprint (at position +297) overlap with the 59 ends of transcripts encoded by the downstream gene, atpF. When primers were designed inside the coding region (f2 and f3 primers, Figure 1A), a sharp RT-PCR product was obtained with both TAP untreated and treated RNA samples ( Figure 1B, lanes 10 and 11), indicating that the transcripts result from a processing event at the 59end. As the cRT-PCR indicated only one mRNA type we directly sequenced the PCR product. We found that the atpF mRNAs displayed a 59 end at position 245 from the atpH AUG and a 39end in the intergenic region between the atpF and atpA ORFs, at position +44 downstream to the atpF stop codon (f in Figure 3A and bottom panel). These data indicate that the atpF mRNAs are mostly dicistronic (together with the atpH cistron) and contain very stable 59 and 39 ends. Therefore, we could not detect mono-cistronic atpF mRNAs with 59-ends overlapping with the atpH 39-ends. The existence of atpH/F transcripts was further confirmed by Northern hybridization, that revealed a very abundant band (f) of around 1300 nucleotide in size using specific probes hybridizing to the atpF coding region (Figure 2A and B, probe 5) and also to the atpH 39 UTR (probes 2, 3, 4). No smaller bands corresponding to atpF mono-cistronic transcripts could be observed. We could not find the transcripts ending at position 230 upstream to the atpF ORF that were previously identified by primer extension [7], probably because they are very rare. In any case, no overlapping is possible between the 39 ends of the atpH mRNAs ending at +297 (corresponding to position 2166 upstream to the atpF start codon) and the 59 ends of the atpF mRNAs starting at 230 (see Figure 3A).
On the other hand, a sRNA reflecting the footprint of a RNAbinding protein was identified at the 39 end of the atpH/F cotranscripts at position +44 (corresponding to position 227 from the atpA AUG) [4]. To determine if the overlapping model applies to transcripts of the atpF/A adjacent genes, we next mapped the atpA mRNA ends. A weak and fuzzy PCR product was obtained ( Figure 1B, lanes 12 and 13) when using the a1 and a3 primers ( Figure 1A). We found that out of 8 clones, 5 clones presented the 59 end at position 2237 from the atpA start codon, inside the atpF coding region. The remaining 3 clones (3/8) presented the 59 end at the 252 site from the atpA start codon, just upstream to the protein footprint, as expected by the overlapping model ( Figure 3A, bottom panel). Both the 2237 and the 252 ends result from a 59 processing event as suggested by the presence of the amplified products also in the TAP-untreated sample ( Figure 1B, compare  lanes 12 and 13). However, surprisingly, all the clones presented truncated 39 ends inside the atpA coding sequence (h and h9 in Figure 3A). Thus, we conclude that we could detect atpA monocistronic mRNAs with 59 ends overlapping with atpF 39 ends ( Figure 3A, bottom panel). However, these transcripts are all truncated at their 39 ends and cannot be translated into functional protein products. Other short atpA mRNAs initiating at position 274 from the AUG were previously mapped by RACE but were detected only at a very low level in primer extension experiments [7]. Concordantly, when a Northern blot hybridization was performed using a double stranded DNA probe hybridizing to the atpA coding sequence (Figure 2A and B, probe 6), no bands of the expected size for mono-cistronic atpA transcripts (around 1500 nucleotides) could be detected. Only longer RNAs of around 3500 nucleotides were clearly revealed (band g). Detection of the same RNAs (g) with probes 2-6 indicates that they correspond to  Figures 1A and 2A. Downward and upward arrows indicate the position of 59 and 39 ends of the atp transcripts, respectively, on the bases of cRT-PCR (this manuscript) and primer-extension data [7]. For the 39 ends of the atpI/H (b9), atpH (e) and atpH/F (f) mRNAs, both the distance from the corresponding stop codon and from the start codon of the downstream cistrons is indicated. Middle panel: transcripts are labeled by letters as in Figure 2B. The thick continuous lines correspond to atp transcripts containing defined ends; the discontinuous lines of the (a9, b, c and h9) atp mRNAs indicate the presence of dispersed 39 ends. Distinct RNA-binding proteins that are likely involved in definition of transcript ends are depicted as circles with different fillings. 39 ends that are likely protected by stem-loop structures are depicted as stem loops. The triangle (dotted line) within the atpF mRNA indicate splicing of the intron. Bottom panel: the sequence of overlapping regions between atp transcript ends is indicated. The numbers below empty or full arrowheads correspond to the ratio of the clones displaying a specific 59 or 39 transcript end, respectively. The term ''PCR'' indicates when sequencing was performed directly on the PCR product. The sequence corresponding to the sRNAs, as reported in [3,4]  poly-cistronic atpH/F/A mRNAs. The presence of abundant polycistronic atpA mRNAs is also supported by our previous analyses using the primer extension technique [7].
To map both ends of the poly-cistronic atpH/F/A mRNAs, we designed primers inside the coding sequences of the atpH (h2) and atpA (a2) genes ( Figure 1A). We obtained a very sharp PCR band that could be directly sequenced ( Figure 1B, lanes 14 and 15). The 59 end of the atpH/F/A mRNAs maps at positions 245 upstream to the atpH AUG ( Figure 3A, bottom panel) and the 39 end maps at position +448 downstream to the atpA stop codon (g in Figure 3A). No sRNAs were found close to the +448 position [3]. However, the upstream 50 nucleotides can fold in a very stable stem loop structure with a predicted DG of 229.49 kcal/mol (using the MFOLD 3.2 program, http://mobyle.pasteur.fr/cgibin/portal.py?#forms) ( Figure 3B, right panel). These data suggest that the 39 end of the tri-cistronic atpH/F/A mRNAs is stabilized by a secondary structure rather than by a protein barrier.
Finally, we analyzed the atpI/H/F/A co-transcripts. Long transcripts of the expected size (around 4600 nucleotides) could be detected in Northern blot analyses exclusively with the atpI probe ( Figure 2A and B, probe 1, star). These data suggest that the amount of poly-cistronic mRNAs covering the entire large atp operon (from atpI to atpA genes) is probably very small in relation to the total amount of different atp mRNAs. Concordantly, no PCR product was amplified when the primers were designed inside the atpI (i2) and the atpA (a2) coding regions ( Figure 1A).

Mapping of the primary atpI/H and atpH transcript ends
Finally, to determine if primary and 59 processed transcripts differ at their 39 ends, we mapped the termini of primary transcripts which are initiated from both the atpI and the atpH promoters. We first characterized the atpI/H co-transcripts, which are initiated at the position -229 from the atpI promoter [7]. We designed a reverse primer, i4 ( Figure 1A), that was specifically hybridizing to the region between the transcription start site (2229) and the 248 processing site. This allowed to exclude all processed mRNAs and to increase the chance to clone the primary transcripts. As expected for primary transcripts, PCR amplification with i4 and h3 primers was obtained when RNAs were TAP-treated ( Figure 1C, compare lane 1 with 2). All the clones that have been analyzed (11) contained 59 ends around position 2229. 7 of the 11 clones contained 39 ends at position +151, where a stem-loop structure is predicted to fold. However, the remaining 4 clones had different 39 ends, which were scattered within and in between the reading frames of atpH and atpF mRNAs (b in Figure 3A). Furthermore, we characterized the primary atpH transcripts which are initiated at the 2418 position from the internal atpH promoter [7]. We designed a reverse primer, h4 ( Figure 1A), that was specifically hybridizing to the region between the internal transcription start site (2418) and the 245 processing site. As expected for primary transcripts, PCR amplification with primers h4 and h3 was obtained when RNAs were TAP-treated ( Figure 1C, compare lane 3 with 4). A smear of PCR products was obtained indicating heterogeneous distribution of mRNAs. All the clones that have been analyzed (10) contained 59 ends at position 2418 but had different 39 ends, which were scattered within and in between the reading frames of atpH and atpF mRNAs (c in Figure 3A). Altogether, these data demonstrated that a large part of the primary atpI/H and atpH transcripts is not stabilized at the 39 end but is rapidly degraded. Degradation is more pronounced for the atpH mRNAs, but it is still detectable for the primary atpI/ H mRNAs. Our results indicate that the primary and the processed mRNAs have different 39ends and different stabilities, suggesting that a processing event at the 59 ends might be required for definition and stabilization of the 39ends.

Discussion
Our study demonstrates that the 59 ends of processed atpI and atpH mRNAs and the 39 ends of processed atpH (at position +297) and atpH/F mRNAs map close to the sRNAs sites ( Figure 3A), which were identified as footprints of RNA-binding proteins [3,4]. Our data therefore support the model that RNA-binding proteins (PPR proteins or other classes of proteins with the same barrier function) are responsible for the protection of the mRNA ends from nuclease degradation. However, only a small proportion of atpI transcripts display overlapping ends with the atpH mRNAs, which are encoded by an adjacent cistron. In addition, the atpA RNAs, which map close to a protein footprint and overlap with the atpF transcripts of the upstream cistron, are truncated at the 39 end and clearly correspond to degradation products. Therefore, the overlapping protection mechanisms appear to have only a limited role in obtaining stable processed mRNAs of the large atp operon in Arabidopsis. Based on our data we propose a new model for end definition of processed transcripts of the atp operon in Arabidopsis chloroplasts. The first site for RNA-binding proteins that follows a transcription initiation site serves to protect and define the 59 end of the nascent poly-cistronic transcripts, while all the following protein-binding sites or hairpin structures that are encountered by the progressing RNA polymerase serve to define the 39 ends of the transcripts. If this model applies also to other operons requires further investigations.
We observe that the processed atpI/H and atpH mRNAs are more efficiently protected at the 39 end than the corresponding primary transcripts. Our data suggest that a processing action at the 59 terminus might be required for definition and stabilization of the 39 terminus, indicating a sort of communication between the ends of an RNA molecule during processing. Interestingly, our observations are reminiscent of the end-protection mechanisms of most eukaryotic nuclear transcripts where the CAP binding protein, protecting the 59 ends, is deposed on the transcripts by the RNA polymerase II in a co-transcriptional manner [12] and enhances the processing events (cleavage and poly-adenylation) at the 39 ends [13]. If also in chloroplasts the RNA-binding proteins are placed on the mRNAs by the RNA polymerase in a cotranscriptional manner is still an open question.
Our study indicates that both mechanisms of mRNA protection, either mediated by RNA-binding proteins or by hairpin structures, can operate in parallel and within the same operon. In addition, the mono-cistronic atpH mRNAs, which are by far the most abundant transcripts of the atp operon [10], are likely stabilized by a RNA-binding protein at the 59 end and both a hairpin and a RNA-binding protein at the 39 end. In case that the 39.59 exonuclease overruns the protein-mediated protection site, the degradation process can be then blocked by a second barrier, the hairpin structure. Thus in addition to transcription from an internal promoter and protein-mediated protection at the 59 end, the amount of atpH mRNAs might be enhanced by a double protection at the 39 end.

Plant growth conditions
Arabidopsis (Arabidopsis thaliana) seeds (ecotype Columbia) of wildtype plants were surface-sterilized for in vitro culture. Seeds were spread on MS agar plates with sucrose, kept for 72 h at 4uC in darkness and then grown at 23uC under a 16 h/8 h light/dark cycle at 70 mmol of photons m 22 s 21 . Plants were harvested after 7 days.

RNA purification
Frozen material of plants was ground in a mortar and resuspended in 3 volumes of solution A (10 mM Tris-HCl pH8; 100 mM NaCl; 1 mM EDTA; 1% SDS) and 2 volumes of phenol/chloroform/isoamyl alcohol (25:24:1; v/v/v). After centrifugation, RNAs in the aqueous phase were again extracted twice with phenol-chloroform and finally once with chloroform. After over-night precipitation in 2 M LiCl at 4uC, RNAs were then precipitated in ethanol.