Four B. aphidicola genomes (BAp, BSg, BBp, BCc) from the following four aphids: Acyrthosiphon pisum, BA000003 1 ; Schizaphis graminum, AE013218 3 ; Baizongia pistaciae, AE016826 4 ; Cinara cedri, CP000263 5 , were used in this work. A fifth genome sequence was published recently from the aphid Cinara tujafilina (CP001817 6 ), very closely related to BCc. As BCt and BCc show very similar genomic repertoires and properties, BCt was not always included in the analyses of this study. The annotations of transcription units (operons) in BAp are those described in 26 .

In this work, the E. coli genome was used as the reference ancestral state of the Buchnera lineage, disregarding the evolution in the branch of E. coli. The analyses were performed on the genomic sequence of Escherichia coli str. K-12 substr. MG1655 (U00096) 45 46 and information concerning the gene regulatory network was taken from the RegulonDB 6.2 database 47 .

The inventory of the transcription factors encoded in the Buchnera genomes was primarily carried out using the close proximity of Buchnera to E. coli and the one-to-one orthology relationship for almost all of the genes of Buchnera, as referred to in BuchneraBase 48 .

We completed this inventory by searching in the Buchnera protein set for all Pfam domains annotated as regions with putative regulatory functions 49 . As helix-turn-helix (HTH) domains are by far the most common protein DNA-binding domains in bacteria 50 51 , a systematic search for these structural domains was also carried out for the complete set of proteins in Buchnera, using the HTH software 52 53 . Non-HTH motif predictions such as zinc fingers, helix-loop-helix, beta-sheet antiparallel or RNA binding domains are available in the generalist DNA Binding Domain Database 54 . Indeed, three non-HTH predictions are available for Buchnera: CspE and CspC each with a cold-shock domain and DksA with a zinc finger domain. These proteins were already detected by their orthologous annotations in E. coli.

Promoters and transcription starts were predicted, in the four Buchnera strains, using the software BPROM^© (http://linux1.softberry.com/berry.phtml) and MacVector (MacVector, Cary, NC, USA) for σ⁷⁰ and σ³² promoters, respectively. BPROM was calibrated for E. coli with a specificity of about 80%. In this work, predictions were made using the E. coli parameters in the 500 bp regions located upstream from all the Buchnera coding DNA sequences (CDS). The motif detection function of MacVector was calibrated using the two consensus σ³²-boxes (CTTGAAAA and CCCCTNT), separated by 11–15 bp 55 , with a tolerance of 50% similarity, according to previous work performed in BSg 12 .

The transcription factor binding sites (TFBS), known in E. coli for the different NAPs, were searched for in the Buchnera genomic sequences using the words.pos and gregexpr R-functions from the SeqinR 56 and base libraries, respectively. The following consensus motifs were searched: GNTYAWWWWWTRANC for FIS 57 , WAT CAANNNNTTR for IHF 58 and TCGWTWAAWW for H-NS 59 .

All the statistical analyses were performed using the R software (http://www.r-project.org). The Gene Ontology (GO) annotations were extracted from the UniprotKB-GOA database 60 . Annotations were compared at the same level (i.e., level 3 or 4) to avoid redundancy-bias linked to the possible non-homogenous depth of the different branches of the GO hierarchy.

The C3P software, developed by Boyer et al. 61 , is a graph-theoretical approach extracting common connected components between two or more graphs for exploring gene neighbourhoods in a genomic and functional context. If we consider two genes, x and y, located in close proximity on the genome, the main principle of this method is that if x and y are co-regulated (i.e., neighbours in the transcription regulatory network), or if the proteins X and Y are interacting directly (i.e., neighbours in the protein-protein interaction network), or are catalyzing successive steps in a metabolic pathway (i.e., neighbours in the metabolic network), the relative positioning of the genes x and y should be preserved during genome evolution if selective pressures are acting on the transcriptional, protein-protein interaction or metabolic networks, respectively. In the present study, we have applied this approach to extract the common connected components between the Buchnera genome, on the one hand, and the transcription network, protein interaction network and metabolic network on the other hand. Two genes are considered as neighbours if they are separated by a maximum of one gene in the molecular interaction networks and by five genes on the genome. A set of neighbour genes (or of proteins encoded by a set of neighbour genes) is called: (1) a synton in the genome; (2) a transcripton (this term was not defined in the original work of Boyer et al. 61 ) in the transcriptional regulatory network; (3) an interacton in the protein-protein interaction network; or (4) a metabolon in the metabolic network. The transcriptional network, together with the protein-protein interaction and the metabolic networks in Buchnera, were inferred from E. coli by direct orthology. In order to test the significance of the structures described in BAp (i.e., to establish whether they could have been observed by chance), we developed a procedure to simulate random transcriptons (r-transcriptons). For that, X genes were randomly selected, within a limited span of 5 genes in the E. coli genome, to generate an r-transcripton. The size (X) of each r-transcripton was defined by sampling the sizes of the true transcriptons of E. coli. The r-transcriptons were retained only if they shared at least two orthologous genes with BAp. Each simulation was ended either when 38 r-transcriptons, sharing at least two orthologues with BAp, were generated or when all the genes of E. coli had been used since gene sampling was performed without replacement. Since a number of r-transcriptons equal to, or greater than 37 were found in only 9 out of the 1000 simulations, a p-value of 9. 10^-3 was estimated (see Results and discussion section and Table 1 for the justification of the thresholds).

^a: Number (#) of Common Connected Components.

We also analysed the evolutionary history of each transcription unit (TU) making up the E. coli transcriptons and conserved in BAp. For that, several types of TUs were defined: identical TUs are BAp TUs with exact orthologous replicates in E. coli whereas similar TUs are BAp TUs that underwent gene deletions in BAp (or more rarely gene insertions in E. coli). Split, merged and reorganized TUs are BAp TUs that were recomposed from different “ancestral” TUs during Buchnera evolution. More details and schemes are given in 26 .

Four physical sequence-dependent properties of the DNA molecule were computed using the GeneWiz software 25 for the four sequenced Buchnera strains: stress-induced duplex destabilization - SIDD 42 , curvature 62 , base stacking energy 41 and base-pair propeller-twisting 63 . The calculation of these different parameters with GeneWiz leads to a single value for each base of the DNA sequence. Non-overlapping sliding windows of different sizes were then defined to calculate the smooth parameter-distributions at different scales (e.g., 300 bp for the whole genome sequence analysis and 20 bp for the promoter region analysis).

For the whole genome comparison between Buchnera and E. coli, two null models of base composition were constructed. The first one (global null model) is a uniform permutation of all the bases of the genome (i.e., preserving only the global GC-content on the genome scale) whilst the second one (local null model) is a uniform base permutation applied to each coding and non coding region (i.e. preserving the local GC-content, as well as the coding versus non-coding GC content).

In order to determine whether protein interactions (across protein, metabolic or transcription regulatory networks) exerted a selective pressure on the conservation of some genomic regions in BAp, we used the C3P program 61 to calculate the number of interactons, metabolons and transcriptons conserved between E. coli and BAp (see Methods section for the definitions). Essentially the idea is that if selective constraints have an effect on the three latter structures, and if the genes are close neighbours on the E. coli genome (supposed, here, to reflect the ancestral state), we expect to see conservation of the corresponding orthologous genes in the equivalent neighbourhood in BAp. Indeed, Table 1 shows that 37/38, 9/10 and 23/23 of the E. coli transcriptons, interactons and metabolons, respectively, found in BAp are conserved as common connected components (so the proximity between genes was conserved between E. coli and Buchnera for these particular genes). These results are significant (i.e., not observed by chance), as revealed by a re-sampling test giving a p-value of 0.009 for the BAp transcriptons (see Methods section). The procedure was not applied to the metabolons and interactons because they are bigger structures, with a lower probability of being observed randomly as conserved associated structures.

Moreover, we have shown that, in addition to the 37 “ancestral” transcriptons, 12 supplementary ones have been produced in BAp, by genomic rearrangements, since its divergence with E. coli. We also analysed the evolutionary history of the TUs composing the transcriptons, found in BAp, following on from our previous work on TUs in Buchnera 26 . Among the 55 TUs composing the 38 transcriptons found in both BAp and E. coli, 16 are monocistronic, 13 are identical or similar TUs (totally or partially conserved from gene deletions) and 26 are TUs that were reorganised during genome evolution, i.e., formed from a fusion or split from different ancestral TUs (see Methods section for definition of the different TU rearrangements). These results reveal that these two bacterial lineages conserved not only some single operonic structures but also some bigger synthenic fragments associating several TUs. Moreover, genomic rearrangements occurring in the BAp lineage during genome shrinkage seem to have clustered some co-regulated genes within the same neighbourhood (i.e., new transcriptons).

During the process of genome shrinkage in the Buchnera lineages, which followed the symbiotic association with aphids, gene losses occurred preferentially within functional classes that escaped from selective pressure due to the new intracellular environment. A systematic search for under- and over-represented functional classes of genes was performed in BAp, relative to those found in E. coli (Additional file 1A), with a specific focus on gene expression regulation. Hence, as previously mentioned in the primary annotation 1 , gene proportions within the GO classes of transporter activity, developmental processes, response to stimulus, localization and biological regulation are significantly lower in Buchnera, compared to E. coli, whereas proportions of many classes corresponding to central core metabolism (e.g. catalytic activity, metabolic processes and cellular processes) are significantly higher.

Moreover, we demonstrate here that the genes preferentially conserved in BAp are mostly the multifunctional ones. Indeed, conserved genes in BAp have a significantly higher association with multiple GO annotations since the distribution of the number of GO terms associated with BAp genes shows more genes with 2 to 8 GO terms, when compared to E. coli (Additional file 1B, Wilcoxon rank test p-value = 2 10^-16). These multifunctional genes are mostly metabolic genes (i.e., associated with the GO term GO:0008152). Indeed, when the analysis is repeated, after removing those genes associated with the corresponding GO term, the distributions remain almost equivalent (Additional file 1C, Wilcoxon rank test p-value = 0.07). This result is important in the context of genome shrinking, demonstrating that genes encoding multifunctional proteins may be favoured in small genomes.

Buchnera have lost a number of genes from the catabolic pathways since most of the end products, synthesized by Buchnera, are exported to the host (only 9% of the E. coli catabolic pathways are present in BAp). Salvage pathways have been interwoven between the two symbiotic partners during symbiotic evolution 8 64 . It is important to note that catabolic genes are generally more highly regulated in E. coli, as compared to anabolic genes 65 . Indeed, 83% of the catabolic genes are regulated in E. coli, versus only 40% for the anabolic genes. Hence, the loss of catabolic genes partly explains the decay of the gene expression regulatory system in BAp.

A systematic search was performed for the complete set of proteins of the four Buchnera strains (BAp, BSg, BBp and BCc) using one-to-one orthology with annotated genes of E. coli, scanning for annotated Pfam domains and HTH domains (see Methods section). Nineteen proteins were detected and are presented in Table 2. The corresponding orthologous genes were searched for within the newly sequenced BCt genome and are presented in the same table. Although no new regulator was discovered, this is the first time that such a global and comparative analysis across the Buchnera strains has been published.

^a: protein sequence identity (%) with the E. coli orthologous gene; ^b: genome sequence of BCt was not analysed systematically, we have only reported here the presence/absence of orthologous genes; ^c: proteomic data were compiled from 66 , +/− (rank) indicates the presence/absence of the protein in partially purified samples of BAp (rank of the quantitative detection out of 400 proteins).

In BAp, 699 σ⁷⁰ promoters were detected, in the 500 bp regions located upstream of the 574 ORF of the genome, which corresponds to 94% of the CDS and 96% of the TUs showing at least one significant promoter prediction. The same observations were made for BSg, BBp and BCc CDS where for 94%, 95% and 95% of the CDS, respectively, we were able to find σ⁷⁰ promoters (Additional file 4). Knowing that BPROM scores are proportional to the similarity with the consensus E. coli boxes, it is worth noting that the scores predicted within the BAp TUs (putative internal promoters) were lower as regards the scores of the promoters located upstream of the TUs (Additional file 5A). As a comparison, in E. coli, a significant σ⁷⁰ promoter prediction was found upstream of 89% of the CDS and 92% of the TUs, and the size distribution of the corresponding 5’ UTRs (5’ UnTranslated Region) was similar in the two organisms (186 ± 127 and 200 ± 128 bp in BAp and E. coli respectively), as shown in the Additional file 5B.

In order to validate our prediction, we analyzed the SIDD profile (see below and the Methods section) of the 400 bp regions located around the start codon of all the CDS found in BAp, differentiating between the intra- and inter-TU regions (Figure 2). Despite the AT-richness of the intergenic regions in BAp, a characteristic profile with an instability sink was observed at about 100–150 bp upstream of the start codon, and this was also observed for the strains BSg, BBp and BCc (data not shown). Likewise, this sink is present in E. coli, where it is located around the start codon. It is important to add that internal promoters located within BAp TUs showed a similar profile centred around the start codon, but with higher SIDD values, i.e., they are more stable (data not shown).

Taken together, all these results (conservation of promoters, intergenic distances and thermodynamic profiles) concur with the prediction for a functional role of the σ⁷⁰ promoters, despite the high rate of sequence evolution in Buchnera.

A similar analysis was performed for the σ³² promoters (located 500 bp upstream of CDS) and for the four Buchnera strains BAp, BSg, BBp and BCc where, respectively, 248, 238, 179 and 98 σ³² promoters were found. Nevertheless, it is significant that, in BAp, among the 45 conserved genes from the σ³² regulon of E. coli, only 19 (42%) had retained a recognizable σ³² promoter. Moreover, when comparing the putative σ³² regulons of these four Buchnera strains, they appeared to be very divergent, revealing that predicted σ³² promoters are not tractable in Buchnera without any experimental data (Additional file 6). These results are quite important as regards the possible role of σ³² in the AT-rich genomes of insect endosymbionts, characterised by mild transcriptional changes in response to heat shock 98 99 . It has been proposed that, in symbiotic genomes, this transcriptional regulator might control currently unknown stress signals, or it may have been transformed into an alternative vegetative σ factor in order to replace the lost σ factors (σ²⁴, σ²⁸, σ³⁸ and σ⁵⁴). Indeed, the very high number of σ³² promoters found in the Buchnera genomes, coupled with the inconsistency of the predicted corresponding regulons, is likely to reflect a high rate of false positives rather than the acquisition of a more generalist role for this σ factor.

A systematic search of TFBS was also performed for the three NAPs for which a known TFBS has been described (see Methods section). However, in Buchnera, all the predicted TFBS are under-represented as regards what would be randomly expected, so it was not possible to extract significant TFBS even for these proteins (data not shown).

The role of DNA negative supercoiling in the regulation of transcription activity in bacteria is now well established 19 . Here, we analysed the sequence dependent features determining some of the physical properties of the DNA molecule of Buchnera, namely intrinsic curvature and SIDD. These parameters enabled us to highlight the most favourable promoter regions for transcription initiation, together with base stacking energy and base-pair propeller-twisting relevant to the flexibility and stability of the DNA. Physical properties of the DNA molecule of BAp were compared to those of E. coli using the global and local null models of base-composition as references (see Methods section). The results are presented in Figure 3. As a direct effect of its AT-richness, the DNA molecule of BAp is much more curved, flexible and shows less base-pair stability compared to that of E. coli. The curvature and the flexibility provided by the propeller twist are more pronounced in BAp than in the global and local null models.

The SIDD is less closely correlated to the AT-composition and it does not allow for discrimination between E. coli and BAp. In order to detect regions more prone to harbour functional promoters, we analysed the SIDD scores within the intergenic regions of BAp. Indeed, divergent intergenic regions (containing two 5’ UTR sequences) encode more promoters than convergent ones (containing two 3’ UTR sequences). Hence, the SIDD scores serve to separate the clearly divergent (less stable) and the convergent (more stable) intergenic regions. Interestingly, tandem regions (containing one 3’UTR and one 5’UTR), that could harbour either a TU external promoter, a TU internal promoter or even no promoter at all, reveal a bimodal distribution separating out the stable and the unstable regions (Figure 4A). We analysed, in more detail, the SIDD distribution within the tandem regions (Figure 4B) and we found that the intra-TU promoter regions were the most stable, whereas the inter-TU promoter regions still showed a bimodal distribution of stable and unstable regions. Referring to our previous work on TUs in BAp 26 , it is probable that stable promoter regions correspond to false predicted inter-TU regions in this strain since long TUs, associating non-functionally related genes, seem to be over-abundant in BAp.

We analysed the correlation between the SIDD values of each gene promoter region (calculated in a window of 150 bp, located upstream of the CDS start) and the corresponding functional gene classes in BAp, using the multiFun classification 100 . Two classes were found to be significantly correlated with unstable promoter regions (Wilcox rank test): the transporter class (p-value = 0.03) and the regulator class (p-value = 2 10^-4). Significant gene lists are given in the Additional file 7. The GC-content of these intergenic regions are not significantly different from the overall intergenic GC-content of BAp, hence this correlation cannot be explained only by a strand bias or a local GC-bias effect (data not shown). The association between unstable promoters and genes involved in transcription or transport was described for 38 out of 43 free living bacteria analysed by Wang and Beham 40 , whereas the authors pointed out that this regulatory mechanism was lost in a group of obligate parasitic bacteria (for 14 out of the 18 analysed) including Chlamydia and Mycoplasma species.

We also looked for associations between unstable promoters and functional gene properties in a data set separating the sensitive from the insensitive genes with respect to supercoiling variations in E. coli 19 23 , as well as in several expression data sets of BAp involving stimulations of the essential amino acid metabolism 14 15 93 . However, no significant correlation was found (data not shown) probably because sensitive genes are different between BAp and E. coli and because the expression data were average for the total Buchnera populations within aphid tissues (i.e., maternal and embryonic populations with different nutritional requirements and stress sensibility).

The inventory of the regulatory elements of Buchnera reveals the very low diversity of specific regulators and the conservation of several NAPs and topoisomerases. These results are also consistent with the attenuated expression profiles observed in BAp and BSg microarray experiments 12 13 14 since NAPs induce continuous changes in the gene transcription rates, whereas local transcription factors induce discrete changes (i.e., On/Off transcription rates) 19 101 . DNA-superhelical density has not, as yet, been measured in Buchnera and changes in the supercoiling induced by variations of the intracellular environment of their bacteriocytes remains speculative. Hence, the question of the functionality of DNA-topological regulators and of their links with gene expression regulation is still open.

Several evidences in favour of a selection of regulatory structures in Buchnera despite genome erosion are provided by this genomic analysis: (1) the conservation and the creation of new transcripton structures during genome shrinkage; (2) the conservation of a few specific transcription factors, corresponding to multifunctional enzymes, that may have acquired a broader regulatory role; (3) the conservation of the regulator DksA, in the absence of the alarmone ppGpp, possibly to enhance transcription elongation in these AT-rich genomes and perhaps to discriminate between stable and unstable promoters, as suggested by Srivatsan and Wang in E. coli 102 ; (4) the conservation of several NAPs and of their positioning organisation along the chromosome; (5) the conservation of unstable promoters upstream of transport and regulatory genes.

This work has allowed us to propose some assumptions about the role of DNA topology in gene expression in Buchnera. Hence, AT-bias is generally considered to be either a consequence of the degeneration of the repair system in Buchnera 103 or for energetic selection linked to the centrality of ATP in the metabolism of the bacterium 104 . We have proposed here that the AT-richness of the DNA molecule of Buchnera might provide a selective advantage, giving more flexibility and curvature to the chromosome and facilitating its regulation with only a small number of global regulators, such as NAPs and topoisomerases.

Finally, the presence of unstable promoters upstream of transporter genes might allow a basal expression of the transporter genes in BAp, even when conditions are unfavourable (potentially for importing metabolic precursors), and an over-expression of these genes when the energetic level of the cell is high (potentially to export metabolites for the host). Indeed, the ATP/ADP ratio, partly controlling the gyrase activity, is linked to the superhelicity of the DNA molecule in that when the ratio is high the level of DNA-supercoiling is high and DNA strands are destabilized by the constraints increasing the activity of most promoters. On the contrary, when the ratio is low the DNA molecule is more relaxed and only unstable promoters would be active.

Experimental studies will be needed to test whether such ancestral gene expression regulation by DNA-topology remains functional in BAp and in other Buchnera with even smaller genomes and could be a preponderant mechanism of gene transcription regulation in these bacterial cells with a very small diversity of transcription factors. Hence, DNA-topology could regulate some important symbiotic functions, as well as whether superhelical changes may have occurred during the few contrasted physiological states imposed on Buchnera by their integration into the aphid life cycle (e.g., the establishment of symbiosis during embryonic development, the growing phase and activation of symbiotic metabolism, decaying symbiosis in old aphids, etc.).