|
|
PLoS ONE. 2007; 2(9): e847. Published online 2007 September 5. doi: 10.1371/journal.pone.0000847. | PMCID: PMC1952141 |
Copyright Moura et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Large Scale Comparative Codon-Pair Context Analysis Unveils General Rules that Fine-Tune Evolution of mRNA Primary Structure Gabriela Moura,1 Miguel Pinheiro,2 Joel Arrais,2 Ana Cristina Gomes,1 Laura Carreto,1 Adelaide Freitas,3 José L. Oliveira,2 and Manuel A. S. Santos1* 1Department of Biology, Center for Environmental and Marine Studies, University of Aveiro, Aveiro, Portugal 2Institute of Electronics and Telematics Engineering, University of Aveiro, Aveiro, Portugal 3Department of Mathematics, University of Aveiro, Aveiro, Portugal Alan Christoffels, Academic Editor Temasek Life Sciences Laboratory, Singapore Received May 25, 2007; Accepted July 31, 2007. Background Codon usage and codon-pair context are important gene primary structure features that influence mRNA decoding fidelity. In order to identify general rules that shape codon-pair context and minimize mRNA decoding error, we have carried out a large scale comparative codon-pair context analysis of 119 fully sequenced genomes. Methodologies/Principal Findings We have developed mathematical and software tools for large scale comparative codon-pair context analysis. These methodologies unveiled general and species specific codon-pair context rules that govern evolution of mRNAs in the 3 domains of life. We show that evolution of bacterial and archeal mRNA primary structure is mainly dependent on constraints imposed by the translational machinery, while in eukaryotes DNA methylation and tri-nucleotide repeats impose strong biases on codon-pair context. Conclusions The data highlight fundamental differences between prokaryotic and eukaryotic mRNA decoding rules, which are partially independent of codon usage. A myriad of evolutionary forces shape the primary structure of coding components (ORFs) of genomes, herein called ORFeomes. These include genome and gene duplication, chromosome rearrangements, DNA recombination, deletions and insertions, transposition of mobile elements, single nucleotide polymorphisms, nucleotide repeats and biased G+C pressure [1]– [4]. Apart from these DNA replication derived phenomena others arising from DNA transcription, mRNA stability and translation [5]– [7] are also likely to fine tune ORFeomes' primary structure, but their significance is not yet fully understood. At the mRNA translation level, synonymous codon usage and codon-pair context (representing the pair of codons located in the A and P- ribosome sites) are expected to be under selective pressure since they affect mRNA decoding speed and accuracy [8]– [15]. Synonymous codon usage biases are explained mainly by G+C content and only secondarily by constraints imposed by mRNA translation variables [4], namely tRNA abundance, efficiency of tRNA charging, mRNA decoding efficiency (speed plus accuracy), mRNA stability and structure, gene expression, and amino acid composition [7], [13], [16]– [18]. The nucleotides surrounding a codon also influence synonymous codon usage, with the strongest influence arising from the interplay between the last nucleotide of a codon and the first nucleotide of the neighbor codon (N 1N 2N3 N1N 2N 3), the so called N 3-N 1 context [7], [19], [20]. Conversely to codon usage, the forces that modulate codon-pair context, with the exception of the context of initiation and termination codons [16], [21], are still poorly understood. The few studies carried out to date show, however, that codon-pair context has a direct impact on missense, nonsense and frameshifting errors [15], [22], [23]. In E. coli, missense error in vivo, under standard growth conditions, is in the order of 10 −3 to 10 −4 per codon decoded [24], [25]. Frameshifting and stop codon readthrough errors happen at levels of 3×10 −4 to 10 −5 and of 10 −3 to 10 −6, respectively [26], [27]. Under stress, namely amino acid starvation, these basal error rates increase significantly [16], indicating that decoding error in nature may be significantly higher than in optimal laboratory conditions. Furthermore, 30% of the newly synthesized proteins in HeLa, lymph node, L-K b and dendritic cells are defective ribosomal products (DRiPs) that arise from missense, frameshifting and ribosome drop off at mRNA pausing sites [28]. Since protein synthesis utilizes 45% of the cell ATP, 30% DRiP rate represents 11% of wastage of total cellular energy [28]. Whether this is a common trend in all type of cells is unknown, however, peptides resulting from proteasome degradation of DRiPs are a major source of peptides for MHC class I molecules, highlighting an unanticipated role of mistranslation in immune cells [28]. It is not yet clear whether the ribosome drops off randomly or preferentially at specific mRNA drop off hot spots. In other words, it is important to elucidate whether average decoding error (10 −4 to 10 −5) is evenly distributed along mRNAs (average error) or whether it fluctuates along the mRNA? If so, how can decoding error hot spots be identified? In order to obtain insight into these questions and identify mRNA primary structural features that influence mRNA decoding error, we have developed a software package, statistical and graphical tools to study codon-pairs corresponding to ribosomal A- and P-site codons, using genome wide approaches (ANACONDA vs 1.0) [20], [29]. ANACONDA 1.0 already allowed us to demonstrate that codon-pair context is weakly modulated by G+C pressure [20]. In the present study, we have significantly improved ANACONDA (creating its version 2.0) and used it to carry out large scale comparative codon-pair context analysis using complete ORFeome sequences of 81 Eubacteria, 18 Archaea and 20 Eukaryota. The data show that i) codon-pair context is species specific, ii) there are general rules governing its evolution in the three domains of life and iii) in eubacteria and archeae codon-pair context is mainly determined by constraints imposed by the translational machinery, while, iv) in eukaryotes the emergence of DNA methylation and tri-nucleotide repeats influenced codon-pair context. The data suggests the existence of fundamental differences between prokaryotic and eukaryotic mRNA decoding rules and shows that codon-pair context is partially independent of codon usage. New tools for large scale comparative codon-pair context analysis The ANACONDA 1.0 algorithm developed previously [20], [29] simulates the ribosome during decoding by reading Open Reading Frames (ORFs) sequences, starting at the AUG initiation codon and moving the reading window three nucleotides at a time (). While doing this, it memorizes all codon-pairs, which represent A- and P-site codons during mRNA decoding. It then builds a codon-pair contingency table () that renders itself to statistical analysis and permits determination of the codon-pair context bias [20]. The existence of association between codon-pairs is determined through the chi-square (χ 2) test of independence and preferred and rejected pairs of codons are identified through the analysis of adjusted residuals for contingency tables. These rejected and preferred pairs of codons are then displayed in a 61x64 green (preferred) and red (rejected) color coded map that generates a global view of the codon-pair context data for any ORFeome (). ANACONDA 1.0 also clusters the data according to the context preferences and rejections (residuals values) and builds Differential Display Maps (DDM), which represent codon-pair context differences between two different ORFeomes (). In an attempt to identify putative general rules that govern codon-pair context, we have carried out large scale codon-pair comparisons, using ANACONDA version 2.0. For this, new algorithms and tools were developed to convert the 61x64 codon-pair context colour-coded maps into a single colour-coded column containing 3904 lines, representing all possible combinations of pairs of the 64 codons (). ANACONDA 2.0 compared these colour-coded columns, clustered the data and highlighted groups of codons that had similar pair preference and rejection patterns (–). Since the size of ORFeomes varied significantly between bacteria and eukaryotes, ANACONDA 2.0 normalized the data using the biggest ORFeome as a reference data set ( Figure S1). This permitted carrying out direct comparisons of large and small ORFeomes and allowed the study of codon-pair context preferences (positive residual value; green color in the map) and rejections (negative residual values; red color in the map) of 119 ORFeomes of Eubacteria, Archaea and Eukarya, including the human and chimpanzee ORFeomes ( Figure S2A– C). Codon-pair context preferences are species specific Codon-pair context maps showed remarkable diversity from bacteria to high eukaryotes (; Figure S3A– J). For example, codon-pair context preferences of the human ( Homo sapiens or H.s) and mouse ( Mus musculus or M.m) ORFeomes showed several differences, which were unveiled by direct comparison of ORFeomes and construction of Differential Display Maps (DDM) (), as described in our previous study [20]. Conversely, the codon-pair context maps for the chimpanzee ( Pan troglodytes or P.t) and human ORFeomes were remarkably similar (), which was in agreement with the high homogeneity found for codon-pair distributions of both ORFeomes (data not shown). The same trend was found in bacteria. Indeed, the Escherichia coli ( E.c) ORFeome codon-pair context map was more similar to that of Salmonella typhi ( S.t) than to Bacillus cereus ( B.c in ). Clustering of the codon-pair context maps showed that codon-pair context follows rRNA phylogeny (), highlighting the possibility of using codon-pair context maps as species specific fingerprints. Furthermore, the overall correlation between the 3 domains of life was lower than that calculated within each domain, as the Spearman's correlations of the ranks ( Table S1) showed low correlation coefficients between species of different domains of life, i.e. 0,452 for Eukarya vs Archaea; 0,450 for Eukarya vs Eubacteria; and 0,500 for Archaea vs Eubacteria. While correlation coefficients calculated between species of the same domain were high, i.e. 0,988 among Eukarya (between H. sapiens and P. troglodytes); 0,823 among Archaea (between P. abyssi and P. horikoshii); and 0,959 among Eubacteria (between E. coli and S. flexneri). The distribution of residual values over the entire set of ORFeomes showed that the 3 domains of life have significantly different codon-pair preferences ( Tables 1, 2). For example, codon-pair contexts with highest and lowest adjusted residual values showed no common codon-pairs in the 3 domains of life, suggesting fundamental differences between eukarya, eubacteria and archeae in codon-pair rules and in the evolutionary forces that shape ORFeomes primary structure. Interestingly, 9 out of the 10 codon-pair contexts with highest residual values (best codon-pairs) of all eukaryotic ORFeomes were pairs formed by identical codons (codon repeats) ( Table 1). The same trend was also detected when the most frequently preferred codon-pair contexts for each domain were compared ( Table 2). With this approach, common codon-pair contexts were identified for the 3 domains of life. For example, AAU-CCA and GGC-UGU had positive residuals in Eubacteria and Archaea. In Eukarya and Archaea ACU-AAG had negative residuals and AGA-AGA had positive residuals in Eubacteria and Eukarya. This suggested that, despite the species specificity of codon-pair context maps, at least some of the evolutionary constraints that shaped codon-pair context are conserved across species in the three domains of life. | Table 1The most biased codon-pair contexts. |
| Table 2General codon-pair contexts. |
Context preferences exist in coding and non-coding sequences A large-scale codon-pair context comparison was carried out to visualize general context patterns, using clustering tools (). Interestingly, a red region, corresponding to negative residual values (rejected context), appeared across the 119 ORFeomes studied (blue box in ). These rejected codon-pairs were of the general type NNU3-A1NN, where N represents any base. Other general patterns in the map represented either preferred codon-pairs in Archaea and Eukarya (; region-Y), rejected codon-pairs in Archaea and Eukarya (; region-Z) or strongly rejected codon-pairs in Eubacteria (; region-X). In order to evaluate whether those general codon-pair context patterns arose from DNA replication biases, a second large scale comparative map was built using complete genome sequences (coding + non-coding) of the 119 organisms under study. For this, ANACONDA 2.0 scanned full chromosome sequences starting at the first six nucleotides and moved the scanning window three nucleotides at each step. In this way, both coding and non-coding sequences were analyzed and the frequency of all hexanucleotides was computed, without worrying about the DNA strand location or the reading frame of coding sequences, i.e. ORFs were scanned randomly in the frames 0, +1 or +2. This full genome context map () showed patterns that were also observed in the ORFeome map (), confirming that DNA replication biases strongly influence codon-pair context. Since the difference between full genome () and ORFeome () codon context maps could separate global genome biases from translational biases, a DDM was built and the differences between the two were colored using a blue color scale (), as before (). The DDM showed significant differences between full genome and ORFeome maps indicating that codon-pair context is also influenced by evolutionary forces that are not related to DNA replication biases. Interestingly, the column corresponding to eukaryotes was generally darker than the rest of the map, meaning that coding and non-coding sequences are similar in eukaryotes (i.e. stronger influence of DNA replication biases). However, the eukaryotic region of the DDM included the highest differences between ORFeomes and genomes (marked with * in ), suggesting that the eukaryotic translational machinery also imposes strong selective pressure on specific combinations of codons, resulting in a localized higher divergence between coding and non-coding sequences. Codon-pair context is influenced by genome and mRNA translation biases Since DNA replication biases are partly visible at the dinucleotide level [30]– [32], we have constructed individual codon-pair context maps in which rows and columns were sorted to separate P-site codons ending with a particular nucleotide (N3; rows) and A-site codons starting with a particular nucleotide (N1; columns) (). These two consecutive positions of codon-pair context discriminated rather well codon-pair preferences and such discrimination was very strong for high eukaryotes and weak for low eukaryotes and bacteria (). In order to determine whether such dinucleotide bias was linked to translational selection or to overall genome dinucleotide preferences, the dinucleotide bias was determined for the full set of 119 genomes under study (). Overall, rejection of UA dinucleotides in the 3 domains of life was evident; a trend that corresponded to the negative codon-pair context rule (NNU 3-A 1NN) described above. The overall dinucleotide biases were also in agreement with the codon-pair context pattern (). For example, the rejection of CpG dinucleotides in higher eukaryotes (with the surprising exception of the honeybee, Apis mellifera), was also observed in NNC 3-G 1NN codon-pairs (). Other examples were UpG and CpA dinucleotides that were strongly preferred in higher eukaryotes (), a characteristic that was also reflected in codon-pair context maps (). Finally, the dinucleotide biases () showed overall preference for ApA and UpU dinucleotides. This feature originated from frequent tandem repeats of 3 and more identical bases ( Figure S4). The only universal rule detected in the large-scale comparison () contained codon pairs of the type NNU3-A1NN. Since this rule included out-of-frame stop codons, namely UAA or UAG (i.e. NNU3-A1A2N or NNU3-A1G2N), we investigated whether NNU3-A1NN rejection was related to premature translation termination. For this, we constructed a subset of codon-pair context maps in which the contexts containing out-of-frame stop codons were represented (). This approach showed that NNU3-A1A2N and NNU3-A1G2N type contexts were indeed the most negative in almost all ORFeomes. However, NNU3-G1A2N; NU2A3-A1NN and NU2G3-A1NN contexts which also contained out-of-frame stop codons had a majority of positive residual values (green), while NNU3-A1C2N and NNU3-A1U2N contexts that did not contain out-of-frame stop codons had a majority of negative residual values (red). Since some of the positive context rules () included the dinucleotide UpA, which was rejected in the total genomes map (), it is likely that dinucleotide bias is not the only cause for the rejection of codon-pair contexts. On the other hand, premature termination was not the only potential problem here, because NNU3-A1C2N and NNU3-A1U2N did not correspond to out-of-frame stop codons and were also strongly rejected in ORFeomes (). General codon-pair context rules In order to highlight the codon-pair context preferences that were exclusive of coding sequences, the original map of ORFeomes () was rebuilt to show in black cells whose residuals values were similar to those of identical contexts in the complete genomes map (). In this filtered map () green and red colored cells corresponded to those context residual values calculated for ORFeomes that were significantly different from those calculated for complete genomes, i.e. cells that were colored in blue in . This large-scale comparative map allowed extraction of ORFeome specific codon-context patterns, while the converse filtering originated a complete genomes map that permitted extraction of genome specific patterns (). This approach highlighted clear codon-pair context differences between ORFeome and complete genome maps (). Interestingly, these patterns corresponded to different sets of codon-pair contexts that could be easily described by the expressions annotated on the side of each map ( see also Figure S5A, B for different thresholds of visualization). In almost all cases, these codon-pair context rules fixed the last nucleotide of the first codon and the first nucleotide of the second codon, confirming that (N 3-N 1) positions shape codon context. Remarkably, the major patterns that appeared in the filtered map for complete genomes () were related to UpA-rich hexanucleotides that produce weak codon-anticodon interactions in coding regions and should hence be under differential selective pressure in both types of sequences. Mistranslation is a poorly understood biological phenomenon which is influenced by various protein synthesis factors and mRNA primary structure features [13], [33], [34]. In order to shed new light on how the later influences decoding error and extend previous studies carried out mainly on the effect of codon usage on mistranslation [25], [35], we are investigating the effect of codon-pair context on decoding fidelity. Our comparative genomics approaches unveiled the effect of both genome replication and translation specific biases on codon-pair context. The few studies carried out to date on codon-pair context were unable to distinguish those two types of biases [12], [36]– [38]. Our large scale approach confirmed the importance of genomic biases but also unveiled important translational biases that shape codon-pair context and should be primary targets for mistranslation hot spots. Large-scale genomic analysis, such as the one that we have performed, allows for obtaining a global view of mistranslation in a way that is totally out of reach from analysis of single ORFeomes. Indeed, comparison of large sets of codon-pair context data unveiled the main codon-pair context patterns that exist in the 3 domains of life. Interestingly, when the most preferred or repressed codon-pair contexts of all organisms were considered ( Table 1), but also when common rules were selected ( Table 2), there was little or no overlapping between the context patterns of the 3 domains of life. This suggests that genome replication and/or mRNA translation in each domain imposes specific constraints to decoding sequences which produce different codon-pair context outcomes. Also, the phylogeny of individual species appeared as an important determinant of its codon-pair context behavior (), in a similar manner to that described for codon usage bias [39] or dinucleotide genome signatures [31], [32]. Influence of genome wide biases on codon-pair context Our observation that ORFeomes and total genomes produce similar patterns of codon-pair context () confirmed previous studies [4], [40], [41]. This implies that most sequence constraints that can be detected in coding sequences are not imposed by the translational machinery, but arise from selective pressure imposed by DNA replication and related biases. That codon usage biases were mainly due to mutational pressure and only secondarily to translational selection further confirmed the relevance of DNA replication biases on codon-pair context [7]. In this scenario, one is prompted to hypothesize that the translational process may work with sub-optimized mRNA sequences since codon-context fine tunes decoding fidelity [15], [22], [23]. Genomes are known to have biased dinucleotide frequencies [31], a feature that has frequently been used to produce genomic signatures of phylogenetical and taxonomical relevance [31], [32]. At the ORFeome level this bias influences codon usage [32] but may also interfere with codon-context, whenever the last nucleotide of one codon is associated with the first nucleotide of the second codon of the pair. Indeed, (N 3–N 1) contexts explained part of our results () and confirmed the good discrimination obtained when one ORFeome map for codon context was arranged according to the last position of the first codon and the first position of the next codon (). The association bias of two consecutive nucleotides is a characteristic of genomes which results from global selective pressures acting upon DNA at the level of repair and replication mechanisms [32] or ecological constraints that may influence, for instance, the overall G+C content of the genome [42]– [44]. Regulatory activity acting upon the entire genome is another cause of dinucleotide bias. CpG dinucleotides, for example, are signals for DNA methylation, a mechanism commonly used by higher organisms to protect their genome from selfish DNA elements and to regulate gene expression [5], [6]. Our dinucleotide bias analysis for the 119 organisms confirmed a clear rejection of CpG methylation in coding sequences of high eukaryotes, as would be expected, since methylated DNA becomes unavailable for transcription and hence translation [5]. On the other hand, UpA dinucleotides are highly repressed in DNA sequences of most organisms [7], [31], [45], [46]. Interestingly, UpA dinucleotides are sites for preferential hydrolysis of RNA by macrophage ribonucleases [45] destabilizing RNA molecules [7] and should hence be avoided [45]. Furthermore, Duan and colleagues [7] proposed that mRNA stability imposes strong selective pressure on synonymous codon usage and it is likely that this is also true for codon-pair context. Our data confirmed that hypothesis since NNU 3-A 1NN contexts were highly repressed in the 119 different genomes analyzed. Influence of translational biases on codon-pair context As already mentioned, the unique universal rule that could be detected in the 119 genomes analyzed was rejection of most codon-pair contexts of the type NNU3-A1NN (). Clearly, this trend is a direct result of repression of the TpA dinucleotide in total DNA sequences (). However, it was surprising that other UpA bearing contexts did not show strong rejection. For example, NU2A3-A1NN contexts are mainly preferred in coding sequences () indicating strong differences between codon-pairs containing UpA dinucleotides and suggesting that translation does influence codon-pair choice. When a large-scale comparison of codon-pair context excluded global genome biases () it became evident that contexts that were truly produced by translation-driven selection were grouped in negative or positive rules depending on the phylogeny of the organisms. This was in agreement with the previous observation that strongly biased codon-pair contexts were different between the 3 domains of life ( Tables 1, 2), and supported the hypothesis that differences in the translational machineries of different organisms reshape mRNA primary structure in different ways. For example, the NNC 3-N 4NN contexts pattern of higher eukaryotes (B1 and B2 in ) could be explained by specific decoding rules of C-ending codons in Eukarya. Indeed, eukaryotic species translate several C-ending codons by wobble pairing rules using inosine [47], which recognizes A, C, or U at the wobbling position [48] while bacterial species decode most C-ending codons with Watson-Crick C-G base pairing between codon and anticodon [47]. As to the other minor rules highlighted on the left side of , namely NNU 3-G 1G 2R, NRU 3-G 1A 2N and NG 2N-NG 2N or NC 2N-NC 2N, they may be related to both canonical decoding of U-ending codons in eukaryotes (A1 and A2 in ) and to the existence of runs of special sets of amino acids, namely serine/proline/threonine/alanine and arginine/glycine (C in ). That contexts of repeated codons are preferred in eukaryotic genes ( Tables 1, 2) and that proline, alanine and glycine are frequently found in amino acid runs of human genes [49] corroborates the above hypothesis. On the other hand, most of the major genomic constraints that were not present in coding sequences, namely NNU 3-A 1NN, NYU 3-A 1RN and N(U/A) 2U 3-U 1(U/A) 2N or N(U/A) 2A 3-A 1(U/A) 2N rules () were associated to weak decoding interactions involving A-U codon-anticodon pairing. Moreover, these rules are produced by either strong genomic dinucleotide bias against UpA (, rule D) or by rejection of error prone UA-rich codon-pair contexts in coding sequences (, rule F), in a clear confirmation of the additive effect of translational and non-translational selective pressures. Finally, we could also see a preference for trinucleotide repeats in non-coding sequences that was not detectable in coding regions, at least in eukaryotes (, rule E). This has already been described in primates and is related to strong mRNA primary structure constraints associated to high mRNA decoding efficiency [50]. Conclusions Codon-pair contexts are biased in ORFeomes and such bias is the result of both translation and non-translation driven processes. Indeed, translational and DNA replication/repair and cis regulatory elements act synergistically on codon-pair context. This myriad of selective pressures creates significant difficulties to the identification of codon-context biases associated to mRNA translation only. Our large scale comparative genome approach indicated that: i) there is a strong influence of non-translational selective pressures upon coding sequences, especially in eukaryotic organisms since these have a higher degree of resemblance between ORFeome and total genome biases; ii) the strongest non-translational selective pressures that could be identified were dinucleotide biases, mainly imposed by regulatory cis-elements linked to DNA methylation or mRNA stability [5], [45], and preference for trinucleotide repeats, usually associated with DNA polymerase slippage during replication [51]; iii) apart from this non-translational noise, DNA coding sequences showed specific features that could be related to mRNA translation, namely repression of usage of premature termination or error-prone contexts associated to weak codon-anticodon interactions. It will now be most interesting to validate these in silico data in vivo, and identify experimentally the codon-pair contexts that are strongly selected for high mRNA decoding fidelity. Primary data sources Nucleotide sequences, of complete genomes and assembled ORFeomes, were downloaded from GenBank or Ensembl Web sites (Genbank: ftp://ftp.ncbi.nih.gov/genomes/; Ensembl: ftp://ftp.ensembl.org/pub/) between December 2005 and January 2006. These included the DNA sequences of 81 eubacterial, 18 archaeal and 20 eukaryotic species. Plasmid sequences were not included in the analysis and all chromosomal sequences from one genome were analyzed together by ANACONDA 2.0. The total set of files downloaded and used in this study is documented as supplementary data ( Figure S2). Statistical analyses Two-codon context bias was studied in complete ORFeome sequences using the residual analysis tools available in the software package ANACONDA 1.0 (a detailed explanation of this software can be found in [20], [29]. ANACONDA is publicly available at http://bioinformatics.ua.pt/submited-papers). Briefly, this methodology counts all consecutive pairs of codons and uses statistical analysis for contingency tables where a multinomial distribution is assumed (). The final result of such statistical approach is the calculation of adjusted residuals for each codon pair present in any ORFeome. The adjusted residuals give direct information about preference or rejection of these codon pairs in relation to what would be expected assuming independence of the distribution (). Since, under independence between two consecutive codons, the adjusted residuals dij have a standardized normal probability distribution [52], we have concluded that:  , as the total number of observations is very high. This means that, for a 99,73% confidence level, an adjusted residue was statistically significant if its absolute value was greater than 3 [20]. However, this approach was based on a local analysis for each residual value. Herein, we considered a global analysis for each species and have thus constructed a simultaneous confidence region for all residual values. Since there are K  = 61×64 different intervals we have introduced the Bonferroni correction to ensure an overall level of significance of α (usually α = 0.05, 0.01, 0.001). The Bonferroni correction is used for correction where each interval is constructed at a 100×(1–α/K) level (see, for example, [53]). Therefore, a– a to a interval at a confidence level of 100×(1−α/(61×64)) was constructed for each adjusted residual value dij. Considering again the asymptotic normal distribution of dij [52] we had a≈4,70341 when 1–α = 0,99, a≈5,15350 when 1–α = 0,999, a≈8,16204 when 1–α = 0.01×10 −10. Thus, we assumed that the codon-pair adjusted residuals that fall within the interval −5 to 5 were not statistically significant, for a global confidence level of 99% (colored in black in all maps shown herein). The final output of residual analysis performed by ANACONDA is a codon-pair context map for each ORFeome being studied (). These maps show one colored square for each codon-pair, the first codon corresponding to rows and the second corresponding to columns in the map. The color scale chosen for this layout determines that preferred contexts are shown in green while repressed ones appear in red (). Taking advantage of the automated statistical analysis performed by ANACONDA, individual maps for all 119 ORFeomes were built (see Figure S3). In order to facilitate large-scale comparison of maps these were converted into single lines and clustered together ( and ). The patterns that appear in the resulting comparative map were then characterized by the codon-pair contexts that were present in each pattern. Also, the values of the adjusted residuals calculated for each species were corrected for ORFeome size to allow direct comparisons among ORFeomes. The above approach was also used to study total genome sequences of the same 119 species in order to differentiate between the effect of translational selection acting upon coding sequences alone and genome mutational biases. With the same purpose, the bias for dinucleotides was studied in total genome sequences, and shown as observed frequencies, colored in green or red whenever the result was 1% above or below the expected value, respectively (). Figure S1 Data normalization. In order to correct the size differences of ORFeomes, particularly between eukaryotes and non-eukaryotes, the adjusted residuals were normalized for 21 million codons which correspond approximately to the larger ORFeome analyzed (X. tropicalis). Normalization of codon-pair data for human chromosomes 1, 2, 3, 22 and ORFeome are displayed. The normalization effect is shown by the brightness of the maps, which is variable in non-normalized maps (above) and constant in normalized ones (below). After data normalization the differences between maps could be compared as shown in the DDM (right end of the Figure). (4.32 MB TIF) Figure S2A List of species used. All species used in the study are listed according to the download order. The database of origin and respective accession numbers are indicated. A - Eukaryotes; B - Archaea and Eubacteria; C - Eubacteria (cont.). (0.54 MB TIF) Figure S3A Individual codon-pair context maps of the 119 species. The codon-pair context maps built with ANACONDA software for individual ORFeomes are shown as ordered in Suppl. Figure S2. (4.32 MB TIF) Figure S4 A and U bases are preferentially arranged in polynucleotide strings. In order to check if the preference detected for AA and UU dinucleotides in total genomes () was due to a tendency for these bases to appear as polynucleotide strings we counted the number of times each individual base appeared isolated or in strings of two, three or more equal bases. The result of this approach points to a clear positive bias towards the accumulation of 3 or more consecutive A or U bases in total genomes. (0.10 MB TIF) Figure S5A Codon-pair context patterns that are exclusive of ORFeomes or genomes. The filtering technique that was used to determine the biases of codon-pair contexts in coding and total sequences () was further explored in here to evaluate the strength of those biases. This was done by gradually increasing the threshold of the residuals (D) that are significantly different in both maps, i.e. those that were allowed to appear in their original colors in the filtered map. When D was increased, only major differences between ORFeomes and genomes were visible, corresponding to differences between the residuals of ORFeomes and genomes maps that stay above 15, 20, 30 or 50. The strongest rules detected by this approach correspond to those identified as B1 and B2 in the filtered map for ORFeomes (map A in ), and F1, F2 and E2 in the filtered map for genomes (map B in ), because they are still visible when D = 50. (0.55 MB TIF) Table S1 Codon-pair distribution similarities between the 3 domains of life. In order to compare the overall distribution of codon-pair contexts among the 119 organisms we have calculated the Spearman's correlation coefficients between all pairs of ORFeomes, producing a triangular colored map. The 119 species were organized by domain of life and sorted alphabetically in each domain. Pairs of species that were not statistically correlated (for a level of significance of 5%) are colored in grey, while green colored cells indicate pairs of species that were highly correlated (correlation coefficient above 0,80), and blue colored cells correspond to the major values fount inside each domain. (0.32 MB XLS) 1. Cliften PF, Fulton RS, Wilson RK, Johnston M. After the duplication: gene loss and adaptation in Saccharomyces genomes. Genetics. 2006;172:863–872. [PubMed] 2. van de Lagemaat LN, Gagnier L, Medstrand P, Mager DL. Genomic deletions and precise removal of transposable elements mediated by short identical DNA segments in primates. Genome Res. 2005;15:1243–1249. [PubMed] 3. Lin YW, Thi DA, Kuo PL, Hsu CC, Huang BD, et al. Polymorphisms associated with the DAZ genes on the human Y chromosome. Genomics. 2005;86:431–438. [PubMed] 4. Chen SL, Lee W, Hottes AK, Shapiro L, McAdams HH. Codon usage between genomes is constrained by genome-wide mutational processes. Proc Natl Acad Sci U S A. 2004;101:3480–3485. [PubMed] 5. Chan SW, Henderson IR, Jacobsen SE. Gardening the genome: DNA methylation in Arabidopsis thaliana. Nat Rev Genet. 2005;6:351–360. [PubMed] 6. Robertson KD. DNA methylation and human disease. Nat Rev Genet. 2005;6:597–610. [PubMed] 7. Duan J, Antezana MA. Mammalian mutation pressure, synonymous codon choice, and mRNA degradation. J Mol Evol. 2003;57:694–701. [PubMed] 8. Berg OG, Silva PJ. Codon bias in Escherichia coli: the influence of codon context on mutation and selection. Nucleic Acids Res. 1997;25:1397–1404. [PubMed] 9. Akashi H. Synonymous codon usage in Drosophila melanogaster: natural selection and translational accuracy. Genetics. 1994;136:927–935. [PubMed] 10. Curran JF, Yarus M. Rates of aminoacyl-tRNA selection at 29 sense codons in vivo. J Mol Biol. 1989;209:65–77. [PubMed] 11. Percudani R, Ottonello S. Selection at the wobble position of codons read by the same tRNA in Saccharomyces cerevisiae. Mol Biol Evol. 1999;16:1752–1762. [PubMed] 12. Boycheva S, Chkodrov G, Ivanov I. Codon pairs in the genome of Escherichia coli. Bioinformatics. 2003;19:987–998. [PubMed] 13. Ogle JM, Ramakrishnan V. Structural insights into translational fidelity. Annu Rev Biochem. 2005;74:129–177. [PubMed] 14. Irwin B, Heck JD, Hatfield GW. Codon pair utilization biases influence translational elongation step times. J Biol Chem. 1995;270:22801–22806. [PubMed] 15. Shah AA, Giddings MC, Gesteland RF, Atkins JF, Ivanov IP. Computational identification of putative programmed translational frameshift sites. Bioinformatics. 2002;18:1046–1053. [PubMed] 16. Buckingham RH, Grosjean H. The accuracy of mRNA-tRNA recognition. In: Kirkwood TBL, Rosenberger RF, Galas DJ, editors. Accuracy in Molecular Processes: Its Control and Relevance to Living Systems. London: Chapman and Hall; 1986. pp. 83–126. 17. Percudani R, Pavesi A, Ottonello S. Transfer RNA gene redundancy and translational selection in Saccharomyces cerevisiae. J Mol Biol. 1997;268:322–330. [PubMed] 18. Curran JF, Poole ES, Tate WP, Gross BL. Selection of aminoacyl-tRNAs at sense codons: the size of the tRNA variable loop determines whether the immediate 3′ nucleotide to the codon has a context effect. Nucleic Acids Res. 1995;23:4104–4108. [PubMed] 19. Fedorov A, Saxonov S, Gilbert W. Regularities of context-dependent codon bias in eukaryotic genes. Nucleic Acids Res. 2002;30:1192–1197. [PubMed] 20. Moura G, Pinheiro M, Silva R, Miranda I, Afreixo V, et al. Comparative context analysis of codon pairs on an ORFeome scale. Genome Biol. 2005;6:R28. [PubMed] 21. Tate WP, Poole ES, Mannering SA. Hidden infidelities of the translational stop signal. Prog Nucleic Acid Res Mol Biol. 1996;52:293–335. [PubMed] 22. Murgola EJ, Pagel FT, Hijazi KA. Codon context effects in missense suppression. J Mol Biol. 1984;175:19–27. [PubMed] 23. Tork S, Hatin I, Rousset JP, Fabret C. The major 5′ determinant in stop codon read-through involves two adjacent adenines. Nucleic Acids Res. 2004;32:415–421. [PubMed] 24. Rodnina MV, Wintermeyer W. Fidelity of aminoacyl-tRNA selection on the ribosome: kinetic and structural mechanisms. Annu Rev Biochem. 2001;70:415–435. [PubMed] 25. Kramer EB, Farabaugh PJ. The frequency of translational misreading errors in E. coli is largely determined by tRNA competition. RNA. 2007;13:87–96. [PubMed] 26. Atkins JF, Weiss RB, Thompson S, Gesteland RF. Towards a genetic dissection of the basis of triplet decoding, and its natural subversion: programmed reading frame shifts and hops. Annu Rev Genet. 1991;25:201–228. [PubMed] 27. Freistroffer DV, Kwiatkowski M, Buckingham RH, Ehrenberg M. The accuracy of codon recognition by polypeptide release factors. Proc Natl Acad Sci U S A. 2000;97:2046–2051. [PubMed] 28. Princiotta MF, Finzi D, Qian SB, Gibbs J, Schuchmann S, et al. Quantitating protein synthesis, degradation, and endogenous antigen processing. Immunity. 2003;18:343–354. [PubMed] 29. Pinheiro M, Afreixo V, Moura G, Freitas A, Santos MA, et al. Statistical, computational and visualization methodologies to unveil gene primary structure features. Methods Inf Med. 2006;45:163–168. [PubMed] 30. Campbell A, Mrazek J, Karlin S. Genome signature comparisons among prokaryote, plasmid, and mitochondrial DNA. Proc Natl Acad Sci U S A. 1999;96:9184–9189. [PubMed] 31. Nakashima H, Nishikawa K, Ooi T. Differences in dinucleotide frequencies of human, yeast, and Escherichia coli genes. DNA Res. 1997;4:185–192. [PubMed] 32. Hooper SD, Berg OG. Detection of genes with atypical nucleotide sequence in microbial genomes. J Mol Evol. 2002;54:365–375. [PubMed] 33. Hooper SD, Berg OG. Gradients in nucleotide and codon usage along Escherichia coli genes. Nucleic Acids Res. 2000;28:3517–3523. [PubMed] 34. Stahl G, McCarty GP, Farabaugh PJ. Ribosome structure: revisiting the connection between translational accuracy and unconventional decoding. Trends Biochem Sci. 2002;27:178–183. [PubMed] 35. Dix DB, Thompson RC. Codon choice and gene expression: synonymous codons differ in translational accuracy. Proc Natl Acad Sci U S A. 1989;86:6888–6892. [PubMed] 36. Gutman GA, Hatfield GW. Nonrandom utilization of codon pairs in Escherichia coli. Proc Natl Acad Sci U S A. 1989;86:3699–3703. [PubMed] 37. Buchan JR, Aucott LS, Stansfield I. tRNA properties help shape codon pair preferences in open reading frames. Nucleic Acids Res. 2006;34:1015–1027. [PubMed] 38. Rocha EP, Danchin A, Viari A. Universal replication biases in bacteria. Mol Microbiol. 1999;32:11–16. [PubMed] 39. Grantham R, Gautier C, Gouy M, Mercier R, Pave A. Codon catalog usage and the genome hypothesis. Nucleic Acids Res. 1980;8:r49–r62. [PubMed] 40. Buckingham RH. Codon context. Experientia. 1990;46:1126–1133. [PubMed] 41. McVean GAT, Hurst GDD. Evolutionary lability of context-dependent codon bias in bacteria. J Mol Evol. 2000;50:264–275. [PubMed] 42. Lao PJ, Forsdyke DR. Thermophilic bacteria strictly obey Szybalski's transcription direction rule and politely purine-load RNAs with both adenine and guanine. Genome Res. 2000;10:228–236. [PubMed] 43. Kennedy SP, Ng WV, Salzberg SL, Hood L, DasSarma S. Understanding the adaptation of Halobacterium species NRC-1 to its extreme environment through computational analysis of its genome sequence. Genome Res. 2001;11:1641–1650. [PubMed] 44. Tekaia F, Yeramian E, Dujon B. Amino acid composition of genomes, lifestyles of organisms, and evolutionary trends: a global picture with correspondence analysis. Gene. 2002;297:51–60. [PubMed] 45. Beutler E, Gelbart T, Han JH, Koziol JA, Beutler B. Evolution of the genome and the genetic code: selection at the dinucleotide level by methylation and polyribonucleotide cleavage. Proc Natl Acad Sci U S A. 1989;86:192–196. [PubMed] 46. Nakashima H, Ota M, Nishikawa K, Ooi T. Genes from nine genomes are separated into their organisms in the dinucleotide composition space. DNA Res. 1998;5:251–259. [PubMed] 47. Marck C, Grosjean H. tRNomics: analysis of tRNA genes from 50 genomes of Eukarya, Archaea, and Bacteria reveals anticodon-sparing strategies and domain-specific features. RNA. 2002;8:1189–1232. [PubMed] 48. Crick FH. Codon–anticodon pairing: the wobble hypothesis. J Mol Biol. 1966;19:548–555. [PubMed] 49. Caburet S, Vaiman D, Veitia RA. A genomic basis for the evolution of vertebrate transcription factors containing amino Acid runs. Genetics. 2004;167:1813–1820. [PubMed] 50. Borstnik B, Pumpernik D. Tandem repeats in protein coding regions of primate genes. Genome Res. 2002;12:909–915. [PubMed] 51. Rocha EP, Matic I, Taddei F. Over-representation of repeats in stress response genes: a strategy to increase versatility under stressful conditions? Nucleic Acids Res. 2002;30:1886–1894. [PubMed] 52. Haberman S. Analysis of residuals in cross-classified tables. Biometrics. 1973;29:205–220. 53. Simenoff JS. New York: Springer-Verlag; 2003. Analyzing categorical data.
|
This article has been 
