DNA replication and RNA transcription share many properties (Little et al., 1993; Hassan & Cook, 1994; Marczynski and Shapiro 1995; Mohanty et al., 1996; Prado & Aguilera 2005), notably in mitochondria (Nass 1995; Lee & Clayton, 1997). The joint occurrence of transcription and replication on DNA apparently necessitates coordination (Gilbert, 2001; MacAlpine at al., 2004), among others because collisions occur between the replication and transcription complexes on the same DNA strand (Mirkin & Mirkin, 2005). This coordination may be part of the regulation of gene expression (Patnaik, 1997) and the rates of both processes (Morton, 1999). This predicts the structural organization of genes on chromosomes around replication origins in relation to functional pressures (Schwaiger & Schubeler, 2006): highly expressed genes are located close to replication origins, those expressed in few tissues are more distant (Huvet et al., 2007). Such functional pressures seem strong enough to cause convergences in genome organization between very distant organisms such as yeast (
1.1. Replication and transcription in mitochondria
Both mitochondrial rRNA genes, the only transcription genes coded by vertebrate mitochondrial genomes, are located next to the mitochondrial control region. Indeed, the mitochondrial D-loop, according to the frequently observed collinearity between replication and transcription in prokaryotes (Nikolaou & Almirantis, 2005), includes the regions that initiate replication of the heavy strand (OH) and transcription (Shadel & Clayton, 1997; Fernandez-Silva et al., 2003), fitting the model that all transcription enhancers also enhance replication and vice versa (Boulikas, 1995). The major regulatory sequences in the D-Loop contain the only mitochondrial promoters, for both light- and heavy-strand transcription, in addition to the origin of heavy strand replication (Chang and Clayton, 1984). Biochemical evidence indicates that the primers generated by the light-strand promoter are used for replication priming (Chang & Clayton, 1985; Chang et al., 1985). This close association between the two processes in mitochondria expresses itself by the fact that some Alzheimer-associated mutations in the mitochondrial control region suppress both mitochondrial replication and transcription (Coskun et al., 2004). Hence, mitochondria, because of their involvement in ageing (Martin & Grotewiel 2006; Yu & Chung 2006), are a good system in which to study the putative relationship existing between the connection (and level of connection) of replication and transcription with ageing processes.
1.2. Mitochondria and ageing
Several properties of mitochondrial genomes have already been shown to associate with lifespan, presumably because of cumulative DNA damage due to free radicals produced by the normal function of mitochondria in the cell’s energetic metabolism (Wiesner et al., 2006). It seems that evolution of vertebrate longevities (and in general, the associated evolution of their life history strategies) causes accelerated rates of amino acid replacements in mitochondrion-encoded protein coding genes (Rottenberg, 2006, 2007). Mutagenesis independent of free oxidative radicals also affects mitochondrion-associated ageing: in mitochondrial genomes, the number of direct repeats, a factor causing deletions, correlates negatively with mammalian lifespan (Samuels, 2004; Samuels et al., 2004; Khaidakov et al., 2006). Using a similar comparative approach, Samuels (2005) showed that lifespan increases proportionally to the stability of hybridization between complementary mitochondrial DNA strands, estimated by free energies. This property is proportional to the probability of opening up and expansion of single-stranded mtDNA bubbles. Negative selection on repeats (Samuels, 2005; Khaidakov et al., 2006) explains the exceptional status of most vertebrate mitochondrial genomes as breaking Chargaff’s second parity rule (complementary nucleotides are met with almost equal frequencies in single stranded DNA), fitting the hypothesis that inversions and inverted transposition could be a major contributing if not dominant factor in the almost universal validity of this rule (Albrecht-Buehler, 2006).
These single stranded bubbles are more likely to occur where direct repeats exist, causing deletions (Khaidakov et al., 2006), but this probably also increases mutation rates in the single stranded sequences forming the bubble when the bubble is not enzymatically excised. This is because hydrolytic deaminations of cytosine to thymine and adenine to guanine, both transitions, occur proportionally to time spent single stranded (Dssh) by the genome (as shown for example during replication in primate mitochondrial genomes, Krishnan et al., 2004a, b).
1.3. Developmental stabilities
Developmental stabilities, typically estimated by the level of symmetry of bilateral traits (Seligmann, 2003) usually increase with correlates of fitness (Moller, 1997, 1999). Hence it is sensible that rationales similar to those for longevity and ageing apply to developmental stabilities. In mitochondria, the chemical stability of rRNA increases developmental stability (Seligmann, 2006a), and the structural stability of the regular light strand origin of replication (OL) also increases developmental stability. So does usage of tRNAs adjacent to the regular OL as additional OLs (Seligmann and Krishnan 2006). Densities of off frame stops in mitochondrial genomes also increase developmental stability, probably because off frame stops stop early translation after unprogrammed ribosomal frameshifts, which produce dysfunctional proteins (Seligmann, 2010a). It hence makes sense to expect that connections between mitochondrial transcription and replication could affect developmental stability as well as lifespan.
1.4. Mutation gradients across genomes
Deamination gradients exist along mitochondrial genomes proportionally to Dssh during replication (Dsshr). In several independent taxa, inversion of the mitochondrial control region inverts the directions of these gradients (Hassanin et al., 2005).
Dssh during transcription (Dssht) also creates deamination gradients along genomes or sections of genomes. The relative dominance of replication- versus transcription-associated gradients varies among genome regions and organisms (Francino et al., 1996; Baran et al., 2003), especially in bacteria (Mackiewicz et al., 1999). It seems that in nuclear vertebrate genomes, transcription-associated gradients mask the replication-associated ones (Hou et al., 2006), and more complex analyses are required to detect them (Touchon et al., 2005). These nuclear vertebrate transcription-associated gradients even reveal genes that are otherwise undetected because of their long introns and low sequence conservation (Glusman et al., 2006). Replication-associated deamination gradients are usually present in bacteria (Mrazek & Karlin, 1998). Transcription-associated ones have also been detected in bacteria (Francino & Ochman, 2001). In an adenovirus, gradients in nucleotide composition asymmetries between strands exist where replication and transcription share directions, but are basically nonexistent where they have opposite directions (Grigoriev, 1999).
In the latter case, this lack of a clear gradient does not signify that no mutations occur during replication and transcription, but rather that maximal and minimal mutation rates of each process coexist in the same region, increasing the range of the genome that is affected by high mutation regimes from at least one of the two processes. This might be adaptive in viruses, by maximizing evolvability (see for example Aldana et al., 2007; Feder, 2007; Jones et al., 2007). But, it is no surprise that both processes are frequently collinear in other organisms, as this limits maximal mutation rates to a specific region that might have relatively high mutational robustness, decreasing mutational constraints on the rest of the genome (Ciliberti et al., 2007; Elena et al., 2007; Wagner & Wright, 2007). It also makes sense that replication and transcription gradients, reflecting the relative frequencies of each process, are approximately balanced in prokaryotes. Hence the tendency for dominance of transcription-related gradients in nuclear genomes of eukaryotes would result from their generally lower cell replication rates. According to this, their mitochondria should resemble bacteria.
1.5. Convergence of replication and transcription in mitochondria
The level of collinearity, also termed here convergence, between replication and transcription is expected to slow ageing and ageing-related processes. Vertebrate mitochondrial genomes seem good candidates for testing this hypothesis, because a) data are available for many species, b) they affect ageing and c) mitochondrial replication and transcription are at least partially collinear.
Indeed, in vertebrate mitochondria, the distance from the D-loop determines Dssht, while Dsshr results from calculating relative distances from the OH, also in the D-loop, and the light strand replication origin (OL, see Seligmann et al., 2006b for details on Dssh calculations, and Seligmann, 2006). Usually, one considers that mitochondrial genomes have a single OL located in the WANCY region, a cluster of 5 tRNA genes (Desjardins & Morais, 1990; Clayton, 2000), resulting in DsshrW. Both processes are only partially collinear when solely the WANCY region functions as OL, but the probabilistic combination of multiple tRNA clusters distributed across the genome that putatively act as OLs (Seligmann et al., 2006b; Seligmann, 2006; Seligmann, 2010b) can result in an overall replication gradient (DsshrX) collinear with the transcription gradient (Dssht). In Figure 1, DsshrX (as it is expected after integrating with equal weights all putative tRNA clusters as OLs into Dsshr calculations) is highly correlated with the distance from the Dloop. As compared to DsshrW, this DsshrX has only one region with high mutation risks (this region codes for ND6 and CytB), while for DsshrW, there is an additional region (coding for ND1 and ND2), ranging
over 4 of all 13 mitochondrially encoded protein coding genes and both rRNAs. Hence, one could expect that evolution of multiple OLs in mitochondrial genomes, especially in taxa with long lifespan, would tend towards increasing collinearity of DsshrX with Dssht, reducing the extent of DNA regions with high mutation risks. Multiple OLs would regulate Dsshr->Dssht convergence. These interactions between mitochondrial replication and transcription would be an additional process interacting with mitochondrial transcription (Bonawitz et al., 2006).
1.6. Alternative replication mechanisms
This is in line with studies suggesting that multiple OLs exist in vertebrate mitochondria (Brown et al., 2005; Brown & Clayton 2006; Clayton & Brown 2006). The hypothesis that mitochondrial light strands are replicated at multiple locations by Okazaki fragments (Holt et al., 2000) as the lagged strand in nuclear genomes is also compatible with the statistical patterns observed by Seligmann et al., (2006b). In fact, the deamination gradients detected by comparative analyses are considered as strong evidence in favor of the unidirectional replication mechanisms (Gibson, 2005). My interpretation is that the unidirectional replication is relatively rare, but it leaves at evolutionary scales a clear imprint on genomes because it causes biases in mutation patterns, and that at least one other replication process, putatively similar to the one in nuclei, exists. That process is more frequent and effective, affecting less the genome at evolutionary scale. Indeed, some evidence on mitochondrial transcription factors suggests that two replication modes coexist, and that the modes of mitochondrial replication are regulated by mitochondrial metabolism (Pohjoismaki et al., 2006). Results and conclusions will be also interpreted according to this hypothesis, considering that only one replication mechanism, the unidirectional one, creates replication deamination gradients.
1.7. Lifespan and convergence of replication and transcription
Heavy strand sequences of mitochondrial tRNA genes tend to form OL-like structures and seem to assist the “recognized” vertebrate mitochondrial OL in the WANCY region (Seligmann & Krishnan, 2006). DsshrX resembles Dssht more than does DsshrW (see Figure 1, and Seligmann et al., 2006b). Pathogenic mutations, as compared to non-pathogenic polymorphisms in human mitochondrial tRNAs, disturb the fine balance of DsshrX by altering which tRNAs function and which do not function as alternative OLs (Seligmann et al., 2006b). These observations strengthen the hypothesis that collinearity between these processes increases longevity by slowing ageing.
A further observation in line with this hypothesis is that nucleotide contents of heavy strand DNA sequences coding for the first and second positions of tRNA anticodons in vertebrate mitochondrial genomes correlate with Dssht calculated according to the highly conserved tRNA arrangement along the vertebrate mitochondrial genome. When mitochondrial replication and transcription are collinear, as observed in
2. Materials and methods
In order to test this ageing-related collinearity hypothesis, I calculated rt, the correlation coefficient between C or T contents at third codon positions for all 13 protein coding genes (methodology as in Seligmann et al., 2006b) and Dssht; and rW, the correlation coefficient between C or T contents as above, and DsshrW. I used the light strand sequences, coding Cs by "1" and Ts by zero, so that the gradient reflects the slower deamination reaction of A to G that occurs during replication on the heavy strand as a function of Dssh. I did also similar calculations for the other gradient, coding light strand As by “1” and Gs by zero, reflecting the heavy strand gradient due to the faster deamination of C to T. Results were generally qualitatively similar for this gradient, but are not presented here.
I used for Dssh calculations the numbering system of Genbank for nucleotide sites, also used by Tanaka and Ozawa (1994) and Seligmann et al., (2006b). Dssht is the relative distance of the base pair from the starting point of the transcription, meaning the number assigned to that base pair following that numbering system divided by the total length of the genome:
where b is the distance in bases of the nucleotide position from the genome numbering starting point and N is the total mitochondrial genome length.
I calculated DsshrW of ND1 and ND2 genes using the equation:
where W is the position at mid-location of the sequence forming the classical light strand replication origin. In species lacking the classical origin, I used for W the mid-location of the sequence located between the two tRNAs that normally flank the regular light strand replication origin, tRNA-Asn and tRNA-Cys. For other genes, I calculated DsshrW according to the equation:
Note that visual examinations of gradients in single species, such as those shown in Figure 2, are based on gene-wise averages of the binary C and T contents of that gene at third codon position. Analyses based on such averages would probably yield qualitatively similar results. I opt for the method using site specific nucleotide contents, without averaging over genes in order to maximize the amount of information used from the raw sequence data. It is possible that reducing data by averaging following the natural units of protein coding sequences might reveal additional phenomena or aspects of the main phenomenon examined, a point that should be kept in mind. The gene-wise averaging method has clear advantages for graphical presentation and is therefore used here in various Figures.
Dsshr calculations for tRNA clusters different from the WANCY region containing the classical light strand replication origin are done following Seligmann et al., (2006b). These calculations were not used besides for Figure 1, for other analyses presented here, only DsshrW and Dsshrt were used. Dsshr calculations for the various tRNA clusters as light strand replication origins, as they were used to estimate DsshrX in Figure 1, are calculated following the principles described by Seligmann (2006).
This was done for a number of mammalian and reptilian taxa, notably for 26 complete Primate mitochondria, and two outgroups,
I also tested for correlations of rt-rW, termed collinearity between gradients or Dsshr->Dssht convergence, with the length of the gestation period. Information on gestation periods is also from
3. Results and discussion
3.1. Replication versus transcription gradients in various species
Examining graphs plotting the mean C/T ratio at third codon position for each gene as a function of Dssht and as a function of DsshrW, one finds that for a majority of species, DsshrW is the better predictor of nucleotide contents at third codon positions and there is no evidence for a gradient resembling the one that would be expected due to transcription, whether due to replication convergent with transcription or transcription itself (for example the greater white-toed shrew
of nucleotide contents with Dssht is better than with DsshrW, indicating that in that species, processes causing deamination gradients due to the time spent single stranded tend to start usually within the Dloop. This could either indicate that in that species the frequency of transcriptions is overwhelmingly larger than of replications, or that replication does relatively rarely start in the region of the regular OL. The latter option is in line with the fact that there is no recognized OL sequence in that lizard species at the regular OL location, between tRNA-Asn and tRNA-Cys, and that in this species, unlike in other lizards lacking a recognized OL sequence between these tRNAs, the adjacent 5’ arm of tRNA-Asn and 3’ arm of tRNA-Cys, including the short intergenic sequence, do not form OL-like structures as those found in
These observations suggest that variation exists among species in the extent that Dsshr converges with Dssht, and that this variation might associate with life history: in the examples presented, regular replication gradients starting at the recognized OL sequence are observed in short lived species with high metabolic rates (shrew), while the convergence between replication and transcription increases for more long lived species with lower metabolisms (gorilla, dolphin, lizard), paralleling the dichotomy noted above for gradients between prokaryotes (where patterns remind more those found in mitochondria of short lived mammals) and eukaryotes (resembling more those found in mitochondria of long lived animals with slower metabolisms). This justifies testing whether the extent of Dsshr->Dssht convergence correlates with lifespan and other ageing-related processes.
3.2. Gradient convergence and lifespan in Primates
In Primates, the strength of the replication gradient that considers only the recognized OL (rW) does not correlate with maximal lifespan (r = 0.11, P = 0.29, one tailed test, not shown); the strength of the transcription gradient (rt) increases with maximal lifespan (r = 0.318, P = 0.049, one tailed test, not shown). This improvement in the correlation with lifespan fits the prediction that the actual replication gradient, calculated having considered all putative OLs and not only the one in the WANCY region, is to some extent collinear with the transcription gradient, and hence the strength of the transcription gradient, rt, is a better estimate of the strength of the replication gradient than rW. In this case, and as expected by the working hypothesis, the extent by which rt is stronger than rW would measure the extent by which DsshrX resembles Dssht.
I quantified this extent by calculating the residuals of rt for each Primate species from the regression between rt (dependent) and rW (independent) (rt = 0.822*rW+0.04, r = 0.83, P < 0.001). These residuals are unlikely, from a statistical point of view, to correlate with lifespan because rt correlates with lifespan: they represent only a small fraction of the variation inherent to rt because rW explains 69% of the variation in rt. Nevertheless, results show that they correlate better than rt with maximal lifespan (r = 0.405, P = 0.016, 1 tailed test; see Figure 3), indicating that the extent of Dsshr->Dssht convergence affects lifespan. Analyses reveal similar patterns in other groups, such as Carnivora (Figure 4, analyses excluding Pinnipedia). In these cases, no residual analyses were done, and rW was simply subtracted from rt. The correlation is positive as expected for a pool of groups excluding Mustelidae and other closely related groups. Patterns in Mustelidae closely resemble those for other Carnivora, besides for an outlier, the Eastern spotted skunk (
3.3. Convergence of replication towards transcription
Analyses between the various life history traits and gradient strengths presented and discussed in the rest of this study did not detect any significant correlation with rW, while those with rt were systematically stronger and sometimes statistically significant. This is despite the strong mathematically trivial correlation existing between rt and rW, which is also apparent from Figure 1. But the strongest correlations were systematically with rt-rW, confirming that the factor that is most relevant to life history is the extent of convergence of replication towards transcription, rather than the extent of the transcription-like replication gradient. This is the main point of the hypothesis presented here.
3.4. Too extreme convergence between replication and transcription decreases lifespan
Closer examinations of Figures 3 and 4 reveal that for species with relatively high (or even extreme) convergence between gradients (rt>rW:
This suggests that at high Dsshr->Dssht convergence levels, another factor decreases lifespan. It is plausible that collinearity between the processes increases the frequency of collisions between replication and transcription forks. This decreases the respective rates of these processes, increasing the overall time spent single stranded, causing more mutations. This increase might be greater than the relative decrease in mutation rate due to collinearity between the processes, especially at high collinearity levels. Figure 5 plots lifespan in
Cetacea as a function of Dsshr->Dssht convergence. At low convergence levels, lifespan increases with convergence until a threshold region in Dsshr->Dssht convergence. Beyond that threshold, lifespan decreases with Dsshr->Dssht convergence. It is hence not a surprise to find a negative correlation between Dsshr->Dssht convergence and maximal lifespans in lizards (Figure 6). Hence the few outliers found in Figures 3 and 4a would reflect the same phenomenon as the one observed for a larger part of species in Cetacea (those for which a negative correlation of lifespan with convergence for high convergence levels exists) or for lizards (Figure 6).
3.5. Developmental stability and convergence between transcription and replication
Analyses testing for correlations between Dsshr->Dssht convergence and developmental stabilities yield qualitatively similar results to those found for associations with maximal lifespan: in some groups, convergence decreases stability (in a pool of lizards from several families, r = -0.52, Figure 7), and in others, convergence decreases instability (Amphisbaenia, r = - 0.76, Figure 8).
3.6. Rates of development and convergence between replication and transcription
As noted above, convergence between replication and transcription increases the frequency of collisions between these processes, hence decreasing their respective rates. Ultimately, decreased replication and transcription rates should impede on an organism’s development, decreasing its differentiation and growth rates. I used the length of the gestation period as an estimate inversely proportional to differentiation rate and tested for the expected positive correlation between gestation period and Dsshr->Dssht convergence levels (see the example for Insectivora in Figure 9). Because maximal lifespan, together with brain size, correlates positively with the length of the gestation period (Sacher & Staffeld, 1974; Jones & MacLarnon, 2004), this result does not independently confirm the Dsshr->Dssht convergence hypothesis, despite that the mechanisms assumed to cause the correlations with lifespan and those with gestation length differ: lifespan is presumed to increase because convergence increases mutational robustness (only extreme convergence decreases mutational robustness and lifespan); at the same time, convergence decreases the rates of replication and transcription,, and presumably also developmental rates.
However, the rationale that Dsshr->Dssht convergence affects both lifespan and gestation yields a prediction that is not trivial, despite the strong positive correlation that exists between lifespan and the length of the gestation period: in groups of species with high fertility and rates of development (short gestation), considered as r-strategists, one expects that Dsshr->Dssht convergence adaptively coevolved with the length of gestation, while in groups of species with low fertility and rates of development (long gestation and lifespan), considered as K-strategists, it makes sense to expect adaptive coevolution between Dsshr->Dssht convergence and lifespan. Hence, despite that lifespan and the length of gestation are highly correlated, a testable, independent, nontrivial prediction exists, which is that correlations between Dsshr->Dssht convergence and lifespan should be weaker in r strategists than those between Dsshr->Dssht convergence and the length of gestation, while in K strategists, the opposite is expected. This is estimated by subtracting the z transformed correlation coefficient between Dsshr->Dssht convergence and the length of gestation from the z transformed correlation coefficient between Dsshr->Dssht convergence and lifespan in that group (z transformation was adjusted for differences in sample sizes between different
taxonomic groups, see method in Seligmann et al., 2006). Figure 10 tests this prediction by plotting this subtraction as a function of the mean maximal lifespan for that taxonomic group, used here as an estimate of the extent that the group is a relatively r- or K-strategist (short and long maximal lifespans, respectively). Results in Figure 10 fit the expectation that correlations with lifespan, relative to those with the length of the gestation period, increase along the r-K gradient. The increase in the subtraction is approximately gradual along the r-K gradient (which is estimated by the mean maximal lifespan in that group). According to this result, patterns from more than 100 mitochondrial genomes follow the complex predictions from a simple hypothesis.
4. General discussion
4.1. Replication versus transcription gradients in various species
Results show that species vary widely in extents of convergence between replication and transcription gradients. In many species, the replication gradient starting at the recognized OL is the only or the major gradient detected, as found for
with replication origins distributed such that the overall replicational gradient (integrating over different replication origins) resembles the one caused by transcription. It is possible that because transcription is much more frequent than replication, the reactions that create the gradients are saturated, and hence no gradient is detected at the time scales the phenomenon is observed here. Hence while the comparative methodology used seems adequate to detect replication gradients, other methods should be used in order to detect transcription gradients. This means that at this evolutionary time scale, replication is the main phenomenon, and transcription is probably a secondary phenomenon, whose detection necessitates more sophisticated methods, as explained for some cases from the literature in the Introduction.
The correlation between convergence of replication gradient with transcription gradient with lifespan was r = 0.80 in Chiroptera (5 species), but this group is not included in Figure 10 because only for 2 Chiroptera both genomic and gestation period data were available. Results presented for lizards (Figure 9) are not included because there is no gestation
4.2. Transcription or replication?
For the sake of simplicity, I consider that the major cause for the observed gradients in nucleotide contents with Dssht is replication, meaning that in these cases replication converged with transcription, but that transcription itself is not responsible for the observed gradients. Although this is at this point a rough simplification, there are several reasons beyond those given above justifying this assumption. It makes sense that the polymerization rates by the gamma polymerase, the enzyme replicating mitochondrial DNA, and by the mitochondrial RNA polymerase (Bonawitz et al., 2006) differ, because these are very different enzymes and the functional requirements differ for each process: the frequency of transcription is much greater than that of replication, and its rate is also probably much greater. However, the impact of errors during RNA polymerization is lower than that during DNA replication and hence RNA and DNA polymerase fidelities are also probably very different. Deamination gradients result from time spent single stranded during these processes, but because one can assume that transcription is much faster than replication, it is likely that the properties of the mutation gradient resulting from transcription differ from those of the replication gradient. Hence effects of one Dssh unit on nucleotide contents should differ between gradients caused by transcription or replication. Examining the various graphs in Figure 2, one can see that this is not the case: the slopes found for gradients with DsshrW and Dssht are very similar when gradients are detected with each DsshrW and Dssht (see for example the western gorilla, Figure 2b). This justifies the simplifying assumption that replication is the major cause of the observed gradients, and this approach should be considered as a satisfying approximation at this point.
This does not mean that this assumption should not be tested later, especially that exploring this issue might yield valuable information on the relative regulations of transcription, replication, and/or various types of replication, which are at the heart of the mitochondrial replication controversy and ageing-related pathologies. Note that even at that level of distinguishing between deamination gradients caused by transcription and those caused by replication in a situation where both are confounded because replication is collinear with transcription, bioinformatics analyses can be helpful. Two deamination gradients exist on the heavy strand, one caused by the chemical reaction C->T, and one by A->G (both hydrolytic deaminations). The former is the faster reaction, and therefore the latter saturates less quickly, also from an evolutionary point of view (see Krishnan et al., 2004a , b). Therefore each of these two mutation types reacts differently to Dssh. Hence the ratio between the slopes of each of these gradients should differ if the gradients are due to transcription (C->T should be less saturated and more similar to the A->G gradient because transcription is faster than replication) than when deamination gradients are due to replication. Hence such analyses could determine which process, transcription or replication, created the detected gradient(s), even when both processes are collinear and apparently confounded.
4.3. Gradient convergence and lifespan in Primates
Results in Figure 3 suggest that convergence between replication and transcription slows ageing-related processes in Primates. Note that Figure 3 shows that relative to other Primates, longevity in
The correlation in Figure 3 also remains significant after applying the method of phylogenetic contrasts (Felsenstein, 1985) to the data, suggesting that the results are statistically valid independently of phyletic constraints (r = 0.50, P = 0.03, one tailed test). It makes sense that results of regular correlations, with and without accounting for phylogenetic contrasts are qualitatively similar because evolution of tRNAs functioning as OLs tends to be saltatory (Seligmann et al., 2006b).
Gestation time, despite its association with maximal lifespan in Primates (r = 0.67, P < 0.01), only slightly increases with the level of Dsshr->Dssht convergence (r = 0.11, P < 0.1). This suggests that collinearity between replication and transcription might cause interferences, slowing down both processes and ultimately developmental rates. Even a weak effect on developmental rates (inversely proportional to gestation length) could be a potent selective pressure in natural populations, counterbalancing pressures against cumulating excess mutations that favor collinearity between the processes.
This effect on growth rates is probably relatively weak in Primates and in general K-strategists, which maximize lifespan rather than developmental rates (Brookfield, 1986). The opposite is expected in groups that are, relatively to Primates, more r-strategy-oriented, a strong prediction corroborated in Figure 10 and discussed below.
4.4. Correlations between molecular and whole organism levels
One should note that several correlations between life history parameters and molecular indices characterizing metabolic strategies of cells have already been described, specifically for Primates: the length of the gestation period with cost minimization of nuclear amino acid usages (Seligmann, 2003), cost minimization of mitochondrial ribosomal frameshifts (Seligmann & Pollock, 2004), slopes of (regular) mitochondrial replication gradients (Raina et al., 2005); and now maximal lifespan with convergence between mitochondrial replication and transcription. Seligmann & Krishnan (2006) discuss how whole organism properties probably result from many different, coadapted cellular processes, so that the wealth of significant correlations detected between molecular properties and whole organism features should be of no surprise. In addition, it is notable that nuclear genome size is not related to life-history traits in Primates (Morand & Ricklefs, 2005), so that effects of mitochondrial properties are more likely to be detected in this group.
4.5. Too extreme convergence between replication and transcription
The examination of Figures 3 and 4 shows that the trend between maximal lifespan and Dsshr->Dssht convergence has outliers, and that these outliers are usually placed in the same relative area of the graph: these are species with relatively high convergence but lower than expected lifespan. It is possible that this situation results from asymmetry in inaccuracies in maximal lifespan estimations, as sampling error can only cause lower values than the real maximal lifespan. However it makes little sense that the well studied
4.6. Rates of development and convergence between replication and transcription
The hypothesis that collisions decrease rates of replication and transcription when both processes are collinear predicts that rates of development decrease with Dsshr->Dssht convergence. The cause for this would differ from the correlation between Dsshr->Dssht convergence and lifespan. For lifespan, convergence decreases cumulation of mutations and in general, increases mutational robustness; for developmental rates, they are the direct result of decreased replication and transcription rates because of increased collision frequencies between replication and transcription forks. It is notable that this rationale yields a molecular mechanism for the well known negative association between metabolic rates and longevities, as described in Insects (Antler flies, Bonduriansky & Brassil, 2005; Drosophila, Marden et al., 2003; Novoseltsev et al., 2005; Mockett & Sohal, 2006), nematodes (Jenkins et al., 2004; Chen et al., 2007; Lee et al., 2006; Hughes et al., 2007) and mice (Cargill et al., 2003; and others, Bonsall, 2006). Some ecological data explaining the tradeoffs exist (Bonduriansky & Brassil, 2005), and results suggest the tradeoff is due to dietary metabolism (Partridge et al., 2005a,b; Speakman, 2005a,b; Kaeberlein et al. 2006; Ruggiero & Ferrucci, 2006; Szewczyk et al., 2006; Wolkow & Iser, 2006). Other evidence shows that this rule might not be universal (Van Voorhies et al., 2004; Khazaeli et al., 2005; Johnston et al., 2006), stressing the need for unifying hypotheses. Several molecular or biochemical mechanisms have been proposed (Balaban et al., 2005; Bartke,2005; Knauf et al., 2006; Powers et al., 2006 ) but no general molecular model exists, stressing the importance to link the Dsshr->Dssht convergence hypothesis with the lifespan-growth rate tradeoff. Making a meaningful test for this prediction that Dsshr->Dssht convergence decreases developmental rates (hence increases the length of gestation) is not straightforward because of the strong positive association that exists between maximal lifespan and gestation length. However, using evolutionary ecology theory on r and K strategists, the simple molecular mechanism makes complex predictions on the relative strengths of association of Dsshr->Dssht convergence with lifespan and gestation length, respectively. The fact that these predictions are overall verified by the analysis of a large number of species and groups of species in Figure 10 is strong support for the Dsshr->Dssht convergence hypothesis and its coevolution with major life history traits.
4.7. Causal interpretations of correlations between lifespan and convergence between replication and transcription
The hypothesis that collisions decrease rates of replication and transcription when both processes are collinear enables to predict the occurrence of species that seem outliers in graphical analyses. However, the longevity-growth rate tradeoff hypothesis suggests the possibility that the causal interpretation of the association of Dsshr->Dssht convergence with maximal lifespan is opposite to the direction assumed. This is because offspring fitness decreases with parental age (Kern, 2001; Priest et al., 2002; Moore & Harris, 2003; Moore & Sharma, 2005), putatively due to ontogenetic cumulation of mutations, especially in mothers (McIntyre & Gooding, 1998; Hercus & Hoffmann, 2000), which are inherited by offspring. This issue is particularly relevant to mitochondria. Indeed, species with long lifespan probably have relatively high transcription/replication ratios. Hence what appears to be convergence of replication gradients towards transcription gradients could be the result of increased lifespan, rather than its cause. This interpretation assumes that the gradients observed are transcription-, rather than replication ones, which remains possible despite the arguments against this in previous sections. Notwithstanding these arguments, this interpretation is not compatible with other predictions presented here about developmental rates, the relatively frequent outlying species characterized by high convergence and lower than expected lifespan, and the threshold phenomenon observed in Figure 5. In addition, this individual-based observation is a stabilizing feedback mechanism where increased longevity causes inheritance of mutations that decrease offspring longevity. This would rather predict negative correlations, or no correlation at time scales larger than that of single generations, such as in the inter-species comparisons described in the Results.
The specific situation in
4.8. Developmental stability and convergence between transcription and replication
It is interesting to note that the principles observed for the association between Dsshr->Dssht convergence and lifespan are also valid for that between Dsshr->Dssht convergence and developmental stability. This observation fits the general trend that developmental instability associates with low fitness and pathologies. It would be interesting to explore whether this hypothesis of convergence of replication with transcription fits with the “double-agent” unifying hypothesis of ageing and diseases based on the tradeoff between oxidative stress inducing genetic reaction mechanisms against stress and its effect on ageing and age-related disease (Lane, 2003). The molecular processes presented provide mechanistic explanations for these similarities.
I present the original hypothesis that heavy strand sequences of tRNA-coding genes functioning as additional light strand replication origins tend to increase the similarity of mutational patterns resulting from replication with those due to transcription, putatively decreasing cumulation of mutations during the two processes. Variation exists among mitochondrial genomes in the extent that replication mutation gradients resemble transcription gradients; in most species (mainly short lived with high metabolism), replication gradients do not resemble transcription gradients. The similarity of replication mutation gradients to transcription ones correlates positively with maximal lifespan in Primates and other taxa. Systematically, outliers to these trends have replication mutation gradients relatively resembling transcription gradients but are for short lived species, the opposite (long lived outliers with replication gradients not resembling transcription gradients) does not occur. In some taxa such as Cetacea, this phenomenon is enhanced with two clearcut ranges in similarity between replication and transcription, one with relatively low similarities, where maximal lifespan increases with the similarity of replication gradients to transcription gradients, and another region where similarities are highest and maximal lifespan decreases with similarity. These patterns suggest that low convergence does not enable high maximal lifespans, but too high convergence limits lifespan, probably because too many collisions between replication and transcription forks decrease both replication and transcription rates, increasing durations spent single stranded, and mutation frequencies. The length of gestation periods increases also with convergence, notably, in r strategists; in K strategists, the convergence levels coevolve more with maximal lifespan, fitting the rationale that the molecular machinery is adapted for high metabolism and fertility in r strategists, and high survival in K strategists. Results are interpreted assuming that the observed phenomena are due to replication that sometimes resembles transcription, but are not due to transcription. Evidence supporting this is presented: in species possessing two gradients, one according to the classical replication origin, and one resembling transcription, both mutation gradients have very similar slopes, which is more compatible with a single enzymatic machinery (the mitochondrial gamma DNA polymerase) causing both gradients, rather than each due to a different polymerase. A method based on differences in the respective rates of replication and transcription for distinguishing between replication and transcription gradients is suggested, where the ratios between slopes of mutation gradients of purines versus pyrimidines should vary when mutation gradients are due to replication resembling transcription rather than transcription itself. The hypothesis that results are due to a causal relationship opposite to the one proposed (high longevity causes high transcription/replication ratios and hence transcription gradients dominate replication ones) is examined and discussed. This interpretation is unlikely, not only because gradients seem to be due to a single enzymatic process, but also because this hypothesis is less compatible with patterns in the data: among others it does not predict the patterns observed for outliers and the differences between r- and K-strategists.
I am very thankful to Neeraja M. Krishnan for manifold help in the preparation and improvement of this manuscript. Morphological data could not have been gathered without considerable curatorial help at the various herpetological collections I visited: American Museum of Natural History, New York; California Academy of Sciences, San Francisco; Field Museum, Chicago; Hebrew University of Jerusalem; Museum of Natural History, London; Museum of Vertebrate Zoology, Berkeley; Peggy Notebaert Nature Museum, Chicago Academy of Sciences; Tel Aviv University.