Performance of the Translational Apparatus Varies with the Ecological Strategies of Bacteria^▿

MATERIALS AND METHODS

RESULTS

The ecological strategies and rrn copy numbers of the strains examined in this study and their phylogenetic relationships (inferred from 16S rRNA gene sequence analysis) are shown in Fig. 1. From a previously described collection of recently isolated soil bacteria (31), we chose both a rapidly responding isolate and a slowly responding isolate belonging to the Actinobacteria, to the Bacteroidetes, and to the β-Proteobacteria. We also included a rapidly responding member of the γ-Proteobacteria and a slowly responding member of the α-Proteobacteria, but for these two strains our collection did not contain a matching strain of the contrasting strategy in the same phylogenetic subdivisions for these two strains. Nonetheless, the eight strains that we examined must represent at least four evolutionary transitions between the two ecological strategies, so any correlation between traits of the organisms and their ecological strategies cannot be explained simply by the shared inheritance of an ancestral phenotype. The γ-Proteobacteria strain E. coli REL607 and the α-Proteobacteria strain S. alaskensis RB2256 were included as well-characterized representatives of bacteria with the rapid and slow response strategies, respectively; these strains are laboratory-adapted strains derived originally from the human gut (40) and oligotrophic ocean water (15).

The taxonomic affiliation of each strain and its growth rate, biovolume, and culture density in two media are shown in Table 1. As expected, the rapidly responding bacteria grew faster than the slow responders; in fact, the growth rates of strains with different strategies barely overlapped under the conditions examined. The range of growth rates obtained in the two media was greater for the slow responders (mean, 1.7-fold difference) than for the rapid responders (mean, 1.2-fold difference). The biovolume generally increased with the growth rate for a strain (as expected), although the relationship between the biovolume and the growth rate varied considerably between strains.

The macromolecular contents per cell and per biovolume are shown for each strain-medium combination in Table 2. As expected, the rank order of macromolecular abundance was protein > RNA > DNA for all strains except the slowly responding mycobacterium strain PX3.15, for which the RNA and DNA contents were not statistically distinct. The macromolecular content per cell was quite variable among the strains, primarily reflecting the differences in cell size. The macromolecular content per biovolume was more consistent across the strains.

Our measurements for the protein and RNA contents per cell at a given growth rate for E. coli were comparable to those reported by Bremer and Dennis (3), once a correction was made for the expected 2.3-fold difference in growth rate due to incubation at 25°C instead of 37°C (17). Our temperature-corrected values for DNA content per cell for E. coli were about 1.7-fold higher than those of Bremer and Dennis (3), which could have been due to differences in the strains examined. However, we noted that an increase in the DNA content per cell at lower growth temperatures was also found in three of the five media tested by Schaechter et al., who performed the original study that established the conventional wisdom that temperature influences bacterial growth rates without altering the macromolecular content (60).

Calculations of macromolecular content per cell or per biovolume facilitate comparisons with previously published work, but they necessarily introduce additional experimental uncertainty by incorporating empirical estimates of culture density and biovolume. To avoid such errors, for our calculations for all quantities involving macromolecular ratios we used measurements for the protein, RNA, and DNA contents for the same cell pellet. The standard errors of estimates of such ratios were obtained from independent estimates for three or four replicate cultures of each strain-medium combination. Figure 2 shows the proportional content of each macromolecule (DNA, RNA, or protein mass divided by the total mass) for all strains and media as a function of the growth rate. The expected trends (an increasing proportion of RNA and a decreasing proportion of protein with increasing growth rate) are apparent across all strains and within most strains (3). However, the macromolecular compositions of all bacteria are generally similar regardless of the ecological strategy or the growth rate, with protein comprising 65 to 85% of the total macromolecular mass, RNA comprising 8 to 28% of the total macromolecular mass, and DNA comprising 3 to 10% of the total macromolecular mass.

The rate of protein synthesis (μP) is plotted against the RNA content for all strain-medium combinations in Fig. 3. The y axis represents the numerator of translational power, while the x axis represents the denominator, so higher translational power is represented by points toward the top left in Fig. 3, and any straight line through the origin connects points of equivalent translational power. As predicted, the translational power is higher for the rapid responders and lower for the slow responders. In the region where the RNA contents of the rapid responders and slow responders overlap, the rate of protein synthesis is three- to fourfold higher for the rapid responders, consistent with an analysis of previously published data for bacterial macromolecular contents (12).

Codon analysis provides corroborating evidence that the importance of translational power for bacterial fitness varies with the ecological strategy, using rrn copy number as an index of ecological strategies ranging from slow responders (low rrn copy number) to rapid responders (high rrn copy number). Translational selection for biased codon use is based on the contribution of rapid exponential growth to bacterial fitness (22, 39, 65), so it ought to be stronger among rapidly responding bacteria with high rrn copy numbers. However, to test this hypothesis, the effects of translational selection on codon use must be distinguished from the effects of mutation bias, which is the major source of codon use variation across genomes (6). We used the average codon frequencies across all genes in a genome to establish the expected frequencies due to the mutational bias of a bacterial strain, and we measured translational selection as the deviation of codon use in highly expressed genes from the expected values, quantified as ΔN′_C (see Materials and Methods). The total number of tRNA genes per genome provides an independent measure of translational selection that does not depend on codon analysis, because the benefit of using a preferred codon depends in large part on the abundance of its cognate tRNA (2), which is related in turn to tRNA gene dosage (14).

Translational selection in 76 fully sequenced bacterial genomes (58), expressed as ΔN′_C, is positively correlated with the number of rrn operons per genome, as predicted (Fig. 4a). Although ΔN′_C measures the difference in codon use between highly expressed genes and all genes of a genome, the number of rrn operons is not related to the degree of codon bias in either set of genes individually (Fig. 5). Thus, ecological strategies are related specifically to the strength of translational selection, not to mutational bias or other influences on codon use. The same correlation with ecological strategy is found when translational selection is measured as the number of tRNA genes per genome (Fig. 4b). Although Fig. 4 and 5 show each bacterial genome as a separate data point, in reality the bacteria have shared evolutionary histories to a greater or lesser extent and thus are not statistically independent. Hence, the significance and strength of these correlations were evaluated using maximum likelihood methods and ratio tests that accounted for the evolutionary history of the organisms (54).

DISCUSSION

The macromolecular data summarized in Fig. 3 confirm that the translational power is higher for rapidly responding bacteria than for slow responders, as suggested by previous results (5, 12, 18, 55, 63). This study extended previous pairwise comparisons of E. coli with other strains by including phylogenetically diverse representatives of both ecological strategies. Furthermore, measuring the macromolecular content of multiple species in a single study eliminated the possibility that the conclusions were influenced by methodological differences between laboratories. Although the generality of the link between translational power and bacterial ecological strategy has previously escaped notice, the result is consistent with analyses of codon usage bias across many species, which indicate that the strength of translational selection is correlated with rrn copy number and hence ecological strategy (Fig. 4) (64).

Paul Sharp and his colleagues have long argued that differences among bacteria in the strength of translational selection reflect, in part, ecological variation in the contribution of rapid exponential growth to fitness (22, 39, 65). Sharp et al. recently published an estimate of the fitness benefit conferred by preferred codon use in 80 bacterial species, based on four amino acids that have twofold codon degeneracy and the same preferred codon in all bacteria (64). As predicted, these authors found that the benefit of codon bias was correlated with rrn copy number, showing that selection for translational power is stronger in strains adapted to the rapid response strategy. We took a less precise but more inclusive approach, relying on an index of translational selection calculated for all 18 amino acids with multiple codons, and reached the same conclusion.

Could differences in codon usage across bacterial species be a sufficient explanation for the observed variation in translational power? Preferred codons are translated more quickly than nonpreferred synonyms are translated (9, 57), so a greater bias toward preferred codons in rapidly responding bacteria than in slowly responding bacteria should contribute to the high translational power of the former organisms. Unfortunately, the data required to quantify the effect of codon bias on the average translation rate are incomplete even for E. coli and are not available for other bacteria. Nonetheless, we have published a mathematical model comparing the translation rates of E. coli and a hypothetical bacterium that differs from it only by the absence of codon bias, bridging gaps in the data with assumptions that deliberately overestimate the effect of bias (12). This model shows that codon bias alone can accelerate translation no more than 0.6-fold in E. coli (12). We concluded that most of the three- to fourfold variation in translational power between rapidly and slowly responding bacteria must be attributed not to codon bias, but to differences in the performance of the translational apparatus itself.

Translational power differences cannot be explained by a complete absence of selection for high translational power among slowly responding bacteria. Codon bias is only weakly sensitive to natural selection for increased power, because any individual preferred codon has at most a small effect. Nonetheless, Sharp et al. detected translationally selected codon bias in 10 of 23 bacterial genomes with a single rrn operon (64). Selection for high translational power in other slow responders could be too weak to maintain hundreds of preferred codons and thus generate detectable bias, but it could still be strong enough to select for single mutations that directly influence the translation rate or the active fraction of ribosomes. Furthermore, the Ehrenberg-Kurland model shows that selection for high translational power is proportional to the fraction of cell mass invested in the protein synthesis system, but rapid and slow responders with similar levels of investment in protein synthesis have dramatic differences in translational power (Fig. 3) (5, 12, 18, 55, 63). Hence, we suggest that the observed differences in translational power are explained not only by a reduced benefit of high power for slow responders but also by a cost of high power that is particularly acute for these strains. Specifically, we propose that there is an evolutionary tradeoff between translational power and translational yield.

All known organisms consume four high-energy phosphate bonds per amino acid during ribosomal protein synthesis, so we expect the variation between strains in the translational yield of functional protein to be due entirely to variation in translational error rates. Kurland and his colleagues have argued that missense errors, which generally produce proteins with at least partial functionality, are much less costly to a cell than ribosomal processivity (i.e., movement) errors, which most often produce completely nonfunctional polypeptides (35, 37). Processivity errors include both frameshifts and dropoff events, the accidental release of peptidyl-tRNA from the ribosome at a sense codon (35, 37). Dropoff is much more frequent than frameshifting, at least for laboratory-adapted strains of E. coli, the only organism in which they have been measured (25, 37). Estimates obtained with different experimental approaches have suggested that the average dropoff rate is roughly once per 2.5 × 10⁴ codons, implying that about 10% of the ribosomes initiating translation on a typical gene fail to reach the end (37). The resulting dropoff products can be hydrolyzed to recover the constituent amino acids, but the energy spent on partially synthesized polypeptides is lost.

The possibility of a power-yield tradeoff mediated by dropoff errors is consistent with studies of wild-type and mutant ribosomes that have shown connections between the rate, accuracy, and processivity of translation (1, 13, 25, 26, 35, 37). Selection in the lab for high translational power in natural E. coli strains also reduces their tolerance for starvation, which could be a direct result of reduced translational yield (38, 48, 49). One mechanistic basis for the tradeoff could be the affinity of the ribosome for peptidyl-tRNA, which affects both the translation rate and the dropoff frequency (26, 37, 41). A second potential mechanism relates dropoff events to the fraction of translationally active ribosomes: the same factors that recycle posttermination ribosomes to the active pool also promote dropoff events (23, 56).

The waste of energy due to translational errors may not matter to rapid responders growing in resource-rich environments; there is little evidence that energy limits the growth of E. coli or other fast-growing strains in typical laboratory media (46, 50, 59). In nature, however, resource levels often fluctuate around the minimum thresholds for bacterial growth or viability, thresholds which are determined in part by the resources required for essential protein synthesis (21, 52, 66). In this situation, selection favors strains that use resources efficiently enough to grow or at least survive, while competitors remain in stasis or die. The spatial structure of environments such as soil could also result in selection for resource use efficiency and high translational yield, if the resources spared by efficient growth are not shared among all competitors but are used primarily by the more efficient cell and its descendants. In contrast to the fitness of rapid responders, which depends primarily on the rate at which the cells convert resources into progeny, we believe that the fitness of slow responders depends primarily on the number of viable progeny generated from a given allocation of resources.

Thus far, dropoff rates have been estimated only for laboratory-adapted strains of E. coli (25, 45, 47). Such measurements have to be extended to phylogenetically diverse microbes with both rapidly and slowly responding ecological strategies in order test our hypothesis that dropoff errors mediate an evolutionary tradeoff between translational power and translational yield. Whether or not this hypothesis is confirmed as an explanation of low translational power in slowly responding bacteria, the results reported here show that translational power varies among bacterial species in a manner that reflects ecological factors and not phylogeny. Recognition that the translational apparatus responds to selection has the potential to increase our understanding of translation itself, via comparisons of the structures and functions of the translational apparatus among microbes adapted to different ecological conditions. Furthermore, the adaptive nature of translational performance is likely to be an important factor contributing to the diversity and distribution of microbial life on our planet.