The principle of parsimony suggests that the most likely evolutionary path between two points in a sequence space is the one which requires the least number of events. Although parsimony is primarily used to construct phylogenetic trees, it and other methods such as maximum likelihood, also provides predictions of ancestral sequences. Traditionally these were ignored. This began to change with the demonstration that parsimony predictions of ancestral sequences on a viral system of known genetic history were largely correct 1 . Subsequently, there has been an increasing interest in what ancestral sequences might reveal about ancient macromolecules 2 . Thus, soon thereafter parsimony analysis was applied to reconstruct the evolutionary history of the artiodactyl ribonuclease superfamily using sequence data 3 . The resurrection of ancestral proteins and viruses 4 5 coupled with demonstration of functionality is now increasingly common and such studies are providing important insights to molecular evolution 6 7 8 .

The primary focus of ancient sequence resurrection has been proteins 9 10 but the increasing realization that RNA played a key role in the early history of life and is intimately involved in the regulation of gene expression suggests that resurrection of ancient RNAs will likely be informative too 11 . Of particular interest is the ribosome whose history is closely intertwined with the early history of living systems 12 13 14 15 . Functional RNAs such as the ribosomal RNA, (rRNA), typically have complex secondary and tertiary structures that must be preserved over evolutionary time to facilitate function.

5S rRNA is the smallest RNA component of the large ribosomal subunit and typically contains approximately 120 nucleotides. Its secondary structure 16 17 is known and in the case of Escherichia coli its tertiary structure in the context of the ribosome is also available 18 . In an earlier investigation, all the sequences along the equally and most parsimonious paths between two pairs of modern 5S rRNA sequences were examined 19 . Each intermediate sequence was tested for validity as a functional 5S rRNA and it was thereby shown that some of the intermediates were not functional. Thus, some evolutionary paths through the sequence space are preferable to others despite the fact that all appear to be equally parsimonious. In order to be useful, methods for constructing ancestral sequences should to the extent possible avoid producing nodal sequences that are not biologically functional and hence are unlikely to have existed in the past. Fortunately, the usual parsimony procedure will automatically conserve sequence correlations between pairs of positions and hence the predicted nodal sequences will usually satisfy secondary structure requirements. Thus, there is good reason to anticipate that many nodal sequences will be functionally valid. Herein, we successfully test this conjecture by resurrecting an ancestral 5S rRNA sequence that is predicted by parsimony and showing that it is in fact a functional 5S rRNA.

The experimental system employed here utilizes a plasmid encoded 5S rRNA derived from the marine bacterium Vibrio proteolyticus that is expressed in an E. coli host. This 5S rRNA was initially chosen because its 5' and 3' terminal sequence regions are identical in sequence to those of E. coli 5S rRNA. The plasmid used, pCV251, was derived from a plasmid pKK5-1 20 , which carries a large portion of the E. coli rrnB operon. Additional file 1, Figure S1 illustrates the composition of plasmids pKK5-1 and pCV251. Plasmid pCV251 has two ribosomal RNA promoters, P1 and P2, a small fragment of the 16S rRNA sequence, a chemically synthesized V. proteolyticus 5S rRNA gene and the two rrnB terminators, T1 and T2 21 . Thus, the V. proteolyticus 5S rRNA is surrounded exclusively by E. coli sequences thereby ensuring that all signals for correct transcription, processing, and maturation of variants of the V. proteolyticus 5S rRNA in E. coli are present. The validity of the V. proteolyticus 5S rRNA sequence and its variants is tested in the E. coli cellular context. There are two major assays employed for this purpose. One is an assay that allows determination of the quantitative level of each variant in three cellular fractions; e.g., total 5S rRNA, 50S ribosomal subunits, and 70S ribosomes. The other assay is a competitive growth rate assay that competes each variant with a strain carrying the wild-type V. proteolyticus 5S rRNA. The V. proteolyticus 5S rRNA has previously been shown to be expressed to significant levels and incorporated into ribosomes without significant detriment to growth 22 . If a variant is not functional, it will either be excluded from active ribosomes or if incorporated into large numbers of ribosomes, substantially impact cell growth rate.

The parsimony algorithm (PHYLIP version 3.5) available in GDE (Genetic Data Environment) 23 was used to construct phylogenetic trees for 32 V. proteolyticus 5S rRNA sequences 24 . Bacillus subtilis 5S rRNA was used as the outgroup sequence. DNAPARS, estimates phylogenies by the parsimony method using nucleic acid sequences and actually provides predicted sequences at each node on the phylogenetic tree.

Escherichia coli strain JM109 was used for the growth of bacteriophage M13. E. coli HB101 was used for rRNA preparation and in growth rate assays as lac^- control. E. coli ML401 was used in growth rate assay as a lac⁺ competing test strain. The E. coli TG1 strain was supplied with Sculptor™in vitro mutagenesis system (Amersham Life Science Corp.) for the mutagenesis reactions. V. proteolyticus ATCC15338 carried the wild-type 5S rRNA. Plasmids pKK5-1 and pCV251 carry the E. coli 5S rRNA gene and V. proteolyticus 5S rRNA gene respectively 21 .

All newly described Vibrio 5S rRNA variants were constructed by in vitro mutagenesis using the manufacturer's protocol provided with the Sculptor™mutagenesis kit. The mutants were verified by dideoxynucleotide sequencing. Each V. proteolyticus 5S rRNA variant was cloned into pKK5-1 using the HindIII sites, and subsequently transformed into E. coli strain JM109.

Total cellular RNA was isolated by low-pH phenol extraction 25 . Ribosomes were separated by sucrose gradient method 26 . The ribosomal subunits, 70S ribosomes, and polyribosomes were separated by ultracentrifugation at 160,000 × g at 0°C for 4 hours using a Beckman SW41 Ti swinging bucket rotor. After ultracentrifugation, all gradients were analyzed at 280 nm using an LKB detector (BROMMA) Model 2112 REDIRAC fraction collector. Fractions containing 70S ribosomes, 50S ribosomal subunits and 30S ribosomal subunits were prepared individually, followed by ethanol precipitation, and phenol/chloroform extraction.

The extent of RNA incorporation into various ribosome fractions was examined. To this end, 5S rRNAs from V. proteolyticus and E. coli were purified respectively from their total RNAs by electrophoresis on two individual 10% polyacrylamide gels 21 . The concentration of purified 5S rRNA was measured by Stratagene's EAGLE EYE™II after running on an agarose gel. Northern Blotting was performed by generating a standard 13% polyacrylamide gel, which was loaded with V. proteolyticus and E. coli 5S rRNA in various calculated ratios. The RNA was then transferred from the gel to a charged nylon filter-Hybond™-N⁺ (Amersham), which would be first hybridized with a probe (HV2) complementary to V. proteolyticus 5S rRNA. The relative gel band density, which reflects radioactivity, was measured using a phosphoimaging system (Fuji, BAS-MP). The Macintosh program MacBas Version 2.1 was used to process the data. Subsequently, the filter was stripped completely before hybridizing a second probe (HE2) complementary to E. coli. The V. proteolyticus specific probe used (HV2, 5'-GTCCAAATCGCTATGGTTGC-3') exhibits seven nucleotide differences relative to the E. coli specific probe (HE2, 5'-GACCACCGCGCTACTGCCGC-3') that targets the same region of 5S rRNA., This level of difference ensures hybridization to V. proteolyticus 5S rRNA without cross hybridization to the host E. coli 5S rRNA 19 . Both HV2 and HE2 probes were [γ-³²P]-ATP end-labeled. Once the radioactivity of each band from the standard gel was measured, a standard curve relating the 5S rRNA concentration ratio to probe intensity ratio of V. proteolyticus versus E. coli was constructed. The amount of each V. proteolyticus variant in a sample can be obtained relative to E. coli, e.g. as a ratio, by comparing the V. proteolyticus variant band density with those obtained on the standard curve. Probes HV2 and HE2 were used in this manner to quantify each variant of the V. proteolyticus 5S rRNA in 50S subunits, 70S ribosomes, and total 5S rRNA. 5S rRNA is not present in the 30S subunits.

Competition growth assays were used to compare the growth rate of the strain carrying each variant 5S rRNA relative to a strain carrying the wild-type V. proteolyticus 5S rRNA. E. coli strains HB101 and ML401 were transformed with unmodified pCV251 (control) and pCV251 carrying the variant 5S rRNA instead of the wild type V. proteolyticus 5S rRNA (test). The growth rate difference was measured by following the ratio of cell numbers between the Lac^- HB101 (control) and Lac⁺ ML401 (test) when mixed cultures were grown on indicator plates 27 28 29 30 31 .

Growth rate assays were initiated by inoculating each strain from single colonies into 10 ml of LB medium containing 50 μg/ml ampicillin and growing at 37°C for approximately 12 hours. Subsequently, the mixture of two strains was diluted to 10^-3 in M9 media. Two sequential subcultures were grown. Samples were taken from each subculture, serially diluted, and plated on MacConkey indicator agar. After incubation at 37°C for 24 hours, Lac^- and Lac⁺ colonies were scored on the basis of colony color (Lac^- : white, Lac ⁺: pink). The growth rate difference or selection coefficient (s) was calculated according to the following equation 26 .

where N_p(t) and N_r(t) represent the relative number of the two competing strains at time t (hr) and s is a measure of differential growth rate (or selection coefficient) per unit time between the two strains. The strain designated by N_p is favored, neutral, or disfavored relative to the strain designated by N_r as s > 0, s = 0, or s < 0. A parallel experiment was performed with the strains carrying the test and control reversed.

One hundred phylogenetic trees were constructed from the Vibrio 5S rRNA dataset using parsimony. Two ancestral nodal sequences, at node 1 and node 2, were chosen for experimental study. The sequences at both nodes were unambiguous and consistently predicted in all 100 trees. Figure 1 is a randomly chosen representative tree. There are 7 sequence differences between the modern V. proteolyticus 5S rRNA sequence and ancestral node 1. The base changes at node 1 are A19U, U21G, U52A, A62C, U64A, U65C, and U88C as indicated on Figure 2. Of these changes, five involve positions involved in secondary structure. Thus, the 21/62 combination converts a UA pairing to a GC pairing, the 19/64 combination converts an AU pair to a UA pair and the change at position 65 converts a GU wobble pair to a Watson-Crick GC pair. In addition to the 7 changes in node 4, an eighth change, C70U, which is in a hairpin loop region, produces the node 2 ancestral sequence.

A previously constructed plasmid, 19 , was used as the starting point for the mutagenesis because it already had three of the required base changes, A19U, U52A, and U64A. This plasmid produces an RNA that is functional in the E. coli cellular context. A series of consecutive mutagenesis reactions were performed to construct the final two plasmids carrying the complete node 1 and node 2 sequences. This mutagenesis process resulted in the formation of two intermediates sequences. Relative to the V. proteolyticus wild type 5S rRNA, the first intermediate had changes A19U, U21G, U52A, and U64A, while the second had changes A19U, U21G, U52A, A62C, U64A, and U65C. In addition to the two ancestral nodal sequences and the two intermediate sequences, two single point mutations (G33C and C49G) and one double mutant (G33C; C49G) were made in the work described here. Positions 33 and 49 are pairing partners in a duplex region of 5S rRNA and are not associated with the resurrection. After mutagenesis, each variant was confirmed by DNA sequencing and cloned into the expression vector as described in the methods. Table 1 shows the results of quantization and growth rate analysis for these seven mutants. In addition, eleven other relevant variants have been previously studied 19 32 and are included in Table 1.

None of the constructs had a substantial affect on the growth of the host strain. However, by examining the accumulation levels of each construct in the total RNA pool and the various ribosome fractions each construct can be classified into one of three major categories referred to as green, red and blue. The green variants are considered to be functional. They are extensively accumulated and incorporated into 50S ribosomal subunits and 70S ribosomes without significant detrimental effect on cell growth rate. The red variants are extensively accumulated in the total 5S rRNA pool but not significantly incorporated into 50S ribosomal subunits or 70S ribosomes. These variants are considered to be non-functional. The blue variants such as A19U accumulate poorly in the total 5S rRNA pool, apparently because the RNA has become unstable. Typically, these variants are expected to be non-functional but in some cases they may actually be functional as they are observed to accumulate at low but nevertheless similar levels in 50S ribosomal subunits and 70S ribosomes. The single base change variant C70U is an example of a blue variant of this type.

The node 1 sequence, which is the predicted common ancestor of 28 Vibrio 5S rRNAs, was in the green category and can therefore be considered to be a valid 5S rRNA. Likewise all the intermediates that did not disrupt secondary structure were also found to be functional in the E. coli context. The additional base change required to create the node 2 sequence resulted in loss of stability such that it was not possible to unambiguously determine whether this nodal sequence would be functional or non-functional, if it accumulated at higher levels. However, it is best concluded that this nodal sequence is likely not a valid ancestor.

It is inherent in the concept of an ancestral sequence that a true ancestor would have been functional in the cellular context that existed at the time. Thus, if a predicted ancestor when resurrected is functional in the modern cellular context, it is far more likely to be a realistic ancestor than one that is not. One can argue that viability of a true ancestor might be lost following resurrection due to changes in the cellular context. In the present instance, we believe this is unlikely to be the case. In prior work, 22 the wildtype sequences of ten different species of modern Vibrio 5S rRNAs were tested for functionality in the E. coli cellular environment. All, were accumulated in ribosomes without significant detriment to growth rate (most were actually slightly favorable) and therefore regarded as functional in the E. coli cellular environment. The evolutionary distance between E. coli and any of these 10 Vibrio species clearly exceeds the distance between the modern Vibrio 5S rRNAs and the ancestors reconstructed here.

The node 1 5S rRNA was found to be functional in the E. coli cellular context. This RNA has seven base changes relative to the modern V. proteolyticus 5S rRNA Given the low probability that this number of changes could be randomly introduced without disrupting function 33 , this should be regarded as a successful resurrection of a realistic ancestor. Of these changes, all have previously been studied as individual point mutations, or in the case of those involving base pairing, as appropriate double mutants, Table 1. When made individually, U52A, U65C and U88C, produced functionally valid 5S rRNAs. Likewise, when both partners in the 19/64 and 21/62 base pairs are changed simultaneously, the resulting 5S rRNA was a functionally valid 5S rRNA. Thus, the individual changes that comprise the ancestral sequence all produce valid 5S rRNAs too. This reaffirms the earlier conclusion 32 that variants which occur frequently in a local region of the 5S rRNA structure space (these seven do) are largely independent of one another and hence can occur separately or together in many sequences, e.g. in this case in the predicted ancestral sequence without disrupting function.

When only one residue in a base pair was changed, e.g. A19U or U64A the individual point mutations did not produce a functional 5S rRNA whereas changing both positions in unison did. Since single changes are more likely to occur than double changes one might expect paired regions to change more slowly than other regions. However, this need not be the case. It was found in the earlier pathway studies 16 that one could bypass the double mutant by making other variants first. In addition, the success of the three double mutants examined here in producing valid 5S rRNAs indicates that there is a very high probability that a double mutant that preserves secondary structure will be accepted, if the bases that comprise the pair are not involved in tertiary interactions. In contrast, a single mutation in a non-paired region may be significantly less likely to be accepted. Thus, the overall rate of acceptance may not be dissimilar even though double mutants are less common. Also, and perhaps more importantly, it should be appreciated that one can potentially transition from one standard pair to another via a GU wobble or GA intermediary.

Although the node 1 sequence is regarded as a valid ancestral sequence, a single additional base change, C70U, made the RNA sequence represented by node 2 unstable. Since the node 2 RNA does not accumulate to a significant extent and is absent from ribosomes, it is at best very suspect as a valid ancestor. As a separate variant, the C70U change alone, Table 1, is also unstable in the E. coli cellular context. It does, however, accumulate in ribosomes enough to suggest that it is likely functional, if it can avoid premature degradation. Thus, the behavior of the multiple variant at node 2 is again strongly influenced by the behavior of the individual variants that comprise it.

From the perspective of the present study, the appropriate conclusion is that node 2 is most likely not a valid ancestor. When one traverses from node 1 to node 2, the new nodal sequence is based on the addition a single additional 5S rRNA sequence, Vibrio damsela whose 5S rRNA has C70U. However, relative to node 2, this 5S rRNA has three additional base changes, A25U, U35A, and C88U. Unless it was erroneously determined, the V. damsela sequence is a valid 5S rRNA. Thus, the comparison suggests that U25, A35, or U88 in some combination or separately compensate for the negative effect of C70U in order to make this 5S rRNA functional. The availability of additional sequences related to V. damsela might have improved the prediction at node 2. The results presented here point out the danger of allowing predictions to be strongly influenced by a single sequence.

In summary, it has been demonstrated that an ancestral 5S rRNA sequence predicted by parsimony can serve as a functional RNA and thus can be regarded as reasonable approximations of a true ancestral sequence. However, a second predicted ancestral sequence did not pass the test of convincing functionality. In both cases, the functional behavior of the ancestral sequence could be largely anticipated based on the behavior of individual base or base pair changes that comprised the predicted ancestral sequence. A strength of the parsimony approach in the present context is that by its nature, meaningful secondary structure will be preserved because paired positions will typically exhibit co-variation when numbers of sequences are aligned. In the future, it should be possible to resurrect other 5S rRNA ancestors as well as other interesting RNAs such interesting regions of the large ribosomal RNAs.

XN, e.g. U88: Uracil at position 88; XNY, e.g. U21G: Uracil at position 21 changed to guanosine; 5S rRNA: 5S ribosomal RNA.

GEF conceived of the study, participated in the interpretation of the results, and prepared the final manuscript. QL designed the experimental system and did all the experimental work, and participated in the interpretation of the results and prepared the first draft of the manuscript. All authors read and approved the final manuscript