Results

- Widespread genetic exchange among terrestrial bacteriophages

Frequency of Cystoviridae Among All Phage Isolates. Of the plant material collected in CA, 40% of all phages that formed plaques on Pp had tripartite dsRNA genomes, identifying them as members of the Cystoviridae.

Phylogenetic Reconstruction. In total, the sequence data encompassed 413 bp of the L segment, 296 bp of the M segment, and 315 bp of the S segment for each isolate. These sequences were orthologous to base pairs 2,293-2,705, 70-362, and 240-554 in the L, M, and S segments, respectively. In the L segment, this range corresponds to the middle of the coding region of gene 2 (part of the RNA polymerase), whereas, in the M and S segments, it corresponds to part of the pac sequence and part of the first ORF. The pac sequence of each segment is important for segment packaging into the viral procapsid. We constructed Bayesian phylogenies using the sequenced portions of each genomic segment and the published sequences of four previously isolated Cystoviruses (25, 39, 40). The 50% majority-rule consensus trees are illustrated in Fig. 1. These phylogenies contain extensive polytomies, because many of the phage sequences are nearly identical. It is possible that the segments differing by 1-2 mutations were clonal before isolation. Approximately 20 phage generations occurred between isolation and sequencing, and, given a mutation rate of 10^-5 per generation per nucleotide (44, 45), the probability of a single mutation among the 1,024 base pairs sequenced is ≈20%.

Segment divergence peaked at 19.2% (S segment), 33.2% (M segment), and 31.4% (L segment). At 4-fold degenerate sites within the L segment, divergence was >50%. The major portions of the S and M sequences were noncoding, so an analysis of polymorphism at 4-fold degenerate sites was not possible. Interestingly, the recently described Cystoviruses that grow on Pp (25) (isolated in 1999 in NY) fell within the M-segment diversity we observed, and two were slightly (with low posterior probability) more diverged on the S segment (Fig. 1). An orthologous sequence from the L segment for these clones is not available, preventing any phylogenetic comparisons. All of the environmental isolates were found to be at least an order-of-magnitude more diverged from 6 than any other phage samples currently in our laboratory, ruling out laboratory contamination. Additionally, sequencing was done at two different laboratories: one in Cambridge, MA, and the other in San Diego. Finally, the phage isolation and sequencing done in San Diego were performed at two time periods approximately 1 year apart.

Population Structure. From the sequence data and phylogenetic reconstructions, it was apparent that migration and reassortment occurred frequently. Considering the three sets of multiple clones isolated from single clovers (14 viruses total), 2 viruses were identical for all three segments (CA/KW66b and CA/KW66d), and two viruses differed by two mutations (CA/KW64d and CA/KW64e). Each of the other 10 appeared to have undergone at least one reassortment event. In several other cases, phages that shared identical S or M segments had only distantly related L segments.

We tested all our isolates for an association between genetic and geographic distance by using a Mantel test. We removed from this analysis one of the two clonal isolates (CA/KW66b and CA/KW66d), taken from a single clover, because these phages were isolated from a single plate, and we therefore suspected that the two isolates may have been derived from a single progenitor phage. Including both of these isolates in the analysis did not qualitatively change the results. All other identical or nearly identical phages were isolated from different plates (e.g., CA/KW66c and CA/KW68), so they could not have been derived from a single progenitor phage. We also excluded previously described Cystoviruses (25), because they had been isolated at least 4 years previously, their geographic origin is unknown, and, for all of these phages, no orthologous L-segment sequence is available. Values of r, Pearson's correlation coefficient, together with the associated P values are shown in Table 1. No statistical evidence of geographic population structure was observed for the L or S segments, although a significant correlation between geographic and genetic distance was detected for the M segment.

Intrasegment Recombination. We tested for intrasegment homologous recombination by examining the relationship between nucleotide distance and LD, using D′ and r² as metrics; recombination should manifest as a significant negative relationship between distance and either metric. In testing for recombination, we used all available sequences. We used the permutation test implemented in the pairwise program from the LDHat package to test for significance of the relationship between D′ or r² and nucleotide distance. Previous experimental evidence has indicated that homologous recombination is extremely rare in Cystoviruses, occurring at a rate of ≈10^-7 per segment per generation (26). The results here suggest that homologous recombination is also rare in natural populations. In two cases (L segment vs. D′, and M segment vs. r²), there was a marginally significant negative relationship that did not remain after correcting for multiple tests (Table 2). Additionally, in both cases, the other metric held no significant relationship with nucleotide distance.

Reassortment. We also looked at levels of LD between segments. LD may exist between phage segments for three reasons: (i) Phages may not frequently migrate over large distances; (ii) if they do, coinfection and reassortment may be rare; and (iii) reassortants may have low fitness (on average) and thus experience negative selection. For all tests of LD between segments, we treated each segment as a single locus, because the above analysis suggested that intrasegment recombination is extremely rare or nonexistent. In the case of the Mantel test, LD was tested for by using the genetic distance between segments; for the r² and D′ metrics, it was done by averaging each metric over all pairs of nucleotides between segments to obtain a single measure of average LD between segments; for the phylogenetic test, we used phylogenetic congruity between segmental phylogenies. We performed the analyses using all of the recently isolated samples, except CA/KW066d, for the reasons discussed above.

Using a Mantel test, we looked for a positive correlation of genetic distance between segments, which would indicate LD. No association was found between the L segment and any other segment. A marginally significant correlation between S- and M-segment similarity was found (P = 0.031, Table 1).

We used a potentially more powerful test of LD in which we calculated the average value of r² and D′ over all polymorphic sites between segments. We used a permutation test to assess whether this value was significantly greater than the expected average r² or D′ if all segments were in linkage equilibrium. The majority of these tests were not significant. P values for the observed D′ values were 0.095, 0.318, and 0.261 for the S-M, M-L, and S-L segment comparisons, respectively (see Fig. 3, which is published as supporting information on the PNAS web site). P values for the observed r² values were 0.015, 0.337, and 0.306 for the S-M, M-L, and S-L comparisons, respectively (see Fig. 4, which is published as supporting information on the PNAS web site). In one instance (between the S and M segments), the observed LD metric was significantly higher than the observed metric between unlinked segments. However, this observation was not consistent for both measures of LD. The power of both this analysis and the Mantel test of LD may be compromised by homoplasy, which may obscure the true historical relationship between alleles.

In an attempt to account for this situation, we also used a phylogenetic analysis. Such an analysis would suggest LD between segments if the segment phylogenies were significantly more similar to each other than to a random set of phylogenies. Five-hundred random trees from the set of 5,000 most-recently visited trees in the MCMC chain were used for the analysis. The frequency of each tree in this data set is proportional to the probability that it is the true tree. This set of 500 trees was compared to two sets of other trees: five hundred of the most-recently visited MCMC trees from another segment, or 500 randomly constructed trees (43). We used two measures of tree-to-tree distance, the number of resolved identical triplets and the number of resolved different triplets between the sets of phylogenies (43).

By using the median number of identical triplets as an estimate of the true distance between phylogenies, we found that none of the segment phylogenies were significantly more closely related to each other than they were to a random set of trees (Fig. 2A, mean number of S-M identical triplets: 834, P = 0.940; M-L, 763, P = 0.878; S-L, 641, P = 0.552). However, there were significantly fewer different triplets between the S and M segment phylogenies (Fig. 2B, mean number of S-M different triplets, 1,119; P = 0.047; M-L, 1,143; P = 0.074; S-L, 1,214; P = 0.211).