In the spring of 2004, 2005 and 2006, several diseased wheat plants showing extreme dwarfing, various types of yellowing, and reduced or no heading were found during field surveys in many wheat fields of China 32. Wheat samples collected from northern, central, northwestern and southwestern areas tested positive by PCR for WDV, suggesting that WDV was widely distributed throughout China. The full genomes of 28 isolates from different regions of China were sequenced in this study. Details of these, together with those of the 18 complete WDV, BDV and ODV genomes already published, are provided in supplemental material [see Additional file 1 and 2]. Phylogenetic trees were constructed by neighbor-joining (NJ) (Figure 1) and maximum-parsimony as described in Methods using Maize streak virus as an outgroup. The topologies of the two types of trees were identical at all branch points that were well supported by bootstrap analysis (> 70%) but differed at some branch points with low statistical support (Figure 1).

Bootstrap analysis supported an unambiguous host-dependent clustering of wheat-, barley-, and oat-derived samples. The isolates also fell into subpopulations seemingly structured according to geographic origin. Wheat isolates from China were > 94.8% identical in nucleotide sequence, whereas > 92% identities were seen between the wheat isolates of China and those of other countries. Greater than 94% nucleotide sequence identities were found among barley isolates. Nucleotide sequence identities were < 69% between wheat and barley isolates. The oat-derived isolate exhibited 59.7%–69.3% identities to those of barley, and 68.3%–69.4% to those of wheat. The values for wheat-barley, barley-oat and oat-wheat comparisons were below the threshold for Mastreviruses species (75%) as proposed by the ICTV 1825.

Unscaled evolutionary distances from the common ancestors to their sequence progeny were deduced from the NJ tree. The distances indicated that the barley isolates diverged from wheat isolates much more recently than the oat isolate did. The most basal barley isolate was TR2 from Turkey, followed by those from Germany and the CSR. Wheat isolates were separated into two groups, the first one from Europe and the second one from France, the CSR, and China. The grouping of all of the Chinese isolates in one clade with the CSR and French isolates suggested that they evolved from the same ancestor. The Chinese wheat isolates were divided into two clades, but were not clustered by geographical source or collection time (Figure 1). For example, 10 isolates from Shijiazhuang, Hebei province, were classified into four different sub-clades in the phylogenetic trees (Figure 1).

Barley and wheat are members of the tribe Triticeae. Together with the Aveneae, which includes oats, they belong to the BEP clade of the Poaceae, while maize represents the PACCAD clade of the Poaceae 33. Maize and wheat lineages have been suggested to split between 50 and 80 Mya 34 and between 44 and 60 Mya 35. The estimate for the wheat-barley divergence from a common ancestor is 11.4 +/- 0.6 Mya 36. The divergence of Aveneae (oat) and Triticeae (barley and wheat) has been placed at 25 Mya 37. In the absence of provision of an error for this estimate, we assumed a range of possible times from 22 to 28 Mya. Thus, the relationships of the maize, oat, barley and wheat mastreviruses revealed in the phylogenetic tree of Figure 1 appeared to mirror the taxonomic positions of their hosts, which observation caused us to test whether their relative divergence times agreed with estimates for divergences of the host species. Relative divergence times for the respective mastreviruses were calculated by MEGA and the correspondence between host and virus times was examined by linear regression. Table 1 showed that the r² values were greater than 0.99 regardless of whether middle, high, or low values for host divergence times were used, with the middle values giving the best fit. Thus, we cannot reject the hypothesis that the mastreviruses codiverged with their hosts. Relative divergence times for viral evolution were converted to Mya times using the slope of the relationship of the middle host divergence times with virus distance, using the assumption that variation occurs according to a linear molecular clock. The results (Figure 1) suggest that the CSR, French and Chinese WDV isolates diverged from others 1.5 Mya and the Chinese isolates split off from the other two 0.4 My later. The radiation of sequences of Chinese isolates was predicted by this analysis to have occurred 0.6 Mya. Of the sites in the mastrevirus genome, 32% were substituted during evolution from the common ancestor of WDV and BDV, leading to a calculated evolutionary rate of 1 × 10^-8 (substitutions/site)/year.

To explore further the possibility of the codivergence of virus and host, we included additional members of the Mastrevirus family in the phylogenetic analysis. To align these additional sequences with the previously analyzed sequences in an unambiguous and reliable fashion, we considered elimination of regions of high diversity and regions prone to be different due to recombination. Regions with high nucleotide diversity were difficult to align reliably. To identify regions that could be reliably aligned, the level of genetic diversity (θw and π) 3839 of WDV was examined in genomic regions (Table 2) or along the entire genomes of Chinese isolates (Figure 2). Examination of nucleotide diversity values (Table 2) revealed few differences among the coding regions. For all WDV isolates, θw was significantly higher for the Rep region than for the other three regions, but this difference was not significant for the π nucleotide diversity. The higher Rep diversity was most apparent among non-Chinese isolates, for which both measures showed significance. Overall, as expected from Figure 1, diversity values were lower for Chinese isolates than for non-Chinese isolates. For the Chinese isolates, the π diversity values were slightly higher for Rep and RepA regions than for CP and MP regions. Theta values were not significantly different from one another. As is generally expected for mastreviruses, the most diverse regions over the 37 genomes were LIR and SIR 2528. Considering only the Chinese isolates, however, the diversity of the SIR region was not distinguishable from those of the coding regions. In the non-Chinese isolates, MP diversity could not be distinguished from diversity for the non-coding regions. Among the non-Chinese isolates, the SIR region was more diverse than LIR, while the opposite was true for the Chinese isolates. Overall, SIR was more diverse by the θw measure, but not significantly so using the π diversity value.

A plot of the diversity values as a function of position in the genome revealed that determinations of diversity over large genomic regions obscure regions of high diversity embedded in a background of low diversity (Figure 2). The diversity distribution presented a fluctuating pattern: (1) high diversity regions were seen in both non-coding regions and coding regions; (2) divergence in the movement protein (MP) region was low except for the extreme 5' and 3' terminal regions of it; (3) highly diverse portions were also located at the 5' terminal regions of coat protein (CP), replication-associated protein (Rep) and Rep A genes; (4) in contrast, the lower diversity portions were located in the central and 3' terminal regions of CP genes and in the overlapping part of Rep and Rep A genes; (5) there were two additional high diversity stretches in the complete genome, one was in the intron, the other was in the LIR (Table 2). The smallest nucleotide diversity along the entire genomes of WDV isolates (Table 2) was located in the CP gene, but the gene also had 5' and internal regions of elevated diversity.

That regions with relatively low nucleotide diversities represented protein coding regions (Figure 2) suggested that these regions evolved under negative selection. To test this hypothesis, frequencies of synonymous and non-synonymous substitutions at sites in each of the four protein coding regions were calculated. Nucleotide substitutions at non-synonymous positions of non-overlapping regions (MP and CP) were less frequent than those of overlapping regions (Rep and RepA, Table 3). The K_a/K_s ratios of all coding regions were lower than 0.22, which suggested negative or purifying selection (K_a/K_s ratios < 1 indicating purifying selection) acting on the sequences, with no genes under positive selection (K_a/K_s ratio > 1 indicating positive selection). The CP gene had the lowest ratio, whereas that of the Rep A gene was highest (Table 3).

Mastreviruses are known to have experienced recombination in their evolution 40. Phylogenetic analysis of sequences containing recombination junctions can result in misleading interpretations. Visual inspection of the alignment of Chinese WDV nucleotide sequences suggested that parts of the genomes of HBSJZ06-11 and YNKM06-2 isolates derived from a genome or genomes not included in the analysis. Analysis by algorithms contained in RDP3 41 confirmed the visual observation. The result of SiScan 42 analysis (Figure 3) illustrates the possibility that these isolates were recombinants. The sequence of isolate GSGG05-2 was used as reference and is typical of all other Chinese WDV sequences. A Z-value (significance score) > 3 indicates reliably that two of the three sequences are more closely related to one another than either is to the third, identifying the third as a probable recombinant. HBSJZ06-11 was identified as having a recombinant segment in the LIR on one side of the nick site, while YNKM06-2 appeared to have an LIR segment derived from an unknown genome on the other side of the nick site. Since the recombined segment was nevertheless similar to WDV sequences, the recombination events likely were intraspecific. Work of others identifies the LIR and the SIR as frequent sites of recombination. Because of diversity hot spots in intergenic regions and the tendency for those regions to derive from recombination, further phylogenetic analysis was restricted to the coding regions.

To facilitate alignment, consensus MP-CP and Rep-RepA sequences were generated for mastreviruses with multiple sequenced representatives (WDV, BDV, MSV, Panicum streak virus (PanSV), Sugarcane streak virus (ScSV), and Urochloa streak virus (USV)). Single available sequences of Chloris striate virus (ChlStrV), Digitaria streak virus (DigSV), Miscanthus streak virus (MisSV), Eragrostis streak virus (ESV), Tobacco yellow dwarf virus (TYDV), Bean yellow dwarf virus (BeYDV) and Chickpea chlorotic dwarf virus (ChPCDV) also were included in the alignment [see Additional file 2]. NJ distance trees were constructed for Rep-RepA (Figure 4A) and MP-CP (Figure 4B) regions. A similar distance tree was constructed from an alignment of NCBI GenBank rbcL gene sequences from the plastids of respective host plants for which the virus was named, recognizing that the plant named in the virus name is not necessarily the one from which the virus is most commonly isolated. The rbcL tree is shown in each Figure 4A and 4B to the left of the virus trees. In both Rep-RepA and MP-CP trees, the isolates named for hosts in the BEP clade formed a monophyletic cluster, consistent with branching patterns of Figure 1. Similarly, the three mastreviruses of dicotyledonous plants formed a monophyletic grouping. The branching structure within these groups mirrored precisely the structure of the rbcL trees of the host plants. Viruses named for hosts in the PACCAD clade were not monophyletic since MisSV was basal to the clade of BEP clade viruses in both Rep-RepA and MP-CP trees. Among the remaining PACCAD clade viruses, ChlStrV was in basal position and ESV and ScSV branched together in both viral trees. In the MP-CP tree, insufficient bootstrap support among the other PACCAD clade viruses prevented further comparison. Relationships between viruses with PACCAD host names and the hosts are not straightforward as noted by the off-vertical lines in Figure 4. A comparison of branch lengths among the Rep-RepA, MP-CP and rbcL trees suggests that a uniformly ticking molecular clock should not be supported. The dicotyledonous plants were on a relatively longer branch than the viruses bearing their names. MisSV had diversified more than the BEP clade viruses in the MP-CP region, but the opposite was true in the Rep-RepA region. ChlStrMV also showed anomalous branch lengths. Such anomalies are signs of recombination events after which adjustment of the recombined segments to one another is needed 8. A series of molecular clock tests supported these observations. A clock could be rejected if all viral sequences were considered, regardless of which sequence was chosen as outgroup for the tree to be tested. However, when ChlStrV or the viruses of dicotyledonous plants were omitted with MisSV as outgroup, a clock could no longer be rejected. Operation of a clock was predicted with confidence (p > 0.9) for a dataset containing WDV, BDV, ODV, MisSV and ChlStrV with ChlStrV as outgroup.

Distance trees constructed from an alignment of a subset of rbcL gene sequences and of the respective viruses were analyzed (Figure 5) by TreeMap 4344. The algorithm resulted in reconciliation of the two trees by the addition of two host species jumps by viruses (Figure 5). One was of the lineage that gave rise to the Egypt species of ScSV from a non-sugarcane lineage to the sugarcane lineage. The second jump was of MSV to maize, consistent with the long term existence of the virus in Africa before the arrival of maize in the continent 45. The confidence level assigned to the reconciled tree was p < 0.01. Neither of the two trees had been rooted, though it is known that the host tree has its root between Chloris and Avena branches.

WDV was collected throughout China during field surveys in the growing seasons 2004 to 2006. The 28 isolates described here originated from wheat planted in different agro-ecological areas in China, including the northwestern (Shaanxi and Gansu provinces), northern (Shanxi and Hebei provinces), central (Henan province), and southwestern (Yunan province) areas. All the field isolates were inoculated to the susceptible wheat (Triticum aestivum L.) cultivar Fengkang No. 8 by vector leafhoppers (Psammotetix alienus L.) to increase virus concentration and to allow serological typing or sequencing of polymerase chain reaction (PCR) products. The wheat plants were later tested for WDV with ELISA using an antiserum (Bio-Rad, Marnes la Coquette, France). Leaves were collected from WDV-positive plants displaying typical symptoms of WDV infection, and stored at -80°C. Details of the isolates, their names, provinces of collection, original host plant, and years of collection are shown in additional file [see Additional file 1].

Total DNA was extracted from systemically WDV-infected wheat leaves 50. DNA extracts were used as template for PCR amplification, performed in a 50 μL reaction solution containing 10×Taq Buffer, 2.5 mM dNTP (each), 0.4 mM of the viral sense and complementary sense primers designed according to the conserved sequences of WDV genomes [see Additional file 3], and Ampli Taq DNA polymerase (Applied Biosystems, Foster City, CA, USA). PCR reactions were carried out for 35 cycles, each consisting of denaturation at 94°C for 1 min, annealing at 55°C for 1 min, and extension at 72°C for 1 min, with 95°C for 2 min at the beginning and 72°C for 10 min at the final step. The expected PCR products were 767 bp, 1152 bp and 1041 bp, using the primer pairs of 40F/806R, 735F/1886R and 1828F/118R, respectively, and together covered the entire length of the viral genome. The PCR product segments were electrophoresed in 1.0% agarose gels, bands were excised using a razor blade and purified using the BioTeq PCR quick Gel Extraction Kit (BioTeq, Inc, USA).

Nucleotide sequences of the entire genome of each isolate were determined using the above PCR fragments. The purified fragments were cloned into the pMD18-T vector (Takara, Dalian, China). The plasmids were transformed into Escherichia coli strain JM110 and plasmid DNA was isolated from overnight cultures by alkaline lysis. Insert sequences were determined on at least three clones for each PCR fragment using the dideoxynucleotide chain termination method by an automated sequencer (ABI BigDye 3.1, Applied Biosystems, Foster City, CA). Sequence data were assembled using DNASIS version 3.5 (Hitachi) or BIOEDIT version 5.0.9 51. The nucleotide sequence data have been submitted to GenBank databases and assigned accession numbers EF536859 through EF536886.

Complete genomes of the 28 WDV isolates sequenced in this study and 18 entire sequences of other WDV, BDV and ODV isolates obtained from the NCBI database (National Center for Biotechnology Information, Bethesda, MD, USA) were analyzed. The coding and intergenic regions were annotated by reading frame or following NCBI's annotations. The complete genome sequences of the WDV, BDV and ODV genomes were aligned with CLUSTAL W V.1.8. MEGA V.4.0 52 determined the number of nucleotide substitutions per site (evolutionary distance) between the strains. Phylogenetic trees were constructed by neighbor-joining (NJ), and maximum parsimony (MP) as implemented by MEGA version 4.0 52 and DNAPARS of PHYLIP package version 3.5 53, respectively, based on the Kimura 2-parameter distance matrix model. Bootstrap confidence values were obtained for 1000 replicates (Figure 1). The homologous regions of the genome of an isolate of Maize streak virus (MSV) (NC_001346) 54 were used as the outgroup for these analyses, as BLAST searches had shown them to be the sequences in the international sequence databases most closely related to those of MSV. Treemap 4.1.1 44 was used to test and display the correspondence between plant and virus trees. The Watterson's estimator of θ (θw) 38 and the average pairwise nucleotide diversity Pi(π) 39, were estimated using DnaSP version 4.10.2 55. Also, the program was used to estimate the proportions of synonymous and nonsynonymous substitutions by the Jukes-Cantor one-parameter model.

To evaluate the sequence relationships among mastrevirus genomes, a selection of mastrevirus sequences available at the time was obtained [see Additional file 2]. A manually adjusted multiple sequence alignment based on encoded amino acid sequences was generated using Se-Al 56such that MP-CP and Rep-RepA regions were satisfactorily aligned. These regions were separately excised from the alignment for further manipulation. In cases where multiple sequences were available for the same named virus, a consensus sequence was generated by Se-Al. For examination of host phylogenetic relationships, rbcL sequences were retrieved from GenBank/DDBJ/EMBL. They are identified in the legend of Figure 4. These were also aligned. Aligned sequences were examined using PAUP 57 by testing models supplied by Modeltest 58. The parameters for the best model were used to construct neighbor joining trees as implemented in PAUP. To test for consistency, bootstrapped neighbor joining was also performed using Phylip package programs, Seqboot, DNAdist, Fitch, and Consense 53. Resulting trees were manually manipulated to minimize tangles between host and virus trees. The validity of a molecular clock for several assemblages of sequences was tested using the log ratio test as described by Posada 59. To determine selection models acting on the WDV, BDV and ODV genes, nucleotide substitutions at synonymous (K_s) and non-synonymous (K_a) positions of genes were calculated by DnaSP version 4.10.2 55.

The average frequency with which mastrevirus sequence sites mutate in evolution is unknown. For highly mutable sites the substitution frequency has been estimated 47 at 3 × 10^-4/site during a 60 day growing period for Tomato yellow leaf curl China virus (TYLCCNV). That value is clearly an overestimate of the average frequency.

Nevertheless, since our interest was in determining the relative ratios of divergence times of WDV, BDV and ODV from MSV, of ODV from WDV and BDV, and of WDV from BDV, the number was used to obtain divergence time estimates with MEGA software 52. Host divergence times were obtained from literature. Resulting calculated virus divergence times were normalized to 100 for the (WDV-BDV-ODV)-MSV split and values were plotted against corresponding host divergences. Linear regression was used to evaluate the correspondences and to determine an appropriate conversion factor that could be applied, assuming uniformity of the molecular clock, to the relative divergence times of the viruses.

To investigate the extent of recombination within the data set, the aligned sequences were examined using the Recombination Detection Program (RDP3) 41, GENECONV 60, BOOTSCAN 61, MAXIMUM CHISQUARE 62, CHIMERA 41, SISTER SCAN 42 and and phylpro 63 recombination detection methods as implemented in RDP3 41, (details of program settings available from http://darwin.uvigo.es/rdp/heath2006.zip). The transversion and transition differences of all pairs of sequences were calculated using the discalc program (kindly supplied by G. F. Weiller, Australian National University) and these were compared in diplomo scatter plots 64.