Molecular evolution of two asymptomatic echovirus 6 strains that constitute a novel branch of recently epidemic echovirus 6 in China

Stool samples of children were collected from three kindergartens in Kunming city, Yunnan province of China. The three kindergartens are located in the east, north and south of Kunming. The children were asymptomatic for virus infection and did not present with fever, cough, diarrhea or rash in the 2 weeks before and after sample collection. These children did not develop any diseases known to be caused by EV infection and did not have contact with patients recently infected by EVs. The children’s ages ranged from 3 to 6, and stool samples were collected between June and August in 2013.

The stool samples were collected and processed according to standard procedures recommended by the World Health Organization (WHO) [13]. RD (human rhabdomyosarcoma cells), KMB17 (human embryonic lung diploid fibroblast) and A549 (human lung cancer cells) cell lines were used to isolate the viruses [14]. Within three passages, samples that induced a cytopathic effect (CPE) were considered positive, and the supernatants were harvested and stored at −80 °C.

The Viral RNA was extracted from cell culture supernatants using a QIAamp Viral RNA Mini Kit (QIAGEN, USA). RT-PCR was carried out using the PrimeScript One Step RT-PCR Kit Ver. 2 (TAKARA, China). Primer pairs 222 and 224 were used to amplify the partial VP1 gene [15], and primers for amplifying the complete genome were designed based on the published sequence of the D’Amori strain, and some primers were designed using the “primer-walking” strategy. The positive PCR products were sequenced at BGI Sequencing Company (Beijing, China). The whole genomes of these two strains were submitted to the Enterovirus Genotyping Tool v0.1 [16] for classification.

Viruses belonging to the same serotype as the newly identified E6 strains were chosen to construct the phylogenetic trees, and the sequences of these viruses were retrieved from GenBank. Their entire coding region of VP1 genes were used to construct the VP1 phylogenetic tree. For construction of the coding sequences (CDS) tree, the Enteroviruses that did not belong to the E6 serotype but whose gene was highly similar were also included. These non-E6 strains were selected by using each gene of K727 strain to compare with the sequences of EV database by BLAST. If the non-E6 strain contained a gene which has above 90% similarity with K727, this non-E6 strain was selected. The whole CDS sequences of these EV strains were aligned using Geneious 4.8 software [17]. Phylogenetic analyses were conducted using Molecular Evolutionary Genetic Analysis (MEGA) version 6 software [18]. The modified Nei-Gojobori method with a fixed Transition/Transversion ratio (0.5) was used to setup the model because the incidence of synonymous mutations and nonsynonymous mutations were significantly different in these viruses [19]. The neighbor-joining method was used to build the trees using 2000 bootstrap replicates. The nucleotide similarities among the strains were plotted with Simplot software version 3.5.1 [20], using a sliding window of 400 nucleotides moving in steps of 20 nucleotides by the Kimura 2-parameter method. The permuted tests were performed with the same software.

For discerning selection, a likelihood ratio test (LRT) [21] was used to compare the differences between non-synonymous and synonymous substitution rates. Non-synonymous mutations are much more likely to have an effect on fitness than synonymous changes. Thus, the ratio (ω = dN/dS) of the number of nonsynonymous substitutions per nonsynonymous site (dN) and the number of synonymous substitutions per synonymous site (dS) provides a sensitive measure of selective pressure at the protein level, with ω values <1, = 1, and >1 indicating purifying selection, neutral evolution, and positive selection, respectively [22]. This test was performed using CodeML of the PAML 4 software [23]. The phylogenetic trees required for the analysis were generated using MEGA6 via the neighbor-joining (NJ) method and the modified Nei-Gojobori method. The SITE, BRANCH and BRANCH-SITE models were all used in the analyses. The CodonFreq was set as F3X4, and the Small_Diff was set as 0.45e-6 in the CodeML model.

After culturing 96 stool specimens of healthy children with RD (human rhabdomyosarcoma cells), KMB17 (human embryonic lung diploid fibroblast) and A549 (human lung cancer cells) cell lines, two E6 strains were isolated. One was named K727/YN/CHN/2013 (abbreviated to K727), which was isolated from a 4.5-year-old boy using the KMB17 cell line. The other one was named K843/YN/CHN/2013 (K843), which was also isolated from a 4.3-year-old boy using the KMB17 cell line. The whole genomes of these two strains were sequenced. When the two sequences were submitted to the Enterovirus Genotyping Tool v0.1 [16], they were both classified into the E6 serotype with 100% bootstrap support. In addition, one Coxsackievirus A24 strain, 8 strains of Echovirus 12, one Echovirus 20 strain and one Enterovirus C99 strain were also isolated from these stool specimens.

Three phylogenetic trees were constructed with the Neighbor-joining method using the modified Nei-Gojobori model [19]. One tree was built based on the VP1 sequences (Additional file 1: Figure S1). The VP1 sequences of 279 E6 strains were used to construct the tree, including K727, K843 and most E6 strains known in China. The K727 and K843 strains clustered together, and clustered with several KJ7724XX strains with 95% bootstrap support. These KJ7724XX strains were found in the AFP patients in Shangdong (a north province of China) in 2011, and the E6 strains were predominant in Shangdong in that year. In this branch, the other E6 strains in China all were found after 2011. Such as: KT633557 and KT633559 strains were found in Zhejiang province (central China) in 2013, LC120880 was found in Yunnan province (southwest China) in 2014 and KF487210 was found in Guangdong province (southern China) in 2012.

Another two trees were built using viral CDS of capsid coding region and 3D region separately (Fig. 1), because a potential recombination occurred in the 2C gene of K727 and K843. This tree included all E6 strains that have a complete CDS in GenBank as well as EV strains whose genes have high nt similarity with K727 and K843 (See Method). K727, K843 and E6SD11CHN clustered together as a novel branch with a bootstrap value of 99% for both trees. The capsid coding region of these three strains clustered with the E6 strains, while their 3D regions clustered with the non-E6 strain. So it looks like a recombination between EV types happened. Additionally, another E6 strain, E6/P735/2013/China, is showed the similar pattern.

A similarity comparison was plotted using simPlot 3.5.1 software [20], and the result from using K727 as a query is shown in Fig. 2a. Seventeen representative strains whose genes were most similar to K727 were selected in the comparison because if too many strains are used, the colors of some strains were too similar to be easily distinguished. In Fig. 2a, the region before the 2C gene of K727 is similar to that of the E6 serotype, and most E6 strains and K727 are more than 90% similar. In contrast, the region after the 2C gene of K727 has more than 90% similarity to non-E6 strains. The similarities of this region between most E6 strains and K727 are under 90%, which suggests a potential recombination in the 2C gene of K727.

Two E6 stains, E6SD11CHN and E6/P735/2013/China, which clustered with K727 and K843 in the both phylogenetic trees, were also highly similar to K727 on the whole genome level, especially E6SD11CHN (red line in Fig. 2a). Most of its genes had more than 95% similarity to K727 except for 3AB and 3C.

A bootscanning analysis was also performed on these strains. (E6SD11CHN was excluded because its high similarity could disrupt the analysis.) The result is shown in Fig. 2b. A significant change in the permutation percentage was observed in the middle of the 2C gene. Before this recombination site, strains K727 appears to share a common ancestor with another E6 strain (FJLY2010327). However, after this point, the K727 sequence becomes similar to a coxsackievirus A9 strain (CVA9-Alberta-2010), suggesting that a potential recombination occurred in the 2C gene of K727. The results of the similarity and recombination analyses for K843 were similar to those of K727 and can be found in Additional file 1: Figure S2.

Likelihood ratio tests (LRTs) were used to compare the different evolutionary scenarios with preset parameters to determine whether positive selection occurred. These were performed using PAML 4 software [23]. The strains which were used to construct the phylogenetic trees were used to perform the test. To begin with, two scenarios were envisioned. One (scenario 1) assumed that the positive selection occurred at some sites of the gene (positive selection model M2a), while the null model was M1a (nearly neutral), which assumed that there were no positive selection site in the gene, and only two kinds of sites in proportions p0 and p1 = 1 − p0 with 0 < ω0 < 1 (under purifying selection) and ω1 = 1 (neutral sites) [24]. The log-likelihood values, p and ω of these two models, were calculated using the SITE model of PAML 4, and the results are listed in Table 1. For every gene, the log-likelihood values of these two models were identical. Therefore, no positive selection site was found when the whole group was used. Actually, we found that most of the virus’ genes were under a strong purifying selection. The ω0 (Table 1) of these genes were only 0.006 to 0.022. Only 0.6 to 3.8% (p1) of the amino acids were nearly neutral (ω1 = 1), and they are also listed in Table 1.

Positive selection often acts on only a few sites and only for a short period of evolutionary time. Scenario 1 tested positive selection in all individuals of every branch, and the signal may be swamped by ubiquitous negative selection. Scenario 2 was considered because we wanted to determine whether adaptive evolution occurred in the asymptomatic E6 strains. The hypothesis in scenario 2 is that the ω changed when the asymptomatic branch (KM branch clustered with K737 and K843 in Fig. 1) grew, and this model was called the Branch (ω changed) model. Its null model was the M0 model (one ω), which assumed that ω was unchanged in the whole tree. The log-likelihood values, p and ω of these two models, were calculated using the BRANCH model [25] of PAML 4, and the results are listed in Table 1. The two models’ log-likelihood values for VP4, VP2, VP1, 2B and 3D were almost the same, so the ω for these genes in the KM branch were not significantly different. The log-likelihood values of VP3 and 2A have slight differences, but they are not significant. For example, the VP3 gene has an LRT statistic of 2Δℓ = 2 × ((−3351.5) - (−3353.0)) = 2 × 1.5 = 3, with a P of 0.08 and a df = 1 using the chi-square test [26]. On the contrary, the LRTs of 2C, 3AB and 3C genes had significantly different ω values on the KM branch. The LRT statistic of 2C was 2Δℓ = 2 × ((−6882.4) - (−6889.1)) = 2 × 6.7 = 13.4, and P = 2.5 × 10⁻⁴ with df = 1. The LRT statistic of 3AB was 2Δℓ = 2 × ((−2348.1) - (−2356.4)) = 2 × 8.3 = 16.6, and P = 4.6 × 10⁻⁵ with df = 1. The LRT statistic of 3C was 2Δℓ = 2 × ((−3923.8) - (−3929.2)) = 2 × 5.4 = 10.8, and P = 0.001 with df = 1. We also noticed that the ω of the KM branch was 8 to 20 times the ω of the whole tree, which may imply that evolution accelerated when the asymptomatic branch developed.

Since the ω increased on the KM branch for 2C, 3AB and 3C genes, a test was performed to determine whether there were sites under positive selection along this lineage in these three genes. Thus, scenario 3 was set up. The branches of the tree were divided a priori into foreground and background lineages, and their evolution parameters were set according to the branch-site model A, suggested by Yang [27]. The hypothesis (H1) in scenario 3 is that there were some sites under positive selection when the KM branch (foreground lineage) formed, and its null model (H0) assumed that positive selection did not happen along this lineage, in other words, there is only one ω for the whole tree. The simulations were conducted using the BRANCH-SITE model in PAML 4, and the results are listed in Table 2. Unfortunately the log-likelihood values between H0 and H1 were almost identical, and we did not find sites under positive selection along this lineage for these three genes. However, we did find that some sites that were under a strong purifying selection became neutral on this branch. The ω (ω2a in Table 2) of these sites increased from 0.01 or 0.02 (foreground) to 1 (background). The sites with altered ω values were tested using a Naive Empirical Bayes (NEB) method with probabilities above 90% and are listed in Table 2. Eleven were only found in the genes of the K727 and K843 strains and are shown in bold.