Metagenomic analysis of nepoviruses: diversity, evolution and identification of a genome region in members of subgroup A that appears to be important for host range

The complete nepovirus ORF1 and ORF2 sequences were retrieved from NCBI as of January 2020, our own curated nepovirus sequence repository obtained by analysis of high-throughput or Sanger sequencing datasets, and a selection of Sequence Research Archive datasets from GenBank [19]. In total, 110 ORF1 sequences and 167 ORF2 sequences were used in this study (Supplementary Tables S2 and S3). In addition, sequences of specific domains were retrieved from NCBI (see Supplementary Table S8).

Codon-based multiple sequence alignments and maximum-likelihood (ML)-based phylogenetic trees were prepared using MUSCLE [7], implemented in MEGA7 and MEGAX software [26, 27], excluding the viral untranslated regions (UTRs). The best ML-fitted model for each sequence alignment was used, and nodes in phylogenetic trees were validated by bootstrap analysis (100 replicates). For visualization effects, FigTree v. 1.3.1 was used (http://​tree.​bio.​ed.​ac.​uk/​). The diversity index (π), which is the average number of nucleotide substitutions per site between any two sequences in a multi-sequence alignment, and the variation of π along genome sequences was evaluated by sliding window analysis (length, 80; step size, 20) using DnaSP v.6.12.03 [29] and MEGA X.

A search for potential recombination signals was performed using all seven algorithms implemented in RDP v4.97 (RDP4) [32]. The default settings were used for each algorithm, and only recombination events detected by five or more methods were considered.

Differences in nucleotide sequence diversity of viral populations defined using different modalities were tested by analysis of molecular variance (AMOVA), as implemented in Arlequin v. 5.3.1.2 [10]. AMOVA calculates the Fixation index, F_ST index explaining the between-groups fraction of total genetic diversity. The significance of these differences was evaluated by performing 1000 sequence permutations.

Tajima’s D (D_T) and sliding window analyses were conducted using DnaSP v. 6.12.03 [29] in order to distinguish the viral populations evolving randomly (per mutation-drift equilibrium; D_T = 0) from those evolving under a nonrandom process (D_T > 0: balancing selection, sudden population contraction; D_T < 0: recent selective sweep, population expansion after a recent bottleneck).

Only complete open reading frame (ORF) sequences of RNA1 (ORF1) and RNA2 (ORF2) of nepoviruses were considered in this study. All sequences were retrieved from the NCBI database as of January 2020, our own curated nepovirus sequence repository, and a selection of Sequence Read Archive (SRA) datasets from the GenBank database. Data mining was performed to increase the number of sequences for ArMV, GFLV, and mulberry mosaic leafroll-associated virus (MMLRaV), a novel nepovirus [31], as described previously [19]. These data mining, Sanger, or Illumina sequencing efforts resulted in 46 new sequences (24 for RNA1 and 22 for RNA2) of ArMV, GFLV, and MMLRaV. New sequences were deposited in the GenBank database (Supplementary Table S1). In total, both genomic RNA sequences were recovered from members of 29 nepovirus species, except from olive latent ringspot virus, for which only a single RNA2 sequence but no RNA1 sequence is available (Table 1). Two nepoviruses (GFLV and ArMV) made up the majority of sequences analyzed in this study, while most species were represented by one or a few sequences of either genomic RNAs (Table 1). Novel nepoviruses used in this study included MMLRaV [31], caraway yellows virus [12], potato virus B [4], and red clover nepovirus A [25]. A few new viruses and isolates belonging to the genus Nepovirus have been identified since we last consulted NCBI (January 2020). The corresponding sequences were not included in this study (Supplementary Table S10). In addition, a few viruses described in the literature as potential members of new nepoviral species, such as Hobart nepovirus 3 [42] or Zhuye pepper nepovirus 1 [3], were not taken into account in this study because the sequences were incomplete or discrepancies were observed between the datasets available at NCBI and the publications. Furthermore, only a single sequence was chosen from a group of sequences displaying nucleotide sequence identity higher than 99% unless the isolates were from different hosts and/or different countries. Complete lists of the 110 ORF1 and 167 ORF2 sequences selected for this study are provided in Supplementary Tables S2 and S3, respectively.

Nucleotide sequence comparisons confirmed the classification of nepoviruses into three subgroups with higher inter-subgroup than intra-subgroup mean distance values (Table 2). Subgroup B sequences displayed the lowest maximum pairwise distance values, which were well below the inter-subgroup mean distance values, suggesting a well-defined group of virus isolates (Table 2). The inter-subgroup mean distance value was lower than the maximum intra-subgroup mean distance value for subgroup A and C sequences, revealing a greater variability and less well-defined groups of virus isolates (Table 2). After performing an alignment of ORF1 and ORF2 sequences, phylogenetic trees were constructed by the maximum-likelihood (ML) method, using the best-fit model (GTR+G+I) (Fig. 1). Interestingly, the members of each subgroup were separated better in the tree based on ORF1 than the one based on ORF2 sequences (Fig. 1 and Supplementary Fig. S1). Indeed, the ORF1 nucleotide sequences of virus isolates of subgroups A, B, and C clustered in separate and well-supported clades in a tanglegram (Fig. 1) and in an unrooted cladogram (Supplementary Fig. S1). Nucleotide sequences of nepovirus isolates of subgroup B were also well defined when using ORF2 sequences, but subgroup A and C ORF2 sequences were scattered in different clades in a tanglegram (Fig. 1) or unrooted cladogram (Supplementary Fig. S1). These results suggest that the classification of nepoviruses into distinct subgroups is more robust when based on ORF1 sequences than when based on ORF2 sequences. This finding should be considered by the International Committee on Taxonomy of Viruses (ICTV) Secoviridae Study Group to eventually define new demarcation criteria for nepoviruses when using pangenome information.

Species demarcation criteria for nepoviruses have been defined by the ICTV (https://​talk.​ictvonline.​org/​ictv-reports/​ictv_​online_​report/​positive-sense-rna-viruses/​picornavirales/​w/​secoviridae). These include CP amino acid (aa) sequence identity less than 75% and conserved protease-polymerase (Pro-Pol) region aa sequence identity less than 80%, among other criteria. We assessed whether these two major demarcation criteria are applicable to the corresponding complete ORF1 and ORF2 aa sequences. Some discrepancies with regard to the intra-species aa sequence identity falling outside the species demarcation were observed for PRMV ORF1 sequences (78.99%) and ORF2 sequences of cherry leaf roll virus (CLRV), ToRSV, cycas necrotic stunt virus, and ArMV (below 74.05%) (Supplementary Table S4). These results revealed that analyzing complete ORF sequences may be problematic for the establishment of new virus species and the classification of new genetic variants of members of existing virus species if the current demarcation criteria pertaining to partial genome sequence information were to be applied. The results further suggest that the demarcation criteria for species in the genus Nepovirus should be amended to accommodate pangenome information. In addition, the ORF2 sequence of ArMV isolate Butterbur was a clear outlier among the ArMV isolates with lower identity values at both the nucleotide (Fig. 2, Supplementary Figs. S3 and S5) and amino acid (70.25%, Supplementary Table S4) levels. According to the original report [21], the pathological and serological features of ArMV-Butterbur are unique, and its CP is 504 aa long (as for all GFLV CPs), while all other ArMV CPs are 505 aa long. These features underscore the need for additional work to ascertain the taxonomic position of ArMV-Butterbur and its recognition as an isolate of ArMV, in particular since no RNA1 sequence is available.

Similarly, the ORF1 aa sequence identity between some isolates belonging to different species was higher than 80%, for example, for beet ringspot virus (BRSV) and tomato black ring virus (TBRV), as well as for BRSV and artichoke Italian latent virus (AILV) (Supplementary Table S4). This high level of sequence similarity could also explain the large number of inter-species recombination events identified between members of these particular species (see the dedicated section below). However, inter-species diversity was below the species demarcation level (< 75%) for ORF2 sequences, unambiguously defining BRSV, TBRV, and AILV as members of different species (Supplementary Table S4). One particular case of interest is grapevine deformation virus (GDefV) [20], a subgroup A nepovirus. GDefV ORF2 aa sequences display 73 and 71% identity to those of GFLV and ArMV, respectively [16], and GDefV ORF1 aa sequences have higher identity to those of GFLV (86-89%) than to those of ArMV (73-74%) [8]. According to the species demarcation criteria for nepoviruses, GDefV would be classified as a highly divergent variant of GFLV when focusing on the ORF1 sequences but as a member of a new species based on ORF2 sequences.

Putative intra-species recombination events have been extensively reported for nepoviruses, mostly in the GFLV RNA2-encoded MP and CP domains [34, 35, 38, 49, 52‐54]. Recombination events have also been described for ToRSV [56] and grapevine chrome mosaic virus (GCMV) [5]. In addition, many inter-species recombination events have been detected, mostly between ArMV and GFLV [9, 34, 35, 54], but also between GCMV and TBRV [5, 28]. With the use of HTS and the recovery of complete virus genome sequences, recombination events can be detected all along the two genomic RNAs [18]. Here, we used the same corpus of nepovirus sequences and searched for potential recombination events using the RDP4 program. Recombination events were only considered when predicted by five or more algorithms with P-values < 10^-3 (Table 3, Supplementary Tables S5 and S6).

Potential intra-species recombination events were identified in ORF1 and ORF2 sequences, mostly of subgroup A members (Table 3, Supplementary Table S5 and Supplementary Fig. S2). Almost twice as many putative recombination events were predicted in ORF2 sequences than in ORF1 sequences (Table 3). Both of these observations most definitely reflect the total number of sequences being recovered and used in this study. Many recombination events were predicted in GFLV and ArMV sequences, with some hotspots, i.e. more than one putative recombinant per site (Supplementary Table S5 and Supplementary Fig. S2). In addition, putative recombination events were also identified for the first time for AILV, CLRV, and raspberry ringspot virus.

All inter-species recombination events predicted in this study strictly involved members of the same subgroup (Supplementary Table S6). Surprisingly, the number of inter-species recombination events was higher than the number of intra-species recombination events within ORF2 sequences (Table 3). For example, 31 inter-species and two intra-species recombination events were detected for subgroup B ORF2 sequences (Table 3 and Supplementary Fig. S2). It should also be noted that all putative ORF1 recombination events detected for subgroup B involved members of different species (Supplementary Table S6). In contrast, all subgroup A recombination events that were predicted involved only ArMV, GFLV, and GDefV (Fig. 1 and Supplementary Fig. S1), again emphasizing their kinship. Recombination may have been facilitated for these three viruses because they have the potential to co-exist in grapevine, a common host, for long periods of time, thus increasing the likelihood of an potential encounter in the same host cell.

ArMV is a ‘generalist’ with a very broad natural host range, including winter barley, narcissus, Ligustrum vulgare, weeds, hops, berries, olive trees, apricot trees, and grapevines, among other species [14, 33, 40]. Our data mining efforts resulted in the retrieval of 17 complete ORF1 sequences from seven monocotyledonous plants and 10 dicotyledonous plants, as well as 21 complete ORF2 sequences from eight monocotyledonous plants and 13 dicotyledonous plants (Supplementary Tables S1, S2, S3, and S8).

The overall genetic diversity (π) of ArMV ORF1 and ORF2 was 0.166 ± 0.003 and 0.133 ± 0.003, respectively (Table 4). As observed previously [57], the coding region 2A is the most divergent genomic region, showing the highest diversity at the extreme 5’ end of ORF2 (Fig. 2), mostly due to size differences among isolates. For ArMV ORF1, the extreme 3’ end is the most divergent genomic region. A comparative analysis of ArMV sequences obtained from mono- and dicotyledonous plants revealed a significantly higher diversity in sequences from isolates infecting dicotyledonous plants compared to isolates infecting monocotyledonous plants (0.170 ± 0.003 vs. 0.109 ± 0.003 and 0.145 ± 0.003 vs. 0.093 ± 0.003, for ORF1 and ORF2 sequences respectively; Table 4). When looking at the evolution pattern (Tajima’s D) of ORF1 sequences (Fig. 2), values were negative but close to 0 (D_T =  – 0.344; P > 0.1), suggesting that the population of ArMV is evolving as per mutation-drift equilibrium with no specific region under selection. On the other hand, two distinct regions of ORF2 sequences were under selection with an overall D_T value of  – 0.994 (P > 0.1) (Fig. 2). One of these two regions covers an aa stretch between two proline-rich segments of the central coding region of the 2A domain. The other region is a specific segment of the CP coding region that overlaps the previously defined R4 region, which is involved in specific transmission of ArMV by the nematode vector Xiphinema diversicaudatum [47].

In a previous study [57], ArMV isolates were separated by the size and aa sequence identity of protein 2A into four groups (I to IV). Here, we recovered 43 ArMV 2A nucleotide sequences from GenBank (Supplementary Table S8) and confirmed the existence of three major clades corresponding to groups II, III, and IV (Supplementary Fig. S3). Group I was composed of a single sequence located within the group II clade. The sequences belonging to each group were genetically different with a high fixation index (F_ST ≥ 0.530) and strong statistical support (P ≤ 10^-5) (Table 5). However, the size of the 2A domain was not linked to the plant host, with ArMV isolates from grapevine belonging to all four groups. A comparative analysis of 2A coding sequences from mono- and dicotyledonous plants documented a statistically supported genetic differentiation (F_ST) (Table 5). Genetic differentiation according to mono- and dicotyledonous plants was also observed when looking at other RNA1 or RNA2 coding region sequences or the complete ORF1 and ORF2 sequences (Table 5, Supplementary Tables S7, S8 and Supplementary Fig. S3). Distinct F_ST values between mono- and dicotyledonous plants were also found at lower cladistic levels, strongly suggesting a likely genetic bias based on the plant host (Supplementary Table S7). Interestingly, similar results have been reported for CLRV, another generalist virus within the genus Nepovirus for which a host-species-dependent population structure was documented using only a short 375-bp sequence corresponding to the extreme 3’ part of the 3’ untranslated region [39].

GFLV primarily infects Vitis spp., making the virus very specialized to this woody plant. The overall nucleotide sequence diversity for GFLV ORF1 (π = 0.127 ± 0.002) and ORF2 (π = 0.130 ± 0.005) sequences was very similar (Table 4). Plotting π along ORF1 sequences (Fig. 3) showed a highly divergent region at the 3’ end of the Pol domain. This result is consistent with other analyses of this particular aa stretch of P1, which was predicted to form an α-helix [18, 36]. Another highly polymorphic region was detected at the extreme 5’ end of ORF2 (Fig. 3), corresponding to a region where intra- and inter-species recombination events have been predicted (see above section and [54]). On the other hand, similar to ArMV, a significant drop in nucleotide sequence diversity is observed within a segment of 2A sequences located between two highly conserved proline-rich regions. The evolution of this particular domain of ORF2 sequences is not neutral, with statistical D_T values well below 0 (Fig. 3), indicating conservative selection with regard to the remainder of the ORF2 sequence. Similarly to ArMV, the same trend was observed for the R4 region of the CP domain [47]. Interestingly, these two regions, which display the lowest D_T values, suggesting a recent selective sweep, were mostly located in sequences recovered from grapevines from the New World (Supplementary Fig. S4). Regarding the evolution pattern of GFLV ORF1 sequences, values were negative but very close to 0 (D_T =  – 0.758), with two sites under selection (P > 0.1). The first site is located at the extreme 5’ end and the second site is positioned within the Hel domain (Fig. 3). When looking at the evolutionary pattern of P1 and P2 (dN-dS), most of the codons were under negative or neutral pressure (data not shown), as described previously [51].

No major differences were observed when separating GFLV sequences by geographic region (France vs. the rest of the world or Old vs. New World), with very similar π and D_T values (Table 4 and Supplementary Fig. S4). However, a genetic structuration between geographic regions was observed, although the F_ST values were extremely low (Supplementary Table S9), with differences in the evolution pattern of GFLV isolates from different parts of the world. This was even more noticeable when separating ORF2 sequences by continent (i.e., Europe, Americas [combining North and South America] and Asia [Far East, Turkey, and Russia]). All F_ST values were statistically supported (P < 0.001). However, the disparity in F_ST values indicated that European and American GFLV variants were more closely related to each other than to the Asian variants. This observation was confirmed when grouping sequences into seven countries or specific regions of the world (France, Slovenia, Italy, USA, Chile, Far East and Switzerland). Some F_ST values were very high, underlying a strong genetic structuration among regions of the world, as confirmed when comparing sequences from the Far East (Iran and China) with other regions of the world (Table S9). This genetic differentiation according to Far East GFLV populations was previously described using GFLV MP sequences [49]. Altogether, these observations suggest a specific geographic evolution and genetic structuration of the virus.

ArMV and GFLV are very closely related but belong to different species (Fig. 1). They share many characteristics such as hosts (i.e., grapevine), closely related vectors (Xiphinema spp.), similar symptomatology, and many natural inter-species recombinants. While genetically different (Supplementary Table S4, Figs 4 and 5), similar patterns in their respective genetic diversity were observed along ORF1 sequences, especially when separately analyzing sequences from Vitis-infecting ArMV isolates from non-Vitis-infecting ArMV isolates. As detailed above, one of the hallmarks of GFLV is a higher π value at the C-terminus of Pol. A higher π at the C-terminal end of Pol was also clearly identified in Vitis-infecting isolates, but not in non-Vitis-infecting ArMV isolates (Fig. 4). Such specific increased genetic diversity in Vitis-infecting ArMV and GFLV isolates was not due to the overlap of a hidden ORF (Supplementary Fig. S6), as described for sobemoviruses [30]. This diversity was also observed at the aa level, with a percent identity higher than 80.41% in the case of non-Vitis-infecting ArMV isolates, but as low as 65.54% and 56.08% for Vitis-infecting ArMV and GFLV isolates, respectively (Fig. 4). Such high divergence was not found when specifically looking at the first 148 aa of the Pol domain, where the sequence identity was above 82%. While highly divergent between ArMV and GFLV (Supplementary Fig. S7), the Pol C-terminus has only two amino acids that are mostly conserved between Vitis-infecting ArMV and GFLV isolates, at position 683 and 746 (Supplementary Fig. S8). Could these two residues be implicated in host adaptation mechanisms? More work is needed to address this hypothesis.

The ORF2 sequences of ArMV and GFLV have many characteristics in common (Fig. 5). For example, higher genetic diversity is detected at the 5’ end of 2A, partly as a result of indels. However, major differences between these viruses were found when focusing specifically on the CP domain, with a clearly different level of genetic diversity in the R4-R5 region between ArMV and GFLV (Fig. 5 and Supplementary Fig. S5). This region is important for vector transmission [47]. When looking at the evolution pattern, most of the ORF2 sequences seem to be evolving randomly, while two regions display a non-random evolution pattern. One of these two regions corresponds to the 2A domain, and the other to the R4-R5 region within the CP domain, both showing strong constraints.