Viral metagenomics of aphids present in bean and maize plots on mixed-use farms in Kenya reveals the presence of three dicistroviruses including a novel Big Sioux River virus-like dicistrovirus

The aphid samples were collected from a total of nine sites over four farms in four bean growing agro-ecological zones in Kenya based on different altitudes in meters above sea level (m asl) and known climatic conditions. These were Ndeiya (Humid, highland), Oloirien (semi-arid, highland), Katumani (semi-arid, lowland) and Kaiti (Humid, lowland) (Fig. 1). Apart from the Katumani farms where beans were grown exclusively, all other farms had beans intercropped with a variety of other food crops. Aphid samples were collected in November 2014, which coincided with the October–November planting season.

Maize leaf samples were collected from 14 sites in three counties in Kenya (Fig. 1). These were collected over the months of June–August from Baringo (semi-arid, highland), Machakos (semi-arid, lowland), Kitui (semi-arid, lowland) and Nakuru (humid, highland) (Additional file 1: Supplementary Table S1).

To trap winged (alate) aphids, five yellow bowls (22 cm diameter) containing approximately 300 ml of 1% (w/v) SDS mixture were set up at the corners and the center of a 10x10m area within a bean plot. These traps were placed on stilts raised to the same height as the bean plants and left for two hours during the warmest part of the day (approximately 12:00 to 14:00 East Africa Time). Aphids were recovered from traps using plastic microbiological inoculation loops (VWR International bvba, Leuven, Belgium) and transferred individually to 1.5 ml micro-centrifuge tubes containing 500 μl of RNALater (Thermo Fisher Scientific, Waltham, Massachusetts, USA). For collection of wingless (apterous) aphids, bean plants within the proximity of the traps were selected at random and examined for wingless aphids for capture and preservation in RNALater. Aphid samples were transported in racks and stored at 4 °C for up to 4 days before processing. Apterous aphids were collected at all four sites (Fig. 1). Alate aphids were available only at the Ndeiya site (Fig. 1).

Cytochrome oxidase subunit 1 (CO1) gene sequencing using template DNA from individual aphids was conducted using the aphid-specific primers, Favret F and Favret R (sense 5′-ACC AGT TTT AGC AGG TGC TAT TAC-3′ antisense, 5′-GTA TAT CGA CGA GGT ATA CCA TTT-3′) using DNA from individual aphids as template (S16-S45). The primers bordered a 700-base pair (bp) region of the CO1 gene [19]. Purified PCR products were sequenced in both directions by automated Sanger sequencing [20, 21].

Thirty-nine libraries were prepared for sequencing using the Illumina MiSeq system. Of these libraries, nine consisted of pools of ten aphids taken from the area sampled. The remaining 30 were made from individual aphids caught in traps. Libraries were prepared using the TruSeq RNA v2 kit (Illumina, San Diego, CA) protocol as per the manufacturer’s instructions. Libraries were normalised, pooled and sequenced in two runs using the Illumina MiSeq system. Paired end reads were generated using the 2 × 250 cycle V2 kit on the MiSeq.

RNA was extracted using Trizol (Ambion) as per manufacturer’s instructions. Ribosomal RNA (rRNA) depletion was performed with Ribo-Zero Magnetic Kits (Plant Leaf – Epicentre), and depletion confirmed on a pre-cast 6% TBE-urea gel (Novex, Life Technologies). Indexed stranded libraries were constructed using Scriptseq V2 RNA-Seq Library Preparation kits (Epicentre) and Scriptseq Index PCR primers (Epicentre). Purification steps were performed using Agencourt AMPure XP beads. Library quantity and quality were checked using Qubit (Life Technologies) and on a Bioanalyzer High Sensitivity DNA Chip (Agilent Technologies). Libraries were sent to Beijing Genomics Institute for 100 bp paired-end sequencing on one lane of a HiSeq 2000 (Illumina).

The quality of reads generated was checked using FastQC (available online at: www.​bioinformatics.​babraham.​ac.​uk/​projects/​fastqc/​). The reads were trimmed to attain optimum quality using the dynamic trim function of SolexaQA [22] and FASTX-Toolkit (http://​hannonlab.​cshl.​edu/​fastx_​toolkit/​). A de novo assembly of the reads was done using Trinity (http://​trinityrnaseq.​github.​io/​) based on the de Bruijn graph algorithm. The resulting contigs from de novo assembly were analysed by BLASTn against a virus database to detect all viral sequences present in the data. In the Virfind online analysis pipeline, the sequences were trimmed (default setting ~5 bases). De novo assembly was performed using Trinity (v.2.0.2) but where the average sequence length was ≤80 nucleotides, de novo sequence assembly was done by Velvet with kmer = 15, 19, 23, 27, to generate sequences ≥90 nucleotides. Where the average sequence length ≤ 40 nucleotides, de novo sequence assembly was done by Velvet with kmer = 11, 13, to generate sequences ≥90 nucleotides. Assembled contigs were subjected to a BLASTn search in NCBI at the default e-value (0.01) to generate annotated sequences. All sequences not detected as ‘virus’ by BLASTn were subjected to a BLASTx search against all GenBank virus proteins with the e-value at the default (0.01). For the maize samples, libraries were de-multiplexed allowing one error within the index sequence and de novo assembly done with Trinity using a custom script.

To confirm the presence of ALPV, RhPV and BSRV-like viruses in the pooled samples, cDNA was made by reverse transcribing 500 ng of total RNA extracted from aphids in Pools 1–9 using GoScript First Strand synthesis system (Invitrogen). Random hexamers were used as primers in a total volume of 20 μl and reverse transcription done at 42 °C for 50 min as recommended by the manufacturer. Primers for ALPV based on the KE Aphid P7 and KE Aphid P9 sequences were ALFWD1 5′-CAA AAC AAT ACA AAA TGTC-3′ and ALRV1 5′-GGT TTG TTT AAA ATC GTT GCC-3′ targeting a 511 bp region corresponding to the sequence positions 511 and 1022 in ORF1 of the ALPV genome. The primers for BSRV-like sequences were based on sequence S37 and were: BSRV1 FWD-5′-ACA ATT TAT ATC GTT TAG GTT-3′, and BSRV1 REV-5′-TTA CTA AGG TTT AAA TCT TTA-3′ amplifying a 511 bp region between (approximate 511–1015) in ORF1 of the S42 BSRV genome. For ORF2, primers BSRV2 FWD-5′-GTC AAA ACT AAA TTT CAT TCA-3′ and BSRV2 REV-5′-GG TGT AAT CAT GTG AAA TCT T-3′ were used targeting a 555 bp region (approximate positions 8467–9022). Primers for RhPV detection in aphid samples were RhPVF1 5′-GCA AAC TCA GTA TCT TCA GC-3′ and RhPVR1 5′-TTT GAT TTA TGG CGT GGT GG-3′, which have previously been used in RhPV detection [23]. For RhPV detection from maize samples, custom primers were designed from the assembled RhPV sequence from sample T2F2S4 (sequence available in Additional file 2: Text file S1). The two primer pairs RPV_200_F1–5′-TTT GGA AGA CGT GTG CGAGA-3′/ RPV_1100_R1–5′-TCG TGC AGC TG GAA CGAAT-3′ (approximate positions 228–1016) and RPV_1350_F1–5′-GAT GGG TAC ACT GGA CAG CC-3′/ RPV_ 2300_R1–5′-CTC TCG CTC GCA GCA AATTC-3′ (approximate position 1401–2236) targeted 789 bp and 836 bp long regions of the assembled sequence, respectively. The primer sequences and sequences from the amplicons generated are shown in the additional information (Additional file 3: Table S2).

For PCR, 1 μl of cDNA solution was used in the reaction using the various primers and Biomix Red® PCR premix (Bioline, London, UK) in a 20 μl reaction. Cycling conditions for all primers were 3 min at 94 °C followed by 35 cycles of 30s at 94 °C/ 30s at 55 °C/ 1 min at 72 °C and a final extension of 7 min 72 °C. PCR products were analysed on a 2% agarose gel in 1X TBE containing Gel Red and electrophoresed in 1X TBE buffer. PCR products were purified and sequenced in both directions by automated Sanger sequencing [20, 21] at the Biosciences eastern and central Africa Hub at the International Livestock Research Institute in Nairobi, Kenya or Source Biosciences (Cambridge UK).

Multiple sequence alignments were done using MUSCLE [24] in Molecular Evolutionary Genetics Analysis version 6.0. (MEGA 6) [25]. The evolutionary history was inferred using either the Neighbor-Joining method [26]. ORF Finder (https://​www.​ncbi.​nlm.​nih.​gov/​orffinder/​) was used for ORF prediction. To annotate the proteins encoded by the ORFs, an online search was conducted in the NCBI conserved domain database (https://​www.​ncbi.​nlm.​nih.​gov/​Structure/​cdd/​cdd.​shtml) for the location of conserved domains.

The percentage differences between the nucleotide sequences of the isolates was done by doing a multiple sequence alignment using MUSCLE in the Sequence Demarcation Tool software version 1.2 [27].

Analyses to detect recombination was done using RDP4 software package. Alignments and analyses were performed for ORF 1 and ORF 2 separately. Default parameters were used for the RDP [28], GENCONV [29], MAXCHI [30], 3Seq [31], SiScan [32] and CHIMAERA [33] methods selected for the analysis. A Bonferroni corrected p value of 0.05 was considered significant. Evidence of recombination was accepted if it was supported in at least three different methods with a p value >10⁻⁶.

Aphids were trapped in bean fields at the locations shown in Fig. 1. Nucleic acids were extracted from aphids for PCR-based aphid species identification and for Illumina sequencing of RNA. Following PCR of aphid DNA using primers specific for CO1, a basic local alignment search tool (BLAST) [34] search of the generated CO1 gene sequences showed that the samples were all A. fabae. The CO1 sequences generated from Sanger sequencing are available in the additional files (Additional file 4: Text S3). Confirmatory results were provided by analysis conducted using Virfind online bioinformatics platform (http://​virfind.​org/​j/​virfind-pipeline) [35]. This analysis yielded longer CO1 gene sequence reads than from the automated Sanger sequencing and likewise identified the sequences as originating from A. fabae (Additional file 5: Text S4). A summary of results of the CO1 gene identification using the BLAST tool using sequences from Sanger sequencing and from Virfind are shown in the additional files (Additional file 6: Table S3).

RNA isolated from aphids sampled in bean plots and isolated from maize leaves was subjected to Illumina sequencing. Two sequence analysis pipelines were used to analyse the sequence data from the pooled and individual aphid samples (see Materials and Methods). Results from analysis of the assembled sequences revealed the presence of three known or putative dicistroviruses: ALPV, RhPV, and a BSRV-like virus in both pooled and individual aphid samples. The samples where these viruses were detected are shown in Table 1.

Three aphid samples (P7, P9 and S45) and sample T2F3S4 from maize had ALPV sequences approximately 9.6 Kb in length (Table 1). This genome size was comparable to other full-length ALPV genomes in GenBank (Table 2). Sequences from samples P7 and P9 were mapped against an ALPV genome from GenBank (reference number NC004365) in CLC workbench version 5.1. to get deeper sequence coverage (Fig. 2). The upper panel in Fig. 2 shows diagrammatically the arrangement of putative functional domains in the proteins encoded by the genome but showing only those predicted with highest confidence. These two sequences from aphid samples (P7 and P9) were renamed KE (Kenya) Aphid P7 and KE Aphid P9 while sample T2F3S4 from maize was renamed KE Maize in further analyses described in this article. ALPV sequences were also detected in two other maize samples (T1F4S3 and T1F5S2) but these were partial genomes and were not used in subsequent analyses. The ALPV sequences from these samples are available in the additional materials (Additional file 7: Text file S4).

One maize sample, T2F3S4, contained a 6816 bp sequence annotated as RhPV (Additional file 2: Text file S1). Additionally, there were shorter (~550 bp) assembled sequences annotated as RhPV from aphid samples (Additional file 8: Text file S5). By comparison, the full-length RhPV genomes available in GenBank are 10Kb in length. For the purposes of this study, RhPV sequences were not analysed further except by confirming the presence of the viral RNAs using RT-PCR and sequencing of the DNA products.

Eleven samples had sequences ranging from 9640 to 10,281 nucleotides long annotated as ‘Big Sioux River virus’ following de novo assembly (Table 1). The variation in the length of the genomic RNA sequences was mainly due to partial sequence assembly of the 3′-untranslated region. Of the 11 sequences, sample S37 (GenBank number KY933255) was the longest assembly at 10,281 nucleotides long including a 15 nucleotide polyA tail and is most likely the most complete of the BSRV-like genomes assembled. Though GenBank contained no full-length genomes for BSRV to allow complete sequence comparisons, the available sequences were sufficient for authentication of the BSRV-like virus RNAs. A list of other virus-like sequences detected from the aphid samples following BLASTx are in the additional information (Additional file 9: Table S4).

We used RT-PCR and DNA sequencing to confirm the presence of the three putative dicistroviruses. For the aphid samples, cDNA templates were synthesized using RNA from samples P4, P5, P6, P7 and P9 and virus-specific primers used to confirm presence of the three viruses (Additional file 3: Table S2). PCR products generated using ALPV- and BSRV-specific primers from the pooled aphid samples migrated to approximately 500–550 bp on agarose gels, which was within the expected size range of the primers used (Additional file 10: Fig. S1). A BLAST search for homology confirmed that the amplified sequences had similarity to ALPV and BSRV-like viruses. The identity of RhPV was confirmed in maize samples by RT-PCR and Sanger sequencing (Additional file 3: Table S2; Additional file 11: Fig. S2). However, we were unable to confirm presence of RhPV from the aphid samples by RT-PCR using either the custom primers used to detect the virus in the maize sample or by primers used by other groups [23].

To predict the domains encoded by the ORFs, analysis of the ORF sequences was conducted in the NCBI conserved domain database. Results revealed the predicted regions coding for the helicase and RdRp domains in ORF1 and four capsid proteins in ORF2. The position of the protease domain was determined by searching for the conserved protein sequence ‘VVV QNR GSY TYH AVT FFG DCG SIL IAS NAA ITQ KIM GMH IAG ITH MNKG’ for this region [36] at residues 1395–1443 of the ORF1 protein sequence when translated from the genomic RNA. A schematic of the ALPV genome and the mapping depth is shown in Fig. 2.

Three ALPV genomic RNA sequences from this study and six others from GenBank were used for phylogenetic analysis (Table 2). Results revealed two main clades (Fig. 3a). The isolates from our study (KE Aphid P7, KE Aphid P9 and KE Maize) were in one of the major clades, grouping with samples from China, South Africa, and East Timor. The two isolates from the USA and Spain formed the other clade. Analyses of nucleotide sequence similarity revealed genetic variation within the subclades and between the clades detected in the phylogenetic analysis. The KE Aphid P7 and KE Aphid P9 ALPV isolates shared 98.9% nucleotide sequence identity with each other and 93% similarity to the isolate from Israel when whole genomes were compared. When compared to other isolates in the subclade (KE Maize, S. Africa, China and E. Timor), sequence similarity was about 89%. Similarity to US and Spanish isolates was 82%. The USA and Spanish isolates shared 94% nucleotide sequence similarity (Fig. 3b).

When we compared the similarity of ORF1 and ORF2 sequences from the genomic RNA of the ALPV isolates we found out that KE Aphid P7 and KE Aphid P9 shared 98.6% and 99.3% similarity in ORF1 and ORF2, respectively. When the two isolates were compared to KE Maize, the similarity was about 90.3% and 86% in ORF1 and ORF2 respectively. All three isolates from Kenya had least similarity to the isolates from Spain and the USA (approximately 83%) in both ORFs (Additional file 12: Table S5).

Analysis of the ORF1 and ORF2 nucleotide sequences using six different algorithms showed evidence of recombination in ALPV. ORF1 showed evidence for three recombination events (Table 3). Event 1 identified three isolates (Israel, KE Aphid P7 and KE Aphid P9) as recombinant. In this event, isolate KE Maize was identified as the most probable major parent alongside an unknown minor parent. Event 2 identified the E. Timor isolate as recombinant and KE Maize was identified as the most probable major parent alongside an unknown minor parent. Event 3 identified the isolate from China as recombinant with isolates KE Aphid P9 and E. Timor as the most probable major and minor parents respectively. ORF 2 showed one recombination event in the isolate from China isolate and KE Aphid P9 and E. Timor as the most probable major and minor antecedents, respectively.

From BLAST analyses, the 11 nucleotide sequences from this study had 87% similarity to isolate BSRV1 (GenBank number: JF423195.1), which is 1473 nucleotides long and encodes part of the non-structural polyprotein (ORF1) of BSRV. A BLAST analysis comparing ORF2 sequences from our BSRV-like sequences and an ORF2 sequence from GenBank, BSRV4 (GenBank number: JF423198.1) revealed 77% and 86% similarity at nucleotide and protein sequence comparisons respectively. The nucleotide similarity among the BSRV-like viruses was 98–99%.

Comparison of the 11 BSRV-like sequences to two ALPV (KE Aphid P9, KE Aphid P7) and three RhPV sequences from GenBank revealed 57–60% similarity to ALPV and 70–72% to RhPV. Among the BSRV-like sequences the similarity was 98–99% (Fig. 4). In phylogenetic analyses, the BSRV-like sequences clustered in a clade containing members of the genus Cripavirus (Fig. 5). This demarcation confirmed that the BSRV-like isolates detected in our samples were distinct viruses unrelated to ALPV or RhPV and were likely to be a novel BSRV-like dicistrovirus.

The sequencing of BSRV-like RNA revealed two ORF. The 5′ proximal ORF 1 was predicted to encode a translation product of 2007 amino acids while ORF2 was predicted to encode a product of 753 amino acids (Fig. 6). Analysis of the genomic RNA sequence for ORF1 in the NCBI conserved protein database predicted regions with similarity to helicase and RdRp domains. The ORF2 sequence was predicted to encode a precursor for three structural proteins (Fig. 6). The length of the intergenic region of the BSRV-like viruses was 594 nucleotides long, which is longer than in any other previously reported dicistroviruses. Only protein functional domains predicted with the highest confidence are shown in the upper panel of Fig. 6.