BCMV isolate PV-0915, BCMNV isolate PV-0413, and a bean-infecting isolate of CMV (PV-0473) were obtained as freeze-dried infected leaf tissue from the Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH (DSMZ) (German Collection of Microorganisms and Cell Cultures). Bulking of inoculum was done by sap inoculation of Nicotiana benthamiana plants before passaging the viruses to common bean. Virus-inoculated N. benthamiana plants were cultivated for least 3 weeks following inoculation and systemically infected leaves were harvested and stored at −80°C for use in subsequent sap inoculation of bean plants.

Experiments were conducted with the common bean (Phaseolus vulgaris L.) variety Red Haricot-GLP 585 cv. “Wairimu” (SimLaw Seeds, Nairobi, Kenya), which is susceptible to infection by BCMV, BCMNV, and CMV. Bean seeds were germinated at 25°C for 5 days in a Petri dish lined with moistened filter paper and after germinating, single beans were planted in Levington M3 compost (Scotts, Chilworth, UK) mixed with sand (J. Arthur Bowers, Lincoln, UK) in a 4:1 ratio in round pots 100 mm × 90 mm (diameter × depth). Bean plants were grown in a growth room (Conviron, Manitoba, Canada) under a long photoperiod (16-h light and 8-h darkness), with a light intensity of 200 µE.m⁻².s⁻¹ (Sylvania Activa 172 Professional 36-W bulbs), at 20°C –22°C.

Once grown to the two-leaf stage (5 days post potting), plants were either inoculated with virus or mock-inoculated with water. For sap inoculation, 50-mg frozen leaf was ground in 1-ml distilled water using a pestle and mortar. Two lower leaves were dusted with Carborundum (SiC), which was used to abrade the leaf during mechanical inoculation to aid virus entry. The sap was rubbed gently onto the two leaves and then the excess sap and ground leaf debris cleaned off by spraying the leaf with distilled water. The plants were left to grow for another 10 days and by then the virus-infected plants displayed clearly observable disease symptoms (see Supplementary Figure 1 ). EPG experiments were conducted at 10 days post-inoculation.

For further confirmation of infection, RT-PCR was done using plant leaf samples post-EPG experiments. Testing prior to the experiment was avoided as introduction of injury to the leaves could have triggered wound-induced changes to the plant physiology that could affect aphid herbivory. Briefly, approximately 50 mg of fresh symptomatic, leaf samples were obtained by using a 1-cm-diameter cork borer. RNA extraction was done using a total RNA purification kit (Norgen Biotek, Thorold, Ontario, Canada) using the manufacturer's instructions. RNA quality and quantity were measured using a Nanodrop^® ND1000 spectrophotometer (Thermo Fisher Scientific, Waltham, Massachusetts, USA). RNA was reverse-transcribed using GoScript™ (Promega) reverse transcription kits as per manufacturer instructions.

For BCMNV (isolate PV-0413; GenBank number HG792063), the sequences for the primers used to detect the BCMNV coat protein coding sequence (CP) were reverse primer 5′-AGA GAA TAT TCA TAC CCGC-3′ and 5′-ACA CAA GAG CTA CCA AG-3′ as forward primer. For BCMV (isolate PV-0915, GenBank Number: HG792064), the sequences for the BCMV CP gene-specific primers were forward primer 5'-TGA CAA TGG CAC TTC ACC-3' and reverse primer 5'- AACAAACATTGCCGTAGC-3'. The technique used to design these primers is described in an earlier publication (Mutuku et al., 2018). The primers for the CMV CP gene (isolate PV-0473; GenBank number; MH748553.1) were forward primer 5'-ACC ATC TCC TAG GTT TCT TCGG-3' and reverse primer 5'- GTC TCC TTT TGG AGG CCC-3'. Another CMV CP gene-specific primer set was also used: forward primer 5′-ATG GAC AAA TCT GAA TCA ACC AGT GCT-3′ and reverse primer 5′-TCA GAC TGG GAG CAC TCC AGA TGT GGG-3′ (Kwon et al., 2016). PCR conditions were 94°C for 3 min followed by 35 cycles of 94°C for 30 s, 56°C for 30 s, 72°C for 1 min, and a final 5-minute extension at 72°C. For visualization of successful PCR amplification, the amplicons were loaded into wells of a 1% (w/v) agarose gel in TAE containing 0.05 µg. ml-1 ethidium bromide stain. The gels were submerged in TAE buffer and run in an MHU–1010 gel rig (Flowgen/Scientific Laboratory supplies, Hessle, UK) at 100 V using a Power-Pac 3000 (Bio-Rad, Hemel Hempstead, UK). A 100-bp DNA ladder (Bioline, London, UK) was used to facilitate size estimation. Gels were examined under UV illumination on a gel documentation system to determine if the amplicons were of the expected product sizes (approximately 800 bp).

Experiments used two aphid (Aphididae: Hemiptera) species: Myzus persicae Sulzer (common names: peach-potato or green peach aphid), and Aphis fabae Scopoli (common name: black bean aphid). Both colonies were generated by transferring a single aphid to a host plant. The insecticide-sensitive M. persicae clone US1L (Devonshire and Sawicki, 1979) was maintained on Chinese cabbage (Brassica rapa subspecies pekinensis) cv. Green Rocket (Tozer Seeds, Cobham, UK). The Kennedy and Booth clone of A. fabae (Kennedy and Booth, 1950) was maintained on broad bean (Vicia faba L.) cv. Sutton dwarf (King Seeds, Essex, UK). Aphid stock colonies were maintained on plants in individual pots in a growth chamber at 22°C under long day conditions and subsequently passaged to new plants every 2 weeks. Infested plants were covered with micro-perforated bread bags (Seal Packaging, Luton, UK) secured around the pots with rubber bands to contain the aphids.

To observe aphids' initial probing behavior, we used a previously described direct observation method (Caillaud et al., 1995). Adult A. fabae or M. persicae were starved for 30 min prior to experimentation. Single aphids were placed on the adaxial surfaces of leaves, observed under magnification, and the time taken for each aphid to first insert its stylet was measured.

More detailed observations of aphid feeding behavior were done using the EPG method (Tjallingii, 1978; Tjallingii and Esch, 1993; Tjallingii, 2006) as previously described (Ziebell et al., 2011; Westwood et al., 2013). Briefly, individual aphids were starved for 30–60 min before being tethered to approximately 4 cm lengths of 20 µm diameter gold wire (EPG systems, Wageningen, The Netherlands) using conductive silver paint (EPG systems), which was soldered to a 1-cm brass pin, connected to an amplifier with 1 GΩ resistance and 50–100X gain. Connected aphids were placed on individual plants inside a Faraday cage and signals received from the EPG monitor taken over 8-h recording periods and 15 aphids per treatment. EPG signals were analyzed using A2EPG software (Adasme-Carreno et al., 2015) and automated EPG parameter calculations used Microsoft Excel-based spreadsheets (Sarria et al., 2009).

Two types of statistical analyses were used. In the first, generalized linear models (GLMs) were used to model the number of occurrences of a waveform and to model the duration of waveforms. In the second, survival analysis was used to model the rate at which waveforms were entered.

We modeled EPG data that took the form of waveform durations using GLMs with Gamma-distributed response variables. We modeled EPG data that took the form of the number of occurrences of waveforms using GLMs with either Poisson- or negative binomial-distributed response variables. In all cases, the GLM included virus treatment as a fixed effect. In preliminary analyses of the count data (number of occurrences), dispersion tests were conducted for each waveform, and where the data was found to be significantly over-dispersed a negative binomial distribution GLM was used (otherwise a Poisson GLM was used). In addition, we modeled potential E2 index (PEI) (i.e., the proportion of the EPG recording time remaining to an aphid, following completion of a first E2 waveform, that is spent in subsequent E2 waveforms) using a GLM with Beta-distributed response variable. Note, that while a Beta distribution was the most appropriate for proportional data of this type, it is, however, necessary to perform the data transformation,

where Y is the transformed response variable, y is the original response variable and n is the sample size (Smithson and Verkuilen, 2006). The transformation is necessary because the Beta distribution is defined for the open interval (0, 1) while the PE2 data contains “zeros” and “ones”. In all of the above cases (duration, count, and proportion data), post hoc comparisons were made using the multcomp package in R (Hothorn et al., 2009), through the extraction of Dunnett contrasts with Bonferroni adjustments (i.e., to test for significant differences between counts, durations, or proportions, on virus-treated plants compared with mock-inoculated plants).

Data of the “time-to” form, i.e., time-to-first probe, and time to phloem EPG phases E1 and E2, were analyzed by survival analysis: a method of analyzing the occurrence of an event during an observation period. Survival curves were estimated by the Kaplan-Meier method (survival package in R, surv, and survfit functions: Therneau, 2015). Virus treatment curves were compared with mock treatment curves using the Peto-Peto test (survminer package in R, surv_pvalue function: Kassambara et al., 2017) to evaluate if virus infection affected the time taken by aphids to begin probing on the leaf epidermis and initiate phloem ingestion respectively. All analyses were conducted in R (version 3.5.0) (R Core team, 2014).

Aphid activities associated with probing, salivation, feeding from the sieve elements, mechanical stylet difficulties, and drinking from the xylem were recorded by EPG for 8 h. The different waveforms observed and the associated aphid behaviors are shown in Table 1 and Supplementary Figure 2 . For aphids of both species, the combined probing (Pd) and stylet pathway (C) activities (indicated in blue) accounted for over 50% of total recorded activity in all treatments ( Figure 1 ). For A. fabae, the highest proportion of time spent in stylet pathway activity (waveform C) was on CMV-infected plants (80.5%), while for M. persicae, it was highest on mock-inoculated plants (66.3%). Over the total EPG recording period, A. fabae spent proportionally more time in phloem related activities (30.6%) than M. persicae (7.2%) on mock-inoculated plants ( Figure 1 ). The calculated statistical outputs from the GLM analyses of the occurrence and duration of the different activities inferred from the recorded waveforms for A. fabae and M. persicae is provided in the additional material ( Supporting Information: Excel Files 1 and 2 , respectively). In addition to EPG studies, visual observations of aphid probing were carried out. These showed that individuals of both aphid species started probing earlier on virus-infected plants than on uninfected plants. Pairwise post hoc comparisons showed significant differences in time taken by aphids to begin probing on virus-infected plants when compared to mock-inoculated plants but not among the virus treatments ( Figure 2 ).

From the EPG observations, the periods of time during which aphids were not probing were significantly shorter for A. fabae on BCMV-infected plants (p = 0.00057) when compared to mock-inoculated plants ( Table 2 ). This was attributed to a significant reduction in total duration spent in the pathway phase (C) (p < 0.001), as there were no significant differences between the different treatments in the number of potential drops (Pd). M. persicae probed less frequently on virus-infected plants with statistically significant reductions for those individuals placed on plants infected with BCMNV (p = 0.022) or BCMV (p = 0.024) when compared to those monitored on mock-inoculated plants.

More detailed examination of EPG data for A. fabae pre-ingestion salivation activity into the phloem (E1 waveform) and subsequent phloem ingestion (E2 waveform) showed that on CMV-infected plants there were significant changes to feeding behavior when compared to aphid activity on mock-inoculated plants. Though the number of E1 events were not significantly increased for A. fabae aphids foraging on CMV-infected plants, there were significant increases in both the total duration (p = 0.0161) and mean length (p = 0.00236) of E1 activities ( Figure 3 ). This is indicative of difficulty feeding from the phloem. Also, both the total duration and mean length of E2 was significantly reduced for A. fabae on CMV-infected plants ( Figure 3 ).

An analysis of the likelihood that sustained continued phloem feeding would occur (PEI), showed that A. fabae aphids were less likely to feed from the phloem of virus-infected plants. The likelihood of recurrent phloem feeding on uninfected plants was highest for aphids on uninfected plants (39.85%) as compared to BCMNV (32.22%), BCMV (28.31%), and CMV (8.97%) ( Figure 4 ). Pairwise comparisons were significantly lower for aphids on CMV-infected plants (p = 0.040) ( Figure 4 , Supporting Information Excel File 3 ). Notably, the E1 events exhibited by A. fabae contributed to more than half of all phloem phase activity for aphids on CMV-infected plants. A. fabae took longer to transition from phloem salivation to sustained phloem feeding (where E2 lasted longer than 10 min). On mock-inoculated plants, A. fabae averaged 3.11 h before their first sustained phloem feeding event, which increased to 4.11, 4.67, and 4.92 h on BCMNV, BCMV, and CMV-infected plants, respectively ( Figure 4 ).

By comparison, there was no discernible trend in the PEI from recordings of M. persicae on virus-infected and mock-inoculated plants ( Figure 4 ). The mean PEI for M. persicae on mock-inoculated plants was 5.6%, which is lower than the lowest PEI for A. fabae, which was 8.8% on CMV-infected plants. Thus, the generalist aphid M. persicae is probably less likely to settle and feed on common bean plants than the bean specialist A. fabae, regardless of the host's virus infection status. The E1 salivation events exhibited by M. persicae contributed to more than half of all phloem phase activity for all four treatments (Mock: 65%, CMV: 61%, BCMNV: 58%, and BCMV: 71%), and sustained phloem feeding events were recorded towards the end of the 8-h recording period ( Figure 4 ).

Patterns of salivation into the phloem and ingestion of phloem sap (indicated by, respectively, waveforms E1 and E2) were further analyzed by survival analysis. For this analysis, the event of interest was the entry of M. persicae or A. fabae mouthparts into the phloem of mock-inoculated or virus-infected bean plants. A comparison of the two aphid species on uninfected plants, in terms of species differences in phloem salivation (E1) and phloem ingestion (E2) activity, showed that while aphids of both species were likely to penetrate and salivate in the phloem, M. persicae seemed innately reluctant to feed from common bean phloem compared to A. fabae and that this difference in species behavior was statistically significant (p = 0.00037) ( Figure 5A ).

For A. fabae, entry into the E2 phloem phase was significantly delayed on virus-infected plants (Kaplan-Meier: p = 0.017), and pairwise comparisons of A. fabae on CMV-infected compared with mock-inoculated plants were highly significant [Pairwise Peto-Peto: p = 0.0099 ( Figure 5B , Supporting Information: Excel File 3 )]. By comparison, there were no appreciable differences in the times taken by M. persicae to initiate phloem salivation or to initiate phloem feeding on mock-inoculated or virus-infected plants ( Figure 5C , Supporting Information Excel File 3 ). Fewer A. fabae transitioned to sustained phloem-feeding on virus-infected plants than on mock-inoculated plants ( Figure 5B ). Of 15 aphids recorded in each treatment, 12 aphids on uninfected plants transitioned to sustained phloem feeding. In comparison, only 7, 8, and 5 aphids on BCMNV-, BCMV-, and CMV-infected plants, respectively, transitioned to sustained phloem feeding over the 8-h recording period ( Supporting Information: File 3 Sheet 2 ). For M. persicae, only two aphids out of 15 in each treatment transitioned to sustained phloem feeding ( Figure 5C , Supporting Information: File 3 Sheet 2 ).

The EPG F waveform registers mechanical stylet activities occurring extracellularly and can indicate resistance to feeding. For A. fabae, more aphids showed mechanical stylet probing difficulties while feeding on virus-infected plants during the first 6 h of the recording period ( Supplementary Figure 3 ). The total and mean duration of F waveform events were significantly increased in all virus treatments when compared to recordings done on mock-inoculated plants ( Figure 6 , Supplementary Table 1 ). There was a trend towards increased duration of feeding difficulties for M. persicae on virus-infected plants over the course of recording though this was not statistically significant. However, analysis of the mean duration of waveform F incidents indicated that feeding difficulties were significantly increased for M. persicae on plants infected with BCMV and CMV ( Figure 6 ).