High-throughput phenotyping of plant resistance to aphids by automated video tracking

Piercing-sucking insects are major vectors of plant viruses causing significant yield losses in crops. Functional genomics of plant resistance to these insects would greatly benefit from the availability of high-throughput, quantitative phenotyping methods. We have developed an automated video tracking platform that quantifies aphid feeding behaviour on leaf discs to assess the level of plant resistance. Through the analysis of aphid movement, the start and duration of plant penetrations by aphids were estimated. As a case study, video tracking confirmed the near-complete resistance of lettuce cultivar ‘Corbana’ against Nasonovia ribisnigri (Mosely), biotype Nr:0, and revealed quantitative resistance in Arabidopsis accession Co-2 against Myzus persicae (Sulzer). The video tracking platform was benchmarked against Electrical Penetration Graph (EPG) recordings and aphid population development assays. The use of leaf discs instead of intact plants reduced the intensity of the resistance effect in video tracking, but sufficiently replicated experiments resulted in similar conclusions as EPG recordings and aphid population assays. One video tracking platform could screen 100 samples in parallel. Automated video tracking can be used to screen large plant populations for resistance to aphids and other piercing-sucking insects.


Background
More than 100 aphid species (Aphididae) are economically significant pest insects and most crops are host to at least one species [1]. Aphids feed on phloem sap, and to reach the phloem they move their stylets between plant cells towards a sieve element, making short punctures in cells along the way. Most probes are prematurely interrupted in the epidermis and mesophyll. When, however, a phloem vessel is reached, aphids can ingest phloem sap continuously for many hours or even days [2]. Although aphids inflict little tissue damage, they transmit plant viruses and deplete host plants of photoassimilates and free amino acids [3,4]. In wild plant populations aphids rarely constitute pests due to effective natural defence strategies, such as epicuticular waxes, protease inhibitors, and induced production of secondary metabolites [4][5][6][7][8][9][10][11][12]. After generations of domestication many of these defence traits have been diminished or lost in cultivated plants, making them vulnerable targets of herbivorous insects [13,14]. The genetic backgrounds of resistance mechanisms still remain largely elusive and genomics studies strongly depend on the capacity for phenotyping large panels of plants. Few high-throughput methods have been established for assessing plant resistance to insect herbivores, such as aphids or other piercing-sucking insects [15][16][17][18][19][20]. Generally, two approaches are used to quantify the level of plant defence against aphids; either assessment of aphid population development or investigation of aphid feeding behaviour. Aphid population assays are generally the most demanding in terms of time and space, since they require the availability of a climatecontrolled compartment for 1 or 2 weeks and extensive manual work [21][22][23]. On the contrary, aphid feeding behaviour can be measured within a couple of hours via the Electrical Penetration Graph (EPG) technique. EPG recording delivers electrical waveforms comprising information on the plant tissue that is penetrated (phloem vessel, xylem vessel or other cells) and the stylet penetration activity (cell puncture, salivation, ingestion, penetration difficulties) [24,25]. EPG studies have shown that aphids prolong phloem ingestion on suitable host plants and delay and reduce feeding on resistant or non-host plants [26][27][28][29][30][31][32][33]. The high specificity of the information about plant tissue and key components of aphid behaviour, makes this methodology appealing for exploring defence mechanisms. A drawback of EPG is, however, the restricted capacity, generally 8 plants per setup [34], and the labour-intensive nature of wiring aphids and annotating electrical signals.
Here, we present the methodology and validation of image-based tracking of aphid feeding behaviour. Automated video tracking was introduced in the early 1990s and has since been used in many animal behaviour studies [35][36][37][38][39][40][41]. Video tracking involves software-engineered pattern analysis of grids of pixels in order to quantify the location and movement of subjects over time. In this study, we used movement patterns of the central body point of aphids to estimate the duration of plant penetrations made by the aphid's mouth parts. Previous EPG studies showed that probes shorter than approximately 3 minutes represent penetrations in the epidermis and/ or mesophyll [26], and that probes involving phloem uptake last on average at least 25 min [27,42,43]. This allowed us to discriminate test probes from putative phloem uptake events in video observations in order to identify plants that are resistant to aphids. We benchmarked the performance of the high-throughput video tracking platform against EPG recordings and aphid population development assays, using natural accessions of Arabidopsis thaliana, and lettuce cultivars, Lactuca sativa, in combination with the green peach aphid, Myzus persicae (Sulzer), and the black-currant lettuce aphid, Nasonovia ribisnigri (Mosely) (Hemiptera: Aphididae), respectively.

Results
Tracking aphid feeding behaviour Automated video tracking of aphid feeding behaviour was performed using video tracking software and a stationary camera mounted above 20 no-choice arenas. We introduced one aphid onto each arena, consisting of an agar substrate almost completely covered by a leaf disc, and recorded 20 arenas simultaneously with a frame rate of 25 frames s −1 (Figure 1, Additional file 1: Figure S1). Because the aphid's mouthparts were not visible in the multi-arena setup, we made the assumption that when the aphid's centre point was located on the leaf disc and did not move, the aphid was penetrating the leaf tissue with its stylets. By assessing video images by eye, we defined velocity thresholds for the start and end of probing events of two aphid species, M. persicae and N. ribisnigri ( Figure 2, Additional file 1: Figure S2). According to our observations, the software was more vulnerable to premature probe endings of N. ribisnigri due to body movements during probing (such as event γ in Figure 2). As this aphid species is somewhat larger (±1.9 mm body length, versus ± 1.7 mm for M. persicae), movements around the fixated mouth resulted in a higher tangential velocity, and therefore required a higher velocity threshold.

Accuracy
To test the accuracy of the platform, we performed automated video tracking and human observations simultaneously. A camera was attached to a stereo microscope to Figure 1 Video tracking platform. A stationary camera is mounted above a microtitre plate which is placed on top of a backlight unit with ventilation at the left. Wells in the microtitre plate contained a leaf disc and an aphid (a). Cling film was wrapped around the plate to prevent aphids from escaping. The camera was connected to a computer with EthoVision® XT video tracking software (b). Movements of the aphid's centre point were automatically tracked (red track shows movements across 30 seconds). With this information the software calculated aphid velocity (line graph) and estimated probing (purple bar) and non-probing events (green bar). deliver a side-view on the arena for manual scoring of probes (Additional file 2). Among a total of 139 probes of 16 different M. persicae aphids scored by hand, 88% was detected with video tracking (Figure 3a). Undetected and false positive probes involved only short events (<3 min). Of the detected probes, 19% was either underrated (multiple 'true' probes were considered as one probe), or overrated (one 'true' probe was translated into multiple probes by the software). Underrated samples were caused by undetected probe stops due to slow movements below the velocity threshold. Overrated samples were caused by false probe stops when, for example, the aphid was immobile on the edge of the leaf disc and the assigned position continuously switched between an "on the leaf disc" and "off the leaf disc" status ( Figure 3b). Three times this incident occurred, leading to 17 redundant probes of which 10 were filtered out automatically (see Methods, section Software settings). Other reasons for premature probe stops were abdominal movements during probing related to e.g. reproduction or honeydew excretion. The longer probes lasted, the higher the risk was of encountering such incidents. Indeed automatically tracked probes were in general biased to end 73 to 12 seconds too early (Figure 3c), and the total duration of probing was underestimated (on average 46 min ± 2.5 min standard error, versus 50 min ± 1.9, P = 0.01, Mann-Whitney U test, total observation duration: 55 min). Nevertheless, the recorded number and duration of probes were highly correlated to human observations ( Figure 4, average r 2 = 0.7 with 275 pixels per mm 2 ). Other parameters, such as distance moved, were also highly correlated with feeding behaviour in general, but were less informative with regard to long probes ( Figure 4l). Although automated video tracking did not achieve a precision as high as manual scoring, it enabled observing multiple arenas simultaneously. In the above described tests, we used 275 pixels per mm 2 , equal to a coverage of 20 arenas with our 768 × 576 pixels camera. To determine whether the capacity could be increased, we repeated the experiment with only 155 pixels per mm 2 , equal to a coverage of 35 arenas, but found that reduced resolution resulted in decreased correlations with human observations (average r 2 < 0.5).

Benchmarking against EPG recording with Arabidopsis
To validate whether automated video tracking delivered a reliable proxy for plant resistance, feeding behaviour of M. persicae was measured during 8 hours continuous recording on two natural accessions of Arabidopsis, Co-2 and Sanna-2 (Additional file 3). These accessions were selected from a population of hundreds of accessions based on preliminary video tracking data. Automated video tracking showed that M. persicae aphids walked larger distances on Co-2 and reduced the mean duration of long probes (Table 1). EPG recordings on intact plants confirmed shorter durations of (sustained) phloem ingestion, and additionally revealed more short probes, non-probing behaviour and a delayed phloem uptake on Co-2 (Table 1). This behaviour is an indication of both epidermis/mesophyll-located and phloemlocated resistance in Co-2 against M. persicae. All aphids ingested phloem, but quantitative differences in feeding Figure 2 Velocity thresholds for registration of probes. An example of how aphid feeding behaviour was measured using a resolution of 275 pixels per mm 2 . Subject states can be defined as 'moving' or 'not moving' by means of two thresholds: the start velocity at which the subject begins to move, and the stop velocity at which the state changes from moving to not moving. Probe starts were recorded if the velocity of the aphid's centre point dropped below 0.02 mm/s for at least 10 seconds (α). Probe stops were recorded if the velocity of M. persicae aphids exceeded 0. 3 mm/s for at least 2 seconds (β), or 0.35 mm/s for at least 2 seconds in the case of winged N. ribisnigri aphids. To avoid premature probe endings due to short movements during probing (event γ), probe stops were only recorded when velocity increased above 0. 1 mm/s for more than 2 seconds. Figure adjusted from the EthoVision XT Reference Manual (version 8) [44].
behaviour between aphids on Co-2 and on Sanna-2 were already apparent in the first hour ( Figure 5). A reproduction assay on intact plants confirmed that Co-2 was indeed more resistant than Sanna-2, although the resistance was not absolute. Depending on plant age, aphids either started reproduction later or produced fewer offspring ( Figure 6). Although we had been able to correctly identify a quantitative difference in resistance with automated video tracking, the effects were smaller than in EPG recordings on intact plants. To verify whether the plant line effects in the video tracking assay were attenuated due to the use of excised plant tissue, the EPG experiment was repeated with leaf discs. Particularly for the resistant accession, aphid feeding behaviour was different and involved more phloem uptake and fewer short probes on leaf discs compared to intact plants (Additional file 1: Table S2). The only significant difference between the accessions that remained was a reduced duration of phloem uptake events on Co-2 (Table 1). In addition, contribution of salivation to the phloem phase, required to suppress (callose-mediated) sieve-plate occlusion [45], was equal on leaf discs but higher on intact plants of Co-2 ( Figure 7). This indicates that the resistance mechanisms in intact plants were partially lost in leaf discs.

Benchmarking against EPG recording with lettuce
Apart from a study system with partial resistance, an example of near-complete resistance was tested with the video tracking platform. The behaviour of black-currant lettuce aphids, N. ribisnigri, biotype Nr:0 was recorded on two near-isogenic lettuce cultivars, the resistant 'Corbana' and susceptible 'Terlana'. Previous studies showed Figure 3 Accuracy of automated tracking in comparison to human observations. M. persicae feeding behaviour was measured on Arabidopsis leaf discs by automated video tracking and human observations simultaneously. (a) Out of 139 probes of 16 aphids scored by hand, 88% was detected by automated video tracking. Probes were considered a match when their duration overlapped at least partially. Some of the detected probes were matched by too few (underrated) or too many (overrated) probes. For these situations, the amount of missed or redundant probes is shown. 17 Probes went undetected and 20 false probes were recorded. Mean duration per probe is shown above the bars. (b) Screenshots of the top-view video used for automated tracking. The lower image (σ) shows an aphid positioned on the edge of the leaf disc for more than 20 min, causing overrated probe counts by the software due to continuous switching between an "on the leaf disc" and "off the leaf disc" status. (c) Differences between software and human observations per matched probe. 95% Confidence Intervals are shown above the histograms. Negative values correspond to too early probe starts, too early probe endings, resp. too short duration of probes compared to the human observations. In case of overrated probe counts, the probe with the most similar duration as the manually scored probe was included. The outlier caused by the example in (b) is annotated with σ.
(See figure on previous page.) Figure 4 Correlation between automated video tracking and human observations. M. persicae behaviour was measured by automated video tracking (x-axes) and human observations simultaneously (y-axes). Three categories of probes were distinguished: All probes, Long probes (>15 min), and Short probes (<3 min). The duration (min) and number of probes measured by human observations were compared to: (a,d,g,j,m) the duration (min) and number of probes (all, long, and short probes) measured by video tracking, (b,e,h,k,n) the total time not moving (min), and (c,f,i,l,o) the distance moved by the aphids (cm) (*P < 0.05; **P < 0.01; ***P < 0.001, Pe = Pearson correlation test, Pl = Pearson correlation test on log transformed data, S = Spearman correlation test, dashed lines represent a hypothetical r 2 = 1, n = 16 recordings of 1 aphid for 55 min, 275 pixels per mm 2 ). that the Nr gene is responsible for near-complete resistance in Corbana against this biotype of aphids, mainly due to a phloem-located mechanism [34,46]. Our video tracking observations on leaf discs were compared to EPG recording data by ten Broeke et al. [47]. Seven out of nine video tracking variables confirmed that cultivar Corbana was more resistant than cultivar Terlana (Table 1). Aphids on Corbana spent less time on long probes and more time on shorter probes and other activities. In addition, aphids increased their walking activity over  time on both cultivars, but generally covered larger distances on Corbana leaf discs (mixed linear model: time effect: P = 0.00, cultivar effect: P = 0.03, time × cultivar interaction: P = 0.77, Figure 8). Yet, the resistance effect was less pronounced in video tracking compared to EPG recording on intact plants: only 11% of the aphids in EPG recordings showed phloem ingestion on Corbana plants, while 78% of the aphids in the video assay performed long probes on Corbana. These long probing events could include other activities, such as water ingestion from xylem vessels, since EPGs showed that on Corbana plants more aphids penetrated xylem sieve elements (12 aphids on Corbana versus 2 aphids on Terlana).

Throughput
Using simulated data with a similar plant line effect as the data sets from the plant-aphid assays described here, we assessed the required sample size and recording duration for automated video tracking ( Table 2). With 20 replicates of 8-hour observations, significant resistance was detected in more than 80% of the cases for the Arabidopsis plant line effect on M. persicae (2 response variables tested per simulated data set, Bonferroni correction: P < 0.025). The near-complete resistance of Corbana lettuce against N. ribisnigri biotype Nr:0 was detected in more than 80% of the cases with 15 replicates of 4 hours of video observations. Subtle differences in resistance in Arabidopsis were more difficult to detect when video observations were shorter than 8 hours (Table 2). On the other hand, reducing the video duration to the first 4 hours improved the detection of near-complete resistance, as with N. ribisnigri biotype  Nr:0 on the Corbana lettuce cultivar. Apparently, in this case, the precision of video tracking decreased over time.
While the EPG recording with lettuce did not reveal an increase in aphid activities in the xylem or mesophyll over time (Figure 8), the last stretch of the video observation was likely confounded by sessile behaviour other than probing. The risk of falsely rejecting the null hypothesis was limited to 1% (M. persicae on Arabidopsis accession Col-0). Overall, video tracking required similar observation durations as EPG recording, but a larger sample size to detect significant plant effects ( Table 3). The required amount of replicates was, however, compensated by screening many samples simultaneously and automated data annotation.

Leaf discs
The effect sizes measured in video tracking with leaf discs were substantially smaller compared to EPG recording on intact plants. EPG recording on leaf discs confirmed that the application of excised plant tissue partially impaired plant resistance [47,48], possibly due to the interrupted supply of ions and metabolites in the phloem, or due to the interference by jasmonic acid and ethylene mediated wound responses [49]. Furthermore, aphids can be disturbed by the decrease in pressure in the sieve elements of excised plant tissue, although they are well capable of active uptake of sap [50,51]. The increase of coagulating proteins and cellular debris in the phloem after plant wounding may plug sieve plates and the aphid's food canal in the stylets [28,52,53]. To prevent such potential clogging of sieve elements, aphids might increase the injection of watery saliva into the phloem or shorten their feeding events, but neither of these effects were observed consistently. To maintain turgor better the use of leaves still connected to intact plants would be favourable, but this is currently not feasible in view of poor detection of aphids in more complex environments. Arenas designed to hold entire detached leaves or seedlings on agar could, however, be a feasible alternative to leaf discs.

Application
High-throughput phenotyping techniques of sucking insect species are urgently needed in view of functional genomics studies aiming to find subtle allelic differences in plant populations measuring many hundreds of plants. Conventional methods, like EPG and population studies, are less scalable for this purpose and carry much higher investments in terms of time (labour, duration) and costs (equipment, greenhouses). In this study, automated video tracking was used to study aphid feeding behaviour, but it could as well be applied to track the behaviour of other piercing-sucking insects. We recommend to validate the velocity thresholds for each species first, by checking several video files by hand. As shown here with two aphid species, size and velocity can differ and will affect the accuracy of probe estimations. When studying plants with thick or dark leaves, increased resolution, better (macro) lenses, and lateral light sources instead of backlight can help to improve the detection of insects. We worked with EthoVision XT video tracking and analysis software, but other programs or programming environments, such as MatLab and ImageJ, could as well serve as robust video tracking tools [37,40,54].

Conclusions
The aim of this study was to develop a high-throughput method to screen large plant populations for resistance to aphids and other piercing-sucking insects. For the first time it is shown that automated video tracking of aphid body movement can be used to estimate how often the insects are penetrating plant tissue and are reaching the vascular bundle. The use of leaf discs instead of intact plants enhanced the throughput of the video tracking platform, but EPG recording illustrated that resistance effects were partially lost in leaf discs. Nevertheless, we could identify both intermediate and extreme levels of resistance with video tracking. In Arabidopsis accession Co-2, we found a quantitative resistance level. This was confirmed in additional bioassays, Student's t-tests have been applied to subsamples of two simulated data sets with a similar mean and standard deviation as two response variables from the video tracking assays. The percentage of tests with a significant outcome in at least one of the two variables is represented as the detection rate (Bonferroni correction: P < 0.025). suggesting the involvement of constitutive or rapidly activated resistance mechanisms in both epidermis/mesophyll and phloem, resulting in a small detrimental effect on the aphid population. The video tracking platform also confirmed the near-complete resistance of the lettuce cultivar Corbana to N. ribisnigri biotype Nr:0. Although video tracking requires more replicates to identify resistant plants than the conventional EPG technique, it can screen many samples simultaneously in a confined space. In addition, computerized data acquisition reduces laborious exercises, such as annotation of electrical patterns or counting of aphid populations, and only little plant material is required which can be advantageous when studying segregating populations with only one plant per genotype. These features make automated video tracking a valuable phenotyping method for screening large plant populations for resistance to piercingsucking insects that are serious pests in our crops.

Plants and insects
Arabidopsis, Arabidopsis thaliana (L.) Heynh., plants were grown for 4-5 weeks in pots (5 cm diameter) with pasteurized potting soil (4 h at 80°C; Lentse potgrond, Lent, The Netherlands) in a climate room at 21 ± 1°C, 50-70% relative humidity, an 8/16 h day/night cycle, and a light intensity of 200 μmol m −2 s −1 . Four natural accessions of Arabidopsis were used throughout this study: Col-0 (CS76113), Van-0 (CS76297), Co-2 (CS28163) and Sanna-2 (CS76223). Seeds were acquired from the European Arabidopsis Stock Centre and propagated by the Laboratory of Genetics, Wageningen University. Lettuce, Lactuca sativa (L.), cultivars Corbana (resistant) and Terlana (susceptible) were grown for 3 to 4 weeks in a greenhouse compartment at a temperature of 20 ± 3°C during the day and 18 ± 3°C during the night, 50-70% relative humidity and a 14/10 h day/night cycle using artificial lighting. Seeds were acquired from Enza Zaden bv. Myzus persicae (Sulzer) aphids were reared in a climate room on radish plants at 19°C, 50-70% relative humidity and a 16/ 8 h day/night cycle. Nasonovia ribisnigri (Mosely) biotype Nr:0 aphids were reared on the susceptible lettuce cultivar Terlana in a greenhouse compartment at a temperature of 20 ± 3°C during the day and 18 ± 3°C during the night, 50-70% relative humidity, and a 14/10 h day/night cycle.

Video tracking platform
Aphid behaviour was recorded with an analogue, monochrome camera (Ikegami, model: I CD-49E, type: REV, 768 × 576 pixels) with a varifocal lens (Computar H3Z4512 CS-IR, 4.5-12.5 mm F1.2) mounted above the arenas (Figure 1). An arena consisted of a well in a 96-well microtitre plate, having a 6.5 mm inner diameter (Sarstedt, sterile flat bottom suspension cells. No. 831835500), containing a leaf disc with the abaxial side up on a substrate of 1% agar (technical agar no.3, Oxoid). One aphid was introduced per arena and cling film was tightly wrapped around the plate to prevent aphids from escaping. The microtitre plate was placed on a platform, 1 cm above a backlight unit (FL tubes, 5000 K). A fan was attached between the platform and backlight unit to prevent water condensation inside the arenas. Room temperature was controlled at 21-22°C.

Software settings
EthoVision XT 8.5 video tracking and analysis software (Noldus Information Technology bv, Wageningen, The Netherlands) was used for automated video tracking of aphid feeding behaviour in multiple arenas simultaneously [41,55]. Subject detection was achieved with grey scaling (Additional file 1: Table S1). Arenas contained two zones: the leaf disc (zone 1) and the space surrounding the leaf disc (zone 2) (Additional file 1: Figure S1). Zone 1 had a diameter of approximately 5 mm, excluding the outer edges of the leaf disc to prevent aphids on the arena wall to be falsely assigned to the leaf disc. Because zone 1 and zone 2 required different grey scale thresholds, optimal thresholds for zone 1, the leaf disc, were chosen. Consequently, only behavioural data acquired in zone 1 were used throughout this study. Velocity and time thresholds appropriate to starting and ending a probe were fine-tuned using simultaneous observations of the top-view camera (275 pixels per mm 2 ) and a side-view camera attached to a stereo microscope (20-40 × magnification), capturing close-up recording of proboscis and antennae movements of M. persicae aphids (Additional file 2). A probe start was automatically recorded when the aphid was positioned on the leaf disc and its velocity dropped below 0.02 mm/s and did not exceed 0.3 mm/s for at least 10 seconds (Figure 2, Additional file 1: Figure S2). A probe stop was recorded when aphid velocity exceeded 0.3 mm/s for the relatively small wingless M. persicae or 0.35 mm/s for the larger winged N. ribisnigri and did not decrease below 0.1 mm/s for at least 2 seconds. Confounding movements during probing were generally characterized by a repetitive pattern of short movements. The 2 seconds time delay prevented that these movements resulted in false probe stops. Zone-transition problems, which occurred when aphids were positioned exactly on the edge of zone 1 and zone 2, were filtered from the data set after acquisition in EthoVision XT, with the statistical computing program R (Additional file 4). These incidences, characterized by a train of consecutive short probes in the output, were filtered out by excluding probes with a duration of less than 3 seconds that were preceded by a very short non-probe bout of maximally 15 seconds. These thresholds were selected by hand using some examples of zone transition problems in this study.

Video recording versus human observations
To validate automated tracking of probes with manual scoring, we used a camera mounted on a stereo microscope (20-40×) with a side-view on a single arena (n = 16) (Additional file 2). Each replicate consisted of a 55 min continuous recording of one arena with a single adult M. persicae aphid and an Arabidopsis leaf disc, by both the top-view and side-view camera. Aphids were starved between 30 minutes and three hours before the experiment. Recordings with the top-view camera were performed at two distances: capturing 20 arenas with 275 pixels per mm 2 , and capturing 35 arenas with 155 pixels per mm 2 . Leaf discs of 6 mm in diameter were cut just below the leaf apex of 4-5 week old Col-0 and Van-0 plants. The Observer® XT 10 software (Noldus Information Technology bv, Wageningen, The Netherlands) was used for manual scoring of probes. Probe starts were manually recorded when body movement stopped, the proboscis was touching the leaf and antennae moved backwards. If the aphid's proboscis was obscured, body arrestment on the leaf disc with subsequent backward movement of antennae was defined as a probe start [56,57]. Probe endings were manually recorded when antennae moved upward and the aphid removed its proboscis from the leaf, or, when the latter was not visible, when the antennae moved upwards followed by locomotion. Apart from probe estimations, we also tracked the "total time not moving" across the whole observation, using a start velocity of 0.3 mm/s and a stop velocity of 0.02 mm/s. Velocities were averaged across 5 frames, using a sample rate of 5 frames per second.

Video tracking assays
In each recording twenty arenas were tracked simultaneously for 8 hours, with a frame rate of 25 s −1 , and a resolution of 275 pixels per mm 2 (Additional file 3). All arenas consisted of a different plant and aphid individual and within each recording the 2 involved plant lines were equally represented. For Arabidopsis accessions Co-2 and Sanna-2, automated video tracking was performed with 7 to 8 day old wingless M. persicae aphids (Co-2 n = 20, Sanna-2 n = 17). Leaf discs of 6 mm in diameter were made just below the apex of intermediately aged leaves. Aphid survival was checked the day after recording. Subject detection was checked after data acquisition on 6 time points across the video. Three samples with no or low quality detection were excluded from the analysis. Video tracking of winged N. ribisnigri biotype Nr:0 on lettuce cultivars Terlana and Corbana was performed with 4 mm leaf discs (Corbana n = 27, Terlana n = 28). In view of the large contour of winged N. ribisnigri aphids, we used arenas with leaf discs of 4 mm diameter and a 3-4 mm leaf edge-to-wall distance in order to have a clear distinction between aphids on the leaf disc and aphids on agar or the arena wall. Leaf discs were made near the leaf base of the third oldest leaf, next to the mid vein. None of the aphids had died the day after recording. Five samples with no or low quality detection were excluded from the analysis. The response variable "duration not moving" was measured using a start velocity of 0.3 mm/s and a stop velocity of 0.02 mm/s. Velocities were averaged across 5 frames, using a sample rate of 5 frames per second.

EPG recording
Feeding behaviour of the green peach aphid, M. persicae, was analysed with EPG recording on two natural accessions of Arabidopsis, Co-2 and Sanna-2, during 8-hour observations. EPG recording was made on both intact plants (Co-2 n = 19, Sanna-2 n = 18) and leaf discs (Co-2 n = 9, Sanna-2 n = 8), using direct currents (DC) according to the methodology of ten Broeke et al. [34]. An electrode was inserted in the potting soil or agar respectively, and a thin gold wire (1.5 cm length for intact plants, 1 cm length for leaf discs) was gently attached to the dorsum of 8 to 11 day old wingless aphids with silver glue. The electrical circuit was completed when the aphid's piercing-sucking mouthparts penetrated the plant cuticle and the electrical signals, correlated to stylet activities, were recorded instantly [25]. Each replicate consisted of a different aphid and plant individual, employing one leaf disc per plant. Leaf discs of 9 mm in diameter were processed just below the apex of intermediately aged Arabidopsis leaves and placed abaxial side up in a Petri dish on a 1% agar substrate. A transparent plastic sheet covered the agar surrounding the leaf disc to prevent aphids to get stuck or make probes in the agar. Aphids that did not start probing within the first 3 hours of the observation were excluded from the analysis. EPG recording of winged N. ribisnigri biotype Nr:0 on lettuce cultivars Corbana and Terlana has been made in a previous study by ten Broeke et al. [47] (8hour recording, n = 19).

Aphid population development
One M. persicae neonate (0 to 24 h old) was transferred to each Arabidopsis plant in a climate chamber (21 ± 1°C, 50-70% relative humidity, an 8/16 h day/night cycle, light intensity of 200 μmol m −2 s −1 ). In the first assay 2.5-weekold plants were infested, in the second assay 3.5-week-old plants. A soap-diluted water barrier prevented aphids from moving between plants. Six, seven, and eight days after introduction the presence of the aphid and its offspring was checked. None of the aphids developed wings.
14 Days after infestation the number of aphids was counted per plant. Plants without an adult aphid 8 days after introduction, and plants without any adults or neonates 14 days after introduction were excluded from the analysis (assay 1: Co-2 n = 19, Sanna-2 n = 15; assay 2: Co-2 n = 14, Sanna-2 n = 13).

Simulations
In simulations, 10 4 random draws were taken from a normal distribution with the mean and standard deviation of a response variable of the Arabidopsis-M. persicae and lettuce-N. ribisnigri data sets (Additional file 1: Table S3). For video observations data was simulated with two probing variables: the mean duration of long probes and the total duration of sustained probes. For EPGs the total duration of phloem ingestion was simulated. Random draws were excluded when values were below zero, below the minimum duration of the probe category, or above the maximum recording duration. The generated data sets were subsampled with 1000 iterations without replacement for several replicate levels (n = 10,15,20,25,30,35,40). Student's t-tests were executed for each iteration and the percentage of significant p-values per replicate level was calculated. Video tracking simulation tests were defined significant if they had a P-value below α = 0.025 for at least one of the two probing variables (Bonferroni correction: α = 0.05/2). For EPG simulations one variable and P-values below α = 0.05 already delivered maximum detection rates. This process was performed on complete data sets of EPG and video recording (8 h observations) and on data sets rescaled to shorter durations (6 and 4 hour observations). The proportion of tests where the null hypothesis is incorrectly rejected, was calculated with simulations based on a data set of 8-hour video recording of M. persicae on Arabidopsis accession Col-0 (data set n = 53, replicate levels n = 15 and n = 20, two variables, P < 0.025, Additional file 1: Table S3).

Statistical analysis
An R script was written to calculate response variables of video tracking, such as the total number and total duration of short and long probes in each observation and for each hour (Additional file 4). For EPG recording, the start time and duration of waveforms were analysed with the EPG PROBE 3.0 software (EPG-Systems, Wageningen, The Netherlands). Further calculations and analyses of EPG data were performed with the statistical computing program R. The duration of phloem ingestion events in EPG recording were calculated as the sum of three subsequent waveforms: (a) inter-and intracellular penetrations followed by (b) phloem salivation and (c) phloem ingestion. Bar graphs were produced with the R package sciplot version 1.1-0 (Morales 2012) [58].
Data distributions and homogeneity of variances were tested with a Shapiro test and a Levene's test. In case data transformations (square root, log, logit, arcsine) did not result in a distribution that approaches a normal distribution, non-parametric tests were applied. Human observations were compared to video tracking parameters with a paired t-test or, when data were not normally distributed with a Wilcoxon signed ranks matched pairs test. Correlations were tested with a Pearson correlation test or, when data were not normally distributed, with a Spearman correlation test. For benchmarking of video tracking against EPGs with susceptible and resistant Arabidopsis and lettuce lines and for the reproduction assay, response variables were tested with a Student's ttest, or when the data were not normally distributed with a Mann-Whitney U test. Walking activity of aphids was tested across 8 time bins of 1 hour. The distance moved was not normally distributed and, therefore, transformed to ranks ranging from the lowest to highest value within the complete data set. A mixed linear model was applied on the ranks, using plant line, time bin, and plant line x time bin interaction as fixed effects and plant/aphid individual as a random effect.

Additional files
Additional file 1: Figure S1. Arena settings. Figure S2. Trial Control Settings. Table S1. Subject detection settings for M. persicae. Table S2. Absolute differences between EPG recording and automated video tracking. Table S3. Video tracking variables used to generate simulations.
Additional file 2: Movie of aphid feeding behaviour measured by manual annotations and automated video tracking simultaneously. Events were estimated where the M. persicae aphid penetrated the Arabidopsis leaf disc with its piercing-sucking mouth parts. Manual scoring of probes was achieved with a side-view camera mounted on a stereo microscope and The Observer XT software. Automated video tracking was performed with a top view on the arena (275 pixels/mm) and EthoVision XT software (Noldus Information Technology bv, Wageningen, The Netherlands). EthoVision XT recorded a probe when the aphid's centre point was located on the leaf disc and did not move. Other motion parameters, such as the distance moved (cm) and velocity (cm/s), were also acquired with automated tracking. The movie is displayed at double speed.
Additional file 3: Movie of automated video tracking of 20 aphids simultaneously. M. persicae aphids were tracked on Arabidopsis leaf discs using EthoVision XT video tracking software (Noldus Information Technology bv, Wageningen, The Netherlands). The aphid's centre point (red) was automatically detected and tracks of the previous 30 seconds are visualised in red. Arenas are numbered from 1 (top left) to 20 (bottom right). The movie is displayed at 16× speed.
Additional file 4: R script for feature extraction from EthoVision XT output.

Competing interests
Two authors, OK and LN, contributed to the development of the video tracking platform and are affiliated with Noldus Information Technology bv. All involved authors confirm that the contents of this manuscript have not been affected by any competing interest whatsoever.