High throughput real-time RT-PCR assays for specific detection of cassava brown streak disease causal viruses, and their application to testing of planting material

Authors


E-mail: ian.adams@fera.gsi.gov.uk

Abstract

Cassava brown streak disease (CBSD) caused by Cassava brown streak virus (CBSV) and Ugandan cassava brown streak virus (UCBSV) is causing severe losses in cassava production in Kenya, Tanzania and Uganda. Two real-time RT-PCR assays based on TaqMan chemistry capable of detecting and distinguishing these two viruses are described. These assays were used to screen 493 cassava samples collected from western and coastal Kenya, the main cassava regions of Uganda and inland Tanzania. Both viruses were found in all three countries and across regions therein. Association of CBSD leaf symptom status with CBSV and UCBSV assay results was weak, confirming the need for a diagnostic assay. For leaf samples that were observed with CBSD-like leaf symptoms but shown as CBSV and UCBSV negative by the RT-PCR assay, deep sequencing using a Roche 454 GS-FLX was used to provide additional evidence for the absence of the viruses. The probability of the CBSD associated diagnostics detecting a single CBSV or UCBSV positive sample amongst other non-CBSD samples was modelled. The results of this study are discussed in the context of the application of diagnostics of CBSD-associated viruses under the Great Lakes Cassava Initiative and the need to minimize the risk of further spread of the viruses with cassava multiplication material. It is shown that high throughput testing undertaken at Fera of 300 cassava leaves taken from fields for seed multiplication, when analysed in pools of 10, has given a 95% probability of detecting 1% infected plants in the field.

Introduction

Cassava is a major subsistence crop in many parts of the world, providing more than half of the dietary calories for half of the population of sub-Saharan Africa (Abarshi et al., 2010). In recent years, cassava crops in the east African countries of Kenya, Tanzania and Uganda have seen an emergence of cassava brown streak disease (CBSD), with reported yield losses up to 70% (Hillocks & Jennings, 2003; Mbanzibwa et al., 2011). The spread of the virus within these countries through the movement of planting material is very evident (Alicai et al., 2007) and there is a high likelihood of further spread beyond these countries.

Cassava brown streak disease was first described in the 1930s (Storey, 1936). However, the causal organism was not identified until the end of the 20th century with the molecular characterization of a single virus: Cassava brown streak virus (CBSV) (Monger et al., 2001b). Recent studies have shown that CBSD is caused by two viruses that are sufficiently genetically distinct to be given species status (Winter et al., 2010; Mbanzibwa et al., 2011). These two species have been described in the literature as CBSV for the Tanzanian or coastland strain and Ugandan cassava brown streak virus (UCBSV) for the Ugandan or highland strain (Mbanzibwa et al., 2009; Monger et al., 2010; Winter et al., 2010). The name UCBSV has recently been approved by the International Committee on Taxonomy of Viruses, with CBSV retaining its original species nomenclature. CBSV and UCBSV supersede other CBSV strain terminologies and the two viruses are collectively termed CBSVs (Legg et al., 2011).

Identification of CBSVs based on symptoms is reported as unreliable because the symptoms are inconsistently expressed in leaf, stem and root and are difficult to distinguish from mite damage and nutrient disorders (Hillocks & Jennings, 2003). Also, there is no evidence of any symptom differences between CBSV and UCBSV, making differentiation of CBSVs by visual symptoms impossible. For these reasons a diagnostic test is required to accurately identify the presence of viruses for research, policy (e.g. quarantine) and planting material multiplication purposes.

The first RT-PCR assay for the detection of CBSV was developed using only the small number of sequences available at the time (Monger et al., 2001a), and as a result this assay has proven to not be comprehensive for all strains of CBSVs. Recently, gel-based conventional RT-PCR assays for the detection of both viruses (Mbanzibwa et al., 2011) and a real-time RT-PCR assay for CBSV have been developed (Moreno et al., 2011). An ELISA based assay is also marketed by DSMZ. This paper describes the first real-time RT-PCR assay for the detection of CBSV and UCBSV. Real-time PCR is widely recognized as providing greater sensitivity than both ELISA and conventional PCR. However, the most significant advantages of the technique are realized in routine testing of samples: the absence of a gel electrophoresis step enables larger sample numbers to be processed at a reduced cost and the closed tube system effectively eliminates post-PCR contamination and resultant false positive results. Next generation 454 sequencing was also used to investigate the presence, or otherwise, of known, or new, CBSV genotypes in samples which appear to have CBSD-like symptoms, but tested negative using the real-time RT-PCR assays.

Aside from the specificity of the diagnostics, consideration has been given to the CBSVs assay application in establishing the lower limits of virus detection and to extrapolate this sensitivity into real sampling regimes and probabilities of detecting infection. In the case of applying diagnostics to certification, it is required at the field scale to look at a relatively large number of samples to have sufficient statistical power and to be giving assurances of very low levels of infection, often zero tolerance. In these cases, to address the sample size and economies of budget, pooling of samples is required, and the question posed is the probability of detecting a single low-titre infection within a pooled sample.

The diagnostics described herein have been developed under the Great Lakes Cassava Initiative (GLCI; http://iglci.crs.org/Pages/Default.aspx) for the high throughput testing of cassava samples sourced from plants intended for vegetative seed production or farmer field survey work. Specifically, the GLCI has sort assurances on the health status of cassava material within its cassava vegetative multiplication source sites prior to dissemination to farmers. The work described provides a full investigation on the systems developed in addressing sensitivity and the probable likelihoods of detecting a single leaf with low titre of CBSVs when pooled with various numbers of healthy (non-CBSD) leaf samples. Accordingly, the diagnostics have been assessed so as to measure uncertainty in detecting plants positive for CBSV or UCBSV in a GLCI source site.

Materials and methods

Sample collection

A total of 493 leaf samples were collected in September and October 2009 from the coastal region of Kenya north of Mombasa (37); western Kenya (305); the Mwanza province of Tanzania (56); and the main cassava growing regions of Uganda (95). Figure 1 shows a map of all sample origins. All collections were undertaken with the assistance of national researchers familiar with cassava and CBSD symptoms. Samples were collected from each sampled field based on CBSD leaf symptom severity only: rated as symptomless (class 1), mild (class 2), moderate (class 3), moderate-severe (class 4) and severe (class 5). Other CBSD symptoms of root necrosis and stem (lesion and dieback) were not considered in the classification. The samples were labelled with GIS coordinates, location name, farmer name and CBSD severity class. The samples were dried prior to shipping to the UK and stored dry prior to extraction. Samples from western Kenya had retained some moisture during drying in Kenya and had started to degrade during transit to Fera. These samples were frozen at −80°C to prevent further degradation and to kill contaminating insect pests. At the time of sampling it was noted that CBSD symptoms were clear in Uganda, the coastal area of Kenya and Tanzania, but were less clear in western Kenya as a result of conspicuous green mite damage.

Figure 1.

 Geographical origins of the cassava samples used in this study.

CBSV positive control material was obtained from DSMZ and N. M. Maruthi (NRI, University of Greenwich, UK).

RNA extraction

Samples (100 mg) of dried leaves were extracted as described in Boonham et al. (2009). Briefly, samples were placed in 5 mL tubes containing a single ball bearing (diameter 11 mm) and shaken at maximum speed in a Merris automixer (Merris Engineering Ltd) for 2 min, prior to centrifugation at 6000 g for 1 min. Lysis buffer (5 mL) containing: 8 m guanidine–HCl, 25 mm EDTA, 3% w/v polyvinylpyrrolidone, 0·5% v/v Triton X-100, 25 mm citric acid, 1 mm tri-sodium citrate, 1% antifoam B (Sigma) was added and the sample shaken for a further 2 min. The tubes were centrifuged for 5 min at 6000 g and the supernatants transferred to a 96 deep-well plate using a Tecan freedom evo robot (Tecan UK Ltd). MAP beads (40 μL; Invitek) were added to each sample and the plates placed on a Kingfisher 96 robotic magnetic particle processor (Thermo Scientific). The robot was then used to pass the magnetic beads through a series of 96-well plates containing per sample: 1 mL lysis buffer; 1 mL 70% v/v ethanol; 1 mL 70% v/v ethanol and 200 μL water. The RNA was diluted to a final volume of 1 mL in water to allow PCR setup with robotic liquid handling.

The efficacy of the extraction was examined by testing each sample using a real-time assay designed to detect expression of the plant cytochrome oxidase gene (Weller et al., 2000). Samples which gave negative results using this assay were removed from further analysis to avoid false negative results.

Real-time RT-PCR

Real-time RT-PCR assays based on TaqMan chemistry were designed using an alignment of published CBSVs genomes (Mbanzibwa et al., 2009; Monger et al., 2010; Winter et al., 2010) along with a sequence of the Nampula isolate obtained from N. M. Maruthi (NRI, University of Greenwich, UK) and sequenced at Fera using the methods described by Monger et al. (2010). The alignment was created with the software package mega 4.1 (Kumar et al., 2008). Primer and probe sequences were chosen using the software package primer express 2 (Applied Biosystems). Prospective primers were analysed using blast (Altschul et al., 1997) against the GenBank nr/nt DNA database (Benson et al., 2010) to investigate unwanted cross reactions to plant and other viral sequences and confirm specificity to CBSV sequences. The primer sequences for CBSV were: forward 5′-GCCAACTARAACTCGAAGTCCATT-3′, reverse 5′-TTCAGTTGTTTAAGCAGTTCGTTCA-3′ and the TaqMan probe 5′-FAM- AGTCAAGGAGGCTTCGTGCYCCTC-BHQ1-3′; those for UCBSV were: forward 5′- GATYAARAAGACITTCAAGCCTCCAAA-3′, reverse 5′- AATTACATCAGGRGTTAGRTTRTCCCTT-3′ and the TaqMan probe 5′-TET- TCAGCTTACATTTGGATTCCACGCTCTCA-BHQ1-3′.

Real-time RT-PCR was performed on 384 well plates set up using a Micro Star liquid handling robot (Hamilton), using core reagent kits (Applied Biosystems). Reactions consisted of 1 ×  buffer A, 0·025 U μL−1 AmpliTaq Gold (Applied Biosystems), 0·4 U μL−1 Revertaid (Fermentas), 0·2 mm of each dNTP, 5·5 mm MgCl2, 300 nm of each primer, 100 nm of probe and 10 μL of extracted and diluted RNA, to give a final reaction volume of 25 μL. The reactions were incubated for 30 min at 48°C, 10 min at 95°C, then cycled 40 times for 15 s at 95°C and 1 min at 60°C on an Applied Biosystems 9700HT, with data collected during the 60°C stage. Negative controls consisted of water replacing the RNA, and RNA from plants which had previously tested negative for CBSVs, and positive controls consisted of RNA from plants known to be infected with CBSVs.

PCR product sequencing

PCR products were cloned using the pGEM-T Easy cloning system (Promega) and sequenced using Sanger sequencing by Eurofins.

Next generation sequencing

Double-stranded cDNA for sequencing was produced from RNA extracted from plants using the cDNA amplification method described by Adams et al. (2009) with the modification that only the Tag dT primer was used for first strand synthesis to target polyA-tailed viral RNA and avoid the large quantities of plant ribosomal RNA present in the extracts. Sequencing was performed on a GS-FLX 454 sequencer using rapid libraries and titanium chemistry in accordance with the manufacturer’s instructions (Roche). Multiplex identifiers (MiDs) were used to mix multiple samples in the same region of the sequencer plate. Sequencing analysis was by reference assembly of the sequences to known CBSV and UCBSV genomes using gsMapper v. 2.4 software (Roche).

Estimates for limits of detection and false positive probability

An upper limit (95% confidence) for the false positive probability was estimated from results generated by the analysis of cassava leaves taken from a separate study of the GLCI from Rwanda and Burundi, countries believed to be free of CBSV at the time that the samples were taken (2008–2011).

The purpose of the analytical method is to detect the presence of infected plants in cassava fields. Two factors affect the limit of detection when the analytical method is used for this purpose: the number of plants tested and the analytical false negative rate. If n plants are tested selected from a field at random, and the analytical method is perfect, then the limit of detection (for a 95% probability of detection) is given by 1–0·051/n. Hence, for a sample of 300 leaves with a perfect analytical method, the limit of detection is a little under 1% of plants in a field (0·99%). The question then arises, ‘How should 300 leaves be analysed?’ At the extremes, one method would be to put the 300 leaves in a single analytical sample and try to detect the presence of a single positive leaf in the pool of 300 leaves (at the risk of the dilution effect rendering the positive leaf undetectable). Alternatively, each of the 300 leaves could be tested individually (300 tests are expensive). The aim is to find the largest pool size that maintains a limit of detection close to 1%.

The effect of dilution by pooling leaves was estimated by mixing leaves with CBSV or UCBSV with uninfected leaf material. Mixtures were homogenized in buffer to simulate pooling of leaves from a field at five levels (1 positive leaf per 1, 4, 16, 64 and 256 leaves). The pooled homogenates were then processed and tested using the real-time assays. Four replicates were performed at each level.

Data from the screening experiment was used to estimate the Ct variation that may normally be expected amongst samples positive for CBSVs.

Results

Development of assays

A number of different primer and TaqMan probe sets were designed to genome regions encoding Ham1, coat (CP) and P1 proteins that were conserved within the specific CBSD-associated viral species but divergent between them. These assays were tested on RNA samples extracted from CBSV and UCBSV infected cassava. Efficacy of the primer sets was based on (i) ability to detect CBSVs independently and (ii) signal to noise ratio (the lowest Ct values with the highest ΔRn values when compared to assays designed to the other genes). From this assessment the CP based assays were shown to possess the optimal qualities (data not shown). The assays selected were specific for their particular CBSV species with no increase in fluorescence associated with the presence of RNA of the other species (Fig. 2). A nucleotide sequence alignment of the CP-encoding region chosen is shown in Figure 3.

Figure 2.

 Amplification plots following real-time RT-PCR using the CBSV and UCBSV assays testing CBSV and UCBSV infected cassava material.

Figure 3.

 Alignment of sequence for the regions of the CBSV and UCBSV genomes, highlighting the location of the primers and probes for (a) the UCBSV and (b) the CBSV assays. Areas in black are conserved.

To enable multiplexing of the two assays, the optimum concentrations of primers and probe (300 nm of each primer and 100 nm probe) were determined as described in Mumford et al. (2000). When the assays were combined and used to test material containing CBSV and UCBSV, either at similar concentrations or with one species at 100 times lower concentration, it was found that the presence of CBSV reduced the limit of detection of the UCBSV assay. However, the presence of UCBSV did not adversely affect the detection of CBSV (data not shown). Based on these data all samples were subsequently tested using the assays in simplex to avoid false negative reporting on UCBSV.

Screening of samples and validation of assays

The results following testing of all the RNA samples with the cytochrome oxidase real-time assay indicated that all extractions of RNA from samples from coastal Kenya, Uganda or Tanzania were successful, whereas 21% of the samples from western Kenya failed to provide PCR amplifiable cDNA, presumably due to the degraded condition of the samples. In addition, a higher proportion of samples from western Kenya were CBSV-negative compared to plants with similar symptoms from other regions and, where CBSV was detected, the Ct value was generally between 2·5 and 6 Ct higher than results generated from leaves with the same symptom score taken from other regions. Hence, results from western Kenya were not used in the assessment of the assays practical limit of detection when applied to the analysis of pools of leaves because they were not representative of the performance of the method on non-degraded leaves.

All samples that were cytochrome oxidase-positive went forward for CBSVs testing. In total 429 samples were analysed by the real-time RT-PCR CBSV assay (Table 1). UCBSV and CBSV were detected in samples from all areas, with greater or lesser frequency: CBSV positive samples ranged from only one sample (from 89 CBSVs positives) in Uganda, through to approximately 20% (5 from 26) of positive samples in coastal Kenya, 33% (12 from 36) of positive samples from western Kenya, and 55% (25 from 45) of positive samples from inland Tanzania. Some samples from coastal Kenya (4 from 26) and inland Tanzania (21 from 45) gave positive PCR results with both primer sets, whereas no dual positives were recorded from samples from western Kenya and Uganda.

Table 1. Numbers of samples infected/uninfected with CBSV or UCBSV found in different regions of eastern Africa. The dual infected samples are also included in the species-specific columns
RegionSymptom classaCBSV positiveUCBSV positiveDual infectionNegativeTotal
  1. aClass 1, symptomless; class 2–3, mild-moderate symptoms; class 4–5, severe symptoms.

Kenya-coastal (5 sites)1383816
2–32111215
4–506016
Total52541137
Kenya-western (6 sites)1040101105
2–312180104134
4–502002
Total12240205241
Tanzania (13 sites)112146
2–311199526
4–5132011224
Total2541211156
Uganda (20 sites)102057
2–30800180
4–516007
Total1880695

For plants showing symptoms, RT-PCR-based disease status and disease status based on CBSD symptom classification agreed in most cases, with the exception of results produced by the analysis of the degraded leaves from western Kenya (Table 2). In general, plants assessed as showing symptoms reliably produced positive test results; 2–9% of plants with symptoms can be expected to produce a negative test result. However, 25–59% of symptomless plants taken from the sampling areas used in this study can be expected to give a positive test result. The PCR products from a number of these symptomless test-positive samples were DNA sequenced, confirming the presence of CBSV or UCBSV.

Table 2. Estimates of extent of agreement between observed disease symptoms and CBSV test results in four East African areas (ranges give 95% confidence interval)
 Kenya-coastalKenya-westernaTanzaniaUgandaCombined (excluding K-w)
  1. a(low), significantly lower than each of the proportions observed in all other regions ( 0·05); (high), significantly higher than each of the proportions observed in all other regions ( 0·05). All other proportions that describe analytical performance are consistent across regions.

% with symptoms in sampling frame41–7250–6374–9483–96 
% positive test in sampling frame54–8311–20 (low)74–9485–97 
% of symptomless with positive test27–731–9 (low)8–716–6525–59
% with symptoms with negative test4–3369–83 (high)1–160·2–62–9

The infection status of samples that were assessed to have visible CBSD symptoms but which gave negative results following real-time RT-PCR testing was assessed by subjecting the RNA from these plants to analysis by 454 sequencing. Twenty-one samples were analysed: 12 with symptoms but real-time assay negative samples, along with nine medium and low titre (based on Ct values) positive samples containing CBSV and UCBSV. In total the next generation sequencing produced 94 771 individual DNA sequences (Table 3). No virus sequences were found in any of the 11 symptom-positive/assay-negative samples, whereas sequences originating from CBSVs were detected in all the nine assay-positive samples.

Table 3. Details of the cassava samples used for deep sequencing of CBSV and UCBSV, and number of sequencing reads obtained
Origin Ct ReadsCBSVs reads
UCBSVCBSV
Visual symptom positive/assay negative samples
 Uganda4040  1819 0
 Tanzania4040  2158 0
 Tanzania4040  8949 0
 Tanzania4040  4281 0
 Tanzania4040  6489 0
 Tanzania404016 244 0
 Tanzania4040  2946 0
 Tanzania4040  1780 0
 Kenya4040  1070 0
 Kenya4040  1070 0
 Kenya4040  1070 0
Positive control samples
 Uganda33·640  2892 3
 Uganda23·340  2691 2
 Uganda23·840  694812
 Tanzania4025·8  3202 2
 Tanzania2227  820225
 Tanzania242415 012 1
 Kenya2540  3052 1
 Kenya4027·7  3019 1
 Kenya29·540  1877 4

Limit of detection and false positive rate of the assay

A total of 1465 samples from Rwanda and Burundi were analysed with no positive results. Hence, the probability of a negative sample producing a false positive result is estimated to be no greater than 0·20% with 95% confidence.

Effect of dilution

Results produced from the pooling experiment were analysed by linear regression of the Ct score against log2 of the dilution factor to produce a prediction interval (based on one-tailed 95% confidence) for new observations (Fig. 4). From this analysis it is seen that both CBSV and UCBSV describe a similar dilution profile and can be considered as equivalent. Accordingly, it is shown that for both CBSV and UCBSV the expected change in Ct for a factor 2 dilution is 1·7 ± 0·2 and more generally log2 (d) × (1·7 ± 0·2) for a dilution factor d.

Figure 4.

 Relation between Ct and dilution factor of CBSV/UCBSV infected cassava material diluted in uninfected material. □ Tanzania leaves, ○ Uganda leaves, — expected Ct,- - - - - - prediction interval for new observations (95%, one-tailed).

Relation between symptom status and CBSV Ct

There was a weak relation between symptom class and CBSV Ct, with lower Ct in positive samples being associated with higher symptom class (average Ct of 27·9, 26·8, 26·1 and 25·5 for symptom classes 1, 2, 3 and 4/5, respectively) but with a large amount of variation within each symptom class (Fig. 5).

Figure 5.

 Boxplot showing the distribution of CBSV or UCBSV Ct results from the different regions, separated by visual symptom class (CBSD score).

False negative rate where a positive leaf is analysed in a pool of uninfected leaves

The distribution of Ct values <40 (i.e. all positive samples) from screened samples with infection status 1 or 2 (symptomless or mild symptoms) was found to follow an approximately normal distribution with mean 27·1 and standard deviation 2·7 with upper limits at 95% confidence for the mean and standard deviation at a mean = 27·8 and standard deviation = 3·2. Hence, the probability (PFN) of a pool of d leaves containing a single leaf either symptomless or with mild symptoms giving a false negative result (Ct ≥ 40) was estimated using:

image(1)

where m is the mean Ct of a positive leaf (using the upper limit of the estimate), s is the standard deviation of Cts of positive leaves (using the upper limit of the estimate), g is the gradient of Ct against dilution on the log2 scale (determined above), and Φ() is the standard normal cumulative distribution function.

Hence:

image(2)

Limit of detection

The effect of pooling leaves on the limit of detection LD of a testing protocol was assessed by calculating the smallest proportion of positive leaves in a field that will be detected with a probability of at least 95% where 300 leaves are selected at random and put into P pools each containing d leaves using:

image(3)

where LD is the minimum proportion of infected plants that will be detected with probability 1–β and P pools are tested each containing d leaves, and the probability of a pool of d leaves which contains a single positive leaf giving a false negative result (PFN) is calculated as described above using Eqn 2.

Hence, if: (i) estimates are used for 1–PFN based on the observed assay performance and Cts of leaves either symptomless or with mild symptoms, (ii) 300 leaves per field are tested and (iii) a 95% probability of detection is required (β = 0·05), then the limit of detection for the testing protocol is: (i) 0·99 infected plants in a field per 100 plants if a hypothetical perfect assay (PFN = 0) is used, (ii) between 1·01 and 1·03 infected plants per 100 using the real assay with 30 pools of 10 leaves each, from each field, (iii) between 1·04 and 1·13 infected plants per 100 using the real assay with 15 pools of 20 leaves each, from each field and (iv) between 1·07 and 1·26 infected plants per 100 using the real assay with 10 pools of 30 leaves each, from each field.

Table 4 shows the relation between pool size, false negative probability and limit of detection in a field for leaves either symptomless or with mild symptoms. It should be noted that the estimates of limit of detection are quite robust against uncertainty or variation in the false negative probability for a single pool of leaves. For example, for testing 300 leaves in 30 pools of 10 leaves, a false negative probability for a single pool is estimated as 2% and a consequent limit of detection of 1·02 infected plants per 100 in a field. However, even if the false negative probability is a factor of 10 larger (20%), the effect on the estimated limit of detection (Eqn 3) is to increase it from 1·02 infected plants per 100 to 1·26 infected plants per 100.

Table 4. Pool size applied to 300 plants, the false negative probability for a single pool, and limit of detection for CBSV in fields
Pool size (leaves)False negative probability for one poolLimit of detection (infected plants per 100 plants in field)
Estimate (%)95% confidence intervalEstimate (%)95% confidence interval
 1 0·01 0·01 0·010·990·990·99
 5 0·50 0·32 0·751·001·001·00
10 2·0 1·2 3·31·021·011·03
20 6·5 3·7111·071·041·13
3011 6·5181·151·071·26
502112321·361·181·72

Discussion

This research spans from late 2007 to 2011, during which time knowledge of the CBSVs genomes has changed substantially, from a few CP-encoding sequences to over 10 whole genomes and many more partial sequences. Most significantly during this period CBSD has been recognized as being caused by two highly diverse species, CBSV and UCBSV (Mbanzibwa et al., 2009; Monger et al., 2010; Winter et al., 2010). The RT-PCR assays described herein are the first for both viruses using real-time technology. To have the facility to test for the separate species is of evident importance for research on epidemiology and breeding, and policies on quarantine and certification. The significance of the two species is only now beginning to be recognized in current and future control strategies for CBSD.

This research has formed a body of evidence in support of the Great Lakes Cassava Initiative, a major cassava multiplication and dissemination project for the Great Lakes region of East Africa, that set out a zero tolerance standard in testing all cassava material for CBSV before further multiplication from the project source sites. The scale of the GLCI has been unprecedented in an African context, and a method for testing cassava material was needed that was specific and sensitive and able to handle large numbers of samples. This challenge was compounded by the uncertainty that surrounds the expression of CBSD symptoms, with the recognition that cassava material may be infected by CBSVs but apparently healthy by visual analysis.

The work described here has validated separate assays for CBSV and UCBSV. It has shown how the assays can be combined with an appropriate sampling scheme to provide a limit of detection of approximately 1% of infected plants in a population by testing 300 plants in pools of up to 10 plants, and how this limit of detection rises with pool size (Table 4). Thus these results allow for informed decision-making about the level of risk that may be associated with the application of the assays. Acceptance around a risk will vary by application and in the context of the resources available to support testing. The lowest levels of risk-tolerance are likely to be applied to adherence with policies on quarantine and seed certification. In the context of the GLCI and the testing of cassava material for further multiplication, the tolerance around detecting CBSVs has been set at a 95% confidence in detecting infected plants that make up 1% of the plants in a field. From the data here, this translates to testing in pooled subsamples of 10. Accordingly, the GLCI has tested 300 leaf samples (plants) from each of its source sites for multiplication, translating to 30 CBSV and 30 UCBSV tests when sub-pooled into 10 samples. To date Fera, under the GLCI, has tested approximately 500 multiplication fields, some 150 000 leaf samples requiring in excess of 45 000 PCR reactions.

A particular challenge to address was how to substantiate that the CBSV negative samples were free of CBSVs, especially when these were recorded as with CBSD symptoms by visual analysis by experienced field operatives. There remains a possibility that the assays may miss a CBSV variant given what is known about the unusual levels of genetic diversity associated within the species (Mbanzibwa et al., 2009; Monger et al., 2010; Winter et al., 2010). Next generation sequencing techniques were applied to investigate the presence of virus within these samples. The technique deployed (Adams et al., 2009) generates de novo sequence and as a result it would enable the identification of anything from minor sequence variants of CBSV/UCBSV that prevent amplification of the sequences using real-time RT-PCR to completely undescribed viruses that produce symptoms similar to CBSD. The data showed complete absence of CBSV sequences in the real-time RT-PCR negative samples, yet in all the low titre samples tested, CBSV sequence was recovered. To the authors’ knowledge this is the first time that deep sequencing has been applied in this way and although not providing absolute proof of absence of virus in assay negative samples, has provided an additional layer of supportive evidence.

Discounting the degraded western Kenyan samples, (Table 2; Fig. 5), the data reported here (high agreement between positive symptoms and test results; lower agreement between negative symptoms and test results) present some of the most robust evidence about the relation between CBSD, CBSV assay results and measures of Ct values, and support the description that CBSD is an elusive disease and that visual estimates of the absence of leaf symptoms alone can be an inadequate measure of virus absence in a particular field. Visual assessment is generally reliable when it has recognized that the disease is present. This is obviously an important outcome to quantify, and information that can be used to design integrated inspection and testing regimes when considering a research or phytosanitary question and the resources available to answer it. Under the GLCI at the level of multiplication for further multiplication, a zero tolerance was maintained for CBSV. However, as the scale of the multiplication increased, material that was for direct dissemination to farmers was inspected for CBSV by visual symptoms alone as it was not practicable to test by RT-PCR at the scale realized at this point of the scaling-up programme.

Whilst not based on an exhaustive sampling regime, the research is in broad agreement on a region by region basis with other recent surveys (Mbanzibwa et al., 2011) and has also provided further information on the distribution of CBSVs. Both studies found UCBSV to predominate in Uganda with a more even distribution between UCBSV and CBSV in the Lake Victoria region of Tanzania. The current study found UCBSV to predominate in the Kenyan lowlands and Mbanzibwa et al. (2011) found CBSV to predominate in the Tanzanian lowlands.

With the facility provided by the assays described here to undertake pathogenicity tests and other epidemiological studies directed at the level of CBSVs, it can be expected that knowledge on the biological significance of CBSV and UCBSV will begin to be understood.

The idea of variants of CBSVs has also been expanded on by these studies. For example, it was found that the CBSV samples from Uganda (but not the other regions) were largely not detected by an earlier iteration of the CBSV and UCBSV primers (data not shown). How distinct these CBSV isolates are over the complete genome has yet to be determined. The frequency of dual infections also presents interesting questions on association. Note that whereas the TaqMan primers developed here target different regions of the CBSV genome and therefore could be detecting a single hybrid strain, the occurrence of dual positives has been reported by Mbanzibwa et al. (2011) using universal primers and gel electrophoresis where the outcome can only be accounted for by the presence of both species.

Acknowledgements

This work was supported through a sub-grant with Catholic Relief Services as the grantee of the Bill and Melinda Gates Foundation award for the Great Lakes Cassava Initiative. Further support was received from grant number MSI/WA1/2/16/08 under the Uganda Millennium Science Initiative. The authors gratefully acknowledge the substantial assistance received from the staff of CRS and the national programmes of the GLCI countries.

Ancillary