A new real-time PCR detection system was developed for grapevine yellows (GY) using TaqMan minor groove binder probes and including two amplicons for group-specific detection of Flavescence dorée (FD) and Bois noir (BN) phytoplasmas, plus a universal phytoplasma amplicon. FD and BN amplicons were designed to amplify species-specific genomic DNA fragments and the universal amplicon to amplify the 16S ribosomal DNA region. Efficiency of PCR amplification, limit of detection, range of linearity and dynamic range were assessed for all three amplicons. The specificity of detection systems was tested on several other isolates of phytoplasmas and bacteria and on healthy field grapevine and insect samples. No cross-reactivity with other phytoplasma strains, plant or insect DNA was detected. The assay was compared with conventional PCR on more than 150 field grapevine, insect and field bindweed samples. Real-time PCR showed higher sensitivity as phytoplasmas were detected in several PCR-negative and in all PCR-positive samples. A data-mining analysis of results from both detection approaches also favoured real-time PCR over conventional PCR diagnostics. The developed procedure for detection of phytoplasmas in grapevine also included amplification of plant DNA co-extracted with phytoplasmic DNA, providing additional quality control for the DNA extraction and PCR amplification for each sample. The newly developed assay is a reliable, specific and sensitive method easily applicable to high-throughput diagnosis of GY.
Phytoplasmas are found in the phloem sieve tubes of plants (Lee et al., 2000; Garnier et al., 2001; Christensen et al., 2005) and in phloem-sucking homopterous insects (Webb et al., 1999; Christensen et al., 2005). Several distinct phytoplasma groups have been identified on the basis of RFLP or phylogenic analysis of 16S rRNA sequences (Lee et al., 1998; Seemüller et al., 1998). FD belongs to group 16SrV (elm yellows group) and BN to group 16SrXII-A (stolbur group). In 2004, the ‘Candidatus (Ca.) Phytoplasma’ taxon was introduced at the genus level and several taxons were introduced at the species level, based on genetic (16S rRNA sequence similarity) and biological properties (e.g. diseases caused in plants). The Flavescence dorée phytoplasma was classified into taxon ‘Ca. Phytoplasma vitis’ and Bois noir into ‘Ca. Phytoplasma solani’ (IRPCM, 2004; Lee et al., 2004).
FD has the greatest economic impact among GY-causing phytoplasmas. The most important reasons are high yield loss and high epidemiological potential because of the ubiquitous presence of the cicadellid Scaphoideus titanus, its transmission vector in European vineyards. In addition, the application of insecticides to control the insect vector adds to the costs of production (Lee et al., 2000; Seemüller et al., 2002; Boudon-Padieu, 2003; EPPO/CABI, 2003). FD is recognized by the European and Mediterranean Plant Protection Organization (EPPO) as a quarantine pest (EPPO/CABI, 2003). Although not as economically important as FD, BN is present in most wine-producing regions of Europe. Its main vector is a planthopper, Hyalestes obsoletus (Boudon-Padieu, 2003). Reliable and sensitive methods of detection and differentiation of FD from other GY-causing phytoplasmas are therefore required for efficient control of the disease.
Several methods for FD and BN detection have been developed:
(iv) More recently, real-time PCR based assays for universal phytoplasma detection (Christensen et al., 2004), as well as apple-proliferation-group-specific (Baric & Dalla-Via, 2004; Torres et al., 2005) and FD-specific (16SrV-C and 16SrV-D subgroup-specific) (Bianco et al., 2004) detection methods. Additionally, Galetto et al. (2005) developed real-time PCR assays for the specific detection of FD, BN and apple proliferation phytoplasma (SYBR® Green), and an assay for common detection of 16SrV, 16SrX and 16SrXII phytoplasma groups (TaqMan®).
In comparison to PCR/RFLP assays, real-time PCR methods may allow faster and more sensitive detection of phytoplasmas. Their levels of specificity and sensitivity can vary greatly depending on the real-time PCR chemistry chosen. For example, the SYBR® Green chemistry used by Galetto et al. (2005) is less specific than TaqMan®, which was used by Bianco et al. (2004) and in the common assay by Galetto et al. (2005). The use of TaqMan® MGB Probes (Minor Groove Binder Probes, Afonina et al., 1997) in this study greatly diminishes the possibility of detecting non-specific target DNA. Sensitivities of all published real-time PCR assays for detection of phytoplasmas were either not compared with the conventional phytoplasma detection (PCR/RFLP) or showed even lower sensitivity.
The aim of this work was to develop fast, reliable and sensitive real-time PCR detection systems based on TaqMan® MGB Probe chemistry for specific detection of BN and FD phytoplasmas, and for universal phytoplasma detection, which would enable enhanced diagnostic sensitivity and specificity compared to previously published detection systems. The approach included additional quality control steps to increase the confidence in results. Furthermore, a data-mining technique was used for classification and construction of decision trees in order to evaluate and compare the newly-developed real-time PCR and conventional PCR-based detection systems.
Materials and methods
Plant, insect and bacterial material
Phytoplasma strains for specificity testing were cultivated in Catharanthus roseus and were kindly provided by Dr R. Osler from the University of Udine, Italy. The strains included: elm yellows phytoplasma strain EY1 (EY1, 16SrV-A), aster yellows phytoplasma (AY, 16SrI-B), a stolbur phytoplasma isolate SE (SE, 16SrXII-A, original host: celery), Western X-disease transferred to C. roseus (WX, 16SrIII-A), apple proliferation phytoplasma strain AP15 (AP, 16SrX-A), pear decline phytoplasma (PD, 16SrX-C) and the European stonefruit yellows phytoplasma transferred to C. roseus from plum (ESFY, 16SrX-B). MA phytoplasma (16SrIII-B), originating from Chrysanthemum leucanthemum (Marguerite Daisy) cultivated on Tanacetum cinerariifolium, was also included in the study.
Several strains of plant pathogenic bacteria were obtained from the National Collection of Plant Pathogenic Bacteria (NCPPB), York, UK for specificity testing. These were: Xylophilus ampelinus (NCPPB 2217, type strain of the species), Pseudomonas syringae pv. syringae (NCPPB 281), Erwinia amylovora (NCPPB 683, type strain of the species) and Clavibacter michiganensis subsp. sepedonicus (NCPPB 4053). Agrobacterium vitis (KIS Av 13-2) was obtained from the collection of the Agricultural Institute of Slovenia, Ljubljana, Slovenia. Some of these strains are present in the grapevine microflora. Additionally, 29 bacterial isolates from plant extracts of various grapevine cultivars grown on nutrient agar (NA) (Bacto Nutrient Agar, Difco) or YPGA (yeast extract 7·0 g, proteose peptone 5·0 g, glucose 10·0 g, agar 15·0 g, distilled water 1 L, pH 7·0) were used. An additional five strains of grapevine pathogenic bacteria employed in this study were obtained from the collection of the Instituto Valenciano de Investigaciones Agrarias (Valencia, Spain) and were characterized as Agrobacterium vitis, A. tumefaciens, A. rhizogenes, Xylella fastidiosa Fetz and X. fastidiosa Staps.
Grapevine leaf samples were collected in the field from 148 grapevine plants (17 cultivars: Chardonnay, Beli Pinot, Sivi Pinot, Modri Pinot, Merlot, Refošk, Modra Frankinja, Sauvignon, Rebula, Šipon, Renski Rizling, Rumeni Muškat, Žametovka, Malvazija, Kerner, Zweigeld and Syrah). All the main coastal and continental winegrowing regions of Slovenia (Primorje, Posavje and Podravje) were represented. Three field-collected samples of field bindweed (Convolvulus arvensis) were also tested. Healthy field-collected grapevine plants were used as negative controls in real-time PCR reactions (cvs Barbera, Chardonnay, Sivi Pinot, Merlot and Refošk). The following sweep-captured insects were also tested: Hyalesthes obsoletus (four samples of at least 20 adults), Scaphoideus titanus (10 samples of at least 20 adults and one sample of at least 20 nymphs), Euscelis incisus (one sample of seven adults) and Reptalus panzeri (three samples of at least 20 adults). Sampling of grapevine plants was carried out from July to October 2005. An additional 12 samples of field-collected apple trees (Malus domestica, leaf veins) that contained apple-proliferation-group phytoplasmas (16SrX) were tested. An additional 159 grapevine leaf samples (the same range of cultivars and geographical locations as in 2005) were collected from July to October 2006 and were analysed with real-time PCR only.
DNA samples from FD-infected (four samples) and BN-infected grapevine plants (one sample) were provided by Dr P. Ermacora of the University of Udine, Italy.
The phytoplasma-enrichment procedure described by Ahrens & Seemüller (1992) was used, with some modifications; 1·5 g of leaf veins (whole leaves in the case of C. roseus and C. arvensis) were used as starting material, homogenized in 20 mL ice-cold grinding buffer and centrifuged for 5 min at 2500 g (4°C) followed by centrifugation of the supernatant for 25 min at 18 000 g (4°C). Lysis was performed in modified CTAB buffer (4% CTAB, 1% PVP MW 10 000, without 2-mercaptoethanol) at 65°C. After chloroform/isoamyl alcohol (24:1) extraction the DNA was precipitated with cold (–20°C) isopropanol, incubated at room temperature for 20 min and centrifuged for 15 min at 15 000 g. The pellet was then washed with cold 80% ethanol (–20°C), dried at room temperature and re-suspended in 50 µL TE buffer (10 mm Tris-HCl, 1 mm EDTA, pH 8·0) overnight at 4°C. Quantity and quality of DNA were analysed using gel electrophoresis (8 µL). The DNA concentrations of all phytoplasma and bacterial strains used in specificity testing and of several diagnostic samples (to estimate the range of concentrations) were measured spectrophotometrically with NanoDrop (NanoDrop Technologies). Tenfold dilutions of the samples were shown to be sufficient to keep the DNA load below 100 ng in real-time PCR reactions.
Conventional PCR assays
Each field-collected DNA sample from 2005 was tested for the presence of FD and BN using a set of PCR reactions:
(i) Specific detection of BN (Stol11 genomic fragment) using a STOL11f2/r1 primer pair (Daire et al., 1997) followed by a nested PCR reaction with STOL11f3/r2 (Clair et al., 2003);
(iii) Using the universal primer pair F6/P7, which is actually a modified P1/P7 primer pair (Schneider et al., 1995): F6 is a modified P1 primer, shortened by two nucleotides at the 5′ and one nucleotide at the 3′ end (Dr P. Ermacora, University of Udine, Italy, personal communication). In the absence of a positive result, a nested PCR reaction with a U5/U3 primer pair (Lorenz et al., 1995) was performed on F6/P7 products.
All PCR assays were performed in 40-µL reactions with 2 µL of DNA (in the case of nested PCR reactions, original PCR products were diluted 100-fold) on a PCR System 9700 GeneAmp PCR cycler (Perkin Elmer). The manufacturer of all chemicals used in this study was Applied Biosystems. Reaction mixtures and conditions for amplification are summarized in Table 1.
Table 1. Amplification conditions of conventional PCR assays for phytoplasma detection
All real-time PCR reactions were performed on an ABI PRISM® 7900 HT Sequence Detection System (Applied Biosystems) in optical 384-well plates with optical adhesive covers (both Applied Biosystems) using universal cycling conditions (2 min at 50°C, 10 min at 95°C, followed by 45 cycles of 15 s at 95°C and 1 min at 60°C, with 9600 Emulation mode) which allowed running of all reactions on the same plate. Real-time PCR was performed in a final reaction volume of 10 µL containing 2 µL of sample DNA, 900 nm primers, 250 nm probe and 1× TaqMan® Universal PCR Master Mix (Applied Biosystems), which includes ROXTM as a passive reference dye. Each sample DNA was tested with all three amplicons for phytoplasmas and an amplicon for an endogenous control (in separate real-time PCR reactions): cytochrome oxidase (COX) for plants (Weller et al., 2000) or 18S rRNA for insects (eukaryotic 18S rRNA TaqMan endogenous control, Applied Biosystems). Endogenous controls amplified plant or insect DNA co-extracted with phytoplasmic DNA. All reactions were performed in two replicate wells in two dilutions (10- and 100-fold) on the same 384-well plate. An automated liquid handling system (Multiprobe® II PLUS EX, PerkinElmer) was used for pipetting a large number of DNA samples and master mixes onto the 384-well plates.
The software sds 2·2 (Applied Biosystems) was used for fluorescence acquisition and calculation of threshold cycles (Ct). For this calculation, the baseline was automatically set and the fluorescent threshold was set manually to intersect with the linear part of amplification curves of all amplicons, resulting in the final Ct value for each well.
Specificity of amplicons
The specificity of all three real-time PCR amplicons was determined by testing the cross-reactivity with: (i) several phytoplasma strains cultivated in C. roseus; (ii) plant DNA extracted from leaf veins of several cultivars of symptomless field-grown grapevine plants; and (iii) several bacterial strains that can be present as epiphytes or saprophytes on grapevine plants and their tissues. For five of the bacterial isolates, DNA extracted from pure cultures was used (four from NCPPB and one from the Agricultural Institute of Slovenia). For 34 other isolates from grapevine plants, pure cultures grown on NA or YPGA were transferred to 100 µL of water and used in colony real-time PCR.
Real-time PCR products were visualized under UV light on 10% polyacrylamide gel.
Efficiency, limit of detection (LOD) and sensitivity of amplicons
Elm yellows phytoplasma strain EY1 is an American isolate of the elm yellows phytoplasma (‘Ca. Phytoplasma ulmi’). It was used to develop the real-time PCR tests because it has the same target sequence as FD (Figs 1 and 3) and because no FD isolate was available.
Limit of detection (LOD) for all three real-time PCR amplicons was performed on 10-fold serial dilutions of EY1 DNA (340 ng µL−1) and SE DNA (830 ng µL−1) in water ranging from undiluted DNA to a 109-fold dilution factor. Five replicate wells per dilution were used. The slope (k) of the linear regression line between logarithmic values of relative DNA concentrations (y-axis) and Ct values (x-axis) was used to calculate the amplification efficiency, E = (10[–1/k])–1, where a value of one corresponded to 100% amplification efficiency (Pfaffl, 2001). The squared regression coefficient after the linear regression (R2) was also determined. Dynamic range, i.e. the range of concentrations for which Ct values were in linear relationship with logarithms of concentrations and range of detection, were also determined.
Sensitivities of PCR and real-time PCR were compared on EY1 and SE dilution series. Each dilution series was prepared by mixing the diluted phytoplasmic DNA with undiluted DNA isolated from leaf veins of several healthy grapevine plants. This healthy grapevine material had already been tested with real-time PCR and confirmed to be BN- and FD-negative. Thus, PCR sensitivity was evaluated for potential effects of host-material inhibition. Two separately prepared sets of dilution series were used, employing the same EY1 and SE DNA, but adding these to a different pool of DNA from healthy grapevine leaf veins. The concentration of DNA in both pools was 300–400 ng µL−1.
Data mining approach
Decision trees were generated using the weka machine learning workbench (Frank et al., 2004) and its algorithm J48, which is weka's implementation of the C4·5 algorithm. As described in Larose (2005), the algorithm visits each possible decision node, in this case all conducted assays, selecting the optimal split, until no further splits are possible. Decision trees were drawn for the PCR-based detection system and the newly-developed real-time PCR assays, and evaluated using 10-fold cross validation. The percentage of correctly classified instances was calculated for both methods (PCR and real-time PCR).
Diagnostic sensitivity and specificity were also calculated for evaluation of both detection systems according to Dinnes et al. (2005). Diagnostic sensitivity (proportion of diseased samples that gave positive results) was calculated as true positives/total diseased and diagnostic specificity (proportion of non-diseased samples that gave negative test results) as true negatives/total non-diseased. Results from all three amplicons were used for calculations.
Design of real-time PCR assay
All amplicons were designed using the nucleotide sequences available in GenBank®. Schematic representations of each amplicon within the aligned nucleotide sequences used for amplicon design, including their GenBank accession numbers, are presented in Figs 1–3 and Table 2. The FDgen amplicon was designed within the SecY gene which codes for a translocation protein (GenBank Accession No. AY197685) using nucleotide sequences of several Flavescence dorée phytoplasma isolates including FD-C (AF458382, AY197688), FD-D (AY197685), FD70 (AF458383, AY197686), FD92 (AF458384) – similar to FD-D (AY197685), FD2000 (AY093581) and HD1 (AF458381). This amplicon is situated within the primer pairs FD9f/r (Daire et al., 1997) and FD9f3b/r2 (Clair et al., 2003) previously used for detection of FD, and contains two polymorphic nucleotides positioned between the 3′ end of the MGB probe and the reverse primer, which therefore do not affect the specificity of the amplicon. An additional polymorphic nucleotide appears in the forward primer, but only in the isolate HD1 and at the 5′ end of the primer, which should not substantially affect the annealing properties (Fig. 1). The BNgen amplicon was designed within the Stol11 genomic fragment (AF447594) using nucleotide sequences that showed no polymorphism within the amplicon region. It is situated within the primer pairs STOL11f2/r1 (Daire et al., 1997) and STOL11f3/r2 (Clair et al., 2003), which were previously used for detection of BN (Fig. 2).
The universal amplicon UniRNA was initially designed to amplify the FD-specific region of the 16S rRNA. Although a high number of ribosomal sequences from the stolbur (16SrXII) and elm yellows (16SrV) groups were used in the design process, the number of polymorphic nucleotides between those two groups was not high enough to even allow the specific discrimination of group 16SrV (Fig. 3). The universal amplicon UniRNA amplifies FD and BN phytoplasmas (along with other phytoplasma strains), but with a distinct preference for FD. Therefore it was included in the protocol as a tool to increase the sensitivity of FD detection and other phytoplasma types. In silico testing of specificity of all three amplicons using a blast search of public databases showed no unspecific hits with high similarity to the amplicon sequences (E > 0·011).
Performance characteristics of real-time PCR assays
Performance characteristics were determined by amplifying serial dilutions of phytoplasma-infected plant DNA in water (Table 3). Phytoplasma-infected C. roseus plants were used as a source of phytoplasma because of their high phytoplasma content. All designed amplicons show a broad dynamic range (at least five log orders of magnitude) and were very sensitive. High variability in Ct values observed in the last dilution, still giving positive results, was attributed to the stochastic effect in target copy distribution in replicate reactions. The occurrence of such stochastic effects in real-time PCR reactions is considered to occur when less than 10 copies of the target are present in the reaction (Morrison et al., 1998; Ellison et al., 2006). Therefore, it was concluded that the limit of detection was below 10 copies per reaction. Intra-run reproducibility was very high for all three amplicons as coefficients of variations of Ct values (cv, results not shown) between the parallels did not exceed 10% when the amplicon copy number was within the dynamic range of the method. The efficiency of amplification was above 89%. Parameters were also evaluated for the serial dilutions of phytoplasmic DNA mixed with grapevine DNA isolated from healthy, phytoplasma-uninfected grapevine leaves to imitate real grapevine samples (results from the first set of the dilution series are shown). No significant effects on sensitivity, limit of detection, repeatability of the method or efficiency of amplification were observed.
Table 3. Performance characteristics of real-time PCR measured for all three amplicons used in this study in water dilutions and for the FDgen and BNgen amplicons in dilutions prepared with phytoplasma-negative grapevine DNA isolated from healthy grapevine leaf veins (approximately 100 ng of genomic DNA per reaction in undiluted samples)
Real-time PCR products were visualized on a 10% polyacrilamide gel stained with ethidium bromide. Single bands of expected size in samples containing phytoplasmic DNA confirmed the absence of unspecific amplification (data not shown).
Specificity of real-time PCR phytoplasma detection
The universal amplicon (UniRNA) successfully amplified all phytoplasma strains, albeit with different sensitivities. The FD-specific amplicon (FDgen) amplified only elm yellows phytoplasma EY1. The BN-specific amplicon (BNgen) successfully amplified only stolbur phytoplasma SE. No cross-reactivity with DNA of healthy grapevine plants was observed (Table 4). Additionally, several grapevine epiphytic bacterial strains were isolated and tested for cross-reactivity with these amplicons. Only one out of 34 bacterial isolates gave a weak signal with the UniRNA amplicon (Ct values above 34·2) at more than 105 genome copies loaded per reaction, meaning that these results are unlikely to confound real diagnostic tests.
Table 4. Specificity of the real-time PCR assay for phytoplasma detection
DNA from organism (strain)
Average Ct value of 10-fold dilutions (c. 50–100 ng genomic DNA)
Type strain for the species.
ND, not detected.
EY1 phytoplasma isolate (16SrV-A, group reference strain)
BN-infected grapevine plants (range of Ct values from all BN-positive samples)
FD-infected grapevine plants (range of Ct values from all FD-positive samples)
Scaphoidens titanus (phytoplasma-free)
Reptalus panzeni (phytoplasma-free)
Sensitivity of real-time PCR compared to conventional PCR
Artificial samples imitating infected grapevine samples were prepared by serial dilutions of EY1 and SE phytoplasmic DNA mixed with healthy grapevine DNA. Detection of phytoplasmas by conventional nested PCR and real-time PCR was performed on two independent sets of artificial samples. SE was detected with the same level of sensitivity by both conventional nested PCR and the real-time PCR assay with the BNgen amplicon, the last signal being observed at the 106-fold dilution factor (weak band on agarose gel and average Ct value of 34·6 in the first and 35·8 in the second set, Fig. 4a). In the case of EY1, however, real-time PCR with the FDgen amplicon showed a similar sensitivity to that with the BNgen amplicon, with the limit of detection occurring at a 106-fold dilution factor (the last weak band on agarose gel was observed at an average Ct value of 37·3 in the first and 35·5 in the second set at this 106-fold dilution factor; Fig. 4b). Conventional nested PCR showed the final signal at the 103-fold dilution level in the first set (Fig. 4b) and at the 107-fold dilution level in the second set (results not shown).
Comparison of real-time PCR detection system and conventional PCR on field samples
A total of 153 field grapevine samples collected in 2005 were analysed for the presence of different phytoplasmas, using both conventional nested PCR and real-time PCR detection systems. A detailed comparison of the PCR vs. real-time PCR results is shown in Table 5.
Table 5. Comparison of real-time PCR and conventional PCR detection of phytoplasmas in 153 field samples of grapevine collected in 2005
In addition, three field bindweed and 19 insect samples were analysed. Two BN-positive bindweed samples by conventional PCR were also detected as BN-positive with real-time PCR. The third sample was only positive with nested PCR (conventional) or in the UniRNA assay (real-time). Further analysis of U5/3 PCR products by RFLP confirmed BN infection (data not shown). All five BN-positive insect samples were confirmed with real-time PCR (three samples of H. obsoletus, one sample of R. panzeri and one sample of E. incisus) and one additional BN-positive sample (H. obsoletus) was discovered among the 13 PCR-negative insect samples.
The results show that the real-time PCR phytoplasma detection system developed was even more sensitive than conventional PCR methods, as additional phytoplasma positives were confirmed in grapevine plants (14/38, 13 samples showed weak symptoms and one lacked all disease features) and in one insect sample. Ranges of Ct values (Table 6) obtained in different experiments showed that the concentration of phytoplasma could vary more than 100-fold from sample to sample. This was true for samples of grapevine leaves, insect vectors and field bindweed.
Table 6. Range of Ct values for PCR phytoplasma detection in various sample types using different amplicons. Bois noir (BN)-positive and Flavescence dorée (FD)-positive samples were classified according to conventional PCR results. Ct values are given for 10-fold sample dilutions. Mixed FD/BN infections (four samples) were counted both as FD-positive and BN-positive samples
Real-time PCR amplicon
BN positive samples
FD positive samples
18S rRNA amplicon was used as the endogenous control for insect samples and COX for plant samples.
Ct values for UniRNA showed that the amplicon detected FD and BN with different sensitivities. In the case of an FD-positive sample, FDgen Ct values should be at least two cycles below UniRNA Ct values and in the case of a BN-positive sample, BNgen Ct values should be at least four cycles above UniRNA Ct values. If these differences in Ct values deviate from this rule it can concluded that an additional type of phytoplasma is present in the sample. Thus, the presence of another phytoplasma type was confirmed in three BN-positive samples.
The difference in Ct values between 10-fold and 100-fold dilutions (ΔCt value) was monitored for each amplicon. As no inhibition was detected whilst using the protocols described in this paper (results not shown), the use of only 10-fold diluted DNA is proposed. When introducing new DNA extraction methods, ΔCt should, however, be monitored to assess the inhibition and to avoid false negative results caused by the inhibition of amplification.
Twelve samples of apple trees infected with phytoplasma belonging to the apple proliferation group (16SrX) also tested positive with the UniRNA amplicon, with Ct values ranging from 19·6 to 36·5. Grapevine samples collected in 2006 were analysed only with the real-time PCR detection system and produced the following results: 45 samples phytoplasma-negative, 96 samples BN-positive, 12 FD-positive and 6 phytoplasma positive containing an undetermined phytoplasma type (non-BN and non-FD).
A data-mining approach was used to produce decision trees for conventional PCR and newly developed real-time PCR detection systems for grapevine samples (Fig. 5a,b). The algorithm produced decision trees that showed which tests were crucial and used cross-validation to calculate how many samples (instances) were classified correctly/incorrectly for each result option (numbers in brackets) and to calculate the overall percentage of correctly classified samples. These results could be used to estimate which detection system performed better. The decision trees also offered a graphic overview over the complexity of the detection system and could be used to make the detection system more efficient. It should be noted, however, that the schemes did not represent the actual experimental design but only assisted the interpretation of results.
The decision tree in Fig. 5a was generated using test results from the conventional PCR detection system (samples collected in 2005). It showed that four different PCR tests would have to be used for classification of an unknown sample. Apart from specific primer pairs for nested PCR-FD9 (FDpf3b/r2), PCR-Stol11 (STOL11f2/r1) and nested PCR-Stol11 (STOL11f3/r2), the primer pair for universal detection of phytoplasmas, U5/3, was also shown to be vital for the correct classification of samples. Because nested PCR-FD9 and U5/3 required a preceding PCR step, the number of required PCR assays was increased to six (all PCR assays used in this study). Tenfold cross-validation showed that the proposed decision tree for PCR assays correctly classified 93·9% of the samples tested and that tests leading to BN-positive results were the most problematic (Fig. 5a).
The decision tree in Fig. 5b was generated using real-time PCR results on the same samples. It was simpler than the decision tree for conventional PCR and showed that all three real-time PCR amplicons were needed for classification of unknown samples. The fact that UniRNA was not positioned at the top of the decision tree reflects its unequal sensitivity towards FD and BN phytoplasmas. Tenfold cross-validation showed that the decision tree for real-time PCR assays correctly classified 100% of the samples. When an additional 156 grapevine samples collected in 2006 were included in the dataset for analysis with data mining, the same percentage of correctly classified samples was calculated (100%).
Diagnostic sensitivity and specificity were also used to evaluate both detection systems on samples collected in 2005. Both parameters supported the results from the data-mining approach, i.e. favouring the real-time PCR detection system: the diagnostic sensitivity of conventional PCR compared to real-time PCR was 88·0% and diagnostic specificity 92·3%.
Diagnosis of GY is currently predominantly based on PCR and RFLP analysis. Although real-time PCR for detection of GY has already been reported (Bianco et al., 2004; Galetto et al., 2005), a complete system for the specific detection of GY with similar or better sensitivity than conventional PCR/RFLP had not yet been made available. This paper reports TaqMan® MGB-based phytoplasma detection systems targeting ribosomal and non-ribosomal phytoplasma genes.
Quality and quantity of isolated DNA greatly influence the detection of GY. Monitoring the Ct values of endogenous controls (COX) allows the estimation of both parameters, therefore increasing the confidence in results. In this way each reaction with a low amount of DNA can be identified. Ct values for COX in the DNA extraction procedure used in this research work varied from 17 to 25. Higher Ct values were an indicator of lower amounts of plant DNA and presumably lower amounts of phytoplasmic DNA present in the reaction, which therefore decreased the confidence in results. Higher COX Ct values indicate that the results are not reliable and the DNA for that sample should be re-extracted and re-tested. 18S rRNA was used in the same way with insect samples. Ct values reached as low as 15·1 and depended greatly on the number of insects used for DNA extraction.
Despite the comparable levels of sensitivity assessed in evaluation of each assay on artificial samples, the analysis of field-collected grapevine samples showed significantly higher diagnostic sensitivity of real-time PCR in comparison to conventional PCR by revealing some false negative PCR results. Ten out of 38 samples that were detected as phytoplasma-negative by PCR were shown to be BN-positive by real-time PCR. All 10 samples showed weak symptoms, thereby eliminating the possibility of false positive real-time PCR results. Additionally, four out of 38 PCR-negative samples were phytoplasma-positive by real-time PCR, i.e. the presence of an undetermined type of phytoplasma was confirmed (only the UniRNA amplicon produced a signal). The increased sensitivity of real-time PCR can be partly attributed to the short regions of target DNA that are amplified in TaqMan® MGB probe-based real-time PCR assays (at least 10 times shorter than conventional PCR and approximately 50% shorter that TaqMan® amplicons). Therefore real-time PCR reactions might be less sensitive to DNA degradation or DNA-binding inhibitors present in the samples.
The robustness of the real-time detection system was additionally confirmed with the field-collected grapevine samples in the following ways: (i) in the consistent ability to amplify GY phytoplasma in field-collected grapevine from various geographical locations in Slovenia and Italy throughout the sampling season (July–October) and (ii) in the ability to confirm negative results of conventional PCR (healthy samples) without producing false positive signals, despite the seasonal variation of epiphytic microflora on grapevine plants. Although one of 34 bacterial isolates from grapevine microflora was detected with high Ct values with the UniRNA amplicon, the 105 bacterial genome copies per reaction is a situation highly unlikely to occur in grapevine samples.
Based on the results from a complete sampling season, the following approach is proposed for GY detection with real-time PCR using 10-µL reactions (2 µL DNA per well) on 384-well plates. Each sample (10-fold diluted DNA) should be tested with the following amplicons in three replicate wells: BNgen, FDgen, UniRNA and COX or 18S rRNA. UniRNA should be used to confirm the presence of FD and BN phytoplasmas and to better analyse mixed BN/FD infections. In the case of FD-only positive samples, Ct values for UniRNA should be two to three cycles below the FDgen Ct values. In the case of BN-only positive samples, the Ct values for the UniRNA amplicon should be more than four cycles higher than for BNgen. This differential response of UniRNA towards BN/FD is a result of higher specificity of the amplicon for the FD phytoplasma (Fig. 3). For three BN-positive samples the Ct values for UniRNA were higher than for BNgen, indicating the possibility of the presence of another type of phytoplasma alongside BN. When only the UniRNA amplicon produced a signal, the sample was considered to be infected with an undetermined phytoplasma type. In all such cases (four grapevine samples) Ct values were high (above 30), even in undiluted DNA, and the plants showed only weak symptoms or lacked them entirely. Therefore, it must be conceded that FD might be present in a very small concentration as the FD-specific amplicon is approximately 10 times less sensitive than the universal one.
Decision rule building has been used in plant pathology diagnostics since the 1970s (Michalski & Chilausky, 1999). Decision trees are often used in other fields of diagnostics because they represent an easily understandable method for extracting important knowledge from data which would otherwise remain confusing because of a large number of samples and assays requiring analysis. A common approach when comparing two (or more) diagnostic methods is the calculation of diagnostic sensitivity and specificity (Dinnes et al., 2005) for each pair of the compared methods. Such an approach requires one diagnostic method to be set as the reference standard. On the other hand, data mining allows simultaneous comparison of two or more diagnostic methods without having to select one as the reference standard. Thus all prejudice about which method is the most suitable for the reference standard is excluded. From decision trees for the PCR-based detection system and the newly-developed real-time PCR assays (Fig. 5a,b) it can be seen that three assays with real-time PCR produced a more accurate classification (100% correctly classified samples) than did six conventional PCR assays (93·9% correctly classified samples). When real-time PCR data from 307 grapevine samples collected in 2005 and 2006 were used in data mining the percentage of correctly classified samples remained 100%, again confirming the robustness of the real-time PCR detection system.
In summary, a method has been developed that has a better sensitivity and robustness than conventional PCR detection of GY and provides additional information about each sample that enhances the reliability of results. Results can also be interpreted in a quantitative way (e.g. with a relative quantification approach) if such data are needed for the study of epidemiology or physiology of the disease. Furthermore, the new method is substantially less laborious and time consuming.
This work was partially financed by the Slovenian Research Agency (V4-0872). We thank the Phytosanitary Inspection unit of the Plant Protection Service of Slovenia for collecting the field samples, Dr Ruggero Osler and Dr Paolo Ermacora from the University of Udine for kindly providing DNA samples and phytoplasma strains on periwinkle, Dr Maria M. Lopéz from the Instituto Valenciano de Investigaciones Agrarias (Valencia, Spain) and Tanja Dreo for providing bacterial strains.