Nanoplate digital PCR assays for detection and quantification of Xylella fastidiosa

.


INTRODUCTION
Xylella fastidiosa (Xf) is an important plant pathogen (Mansfield et al., 2012). Xf is a Gram-negative bacterium that colonizes plants, particularly their xylem vessels, as well as insects. This pathogen is difficult to culture, which led to the species name of fastidiosa (Wells et al., 1987). Xf can cause severe damage when introduced in new environments. The pathogen was initially known in North and South America, associated with grapevine Pierce's disease (Davis et al., 1978). However, Xf has been identified in numerous outbreaks, and many studies have determined that Xf has a wide host range, including 655 plant species in 88 families (EFSA, 2022). To date, the main crops affected by Xf are olive trees (Olea spp.), grapevine (Vitis spp.), citrus (Citrus spp.), coffee (Coffea spp.), peach (Prunus persica) and almond (Prunus dulcis) (Chang et al., 1993;Rodriguez et al., 2007;Saponari et al., 2013;EFSA 2022), but the pathogen has also been found in forest trees, including American elm (Ulmus americana), American sycamore (Platanus occidentalis) and northern red oak (Quercus rubra) (Desprez-Loustau et al., 2020), as well as in common ornamental, wild and crop plants including fleabane (Erigeron sp.), Helichrysum stoechas, pistachio (Pistacia vera), and persimmon (Diospyros kaki) (EFSA 2022). This host list is being often updated, as EFSA has been mandated by the European Commission to publish biannual updates of Xylella hosts species during 2021 to 2026.
Many asymptomatic hosts of Xf have been discovered, because the pathogen is commensalist with its host plants, and only a subset of interactions between specific hosts and clades of Xf can result in the disease development (Sicard et al., 2018). This phenomenon is also because Xf is genetically diverse, and its classification has been a matter of opinion with as few as two and as many as five subspecies being recognized. Among these, three subspecies are widely supported: Xylella fastidiosa subsp. fastidiosa, Xylella fastidiosa subsp. multiplex, and Xylella fastidiosa subsp. pauca (Potnis et al., 2019;Vanhove et al., 2019). Xylella fastidiosa subsp. pauca is gaining importance in Europe as the cause of olive quick decline syndrome in Apulia (Saponari et al., 2013;. As well as high genetic variability and wide host range, management of Xf is further complicated because the pathogen is transmitted by insect vectors, which harbour the bacterium in their foreguts (Purcell, 1979, Backus andMorgan, 2011). This is different from most other pathogens transmitted persistently, that instead perform complex molecular interactions with their insect vectors and have limited host ranges (Redak et al., 2004). In practical terms, this means that sharpshooter leafhoppers (Hemiptera, Cicadellidae, Cicadellinae) or spittlebugs (Hemiptera, Cercopoidea, Aphrophoridae, Cercopidae, Clastopteridae), which are xylem-sap feeders, could potentially be Xf vectors (Frazier, 1965). The most important known vectors of Xf are Homalodisca vitripennis (Hemiptera, Cicadellidae) in America, and Philaenus spumarius (Hemiptera, Aphrophoridae) in Europe (Cornara et al., 2016).
Taking these biological features of Xf into consideration, which are further complicated by ecological, social, and economic factors, introduction of Xf into new areas where novel vectors and hosts occur, can have severe consequences, leading to inclusion of Xf into the quarantine pest category (NAPPO, 2004;EPPO, 2021). The European Commission has published a list of priority regulated quarantine pests, including Xf, which every Member State must implement all possible actions to avoid introduction and spread of this organism, and formulate contingency plans, simulation exercises, and action plans for the eradication of these pests (Regulation (EU) 2016/2031).
For this reason, early and precise diagnoses of the presence of Xf are important, particularly because this pathogen causes non-specific symptoms which can be mistaken for abiotic stresses (Thorne et al., 2006), particularly water or nutrient stress, and these symptoms develop months after infections (Baldi and La Porta, 2017). In the periods between infection and symptom development, the pathogen can be acquired by vectors and transmitted to new hosts, so diagnoses based on symptom observation are inefficient for the containment of the pathogen. However, Xf can be detected by molecular assays before host symptoms develop, so molecular assays have dominated diagnosis of this pathogen. These methods include traditional assays such as ELISA and conventional PCR, as well as more modern approaches such as real-time PCR (qPCR) and LAMP. The main methods for the detection of the pathogen are indicated in the standard EPPO PM 7/24 (EPPO, 2019), while in Regulation EU 2020/1201 (Annex IV) official test methods are indicated that must be used by laboratories for the identification of X. fastidiosa and its subspecies.
Despite being accurate and efficient, the current methods used for detection of Xf suffer from low throughput rates and being time-consuming and labourintensive. These methods also allow precise pathogen quantification only when using standards of known concentration as references. Digital PCR (dPCR) can increase analysis throughput by removing the need for technical replicates and allowing absolute pathogen quantification without the need for reference material or standards (Selvaraj et al., 2019). This reduces pathogen detection times. Early pathogen detection is important in areas where the pathogen is yet to establish, and at important entry points of plant material into a country, such as harbours and airports. In global trade, the availability of fast, reliable, and quantitative diagnostic assays for pathogens and pests has become are urgently required (Faino et al., 2021), as testified by the reports of impor-tant pests being recently found in new areas, for example Erwinia amylovora, another relevant bacterial pathogen, reported in Tuscany in 2020 (Migliorini et al., 2021).
The present study aimed to evaluate the performance of primers and probes currently employed for diagnosis of Xf through qPCR when used in a nanoplate-based dPCR assay, and to determine if the qPCR methods could be directly transposed to this new dPCR technology. Similar studies have been carried out to test these methods in droplet-based dPCR (Dupas et al., 2019), but considering the technical differences between dropletand nanoplate-based dPCR, an entirely separate set of tests for initial validation of these methods in nanoplatebased dPCR was necessary. All tests were carried out on experimental samples obtained by adding known quantities of DNA from Xf subsp. fastidiosa, Xf subsp. multiplex, or Xf subsp. pauca in the nucleic acid extracts from healthy plants of Nerium oleander, an ornamental host of Xf. These tests underwent preliminary validation by comparison with the current qPCR assays.

Bacterial strains and DNA used in this study
Total nucleic acids from three X. fastidiosa strains and one strain of a non-target, non-pathogenic Pseudomonas syringae pv. syringae, belonging to a genus commonly found in healthy plant tissues, were used in this study.
The nucleic acids of pure cultures of X. fastidiosa subsp. fastidiosa strain DSM 10026 (indicated hereafter as Xff) and X. fastidiosa subsp. multiplex strain DSM 103418 (Xfm) were supplied by DSMZ GmbH. The nucleic acids of X. fastidiosa subsp. pauca strain ST53 (Xfp) were kindly provided by the Phytosanitary Service of Lombardy region.
After acquisition or extraction, all bacterial nucleic acids were stored at -30°C.
Nucleic acids from oleander (Nerium oleander) were extracted from asymptomatic whole leaf samples or midribs (0.5-1 g) using a CTAB method, described in EPPO standard 7/24 (EPPO, 2019). Oleander leaves were sampled in spring 2020 from three different asymptomatic plants, the samples were pooled together, and then kept frozen at -30°C until extraction in September 2020. Three samples of whole leaves were obtained and pooled together, while 4 samples of midribs only were obtained and assessed individually, as reported in the following paragraphs and Tables 1 and 2. Absence of X. fastidiosa in these samples was confirmed by including the nucleic acids from the plants, without spike of any kind, in all subsequent molecular assays: no amplification due to X. fastidiosa presence was detected in any of the nucleic acid samples extracted from asymptomatic N. oleander plants used in this study.
All quantifications of nucleic acids were carried out using a Nanodrop 1000 spectrophotometer.

Preparation of experimental samples and controls
Two sets of experimental samples and controls were prepared.
The first set was prepared by adding nucleic acids from oleander (using samples obtained from whole leaves, final concentration = 10 ng µL -1 ) with nucleic acids from one of the three X. fastidiosa strains (final concentrations down to ten copies per μL, as shown in Table 1), or the 260-02 strain (final concentration = 10 5 copies per μL). The theoretical copy numbers of the bacterial genomes were estimated using concentration of the nucleic acids, the size of the genome, and the average molecular weight of a DNA base pair to calculate molarity, and, therefore, the number of molecules in the volume. Positive controls included nucleic acids from one of the three X. fastidiosa strains at different final concentrations, as shown in Table 1. Negative controls included nucleic acids either from the 260-02 strain, only from oleander, from oleander with 260-02 added, or no nucleic acids (NTC), as shown in Table 1.
The second set of samples and controls was prepared to further assess possible interference from the plant matrix in a sample more closely resembling those obtained during actual surveys for X. fastidiosa diagnoses. DNA was extracted from four samples each of 0.5 g of midribs from asymptomatic oleander plants. Each of the three Xf subspecies was added to each DNA sample at the following concentrations: Xff approx. 250 or 125 copies per µL, Xfm approx. 150 or 75 copies per µL, Xfp approx. 275 or 140 copies per µL, for a total of 6 experimental samples from each single starting DNA. DNA samples without added Xf DNA were also employed as negative controls in this experiment. All the samples and controls included in this second set are reported in Table 2.

Xylella fastidiosa detection using qPCR
The presence of Xf in the samples was determined using the SYBR Green assay (Francis et al., 2006), and a TaqMan assay (Harper et al. (2010), erratum 2013). The methods for these two assays are outlined in EPPO standard PM 7/24 (EPPO, 2019).
The first method uses a pair of specific primers for XF, that was designed on the sequence of a conserved hypothetical protein gene: HL5 (5'-AAGGCAATAAACGCG-CACTA-3') and HL6 (5'-GGTTTTGCTGACTG-GCAACA-3'). This primer pair amplifies a segment of length 221 bp. The reaction mix was prepared as indicated in the EPPO standard PM 7/24 (4) (Francis et al., 2006;EPPO, 2019), modified as follows: the volume of template nucleic acids was doubled and the total reaction volume was raised to 12 µL. The final composition of the mix was as follows: PowerSYBR master mix (Applied Biosystems) 1×, primer HL5 0.28 µM, primer HL6 0.28 µM, DNA template 2 µL, and water up to a volume of 12 µL.
For both methods, the changes in volumes of the reactions in comparison with that described in the Table 1. List of samples, positive controls, and negative controls included in the first set of analyzed material. The table reports for each sample/control the name, composition of host DNA and the source of added DNA. The concentration of the nucleic acids is expressed either as ng µL -1 or copy number per µL, depending on which parameter was more relevant during the preparation of the sample. This concentration value is an approximation used to express the order of magnitude of the target (for copies per µL) or rounded to the nearest multiple of 5 (for ng µL -1 ). EPPO standard (EPPO, 2019) were carried out to be more similar to the operative conditions of the dPCR assay, which uses a total volume of 12 µL, as reported in the handbook for the reaction mixes. Using the same total reaction and experimental sample volumes assures that the concentration of target DNA is the same in both assays, making the results directly comparable. These reactions were carried out in StepOnePlus Real-Time PCR thermocycler (Thermo Fisher Scientific).

Sample
The thermal cycling profile used for both methods was as described in the EPPO standard (EPPO, 2019).
All the samples reported in Table 1 were tested with both methods. Each sample and control were analysed in triplicate.
For assessing the results obtained, in accordance with two methods described in the EPPO standard (EPPO, 2019), the quantification cycle (Cq) was evaluated for both the Taqman and SYBR Green methods,

Xylella fastidiosa detection and quantification using dPCR
The presence and quantity of XF in the samples and controls were assessed in nanoplate-based dPCR, using the QIAcuity instrument (Qiagen). The results obtained were analyzed using the QIAcuity Software Suite version 2.0.20.
The methods of Francis et al. (2006) and Harper et al. (2010) were carried out in dPCR. The reaction mixes were set up with the same composition as the corresponding mixes used for the qPCR assays, but using the EG PCR Master Mix 3× (Qiagen) in place of the Pow-erSYBR mix, and QIAcuity Probe PCR Master Mix 4× (Qiagen) in place of TaqMan Universal Master Mix No Amperase. Both master mixes were used at final concentrations of 1×.
To keep these reactions as close as possible to the original methods used in qPCR, no restriction enzyme was added to the mixes, although use of a restriction enzyme is normally suggested in instructions for the master mixes used in dPCR.
The reactions were carried out in 96 well QIAcuity Nanoplates, with 8.5K partitions per well. All the samples and controls reported in Table 1 were assayed with both methods. Each sample and control were tested in duplicate. All the samples and controls reported in Table  2 were assayed with both methods, without carrying out technical replicates, with the exception of NTC, which was assayed in triplicate.
For the Eva Green (EG) protocol, thermal cycling was as follows: one cycle of incubation at 95°C for 2 min; 40 cycles of 95°C for 15 sec, 60°C for 15 sec, and 72°C for 15 sec; then incubation at 40°C for 5 min. This cycle followed instructions in the EG PCR Master Mix 3× (Qiagen) handbook, changing only the annealing temperatures to match that of the primers employed. Imaging was carried out with 700 ms of exposure and a gain value of 8.
For the Probe protocol, thermal cycling was as follows: one cycle of incubation at 95°C for 2 min; 40 cycles of 95°C for 15 sec, and 62°C for 30 sec. This followed the instructions of the Probe PCR Master Mix 4× (Qiagen) handbook, changing only the annealing temperature to match that for the primers employed. Imaging was carried out with 500 ms of exposure and a gain value of 6.

Evaluation of diagnostic performance parameters
For each test, parameters were calculated as follows: Accuracy = 100 × (PA + NA) / (PA + NA + PD + FD); Sensitivity = 100 × PA / (PA + ND); Specificity = 100 × NA / (NA + PD); where PA was positive agreement (a positive result is obtained when a positive result is expected), NA is negative agreement (a negative result is obtained when a negative result is expected), PD is positive deviation (a positive result is obtained when a negative result is expected), and ND is negative deviation (a negative result is obtained when a positive result is expected).

Xylella fastidiosa detection through qPCR
The qPCR assays correctly detected the presence of Xff, Xfm and Xfp in all the experimental samples and positive controls (Table 1, Figure 1), regardless of the primer pairs employed. For all samples and positive controls, the qPCR assays yielded the expected results, with Cq increasing as the concentration of pathogen decreased and, for the SYBR Green assay, giving a single, recognizable Tm peak for each sample. The assays therefore detected the pathogen at an order of magnitude as low as ten copies per µL.
The TaqMan assay gave an expected result of "undetected" for all the negative controls. On the other hand, the SYBR Green assay gave as result a Cq also on negative samples and controls. The NTC gave an average Cq of 37.32, which is indicated as an undetermined result, but did not show exponential amplification nor the correct Tm of the amplicon and are therefore considered as negative results as indicated in the protocol (EPPO, 2019). However, other negative controls gave results that could be mistaken as positive based on Cq alone. These included N. oleander with added P. syringae nucleic acids (OPSS_5), or without these nucleic acids (Oleander), gave Cq values between 32 and 34, regardless of the presence of added nucleic acids, while nucleic acids from P. syringae pure culture (PSS) gave an average Cq of 33.82. Analysis of the melting curves showed that these results were due to non-specific amplification: Tm was consistent between samples with added Xf subspecies (average 83.15°C), while in the samples without Xf DNA the Tm varied between 61-93°C, without clear, defined peaks in the melting curves.

Xylella fastidiosa detection through dPCR
Results of dPCR assays from the Eva Green (EG) and the Probe methods for each control and sample reported in Table 1 indicated that both primer pairs amplified as intended. Some examples of the outputs produced by the dPCR are presented in Supplementary  Figures 1 and 2.
For the EG method, a strong background fluorescence was detected, with a fluorescence for negative partitions ranging from 50 to 135 RFU, while the positive partitions had intensity of 200 to 250 RFU (Figure 2A). It is important to note that the RFUs for negative controls are different between the NTC or P. syringae DNA, and those that contain DNA of N. oleander, being greater for N. oleander DNA. This could be due to some non-specific annealing of the primers producing some DNA amplicons, in line with the results obtained with qPCR. Also, there is likely presence of some 'rain' in the samples and positive controls, defined as partitions that give intermediate fluorescence between negative and positive (in this case 135-200 RFU), which were found with high frequency in the XFM samples and controls. To determine the effect of the threshold level on detection and quantification of XF, results include two different thresholds. The first threshold was set just above the RFU of the negative cloud (dMIQE group, 2020) which is presented under the code EG_135, since the threshold is at 135 RFU. The sec-ond threshold was set at the lowest border of the positive cloud, effectively removing the 'rain' partitions, which is presented under the code of EG_200, since the threshold is set at 200 RFU (Table 3). The Probe method showed an overall lower level of fluorescence, both for negative and positive partitions: negative partitions had fluorescence of 10 to 20 RFU, while the positive partitions ranged from approx. 40 to 60 RFU ( Figure 2B). This result is in line with the lower exposition time and gain utilized for the imaging. Although less common than in the EG method, also in this assay there was some 'rain', with fluorescence between 20 and 40 RFU. In order to determine the effects of the threshold level on detection and quantification of XF, results include two different thresholds: one set just above the RFU of the negative cloud (dMIQE group, 2020) which is presented under the code of Probe_20, since the threshold is at 20 RFU; the second threshold was set at the lowest border of the positive cloud, effectively removing the 'rain' partitions, which is presented under the code of Probe_40, since the threshold is set at 40 RFU (Table 3).
Detection and quantification of the pathogen was possible in all samples and positive controls (Table 3),  (Table 1) on the X-axis, and quantification cycles (Cq) obtained on the Y-axis. For each sample, the grey bars are results of TaqMan assays (Harper et al., 2010) while the white bars are results of the SYBR Green assay (Francis et al., 2006). The represented number for each Cq is the average obtained from three replicates, and the error bar represents the standard deviation. Where no bar is present, the result of the assay was "undetected". regardless of the threshold setting, for both the EG and Probe methods. Taking into consideration that there was no standardized protocol to follow for these methods for nanoplate-based dPCR, and that the present study was carried out only on experimental samples and controls of known status (positive and negative), also certified by two different qPCR diagnostic methods, the parameters to discriminate between positive or negative results were not decided beforehand, but were instead formulated by analyzing the present results. For copy numbers of targets per µL and confidence intervals, it was possible to discriminate between positive and negative results by setting threshold copy number for a positive result as greater than the average target copy number detected in negative controls, taking into consideration the greatest bounds of the confidence intervals. Due to the rare cases of amplification in the negative controls, this threshold was calculated as 12 for EG_135, 2 for EG_200, 3 for Probe_20, and 2 for Probe_40 (Table 3). The numbers of copies to consider a sample positive were greater in the cases of lower thresholds (EG_135 and Probe_20), as there was presence of 'rain' partitions in the negative controls, but there was negligible influence of the threshold setting in the Probe method. Also when considering 'rain' partitions as positive, in line with dMIQE guidelines, in no case was there a copy number of target sequence greater than 10 for negative controls, or less than 10 for samples and positive controls (Table 3). Nevertheless, considering that ten target copies is close to the greatest registered also for negative controls, an order of magnitude of ten copies of target per µL of sample was the limit of detection for these assays. The results obtained with both assays confirmed the theoretical concentration of Xf in the experimental samples and positive controls, as the concentration of target was always in the expected order of magnitude. The positive controls with the greatest concentrations of target, namely XFM-2 and XFF-2, caused the respective wells to be oversaturated. This result, while positive, did not allow quantification of the target concentrations, so was not optimal. While the order of magnitude of target copy number in the samples and positive controls was confirmed by the two different assays, there were differences in the results. samples and positive controls contain-ing Xfm gave much greater copy numbers per µL when tested with the EG method than with the Probe method. This result is partially explained by the high number of 'rain' partitions in the Xfm samples tested with the EG dPCR method. The overestimation in this sample type was exemplified by the XFM-3 positive control. The Probe_20 and Probe_40 results were similar, with, respectively, 26,173 and 24,891 copies per µL. In contrast the results for EG_135 were 88,351 copies per µL, and 67,076 for EG_200. While the higher threshold caused overestimation of the target's quantity by 2.7 times, the lower threshold raised this error further to around 3.5 Table 3. Results of dPCR assays on the first set of samples, positive and negative controls. The table reports the target copy number/µL for each sample, the error is calculated as a CI of 95%. The columns report the results obtained with either the EG or Probe method with the two considered thresholds. Copy number was rounded to the closest unit. The error is rounded to the closest decimal.

Sample
Digital PCR results (copies/µL of template) EG (Francis et al., 2006) Probe (Harper et al., 2010 times greater. The positive control XFM-4 gave a similar result, with EG_200 giving a result two times greater than that obtained with the Probe method, and with EG_135 giving a result four times greater. For the samples and positive controls containing Xff and Xfp, these differences were not as pronounced, especially when the copy number was at 10 3 or less. While this overestimation of the target abundance for XFM did not interfere with detection of the pathogen, it would become an issue when accurate quantification is necessary.

Comparison between different diagnostic methods
All four methods employed in the study (qPCR SYBR Green, qPCR Probe, dPCR EG, dPCR Probe) detected presence of the target pathogens in the samples and positive controls, and did not detect the target pathogens in the negative controls. Having detected no differences between the expected results and experimental results, all four methods showed a 100% accuracy, diagnostic sensitivity, and specificity. All the methods also detected X. fastidiosa at all the tested concentrations, down to an order of magnitude of ten target copies per µL.
Since both the qPCR and dPCR approaches are quantitative, a direct comparison of the results obtained by the methods is possible. Linear regression analyses showed, that there was strong statistical significance (P = 0.000) for the correlations between results from the different diagnostic assays (Figure 3). Correlations for the different methods were especially high when comparing the two dPCR methods ( Figure 3D), and the qPCR and dPCR using probes ( Figures 3E and 4F), and with overall R 2 values greater than 0.9 and close to 1 when considering the individual Xf subspecies. The least correlations were measured for the comparisons between the qPCR with SYBR Green, and the dPCR with EG and threshold of 135 ( Figure 3B). This was due to two distinct forms of bias that have been highlighted by the comparison of distinct methods. Firstly, Cq registered for Xff using the SYBR Green method was slightly less than those for the other two subspecies at equal concentrations of target. Secondly, the dPCR quantification for Figure 3. Scatterplots showing the correlations between results obtained by different diagnostic assays in the following pairs: (A) qPCR with SYBR Green on the X-axis and qPCR with Probe on the Y-axis; (B) qPCR with SYBR Green on the X-axis and dPCR with EG, threshold set at 135 RFU, on the Y-axis; (C) qPCR with SYBR Green on the X-axis and dPCR with EG, threshold set at 200 RFU, on the Y-axis; (D) dPCR with EG, threshold set at 135 RFU, on the X-axis and dPCR with Probe, threshold set at 20 RFU, on the Y-axis; (E) qPCR with Probe on the X-axis and dPCR with Probe, threshold set at 20 RFU, on the Y-axis; and (F) qPCR with Probe on the X-axis and dPCR with Probe, threshold set at 40 RFU, on the Y-axis. Markers on the plots indicate single samples, and their colours and shapes indicate the subspecies of Xylella fastidiosa they belong to, as reported in the legend. Each graph contains three dashed trendlines, which are calculated on samples belonging to a single subspecies, and a solid-line trendline, calculated on all the samples for each combination of the diagnostic assays. Each graph also reports the overall R 2 value and the P value obtained from linear regression analysis, as well as the R 2 value for each X. fastidiosa subspecies.
Xfm gave greater copy number per µL than expected, mostly due to presence of many 'rain' partitions. These two effects reduced the R 2 values for the subspecies in which they were identified by small amounts (to 0.862 for Xff and 0.754 for Xfm). Overall, when both effects are considered together by evaluating all three Xf sub-species at the same time, they caused high reductions in correlations between the two methods, with an R 2 value of 0.678. The same analysis with the threshold at 200, ignoring the effect of the 'rain' partitions on Xfm, gave greater correlation with the qPCR results, with an R 2 values of 0.861 for Xff and 0.826 for Xfm, and an overall correlation of 0.739 ( Figure 3C). Xfp gave consistently greater correlation values compared to other Xf subspecies, regardless of which two methods were being compared. A list of all R 2 values for the correlations of each pair of methods is presented in Supplementary Table 1.

Evaluation of plant matrix effect on detection through dPCR
The methods previously described and tested on samples that contained N. oleander DNA diluted to 10 ng µL -1 were tested on a set of samples and controls that resembled real samples that could be obtained by DNA extraction from infected host material. Results of dPCR EG and dPCR Probe from this set of samples and controls are shown in Table 4.
dPCR EG failed to provide useful information for this set of samples. The positive controls that contained only Xf were correctly amplified, but all samples and negative controls that contained undiluted N. oleander DNA resulted in oversaturation, probably caused by the high quantity of DNA that can bind to the non-specific EvaGreen reporter.
In contrast, the dPCR Probe method correctly detected the presence of Xf subspecies in all samples and positive controls, while giving negative results for the negative controls. The negative control O3_C with the threshold set at 20 had a total of three copies per µL. However, as the positive threshold was determined to be more than three for the Probe_20 method, it was still considered negative. With the threshold set at 40, all negative controls returned a concentration of zero copies of the targets.
Quantification of the pathogen concentrations was adequate with the dPCR Probe method, as the results were in line with the expected orders of magnitude for each sample and control.

DISCUSSION
Digital PCR (dPCR) is a technique that is recently being adopted in research laboratories and will probably not rapidly become widespread for diagnostic applications. Nevertheless, its potential benefits for detection and quantification of pathogens are many, and rapid integration of this molecular technique in diagnostics could prove beneficial for early detections of pathogens, particularly for quarantine pests. Several examples of the use of dPCR as diagnostic tools are available in human medicine (Sedlak et al., 2014;Mangolini et al., 2015;Devonshire et al., 2016), indicating the benefits that can be obtained from the use of this technique. While it is still a budding technology in plant health, there are some studies that have showed adaptation of qPCR assays to dPCR (Dreo et al., 2014;Lu et al., 2019;Maheshwari et al., 2017;Zhao et al., 2016;Dupas et al., 2019). These studies utilized the droplet-based dPCR technology, which uses oil emulsions to create partitions as small droplets, which then each undergo amplification and analysis. This technology needs additional steps in adaptation of qPCR protocols, due to different working environments caused by the use of oil, and the available systems for droplet-based dPCR suffer from lower throughput compared to qPCR while requiring more steps for sample preparation (Dupas et al., 2019). The system used in the present study was a nanoplate-based dPCR, which obtained the partition of the reaction mixture through the physical conformation of the reaction plate. This allowed direct adaptation of qPCR protocols, and could process a high number of samples at once. The present study used qPCR and dPCR which gave comparable results for detection of X. fastidiosa. Both approaches showed high performance criteria, with accuracy, specificity, and diagnostic sensitivity equal to 100% in the first set of experimental samples. While some differences based on the pathogen subspecies were highlighted in these analyses, all the methods detected the pathogen, regardless of subspecies. All the results confirmed that dPCR and qPCR had similar performance criteria for detection of Xf, which was in accordance with a previous reported by Dupas et al. (2019). That study compared the qPCR diagnostic method of by Harper et al. (2010) for Xf with droplet-based dPCR. Both methods were effective for detection of the pathogen, and there was high correlation between copy numbers detected by droplet-based dPCR and the Cq values from qPCR. Since it is reported that the benefits obtained from the use of dPCR can be dependent on the studied pathosystem (Dreo et al., 2014), it is significant that the present results confirm that dPCR can be suitable for the detection of Xf, using different plant matrices and dPCR technology than those tested by Dupas et al. (2019). Quantification of Xf through dPCR was an improvement compared with qPCR, which would result in increased throughput in the pathogen diagnoses. Since qPCR can only achieve relative quantifications in comparison to reference material, using this technique to quantify target copy numbers requires the inclusion of standards with known concentrations, which uses up several wells in each assay plate. Obtaining accurate, absolute pathogen quantification without using such standards allows the processing of large numbers of samples per reaction. Such standards may not always be available. For Xf, axenic culturing is possible, so standards of known quantity can be developed to make absolute quantifications through qPCR. This is not the case for many other plant pathogens, for which the use of dPCR for quantitative assays offers benefit (Gutierrez-Aguire et al., 2015).
Considering that standard qPCR procedures analyze two or three replicates for each sample, use of dPCR, with results from thousands of repetitions in each well, can increase the numbers of samples processed per reaction, removing the limitation of employing several wells per sample. The increase of throughput can translate into shortening of the technical time needed to carry out diagnoses for many samples and also in reduced per sample analysis costs.
As dPCR is a quantitative and highly sensitive assay, it rarely gives exclusively negative results (zero positive partitions), and could also show some partitions with fluorescence from negative samples, in a way similar to how qPCR can give late Cq values greater than 37 for negative samples. Therefore, it is necessary to set thresholds to discriminate between positive and negative results, especially for EG methods that do not use specific probes. In the present analyses, setting the threshold at the highest copy number per µL detected in a known negative control (adjusted to the highest range of the confidence interval) allowed discrimination between positive and negative samples. This result will need further confirmation and validation before it can be used as a threshold for true diagnostic tests, especially considering that for EG_135 the threshold was high, at 12 copies per µL and could theoretically cause positive samples with ten copies per µL to be incorrectly classified as negatives. Also, the possibility of adding more negative and non-template controls to each plate could be considered, to more precisely estimate how many positive partitions are detected in negative samples when working with environmental samples.
In conclusion, EG and Probe dPCR methods, with either low thresholds set just above the cloud of negative partitions or higher thresholds set just below the cloud of positive partitions, were able to detect the presence of three different Xf subspecies in the analyzed samples. For the quantification, the EG method was not reliable for Xfm, and, especially with the lower threshold (EG_135), it detected several positive partitions in negative controls. Concern regarding fluorescence derived from non-specific amplification was more relevant in EG dPCR. as, compared to SYBR Green assays carried out on qPCR, the lack of a melting curve step might contribute to incorrect positive detections in samples that do not contain the targets organisms. This concern was a major downfall of the method for the second set of samples containing large quantities of non-target DNA, in line with what would be obtained from extractions from infected plant material. This caused oversaturation of the wells even in the negative controls. Also, this method showed relevant production of 'rain', in particular for Xfm, that could contribute to uncertainty in the data analyses. In contrast, having lower values for positive partitions in the negative controls, guaranteed by using specific primers and a specific probe, the present study results suggest that using the protocol described by Harper et al. (2010) could give more reliable results. These considerations, in particular the completely unreliable results obtained when using samples that contained 300-600 ng per µL of host plant DNA, suggest that the EG method should not be employed for actual diagnostics, while the Probe dPCR method was reliable.
This study indicates the potential benefits of using nanoplate-based dPCR as a technique that can substitute traditional qPCR assays for detection of Xf, offering comparable performance criteria and the possibility of increased sample throughput, lowering the time and cost of analyses. This study is a first step demonstrating the possibility of using this technique for diagnostic applications in plant pathology. Before being utilized in actual diagnoses, these results should be validated by accredited laboratories, using more samples and including naturally infected plants.