The spatiotemporal spread of esca disease in a Cabernet Sauvignon vineyard: a statistical analysis of field data



The occurrence and spread of plants showing esca symptoms were assessed in a vineyard located on the plains of a northern Italian wine-growing region. Esca disease symptoms were assessed over 16 consecutive years, beginning one year after planting. The number of plants with symptoms was recorded over time, considering both vines with foliar symptoms in the year of assessment (manifest esca) and vines with foliar symptoms in previous years (hidden esca). The sum of manifest and hidden esca was indicated as cumulative esca. The first symptoms of esca appeared in the sixth year of cultivation, with the incidence of manifest esca increasing to approximately 3% nine years after planting. The number of cumulative plants with symptoms increased exponentially in the final period of observation. The aim of this work was to investigate the spatiotemporal spread of esca infection throughout the vineyard and to assess the distribution pattern of plants with symptoms using Bayesian spatiotemporal models. The research results seem to support a higher probability of infection along rows rather than among adjacent rows. This observation may have implications for the technical management of the vineyard.


Esca is one of the most important fungal grapevine trunk diseases and, over the last decade, it has spread throughout the world. The primary fungal taxa contributing to esca are the tracheomycotic fungi, such as Phaeomoniella chlamydospora and Phaeoacremonium aleophilum (teleomorph = Togninia minima), and ligninolitic species, such as Fomitiporia mediterranea (Larignon & Dubos, 1997; Mugnai et al., 1999; Fischer, 2002).

Esca-inducing fungi can be spread by airborne inoculum and infect the vines through existing wounds (Eskalen & Gubler, 2001; Quaglia et al., 2009). For example, P. chlamydospora produces pycnidia or simple conidiophores on pruning wounds, wood cracks and crevices (Edwards et al., 2001a) and T. minima can produce inoculum in the form of ascospores (Rooney-Latham et al., 2005). The susceptibility of annual pruning wounds on grapevines to P. chlamydospora and P. aleophilum has been confirmed by some authors (Larignon & Dubos, 2000; Eskalen et al., 2007; Serra et al., 2008; Rolshausen et al., 2010; van Niekerk et al., 2011).

To better define the development of esca in vineyards, visual assessments have been conducted in many wine-growing regions. Disease assessment is complicated because foliar symptoms appear discontinuous: plants recorded as having symptoms one year (manifest esca) can be symptomless the next (hidden esca) (Mugnai et al., 1996; Serra et al., 1998). These fluctuating symptoms could also be influenced by exogenous factors such as rainfall and temperature (Surico et al., 2000a; Marchi et al., 2006).

Long-term assessments were used to create two-dimensional maps showing the plants with symptoms in space and time (Pollastro et al., 2000; Surico et al., 2000a; Marchi et al., 2006; Romanazzi et al., 2009). These maps were used to analyse spatial patterns to identify potential sources of primary contamination and evaluate the secondary spread of the disease. For example, external, airborne sources typically result in random distribution patterns, while internal or nearby sources typically produce aggregated patterns. In addition, aggregation along plant rows could indicate the spread of esca by human practices.

Two-dimensional maps based on vineyard observations only allow spatial patterns to be considered subjectively. Therefore, statistical and geostatistical analyses were conducted to extrapolate objective and measurable patterns of infection.

Surico et al. (2000b) analysed the distribution of grapevine esca symptoms in a vineyard using indices of dispersion, such as Morisita's index, Lloyd's index of patchiness (LIP), ordinary runs analysis and two forms of two-dimensional distance class analysis. The authors reported that, in half of the four examined vineyards, infected vines showed a random distribution, while in the remaining half, an aggregated pattern was observed. Aggregation was identified along rows in one case, a pattern that was attributed to differences in varietal susceptibility between adjacent rows, as each row consisted of a different cultivar; however, in the other case, aggregation was observed both along and across rows.

Ordinary runs analysis was also used by Edwards et al. (2001b) to evaluate the distribution patterns of vines with symptoms in two vineyards under study. The analysis showed that vines with symptoms were distributed at random in one site and were aggregated in the other.

Stefanini et al. (2000) attempted to explain the dynamics of symptom expression using a longitudinal model based on the probability that esca symptoms expressed in the year t were dependent on the symptoms expressed in the year t−1. They observed an association between a plant's vicinity and the expression of symptoms. In another study, Sofia et al. (2006) used two indices of dispersion (variance-to-mean ratio and Morisita's index) and a two-dimensional distance class analysis (2dclass software) to show that, in the observed vineyard, the infected vines were distributed at random.

Variation in the spatial patterns of grapevines with esca symptoms among vineyards could be attributed to different environments, which are often related to the vineyard's age. For example, in young vineyards, a random pattern of infection could be explained by primary inoculum (i.e. infected grafts or cuttings and airborne spores). Once primary infections are established, the spread of secondary inoculum (i.e. splash-borne conidia, cultural practices) could then produce an aggregated pattern. Data collection may also contribute to incongruent reports in the literature, as discontinuity in the appearance of symptoms precludes the evaluation of the true incidence and spread of affected plants. This is particularly true when symptom assessments begin after vineyards are mature.

For all of these reasons, this study investigated the onset of esca symptoms in a newly planted vineyard. The spread of symptoms was evaluated through space and time, beginning the year after planting and continuing over the next consecutive years. The analyses focused on identifying the process responsible for the secondary spread of esca. Different symptom distribution patterns were compared using hierarchical Bayesian spatiotemporal models, and the parameters were estimated via Monte Carlo Markov chain (MCMC) algorithms. Bayesian techniques combined with MCMC algorithms allowed for a unified inference approach. Indeed, once MCMC samples from the posterior distribution are drawn, the same procedure can be followed to answer almost every question. However, frequentist approaches are less flexible and often require asymptotic approximations or ad hoc procedures. As an example, based on the MCMC output presented in the results section, a method is proposed for assessing the contributions of different sources of variability to the spread of esca.

Materials and methods

Characterization of the plot

Esca disease incidence and severity was monitored during 16 consecutive growing seasons in an experimental vineyard planted in 1992 in Spresiano, Treviso (45°47′57″ N; 12°15′10″ E), in the Venetian plains region of Italy (Table 1).

Table 1. Description of the analysed vineyard
Planting yearVarietySoilPruning systemVine × row space (m)RootstockNo. of plants
1992Cabernet SauvignonMedium texture with calcareous pebbles of alluvial originSylvoz1·5 × 3·0K5BB3440

The vineyard consisted of a unique block (40 rows of 86 plants each) on homogeneous soil, and it was cultivated according to the local agronomical practices, i.e. weed management along the rows using herbicide distribution, mechanical trimming, manual desuckering, chemical fertilization, fungicide sprays against downy and powdery mildew and mechanical harvesting.

Disease assessment

The observed symptoms were characteristic of esca: foliar intervein chlorosis and necrosis and the wilting of either some or all of the branches on the whole plant (apoplexy).

The assessments of the vineyard began one year after initial planting (in the second year of cultivation) and were conducted in each of the 15 subsequent growing seasons. All of the 3440 plants were visually inspected every year, in late August or early September, for the occurrence of esca foliar symptoms. Observed plants were divided into six symptom severity classes: 0 = symptomless, 1 = symptoms on a maximum of 10% of the grapevine canopy (few leaves), 2 = symptoms on 11–25% of the canopy (leaves of a few branches), 3 = symptoms on 26–50% of the canopy, 4 = symptoms diffused to more than half of the canopy, 5 = apoplexy.

As stated in the introduction, it is possible that plants that were recorded as having symptoms one year (manifest esca) were symptomless the next (hidden esca). To avoid underestimation of the true disease incidence in the vineyard, the sum of plants with both manifest and hidden esca (cumulated esca) was considered. The incidence of esca disease in the vineyard was calculated as the percentage of the total number of plants that had symptoms (manifest or cumulated).

Spatiotemporal statistical analyses

Spatiotemporal models were specified with the aim of explaining the variation of infection probabilities in both space and time. Because the vines are disposed on a regular grid and symptom values cannot be observed in the space between contiguous plants, the data were considered as realizations of a spatiotemporal process on a spatial discrete domain (lattice). The vineyard consisted of = 40 rows, each composed of = 86 vines, with a total of × J vines.

When analysing a set of N lattice data, the first step is to build an × N -dimensional adjacency matrix W that specifies the spatial structure used for modelling. In the matrix, entries wn,m (m,= 1,…,N) connect units m and n in some fashion and diagonal entries are expected to be wn,m = 0.

To identify potential spatial patterns of infection, three main specifications for W were provided; k orders of neighbourhood were considered.

The first specification, known as the Queen rule, was denoted as WQ(k) with inline image if vines n and m were (k-th order) neighbours or inline image otherwise. In Figure 1, according to the Queen rule, neighbouring structures a, b and c correspond to the orders = 1, 2 and 3, respectively.

Figure 1.

Models of the three main specifications, with three neighbourhood orders, used to identify potential spatial patterns of infection (a, b, c: Queen rule; d, e, f: Rook rule; g, h, i: Row rule). Rows are represented by horizontal lines. Vine positions along the rows are shown as vertical lines.

The second specification, known as the Rook rule, was denoted as WR(k) with inline image if vines n and m were (k-th order) neighbours or inline image otherwise. In Figure 1, according to the Rook rule, neighbouring structures d, e and f correspond to orders = 1, 2 and 3, respectively.

The third specification, which is called the Row rule, sets as neighbours only the vines belonging to the same row. The matrix was denoted as WRow(k) with inline image if vines n and m were (k-th order) neighbours or inline image if otherwise (Fig. 1 g,h,i).

The set of neighbouring vines constructed according to the Row rule is contained in the set of neighbouring vines built with the Rook rule, which, in turn, is included in that created by the Queen rule (Cliff & Ord, 1981). To evaluate spatial correlations in all directions, the following adjacency matrices were defined:

display math
display math

Let Yn,t be the dichotomous random variable associated with the n-th vine at year t and let yn,t be its realization. All of the plants showing esca symptoms were classified as ‘diseased plants’ (yn,t = 1), while the symptomless plants were classified as ‘healthy’ (yn,t = 0). Moreover, it was posed that yn,t = 1 if a plant showing esca symptoms was cut, headed, dead or substituted. For = 1,…,I, Yn,t refers to vines in the 1st row at year t, for + 1,…,2I, Yn,t refers to vines in the 2nd row at year t, and so on.

A cumulative incidence of the disease was assumed. In other words, if Yn,t = 1 at time t, then Yn,k = 1 for k = t + 1,…,T, such that the vine was considered to be infected in all the years after year t.

Given a generic neighbouring structure W(k), the following vectors were defined:

the N-dimensional vector of the observed values of the random variables Yn,t−1 was defined as yt−1; inline image was denoted as the N-dimensional vector whose generic element inline image denotes the observed number of infected plants among the k-th order neighbours of plant n at year t−1; the N-dimensional vector zt(k) = W(k) (ytyt−1) is the vector whose generic element inline image denotes the observed number of plants among the k-th order neighbours of plant n, infected at year t, that were not infected at year t−1.

Because no diseased vines were observed in the first five years of cultivation, in what follows, a model is proposed for the probability of infection in the last 13 years. In other words, = 5,…,17 was set because the first variation in rates of infection over time was observed between years 5 and 6. Let Ht be the set of healthy vines at year t−1, i.e. the set of vines at risk of being infected due to exposure during the year t. Nt is denoted as the number of vines belonging to the set Ht. The aforementioned cumulative assumption implies that inline image. The interest is in studying inline image, i.e. the probability of infection in plants at risk due to exposure in each year t.

For, inline image it is assumed that:

display math

As a starting point, the following model is proposed:

display math(1)

Consistent with the previously introduced notation, this model assumes the Queen rule of neighbouring. The coordinates (x1, x2) have been centred, i.e. the coordinate point (0, 0) occupies the centre of the vineyard. The regression coefficients β1 and β2 each measure linear, large-scale spatial variation in the row and column directions, respectively. Finally, γ1k is the autoregressive parameter capturing the effect of k-th order neighbour infected vines at year t−1 on the probability of infection at year t, and γ2k is the autoregressive parameter capturing the effect of k-th order neighbours infected vines at year t (which were not infected at year t−1) on the probability of infection at year t.

The parameter θt represents the logit of the probability that a plant in the middle of the vineyard, with no infected neighbours, is infected at time t, i.e. the baseline probability of being infected at time t. A random walk model is specified for this time-varying intercept to capture how infection incidence evolves over time, i.e. inline image, where inline image. This represents a first-order non-stationary temporal model. Such a process adequately captures the non-stationary temporal behaviour of the incidence of infection that was highlighted by the preliminary analysis.

Because a Bayesian approach to inference is adopted, the prior specification of model parameters is required. The following reasonably non-informative prior distributions were specified:

display math

A variety of alternative models (see Table 2 for a summary of the models considered in the application) can be derived from Eqn (1) for different values of K and different neighbouring structures. These models can be viewed as a time-varying generalization in a hierarchical framework, of the autologistic model (see e.g. Griffith, 2004) and of the model proposed by Stefanini et al. (2000). The proposed models allow all of the data to contribute simultaneously to the estimation of model parameters activating the process of borrowing strength in both space and time. For this reason, these models cannot be interpreted as a set of independent autologistic models applied at different times. Moreover, because of the need to condition the analysis on a time-varying set of exposed vines, spatiotemporal autocorrelations estimation as proposed in Reynolds & Madden (1988) cannot be adapted to this context.

Table 2. Model structure and selection
Adjacency structureNeighbouring order
ModelQRRowCol K z z*DIC
  1. Q, queen rule; R, rook; Col, column; DIC, deviance information criterion.

1   411201
2 411172
3  411169
4   411161
5   311150
6   211159
7   3 11164

Parameter estimation for these models is not straightforward under a frequentist approach. In the Bayesian framework, inference is based on the posterior distribution of the parameters given the data, e.g. inline image where Y denotes the entire observed data set, whereas Ω is the vector of model parameters and π denotes a probability distribution. Because of the complexity of the distributions involved in the proposed hierarchical model, the posterior distribution is not available in an analytical form. The posterior summaries of model parameters are computed from the means of the MCMC routines, as implemented in the OpenBUGS software (Thomas et al., 2006).

In the analysis here, the detection of the spatial patterns is addressed as a problem of model selection, i.e. the most suitable spatial pattern is the one assumed under the ‘best’ model. Model selection was performed based on the deviance information criterion (DIC), a widely used measure of model performance in the Bayesian framework. DIC is defined as the sum of a measure of fit and a measure of complexity: a model is preferred if it shows a lower DIC value (Spiegelhalter et al., 2002).

To evaluate the possibility that disease spread according to one pattern in the young vineyard (5–11 years of cultivation) and a second pattern after the number of plants showing symptoms increased (12–17 years), the analysis was also performed separately for these two time frames.


Disease assessment

No symptoms of esca disease were observed during the first five years of cultivation (Fig. 2). Later, a few vines with symptoms were observed. Initially, the total percentage of plants with symptoms in the vineyard was close to zero; however, in the 10th year of cultivation the first step in disease development was observed, and the manifest incidence of esca reached 3%. Then, a 2-year lag period followed, until another step was observed during the 13th year of cultivation. A pattern of steady incremental increases in the proportion of esca diseased plants continued in the following years. In the last year of assessment, manifest esca incidence was approximately 30%, and the cumulative incidence reached approximately 45% (Fig. 2). Symptom severity increased over time: shortly after planting, few plants in severity classes 1–3 were observed. Later, class 1 and class 3 vines reached a level of 7%, while class 2 vines increased to approximately 10%. In addition, class 4 plants reached a remarkable 3%. The number of apoplectic vines (class 5) increased slightly, primarily during the last years of observation, where incidence passed from 0·06% (13th year of cultivation) to 0·60% (17th year of cultivation). A relatively small percentage of plants suffered from apoplexy or died due to esca disease. These plants were headed or substituted throughout the years.

Figure 2.

The annual incidence of manifest and cumulated esca in the analysed vineyard.

The analysis of two-dimensional maps corresponding to years 7, 10, 13 and 16 (Fig. 3) suggests an increase in cumulated esca incidence and spatial clustering over the years.

Figure 3.

Maps showing the observed patterns of cumulated esca disease in years 7, 10, 13 and 16.

Spatiotemporal analyses

Model selection was conducted with the following aims: (i) to identify the most appropriate neighbouring structure for capturing spatial association; (ii) to identify the optimal order of the neighbourhood (K); and (iii) to verify any temporal patterns for the local spatial dependency, measured by the parameters γ1 and γ2. In Table 2, the structures of the proposed models are summarized, along with comparisons in terms of DIC. Model 1 hypothesizes that the spatial pattern is captured by the Queen rule neighbouring structure with the highest neighbouring order (= 4). Models 1–4 differ only in the specification of the neighbouring structure. The results from the DIC and parameter estimation (not presented) show that, of the directions specified within the Queen rule, only the direction corresponding to rows is relevant. The comparison among models 4–7 revealed that the optimal neighbouring order is = 3 and that only the following parameters are relevant: (i) the autoregressive parameter capturing the effect of k-th order neighbours (infected vines) at year t−1 on the probability of being infected at year t, and (ii) the autoregressive parameter capturing the effect of the k-th order neighbours (infected vines) at year t (which were not infected at year t−1).

The final selected model is:

display math(2)

This corresponds to the model with lowest DIC in Table 2.

The posterior summaries of the model parameters are reported in Table 3: the ‘Mean’ column contains the parameter's point estimates (posterior means), while columns P2.5 and P97.5 give a 95% Bayesian credibility interval for each parameter. The results show a statistically significant large-scale spatial trend in the row direction (parameter β1). Parameters γ, describing the amount of small-scale spatial association, are all statistically significant with the exception of γ22 and γ23. Because a model excluding these parameters was fitted and a higher DIC obtained (11253), it is preferable that these parameters are included in the final model. In Table 4, the posterior summaries of exp (γ1k) and exp (γ2k) are shown: these quantities can be interpreted in terms of odds ratios. The probability that a vine with an infected 1st order neighbour is infected at time t is 57·8% higher than the probability of infection in a vine with no infected neighbours of each order. In fact, the posterior mean of inline image is equal to 1·578. Moreover, the risk of being infected is higher when the neighbouring vine is infected at time t than at time t−1, i.e. the posterior means of E[exp(γ1k)∣Y] are always higher than posterior means of E[exp(γ2k)∣Y].

Table 3. Posterior summaries (mean, percentiles 2·5 and 97·5) of model parameters
 MeanP2·5P97·5 MeanP2·5P97·5
θ 5 −6·699−7·737−5·860 β 1 0·00500·00280·0072
θ 6 −6·614−7·563−5·835 β 2 0·0024−0·00220·0070
θ 7 −6·114−6·836−5·478 γ 11 0·45350·32210·5846
θ 8 −5·006−5·432−4·616 γ 12 0·11310·01540·2376
θ 9 −4·727−5·095−4·392 γ 13 0·18490·03590·3302
θ 10 −3·633−3·839−3·430 γ 21 0·10500·00530·2041
θ 11 −4·244−4·538−3·966 γ 22 0·0141−0·08580·1147
θ 12 −3·710−3·938−3·499 γ 23 0·0489−0·05430·1488
θ 13 −3·039−3·209−2·874 σ θ 2 4·12341·78259·1961
θ 14 −2·930−3·097−2·765    
θ 15 −2·583−2·738−2·431    
θ 16 −2·143−2·297−1·992    
θ 17 −1·920−2·090−1·748    
Table 4. Posterior summaries of inline image

The temporal evolution of the baseline probability of infection in a healthy plant in the middle of the vineyard, obtained as the posterior mean of the inverse logit transformation of θt′ s, is reported in Figure 4, along with its credibility intervals.

Figure 4.

The temporal evolution of the baseline probability of infection for a healthy plant in the middle of the vineyard. The results were obtained from the posterior means of the inverse logit transformation of the parameters θt.

The separate analyses (not shown) conducted for each of the two time frames (5–11 and 12–17 years of cultivation) revealed no appreciable differences between model estimates. In fact the 95% credibility intervals of the model parameters overlap. Indeed, this was expected because most of the information about the spatial parameters is provided by the last period of observation.

In what follows, a strategy is proposed for assessing, at each time t, the relative contributions of different model components to the variation in the individual probabilities of esca infection in the vineyard. In particular, it focuses on assessing the contributions of a large-scale spatial trend and two different clustering effects considered in the model. As a starting point, it is observed that μn,t (see Eqn (2)) can be decomposed as:

display math


display math

As they contribute to the mean value, the parameters θt s capture the year-specific incidence, but they do not help to explain the within-year variability.

Thus, the posterior mean of the variance of the vector inline image can be decomposed as:

display math

If the posterior means of the covariances are negligible, as is the case in this application, the following quantities allow the evaluation of the evolution of the relative contribution of each component to the total variability of μt over time.

inline image(the contribution of a large scale spatial trend);

inline image(the contribution of a clustering effect due to infected vines at year t−1);

inline image(the contribution of a clustering effect due to infected vines at year t)

These measures are analogous to those proposed by Cocchi et al. (2007) for decomposing the variability of atmospheric pollutants in a continuous spatial domain, and they are easily obtained on the basis of the MCMC output.

Figure 5 shows the temporal evolution of r1t, r2t and r3t. It can be noted that the contribution of a large-scale spatial trend is predominant at the beginning of the epidemic, and this effect declines as disease incidence increases. In the last years of the study, a clustering effect becomes more evident, with infected vines at year t−1 becoming the most important component. Indeed, the clustering effect resulting from infected vines at year t increases its contribution over time as well.

Figure 5.

The temporal evolution of r1t, r2t and r3t.


Esca monitoring in the vineyard began shortly after the grape planting, allowing observations to be recorded over the course of disease development. The first symptoms on vines appeared in 1997; however, an increase in esca foliar symptoms (approximately 3% of plants with manifest esca) was noted in 2001. In 1997–98, the plants with symptoms hosted all of the fungal taxa involved in esca disease, even if the growth of deteriorated wood tissues was restricted (Serra et al., 2000). Phaeomoniella chlamydospora and Phaeoacremonium spp. were also isolated from the deteriorated wood of foliar symptomless vines. The predictable growth of deteriorated wood tissues, together with the spread of fungi to new vines, might have contributed to the increase in symptom expression when the vines were 8–9 years old. All of the fungal species contributing to the esca disease complex can produce toxic compounds that are supposed to cause visible foliar symptoms, such as ‘tiger stripes’ (Bruno & Sparapano, 2008). These toxic effects might be most visible in rainy years, when active sap circulation would favour the accumulation of the toxins in the leaves (Marchi et al., 2006). Based on the observations here, the incidence of plants with esca symptoms grew slowly during the early stages of cultivation and increased in 2004 (after 13 years of cultivation), after a rainy summer, and following the 2003 droughts.

The cumulative number of plants showing external symptoms of esca increased through the years. In addition, the symptoms of the affected vines became more severe over time, indicating that the disease is progressive and irreversible. Therefore, the disease progressed even in years where external symptoms were not visible. Apoplectic plants were mainly seen in the last years of observation because they are primarily associated with the extension of wood white rot, which requires a long time to develop (Sánchez-Torres et al., 2008).

With respect to spatial patterns, the statistical model specifically proposed and developed in this work explained the spread of infected plants as being more probable along the rows. Separating the analysis into two phases, to distinguish the young and the mature vineyard, did not detect any aggregation patterns during the first eleven years. When only the mature vineyard phase was considered, the results were the same as those obtained for the overall period of observation.

The spread of infection along the rows could be an important practical consideration because it implies that any man-mediated practices conducted along the rows and causing wounds that can be infected by esca fungi might be involved in the spread of disease. Although their role in the spread of trunk diseases is still debated, these processes include winter pruning, spring desuckering and summer trimming.

The possibility that disease might be spread through pruning practices was rejected by several authors on the basis of spatial patterns and genetic or epidemiological data. Because of the random distribution of plants with symptoms in two Tuscan vineyards, Surico et al. (2000b) proposed that airborne spores of internal or external origin may play a stronger role in the spread of disease than the use of the contaminated pruning tools along the agronomic rows. Cortesi et al. (2000) analysed the spatial patterns of vines with symptoms and did not find any significant aggregation of disease. Moreover, they observed that genotypic diversity was high in Fomitiporia punctata strains isolated from different vines, therefore strengthening the hypothesis that inoculum is not spread clonally by pruning or roots. In a final study, Sofia et al. (2006) concluded that primary infection was probably caused by airborne spores from an infected neighbouring vineyard. The secondary spread by pruning tools was not considered because an aggregated pattern was not observed.

On the other hand, based on the observation that plants with symptoms were grouped along a row in an Italian vineyard, Mugnai et al. (1999) postulated that esca might be spread by pruning tools. In addition, in an analysis of many vineyards over time, Pollastro et al. (2009) observed a trend for the aggregation of vines with symptoms. This led the authors to suggest that infection might be spread by soilborne esca inoculum or from affected to healthy vines by natural or man-mediated processes. Indirect evidence that disease could be spread by pruning practices was reported by Scalabrelli & Ferroni (2010): disinfection of the pruning tools with sodium hypochlorite, together with the replacement of plants with symptoms, led to a significant reduction in the spread of symptoms in the vineyard over an 8-year time period.

In some cases, processes leading to the spread of disease were impossible to infer. Stefanini et al. (2000) demonstrated that a symptomless plant tended to remain symptomless in the following year, while a vine with symptoms was less likely to remain so. However, the obtained results could not provide a basis for additional inference. Finally, a spatial analysis conducted by Edwards et al. (2001b) was inconclusive in terms of extrapolating a general rule to explain the spread of esca disease.

It is supposed that if spores can be dispersed by wind or water then they can be dispersed by technical practices as well. These two possibilities are not mutually exclusive, and both secondary routes of contamination might contribute to patterns of infection observed in any given vineyard. In all likelihood, because cultural practices are consistent, the secondary spread of disease along rows could exceed the contribution of the random spread of airborne inocula as a vineyard matures. Moreover, it has been hypothesized that insects may disperse Pchlamydospora and Phaeoacremonium spp. spores (Edwards et al., 2001a; van Niekerk et al., 2010).

In conclusion, the results of this statistical approach do not reject the hypothesis that cultural practices could contribute to the spread of esca. The use of any device and/or measure to avoid contamination by pruning tools, particularly when they cause large wounds on old wood, should not be ignored in vineyards that are significantly affected by esca disease.


The work has been partially funded by the Project COLLEZIONI E-A-OR (MIPAAF-Ministry of Agricultural, Food and Forestry Policies). The authors would like to thank Dr Michele Borgo for the helpful support and suggestions.