A systematic literature search was conducted in the Web of Science (WoS) Databases (which include the Web of Science Core Collection, BIOSIS, Derwert, KCI-Korean, Russian Science, SciELO, Data Citation index, and MEDLINE^®)(Clarivate Analytics, 2020) using the following search string: Grapevine AND Trunk AND Disease$ AND Wound$ AND (susceptibility OR age). The operator AND indicates that the words must occur simultaneously in the search results, while OR indicates that the selected study must contain any of the terms separated by the operator; the dollar sign ($) represents zero or one character (e.g., “wound” or “wounds”). No restrictions on language or publication date were used. The search detected a total of 79 papers that were exported to EndNote™ online (Clarivate Analytics, online) and that were first screened based on the title and abstract. To be selected, papers had to first satisfy the following criteria: (i) the papers were published in peer-reviewed scientific journals, conference proceedings, scientific reports, or peer-reviewed theses; (ii) they describe experiments in which pruning wounds were artificially inoculated with GTD pathogens at different times after pruning; (iii) they contain quantitative information in the main text, tables, or figures on the percentage of inoculated wounds that were infected (e.g., by observing internal symptoms, or by reisolating the fungal pathogen after artificial inoculation); and (iv) the studies had an experimental design with replicates and were conducted with potted plants or field plants. The papers fulfilling these eligibility criteria were considered to be of potential interest, and the full-text manuscripts were retrieved and reviewed. Additionally, a cross reference-based search of relevant studies was performed to include reports of other studies that might have been eligible for the review and to reduce the risk of missing relevant information due to the choice of search terms. This procedure is shown in Figure 1 as a PRISMA flow diagram (Koricheva and Gurevitch, 2014).

The cut-off date for publications to be considered for inclusion in this quantitative analysis was 25 July 2022.

Data on the percentage of pruning wounds that showed internal disease symptoms or from which the pathogen was re-isolated following artificial inoculation (hereafter referred to as “disease incidence”) at different times after pruning were extracted from the original papers. All the inoculations were done by depositing a spore suspension (concentration provided as conidia/mL, or ascospores/spores per wound) on the pruning wound surface. Additional details can be found in Supplementary Material S3 .

Data in the main text or tables were extracted directly, while data in graphs were extracted with WebPlotDigitalizer version 4.3 (Rohatgi, 2020), an online tool that supports the extraction of numerical data from different types of graphs in a PNG or JPEG format. Unfortunately, it was not possible to extract data concerning the within-study variance because only five original articles contained information on the residual error component and the number of replicates (González-Domínguez et al., 2018). To retrieve missing information, we contacted the corresponding authors of the papers were contacted via email, but we received only a few replies, in which the authors declared they are not able to provide us with the original data or missing information, with only one exception.

Extracted data were organised in a database containing the selected studies (i.e., articles), cases within a study (i.e., single location or year or variety inoculated with a single pathogen), age of the pruning wounds (the time between pruning and pathogen inoculation, as defined before), and the disease incidence (as defined before). The time at which the pathogens were artificially inoculated on pruning woods was expressed as days after pruning (DAP); the day of pruning was considered as time zero (t₀). For instance, in an experiment in which pruning wounds were inoculated with a pathogen immediately after pruning and then 7, 14, 21, and 28 days later, the times were referred to as t₀, t₇, t₁₄, t₂₁, and t₂₈.

Relevant information on each case was also included in the database in order to extract subsets of data for considering the following main factors that could potentially affect the relationship between disease incidence and DAP were the pruning period (season), identity of the GTD, identity of the inoculated pathogen, and grapevine variety. The pruning period included three “categories”: early-season pruning (i.e., November for the Northern Hemisphere, no data for the Southern Hemisphere were found), mid-season pruning, (i.e., December and January for the Northern Hemisphere and June for the Southern Hemisphere), and late-season pruning (i.e., February and March for the Northern Hemisphere and July and August for the Southern Hemisphere). Identity of GTD included three categories: the Esca complex (EC, cases in which pruning wounds were inoculated with Pm. minimum or Pa. chlamydospora), Botryosphaeria dieback (BD, cases in which pruning wounds were inoculated with D. seriata, L. theobromae, N. parvum, or N. luteum), and Eutypa dieback (ED, cases in which pruning wounds were inoculated with E. lata (previously named E. armeniacae). There are eight identities for the inoculated pathogens: Pa. chlamydospora, Pm. minimum, D. seriata, L. theobromae, N. parvum, N. luteum and E. lata. Ten grapevine varieties: Cabernet Sauvignon, Chardonnay, Chenin Blanc, Merlot, Sauvignon Blanc, Thompson Seedless, Grenache, Pinot Noir, Shiraz, or Tempranillo.

Since it was not possible to extract data on the within-study variance of the different studies, it was impossible to fit a multivariate random effects model via linear (mixed-effects) models, and perform a formal meta-analysis. Therefore, we made a quantitative analysis of pooled data coming from the different studies considering the among-study variability only. The weaknesses of this approach are discussed in the next paragraphs.

We first used boxplots (Williamson et al., 1989) to study the effects of the main factors on the disease incidence at the time of pruning (t₀). To assess the differences between the main factors at t₀, we used generalised linear models (GLMs). GLMs provide regression analysis and analysis of variance for one dependent variable by one or more factors, allowing the test of the null hypotheses about the effects of other factors on the means of various groupings of a single dependent variable (Neal and Allgar, 2005). Because there was under-dispersion of data (i.e., lower variability than expected by the binomial model), a quasibinomial family distribution was used. The function glht (general linear hypotheses) was used for post hoc comparisons to compute contrasts between each combination of the factors by using Tukey’s method. The raw data used are available as Supplementary Material S1 .

To study the temporal dynamics of pruning wound susceptibility, the disease incidence at t₀ was set at 1, and the disease incidences at the following inoculation times were rescaled to 1 and expressed on a 0 to 1 scale. For instance, if the disease incidence was 85% for wounds inoculated at t₀, and was 55%, 32%, 10%, and 3% for wounds inoculated at 7, 14, 21, and 28 DAP, then t_{0 =} 1.00, t_{7 =} 0.65 (i.e., 55/85), t_{14 =} 0.38 (i.e., 32/85), t_{21 =} 0.12 (i.e., 10/85), and t_{28 =} 0.04 (i.e., 3/85).

Average (and standard error) of the rescaled disease incidence in the different cases was then calculated for each timing after pruning (DAP) and fit to a negative exponential equation in the following form:

where Y_t is the rescaled disease incidence (dependent variable, from 0 to 1) at time t; t is the time of fungal inoculation after pruning (independent variable, in days: DAP); parameter a is the rate at which Y decreases as t increases, which was estimated for all data, main factors, and interactions (for a total of 71 estimates) by using nonlinear regression procedures (nls); standard errors of the parameter a given by the nls function were used to calculate the confidence interval of the predicted Y values.

Goodness-of-fit of equations [1] were calculated as follows: NSE (Nash-Sutcliffe efficiency), i.e., the ratio of the mean square error to the variance in the observed data, subtracted from unity (when the error is zero, NSE = 1, and the equation provides a perfect fit); W index of agreement, i.e., the ratio between mean square error to total potential error (W = 1 represents a perfect fit); root mean square error (RMSE), i.e., the fit standard error of the regression, which is calculated as the square root of the mean square error (RMSE represents the average distance of real data from the fitted line); coefficient of residual mass (CRM), which is a measure the tendency of the model to over- or underestimate the observed data (a negative CRM indicates a tendency of the model toward overestimation); R², which represents the proportion of experimental variability explained by the selected model (R² = 1 represents a perfect fit); and the concordance correlation coefficient (CCC), which is calculated as a measure of model accuracy (the CCC is the product of the Pearson product-moment correlation coefficient between observed and predicted values, and the coefficient C_b indicates the difference between the best-fitting line and the perfect agreement line; CCC ranges from -1 to 1, with perfect agreement at 1 (Nash and Sutcliffe, 1970; Loague and Green, 1991; Madden et al., 2007).

To determine whether the temporal dynamics of pruning wound susceptibility were significantly affected by the categories of the main factors and their two-way interactions (three-way interactions were not considered because the number of studies would have been too low), the following null hypothesis was tested: the slope a of equation [1] estimated for a category of a main factor (e.g., category EC for the main factor GTD) was not significantly different from that of another category of the same factor (e.g., EC vs. BD). The latter null hypothesis was tested as follows. Equation [1] was expressed in its linear form through data transformation through the natural logarithm function (ln), and the intercept of the linearised equation was set at ln(1 + 1) shifting the y-values by ln(2). Significant differences between slopes of the linearised, shifted functions were finally tested using the lsmeans package; slopes were extracted using the function lstrends and then compared through the function pairs, which returns a matrix consisting of t-tests for each comparison. There were 529 comparisons in total, and these are shown in Supplementary Material S2 .

All statistical analyses were conducted with R studio Version 1.4.1717 (RStudio Team, 2021).

The literature search provided 79 studies, of which 41 were selected for full text analysis. The others were disregarded because they were not considered relevant to this study as they failed to meet the above-mentioned criteria. Two additional papers were selected via cross-referencing. After full text analysis, 16 papers were used for quantitative data extraction. The selected papers were published between 1980 and 2022, and were conducted in Australia, New Zealand, California, the USA, South Africa, Italy, Spain, and France; most were performed in the Northern Hemisphere. The key information of these 16 studies is summarised in Table 1 ; further information is available as Supplementary Material S3 . Since each study considered multiple fungi, years, or vineyards, there were 154 cases in our dataset, for a total of 1110 single data values for disease incidence.

In the selected studies, wounds were inoculated at various times after pruning with a maximum of 10 months (Eskalen et al., 2007) on different grapevine varieties and in experiments conducted either in the field or with potted plants located outdoors, for 1 to 4 consecutive years. Three major GTDs were considered, including six studies for the Esca complex (in which the artificial inoculation was conducted with Pm. minimum or Pa. chlamydospora, nine for Botryosphaeria dieback (with D. seriata, L. theobromae, N. parvum, or N. luteum), and nine for Eutypa dieback (with E. lata) ( Table 1 ). Pruning dates varied between November and March for the Northern Hemisphere, and between June and August for the Southern Hemisphere. In only two studies (Amponsah et al., 2014; Shafi, 2016), vines were pruned during the vegetative period; the data from these two studies were used in the whole dataset but not in the analysis of the pruning period.

In the whole dataset, the median value of disease incidence (wounds being infected or showing GTDs symptoms following artificial inoculation with GTDs fungi) at the time of pruning (t₀) was 64.0%; the variability, however, was very high with extremes ranging from 0 to 100%, and with 50% of the data ranging from 40.3 to 80.0% ( Figure 2 ). When the whole dataset was split based on the pruning period, the disease incidence for wounds inoculated at t₀ was significantly higher for wounds inoculated at mid-pruning than at late pruning (P< 0.001), with early pruning showing intermediate values ( Figure 2 ). When the whole dataset was split based on the GTD, the incidence of successful infections was significantly lower for Eutypa dieback than for Botryosphaeria dieback (P = 0.032), with intermediate values for Esca complex agents. When the whole dataset was split based on grape variety ( Figure 3 ), disease incidence was lower for Chenin Blanc and Merlot than for other varieties; for Chenin Blanc and Merlot,< 40% of inoculated wounds became infected ( Figure 3 ). Because of high variability among experiments and fewer data availability for Grenache, Pinot noir, Tempranillo, and Shiraz, no significant differences were found in the pairwise comparisons between these varieties (P > 0.058).

In the whole dataset, the disease incidence declined over time after pruning (DAP) when the GTD fungi were inoculated ( Figure 4 ). Equation [1] fit the data with a = 0.019 ± 0.002, R² = 0.74, CCC = 0.85, NSE = 0.68, W = 0.92, RMSE = 0.16, and CRM = -0.031. The analysis of data distribution with respect to the fit and its confidence interval showed that > 50% of the observed data overlapped with the predicted interval, with some overestimation between 20 and 30 DAP, and underestimation between 90 and 120 DAP ( Figure 4 ).

When the whole dataset was divided according to the main factors (pruning period, GTD, grape variety, and fungal species) and their interactions, for a total of 70 combinations, equation [1] fit the observed data rather well with the following ranges of values: R² (0.25 to 1); NSE (0.1 to 0.99); W (0.46 to 1); RMSE (0.02 to 0.72); CRM (-0.08 to 0.12); and CCC (0.3 to 1) ( Tables 2 – 5 ).

The pruning period (as a main factor) was not significantly related to the dynamics of wound susceptibility, because the slope parameters calculated for early (a = 0.018), mid (a = 0.020), and late (a = 0.021) pruning were not significantly different from a based on the whole dataset (P > 0.243; Table 2 ) or from each other, with P > 0.485 ( Figure 5 ). Disease as a main factor, on the contrary, showed a significant effect. Wound susceptibility decreased over a shorter period after pruning for Eutypa dieback (a = 0.049) than for the Esca complex (a = 0.016) or Botryosphaeria dieback (a = 0.015), with P< 0.001 ( Table 2 and Figure 6 ). For Botryosphaeria dieback, a significant interaction between pruning period x GTD ( Table 3 ) showed that the susceptibility decreased faster for late pruning (a=0.024) and early pruning (a = 0.019), which were not significantly different from each other (P = 0.999), than for mid-pruning (a = 0.010; P = 0.021 and 0.029, respectively) ( Supplementary Material S2 ).

Wound susceptibility decreased faster when pruning wounds were inoculated with E. lata (a = 0.049) than with other fungal species (a ranged from a minimum of 0.011 for D. seriata to a maximum of 0.023 for N. parvum, Table 2 ; with P< 0.001, Supplementary Material S2 ), with the exception of N. luteum (a = 0.083; P = 0.019). The latter species was also different from the two main species involved in the Esca complex, Pa. chlamydospora and Pm. minimum (a = 0.016 for both species; P = 0.001 and 0.002, respectively), and from D. seriata (a = 0.011; P< 0.001), which is part of the same GTD (i.e., Botryosphaeria dieback).

Wounds were susceptible for a shorter period for Shiraz, Grenache, Pinot noir, and Merlot (in decreasing order, with a ranging from 0.095 to 0.040) than for Cabernet Sauvignon, Chardonnay, Chenin Blanc, Tempranillo, and Thompson seedless (in decreasing order, with a ranging from 0.026 to 0.022); the pairwise comparisons between varieties in the first and second of these groups were in most cases significant groups ( Table 2 and Supplementary Material S2 ). Sauvignon Blanc had the longest period of wound susceptibility (a = 0.010), and was different from the first group of varieties but not from the second. No significant differences were found for the pruning period within each variety (P > 0.593), with the exception of Thompson Seedless, in which the susceptibility of pruning wound declined slower with late pruning (a = 0.019) than with early or mid-pruning (a = 0.036 and 0.060, respectively; P = 0.001) ( Table 3 and Supplementary Material S2 ). Nevertheless, differences were detected between varieties within the same pruning seasons. Wounds of Sauvignon Blanc pruned in mid-season (a = 0.01) were susceptible for a longer time than wounds of Cabernet Sauvignon (a = 0.028; P = 0.02) or Thomson seedless (a = 0.06; P< 0.001) pruned in the same season. Differences among varieties were not significant with late-season pruning ( Supplementary Material S2 ). No data were available for early-pruned Sauvignon Blanc.

Differences were also found for some varieties or pruning seasons among GTDs. For Cabernet Sauvignon, the decrease in wound susceptibility over time after pruning was faster for Eutypa dieback (a = 0.060) than for Botryosphaeria dieback or the Esca complex (a = 0.022 and 0.026, respectively; P< 0.001) ( Table 4 and Supplementary Material S2 ). Similarly, decline in wound susceptibility for Thompson seedless was faster for Eutypa dieback (a = 0.057) than for Esca complex (a = 0.011; P< 0.001) (no data available for Botryosphaeria dieback). For Chardonnay, on the contrary, the decline in wound susceptibility was faster for the Esca complex (a = 0.073) than for Botryosphaeria dieback (a = 0.023; P = 0.003) ( Supplementary Material S2 ). Among the GTDs, the decline in wound susceptibility was faster for Botryosphaeria dieback with late-season pruning (a = 0.024) than with mid-season pruning (a = 0.01; P = 0.002) ( Table 5 and Supplementary Material S2 ).

When the estimates of a for main factors and interactions were compared with the estimate of a for the whole dataset (a = 0.019), significant differences were found as indicated in Tables 2 – 5 .

Our analysis showed that, at the time of pruning, disease incidence of wounds was higher for the fungi associated with Botryosphaeria dieback than for those associated with Esca complex or Eutypa dieback, which generally agrees with previous reports (Rolshausen et al., 2010; Amponsah et al., 2014; Ayres et al., 2016). In addition, pruning wounds remained susceptible to infection longer when inoculated with Botryosphaeria dieback fungi or Esca fungi than with Eutypa dieback fungi, except in the case of Neofusicoccum luteum, for which wound susceptibility decreased faster than for other Botryosphaeriaceae and E. lata. A shorter susceptibility of wounds to E. lata infection than to infection by other GTDs was observed by Ayres et al. (2016) and Herche (2006), but not by Elena and Luque (2016), who observed a similar decrease in wound susceptibility for Esca and Botryosphaeria dieback. Our result overall confirmed the potential threat of Botryosphaeriaceae species once they have established in a vineyard (Rolshausen et al., 2010) as well as the increasing prevalence of these species among the GTDs (Guerin-Dubrana et al., 2019).

An important limitation of our study is that no studies were found for the basidiomycetes involved in the EC. A possible reason is that these fungi have been isolated from infected vine trunks, but often with an incomplete understanding of their role in the disease aetiology (Surico et al., 2006; Bertsch et al., 2013; Gramaje et al., 2018). With the advent of metagenomics, reports on basidiomycetes are increasing (Del Frari et al., 2019; Bruez et al., 2020) as well as the studies on their role in symptom development (Cholet et al., 2021; Moretti et al., 2021; Pacetti et al., 2021); recently Brown et al. (2020) suggested that basidiomycetes may not require infection by P. chlamydospora in order to extensively colonize the wood. Studies on the susceptibility of pruning wounds to infection by basidiomycetes are then needed.

Reports on the effect of pruning season on the susceptibility of pruning wounds to infection by GTD fungi are inconsistent. In California, USA, for instance, Petzoldt et al. (1981) found that pruning wounds made in December remained susceptible to infection by E. lata for a longer time than pruning wounds made in March. Similarly, Úrbez-Torres and Gubler (2011) showed that wounds were susceptible to L. theobromae and N. parvum for a longer time when pruning was done in early winter (with susceptibility lasting up to 84 days) than in early spring (with susceptibility lasting up to 12 days). Furthermore, Munkvold and Marois (1995) found that the percentage of wounds infected decreased faster with time when pruning was done in January or March than in previous months. Studies conducted in Spain, in contrast, documented a faster decrease in wound susceptibility when pruning was done in autumn compared to late pruning, but only when wounds were inoculated with D. seriata (Luque et al., 2014; Elena and Luque, 2016). Some studies reported that fresh pruning wounds are susceptible to infection regardless of the pruning season (Eskalen et al., 2007), and that adjustment of pruning season is ineffective in controlling wood diseases (Serra et al., 2008). Based on the differences among findings from experiments carried out under different conditions, Van Niekerk et al. (2011) concluded that the susceptibility of pruning wounds to infection may be less related to the time of year when pruning is done than to other factors, such as the disease investigated or more specifically the fungi implicated.

In our analysis, however, infection incidence for wounds inoculated at the time of pruning was overall lower when vines were pruned later in the season, and the reduction in the infection incidence over time after pruning for fungi causing Botryosphaeria dieback was faster when pruning was done in the late season than in earlier pruning periods. Úrbez-Torres et al. (2010) found that there was less spore dispersal by the Botryosphaeriaceae in the late dormant season than in winter, and this may contribute to the result of a lower incidence of wound infection with late versus early pruning.

We, therefore, confirm that the time of pruning is relevant for reducing infection of susceptible varieties and by fungi causing Botryosphaeria dieback, with late pruning being preferred, aware that seasonal variation might occur between regions caused mainly by climatic differences (Gramaje et al., 2018). Advantages of late pruning have been associated with “vine bleeding” in early spring, which can temporarily inhibit fungal penetration into the xylem vessels (Larignon and Dubos, 2000; Serra et al., 2008; Martínez-Diz et al., 2020) and can flush away fungal spores from the wound surface (Martínez-Diz et al., 2020). In addition, the low temperatures of winter may reduce the rate at which wounded tissues heal, increasing the time that they remain susceptible to infection (Petzoldt et al., 1981; Munkvold and Marois, 1995; Rolshausen et al., 2010). In contrast to our overall results, our analysis indicated that wounds of Thompson Seedless remained susceptible for a longer time when pruned later than earlier in the season (mid-season pruning), and this deserves further study.

In our quantitative analysis, differences were found among grape varieties for infection incidence of wounds inoculated at pruning (t₀) and after pruning. For the t₀ data, however, data variability and unbalanced replications caused some under-dispersion in the GLM model (Sellers and Morris, 2017). Additional data are needed to determine whether the degree to which the probability of a pruning wound becoming infected is influenced by grape variety, and to which extent.

Our analysis showed that pruning wounds remained susceptible to GTDs for a longer time on Sauvignon Blanc than on other varieties. Sauvignon Blanc was previously reported to be especially susceptible to GTDs (Borgo et al., 2008; Bertsch et al., 2013; Murolo and Romanazzi, 2014; Sosnowski et al., 2022). Similarly, wounds of Cabernet Sauvignon remained susceptible to infection for a longer time than wounds of other red varieties, such as Grenache, Pinot noir, Merlot, or Shiraz. Differences in the susceptibility of grapevines to GTDs may be related to a variety of factors such as differences in sensitivity to fungal toxins, the rate of toxin degradation in wood tissue, and the time required for recognition of the fungal invasion and consequent activation of defense mechanisms (Wagschal et al., 2008), which involve stilbenes, other polyphenols, and proteins involved in primary metabolism (Borgo et al., 2016; Fontaine et al., 2016a; Fontaine et al., 2016b). Xylem morphology, and especially xylem vessel size may also contribute to differences in susceptibility among varieties (Pouzoulet et al., 2014; Fischer and Peighami Ashnaei, 2019; Claverie et al., 2020). It is also important to notice that a higher number of data was available for Cabernet Sauvignon than for Grenache, Pinot noir, Merlot and Shiraz so that our estimation of parameter a would be more accurate for Cabernet Sauvignon than for other varieties. Additional work is then needed to add robustness to our interpretations.