Global predictions for the risk of establishment of Pierce’s disease of grapevines

Thermal requirements to develop PD

We examined the response of a wide spectrum of European grapevine varieties to Xf_PD infection in three independent experiments conducted in 2018, 2019, and 2020. Overall, 86.1% (n = 764) of 886 inoculated plants, comprising 36 varieties and 57 unique scion/rootstock combinations, developed PD symptoms 16 weeks after inoculation. European V. vinifera varieties exhibited significant differences in their susceptibility to Xf_PD (Supplementary Table S1). All varieties, however, showed PD symptoms to some extent, confirming previous field observations of general susceptibility to Xf_PD^9,12,37. We also found significant differences in virulence (χ² = 68.73, df = 1, P = 2.2 × 10⁻¹⁶) between two Xf_PD strains isolated from grapevines in Majorca across grapevine varieties (Supplementary Fig. S1). Full details on the results of the inoculation tests are available in “Methods”, Supplementary Note 1, Supplementary Table S1 and Supplementary Data 1.

Growing degree days (GDD) have traditionally been used to describe and predict phenological events of plants and insect pests, but rarely in plant diseases⁵⁸. We took advantage of data collated in the inoculation trials together with temperature to relate symptom development to the accumulated heat units at weeks eight, 10, 12, 14, and 16 after inoculation (Supplementary Data 1). Rather than counting GDDs linearly above a threshold temperature, we consider Xf ’s specific growth rate in vitro within its cardinal temperatures. The empirical growth rates come from the seminal work by Feil & Purcell³⁸ shown in the inset of Fig. 1a. This Arrhenius plot was transformed, as explained in Supplementary Note 2A, to obtain a a piece-wise function f(T) Eq. (1). Our model and risk maps are based on f(T) (red line in Fig. 1a) because it provides the best fit to the experimental data when compared with the commonly used Beta function (blue line) for representing the thermal response in biological processes^59,60. This Modified Growing Degree Day (MGDD) profile Eq. (1) enables to measure the thermal integral from hourly average temperatures, improving the prediction scale of the biological process⁶¹. MGDD also provides an excellent metric to link Xf_PD growth in culture with PD development as, once the pathogen is injected into the healthy vine, symptoms progression mainly depends upon the bacterial load (i.e., multiplication) and the movement through the xylem vessel network, which are fundamentally temperature-dependent processes^38,62. Moreover, MGDD can be mathematically related to the exponential or logistic growth of the pathogen within the plant (Supplementary Note 2B).

Interannual infection survival in grapevines plays a relevant role when modelling PD epidemiology. In our model, we assumed a threshold of five or more symptomatic leaves for these chronic infections based on the relationship between the timing and severity of the infection during the growing season and the likelihood of winter recovery^38,39,42. This five-leaf cut-off was grounded on: (i) the bimodal distribution in the frequency of the number of symptomatic leaves among the population of inoculated grapevines (Supplementary Fig. S1), whereby vines that generally show less than five symptomatic leaves at 12 weeks after inoculation remain so in the following weeks, while those that pass that threshold continue to produce symptomatic leaves, and (ii) the observed correlation between the acropetal and basipetal movement of Xf along the cane (Supplementary Fig. S1). The likelihood of developing chronic infections as a function of accumulated MGDD among the population of grapevine varieties was modelled using survival analysis with data fitted to a logistic distribution ({{{{{{{mathcal{F}}}}}}}}({{{{{rm{MGDD}}}}}})). A minimum window of MGDD = 528 was needed to develop chronic infections (var. Tempranillo), about 975 for a median estimate, while a cumulative MGDD > 1159 indicate over 90% probability within a growing season (red curve in Fig. 1c and “Methods”).

Next, we intended to model the probability of disease recovery by exposure to cold temperatures. Previous works had specifically modelled cold curing on Pinot Noir and Cabernet Sauvignon varieties in California as the effect of temperature and duration³⁹ by assuming a progressive elimination of the bacterial load with cold temperatures⁴². In the absence of appropriate empirical data to formulate a general average pattern of winter curing among grapevine varieties, we combined the approach of Lieth et al.³⁹ and the empirical observations of Purcell on the distribution of PD in the US related to the average minimum temperature of the coldest month, T_min, isolines⁴¹. To consider the accumulation of cold units in an analogy of the MGDD, we searched for a general correlation between T_min and the cold degree days (CDDs) with base temperature = 6 ^∘C (see “Methods”). We found an exponential relation, ({{{{{rm{CDD}}}}}} sim 230exp (-0.26cdot {T}_{min })), where specifically, CDD ≳ 306 correspond to ({T}_{min } < -1.{1},^{circ }{{{{{rm{C}}}}}}) (Fig. 1b). To transform this exponential relationship to a probabilistic function analogous to ({{{{{{{mathcal{F}}}}}}}}({{{{{rm{MGDD}}}}}})), hereafter denoted ({{{{{{{mathcal{G}}}}}}}}({{{{{rm{CDD}}}}}})), ranging between 0 and 1, we considered the sigmoidal family of functions (f(x)=frac{A}{B+{x}^{C}}) with A = 9 × 10⁶, B = A and C = 3 (Fig. 1c), fulfilling the limit ({{{{{{{mathcal{G}}}}}}}}({{{{{rm{CDD}}}}}}=0)=1), i.e., no winter curing when no cold accumulated, and a conservative 75% of the infected plants recovered at ({T}_{min }=-1.{1},^{circ }{{{{{rm{C}}}}}}) instead of 100% to reflect uncertainties on the effect of winter curing.

MGDD/CDD distribution maps

MGDD were used to compute annual risk maps of developing PD during summer for the period 1981–2019 (see “Methods”). The resulting averaged map identifies all known areas with a recent history of severe PD in the US corresponding to ({{{{{{{mathcal{F}}}}}}}}({{{{{rm{MGDD}}}}}}) , > , 90 %) (i.e., high-risk), such as the Gulf Coast states (Texas, Alabama, Mississippi, Louisiana, Florida), Georgia and Southern California sites (e.g., Temecula Valley) (Fig. 2a), while captures areas with a steep gradation of disease endemicity in the north coast of California (({{{{{{{mathcal{F}}}}}}}}({{{{{rm{MGDD}}}}}} , > , 50 % )). Overall, more than 95% of confirmed PD sites (n = 155) in the US (Supplementary Data 2) fall in grid cells with ({{{{{{{mathcal{F}}}}}}}}({{{{{rm{MGDD}}}}}}) , > , 50 %).

The average MGDD-projected map for Europe during 1981–2019 spots a high risk for the coast, islands and major river valleys of the Mediterranean Basin, southern Spain, the Atlantic coast from Gibraltar to Oporto, and continental areas of central and southeast Europe (Fig. 2b). Of these, however, only some Mediterranean islands, such as Cyprus and Crete, show ({{{{{{{mathcal{F}}}}}}}}({{{{{rm{MGDD}}}}}}) , > , 99 %) comparable to areas with high disease incidence in the Gulf Coast states of the US and California. Almost all the Atlantic coast from Oporto (Portugal) to Denmark are below suitable MGDD, with an important exception in the Garonne river basin in France (Bordeaux Area) with low to moderate MGDD (Fig. 2b).

Figure 2a shows how the area with high-risk MGDD values extends further north of the current known PD distribution in the southeastern US, suggesting that winter temperatures limit the expansion of PD northwards⁹. A comparison between MGDD and CDD maps (Fig. 2a vs. Fig. 2c, Fig. 2e) further supports the idea that winter curing is restricting PD northward migration from the southeastern US. However, consistent with growing concern among Midwest states winegrowers on PD northward migration led by climate change⁶³, we found a mean increase of 0.12% y⁻¹ in the areal extent with CDD < 306 ((sim {T}_{min } < -1.1,^{circ }{{{{{rm{C}}}}}})) since 1981, comprising land areas between 103°W and 70^oW of the US (Supplementary Fig. S4). Such an upward trend corresponds to 5090 km² y⁻¹ in the potential northward expansion of PD due to climate change and an accumulation of ~193420 km² of new areas at risk since 1981.

High-CDD values would also be expected to restrict the potential PD colonisation in continental Europe (Fig. 2d). Unlike North America, the East-West distribution of major European mountain ranges together with the warming effect of the Gulf Stream decreases the likelihood of cold winter spells reaching the western Mediterranean coast. ({{{{{{{mathcal{G}}}}}}}}({{{{{rm{CDD}}}}}})) between 100% and 95% (i.e., recovery probability <5% – low winter curing) are mostly prevalent below 40°N latitude in the southwest Iberian Peninsula and Mediterranean islands and coastlands (<50 km away). Above 40°N latitudes, CDD < 100 are encountered mainly in the Atlantic coast and Mediterranean coast and islands (Fig. 2d). In contrast, central and southeast Europe show high CDD values likely preventing Xf_PD winter survival on infected grapevines.

In Fig. 2e, f, we show the average climatic suitability for PD establishment only from the mechanistic relation between Xf_PD and temperature. Although all areas with current Xf_PD-related outbreaks are identified, risk predictions based only on the combination of MGDD and CDD could lead to overestimations, as this approach overlooks disease transmission dynamics and climate interannual variability.

PD global risk

We ran several simulations of the model Eq. (7) with R₀ values between 1 and 14 to validate PD spatiotemporal distribution in the US. We found R₀ = 8 as the optimal parameter for maximising the area under a ROC curve (Supplementary Fig. S5), returning an accuracy of more than 80%, except for 2006, due to data obtained from an area at the transient-risk zone (Supplementary Fig. S7 and Table 1). For Europe and the rest of the world, we derived a R₀ = 5, as a maximal baseline estimate for modelling PD transmission (see “Methods” and Supplementary Note 2D). These R₀ values should be taken as operating estimates for the model. From the model simulations Eq. (7), we obtained a risk index r that measures the relative exponential growth rate in the population of infected plants at the epidemic onset with respect to the maximum growth, r = 1. This index served to rank the epidemic-risk zones in high (>0.9), moderate (0.66–0.9), low (0.33–0.66), and very low (~0.075–0.33) risks (see Fig. 1f, “Methods”, and Supplementary Note 2E).

To date, PD is mainly restricted to the American continent with some unrelated introductions of Xf_PD to Taiwan and Majorca (Spain) from the United States^12,13. To assess the risk of PD establishment elsewhere, we projected our epidemiological model into the main winegrowing regions of the Northern Hemisphere (US, Europe, and China) and Southern Hemisphere (Chile, Argentina, South Africa, Australia, and New Zealand)(Fig. 3a–e). We found that emerging wine-producing areas in China are predominantly located in non-risk zones, whereas only some vineyards in the Henan and Yunnan provinces fall in transition and moderate-high risk zones (Fig. 3b and Supplementary Data 3). In Europe, 92.1% of the territory is in non-risk zones and 6.1% is included in the epidemic-risk zone, with only 1.9% showing a high-risk index and 1.5% a moderate risk (Supplementary Table S2). The model also reveals a progressive transition from areas without risk (r(t) < 0) before 1990 to epidemic-risk zones with low-risk indexes by 2019⁵⁷ (see Movies), mainly affecting the basins of the rivers Po in Italy, Garonne, and Rhone in France and Douro/Duero in Portugal and Spain. This represents a mean increase of 0.21% y⁻¹ in the epidemic-risk zone, a rate 3.5-times greater than that of the eastern US, which could increase the likelihood of PD establishment in Europe in the coming decades. In the US, most states around the Gulf Coast show high-risk indexes, whereas, around 37.5% of California’s surface is suitable for epidemics with high growth rate incidence (Supplementary Table S3).

In the Southern Hemisphere, vineyards at non-risk or transient epidemic-risk zones predominate—e.g., non-risk in New Zealand and Tasmania (Fig. 3c). Risk indexes in areas where PD can become established (r(t) > 0) range from very low to low for most coastal vineyards in Australia (west, south and east) with somehow more suitable conditions in the interior of New South Wales, Greater Perth and Queensland (Fig. 3c); a general very-low or low-risk indexes are predicted in the Western Cape in South Africa (Fig. 3d); overall very-low but localised low to moderate risk indexes in some areas in Chile; and low to moderate growth of the number of infected vines in most of Argentina, being this the wine-growing country with the highest risk (Fig. 3a). Detailed information on areas with non-risk, transient risk and risk indexes (i.e., disease-incidence growth rates) in areas with the potential risk of establishment by country and regions is provided in Supplementary Table S4 and Supplementary Data 3.

Risk indexes may vary within epidemic-risk zones if any of the epidemiological parameters governing transmission change. As expected, I(t) < I(0) boundaries increasingly displace to northern latitudes in the US and Europe under higher transmission scenarios, increasing the risk-epidemic zones significantly (Fig. 4a–f). The line representing the outbreak extinction i.e., the non-risk zone r(t) < −0.09, in the validated R₀ = 8 scenario for the US, falls at some distance above the isoline ({T}_{min }=-1.{1},^{circ }{{{{{rm{C}}}}}}) in comparison to the R₀ = 5 scenario (Fig. 4c vs Fig. 4a and ref. ⁵⁷, Movies). This distribution pattern holds and moves slightly northward over time in parallel to global warming, although the trend of PD latitudinal change is moderated by high-CDD values (i.e., cold accumulation). In addition, the disease extension also declines due to CDD interannual fluctuations in the simulations. Cold waves periodically occur that reach latitudes close to the Gulf, such as those that occurred in 1983, 1993, 1995, 2000, 2009, and 2013⁵⁷ (see Movies), thus preventing PD expansion northward. The magnitude of this decrease is revealed after comparing the average annual increase of the areas between r(t) > 0 and CDD < 306 lines. From 1981 to 2019, the area with risk r(t) > 0 increased at a rate of 0.05% y⁻¹, while that of CDD < 306 by 0.12% y⁻¹, an important difference not explained alone by CDDs without considering climate fluctuations (Supplementary Fig. S4).

We checked whether using a beta function produces changes in the risk indexes with respect to the Arrhenius-based approach. Firstly, we needed to calibrate the model using the probability of developing chronic infections, as in Fig. 1c, with the values of MGDD obtained with the beta function. We found little differences, mainly a decrease in risk index in the transition zones between risk and non-risk zones ((Supplementary Fig. S12) and (Supplementary Fig. S13)), and non-significant differences in risk zones at the global scale.

PD risk projections for 2050

Global shifts in the risk index r_j(t) between 2019 and those projected for 2050 were calculated under the same baseline scenario (Fig. 5a–f, “Methods”). Our simulation shows a generalised increasing trend mainly due to shifts from transition zones to epidemic-risk zones with very low or low-risk indexes in the main wine-growing regions, except for the US. Here the epidemic-risk zone would increase by 12.8% with the greater increments in the high-risk index category (22.7%) and a decrease in the transition zones (Supplementary Table S5). Much less surface would be included in the epidemic-risk zone in Europe (8.6%) compared to the US (36.5%). However, the epidemic-risk zone would expand by 40.0% with respect to 2020, a rate more than three times higher than that of the US (Supplementary Table S6). Such increases are due to the emergence of previously unaffected areas in 2020 evolving into epidemic-risk zones by 2050, and epidemic growth-rate increases in already epidemic-risk zones in 12 of 42 countries (Supplementary Table S2). Among these 12 countries, however, there is substantial variation in the risk index increments within epidemic-risk zones with respect to 2019 (Supplementary Table S6). While non-risk zones still cover 87.6% of Europe’s land area, epidemic-risk zones with high-risk indexes are expected to be almost two-fold higher than that of 2019, comprising 3.2% of Europe (Table 2).

Risk based on vector information

So far, we have ignored the distribution of known and potential vector species due to their large number in the Americas and the limited quantitative information generally available. In the case of Europe, given P. spumarius prevalence as a potential vector and its wide distribution, we added a vector layer in a spatially dependent ({R}_{0}(j)={R}_{0}^{max },v(j)), where v(j) is the climatic suitability for the vector (“Methods”), v = 1 implies optimal climatic conditions with no constraints for the vector population size, while v = 0 implies unsuitable climatic conditions and its absence (Supplementary Fig. S8). According to the model, no European zone shows a high-risk index and barely 0.34% of the territory falls in areas with potential moderate exponential growth rates in disease incidence (Supplementary Table S7). Irrespective of vineyard distribution, we estimated that PD could potentially become established (i.e., r(t) > 0) at a maximum of 3.1% of the territory, while the area at moderate-risk index would be 5-times lesser than the model without the vector’s climate suitability layer, this latter more in consonance with other proposed risk maps^45,46. Such differences in the projected risks are mainly concentrated in the warmest and driest Mediterranean regions and are due to uncertainties concerning temperature-humidity interactions in the ecology of the vector³⁵.

Combining vineyard land cover across Europe with the model output

When we integrate into the model a layer of vineyard surface from Corine-Land-Cover, we find that PD could potentially become established (i.e., r(t) > 0.075) in 22.3% of the vineyards in Europe. However, no vineyard is in epidemic-risk zones with a high-risk index and only 2.9% of the vineyard surface is at moderate risk (Supplementary Table S8). The areas with the highest risk index (r(t) between 0.70 and 0.88) are mainly located in the Mediterranean islands of Crete, Cyprus and the Balearic Islands or at pronounced peninsulas like Apulia (Italy) and Peloponnese (Greece) in the continent (Fig. 6a and Supplementary Table S8). Most vineyards are in non-risk zones (42.1%), whereas 35.6% are located in transition zones with presently non-risk but where Xf_PD could become established in the next decades causing some sporadic outbreaks. In Supplementary Data 4 and Supplementary Table S8, we provide full details of the total vineyard areas currently at risk for each country and region.

Our model with climate and vector distribution projections for 2050 indicates a 55.8% increase in the epidemic-risk zone in Europe (Fig. 6b). This increment would be mainly due to the extension of epidemic-risk zones with very low and low-risk indexes. However, within the epidemic-risk zones, areas with moderate risk indexes would decrease from 114925 ha in 2020 to 43114 ha in 2050, and no vineyards would be at high risk (Fig. 6b; see Supplementary Table S9 and Supplementary Data 4). Counterintuitively, our model indicates a substantial increase in the area where PD could establish and become endemic for 2050, but a moderate decline in those areas where crop damage could be expected to be significant (e.g., Balearic Islands, Crete, Cyprus, Apulia).

Source: Ecology - nature.com