Understanding the effects of multiple sources of seasonality on the risk of pathogen spread to vineyards: vector pressure, natural infectivity, and host recovery

Authors


E-mail: bgruber@ufl.edu

Abstract

Seasonality plays an important role in the dynamics of infectious disease. For vector-borne pathogens, the effects of seasonality may be manifested in the variability in vector abundance, vector infectiousness, and host-infection dynamics over the year. The relative importance of multiple sources of seasonality on the spread of a plant pathogen, Xylella fastidiosa, into vineyards was explored. Observed seasonal population densities of the primary leafhopper vector, Graphocephala atropunctata, from 8 years of surveys in northern California were incorporated into a model of primary spread to estimate the risk of pathogen infection under different scenarios regarding seasonality in vector natural infectivity (i.e. constant or increasing over the season) and grapevine recovery from infection (i.e. none or seasonal recovery). The extent to which local climatic conditions affect risk estimates via differences in vector abundance was investigated. Seasonal natural infectivity, seasonal recovery, and especially the combination, reduced (up to 8-fold on average) within-season and cumulative yearly estimates of pathogen spread. Estimated risk of infection also differed greatly among years due to large differences in vector abundance, with wet and moderate winter and spring conditions favouring higher G. atropunctata abundance. Seasonal variation of the pathogen–vector interaction may play an important role in the dynamics of disease in vineyards, reducing the potential prevalence from what it could be in their absence. Moreover, climate, by affecting sharpshooter leafhopper abundance or activity, may influence Pierce’s disease dynamics.

Introduction

Disease dynamics in a susceptible host population are inherently a function of time (Madden et al., 2007). For vector-borne pathogens, rates of spread may vary over time due to underlying variability in vector, pathogen, or host interactions. For example, confirmed cases of malaria in French Guiana have been reported to demonstrate a seasonal pattern that follows increases in cumulative rainfall and river levels thought to support Anopheline mosquito populations (Girod et al., 2011). Similarly, black turpentine beetle (Dendroctonus terebrans), a pest of Pinus spp., was most abundant on longleaf pine (P. palustris) during winter months, the same time of year that Ophiostoma ips-like fungal pathogens of pines are isolated most frequently from the beetle (Zanzot et al., 2010). Thus, seasonal variation in vector or pathogen abundance may translate to dramatic differences in the risk of disease development. The current study evaluates the relative importance of different sources of seasonality in affecting the risk of infection by an economically important plant pathogen, Xylella fastidiosa, in coastal northern California vineyards.

Pierce’s disease of grapevine is caused by the xylem-limited bacterium X. fastidiosa. Symptoms of this degenerative disease include progressive chlorosis, necrosis of the leaf blade and defoliation (Hopkins & Purcell, 2002). Typically, diseased grapevines die within a few years. Several species of xylem sap-feeding insects are capable of transmitting X. fastidiosa (Severin, 1949). The most important of these are the sharpshooter leafhoppers. The ability of these vectors to promote rapid pathogen spread has been attributed to the lack of a latent period (Purcell & Frazier, 1985) and persistent infectiousness in adults (Severin, 1949).

The key vectors of Pierce’s disease vary geographically throughout California. In southern and central California, the main vectors are Homalodisca vitripennis and Xyphon fulgida, respectively (Hopkins & Purcell, 2002). In coastal California vineyards, the predominant leafhopper vector of X. fastidiosa is the blue-green sharpshooter, Graphocephala atropunctata (Purcell, 1975, 1981). This insect is an efficient vector of X. fastidiosa to grapevines, more so than other sharpshooter species in the region (Almeida & Purcell, 2003; Daugherty & Almeida, 2009). Like other sharpshooters, G. atropunctata is polyphagous (Winkler, 1949; Freitag & Frazier, 1954) and capable of overwintering in riparian vegetation that is often found adjacent to vineyards in coastal California (Hopkins & Purcell, 2002). Trap catches of G. atropunctata are lower in the vineyard interior, compared to the outermost vineyard rows, indicating vector movement from riparian corridors into vineyards (Purcell, 1975). Plant species found in these riparian habitats have been reported as competent hosts of X. fastidiosa (Baumgartner & Warren, 2005), and grapevines showing symptoms are more common near the riparian vegetation–vineyard interface than in the vineyard interior (Purcell, 1974). Thus, riparian habitats function as a source of vectors and the pathogen, facilitating primary spread of X. fastidiosa into nearby vineyards (Purcell, 1974, 1981).

Sharpshooter populations appear to exhibit highly seasonal dynamics. Purcell (1975) reported that G. atropunctata trapped in riparian habitats peak in the early spring, and then decline sharply through late June. Trap catches in adjacent vineyards showed similar seasonal patterns, with more than 50-fold differences in the number of sharpshooters caught over the season in the interior of vineyards and more than 100-fold difference on vineyard edges (Purcell, 1975). Strong seasonality has also been observed in other sharpshooter species in the region (Blua et al., 2001). Moreover, large differences in sharpshooter abundance have been noted among years (Severin, 1949; Purcell, 1979; Daugherty et al., 2012). Climate has been suggested to underlie some of this variability. Boyd & Hoddle (2006) reported that G. atropunctata abundance was positively associated with degree-day units in late spring, and variability in winter precipitation may explain some of the year-to-year variation (Severin, 1949). Yet the underlying causes and epidemiological consequences of G. atropunctata population dynamics remain elusive.

In addition to variability in vector abundance, the natural infectivity of vectors (i.e. the proportion of vectors that are infectious) can vary seasonally (Adlerz & Hopkins, 1979; Carraro et al., 2004). The natural infectivity of G. atropunctata moving into vineyards is probably an important component to estimating the amount of primary spread of X. fastidiosa. Even if ambient reservoirs of X. fastidiosa are accessible to large populations of G. atropunctata, primary pathogen spread is likely to be constrained by the actual number of vectors that are infectious because X. fastidiosa is xylem-limited and sessile (Purcell, 1981). Two studies have investigated G. atropunctata natural infectivity in northern California vineyards over the season (Freitag & Frazier, 1954; Purcell, 1975). Yet the two studies document qualitatively different patterns over the season. One of them (Purcell, 1975) appears to show little seasonality, whereas the other (Freitag & Frazier, 1954) shows far higher infectivity later in the season. Whether these different assumptions regarding how sharpshooter infectiousness changes over the season appreciably affect disease incidence is not known.

Still another source of seasonality in this pathosystem relates to grapevine infection dynamics. Pierce’s disease is notable because of the propensity for infected grapevines to recover from X. fastidiosa infection (Purcell, 1977, 1980; Feil et al., 2003; Hill et al., 2006). In other words, acute infections do not necessarily develop into chronic infections. The exact mechanism of recovery has not yet been elucidated; it may be that late-season infections do not have sufficient time to move systemically within the grapevine before the onset of inhospitable winter temperatures (Redak et al., 2004). Concentrations of X. fastidiosa vary seasonally, with low infection levels during the early, cooler season (Hopkins, 1981; Hopkins & Thompson, 1984). Recoveries, if they occur, do so at some point between leaf fall and budbreak. Perhaps most importantly though, is that the proportion of grapevines that recover depends on the timing of when they are first infected with X. fastidiosa. Early-season infections are more likely to persist over winter than are later-season infections (Purcell, 1981; Feil et al., 2003; Hill et al., 2006). This recovery should act to temper Pierce’s disease incidence. In addition, the seasonal nature of it indicates that rates of chronic pathogen spillover from riparian habitats should depend on not only the total number of vectors moving into vineyards, but also the timing of when vector incursion occurs.

The sharpshooter–Xylella fastidiosa–grapevine pathosystem appears to include at least three sources of seasonality (i.e. sharpshooter abundance, natural infectivity, overwinter recovery) that may be epidemiologically significant. Yet the degree to which these factors, especially in combination, influence the risk of grapevine infection with X. fastidiosa has not been resolved. Therefore, the goals of the current investigation were to: (i) use a multiyear survey of G. atropunctata seasonal abundance coupled with historic estimates of sharpshooter natural infectivity and grapevine recovery to evaluate the consequences of these sources of seasonality on the risk of pathogen infection; and (ii) quantify the extent to which climate underlies variability in X. fastidiosa spread due to effects on G. atropunctata abundance.

Materials and methods

Modelling risk of pathogen infection

Comprehensive estimates of X. fastidiosa incidence do not exist, which would be needed to determine definitively the importance of seasonality. However, parameter estimates are available in the literature that can be used to generate plausible estimates of risk of infection under different scenarios. These data include a long-term data set on sharpshooter seasonal abundance in vineyards (Daugherty et al., 2012), and epidemiologically important variables such as transmission efficiency (Daugherty & Almeida, 2009), natural infectivity (i.e. the proportion of vectors that are infectious; Freitag & Frazier, 1954; Purcell, 1975), and grapevine overwinter recovery rate (Feil et al., 2003; Hill et al., 2006). Below are descriptions of these parameters and the risk model (Purcell, 1981) used to generate relative estimates of G. atropunctata -mediated X. fastidiosa infection of grapevines.

Field surveys of Graphocephala atropunctata abundance

Between 2001 and 2008 G. atropunctata were monitored regularly in plots in a vineyard located in the Oak Knoll region of Napa County, CA, USA, adjacent to the Napa River (38°20′31·82′′ N, 122°17′18·81′′ W). The goal of the study, described in more detail elsewhere (Daugherty et al., 2012), was to test whether trees planted between the riparian corridor and vineyard edge act as a physical barrier to sharpshooter incursion. Only the control plots, for which there was open space between the vineyard edge and nearby riparian vegetation (i.e. no physical barrier to insect movement), were considered. There were three control plots in the study, each 60 m in length and located approximately 10 m from the riparian corridor.

In spring 2001, three yellow sticky traps were deployed in each of the vineyard plots to monitor G. atropunctata. Traps (23 × 14 cm) were attached to the vineyard trellis system approximately 1·5 m above the ground, a few vines from the end of the row, with one side facing the river. All traps were orientated parallel to the river and were approximately equidistant from the riparian corridor. All traps were checked every 7–10 days from spring to early autumn in most years. At each census, the number of G. atropunctata was recorded and then new sticky traps were deployed. Monitoring concluded in 2008, resulting in a total of 194 censuses over 8 years (Fig. 1). Although no direct observations were made in this study regarding Pierce’s disease incidence, the measures of G. atropunctata temporal dynamics are useful for estimating the risk of pathogen infection. To do this, the sum of the three traps per plot on each census served as an estimate of vector number, which was included in the model of infection risk (Purcell, 1981).

Figure 1.

 Mean (±SE) seasonal number of adult Graphocephala atropunctata per plot for 8 years. Note that the scale of the y-axis differs among panels.

Natural infectivity

Estimates of seasonal G. atropunctata natural infectivity (i.e. the proportion of vectors that are infectious) come from two separate field studies conducted in northern California vineyards (Freitag & Frazier, 1954; Purcell, 1975). In each study, field collected G. atropunctata were confined on healthy grapevines, whose X. fastidiosa infection status was later evaluated. These tests were conducted monthly for 7 months in Freitag & Frazier (1954), and 12 months split over three consecutive years in Purcell (1975).

Observations from the two data sets were used to generate a quantitative description of natural infectivity over the year. Specifically, the observed natural infectivity of G. atropunctata from each study separately was subjected to curve fitting exercises to determine the most parsimonious model fit. Four candidate models were evaluated for each data set separately after visual inspection of the two data sets: a two parameter exponential non-linear regression, three parameter logistic non-linear regression, a linear regression, or an intercept-only model (Table 1). For both data sets it was assumed that the fifteenth day of each month was the trial date, and natural infectivity measures among dates were assumed to be independent of each other. Akaike’s information criterion (AIC) was used to score the relative fit of each model, for the two natural infectivity data sets (Pinheiro & Bates, 2000). The two data sets generated different preferred models for temporal patterns of natural infectivity over the year (Fig. 2). Therefore, the consequences of these two different descriptions of seasonal natural infectivity on pathogen infection were explored as alternatives in the risk modelling.

Table 1. Results of model fitting and selection for the relationship between day of the year (DOY) and observed Graphocephala atropunctata natural infectivity from one of two historical data sets
ModelEquationAIC scoresa
Freitag & Frazier, 1954 Purcell, 1975
  1. aAkaike information criterion; preferred models have smaller values.

  2. bModel did not fit the observed data and was not considered further.

Logistic inline image N/Ab−40·89
Exponential inline image −0·68−42·39
Linear inline image −1·12−42·41
Intercept only y=a09·21−43·57
Figure 2.

 Seasonal proportion of Graphocephala atropunctata that transmitted Xylella fastidiosa from two separate field surveys. Data from Purcell (1975), (a), were best described by constant natural infectivity (0·275 ± SE = 0·021). Data from Freitag & Frazier (1954), (b), were best described as exponentially increasing over the season (= 0·0124e0·0135*DOY).

Overwinter recovery

Feil et al. (2003) examined the phenomenon of grapevine overwinter recovery in detail by inoculating grapevines with X. fastidiosa on different days of the year. The following season vines were assessed again to determine whether they were still infected. Results indicated that the percentage of grapevine recovery increases linearly until approximately the end of May or early June, after which the percentage of recovery reaches an inflection point and approaches 100%. Feil et al. (2003) concluded that a saturating, monotonic non-linear model best described the relationship between percentage of grapevines recovering and day of year that plants were inoculated (DOY):

image(1)

Hill et al. (2006) reported a similar pattern of grapevine overwinter recovery from X. fastidiosa as a function of the day of year that vines were inoculated. The consequence of this seasonal recovery was explored in the risk modelling. It should be noted that modelling endeavours have uncovered the possibility that this overwinter recovery may be a function of geographic location and local climate patterns (Lieth et al., 2011). However, the modelling in the current study did not account for such spatial heterogeneity as only one site is being considered.

Modelling framework

Purcell (1981) described a model for quantifying the risk of infection by X. fastidiosa via sharpshooters. Purcell (1981) assumes that X. fastidiosa is spread to grapevines in a primary manner (i.e. introduction from outside the vineyard). In other words, secondary (i.e. vine-to-vine) spread is assumed to be negligible, which is supported by observed patterns of Pierce’s disease in coastal northern California vineyards (Purcell, 1974, 1975). The Poisson form of this primary spread model (Purcell, 1981) was used to generate risk estimates for G. atropunctata -mediated X. fastidiosa infection:

image(2)

where Pnt = the risk of infection (i.e. the probability that any single grapevine is infected during a given time interval, t), = the number of vectors, = the natural infectivity, = the proportion of vector–plant contacts that result in X. fastidiosa infection, and = the time interval over which pathogen infection occurs. For this study, it was always assumed that = 1 day. It is important to mention that Pnt does not quantify the number of affected plants, but rather represents the risk, categorically, of any of a group of vines becoming infected with X. fastidiosa.

As noted already, X. fastidiosa infection does not necessarily equate to chronic Pierce’s disease. Reduction in the overall risk of disease establishment should be expected if some acute infections are lost over the winter (Feil et al., 2003; Hill et al., 2006). Purcell (1981) included such transient infections into the probability model through the addition of a recovery term:

image(3)

where = recovery proportion (i.e. the proportion of individual grapevines recovered from infection; equal to y from Eqn 1). Thus, recovery is assumed to ameliorate relative levels of risk. In other words, Pnt can be considered to represent the risk of chronic infection. Estimates of risk were then evaluated by substituting different observed patterns of G. atropunctata abundance among years (Daugherty et al., 2012), different scenarios for seasonality in vector natural infectivity (Freitag & Frazier, 1954; Purcell, 1975), and assumptions of recovery (Pnt vs Pnt).

Throughout, the proportion of vector–plant contacts that result in X. fastidiosa infection, E, is assumed to be a constant, whose value was obtained from the literature. Daugherty & Almeida (2009) conducted a meta-analysis of G. atropunctata – X. fastidiosa inoculation rate to grape. Based on that analysis, inoculation rate (i.e. E) was 0·593 per day. This regression coefficient can be substituted into the risk model as an estimate of E. Conversely, vector number, n, natural infectivity, i, and recovery proportion, r, are assumed to vary over time, with values approximated from the data sets described above.

Observations of the temporal dynamics of G. atropunctata over 8 years (Daugherty et al., 2012) were incorporated into the model as a measure of vector number, n. Natural infectivity, i, was described alternately by either the best fit models for the Freitag & Frazier (1954) or Purcell (1975) data sets (Fig. 2). For Eqn 3, recovery proportion, r, was estimated as an increasing function of day of year as described by Feil et al. (2003; Eqn 1).

Data analysis

Observations of G. atropunctata abundance from 2001 to 2008 in the control plots of the field study were used to explore the likely effects of the two sources of seasonality on the risk of infection. For all analyses, the plot was considered the experimental unit for each census date. For each census date, an estimate of risk of X. fastidiosa infection (i.e. Pnt) for each of the three vineyard plots was calculated (Eqn 2). Additionally, for each census date, an estimate of risk of X. fastidiosa infection assuming overwinter recovery (i.e. Pnt) was calculated (Eqn 3). Next, risk estimates from both Eqn 2 and Eqn 3 were calculated using either the assumption that natural infectivity was constant or seasonal (Fig. 2). Thus, four combinations of seasonality were evaluated, assuming: (i) constant natural infectivity and no recovery, (ii) constant natural infectivity and seasonal overwinter recovery, (iii) seasonal infectivity and no recovery, and (iv) seasonal infectivity and overwinter recovery. Though generated using the same values for vector abundance and transmission efficiency, the different natural infectivity and recovery assumptions were treated as separate, independent treatments. These effects of the different combinations on risk of pathogen spread were analysed at two different time scales.

First, comparisons were made of within-season risk among census dates. Given the high number of zero vector catches in most years, these analyses were restricted to 2003 and 2004, the 2 years for which sharpshooter catches were most regular. A linear mixed-effects model repeated measures analysis (proc mixed, sas v. 9.2) was used to evaluate differences in risk among the four different assumption combinations for 2003 and 2004, with each year analysed individually. Census date, natural infectivity (constant or seasonal), and overwinter recovery (Eqn 2 or Eqn 3) were considered fixed effects. A compound symmetry covariance structure was used to account for serial observations made on the same plot; therefore, a unique random effect for plot was not declared, as recommended by Littell et al. (2006). A Kenward–Roger adjustment was used to calculate degrees of freedom. Mean separation t-tests using Tukey-adjusted significance values for multiple comparisons were used when necessary.

Next, comparisons of risk variability among years were made. The risk of pathogen infection was calculated for a given plot for each census date, and then these measures were summed over the year. This was performed for all 8 years of the survey. These cumulative values, which are not bounded between 0 and 1, allow for relative comparisons of yearly risk of pathogen infection among the different assumptions made regarding natural infectivity and seasonality. In order to control for differences in sampling effort and G. atropunctata abundance among years, cumulative vector number in a plot for each year was first regressed against cumulative risk for that year (i.e. eight data points per plot). In all instances, the addition of an autoregressive covariance structure to account for repeated measures did not improve the fit of linear models; therefore, all errors among years were assumed to be independent. The estimated slopes from the regressions, which reflect the per capita risk associated with cumulative vector abundance, were then compared among the four combinations of seasonality assumptions using a two-way analysis of variance (Crawley, 2009). A significant interaction between natural infectivity and recovery assumptions was followed up with pairwise t-tests among the four combinations, with Bonferroni correction for multiple tests.

Climatic effects on Graphocephala atropunctata

Sharpshooter abundance varied greatly among years in the field survey (Daugherty et al., 2012) (Fig. 1). The extent to which differences in climatic conditions influenced the variability in sharpshooter populations was investigated. Precipitation and temperature data were subjected to principal components analysis, then a simplified set of climate variables were used to explain sharpshooter abundance.

Climate data were retrieved from the United States Historical Climatology Network database (http://cdiac.ornl.gov/epubs/ndp/ushcn/ushcn.html) for a weather station nearest to the field site (Napa State Hospital, Napa, CA, USA; 38°16′30′′ N, 122°15′45′′ W), for autumn of 2000 to autumn of 2008. Initially, a wide range of temperature and precipitation metrics from different windows of time over this period were considered. However, due to strong correlations between many of these alternate metrics, the suite of candidate metrics was reduced to four unique climate variables: ‘early precipitation’ (i.e. total precipitation from October to May), ‘early temperature’ (i.e. mean daily average temperature from October to May), ‘growing season maximum temperature’ (i.e. mean daily maximum temperature from June to the last census date), and ‘number of cold days’ (i.e. number of days from November to April with daily minimum temperature within one standard deviation of the coldest temperature recorded during the study period [2·37°C]). Values of these four climate variables were calculated for each year of the field study.

The four climate variables were incorporated into principal components analysis (PCA; Crawley, 2009). The output from the PCA consisted of four components, the first two of which account for more than 84% of the total variation. Thus, further analyses were limited to just these two components. The effects of climate were analysed via linear regression of three different metrics of sharpshooter abundance on the first two principal components, separately. The first two metrics relate to the yearly observed peak in sharpshooters caught in a given plot; the timing of the peak (i.e. day of year) and the size (i.e. number of sharpshooters) of the peak. A third metric calculated was vector density for each plot over the year (i.e. cumulative no. of vectors/no. of censuses), which controls for differences in sampling effort among years. Each test consisted of a mixed effects linear regression, assuming a compound symmetry covariance structure to account for repeated measures on the same plots (Crawley, 2009). Values for the size of the peak and cumulative density were log10 transformed to meet test assumptions.

Results

Modelling risk of pathogen infection

Graphocephala atropunctata natural infectivity

A re-analysis of two prior G. atropunctata natural infectivity surveys indicated that the two studies differed in their temporal patterns. Based upon AIC scores (Table 1), the most parsimonious explanation for the observations from Purcell (1975) showed no evidence of seasonality. Instead, natural infectivity was best described with the intercept-only linear model (Table 1), with a constant of approximately 0·3 over the season (Fig. 2a). The other data set, Freitag & Frazier (1954), was best explained with a simple linear model (Table 1). However, the biological interpretation of the predicted values of natural infectivity from the simple linear model was unreasonable because the predictions yielded negative probabilities until day 160. Thus, the simple linear model was not retained. The next most parsimonious fit to describe the Freitag & Frazier (1954) data set (Table 1) was an exponential function (= 0·0124e0·0135*DOY) for which the proportion of vectors that were infectious increased from approximately 0·1–0·9 between May and October (Fig. 2b).

Estimated risk of pathogen infection

Within-season risk of pathogen infection in 2003 or 2004 was not significantly affected by the census date × infectivity × overwinter recovery interaction (Table 2). However, in both years two-way interactions between census date and overwinter recovery, census date and infectivity, and overwinter recovery and infectivity were all significant (Table 2). Seasonal overwinter recovery reduced estimated risk compared to the no recovery assumption, particularly later in the season (Fig. 3a,b). Seasonal natural infectivity also reduced estimated risk compared to constant infectivity, particularly early in the season (Fig. 3c,d). Finally, the combination of the two sources of seasonality resulted in estimated overall risk that was approximately one quarter to one half of each seasonality source alone and one ninth to one third of the assumption that neither source of seasonality occurred (Fig. 3e,f).

Table 2. Statistical results for the effects of natural infectivity (NI; constant or seasonal), recovery (R; none or seasonal), census date (D), and their interactions on the predicted within-season risk in 2003 and 2004
Source20032004
Fdf, dfe P Fdf, dfe P
NI154·321,102<0·000163·931,254<0·0001
R186·401,102<0·000179·921,254<0·0001
D17·1712,102<0·000114v6731,254<0·0001
NI*R26·491,102<0·00017·711,254<0·0059
NI*D4·2012,102<0·00014·1631,254<0·0001
R*D7·4512,102<0·00013·5931,254<0·0001
NI*R*D0·8112,1020·64420·4931,2540·9909
Figure 3.

 Mean (±SE) predicted risk of Xylella fastidiosa infection among dates for 2003 (a, c, e) and 2004 (b, d, f). Panels (a) and (b) depict a significant date-by-recovery (i.e. none or seasonal) interaction, and panels (c) and (d) depict a significant date-by-natural infectivity (i.e. constant or seasonal) interaction. * denotes a significant pairwise difference between seasonality assumptions on a given date. Panels (e) and (f) depict a significant natural infectivity-by-recovery interaction. Different letters above bars denote significant differences in risk.

The per capita cumulative risk of infection over the year was significantly affected by the assumption of natural infectivity (F1,8 = 93·33, < 0·0001), recovery (F1,8 = 107·39, < 0·0001), and the interaction between them (F1,8 = 13·51, < 0·0063). Again, each source of seasonality lowered the estimated risk compared to the assumption of no seasonality, which was nearly 8-fold higher than for the combination of both sources of seasonality (Fig. 4).

Figure 4.

 Mean (±SE) predicted per capita cumulative risk of Xylella fastidiosa infection (cumulative risk/cumulative no. of vectors) among different seasonality assumptions for all 8 years of the field survey. Different letters above bars denote significant differences in per capita cumulative risk.

Climatic effects on Graphocephala atropunctata

Results from the principal components analysis revealed that the first component (PC1) was characterized as having a strong negative loading of early precipitation, but strong positive loading of the number of cold days. Thus, PC1 reflects the extent of dry, cold winter conditions. PC2 included a strong negative loading of growing season maximum temperature, but a strong positive loading of early temperature.

PC1 had a significant effect on cumulative vector density (F1,19 = 10·39, = 0·0045) and the size of the peak catch (F1,19 = 5·58, = 0·029), but not the timing of peak catch (F1,19 = 0·20, = 0·6587). PC2 did not affect significantly cumulative vector density (F1,19 = 1·24, = 0·28) or the timing of the peak catch (F1,19 = 2·29, = 0·1463), but there was a marginally significant effect on the size of the peak catch (F1,19 = 4·151, = 0·0558). Higher PC1 scores, which are associated with colder and drier conditions, result in lower vector densities and smaller peak catches (Fig. 5).

Figure 5.

 Effect of climatic conditions on (a) cumulative Graphocephala atropunctata density (cumulative no. of vectors/no. of censuses) over the year, and (b) the size of yearly peak in catches. The x-axis represents the most important component (PC1) from a principal components analysis of temperature and precipitation near Napa, CA. PC1 is characterized by conditions October–May that are increasingly dry and cold. Dashed lines denote the fit of mixed effects linear regressions. Note that some points are obscured due to overlap.

Discussion

Vector abundance is widely recognized as a primary determinant of disease risk or the pace of outbreak (Killeen et al., 2000; Napp et al., 2010; Girod et al., 2011). However, if strong seasonality exists, then not only the abundance of vectors but also the timing of vector activity may have strong effects on disease incidence. Here, the likely individual and combined effects of seasonality in vector abundance, natural infectivity, and host-plant recovery on the risk of chronic X. fastidiosa spread in Californian vineyards were explored. The results of this modelling indicate the potential for strong interactive effects among these sources of seasonality. Moreover, an analysis of vector abundance suggests that climate may underlie year-to-year differences in vector pressure.

The current investigation analysed censuses of G. atropunctata abundance in a coastal California vineyard over 8 years, representing the most comprehensive survey for this important vector, to predict the risk of pathogen spread in vineyards. Vector abundance peaked between late April and middle of May, after which abundance declined throughout the remainder of the year. These patterns were similar to the sticky trap catches of G. atropunctata that Purcell (1975) reported. Boyd & Hoddle (2006) monitored adult G. atropunctata activity for 2 years in southern California and concluded that peak abundance occurred in late June and early July, after which abundance declined noticeably for the remainder of summer. Therefore, it appears that G. atropunctata abundance throughout grape growing regions in California is highest between late spring and mid-summer and is rarely found in vineyards the remainder of the year. However, comprehensive studies focusing on how riparian habitats influence G. atropunctata activity are lacking. For example, while it has been noted previously that G. atropunctata activity can differ substantially within and among growing seasons in riparian habitats (Daugherty et al., 2012), it is not yet clear how abiotic and biotic influences help drive G. atropunctata population dynamics in such environments. However, the leafhopper vectors also show similarly strong seasonality in abundance. For example, Circulifer tenellus, an efficient vector of the beet leafhopper-transmitted virescence agent phytoplasma, in the Columbia Basin of Washington and Oregon, show peak abundance in early or late June and then decline noticeably in late June or early July (Munyaneza et al., 2008). Populations of Psammotettix alienus, a vector of Wheat dwarf virus, also peak in middle to late June and decline throughout July (Lindblad & Arenö, 2002). In the absence of other information, targeting these peaks in vector abundance may be a preferred strategy for disease management as they probably represent windows of vulnerability to pathogen spread. However, if other sources of seasonality exist in the pathosystem, as appears to be the case with Pierce’s disease, identifying clear vulnerability periods may be challenging.

Although trap catches of sharpshooter leafhoppers reflect a vector whose abundance oscillates throughout the growing season, it is the proportion of vectors that are actually infectious that is most directly linked to new infections. The current work re-analysed two historical data sets where the proportion of G. atropunctata carrying X. fastidiosa was estimated from field catches. One data set (Freitag & Frazier, 1954) exhibited a dramatic increase in the proportion of infectious G. atropunctata over the course of the growing season. Conversely, the proportion of infectious G. atropunctata in the other data set (Purcell, 1975) exhibited substantial variability, but no clear seasonal trend. The two studies used slightly different methods to estimate natural infectivity, testing individual vectors (Purcell, 1975) or groups of vectors (Freitag & Frazier, 1954), which may marginally affect the observed values on a given date. However, it is not clear that this difference in methodologies would correspond necessarily to differences in overall seasonal patterns of infectivity. Nonetheless, the consequence of realizing one or other scenario was demonstrated clearly: constant vector infectivity resulted in significantly elevated risk of acute disease development early in the growing season. In other systems, natural infectivity of adult arthropod vectors varies over the growing season as a function of insect age (Carraro et al., 2004) and gender (Lessio et al., 2009). Therefore, failing to incorporate vector demography traits into analyses of natural infectivity could obscure seasonal trends. Resolving this issue will probably require more extensive and more detailed surveys of G. atropunctataX. fastidiosa natural infectivity over the growing season at different locations.

The loss of X. fastidiosa infection over the winter has been noted for grapevines (Feil et al., 2003; Hill et al., 2006) and other susceptible plant species (Ledbetter et al., 2009). There is also evidence of grapevine recovery from other pathogens, such as phytoplasma (Bellomo et al., 2007). For X. fastidiosa in grapevines, overwinter recovery rates range from 20% for those vines infected early in the spring to 100% for those vines infected after mid-summer (Feil et al., 2003). Thus, recovery reduces risk of infection, because it constrains the extent to which late season inoculations can develop into chronic disease. In other words, late season arrival of vectors contributes little in promoting chronic pathogen spread. This form of seasonality, in conjunction with the rarity of G. atropunctata in vineyards later in the season, has been argued to explain why G. atropunctata -mediated spread appears to resemble patterns of primary spread (Purcell, 1981). Conversely, expectations for Pierce’s disease spread in southern California, mediated by the invasive glassy-winged sharpshooter, H. vitripennis, may be more consistent with secondary spread due to this vector’s higher activity levels in vineyards throughout the year (Almeida et al., 2005). A broad understanding of the role that recovery plays in Pierce’s disease dynamics will require more extensive documentation of recovery proportions in the field. Currently, only two studies have measured seasonal Pierce’s disease recovery. However, it should be noted that one of those studies was conducted in northern California (Feil et al., 2003), where greater vine recovery proportions were observed compared to a similar study conducted further south (Hill et al., 2006), indicating that local climate may influence the magnitude of overwinter recovery. This hypothesis was clarified recently in an analysis by Lieth et al. (2011), which showed geographic differences in recovery due to differences in local climate.

A clear limitation of the current study is the lack of concurrent measurement of Pierce’s disease prevalence over the years that G. atropunctata were surveyed (Daugherty et al., 2012). Given the absence of available disease prevalence data, probability modelling was used to generate estimates of risk based on known epidemiological components of this pathosystem. As such, this modelling work is not intended to be used for disease forecasting per se. Specifically, a Poisson probability model (Purcell, 1981) was used to estimate the likelihood of X. fastidiosa infection under combinations of different biologically plausible scenarios regarding seasonality in natural infectivity (Freitag & Frazier, 1954) and overwinter recovery (Feil et al., 2003). This modelling approach is bounded, for a given time step, between 0 and 1. Therefore it is perhaps best viewed as a relative estimate of ‘risk’ (i.e. the probability that any single grapevine becomes infected during a given time interval). In other words, the goal of this study was not to forecast the net number of affected plants over time, but rather to identify the relative potential for Pierce’s disease to establish within a group of grapevines.

Similar models of disease risk have been employed for understanding the epidemiology of human pathogens as a function of various parameters related to vector or host abundance and the efficiency of pathogen transmission (Killeen et al., 2000; Girod et al., 2011). For example, Napp et al. (2010) used an approach similar to that of the current investigation in assessing the risk of canine rabies spreading to the European Union from Morocco and the effectiveness of interventions, based on historical records. The risk model used in the current study considered the most germane aspects of Pierce’s disease epidemiology, including grapevine overwinter recovery and either constant or seasonal natural infectivity of G. atropunctata. As noted, the results suggest that seasonality in either of the epidemiological features evaluated constrains the extent to which pathogen spread can occur. Notably, it is clear that multiple sources of seasonality may interact with each other to reduce dramatically the risk of X. fastidiosa infection compared to the case of no seasonality (up to 8-fold on average). Such modest levels of Pierce’s disease incidence in most years are typical for vineyards in coastal California (S. Purcell, ESPM Department, University of California, Berkeley, personal communication). Without disease data it is not possible to evaluate the utility of the model. Nonetheless, these results provide a first hypothesis regarding the moderating influence of seasonality on Pierce’s disease incidence; one which future disease surveys could evaluate.

Collectively, over the 8 years of G. atropunctata surveys, there were dramatic between-year differences in vector pressure, with up to 50-fold differences on average between some years (e.g. 2004 vs 2008). These differences in vector catch contributed to clear differences in the predicted cumulative infection risk among years, regardless of the seasonality assumptions. The results of the principle components analysis indicate that climatic conditions may underlie some of that variability among years. Specifically, G. atropunctata trap catches and estimated infection risk were negatively associated with dry, very cold overwinter to spring conditions. Precipitation was mentioned by Purcell (1979) as being a critical environmental factor favouring G. atropunctata abundance. Moreover, populations of another important sharpshooter, H. vitripennis, in southern California have also been reported to increase with increased degree-day units (Castle et al., 2005). It is not clear whether trap catches represent vector population density or simply vector activity. Yet, distinguishing between them may not be important if a pest’s main source of damage stems from pathogen transmission, as is the case with sharpshooters (Redak et al., 2004). Regardless, this result contributes to a growing recognition of the role that climate (Severin, 1949; Al-Wahaibi & Morse, 2003; Boyd & Hoddle, 2006; Son et al., 2009) may play in the biology of sharpshooter leafhoppers. Similar effects of climate have been associated with differences in disease incidence in other systems, such as flea-mediated Yersinia pestis outbreaks in great gerbil populations (Stenseth et al., 2006). Ultimately, determining the extent to which future climate change will alter Pierce’s disease epidemiology requires integrating these effects of climate on vector populations with climate-dependent effects of X. fastidiosa transmission (Daugherty et al., 2009) and infection dynamics within hosts (Feil & Purcell, 2001).

Acknowledgements

The authors thank M. Cooper, Y. Rasmussen, M. Anderson, and especially E. Weber for making available the sharpshooter census data that was used in this work. Thanks also to R. Almeida and S. Purcell for helpful comments and discussion on this topic. This work was funded by the California Department of Food and Agriculture Pierce’s Disease Research Program and USDA NRI grant #2009-05174 to M.D.

Ancillary