Spatiotemporal pattern analysis of Cacao swollen shoot virus in experimental plots in Togo

Authors


E-mail: christian.cilas@cirad.fr

Abstract

In order to quantitatively analyse the spatial pattern of cacao swollen shoot disease, particularly in cases of re-emergence, three experimental plots were installed in a diseased area of cacao cv. Amelonado in Togo. After thorough cleaning and grubbing, the three plots were planted with less susceptible, hybrid plant material. Twenty years after replanting, a survey of healthy, diseased and dead trees was carried out during 2 consecutive years. Data were analysed using Ripley’s functions and join counts analysis. The re-emergence of the disease occurred in patches or foci: following analyses with the two statistical methods, diseased trees and dead trees were found to be clearly aggregated on the three observed plots for the 2 years. The observed progress of the disease was not the same on the three plots and seemed dependent on the disease state of the first year: the higher the attack rate of the first year, the faster the disease progression. The use of less susceptible plants helped keep the land productive for 15 years. In conclusion, uprooting of the first infection focus can extend the life of cacao plots.

Introduction

Although the cacao (cocoa) plant (Theobroma cacao) is a neotropical, small, evergreen tree native to South America (Motamayor et al., 2002), the main area of cultivation is currently in West Africa (Côte d’Ivoire, Ghana, Nigeria and Togo), where it represents nearly 70% of annual world exports.

Cacao swollen shoot disease (CSSD) is caused by a virus occurring only in the main cacao-growing areas of West Africa. Cacao swollen shoot virus (CSSV) is a member of the genus Badnavirus, family Caulimoviridae, and possesses non-enveloped bacilliform particles and a double-stranded DNA genome (Lockhart, 1990; Lot et al., 1991). Natural transmission is by mealybug vectors in the fields. Fourteen species of mealybugs belonging to the Pseudococcidae family are known to transmit the virus, but the main species present in Togo are Formicococcus njalensis, Pseudococcus longispinus, Ps. hargreavesi, Planococcus kenyae, Pl. citri and Ferrisia virgata (Dufour, 1991). These mealybugs live in association with ants; the ants move the mealybugs from one cacao tree to another, ensuring the spread of virus in the plot (Partiot et al., 1978; Castel et al., 1980; Dufour, 1991). The virus can also be transmitted experimentally to susceptible species by grafting (Posnette, 1940).

Characteristic symptoms of the disease are red vein banding of the young leaves and shoot, stem and root swelling (Fig. 1). The virulent isolate of CSSV Agou1 described in Togo (Lot et al., 1991) causes complete defoliation, small pods and death of the tree within 3 years (Crowdy & Posnette, 1947). This isolate was completely sequenced in 1993 (Hagen et al., 1993). Other isolates from Togo and Ghana were then molecularly studied and analysis of spatial repartition of the different isolates in Togo is in progress (Muller & Sackey, 2005; F. Z. Oro, unpublished results). The epidemic cycle depends heavily on the biology of mealybugs, which can be strongly influenced by climatic factors including rainfall (Bigger, 1981).

Figure 1.

 Characteristic symptoms of cacao swollen shoot disease (CSSD): (a) intense redness along the principal veins and between the secondary veins of the young cacao leaf, (b) swelling of the cacao plant stem.

Historically, CSSD appeared for the first time in Ghana in 1922 but was only described in 1936 (Steven, 1936). Its presence was suspected in 1949 (Meiffren, 1949) near the boundary between Togo and Ghana in a small village called Klo-Moyondi. The presence of this virus was confirmed in 1955 around Mount Agou, where the symptoms had been observed in several old cacao plots of the cultivar Amelonado. The disease was then observed between 1967 and 1977 in the south-east of Kloto province (villages of Tové and Nyive) and the south of the Kpelé area (Mount Toutou). At that time, nearly 21% of the cacao plots in Kloto were affected, but CSSV has not been reported in Litimé, the main cocoa-producing zone of the country (Partiot et al., 1978; Castel et al., 1980; Anonymous, 1986; Cilas et al., 1988). It was only towards the end of the 1990s that the disease was reported in Litimé (Cilas et al., 2005), and the isolates from this new area were genetically close to several from Ghana and not related to those from Kloto (Oro et al., 2011). Current low yields can largely be attributed to the disease, with nearly 70% of areas affected (Bekou et al., 2005).

Cacao swollen shoot disease has already caused enormous economic damage in West Africa and, despite numerous eradication campaigns, especially in Ghana, it is still a serious constraint to cocoa production in Ghana, Côte d’Ivoire and Togo. CSSD is recognized now as the most damaging disease affecting cacao in Togo.

The disease progresses exponentially (Partiot & Agbodjan, 1979) with three major phases: the initial phase, the growth phase and the stable phase. The initial phase corresponds to the beginning of viral infection, and the growth phase lasts at least 6 months until the symptoms appear. The stabilization phase is the final stage characterized by the death of the cacao trees.

Cacao swollen shoot disease has long been considered as an archetypal ‘crowd’ disease (Thresh et al., 1988), a term proposed by Vanderplank (1948) for diseases that do not persist in the soil and do not spread far in any considerable amount. There is much evidence to support this view, which is consistent with the behaviour of the mealybug vectors; they are wingless and have low mobility. Mealybugs walk between the interlocking canopy branches of adjacent trees; they are often moved by ants from tree to tree, and occasionally by wind over greater distances. Consequently, CSSD outbreaks tend to occur in discrete patches that expand slowly by radial spread from around the perimeter and much less frequently by jump spread to initiate new outbreaks, including some far from the source (Thresh et al., 1988; Jeger & Thresh, 1993). Based on these disease characteristics, reinfection of healthy replanted trees by CSSD has already been modelled, including specifically to evaluate the effect of a sanitary cordon separating replanting plots from the surrounding diseased area (Jeger & Thresh, 1993). Eradication methods involving uprooting of diseased trees and their neighbours followed by replanting with less susceptible plant material did not give satisfactory results. There are several reasons for this, ranging from the economic and sociopolitical to the scientific. Additionally, the control strategy involving the isolation of new cacao planting by surrounding with an unplanted area or cordons of immune crops has been discussed and recommended (Ollennu et al., 1989; Dzahini-Obiatey et al., 2006) but scarcely evaluated in natural conditions. The spatiotemporal dynamics of the disease in experimental plots are poorly documented and this knowledge is needed to devise a truly effective eradication and replanting programme with the most resistant cultivars currently available.

The objective of this paper was to evaluate the speed of disease spread and describe the spatial development of CSSD at field level. The spatiotemporal pattern of CSSV dynamics was assessed in three fields of cacao trees in Togo and characterized with two complementary statistical methods.

Materials and methods

History of cacao fields

Three experimental cacao plots were used which had been set up by the Institut de Recherche du Café et du Cacao (IRCC) in 1989 (Dufour, 1988) in order to investigate efficient plant barriers in the process of delaying disease progression, both within and between plots. The three cacao plots were located in the Heheti area (Agou-Apegame district, close to Mount Agou). Heheti is situated in Kloto region, which is characterized by humid forest and mountains. The three plots had contiguous boundaries and their areas were 0·8 ha for plot 1 and plot 2, and 0·5 ha for plot 3. The numbers of planted trees were 747, 597 and 365 for plots 1, 2 and 3, respectively.

The three cacao plots were previously occupied by old virus-infected cacao cv. Amelonado trees. These old cacao trees and all the other trees assumed to be virus reservoirs, such as Adansonia digitata, Bombax buonopozense, Ceiba pentandra, Cola chlamydantha, Co. gigantea and Spondias mombin, were uprooted and burned. The young cacao plants planted were hybrids from hand pollination: hybrid 1: (IFC5 × IMC67), hybrid 2: (T12/5 × IMC67) and hybrid 3: (T20/21 × IMC67). These hybrids were planted because of their lower susceptibility (Cilas et al., 1988) relative to cv. Amelonado. The three cacao hybrids were planted in rows, always in the same order from the first to the last row. The distance between rows was 3 m and the distance between trees in the same row was 2·5 m. Each tree was identified by the row number and the tree number on the row.

Each cacao plot was surrounded by a hedge of three rows of coffee trees. Coffee trees served as an effective neutral barrier because they eliminated any direct contact between the cacao trees located on both sides of the common boundary of neighbouring plots. From an epidemiological point of view, this mode of replanting can isolate each plantation and thus substantially reduce the risk of contamination from one plot to another. These three cacao plots were included in a broad area of food crops and were naturally infected by CSSV between the end of 1990s and 2008. Plot 1 was in a better condition than plots 2 and 3 in 2008 because it was properly maintained (K. Wegbe, CRA/F – ITRA, BP 90, Kpalimé, Togo, personal communication); plot 2 was completely abandoned and plot 3 was intermediate.

Data collection and study area

Cacao trees were monitored once a year in the 2 years 2008 and 2009 on the three cacao plots. Disease incidence was considered, each tree being classified as healthy (no symptoms), diseased (symptoms on leaf or stem) or dead/missing.

Progress of the disease over time

Disease dynamics can be observed by transition matrices. A transition matrix is a square matrix describing the probabilities of moving from one state to another. Each row shows the probability (or percentage) change from one state (healthy, diseased, dead, as shown by the row header) to another state (shown by the column header). The number of trees corresponding to the six transitions healthy–healthy, healthy–diseased, healthy–dead, diseased–diseased, diseased–dead, dead–dead between the first and the second year were estimated for the three plots and were used to compute the corresponding proportions. These transition matrices were used to compare the disease progression from one year to another on the three plots.

Spatial statistical analysis of cacao tree health states

Ripley’s function

Ripley’s K(r) function (Ripley, 1977) allows characterization of the spatial structure of a set of points at different scales. This function has been used in the context of several studies to describe forest stands (Ngo Bieng et al., 2006).

Spatial structure analyses were undertaken by the formalism of point process. In each study plot, cacao trees were considered as ‘points’, and took the form [(xi, yi), mi], along with the location of their stem in the studied area Xi = (xi, yi) and their ‘marks’ mi (healthy, diseased, and dead trees). Thus, the patterns of each set of points were studied separately. Trees having the studied state were considered as events, whereas other trees were considered as locations without events.

When considering an arbitrary event in a set of points, the expected mean number of events within a distance r of this event is equal by definition to λK(r), where λ is the average number of events per unit area and K(r) is the K function of Ripley. λ is estimated as the ratio of the total number of events to the total area, and λK(r) is estimated as the average number of events in discs of radius r centered on any event (Stoyan & Penttinen, 2000). In the case of a random pattern of events, the expected mean number of events within a distance r of any event is equal to λπr2, which is the average number of events per unit area multiplied by the surface of a disc of radius r, so that inline image. The transformed function of Ripley was used: inline image (Besag, 1977). The L function simplifies the comparison with the null hypothesis. For a random pattern of events, a graphical plot of L(d) against d is a horizontal line equal to zero. For an aggregated pattern, an excess of short-distance observations would be expected, compared to the random pattern, so that K(r) > πr2 and L(r) > 0. For a regular pattern, L(r) < 0 (Ripley, 1977; Diggle, 1983; Cressie, 1993).

To assess the significance of departures from the hypothesis of complete randomness (CSR) for each set of points, confidence intervals were computed for L(r) using Monte Carlo simulations of the null hypothesis, corresponding to a random distribution of the analysed set of points (Goreaud & Pélissier, 1999; Pélissier & Goreaud, 2001). Thus, the significance of the L(r) function does not depend on density, as the comparison is made between the studied set of points and the same number of points randomly distributed in the same area.

Upper and lower bounds for L(r) were generated at α = 0·01, and for a range of r values at α = 0·01. The radius r varied from 4 m to 50 m by steps of 2 m.

Join count analysis

Join count analysis (Cliff & Ord, 1981) may be used to analyse spatial association for disease incidence data on a lattice. Sites are coded black (B) or white (W) according to the occurrence of a given event at each site of the lattice. If plants with symptoms in a field are coded B and symptomless plants are coded W, two plants may be classified by the type of join linking them: BB, BW or WW. In an array of plants, one can consider joins at different distances in different orientations: along rows, across rows, diagonally, or a combination of these. Then, the number of joins of the specified types in the orientation of interest are counted and compared with a distribution under the null hypothesis of no spatial autocorrelation.

Here, BB join counts were considered for all the pairs of plants separated by r rows and t trees along the row. Simulation was used to obtain the distributions of BB join counts under the null hypothesis of no spatial autocorrelation. The observed numbers of trees with and without symptoms were assigned to locations in the field to obtain 1000 permutations. For each permutation, the BB joins for each (r, t) class were counted and P-values were obtained by ranking the observed number and the 1000 simulated numbers in the class. The P-values were plotted as a 2D correlogram. Two-sided tests at levels of 0·005 and 0·001 were used to detect significant correlations. Only classes corresponding to −10 < < 10 and t < 10 were represented.

Software

Join count analyses were programmed using the statistical software package r, version 2.10.0 (R Development Core Team, 2009), and the results from the Ripley’s function were obtained using the r package ads (Pélissier & Goreaud, 2010).

Results

Progress of the disease over time

In 2008, the proportion of healthy trees in plot 1 was more than twice that in plots 2 and 3 (Fig. 2). In 2009, 37% of the trees were still healthy in plot 1, compared with only 7% in plots 2 and 3.

Figure 2.

 Percentages of dead, diseased and healthy trees in three plots affected by cacao swollen shoot disease (CSSD) in 2 years.

The transition matrix was used to compare changes in different states of health (Table 1). For plot 1, 69% of healthy trees remained healthy, 29% of healthy trees became diseased and 2% of healthy trees died. Of diseased trees, 84% were still diseased and 16% died. For plots 2 and 3, fewer of the healthy trees were still healthy in the second year: 33% for plot 2 and 37% for plot 3; over a third of the healthy trees became diseased and a sizeable proportion died. Of the diseased trees, 62% and 76% remained diseased in plots 2 and 3, respectively, while 38% and 24%, respectively, died.

Table 1.   Numbers of cacao trees (percentages in brackets) corresponding to the six transitions between the first and the second year in three plots
 20082009
HealthyDiseasedDeadTotal
Plot 1Healthy276 (69·5)114 (28·7)7 (1·8)397
Diseased 84 (84·0)16 (16·0)100
Dead  250 (100)250
Total276198273747
Plot 2Healthy42 (32·6)44 (34·1)43 (33·3)129
Diseased 100 (61·7)62 (38·3)162
Dead  306 (100)306
Total42144411597
Plot 3Healthy25 (36·9)36 (44·4)20 (22·2)81
Diseased 65 (75·6)21 (24·4)86
Dead  198 (100)198
Total25101239365

According to the transition matrices, the three plots had different profiles. The rate of disease progression was lower for plot 1 and the proportion of healthy trees in 2008 that were dead in 2009 was very low or negligible (2%). Plot 1 was the least diseased of the three plots. On the other two plots (plots 2 and 3) the percentages of state changes were higher, which means epidemics were faster. The proportions of healthy trees that were directly killed were sizeable, 33% and 22% for plots 2 and 3, respectively.

Plots 2 and 3 were more affected by the disease than plot 1 in the first year of observation and, in plots 2 and 3, healthy trees were more likely to become diseased and more likely to die than in plot 1.

Spatial statistical analysis of cacao tree health status

Maps of the spatial distribution of the disease suggested aggregation of diseased trees (Fig. 3); statistical methods were used for further analyses.

Figure 3.

 Maps of cases of cacao swollen shoot disease (CSSD) in three plots and 2 years: plot 1, 2008 (a) and 2009 (b); plot 2, 2008 (c) and 2009 (d); and plot 3, 2008 (e) and 2009 (f). Each rectangle represents a tree with its disease status: healthy (white), diseased (grey) or dead (black).

Ripley’s function

Ripley’s transformed functions were always positive for the health states healthy, diseased and dead, regardless of experimental plot in 2008 (Fig. 4) and 2009 (Fig. 5). This result indicates that the different health states show an aggregated pattern for all the studied distances (<50 m).

Figure 4.

 Ripley’s function L(r) for the three cacao plots for 2008. (a,d,g) Healthy trees, (b,e,h) diseased trees and (c,f,i) dead trees. Plot 1 (a–c); plot 2 (d–f); plot 3 (g–i). Solid line, observed values; dashed lines, 0·99 confidence envelope under the null hypothesis of spatial randomness, obtained by simulation.

Figure 5.

 Ripley’s function L(r) for the three cacao plots for 2009. (a,d,g) Healthy trees, (b,e,h) diseased trees and (c,f,i) dead trees. Plot 1 (a–c); plot 2 (d–f); plot 3 (g–i). Solid line, observed values; dashed lines, 0·99 confidence envelope under the null hypothesis of spatial randomness, obtained by simulation.

Join count analysis

For plot 1 in 2008 (Fig. 6a), positive autocorrelation was detected at a level of 0·5 × 10−3 perpendicularly to the rows up to a distance of two rows (6 m), along the row up to a distance of two trees (5 m), and diagonally up to a distance of one row and one tree (4·75 m). These distances represent an isotropic cluster of about 5 m radius. In 2009 (Fig. 6b), the correlation that was perpendicular to the rows was significant only at a distance of one row (3 m), and the other correlations were unchanged or slightly lower. For plot 2, autocorrelation at a level of 0·5 × 10−3 was detected only along the rows up to a distance of four trees (10 m) in 2008 (Fig. 6c) and one tree (2·5 m) in 2009 (Fig. 6d). For plot 3 in 2008 (Fig. 6e), significant positive autocorrelations at levels of 0·5 × 10−3 to 2·5 × 10−3 at distances from 2·5 to 30 m were clustered and oriented at 135° relative to the rows. In 2009 (Fig. 6f), no significant autocorrelation at a level of 0·5 × 10−3 was detected. For plot 2 in 2008, significant negative autocorrelations at 0·5 × 10−3 were detected perpendicularly to the rows and diagonally at distances from four to eight rows (12–24 m). Significant negative autocorrelations at 0·5 × 10−3 were also detected for plot 1 (2008 and 2009) and plot 3 (2008) at distances of approximately 30 m.

Figure 6.

 Spatial dependence between cases of cacao swollen shoot disease (CSSD) in three plots in 2 years: plot 1, 2008 (a) and 2009 (b); plot 2, 2008 (c) and 2009 (d); and plot 3, 2008 (e) and 2009 (f). Each rectangle represents a distance class between pairs of trees. The shade of grey indicates the P-value of the join count analysis obtained from 1000 simulations.

Discussion

The comparison of the three plots’ dynamics showed that plots 2 and 3 were different from plot 1, although the three plots were in the same environment and planted with the same plant material. However, the three plots had not received the same level of maintenance: plot 1 was better maintained and had greater numbers of healthy trees than plots 2 and 3; plots 2 and 3 had more dead trees in 2008 and 2009 than plot 1.

In all three plots, healthy trees could die within a year, but they did so in different proportions related to the level of maintenance. In plot 1, only 2% of the healthy trees died within 1 year, compared with 33% and 22% for plots 2 and 3, respectively. Either the progress of the epidemics was actually higher for plots 2 and 3 or/and some trees died for reasons other than CSSV in these two plots. With regard to other changes of state, the proportions of healthy trees becoming diseased and the proportions of diseased trees becoming dead were also lower for plot 1. Thus, the progress of epidemics was lower in plot 1 than in the other plots.

These differences seem to be related to the lack of regular maintenance and also to a lack of treatment against insects in plots 2 and 3. The absence of regular maintenance of cacao trees and the absence of insecticide treatments (K. Wegbe, CRA/F – ITRA, BP 90, Kpalimé, Togo, personal communication) led to high parasite pressure, in particular populations of mealybugs carrying CSSV.

The positive Ripley function, regardless of health state and the year of observation, indicated an aggregation of the different health states of cacao trees. This aggregation of health states suggests that the disease is distributed in patches of infection. The spatial repartition in patches could have two possible causes: contamination in the plots can occur little by little, or by arrival of aggregated inocula. The gradual spread of CSSV coincides with the biological characteristics of the mealybug vector: mealybugs have low mobility and are, in most cases, moved by ants from tree to tree.

Join count analysis detected positive spatial autocorrelations between neighbouring trees for each of the three studied plots. Significant positive correlations that were grouped around the central point of the 2D correlogram and that constituted the core cluster (Gray et al., 1986) represented correlation between trees separated by 2·5–30 m. The size of the core cluster decreased as the prevalence of the disease in the plot increased. This property can be explained merely by the decreasing power of the test as the proportion of trees with symptoms approaches 0% or 100%. For plot 3 in 2009, where the prevalence was 93%, the core cluster completely disappeared. The size of the core cluster is not, therefore, a characteristic parameter of the epidemic, but can be used to compare plots with identical proportions of trees with symptoms using tests at the same probability level. The shape of the core cluster may also give valuable information on the epidemiological process. However, the three plots had different core cluster shapes. The core cluster of plot 1 was approximately circular, indicating an isotropic aggregation pattern. Conversely, the core cluster of plot 2 was elongated and parallel to the rows, indicating an anisotropic aggregation pattern along the rows. The core cluster of plot 3 in 2008 was larger than the core clusters of plots 1 and 2, and indicated anisotropy in a direction 135° from the direction of the rows. This can be explained by the spatial pattern of plot 3, where dead trees were concentrated in the upper right part of the map (Fig. 3).

The clusters of negative autocorrelations that were not adjacent to the central point and that were detected for the three plots in 2008 and for plot 1 in 2009 indicate the intraplot heterogeneity of incidence. The distance between the central point and these clusters represents the distance that separates areas with different incidences. In 2008, this distance was between 12 m for plot 2 and approximately 30 m for plots 1 and 3.

The three plots, although initially in the same physical environment, had very different dynamics; plots 2 and 3 were more infested than plot 1. This can be explained in two ways: the coffee barrier effect and the level of maintenance. The coffee barriers certainly played an important role in the isolation of plot 1 from plots 2 and 3. Lack of maintenance on plots 2 and 3 relative to plot 1 played a similarly important role.

The methods of Ripley and join count analysis showed that the health states of cacao trees were aggregated. This indicates that disease spread from patches of infection. Indeed, these plots installed in affected areas produced cacao for around 15 years before the disease arrived. Good maintenance could increase this duration of cacao production, and assist especially in the control of mealybugs.

With the perspectives gained from this work, it could be possible to separate the barrier effect from the maintenance effect by specific experiments; plots with a barrier and without maintenance could be compared to plots with a barrier and with maintenance. As the disease spreads little by little in plots, it would be particularly interesting to develop agroforestry systems for new cacao plots, to put barriers of CSSV-immune crops between the plots, as was previously recommended (Ollennu et al., 1989), and thus slow down the disease. Also, because of the nature of the spread, disease control could be improved by early removal of trees in infected patches when the plot is contaminated (Thresh & Owusu, 1986; Sosnowski et al., 2009).

Acknowledgements

The authors acknowledge Institut Togolais de la Recherche Agronomique (ITRA) for technical support and Sarah Bharath for reviewing the English version.

Ancillary