You are on page 1of 32

1 Target Journals: Journal of Plant Pathology, PLOS One (I recommend this journal),

2 Phytopathology Research

3 Determinants of infection of BBTD recovered gardens in a small holder environment – Options for

4 ecologically based management in small holder systems.

6 Aman Omondi Bonaventure1, Dato Geofroy2, Marius Aza3, Innocent Nduwimana3 Abraham

7 Shemezimana4, Emile Nitunga4, Mintodê Nicodème Atchadé2,5 Niyongere Celestin4, Staver

8 Charles1.

10 1. The Alliance of Bioversity International and CIAT, co/ IITA Building, Regina Mundi Road,

11 Bujumbura, Burundi.

12 2. Alliance of Bioversity International and CIAT, c/o IPGRI Building, IITA Campus, Abomey

13 Calavi, Benin.

14 3. University of Abomey-Calavi/International Chair in Mathematical Physics and Applications (ICMPA:

15 UNESCO-Chair), 072 BP 50, Cotonou, Republic of Benin

16 4. Institute of Agronomic Sciences, (ISABU) Burundi, Bioversity France, UAC Benin.

17 5. Faculty of Agronomic Sciences, University of Abomey, Benin

18 6. National Higher School of Mathematics Genius and Modelization, National University of

19 Sciences, Technologies, Engineering and Mathematics, Abomey, Republic of Benin

20

21 Corresponding author: b.a.omondi@cgiar.org

22

23 Abstract
1
24 Spatial factors associated with disease prevalence are important in understanding practical options

25 for integrated pest and disease management. Such packages include the use of ecosystems services

26 in landscape design towards disease management. In this study, landscape diversity, crop

27 production systems and banana bunchy top disease (BBTD) control activities were assessed around

28 re-established banana gardens and correlations with the disease prevalence investigated. The

29 gardens assessed were located in disease endemic landscapes and three years old since re-

30 establishment with clean seed following BBTD infestation. Of the factors assessed the density of

31 banana within 60 meters, symptomatic BBTD prevalence, aphids density within the gardens,

32 number of varieties grown and frequency of alternative seed sources were positively correlated to

33 BBTD infection observed. In contrast, the presence of hedges, altitude, rogueing frequency and the

34 destruction of bananas in and around the replanted fields were negatively associated with BBTD

35 prevalence. The level of infection differed in the four villages. To reveal potential IPDM packages,

36 an analysis of combinations of factors associated with BBTD prevalence showed that aphid

37 prevalence, rogueing frequency banana density and seed sources appeared in half the models

38 suggesting their importance in BBTD control. Similarly, plot size, altitude and gender of the farmer

39 were highly represented underlining context specificity and potential gender gaps in the

40 performance of these packages. These findings support the general thrust of the community

41 approaches in BBTD management and give some dimension of the size of the buffer that could be

42 effective among smallholders in mixed systems. The context specificity and role of rogueing and

43 hedges in and hence crop mixtures suggest the greater host interaction and could be tested for

44 species mixtures.

45

46 Key words: Bunchy top disease, small holder systems, cropping management, spatial modelling

47

2
48 Introduction

49 Invasive species are a leading threat to agricultural production and stability of ecological

50 functioning (Meyerson and Mooney, 2007). They are increasingly common in agriculture today as

51 a result of human and environmental factors (Hulme 2009; Global Invasive Species Database,

52 2018). The successful establishment and spread of an invasive species require the presence and

53 connectivity of suitable habitats. In the case of pests and disease, these include the presence of

54 susceptible hosts and the spatial and behavioural the connectivity between potential hosts in the

55 landscape; and also the efficiency of the dispersal mechanisms that support their dispersal.

56 Understanding landscape connectivity of host vegetation is important in predicting the spread of

57 pests or diseases. On a local scale these factors are influenced by the activities that reduce of

58 inoculum and its infection spread such as cropping matrices and plot scale management. Modelling

59 these and agro-environmental factors might provide important feedback for the performance of

60 specific steps and conditions for the control of these diseases.

61 Banana and plantains (Musa sp) is an important fruit and cash crop worldwide (FAO,

62 2019). It is also an important staple food and an economic security crop to many producers in

63 different parts of the world. In rural small-scale perennial systems, economic and food security

64 crop for smallholder farmers (Marchal, 1990) with potential in promoting equitable resource

65 distribution and economic empowerment in rural areas. It provides benefits to different household

66 members as the fruit and plant parts are used in different ways (Akinyemi et al., 2010; Lokossou et

67 al., 2012; Mulugo et al., 2020). Banana varieties are therefore selected for the major uses and

68 grouped as, beer, dessert, cooking and multi-purpose banana varieties and plantain (highly starchy

69 cooking types). Despite their importance, banana yields in Africa are way below most of the world,

70 due to a combination of field management, biotic and abiotic stress. In Burundi, for example,

71 Fusarium wilt decimated the highly susceptible Kayinja (ABB), previously the dominant beer

72 banana type (Simbare et al., 2020). Its replacement with Yangambi Km5 (AAA) is closely

3
73 associated with the proliferation of the banana bunchy top disease. Other biotic stress factors in the

74 region include Banana Xanthomonas Wilt (BXW), banana weevils and nematodes (Jacobsen et al.,

75 2019).

76 The banana bunchy top disease (BBTD) threatens banana production worldwide. BBTD is

77 now found in most of the major banana growing regions in the world (Staiton et al., 2015). In

78 Africa, it is now reported in at least 18 countries (Ocimati et al., 2021; TZ). BBTD is estimated to

79 affect about 6 to 12 million smallholders (Kumar et al., 2015; Niyongere et al., 2015, Oben et al.,

80 2009). In Malawi and DRC, banana and plantain production has been significantly reduced with

81 significant economic and livelihood impact on farmers (Soko et al., 2009). BBTD is spread through

82 planting infected seed and aphids, and leads to a gradual loss of production and simultaneously

83 decline in the availability of clean seed to rehabilitate lost gardens. Over time, a total production

84 loss is associated with a seed system collapse. Invasion of the continent also threatens two centres

85 of secondary diversity of Musa spp: the East African highlands, with her ca 100 highland AAA-EA

86 bananas and the centre of plantain diversity in West Africa. Such a genetic loss may threaten the

87 long-term sources of genetic materials for banana improvement, against emergent biotic and abiotic

88 challenges. BBTD therefore threatens global banana genetic diversity. Small holder production

89 systems are particularly vulnerable because of the dominance of informal seed sourcing; and the

90 perennial nature of the production system (Simbare et al., 2020). Controlling BBTD in smallholder

91 systems requires knowledge of the epidemiological factors that may promote the disease spread.

92 Banana recovery and BBTD containment strategy has been implemented in Burundi,

93 alongside other eight countries, since 2013, using a community-based collaborative learning and

94 action (Omondi et al., 2020). This approach, based on the knowledge of BBTV epidemiology

95 involves area wide destruction of infected plants, enforcement of a banana free fallow to eliminate

96 inoculum sources and viruliferous vectors, multiplication of disease-free planting material,

97 replanting in well isolated fields and collaborative monitoring of the replanted gardens (Omondi et

4
98 al., 2020). Implementing this strategy requires significant collaboration at landscape level

99 especially in crop monitoring inoculum reduction and seed supply systems. The implementation of

100 the recovery protocol in Burundi was under high population density and existing social and

101 production system. About 130 farmers had been involved in the recovery, representing two

102 community nurseries and four villages at different altitude levels. As all farmers started about the

103 same time with receipt of a number of planting materials following prescribed preparation, a single

104 shot assessment of their relative performance against ecological indicators would inform BBTD

105 spread and epidemiology in small holder banana dominated landscapes. These landscapes may be

106 similar to large plantations in banana canopy continuity, but are more complex, involving multiple

107 varieties, species and farming methods.

108 Understanding reinfection risk of recovered banana gardens in BBTD affected areas is an

109 important component of recovery. A stable minimal risk of reinfection also provides an opportunity

110 for gradual expansion of the recovered area taking advantage of the same buffer zone, and supports

111 the broadening of seed production and exchange within recovered areas. This work sought to

112 evaluate the factors associated with BBTD reinfection of replanted bananas towards identifying

113 key intervention weak links for improvement. In this study, we evaluated the level of BBTD

114 occurrence in farmer fields previously replanted with BBTV-free planting materials in 2013 and

115 2014. This allowed for testing BBTD prevalence determinants, on farms in a comparable

116 background risk with a uniform staring point, hence minimizing potential temporal determinants of

117 the infection. Specifically, this would test the importance of epidemiological and environmental

118 variables associated with BBTD spread in this region, hence a basis of understanding the aspects of

119 the control. The study focused on the landscape characteristics, cropping and vegetation patterns,

120 existing BBTD prevalence and indicators of BBTV management and seed system functioning in

121 predicting the observed BBTD infection rates.

122

5
123

124 Materials and methods

125 A field study was done on 123 farms in four communes Rugombo, Rusagara, Munyika and Mugina

126 communes in Cibitoke Province, Burundi. Cibitoke province is a key banana production and

127 BBTD endemic zone, lying in in the frontier between Burundi, Rwanda and the Democratic

128 Republic of Congo (Figure 1). It covers an altitude between 800 to 1300m asl, including the Rusizi

129 Valley shared between Rwanda, Burundi and the DRC and endemic to the BBTD for the last three

130 decades. Pilot sites had been created in these communes to evaluate community approaches to

131 BBTD recovery in the country, about 2014 (Lepoint 2-16/ 2014; Omondi et al., 2020, 2022). The

132 farms assessed in this study had been planted with BBTD -free planting materials from a tissue

133 culture plantlet hardening nursery located in the village. Although farmers had been guided and

134 encouraged to follow BBTD risk reduction practices and cultural management protocols during this

135 period, actual implementation depended on their means and farm management priorities. Thus,

136 they shared a known starting point for comparing the environmental and management risks of

137 BBTD reinfection in the area, under a matrix of combinations of management practices,

138 preferences and social and ecological realities. They are therefore useful in assessing BBTD risk

139 factors at landscape scale in this area, without the confounding random factor of first infection in

140 an area.

141

142 Data collection

143 A mixture direct observation by field measurements and quantitative surveys were used to collect

144 data. For each plot, geographical details such as altitude and GPS coordinates were taken. A sketch

145 map of each target garden was made to orient the characterization and data taking of the field and

6
146 its effective buffer of 90m (Allen, 1978) (Figure 1). Each plot was assessed for crop density and

147 diversity within replanted gardens, banana cultivar diversity and BBTD occurrence. The proportion

148 of plants with symptomatic BBTD, banana aphids and aphid colony types were assessed by visual

149 count, as the response variable (Omondi et al., 2020). Plot owners were interviewed to collect

150 information on the field history, management during replanting, source of planting material,

151 diversity of cultivars, crop ownership, management and contact with neighbours. The list of the

152 variables taken is summarized in Table 1.

153 Each of the gardens in the commune replanted with TC bananas since the beginning of the

154 project was considered a single focus plot. Data was collected in the focus plot and a 90m buffer

155 around it (Allen, 1978) divided into to three bands of 30m each. The host distribution variables

156 considered included the density of banana, the level of BBTD, cropping plan and vegetation

157 mixtures. Location and social factors influencing landscape-scale disease prevalence included the

158 Altitude and physio-geographic factors; gender of owner, size of farm. The cropping and seed

159 supply systems; varietal diversity and cropping mixtures were evaluated. Finally, the BBTD

160 control strategy was examined including banana free fallow; buffer management, seed sourcing;

161 rogueing and crop management by farmer declaration. All infection observed was considered as a

162 resultant of the risk factors, since the starting point was known and included indexed BBTV free

163 planting materials.

164 BBTV is systemic, thus, the unit of data taking was a mat, comprised of all shoots

165 connected at the rhizome level. A sample of at least 10 mats from different locations in the garden

166 were selected randomly and examined for BBTD symptoms, per section of the plot. Three

167 youngest fully opened leaves per selected shoot were inspected for BBTD symptoms on the leaf

168 lamina and petiole and midrib, according to the existing protocols (Niyongere et al., 2013; Omondi

169 et al., 2020). Mats were marked as diseased if any shoot was symptomatic. The number of diseased

170 plants was assumed to closely represent that of BBTV infected plants.

7
171

172

173 Data exploration

174 The dataset contained missing values which were imputed using the multiple imputation by

175 chained equations (MICE) method from package MICE for categorical variables and median for

176 continuous variables to enable analysis (Azur et al., 2011). The dependant variable of interest was

177 the reported number of mats showing BBTD symptoms per plot, each representing an infection

178 event. Accordingly, count regression (Poisson and negative binomial) models were the most

179 appropriate to assess the statistically significant factors associated with banana’s BBTV infection

180 risk. To select the best count model, the mean and variance of the number of diseased banana

181 plants were used to test the overdispersion of the count variable. This was followed by an analysis

182 of the number of diseased banana plants per plot.

183

184 The response variable showed a positive skewed distribution (skewness=1.6) and extremely

185 different values for the unconditional mean (13.5) and variance (163.6), which had a nonlinear

186 relationship, thus suggesting overdispersion (Figure 2). The log-likelihood ratio test for

187 significance of overdispersion was computed as follows:

188 log −likelyhood ratio=– 2 ×(log−likelihood of Poisson regression – log−likelihood of negative binomial regress

189 The value of the likelihood-ratio test showed that the data was significantly over-dispersed χ2 =

190 759.177, p-value = 0.0001, thus favoured the negative binomial regression model (NBRM), a

191 generalisation of the Poisson model. The NBRM permits sufficient flexibility with only two

192 parameters (the shape parameter and the mean). It subsumes the Poisson distribution while also

193 allowing for overdispersion, where the variance may me greater than the mean.
8
194

195 Choosing Model Predictors

196 To choose the predictors to include in the final model, we used backwards elimination.

197 1. The model containing all the chosen predictors was fitted

198 2. All coefficients from this model were saved for comparison with products of the later steps

199 3. Remove in one go all predictors whose p-value was above the preset threshold ( p=0.1) were

200 removed.

201 4. The coefficients for the remaining predictors were compared with their coefficients in the

202 original model in step 2 (above), till the least possible change was observed. This model was

203 retained.

204

205 The unconditional mean of our outcome variable was much lower than its variance, so, we

206 used a negative binomial model to adjust the parameters (Bruin, 2006). This allowed the retention

207 of the general mean structure of the Poisson regression, with an extra parameter to model the over-

208 dispersion (Bruin, 2006). The negative binomial regression has been used to model the negative

209 binomial distribution arising as a distribution of the number of failures ( X ) before the r th success in

210 independent trials, with success probability p in each trial (consequently, r ≥ 0 and 0 ≤ p ≤ 1)

211 (Lindén and Mäntyniemi 2011).

212 2. Negative binomial regression


213 The negative binomial distribution has been parameterized in several different ways in the

214 statistical and applied literature. Perhaps the most common way to parameterize is to see the

215 negative binomial distribution arising as a distribution of the number of failures (X) before the rth

216 success in independent trials, with success probability p in each trial (consequently, r ≥ 0 and

9
217 0 ≤ p ≤ 1) (Lindén and Mäntyniemi 2011). In such a case the probability mass function can be

218 expressed as

Γ ( x+ r ) r
219 P ( X=x|r , p )= p ( 1− p ) x (1)
x ! Γ (r )

220

221 and the random variable X has the expectation (theoretical mean) μ=r (1 – p)/ p and variance

222 σ 2=r (1 – p)/ p 2. Corresponding to this parameterization, parameters r and p can be written in terms

223 of the mean and the variance as

μ2
224 r= (2)
σ 2−μ

μ
225 p=1− 2 (3)
σ

226 We use Equations. 2 and 3 as a starting point for a more flexible parameterization of the negative

227 binomial distribution that can be used to represent various kinds of over-dispersed Poisson

228 processes. By writing the variance as a quadratic function of the mean,

229 V (μ)❑=ωμ +θ μ2 (4)

230 diverse mean-variance relationships were obtained using the parameters ω and θ , hereafter called

231 overdispersion parameters. This process was defined as long as σ 2> μ , or phrased in terms of the

232 overdispersion parameters, when ω+ θμ>1. With any valid combination of expectation E( X) and

233 overdispersion parameters, the parameters r and p of the negative binomial distribution can be

234 derived as

μ
235 r= (5)
ω−1+θμ

1
236 p= (6)
ω+ θμ

10
237 and then be substituted into Eq. 1 to obtain the relevant likelihood. A linear mean-variance

238 relationship, corresponding to the assumption of quasi-Poisson regression, is obtained by fixing ω

239 to zero. Similarly, fixing ω to 1 gives the quadratic mean-variance relationship of the classic

240 negative binomial regression. The negative binomial distribution can also be seen as an explicit

241 over-dispersed Poisson process, where the Poisson intensity is drawn from a gamma distribution

242 (Gelman et al. 2004).

243 Assessment of best combinations of descriptors.

244 Disease management is typically achieved by a combination of crop management approaches

245 depending on farmer knowledge and access to resources. To simulate a hypothetical management

246 regime, we next selected combination of variables explaining the response (BBTD level of

247 infection by comparing all models subsets using Akaike Information Criterion ( AIC )).

248

249 AIC is given by:

250 AIC=−2 log L ( θ ) +2 k

251 where L (θ ) is the likelihood of the data given each fitted model,

252 k the number of estimable parameters in each model.

253 AIC compares multiple competing models all at once to determine the best approximating model.

254 Model fits were then ranked according to their AIC values, the model with the lowest AIC value

255 being considered the best. Models in which the difference in AIC relative to AIC min is < 2 can be

256 considered also to have substantial support (Burham,2002, Burnham and Anderson,1998) and were

257 retained.

258 For each model i, we calculated the variation in AIC relative to the best-ranking (lowest AIC )

259 model (∆ AIC= AICi −AIC min ); models with ∆ AIC< 2 are generally taken to be substantially

11
260 supported by the data. This was followed by the calculation of Akaike weights ( w i) which are

261 normalized models likelihoods, expressed as

262 w i=exp ( 2 )
−∆ AIC i
(
/ ∑ exp
−∆ AIC
2 )
263 and can be considered as analogous to the probability that a given model is the best approximating

264 model. Management and locational variables were not separated in this analysis.

12
265 Results
266 Table 2 shows the coefficients for the retained predictors with their coefficients from the original

267 model. In total thirteen variables were retained in the final model after backwards elimination

268 (Table 4). The output of the negative binomial model regression shows seven variables

269 statistically significant at 5% threshold: the size of plot, mean of banana density within 30m,

270 hedge presence, number of varieties, type of planting material used, frequency of scouting

271 banana infected, and location; and one variable (farmer declared roguing frequency) at 10%

272 threshold.

273 The Mean of Banana density in 30m has an odds ratio of 108.47 (

274 108.47/1+108.47=0.99). A unit increase in the Mean of banana density in 30m leads 99%

275 chance of increase in the number of infected mats, with a 95% CI (Confidence Interval) of 0.21

276 to 9.16. Plots reporting roguing were half as likely to report banana mats infected compared to

277 those that did not. The probability of finding BBTD in for gardens with a hedge is 41% of those

278 without a hedge. However, the probability of having banana mats infected for farmers who used

279 three varieties compared with those having one variety is 63% higher compared to those with

280 only one variety. Using tissue culture as planting material alone has a 12% of probability of

281 having banana mats infected compare to using suckers alone. The BBTD prevalence was also

282 associated with altitude and hence communes studied (Figure 2). The higher the altitude, the

283 lower was the prevalence of the BBTD observed. Thus, gardens in Munyika had a 35%

284 probability of getting banana mats infected compared to those in Gitebe.

285

286 Most used BBTD management technics

287 A summary of the percentage of farmers using different BBTD management methods is

288 presented in Figure 3. Roguing, Intercropping and Scouting for BBTD were used by a majority

289 of the farmers, while about 25% and 36% of the gardens were established using clean seed alone

290 (tissue culture material) or had a hedge around them. Crop rotation and the enforcement of

13
291 banana free period were only used by a minority. Many farmers used different combination of

292 these practices, with only 2.1% using all seven (Figure 4).

293

294

295

296

297

298 Discussion

299 We investigated the field management and landscape factors associated with BBTD

300 infection rates in a BBTD endemic zone in Eastern Africa. The field conditions within and

301 around smallholder banana gardens, location and cropping systems, vector abundance, disease

302 prevalence, seed systems, and crop management measures taken were assessed. All gardens were

303 established initially using clean planting material, in a BBTD endemic zone; but followed

304 disease and crop management options determined by the farmers themselves. Beginning from a

305 known starting point with comparable production environments, enabled us to assess an

306 ecologically based management regime This analysis provides a first landscape scale assessment

307 of farmer practices and their effectiveness in managing BBTD; and the potential of their

308 adaptation. This assessment provides best-bet approaches to sustaining banana production in

309 BBTD endemic zones.

310 Landscape diversity and management were associated with BBTD control in recovered

311 fields. Banana density around the target field, vector density, number of varieties grown,

312 vegetation barriers, seed sources and management frequency were significantly associated with

313 lower BBTD prevalence. Purely locational factors such as altitude (or village/location) size of

314 holding and gender of garden owner were also associated with differences, implying context

14
315 specificity in in BBTD cases reported. Curiously, the actual BBTD prevalence in the

316 surrounding gardens and distance of the nearest observed infected plant from the target garden

317 were not significantly correlated with BBTD prevalence. These results underline the importance

318 of landscape context hence community-scale operations, in disease management.

319 The importance of clean seed sourcing in BBTD management in small holder systems

320 was evident in this study. Several additional cultivars were found in the replanted fields, yet only

321 five varieties had been given out as indexed tissue culture planting materials, with each farmer

322 receiving a maximum of three. Farmers who sourced additional planting materials were more

323 likely to have greater BBTD levels, suggesting the greater risk of informal seed sourcing in

324 reintroducing BBTD to recovered fields. The informal sourcing of seed is particularly vulnerable

325 to seed-borne pathogens such as BBTD yet are important sources of desired local cultivars

326 (Simbare et al., 2020). Sourcing of seed is motivated by the need for additional cultivars and

327 involves an evaluation of farmers needs and potential associated risks, not limited to disease

328 risks (Nduwimana et al., 2022). Exigence on formal seed systems in BBTD affected areas

329 present both operational difficulties (inadequate seed for desired varieties) and a significant

330 gender and social implementation gap for reaching a majority of the farmers (McEwan et al.,

331 2021). Managing BBTD and practical clean seed systems in farmer conditions require adaptable

332 innovations taking advantage of both demand and matching seed quality classes with distribution

333 limits to deliver low risk seed.

334 Landscape level complexity provided a relevant opportunity for the management of

335 BBTD at a topographical scale. The banana canopy connectivity and density around a recovered

336 plot are associated with increased BBTD risk in the target plot. These variables could influence

337 the vector aphid in BBTD spread at garden level, but could also present opportunities at a

338 topographical scale (Kebede et al., 2018). Insect pest and vector behavioral manipulation through

339 intervening non-host species is a powerful ecological management approach (cite push pull).

340 Presently, banana free periods and adequate buffer zones have been most considered (Allen,

15
341 1987). We observe that hedges around banana gardens are associated with lower levels of

342 infection suggesting that aphid immigration from external inoculum sources played a role in

343 reinfection departing from the assumptions of Smith et al., (1998). Thus, mixed cropping

344 canopies could also support natural enemies. The banana aphid has a restricted host spectrum

345 and is directly affected by host distribution and BBTV (Watanabe et al., 2013, Wessels et al.,

346 2019). Planning cropping matrixes and landscape level would fit well with community

347 approached to BBTD management especially once a choice of the most effective hedge species

348 is identified. The removal of bananas within and around the target plot, and the existence of

349 vegetation breaks (border rows) were negatively associated with infection rate in the target plot.

350 However other factors previously considered useful including intercropping matrices, size of plot

351 and manager gender were not found to affect infection rate significantly. The role of infected

352 asymptomatic banana mats is not clear from our results, and could influence the model if

353 accommodated (Cunniffe et al., 2015). The relationship between infection, infectiveness and

354 symptoms has been studied in BBTD, mostly in Cavendish systems (Allen, 1978). A delay

355 between first symptom appearance and infectivity was found in Cavendish and could also be

356 variable among banana cultivars (cf Niyongere et al., 2012, Ngatat et al., 2022; Chabi 2020). Our

357 study does not make a distinction between infection, infectivity and symptom appearance, and

358 disease pressure in the fields studied, if rogueing were regular. Since disease management is not

359 uniform among the farm households studied realistic estimates were difficult to achieve. Yet, all

360 asymptomatic plants samples appeared negative to BBTD when tested using ELISA. Similarly,

361 disease progress has been shown to vary between cultivars (Ngatat et al., 2017), seasons

362 (Niyongere et al., 2011) and could vary by altitude, and age of crop (Lelerc et al., 2015).

363 Roguing has been an effective approach in BBTD control if done persistently and early

364 (Omondi et al., 2020). Our study did not test the role of persistent rogueing is not considered as

365 this was not measured directly for the static time point considered. Indeed, the observation of

366 BBTD as a response variable could confound both infection rate and response to continuous

16
367 disease management. The closest estimate of rogueing would either be estimated from the

368 number of plants missing since planting which is known; or from declaration from the farmers

369 (rogueing intensity, or mortality figures). Rogueing would also have a variable relationship with

370 infection levels. A positive correlation would be expected under greater reinfection pressure

371 since higher reinfection would expectedly result in greater rogueing. Higher rogueing intensity is

372 however meant to reduce inoculum source hence minimize the resultant infection rate. This study

373 could not test this aspect, which needs time series reinfection studies to estimate.

374 The potential roles of alternative hosts of BBTD and banana aphid were not considered.

375 While congeneric species of the banana aphid (Pentalonia caladii) exists in Burundi, and its

376 hosts common in banana landscapes (Wessels et al., 2019); its role in Banana epidemiology in

377 the field has not been demonstrated (Watanabe et al., 2013). Host susceptibility and symptom

378 expression, could also influence disease dynamics. Although all banana cultivars are to date

379 assumed susceptible, varieties grown in the continent accumulate BBTV titre and express

380 symptoms at different rates (Niyongere et al., 2011; Ngatat et al., 2017; 2022). The infestation

381 and population buildup of vector aphids in different varieties of banana have also been

382 hypothesized to be different (Niyongere et al., 2011). These factors may also contribute to the

383 efficiency of an infected plant as a source of infection. Similarly, host-virus-vector interactions

384 might determine the rate of infection progress. Virus infection sometimes affects vector biology

385 and could influence virus transmission by modifying vector reproductive rate, host attractiveness

386 to incoming vectors or suitability or vector behaviour (Ingwell et al., 2012). These effects could

387 be a direct virus-host interaction or mediated through virus host interaction mechanisms via food

388 suitability of the host. Understanding these interactions might help explain the factors governing

389 epidemic appearance of BBTD. Similarly, vector dispersal and its response to weather factors

390 directly would influence the application of models in predicting BBTD spread. This study

391 assumed similar weather factors for the regions that were in close proximity. Other factors worth

392 investigating include plot sizes and local seed linkages, which have an economic and gender

17
393 component via resource access and social networks (Nkengla et al., 2020; Nduwimana et al.,

394 2022).

395 Overall, our results show the comparative importance of landscape characteristics in

396 BBTD progress. To our knowledge this is the first attempt to model BBTD spread in mixed

397 cropping systems among small holder communities. The studies also identify contextual

398 considerations necessary in studying BBTD epidemiology and control in Africa. Most

399 importantly, the role of landscape complexity in BBTD progress, has been underlined. Further

400 studies will be necessary to put these studies in a dynamic time-scale study. For a gender

401 inclusive assessment of the technologies identified here, an evaluation of the contexts of their

402 application and utility is recommended. This would also include environmental contexts

403 promoting or overriding the benefits of components and packages in an integrated management

404 of BBTD.

405 Acknowledgements:

406 We are grateful to the 132 farmer partners of the Banana Bunchy Top Disease Management

407 Project in Cibitoke, Province, Burundi; whose plots yielded the data sets for their cooperation

408 and maintenance of the study pilot sites. We also thank the leadership of the communes for

409 assisting with mobilization at the beginning of the project. We are grateful for the support from

410 the Ministry of Agriculture, Department of Agricultural extension and ISABU in coordinating

411 BBTD recovery and seed systems projects in Burundi. This study was funded by the CGIAR

412 Cross cutting Research Programme on Root Tubers and Bananas under Cluster 3.4 Banana virus

413 diseases.

414

415 Conflict of Interest:

416 Authors declare no conflict of interest.

18
417

418 Author roles:

419 Conceived the study and designed study tools: ABO, CS

420 Collected data:

421

422

423 Data availability:

424

425

426

427 References

428 Bruin, J. (2006). new test: command to compute new test. UCLA: Statistical Consulting Group.

429 https://stats.idre.ucla.edu/stata/ado/analysis/

430 Chabi M (2022)

431 Dato, K. M. G., Dégbègni, M. R., Atchadé, M. N.,  Hounkonnou M. N., & Omondi, A. B.

432 (2021). Spatial parameters associated with the risk of banana bunchy top disease in

433 smallholder systems. Plos One, 16(12), e0260976. DOI: 10.1371/journal.pone.0260976.

434 Lindén, A., and Mäntyniemi, S., (2011). Using the negative binomial distribution to model

435 overdispersion in ecological count data. Ecology, 92(7), 1414-

436 1421. https://doi.org/10.1890/10-1831.1

437 Gelman, A., J. B. Carlin, H. S. Stern, and D. B. Rubin. (2004). Bayesian data analysis. Second

438 edition. Chapman and Hall/CRC, Boca Raton, Florida, USA.

19
439 Ajambo S, Rietveld A, Nkengla LW, et al., (2018). Recovering banana production in bunchy

440 top-affected areas in Sub-Saharan Africa: developing gender-responsive approaches. Acta

441 Horticultureae 1196, 219-228.

442 Azur, M. J., Stuart, E. A., Frangakis, C., & Leaf, P. J. (2011). Multiple imputation by chained

443 equations: what is it and how does it work?. International Journal of Methods in

444 Psychiatric Research, 20(1), 40–49. https://doi.org/10.1002/mpr.329

445 Leclerc M, Dore´ T, Gilligan CA, Lucas P, Filipe JAN (2014) Estimating the Delay between

446 Host Infection and Disease (Incubation Period) and Assessing Its significance to the

447 Epidemiology of Plant Diseases. PLoS ONE 9(1), e86568.

448 doi:10.1371/journal.pone.0086568

449 Meyerson LA and Mooney HA (2007) Invasive alien species in an era of globalization.

450 Frontiers in Ecology and Environment 5, 199-208.

451 Hulme P. (2009). Trade, transport and trouble: managing invasive species pathways in an era of

452 globalization, Journal of Applied Ecology, 46, 10 – 18.

453 Allen RN, 1978. Epidemiological factors influencing the success of rogueing for the control of

454 bunchy top disease of bananas in New South Wales. Australian Journal of Agricultural

455 Research 29, 535–544.

456 Wessels N., Omondi AB, Kruger K and Jooste E (2019) Non-destructive DNA extraction from

457 aphids: the application in virus - vector studies of Banana bunchy top virus (BBTV)

458 European Journal of Phytopathology, 153,  571–582.

459 Djaillo B.D., Lokana J., Ngama F. et al., (2016). Systemicity of banana bunchy top viral

460 infection in the Kisangani region of the Democratic Republic of Congo. African Journal of

461 Agricultural Research 11, 527-32.

462 Global Invasive Species Database, 2018 Species profile: Banana bunchy top virus (BBTV).

463 http://www.iucngisd.org/gisd/species.php?sc=141 accessed on 21-03-2018.

20
464 Kumar PL, Hanna R, Alabi OJ, Vangu GH, Naidu RA 2011. Banana bunchy top virus in sub-

465 Saharan Africa: Investigations on virus distribution and diversity. Virus Research 159,

466 171–82.

467 FAOSTAT (2013) FAO statistical database. http://faostat3.fao.org/home/E. accessed date?

468

469 Kumar PL, Selvarajan R, Iskra-Caruana ML, Chabannes M, Hanna R, 2015. Biology, Etiology,

470 and Control of Virus Diseases of Banana and Plantain. Advances in Virus Research 91,

471 229-69.

472 Nduwimana et al., 2022

473 Niyongere C, Ateka EA, Losenge T, Lepoint P, Blomme G, 2011. Screening Musa genotypes for

474 banana bunchy top disease resistance in Burundi. Acta Horticulturae, 897, 439–47.

475 Niyongere et al., 2013 XXXXXXXXXX

476 Kebede, Y., Baudron, F., Bianchi, F., and Tittonell, P. (2018). Unpacking the push-pull system:

477 Assessing the contribution of companion crops along a gradient of landscape complexity.

478 Agriculture Ecosystems & Environment. 268, 115-123. 10.1016/j.agee.2018.09.012.

479 Zuur, A.F., Ieno, E.N. & Elphick, C.S. (2010) A protocol for data exploration to avoid common

480 statistical problems. Methods Ecology and Evolution, 1, 3–14.

481 Zuur, A.F., & Ieno, E.N. (2016) A protocol for conducting and presenting results of regression-

482 type analyses Methods in Ecology and Evolution, 7, 636–645.

483 Ngatat S, Hanna, R, Kumar, PL, et al., 2017. Relative susceptibility of Musa genotypes to

484 banana bunchy top disease in Cameroon and implication for disease management, Crop

485 Protection 101, 116 – 122.

486 Simbare, A., Sane, C.A.B., Nduwimana, I., Niyongere, C., Omondi, B.A., (2020). Diminishing

487 Farm Diversity of East African Highland Bananas in Banana Bunchy Top Disease

488 Outbreak Areas of Burundi—The Effect of Both Disease and Control Approaches.

489 Sustainability 12, 7467. https://doi.org/10.3390/su12187467

21
490 Stainton, D, Martin, DP, Muhire, et al. (2015). The global distribution of Banana bunchy top

491 virus reveals little evidence for frequent recent, human-mediated long distance dispersal

492 events. Virus Evolution 1, 1-16.

493 Watanabe S, Greenwell AM, Bressan A, (2013). Localization, concentration, and transmission

494 efficiency of banana bunchy top virus in four asexual lineages of Pentalonia aphids.

495 Viruses 5, 758–775. Accessed on 26 June 2015 http://dx.doi.org/10.3390/v5020758

496 Cunniffe, N.J., Stutt, R.J.O.H., van den Bosch, F., Gilligan, C.A., (2012). Time-dependent

497 infectivity and flexible latent and infectious periods in compartmental models of plant

498 disease. Phytopathology 102, 365–380.

499 Ingwell, L.L., Eigenbrode, S.D., Bosque-Perez, N.A., (2012). Plant viruses alter insect behavior

500 to enhance their spread. Scientific Reports, 2, 578 doi: 10.1038/srep00578.

501 van der Plank, J.E., 1948. The relation between the size of fields and the spread of disease

502 between them. Part I. Crowd diseases. Emp. J. Exp. Agric. 16, 134–142.

503 Omondi, B. A., Soko, M. M., Nduwimana, I., Delano, R. T., Niyongere, C., Simbare, A., &

504 Staver, C. (2020). The effectiveness of consistent roguing in managing banana bunchy top

505 disease in smallholder production in Africa. Plant Pathology, 69, 1754-1766

506 https://doi.org/10.1111/ ppa.13253

507 Quénéhervé P, Barrière V, Salmon F, Houdin F, Achard R, Gertrude J-C, Marie-Luce S,

508 Chabrier C., Duyck P-F, Tixier P, (2011) Effect of banana crop mixtures on the plant-

509 feeding nematode community, Applied Soil Ecology 49: 40– 45

510 Smith, M.C, Holt, J. Kenyon L. and Foot C (1998) Quantitative epidemiology of Banana Bunchy

511 Top Virus, Disease and its control, Plant Pathology, 47, 177–187

512

513

514

515

22
516

517

518

519

520 Table 1: Model selection results for predictors of BBTD occurrence in recovered gardens under study

N Full Variable Name Code Description


1 Aphid presence on plot AphidP Count of aphids on 10 randomly selected mats.

2 Roguing frequency per year Roguing Farmer estimation on the number of times they removed diseased plants in
the past year, on a monthly.
3 Type of planting materials PlantingMtls The choice of suckers, macropropagation of tissue culture plants used for
planting.
4 Source of planting materials SourcePlMtls Number of sources of planting materials used

5 Number of banana cultivars CultNber Number of Banana Cultivars grown on plot

6 Farm owner sex Sex Male or female

7 Plot size PltSiz Area of plot in Square metres (m²)

8 Plot Altitude Altitude Elevation of farm taken at centre in metres above sea level (m asl)

9 Plot location Location Commune where plot is located in Burundi

10 Banana density on plot Density Estimated number of mats per square metre within the plot.

11 Cropping mixture CropSystm Whether monoculture or mixed cropping

12 Contact with neighbors ContcNeigh Extent of information and contact sharing with neighbours

13 Age of banana plantation BanAge Age in years since the plantation was established

14 Distance to nearest infested Disease Distance in metres to the edge of the focus plot from that of a plot in which
plot distance BBTD has been identified

15 BBTD prevalence in 30m BBTD30m Proportion of diseased plants within 30m distance of the plot
band
16 Banana free period BanFallow Did the farmer maintain a period of 3 months or more with no banana
growing on the target plot.
17 Distance from infested mat DistBBTDMat The shortest distance to an identified BBTD symptomatic mat to the
nearest edge of the target garden.
18 Associated crop cycle CyclIntercrop Cycle of the associated crop (eg annual, perennial etc)

19 Seed exchange with SeedExch If informal seed exchange occurred with neighbours since project began
neighbours
20 Presence of hedge SSSSS Was there a defined hedge around the plot ?

21 BBTD prevalence 60m band BBTD60m Proportion of diseased plants within 60m distance of the plot

22 BBTD prevalence 90m band BBTD90m Proportion of diseased plants within 90m distance of the plot

23
23 Banana density 30m band BanDen30m Number of banana plants per square metre within 30m distance of the plot

24 Banana density 60m band BanDen60m Number of banana plants per square metre within 60m distance of the plot

25 Banana density 90m band BanDen90m Number of banana plants per square metre within 90m distance of the plot

521

522 Table 2: comparison of the coefficients for the remaining predictors with their coefficients from
523 the original model

524

Predictors Full Model HR Reduced Model HR  


exp(coef) exp(coef) differences

Plot size in m2 1.00 1.00 0.00


Mean of Banana density 30m 109.16 899.53 790.37
Edge presence 0.66 0.66 0.00
Reference is edge presence

Number of Varieties_two varieties 1.28 1.47 0.19


Number of Varieties_three varieties 1.64 1.85 0.21
Number of Varieties_four varieties 1.31 1.60 0.26
Number of Varieties_more than five varieties 1.53 1.92 0.39
Reference is one variety

Type of planting material_TC 0.12 0.12 -0.01


Type of planting material Suckers+TC 0.92 0.86 -0.06
Reference seed is Suckers

Roguing diseased banana Once a month 0.52 0.58 0.05


Roguing diseased banana _Twice a month 0.61 0.59 -0.02
Roguing diseased banana _Weekly 0.52 0.51 -0.01
Roguing diseased banana _Twice a week 1.05 0.81 -0.24
Roguing diseased banana _Thrice a week 0.69 0.59 -0.09
Roguing diseased banana _Four time a week 0.56 0.47 -0.09
Roguing diseased banana _ > four times a week 0.74 0.61 -0.13
Roguing diseased banana Less than once a month

Location_Kagazi 0.88 0.93 0.05


Location_Munyika 0.53 0.51 -0.01
Location_Rusagara 0.98 1.14 0.16
Location _ Gitebe

525

24
526 Table 3: Ranking of most relevant combinations of predictors for BBTD prevalence in smallholder systems.

Model (Predictor Variables) AIC ΔAIC Akaike weight(wi)


AphidP+Roguing+PlantingMtls+SourcePlMtls+CultNber+Sex+PltSiz+Altitude 189.10 0 0.36
AphidP+Roguing+SourcePlMtls+CultNber+CropSystm 189.21 0.11 0.27
Roguing+SourcePlMtls+CultNber+Sex+Density+BanAge 189.1 0.76 0.19
AphidP+PltSiz+[Altitude+Location]+Density+CropSystm+ContcNeigh+BanAge+Disdist 189.97 1.12 0.18
AphidP+Roguing+PlantingMtls+SourcePlMtls]+CultNber+Sex+PltSiz+BanAge 191.09 1.40 0.14
AphidP+Roguing+PlantingMtls+SourcePlMtls+PltSiz+Altitude 192.49 1.55 0.10
AphidP+PlantingMtls+SourcePlMtls+Sex+BanAge+Disdist 194.04 1.60 0.07
Roguing+Sex+[Altitude+Location]+Density+CropSystm 195.64 1.71 0.07
AphidP+CultNber+PltSiz+Altitude+ContcNeigh+BanAge+Disdist 197.35 1.75 0.04
Roguing+Sex+PltSiz+[Altitude+Location]+Density+ContcNeigh+BanAge 199.1 1.81 0.04
AphidP+Sex+PltSiz+[Altitude+Location]+Density+CropSystm 200.91 1.90 0.02
Roguing+PlantingMtls+[Altitude+Location]+Density 202.81 1.95 0.01

527
528
529 AIC, Akaike Information Criterion; ΔAIC, Difference of AIC between the model and the most parsimonious model. Models shown here are the top 12 best models reported following Burnham
530 & Anderson(2002) and Wood (2006) . The preferred model (simplest model with ΔAIC < 2).

531

532

533

534

25
535 Table 4: output of negative binomial regression

Estimate Odd-Ratio Probabilities of Std. Error z value Pr(>|z|) 2.5 % 97.5 %


event
(Intercept) 2.70 14.94 0.94 0.43 6.30 0.0000*** 1.86 3.55
Plot size 0.00 1.00 0.50 0.00 2.71 0.00667 ** 0.00 0.00
Mean of Nearest BBTD infected banana mat 0.00 1.00 0.50 0.00 -0.46 0.64790 -0.01 0.00
Mean of Banana density 30m 4.69 108.47 0.99 2.28 2.05 0.03997 * 0.21 9.16
Mean of Banana density 90m 0.06 1.06 0.52 0.11 0.53 0.59371 -0.16 0.28
Rogueing per year farmer declaration 0.01 1.01 0.50 0.00 1.73 0.08309 0.00 0.02
Aphids presence on asymptomatic mother plant_No -0.18 0.84 0.46 0.22 -0.80 0.42227 -0.61 0.26
Aphids presence on symptomatic mother plants_No 0.33 1.40 0.58 0.23 1.48 0.13837 -0.11 0.78
Hedge presence East_No -0.38 0.68 0.41 0.15 -2.64 0.00834 ** -0.67 -0.10
Number of Varieties_two varieties 0.28 1.32 0.57 0.27 1.04 0.29784 -0.25 0.81
Number of Varieties_three variety 0.54 1.72 0.63 0.26 2.13 0.03342 * 0.04 1.04
Number of Varieties_four varieties 0.41 1.51 0.60 0.33 1.24 0.21554 -0.24 1.06
Number of Varieties_more than five varieties 0.77 2.16 0.68 0.48 1.61 0.10853 -0.17 1.71
Type of planting material_TC -1.97 0.14 0.12 0.73 -2.71 0.00674 ** -3.39 -0.54
Type of planting material_Suckers+TC -0.08 0.92 0.48 0.14 -0.56 0.57812 -0.36 0.20
Frequency of scouting infected_Once a month -0.57 0.56 0.36 0.36 -1.58 0.11355 -1.29 0.14
Frequency of scouting infected_Twice a month -0.59 0.55 0.36 0.36 -1.66 0.09604 -1.29 0.11
Frequency of scouting infected_Weekly -0.81 0.44 0.31 0.33 -2.44 0.01466 * -1.47 -0.16
Frequency of scouting infected_Twice a week -0.23 0.79 0.44 0.37 -0.64 0.52550 -0.95 0.49
Frequency of scouting infected_Thrice a week -0.55 0.58 0.37 0.36 -1.54 0.12313 -1.25 0.15
Frequency of scouting infected_Four time a week -0.80 0.45 0.31 0.38 -2.12 0.03445 * -1.55 -0.06
Frequency of scouting infected_ > four times a week -0.62 0.54 0.35 0.41 -1.52 0.12837 -1.43 0.18
Banana free period_No -0.13 0.88 0.47 0.20 -0.64 0.52499 -0.52 0.26
Location_Kagazi -0.01 0.99 0.50 0.22 -0.04 0.96760 -0.44 0.42
Location_Munyika -0.64 0.53 0.35 0.22 -2.93 0.00334 ** -1.07 -0.21
Location_Rusagara 0.18 1.20 0.55 0.20 0.90 0.37060 -0.22 0.58

26
Figure 2: Histogram showing the density of the number of BBTD infected banana plants (for the count regression model).

537

538

539

540

541 Figure 3: Regression plot of Plot altitude on BBTD prevalence draw all possible trend (log,
542 geometric, etc)

543

544

545

27
Roguing

Intercropping

Scouting for BBTD

Hedge presence

Use of healthy banana seedling

Crop rotation

Banana free period

0 10 20 30 40 50 60 70 80 90 100

Percentage of farmers
546

547 Figure 3: BBTD management methods reported by farmers.

548 Figure 3: Proportion of farmers reporting the use of specific single BBTD control associated
549 practices in the study area.

550

551

28
552

Roguing+Intercropping

Roguing+Intercropping+Scouting for BBTD

Roguing+Intercropping+Scouting for BBTD+Hedge presence

Roguing+Intercropping+Scouting for BBTD+Hedge presence

Roguing+Intercropping+Scouting for BBTD+Hedge presence+Use of healthy


banana seedling

Roguing+Intercropping+Scouting for BBTD+Hedge presence+Use of healthy


banana seedling

Roguing+Intercropping+Scouting for BBTD+Hedge presence+Use of healthy


banana seedling+Crop rotation

Roguing+Intercropping+Scouting for BBTD+Hedge presence+Use of healthy


banana seedling+Crop rotation+Banana free period

0 5 10 15 20 25 30 35 40 45 50

554

555 Figure 4: Multiple BBTD management methods used by farmers

556

557 Figure 4: The frequency of use of selected combinations of management practices that
558 showed the greatest association with BBTD prevalence reduction.

29
559 Supplementary information 1(?): give a title

560  Nonparametric dispersion test via mean deviance residual fitted vs. simulated-refitted.

561 dispersion=1.0284 , p−value=0.576


562 alternative hypothesis: two.sided

563  Hosmer and Lemeshow goodness of fit (GOF) test

564 X −squared=−4.619 , df =8 , p−value ⋍ 1


565
566 In the Q-Q Residuals (Supplementary Figure 1), we can see that the residual point followed a

567 diagonal (almost perfect) line. Therefore, the residuals are normally distributed. Also, from

568 the above parameter dispersion results, we can see that the parameter dispersion value is

569 1.0284 already close to 1. This means that the negative binomial regression model can reduce

570 the overdispersion problem.

571 Furthermore, the Hosmer and Lemeshow goodness of fit (GOF) test show that the p-

572 value is closed to 1, more than 0.05, so there is good evidence that the model can be adapted

573 to the data.

574

575

576

577

578

579

580

581

582

583

584

30
585

586

587 Supplementary Figure 1. Q-Q Residual graph. BBTD residuals were normally distributed hence, meaning the negative binomial model
588 fits the dataset.

31
589

32

You might also like