Cassava Brown Streak Virus Infection Genome

Cassava Brown Streak Virus Infection Genome Genome-wide prediction and association analysis for sensitivity to cassava brown streak virus infection in Cassava Siraj Ismail Kayondo, Dunia Pino Del Carpio, Roberto Lozano, Alfred Ozimati, Marnin Wolfe, Yona Baguma, Vernon Gracen, Offei Samuel,  Robert Kawuki  and Jean-Luc Jannink ABSTRACT Cassava (manihot esculenta Crantz), a key carbohydrate source faces unprecedented challenge of viral diseases importantly, cassava brown streak disease (CBSD) and cassava mosaic disease (CMD). The economic parts of the crop are rendered unmarketable by these viral diseases resulting into mega fiscal setbacks. The remarkable completion of the cassava genome sequence equips cassava breeders with more precise selection strategies to offer superior varieties with both farmer and industry preferred traits. This article reports genomic segments associated to foliar and root CBSV sensitivity measured at different growth stages and environmental conditions. We identified significant single nucleotide polymorphisms (SNPs) associated to CBSV sensitivity in cassava on chromosome 4 and 11. The significantly associated regions on chromosome 4 co-localises with a Manihot glaziovii introgression from the wild progenitors. While significant SNPs markers on chromosome 11 are in linkage disequilibrium (LD) with a cluster of nucleotide-binding site leucine-rich repeat (NBS-LRR) proteins encoded by disease resistance genes in plants. Genotype by environmental interactions were significant since SNP marker effects differed across environments and years. Key words: Genome-wide association studies (GWAS), virus sensitivity, augmented designs, de-regressed best linear unbiased Predictions (dr-BLUPs), NBS-LRR proteins, QTLs    INTRODUCTION Cassava (Manihot esculenta crantz), is a major source of income and dietary calories for over a billion lives across the globe especially in Sub Saharan Africa (SSA). Edge cutting technologies are rapidly turning cassava into an industrial crop especially tapping into its unique starch qualities hence opening new income opportunities for the poor (Pà ©rez et al., 2011). Cassava brown streak virus disease (CBSD), a leading viral constraint limiting production across SSA is responsible for mega fiscal setbacks estimated at 100 US million dollars per annum at physiological maturity (ASARECA:, 2013; Ndunguru et al., 2015). As a consequence of CBSVs, cassava yields were recorded to be eight times lower than the expected yield potential in Uganda(). Two major strains; Cassava brown streak virus (CBSV) and Uganda Cassava brown streak virus (UCBSV), have successfully colonized both the lowland and highland altitudes across East Africa though newer strains are being reported (Winter et al., 2010; Ndunguru et al., 2015; Alicai et al., 2016a). In addition to uncontrolled exchange of infected cassava steaks among cassava farmers across porous borders, the African whitefly (Besimia tobaci) stands out as the famous semi-persistent virus transmitter under field conditions (Legg et al., 2014; McQuaid et al., 2015). Upon entry, the virus exploits the plants transport system to traverse the susceptible cassava plant resulting into yellow chlorotic vein clearing patterns along minor veins of the leaves. Prominent brown elongated lesions are formed on the stem commonly referred to as brown streaks. While the brown necrotic hard-corky layers are randomly formed in the root cortex of most susceptible cassava clones. In view of the rapid but steadily virus evolution rates and the insufficiency of dependable virus diagnostic tools (Alicai et al., 2016b), breeding for durable CBSD resistance emerges as a timely and economically viable option. Earlier CBSD resistance breeding initiatives have highlighted its polygenic but recessive nature of inheritance in both intraspecific and interspecific cassava hybrids (Nichols, 1947; Hillocks and Jennings, 2003; Munga, 2008; Kulembeka, 2010). The rate of progress to genetic improvement in a traditional cassava breeding pipeline has been slower due to several biology-related opportunities like; shy flowering, length of breeding cycle, limited genetic diversity and slow rate of multiplication of planting materials. Most of the available elite cassava lines have exhibited some level of sensitivity to CBSVs ranging from mild sensitivity total susceptibility. Therefore, a concise but then targeted exploration for potential sources of resistance using the available biotechnology tools could be a promising strategy. The remarkable completion of the cassava genome sequence equips cassava breeders with more precise selection strategies to offer superior varieties with both farmer and industry preferred traits. A study by Bredeson et al., (2016) reports the presence of introgressions segments from the wild progenitors into the elite breeding lines developed by the Amani breeding program in Tanzania. Hence, resistance sources to CBSD exist but may have been reshuffled over generations of recurrent selection thus not fully fixed and need to be exploited. Moving forward, a genome wide survey for existing natural variations as explained by the observed phenotypes for a given series of agronomic traits could facilitate identification of causal loci associated with the inheritance of a trait of interest. This tool, commonly referred to as genome-wide association study (GWAS) exploits the power of statistical analyses to identify such historical recombination events that have occurred over time (Jannink and Walsh, 2002; Hamblin, Buckler and Jannink, 2011). Hence, GWA-studies will complement bi-parental mapping efforts that have been widely applied in cassava breeding in the previous decade (Ferguson et al., 2012; Ceballos et al., 2015). GWA-studies have been widely undertaken by animal, human and plant geneticists to identify quantitative trait loci (QTLs) in close association to several important traits. However, GWAS has been thinly applied in cassava breeding especially in the definition of the genetic architecture of cassava mosaic di sease (Wolfe et al., 2016) and beta carotene (unpublished). In this study, we exploited the reduced genotyping costs using genotyping by sequencing (GBS) to genotype data for our association mapping panel. The goal of this study was to identify genomic regions closely associated with sensitivity to CBSV infection in a diverse regional cassava breeding panel. Fine mapping around the identified regions would guide in marker discovery as well as identification of franking genes for CBSV sensitivity for marker assisted breeding. MATERIALS AND METHODS Plant material The data set comprised of field disease evaluations undertaken across five locations; Namulonge, Kamuli, Serere, Ngetta and Kasese in Uganda. Two different but closely related GWAS panels were evaluated across environments. Between 2012 and 2013, GWAS panel 1 consisted of between 308 to 429 entries that were replicated twice across three locations. Each trial was designed as a randomized complete block (RCB) with two-row plots of five plants each at a spacing of 1 meter by 1 meter. In 2015, GWAS panel 2 consisting of entries ranging from 715 to 872 clones was evaluated in three locations but contrasting sites for CBSD pressure. These entries were evaluated as single entries per site being connected by six common checks in an augmented completely randomized block design with 38 blocks per site (Federer, Nguyen and others, 2002; Federer and Crossa, 2012). The two GWAS panels had one location in common; Namulonge that is regarded as the CBSD hot spot with the highest CBSD pressure. The data was generated from 1281 cassava clones developed through three cycles of genetic recombination with local elite lines by the National root crops breeding program at NaCRRI. These cassava clones had a diverse genetic background whose pedigree could be traced back to introductions from international institute for tropical agriculture (IITA), International center for tropical Agriculture (CIAT) and Tanzania[KI1] breeding program (sup.fig1). Phenotyping protocol for CBSV sensitivity   Ã‚   The key traits were CBSD severity and incidence scored at 3, 6, and 9 months after planting (MAP) for foliar and 12 MAP for root symptoms respectively. CBSD severity was measured based on a 5 point scale with a score of 1 implying asymptomatic conditions and a score 5 implying over 50% leaf vain clearing under foliar symptoms. However, at 12 MAP a score of 5 implies over 50% of root-core being covered by a necrotic corky layer. (fig.1) Clones were classified with a score of 5 if pronounced vein clearing at major leaf veins were jointly displayed with brown streaks on the stems and shoot die-back that appeared as a candle-stick. Clones with 31 40% leaf vein clearing together with brown steaks at the stems were classified under score 4. A Score of 3 was assigned to clones with 21 30% leaf vein clearing with emerging brown streaks on the stems. While a score of 2 was assigned to clones that only displayed 1 20% leaf vein clearing without any visible brown streak symptoms on the stems. Plants classified with a score of 1 showed no visible sign of leaf necrosis and brown streaks on the stems. On the other hand, root symptoms were also classified into 5 different categories based on a 5 point standard scale. Two-stage genomic analyses For the panel 1 which was designed as a randomized complete block (RCB) we fit the model: , using the lmer function from the lme4 R package (Bates et al., 2015).In this model, ÃŽÂ ² included a fixed effect for the population mean and location. The incidence matrix Zclone and the vector c represent a random effect for clone and I represent the identity matrix. The range variable, which is the row or column along which plots are arrayed, is nested in location-rep and is represented by the incidence matrix Zrange(loc.) and random effects vector .Block effects were nested in ranges and incorporated as random with incidence matrix Zblock(range) and effects vector . Residuals were fit as random, with . For panel 2, which followed an augmented design, we fit the model Where y was the vector of raw phenotypes, ÃŽÂ ² included a fixed effect for the population mean and location with checks included as a covariate, The incidence matrix Zclone and the vector c are the same as above and the blocks were also modeled with incidence matrix and b represents the random effect for block. The best linear predictors (BLUPs) of the clone effect (Ć°) were extracted as de-regressed BLUPS following the formula: Broad sense heritability was calculated using variance components extracted from the two step lmer output. SNP-based heritability was calculated by extracting the variance components from the output obtained by fitting the SNPs as a kinship covariate calculated using the A.mat function from the rrBLUP R package and included in a one step model using the emmreml function from the EMMREML R package (Akdemir and Okeke, 2015). DNA preparation and Genotyping by sequencing (GBS) All cassava clones included in the phenotypic data set had their total genomic DNA extracted from young tender leaves according to standard procedures using the DNAeasy plant mini extraction kit (Qiagen, 2012). Genotyping-by-sequencing (GBS) (Elshire et al., 2011) libraries were constructed using the ApeKI restriction enzyme as used before (Hamblin Rabbi, 2014). Marker genotypes were called using TASSEL GBS pipeline V4 (Glaubitz et al., 2014) after aligning the reads to the Cassava v6 reference genome (Phytozome 10.3; http://phytozome.jgi.doe.gov) (International Cassava Genetic Map Consortium, 2014; Prochnik et al., 2012). Variant Calling Format (VCF) files were generated for each chromosome. Markers with more than 60% missing calls were removed. Genotypes with less than 5 reads were masked before imputation. Additionally, only biallelic SNP markers were considered for further steps. The marker dataset consisted of a total of 173,647 SNP bi-allelic markers called for 986 individuals. This initial dataset was imputed using Beagle 4.1 (Browning and Browning, 2016). After the imputation 63,016 SNPs had an AR2 (Estimated Allelic r-squared) higher than 0.3 and were kept for analysis; from these, 41,530 had a minor allele frequency (MAF) higher than 0.01 in our population. Dosage files for this final dataset were generated and used for both GWAS and GS. Structure and Genetic stratification analysis The extent of phylogenetic relationship and degree of family relatedness within the cassava lines was assessed using principal component analysis (PCA) implemented in R. Genome-wide association analysis for CBSV sensitivity The input binary PED files were prepared from the genotype dosage files using PLINK version 1.07 (Renterà ­a, Cortes and Medland, 2013; Purcell et al., 2007). Mixed linear modal association analysis (MLMA) implemented by GCTA version 1.26.0 was used to generated GWAS results (Yang et al., 2011). MLMA was implemented such that in every cycle of analysis, the chromosome on which the candidate SNPs existed got excluded from the GRM calculation using the modal in equation 3. Where y is the phenotype, a is being the mean term, b being the fixed additive effects of the candidate SNP being tested for association, x being the SNP genotype indicator variable and g is the accumulated effect of all SNPs excluding those where the candidate SNP is located making our analysis model more powerful. We estimated variance components using restricted maximum likelihood (REML). Sub population stratification was corrected for by taking GRM as a random effect term in the model during analysis. A more conservative Bonferroni correction method was used to fix genome-wide significance threshold at Pà ¢Ã¢â‚¬ °Ã‚ ¤10-7 as a way of correcting for experimental-wise error. Manhattan and Quantile Quantile plots for all the traits were constructed using R package qqman package implemented in R (Turner, 2014). Genomic prediction models GBLUP. In this prediction model the GEBVs are obtained by assuming , where is the additive genetic variance, and K is the symmetric genomic realized relation matrix based on GBS SNP marker dosages. The genomic relationship matrix used was constructed using the function A.mat in the R package rrBLUP(Endelman, 2011) and follows the formula of VanRaden (2008), method two. GBLUP predictions were made with the function emmreml in the R package EMMREML (Akdemir and Okeke, 2015). RKHS. Unlike GBLUP for RKHS we use a Gaussian kernel function: , where Kij is the measured relationship between two individuals, dij is their euclidean genetic distance based on marker dosages and ÃŽÂ ¸ is a tuning (bandwidth) parameter that determines the rate of decay of correlation among individuals. This function is nonlinear therefore the kernels used for RKHS can capture non-additive as well as additive genetic variation. To fit a multiple-kernel model with six covariance matrices we used the emmremlMultiKernel function in the EMMREML package, with the following bandwidth parameters: 0.0000005, 0.00005, 0.0005, 0.005, 0.01, 0.05 (Multi-kernel RKHS) and allowed REML to find optimal weights for each kernel.For the optimal kernel RKHS we used the kernel weights assigned by emmremlMultiKernel in the first step to construct a single kernel that is the weighted average of the original six. We then used this optimal kernel in single-kernel predictions. Bayesian maker regressions.We tested four Bayesian prediction models: BayesCpi (Habier et al., 2011), the Bayesian LASSO (BL; Park and Casella, 2008), BayesA, and BayesB (Meuwissen, Hayes and Goddard, 2001). The Bayesian models we tested allow for alternative genetic architectures by differential shrinkage of marker effects. We performed Bayesian predictions with the R package BGLR (Pà ©rez and De Los Campos, 2014) Random Forest. Random forest (RF) is a machine learning method used for regression and classification (Strobl et al. 2009, Breiman 2001). Random forest regression with marker data has been shown to capture epistatic effects and has been successfully used for prediction (Sakar et al 2015, Heslot et al 2012, Charmet et al 2014, Spindel et al 2015, Breiman, et al 2001, Michaelson et al 2010, Motsinger-Reif et al 2008). We implemented RF using the randomForest package in R (Liaw and Wiener 2002) with the parameter, ntree set to 500 and the number of variables sampled at each split (mtry) equal to 300. Multikernel GBLUP We followed a multikernel approach by fitting three kernels constructed with SNPs with MAF> 0.01 from chromosomes 4,11 and the SNPs from the other chromosomes. We selected chromosomes 4 and 11 because they contained QTLs for foliar severity 3 and 6 MAP. Multikernel GBLUP predictions were made with the function emmremlMultiKernel in the R package EMMREML (Akdemir and Okeke, 2015) Introgression Segment Detection To identify the genome segments in our germplasm, we followed the approach of Bredeson et al. (2015). We used the M. glaziovii diagnostic markers identified Supplementary Dataset 2 of Bredeson et al. (2015). These ancestry diagnostic (AI) SNPs were identified as being fixed for different alleles in a sample of two pure M. esculenta (Albert and CM33064) and two pure M. glaziovii (GLA XXX-8 and M. glaziovii(S)). Out of 173,647 SNP in our imputed dataset, 12,502 matched published AI SNPs. For these AI SNPs, we divided each chromosome into non-overlapping windows of 20 SNP. Within each window, for each individual, we calculated the proportion of genotypes that were homozygous (G/G) or heterozygous (G/E) for M. glaziovii allele and the proportion that were homozygous for the M. esculenta allele (E/E). We assigned G/G, G/E or E/E ancestry to each window, for each individual only when the proportion of the most common genotype in that window was at least twice the proportion of the second most common genotype. We assigned windows a No Call status otherwise. We also used this approach on six whole-genome sequenced samples from the cassava HapMap II (Punna et al. under Review). These included the two pure cassava and M. glaziovii(S) from Bredeson et al. (2015), plus an additional M. glaziovii, and two samples labeled Namikonga. Because these samples came from a different source from the majority of our samples, we were able to find only 11,686 SNPs that matched both the sites in the rest of our study sample and the list of ancestry informative sites for analysis. Linkage disequilibrium plots LD scores were calculated for every SNP in chromosome 4 with a window of 1Mb using the GCTA Software (Yang et al., 2011). Briefly, LD score for a given marker is the sum of R2 adjusted between the index marker and all markers within a specified window. The adjusted R2 is an unbiased measure of LD: where n is the population size and R2 is the usual estimator of the squared Pearsons correlation (Bulik-Sullivan et al 2015). We calculated the LD between that marker and other markers in a window of 2Mb (1Mb upstream and 1Mb downstream) For the top significant SNP hit in chromosome 11 for the 6MAP GWAS result from panel 1 and panel 2. The LD was evaluated using squared Pearsons correlation coefficient (r2) as calculated with the à ¢Ã‹â€ Ã¢â‚¬â„¢r2 -ld-snp commands in the software PLINK version 1.9. Candidate gene identification To identify candidate genes for CBSD severity in leaves and CBSD root necrosis we used the GCTA mlma GWAS output obtained for each trait. We filtered the SNP markers based on -log10 (P-value)> Bonf, being these values higher than the Bonferroni threshold (~ 5.9). The resulting SNP markers were assigned onto genes using the SNP location and gene description from the Mesculenta_305_v6.1.gene.gff3 available in Phytozome 11 (ref) for Manihot esculenta v6.1 using the intersect function from bedtools (ref). RESULTS Phenotypic assessment of cassava for sensitivity to cassava brown streak virus infection Most clones showed varied responses to CBSV infection spanning from super susceptibility that represented candle-like die-back of the shoot to tolerance (Fig.). Foliar phenotyping clumped these plant responses into five major classes based on a 1 to 5 scale. The broad sense heritability of the studied traits ranged 0.17 to 0.72 for both GWAS panels (Table 1). Analysis of the phenotypic data showed very significant GxE interactions (P Genetic correlations and heritability estimates We found moderate heritability estimates for CBSV sensitivity for foliar phenotypes at 3, 6 and 9 MAP as well as root phenotypes under five environments (fig..) . Genetic correlation on the traits assayed were performed and revealed that ranged from moderate to high positive correlations among traits studied. Assessment of linkage disequilibrium Genome-wide association mapping often explores the benefit of existence of several historic recombination events over time to associate observed phenotypic variation with genome. Detection of candidate QTLs for CBSV sensitivity in cassava To efficiently a run GWAS, we used SNP data to examine the extent of genetic interrelatedness and sub-population structure of the cassava clones. A principal component analysis (PCA) to account for structure showed no distinct clusters implying that the selected clones were not highly structured (Fig.1). Therefore, we did not include PCs in our GWAS linear modal analysis. The Bonferroni suggestive threshold (ÃŽÂ ± = 0.05) was used to identify loci associated to CBSV sensitivity on both chromosome 4 and 11 that had clear peak signals at the different stages of phenotyping (Fig. 2). The observed P-values initially aligned well with the expected P-values but later differed substantially due the large introgression block on chromosome 4 presumably from the wild progenitors of cassava traceable from the AMANI breeding program (Jennings, 1959). The significant signals on chromosome 11 contained loci with strong association with CBSV sensitivity in a 2 Mb region that annotated well with several candidate genes. Genome-wide prediction for CBSV sensitivity in cassava We did genome wide prediction for CBSV sensitivity based on the identified SNPs with the highest effects found on both chromosome 4 and 11 in order to capture most of the genetic variation. We explored several genomic prediction model methods; GBLUP, RR-BLUP, B-LASSO, random forest, BayesA, BayesB, and BayesC. References Akdemir, D. and Okeke, U. G. (2015) EMMREML: Fitting Mixed Models with Known Covariance Structures,. Alicai, T., Ndunguru, J., Sseruwagi, P., Tairo, F., Okao-Okuja, G., Nanvubya, R., Kiiza, L., Kubatko, L., Kehoe, M. A. and Boykin, L. M. (2016a) Characterization by Next Generation Sequencing Reveals the Molecular Mechanisms Driving the Faster Evolutionary rate of Cassava brown streak virus Compared with Ugandan cassava brown streak virus, bioRxiv, Cold Spring Harbor Labs Journals, [online] Available at: http://biorxiv.org/lookup/doi/10.1101/053546 (Accessed 27 June 2016). Alicai, T., Ndunguru, J., Sseruwagi, P., Tairo, F., Okao-Okuja, G., Nanvubya, R., Kiiza, L., Kubatko, L., Kehoe, M. A. and Boykin, L. M. (2016b) Characterization by Next Generation Sequencing Reveals the Molecular Mechanisms Driving the Faster Evolutionary rate of Cassava brown streak virus Compared with Ugandan cassava brown streak virus, [online] Available at: http://biorxiv.org/lookup/doi/10.1101/053546 (Accessed 10 June 2016). ASARECA: (2013) ASARECA Annual Report 2012: Transforming Agriculture for Economic Growth in Eastern and Central Africa, Entebbe, Uganda. Bredeson, J. V, Lyons, J. B., Prochnik, S. E., Wu, G. A., Ha, C. M., Edsinger-Gonzales, E., Grimwood, J., Schmutz, J., Rabbi, I. Y., Egesi, C., Nauluvula, P., Lebot, V., Ndunguru, J., Mkamilo, G., Bart, R. S., Setter, T. L., Gleadow, R. M., Kulakow, P., Ferguson, M. E., Rounsley, S. and Rokhsar, D. S. (2016) Sequencing wild and cultivated cassava and related species reveals extensive interspecific hybridization and genetic diversity., Nature biotechnology, 34(5), pp. 562-570, [online] Available at: http://dx.doi.org/10.1038/nbt.3535. Browning, B. L. and Browning, S. R. (2016) Genotype Imputation with Millions of Reference Samples, The American Journal of Human Genetics, 98(1), pp. 116-126. Ceballos, H., Kawuki, R. S., Gracen, V. E., Yencho, G. C. and Hershey, C. H. (2015) Conventional breeding, marker-assisted selection, genomic selection and inbreeding in clonally propagated crops: a case study for cassava, Theoretical and Applied Genetics, [online] Available at: http://link.springer.com/10.1007/s00122-015-2555-4 (Accessed 25 June 2015). Endelman, J. B. (2011) Ridge Regression and Other Kernels for Genomic Selection with R Package rrBLUP, The Plant Genome Journal, 4(3), pp. 250-255. Federer, W. T. and Crossa, J. (2012) Screening Experimental Designs for Quantitative Trait Loci, Association Mapping, Genotype-by Environment Interaction, and Other Investigations, Frontiers in Physiology, 3, [online] Available at: http://www.frontiersin.org/Journal/10.3389/fphys.2012.00156/full (Accessed 2 September 2014). Federer, W. T., Nguyen, N.-K. and others (2002) Constructing Augmented Experiment Designs with Gendex, [online] Available at: https://dspace.library.cornell.edu/handle/1813/32208 (Accessed 2 September 2014). Ferguson, M., Rabbi, I., Kim, D.-J., Gedil, M., Lopez-Lavalle, L. A. B. and Okogbenin, E. (2012) Molecular Markers and Their Application to Cassava Breeding: Past, Present and Future, Tropical Plant Biology, 5(1), pp. 95-109, [online] Available at: http://link.springer.com/10.1007/s12042-011-9087-0 (Accessed 5 August 2014). Habier, D., Fernando, R. L., Kizilkaya, K. and Garrick, D. J. (2011) Extension of the bayesian alphabet for genomic selection., BMC bioinformatics, BioMed Central Ltd, 12(1), p. 186. Hamblin, M. T., Buckler, E. S. and Jannink, J.-L. (2011) Population genetics of genomics-based crop improvement methods, Trends in Genetics, 27(3), pp. 98-106, [online] Available at: http://www.sciencedirect.com/science/article/pii/S0168952510002301 (Accessed 18 April 2014). Hillocks, R. J. and Jennings, D. L. (2003) Cassava brown streak disease: a review of present knowledge and research needs, International Journal of Pest Management, 49(3), pp. 225-234, [online] Available at: http://www.tandfonline.com/doi/abs/10.1080/0967087031000101061 (Accessed 5 February 2014). Jannink, J.-L. and Walsh, B. (2002) Association mapping in plant populations, Quantitative genetics, genomics and plant breeding, pp. 59-68, [online] Available at: http://books.google.com/books?hl=enlr=id=hMBY_MvS39ACoi=fndpg=PA59dq=%22of+meioses+that+have+occurred+and+(in+the+case+of+advanced+intercross+lines)%22+%22discussion+of+rigorous+methods+to+test+for+marker+association+with+a+quantitative%22+ots=wo7fd. Jennings, D. L. (1959) Manihot melanobasis Mà ¼ll. Arg.-a useful parent for cassava breeding, Euphytica, 8(2), pp. 157-162, [online] Available at: http://link.springer.com/article/10.1007/BF00022435 (Accessed 8 October 2014). Kulembeka, H. P. (2010) Genetic linkage mapping of Field Resistance to cassava brown streak Disease in cassava landraces from Tanzania, Doctoral thesis, University of the Free State. Legg, J. P., Sseruwagi, P., Boniface, S., Okao-Okuja, G., Shirima, R., Bigirimana, S., Gashaka, G., Herrmann, H. W., Jeremiah, S., Obiero, H., Ndyetabula, I., Tata-Hangy, W., Masembe, C. and Brown, J. K. (2014) Spatio-temporal patterns of genetic change amongst populations of cassava Bemisia tabaci whiteflies driving virus pandemics in East and Central Africa, Virus Research, Elsevier B.V., 186, pp. 61-75, [online] Available at: http://dx.doi.org/10.1016/j.virusres.2013.11.018. McQuaid, C. F., Sseruwagi, P., Pariyo, A. and van den Bosch, F. (2015) Cassava brown streak disease and the sustainability of a clean seed system, Plant Pathology, p. n/a-n/a, [online] Available at: http://doi.wiley.com/10.1111/ppa.12453 (Accessed 17 September 2015). Meuwissen, T. H., Hayes, B. J. and Goddard, M. E. (2001) Prediction of total genetic value using genome-wide dense marker maps., Genetics, 157(4), pp. 1819-29. Munga, T. L. (2008) Breeding for Cassava Brown Streak Resistance in Coastal Kenya, Doctoral thesis, University of KwaZulu-Natal Republic of South Africa. Ndunguru, J., Sseruwagi, P., Tairo, F., Stomeo, F., Maina, S., Djinkeng, A., Kehoe, M. and Boykin, L. M. (2015) Analyses of Twelve New Whole Genome Sequences of Cassava Brown Streak Viruses and Ugandan Cassava Brown Streak Viruses from East Africa: Diversity, Supercomputing and Evidence for Further Speciation, PloS one, 10(10), p. e0139321, [online] Available at: http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0139321 (Accessed 18 March 2016). Nichols, R. F. W. (1947) Breeding cassava for virus resistance., East African Agricultural Journal, 12, pp. 184-94, [online] Available at: http://www.cabdirect.org/abstracts/19471600996.html (Accessed 5 February 2014). Park, T. and Casella, G. (2008) The Bayesian Lasso, Journal of the American Statistical Association, 103(482), pp. 681-686. Pà ©rez, J. C., Lenis, J. I., Calle, F., Morante, N., Sà ¡nchez, T., Debouck, D. and Ceballos, H. (2011) Genetic variability of root peel thickness and its influence in extractable starch from cassava (Manihot esculenta Crantz) roots, Plant Breeding, 130(6), pp. 688-693. Pà ©rez, P. and De Los Campos, G. (2014)