Cassava (Manihot esculenta Crantz) is a vital tropical food crop, ranking third in caloric contribution after maize and rice. Its large starchy roots and edible leaves serve as a primary food source for more than 800 million people, primarily in sub-Saharan Africa (Nassar & Ortiz, 2010).

Over the past 90 years, cassava production has faced escalating threats from biotic and abiotic stresses (Tomlinson et al., 2018) exacerbated by climate change (Jarvis et al., 2012). Cassava brown streak disease (CBSD), ranked among the seven most serious threats to world food security (Pennisi, 2010), stands as a significant biotic challenge to cassava production, particularly in East, Central and Southern Africa, with the potential to cause losses of up to 100% in susceptible varieties (Hillocks et al., 2000; Hillocks & Jennings, 2003; Kaweesi et al., 2014). CBSD is caused by a positive sense single-stranded RNA virus belonging to the genus Ipomovirus and family Potyviridae (Winter et al., 2010; Walker et al., 2022). It is caused by two distinct viruses: cassava brown streak virus (CBSV) and Uganda cassava brown streak virus (UCBSV), both collectively referred to as cassava brown streak viruses (CBSVs). Both viruses are transmitted by whiteflies (Bemisa tabaci) in a semi-persistent manner (Maruthi et al., 2005; Mero et al., 2021) and through the movement of infected stem cuttings by farmers, which adds to the complexity of controlling the disease. CBSD symptoms are characterized by the initial emergence of leaf chlorosis along secondary vein margins, which later develop into blotches. Subsequently, brown streaks develop on the stem, accompanied by radial root constrictions and root necrosis (Hillocks & Jennings, 2003; Alicai et al., 2007; Kaweesi et al., 2014). The most destructive symptom is root necrosis, as it makes the cassava tubers unfit for consumption by both humans and animals. Therefore, it is essential to thoroughly investigate CBSD root necrosis to uncover the genetic markers or genetic mechanisms that can be leveraged in breeding for resistance. This can be comprehended by adopting root necrosis image capture and analysis to complement the 1-5 visual severity scores. This incorporation of root necrosis imaging could introduce fresh opportunities for enriching the qualitative 1-5 scoring approach with more precise quantitative analyses.

High-throughput phenotyping based on images holds enormous promise in unraveling the genetic basis of root necrosis. Image analysis enables the dissection of the genetic architecture of root necrosis by harnessing valuable information from variation in root necrosis expression patterns, which are often challenging to characterize with the naked eye (Demidchik et al., 2020). The effectiveness of image analysis for crop phenotyping has been showcased in a variety of species, including beans (Kumar et al., 2013; Sun et al., 2016), rice (Wu et al., 2019), potatoes (Cassells et al., 1999; Siebring et al., 2019). Its success extends to the study of diverse traits, ranging from plant architecture to chlorophyll content (Yadav et al., 2010; Dutta Gupta et al., 2013; Walter et al., 2015). Image analysis in plants has also been expanded to multiple pathosystems, enabling the detection and quantification of plant diseases (Sun et al., 2014; Elliott et al., 2022; Palma et al., 2022; Shi et al., 2023). Success of image analysis relies on a diverse range of image analysis tools, including Image J/Fuji (Schindelin et al., 2012, 2015, 2017), Brushlets (Meyer & Coifman, 1997), PlantCV (Gehan et al., 2017), and many others. Plant Computer Vision (PlantCV) is an open-source, high-throughput image analysis tool that can be used to create customized workflows for segmenting and measuring quantitative characteristics of plants from images (Gehan et al., 2017). PlantCV Version 2.1 is a Python-based package comprised of modular functions, which has remarkable flexibility, usability, and functionality in processing images from multiple platforms including red, green, and blue (RGB), Near-infrared (NIR), PSII fluorescence, thermal, and hyperspectral sensors. PlantCV has found application in the phenotypic characterization of a diversity of plant species including pennycress, Arabidopsis, wheat, teff, rice, common bean, quinoa, and more (Castillo et al., 2022; Knapp et al., 2022; Griffiths et al., 2023; Pierz et al., 2023), underscoring its significance as a valuable tool for quantitative classification of plant phenotypes.

Substantial advancements in next-generation sequencing (NGS) and statistical methodologies have opened new avenues for plant breeders to use state-of-the-art tools and methods to adopt more efficient strategies when developing improved cassava varieties (Wolfe et al., 2016). CBSD root necrosis is a trait in cassava for which the genetic architecture can be explored by leveraging DNA markers spread throughout the entire cassava genome. Integration of DNA markers has gained widespread acceptance in genome-wide association studies (GWAS) and has proven invaluable in revealing the extensive allelic diversity present within natural populations (Wang et al., 2020; Xiao et al., 2022). As a result, GWAS has enabled the identification of genomic regions or QTLs (Quantitative Trait Loci) associated with desirable traits that have been leveraged in implementing marker-assisted selection (MAS) and/or genomic selection (Rabbi et al., 2022). Quantitative root necrosis assessment using image analysis introduces novel traits that provide precise measurements and capture spatial characteristics overlooked by the traditional 1-5 scoring method, which relies on human observation to quantify root necrosis. To implement quantitative root necrosis assessments, Diversity Arrays Technology (DArT) markers, which detect thousands of genetic loci through microarray technology were used to conduct GWAS for root necrosis. These markers rely on hybridization of genomic DNA fragments to microarrays with probes for multiple loci, enabling detection of genetic polymorphisms such as SNPs and indels (Deres and Feyissa, 2023). Given that this marks the first attempt to investigate the genetic architecture of CBSD root necrosis using imaging. There is a necessity to identify genomic regions and alleles linked to these traits, potentially resulting in the discovery of novel genetic loci associated with CBSD resistance. This is because for more than 9 decades, CBSD evaluations have primarily relied on a qualitative 1-5 visual scoring scale (Tomlinson et al., 2018) which hinders the precise quantitative assessment of CBSD root necrosis, thus restricting the potential to reveal functional genomic insights. These insights can play a crucial role in utilizing marker-assisted selection or genomic selection to expedite genetic gains in breeding for CBSD root necrosis resistance.

In this study, we used data from PlantCV and CBSD severity scores to achieve four main objectives: (1) characterizing root necrosis image traits through broad-sense and narrow-sense heritability, genotype by trait (GT), and genotype plus genotype-by-environment interaction (GGE) biplots, (2) determining phenotypic and genetic correlations, (3) identifying genomic regions through univariate and multivariate GWAS analyses, and (4) categorizing candidate genes associated with root necrosis image traits and CBSD severity in the Cycle 2 population of genomic selection.

Using PlantCV to extract root necrosis image traits and the 1-5 visual scoring method for foliar and root severities, we conducted a comprehensive characterization of cassava brown streak disease. Our aim was to identify genomic regions associated with CBSD root necrosis image and severity traits through univariate and multivariate GWAS analyses. Additionally, we investigated the functions of annotated genes associated with these traits. Key traits of interest included solidity of necrosis, convex hull area of root necrosis, ellipse eccentricity of root necrosis, percentage of necrosis, necrotic area fraction, and necrotic width fraction. Broad-sense and narrow-sense heritability estimates are key in providing insights into the genetic gain achievable through selection (Holland et al., 2002; Schmidt et al., 2019). Results showed that broad sense heritability estimates were low to moderate for all the root necrosis images traits while narrow sense heritability estimates were consistently low and ranged from 0.03 to 0.22. The low to moderate broad-sense heritability estimates provide evidence supporting the involvement of diverse levels of genetic factors in the manifestation of these root necrosis mage traits while the narrow sense heritability estimates also provide additional evidence supporting the involvement of diverse levels of additive genetic factors. These heritability estimates indicate some genetic influence, especially for root necrosis traits with moderate estimates. We suggest that random factors may affect the spatial distribution of necrotic lesions, leading to lower heritability estimates. Additionally, we excluded images without root necrosis from our analysis, but these images were given a score of one on a 1-5 scale. This could mean that the set of root necrosis images may not fully represent all the phenotypes, potentially resulting in underestimated heritability for certain image traits compared to the true genetic variation. Broad sense and narrow sense heritability estimates for root necrosis severity scores were similar with those previously reported (Okul Valentor et al., 2018; Ozimati et al., 2019, 2021; Nandudu et al., 2023).

Phenotypic and genetic correlations were analyzed to understand the genetic basis of root necrosis image traits and CBSD severities. We observed varied phenotypic and genetic correlations among root necrosis image traits, including significant positive correlations. Traits such as percentage of necrosis, necrotic width fraction, and necrotic area fraction, exhibited both high phenotypic and genotypic correlations ( Figure 4 ). This means that these traits can be effective mutual predictors for indirect selection when quantifying root necrosis, resulting in cost efficiency, and reduced computational time. Nonetheless, for breeders to effectively utilize highly correlated root necrosis traits in breeding against root necrosis, it is essential to incorporate them into selection indexes (León et al., 2021). This would involve assigning appropriate weights to these traits based on their economic significance and heritability. Such indexes would enable breeders to prioritize and concentrate their efforts on the root necrosis traits that have the most substantial influence on overall crop enhancement. The high positive correlations between root necrosis image traits also signify the existence of common genetic mechanisms (Vattikuti et al., 2012; Sodini et al., 2018; van Rheenen et al., 2019).Correlations between root necrosis traits can be affected by various factors and one influencing factor is pleiotropism and linkage disequilibrium (Walsh & Blows, 2009; van Rheenen et al., 2019) Pleiotropism occurs when a single gene has an impact on multiple traits, leading to correlations between these traits. Another factor is linkage disequilibrium, a phenomenon described by Walsh and Blows (2009). In linkage disequilibrium, certain genetic variants tend to be inherited together, creating dependencies between traits. Both pleiotropism and linkage disequilibrium contribute to the complex relationships observed among different traits, highlighting the intricate nature of genetic interactions and their effects on trait correlations.

Incorporating genomic data into the analysis pipeline constitutes a pivotal strategy for unraveling the genetic foundations of CBSD root necrosis traits, even in scenarios characterized by low to moderate heritability estimates (Shao et al., 1991). We conducted a comprehensive analysis, encompassing both univariate and multivariate GWAS approaches. The univariate GWAS found eight SNP markers on chromosomes 1, 7, and 11, linked to both the ellipse eccentricity of root necrosis and the percentage of root necrosis. The haplotype view ( Supplementary Figure 11 ) highlighted the 6 SNPs identified in chromosome 1 to be in a high LD block region ( Supplementary Figure 11A ). When we zoomed into the high LD block region where the 6 SNPs clustered, three of the SNPs were seen to be in high LD with correlations ranging from 0.9 to 0.97 ( Supplementary Figure 11B ). On average, four SNPs captured the variation explained by the significant SNPs identified on chromosome 1. Despite being significant, these SNPs only accounted for small proportions of phenotypic variance. These small proportions of variance could arise due to the complexity of CBSD root necrosis image traits’ genetic architecture which may arise from the influence of numerous rare variants with minor effects. Furthermore, standard GWAS methods may overlook low-frequency polymorphisms, which could also contribute to CBSD root necrosis image traits. Consequently, the significant SNPs identified in this study might only explain small proportions of phenotypic variance. The limited extent of phenotypic variability also suggests that the underlying causal variants linked to both root necrosis ellipse eccentricity and the percentage of root necrosis, despite their rarity, might reside at substantial distances from the significant SNPs pinpointed in this GWAS (Orozco et al., 2010; Wray et al., 2011). Consequently, the actual effect size could potentially be much more substantial than what is suggested by the significant SNPs identified in this study. To comprehend the impact of multiple rare variants, increasing the population size with diverse individuals is pivotal in enhancing statistical power for their detection. A larger sample size with diverse individuals ensures a more comprehensive representation of rare alleles, facilitating the identification of associations. In our study, the C2 population exhibited moderate genetic diversity ( Figure 7 ). However, our sample size was significantly reduced by excluding root images lacking necrosis, potentially limiting our ability to detect multiple rare variants. Also, to determine causal SNPs linked to CBSD root necrosis, it is advised to establish a sizable biparental cassava population with diverse root necrosis levels for precise genetic mapping (Nzuki et al., 2017; Ferguson et al., 2019). This population should be evaluated in various environments to enhance statistical power. Fine mapping on regions of interest to identify causal variants associated with synthetic associations should be carried out. Also, integrating traditional QTL mapping with association mapping for a thorough understanding of the genetic architecture of CBSD root necrosis traits is key. We highlight that the outcomes of our study relating to root necrosis traits and CBSD severity scores align with earlier GWAS and QTL analyses focused on CBSD severities utilizing the 1-5 scoring method. These genomic regions include (1) chromosome 1 (Nandudu et al., 2023), (2) chromosome 11 (Kayondo et al., 2018), (3) chromosome 11 (Nzuki et al., 2017), chromosome 11 (Masumba et al., 2017), and chromosome 11 (Kawuki et al., 2016). This implies that certain genes or closely linked genes on these chromosomes might play a role in contributing root necrosis image traits and CBSD severity. Notably, such genes may also originate from prominent founders like Namikonga and Kiroba, which have been extensively employed in CBSD resistance breeding across multiple cassava breeding programs. Ultimately, the persistence of these genomic regions across various studies further emphasizes their significance for future research endeavors aimed at comprehending and enhancing resistance against cassava brown streak disease.

A total of 24 potential candidate genes were identified in the 172.2kb region on chromosome 1, with diverse range of functions crucial in the context of CBSD. These functions include ubiquitin-protein ligase activity, potentially involved in degrading viral components; ribosomal proteins, essential for viral replication and translation; transmembrane signal receptors, aiding in the recognition of viral molecules; DNA-binding transcription factors, governing the expression of defense-related genes; and RNA metabolism proteins, influencing viral RNA stability and processing. Molecular characterizations revealed various gene ontologies associated with these candidate genes. Catalytic activity (GO:0003824) indicates their roles in enzymatic reactions that can either facilitate or hinder CBSD viral replication and spread within cassava plants. The identification of transporter activity (GO:0005215) suggests that these genes may be involved in transporting CBSD viral molecules across cellular membranes, potentially increasing the severity of infection. Transcription regulator activity (GO:0140110) highlights the influence of these genes on gene expression regulation during CBSD viral responses. The biological classification of these potential candidate genes also highlighted their involvement in various critical processes critical for plant-virus interactions. These encompass cellular processes (GO:0009987), which are fundamental for viral entry and replication within the host. Metabolic processes (GO:0008152) underscore their roles in managing energy allocation and resources during infection, while biological regulation (GO:0065007) fine-tunes the plant’s responses to viral challenges. Processes related to localization (GO:0051179) are essential for precisely targeting viral components within the cell. Gene ontologies catalytic activity (GO:0003824), transporter activity (GO:0005215), structural molecule activity (GO:0005198) and molecular, cellular processes (GO:0009987), and metabolic processes (GO:0008152) were previously reported (Amuge et al., 2017) and their presence played a crucial role in facilitating susceptibility to CBSD. Therefore, our study sheds light on the complex nature of CBSD, its genetic architecture, and several candidate genes involved in root necrosis and disease severity. These findings have important implications for understanding CBSD and developing strategies for resistance breeding, which is critical for food security in East and Central Africa.

Using PlantCV for root necrosis traits and a 1-5 scoring method for foliar and root severities in CBSD, we aimed to identify genomic regions associated with CBSD root necrosis through univariate and multivariate GWAS analyses. Key traits included solidity, convex hull area, ellipse eccentricity, percentage of necrosis, necrotic area fraction, and necrotic width fraction. Broad-sense heritability was low to moderate, indicating diverse genetic factors, while narrow-sense heritability was consistently low.

Phenotypic and genetic correlations were analyzed, revealing significant positive correlations among root necrosis traits, especially percentage of necrosis, necrotic width fraction, and necrotic area fraction. Integration of genomic data identified 8 significant SNPs associated with ellipse eccentricity and percentage of root necrosis, suggesting potential epistatic interactions with rare variants. A total of 24 candidate genes were found, involved in crucial processes like viral degradation, replication, transcription regulation, and defense-related gene expression. Gene ontologies highlighted their roles in catalytic activity, transporter activity, structural molecule activity, cellular and metabolic processes. These genes also played roles in cellular entry, energy allocation, regulatory responses, localization, homeostasis, reproduction, developmental and multicellular organismal processes, detoxification, and growth. Some of these ontologies were previously linked to CBSD susceptibility.

In conclusion our study has uncovered potential candidate genes linked to CBSD, shedding light on its genetic makeup and the intricacies of the disease. Despite our initial anticipation of enhanced heritability estimates and strong genomic associations through image analysis objectivity compared to the 1-5 scoring method, the results yielded unexpectedly lower outcomes than those achieved with the 1-5 scoring method. These findings hold considerable importance for comprehending CBSD and devising resistance breeding strategies, ultimately contributing to food security in East and Central Africa.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/ Supplementary Material .

LN: Conceptualization, Data curation, Formal Analysis, Investigation, Methodology, Supervision, Visualization, Writing – original draft, Writing – review & editing. CS: Data curation, Formal Analysis, Methodology, Software, Writing – original draft, Writing – review & editing. AO: Formal Analysis, Methodology, Visualization, Writing – original draft, Writing – review & editing. KR: Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – original draft, Writing – review & editing. JJ: Conceptualization, Formal Analysis, Funding acquisition, Methodology, Project administration, Resources, Supervision, Validation, Writing – original draft, Writing – review & editing.