Groundnut [Arachis hypogaea L. (2n = 4x = 40)] is an allotetraploid originating from South America and is a grain legume that is largely grown in the tropical and subtropical regions of the world (Bertioli et al., 2016). It comprises of two genomes with their origins coming from different diploid wild ancestors (Bertioli et al., 2016). Groundnut plays a pivotal role in the life of small-holder farmers in Ghana, and it is a suitable vehicle for making improvements in the areas such as poverty alleviation, food and nutritional security (Tyroler, 2018). It is a good source of plant protein for resource poor-households who are unable to buy animal protein. Groundnut also provides vitamins, minerals and unsaturated oil for most Ghanaians (Asibuo et al., 2008). Increase production and consumption of groundnut will reduce the number of the over 800 million people in many developing countries who are chronically hungry as well as the about 2 billion people who suffer micronutrients deficiencies (FAO et al., 2019). Globally, it provided 8, 3 and 2% of oilseed produced, vegetable oil and protein meal consumed, respectively (http://soystats.com/, accessed 03.07.2022).

Despite the importance of groundnut, its cultivation is hindered by numerous biotic and abiotic factors. Early leaf spot (ELS) caused by the fungus Passalora arachidicola (S. Hori) and late leaf spot (LLS) also caused by the fungus Nothopassalora personata (Berk. & Curt.) are the two most destructive groundnut diseases in Ghana with a potential yield loss of ≤ 70% (Naab et al., 2005; Denwar et al., 2021). Chemical control of these diseases is not feasible in farmers field in Ghana due to their inability to afford these chemicals and therefore ends up not controlling the diseases (Nutsugah et al., 2007; Denwar et al., 2021). Farmers often confuse leaf spots severities with maturity indicator further affecting their mitigation measures. Development and cultivation of leaf spot resistant varieties is cheaper and environmentally friendly (Gaikpa et al., 2017). Resistance is largely controlled by polygenes and influenced by genotype (G), environment (E) and their interactions (G×E) (Johnson, 1984; Wiesner-Hanks and Nelson, 2016). This makes the traditionally commonly used screening methods for identifying leaf spot resistant varieties in Ghana difficult due to the partial and polygenic nature of these diseases (Dwivedi et al., 2002; Pasupuleti et al., 2013). In addition, conventional phenotyping procedures are laborious, time-consuming, destructive, subjective, costly, inefficient and lack inter or intra-rater repeatability (Araus et al., 2018; Awada et al., 2018).

To overcome errors and expenses by manual phenotyping, the red-green-blue (RGB)-image method (which is the science of making measurements through the use of an RGB camera), together with conventional and marker-assisted selection (MAS) may overcome the flaws of current breeding methods (Li et al., 2014; Wang et al., 2017; Yang et al., 2020; Sarkar et al., 2021a). The application of the RGB image method for screening has generated much interest in agricultural research because of its importance in crop production (Sie et al., 2022; Chapu et al., 2022). RGB image method is more efficient, offers inter or intra-rater repeatability, is easy to apply, is less expensive, non-destructive, and offers the chance to take multiple measurements on a specific plant due to the non-destructive nature of the technology (Araus et al., 2018; Awada et al., 2018; Gill et al., 2022). Moreover, the application of the RGB image method in phenotyping will allow the screening of a large set of genotypes using a small fraction of the time that would have been used in conventional phenotyping. For instance, previous studies have shown the efficacy of the RGB imaging for assessment of a number of diseases in several crops verticillium wilt (caused by Verticillium dahliae Kleb) in olive (Olea europaea L.) (Sancho-Adamson et al., 2019), yellow rust (Puccinia striiformis f. sp tritici) in wheat (Zaman-Allah et al., 2015; Zhou et al., 2015), and lethal necrosis (caused by a combination of maize chlorotic mottle virus (MCMV) and sugar cane mosaic virus (SCMV) in maize (Zea mays L.)) (Kefauver et al., 2015).

In the last decade, there has been rapid development of next-generation sequencing, high-throughput genotype data together with phenotypic data for utilization to identify marker-trait associations via genome-wide association study (GWAS) (Varshney et al., 2019; Varshney et al., 2020). Compared to linkage mapping, GWAS has emerged as a powerful tool to detect markers (single nucleotide polymorphisms (SNPs)) closely linked to quantitative trait loci (QTL), based on the principle of linkage disequilibrium (LD) between genetic markers and QTL that affect the trait (Geng et al., 2015). By this strategy, Zhang et al. (2020a) detected a total of 46 and 28 QTL for ELS and LLS, respectively, with Efficient Mixed-Model Association eXpedited (EMMAX), while Pandey et al. (2014) detected 6 QTL for ELS and 1 QTL for LLS. In contrast, seven and five major QTL for ELS and LLS have been detected and reported on chromosomes 2 and 3, respectively (Chu et al., 2019). In addition, Shaibu et al. (2021) detected 25 SNPs for ELS in mini-core collection of 168 accessions in Nigeria by efficient mixed-model association (EMMA) and compression mixed linear model (CMLM).

The statistical model adopted is one of the setbacks to the power of detection in GWAS (Gupta et al., 2014; Ibrahim et al., 2020; Yoosefzadeh-Najafabadi et al., 2022). Traditional popular statistical models (single-marker genome-wide scan models), mixed linear model (MLM), and general linear model (GLM), among others, have a number of limitations such as the stringent threshold of significance and mapping power (Wen et al., 2018). To overcome these limitations, several multi-locus models have been developed and utilized for GWAS in several crops (Zhang et al., 2019; Karikari et al., 2020; Berhe et al., 2021; Vikas et al., 2022). Among them include a multi-locus random-SNP-effect mixed linear model (mrMLM) (Wang et al., 2016), a fast mrMLM (FASTmrMLM) (Zhang et al., 2018), a fast mrMLM efficient mixed-model association (FASTmrEMMA) (Wen et al., 2018), polygene-background-control-based least-angle regression plus empirical Bayes (pLARmEB) (Zhang et al., 2017), Kruskal-Wallis test with empirical Bayes under polygenic background control (pKWmEB) (Ren et al., 2018) and integrative sure independence screening expectation maximization Bayesian least absolute shrinkage and selection operator model (ISIS EM-BLASSO) (Tamba et al., 2017).

The multi-locus models have become the state-of-the-art procedure to identify genetic bases for complex traits due to their power of detection and robustness (Zhang et al., 2019). Therefore, the present study applied the six multi-locus models to identify genomic regions and potential candidates associated with ELS and LLS diseases. A total of 294 groundnut accessions were collected from different African countries and screened in two years with manual scoring of ELS and LLS together with 5 imaged-based indices and area under disease progression curve (AUDPC) for both diseases. The results from this study lay the foundation for MAS to speed up breeding for leaf spot-resistant cultivars.

Leaf spots (ELS and LLS) diseases are the two most destructive groundnut diseases in Ghana resulting in ≤ 70% yield losses which is controlled largely by chemical method (Naab et al., 2005; Denwar et al., 2021). To speed up breeding efforts, the present study was undertaken to identify significantly associated molecular markers and putative candidate genes linked to ELS and LLS diseases’ indicators. Irrespective of the breeding strategy, germplasm serves as lifeblood for breeding effort (Acquaah, 2009; Allier et al., 2020). With this, the present study utilized 294 groundnut germplasm assembled from nine African countries (Ghana, Malawi, Mali, Mozambique, Niger, Senegal, Togo, Uganda and Zambia) ( Supplementary Table 1 ) to assess their response to leaf spot diseases in 2020 and 2021. These germplasm exhibited wide range of responses to leaf spot diseases (resistance to susceptibility) ( Table 1 ; Figure 1 ) of which a portion was recently published by Sie et al. (2022). This suggest that the germplasm hold a promise for breeding groundnut cultivars resistant to leaf spot diseases as well as other demand driven traits. In addition, the high H² ( Table 1 ) suggest that these ELS and LLS as well as other indicators used are largely controlled by polygenes with both major minor effects. This highlights that selection based only on phenotypic variation may be misleading.

One of setbacks in phenotyping large germplasm is the laborious nature coupled with human error among others (Sandhu et al., 2022), which have necessitated a number of phenotyping strategies/platforms which complement the conventional phenotyping. Among these include multispectral imaging (Kobayashi et al., 2001; Chang et al., 2021), RGB imaging (Duan et al., 2018) and others (see Sandhu et al. (2022) for more). The present study manually scored leaf spot incidences and these scores were converted to quantitative scores by AUDPC which were correlated by r=0.47-0.76, giving credence to our scorings ( Supplementary Figure 1 ). In addition, we used four RGB imaging indices (Hue, GA, GGA and CSI) and one vegetation index (NDVI) to confirm both the manual scoring of ELS and LLS as well as ELS- and LLS-AUDPC. It was observed that CSI positively correlated with ELS, LLS, ELS-AUDPC and LLS-AUDPC with r=0.44-0.71 ( Supplementary Figure 1 ), thus the higher the CSI, the more developed the leaf spot diseases, giving an indication that CSI could be used as a screening criterion for leaf spot diseases in groundnut. The predominance and distribution of leaf spots vary according to regions. In most cases there is dual occurrence but the predominance towards physiological maturities varies. For instance, in Eastern Africa, LLS predominates, whereas in Western and central Africa, ELS predominates. However, in the current study, CSI could not distinguish between ELS and LLS, hence further study is needed to develop a model or phenotyping platform to create the distinction.

Recent advances in plant phenotyping involve the use of unmanned aerial vehicles (UAVs) to collect several images generating large amounts of data. Several studies have reported that UAVs are faster and more effective for phenotyping large populations for traits such as height and drought tolerance in groundnut breeding (Sarkar et al., 2021b; Chapu et al., 2022), hence providing the desired high-throughput. This study therefore lays the foundation for investment in such more advanced equipment in groundnut breeding for selection for resistance to late leaf spot and groundnut rosette disease which are the most important foliar diseases in SSA (Chapu et al., 2022).

Leaf spot diseases are well known to affect photosynthetic capacity of leaves (Singh et al., 2011). Three image-based indices, i.e., GA, GGA, Hue and one vegetation index, i.e., NDVI negatively correlated weakly to strongly with ELS, LLS, ELS-AUDPC and LLS-AUDPC ( Supplementary Figure 1 ). However, GA, GGA and Hue seem more better reflective of leaf spot diseases based on our correlation analysis ( Supplementary Figure 1 ) According to Wang et al. (2017), NDVI is a proxy of green biomass, which is linked to canopy photosynthesis. Pronounced changes take place in the visible portion of the electromagnetic spectrum due to the effect caused by a disease or physiological stress to the reflectance properties of the vegetation. Healthy plants absorb both the red and blue light of the electromagnetic spectrum, whiles reflecting near-infrared and some part of the green light (Knipling, 1970). Less of the red light is absorbed by a stressed plant as well as reflecting less of the near-infrared light. Wu et al. (2014) have provided evidence that healthy and unhealthy plants differ in their absorption and reflection of visible and near-infrared lights.

The advancement in high-throughput genotyping, next generation sequencing, bioinformatics tools, statistical models, etc., have served as catalyst to access valuable information from genomic databases and a large number of germplasm, allowing effective harnessing of genetic diversity of a crop (Vikas et al., 2022). Such diversity is vital for broadening the genetic base, as it increases the probability of identifying more unique genes for which two parents have different alleles (Mascher et al., 2019). The 294 germplasm optimally divided into 2 sub-populations/clusters/clade ( Figures 2A-E ). This information is not only useful for the utilization of the 294 germplasm, but also provide valuable information in their conservation as the cost of maintenance and uncertainty about their genetic similarity and dissimilarity (Wambugu et al., 2018; Mascher et al., 2019). The results from this study could contribute to tracking the identity of accessions, avoiding unnecessary duplications within and between genebanks and breeding programmes, while maintaining the genetic integrity of accessions (Mascher et al., 2019). The germplasm in sub-populations/clusters/clade did not follow strictly to a country/sub-region ( Table 3 ; Supplementary Table 1 ), suggesting that some germplasm may have a duplicated version(s) in other country(ies) breeding programmes.

A number of studies by linkage and association mappings have been conducted on leaf spot diseases in groundnut (Pasupuleti et al., 2013; Pandey et al., 2014; Zhang et al., 2020a; Shaibu et al., 2021). However, most of the reported genomic regions/markers are not consistent due to population specific or environment sensitivity (Patil et al., 2018). This necessitated the present study in our attempt to identify potential markers for marker-assisted selection (MAS) (Barmukh et al., 2022). To increase the chances of detecting more possible SNPs that could be associated with the various indices used to assess leaf spot diseases, we applied 6 multi-locus models (mrMLM, FASTmrMLM, FASTmrEMMA, pLARmEB, pKWmEB and ISIS EM-BLASSO) of GWAS. These models differed in their power of detection of associated SNPs, which are in consonance with several earlier studies (Zhang et al., 2019; Karikari et al., 2020). In comparison to study of Zhang et al. (2020a) who reported only SNPs with major effects (PVE ≥ 10%), the present study detected numerous SNPs with both minor (PVE< 10%) and major effects. According to Zhou et al. (2021) SNPs identified by multiple models are usually reliable when several multi-locus models of GWAS are applied on the same dataset. Hence, AX-176801892 (Chr02, LOD= 3.18-3.99; PVE= 0.05-3.49%) linked to ELS, AX-176799357 (Chr04, LOD= 7.73-16.45; PVE= 12.78-20.69%) linked to LLS, AX-176806210 (Chr04, LOD=3.20-9.84; PVE=7.21-38.50%) linked to AUDPC-LLS, AX-147224496 (Chr04, LOD= 4.36-5.46; PVE= 0.01-10.15%) linked to NDVI and several others ( Table 4 ; Supplementary Table 3 ) could be targeted for verification and use for practical plant breeding. The numerous genomic regions/SNPs suggest that leaf spot diseases are regulated by multiple loci with both minor and major effects, hence selection based on only phenotypic data may be misleading (Sood et al., 2020). In addition to the above, a number of SNPs that colocalized with multiple indices assessed in this study could be useful in developing models that could distinguish between ELS and LLS.

One advantage of GWAS is the power to detect SNPs in a narrow genomic regions of which causal variants could result in variation in traits of economic importance such as leaf spot diseases (Jiang et al., 2009; Zhao et al., 2021). However, one of the factors that determine resolution of GWAS is LD (Korte and Farlow, 2013; Ibrahim et al., 2020) which is population specific and influenced by recombination, genetic drift and mating system (Yao et al., 2009). In this study, the LD decay across the two sub-genomes of groundnut was estimated to be about 300 kbp which is nearly 50% higher than the estimation by Zhang et al. (2020a) within 120 kbp in US accessions with 13,382 SNPs. Upon application of LD for the 294 African germplasm, seventeen candidate genes were predicted based on in-silico analyses ( Table 5 ; Figure 4 ). A number of these genes encode for phytohormone/contain CAREs/photosynthesis pathway which could be implicated in groundnut response to leaf spot diseases. For plants to defend themselves from pathogen attack, plants often rely on elaborate signaling networks regulated by phytohormones (Kazan and Lyons, 2014). These genes would be valuable for future functional validation by gene overexpression, CRISPR/Cas9 technology, among others (Zaidi et al., 2020).

In all, ninety-seven SNPs were detected by the six multi-locus GWAS models. Out of which twenty-nine SNPs were detected by at least two models for one or more traits across 16 chromosomes with explained phenotypic variation ranging from 0.01 - 62.76%, with exception of Chr08, Chr10, Chr11, and Chr19. Two hundred and fifty-three genes were located within 300 kbp of 14 SNPs, from seventeen potential candidate genes were predicted. Most of the predicted candidate genes were found to SA, ABA, GA, auxin and MeJA responsive and MBSI CAREs implicated in plant response to biotic stresses. The results from this study would be useful for breeding leaf spots resistance cultivars.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

RO-F, ES, MR, DF, DO, and PO-A conducted the experiments. BK, RO-F, DP, and MBa organized and supervised the overall project. BK, YK, and ES performed the data analyses and wrote the manuscript. RO-F, DO and MBu edited the manuscript. All authors contributed to the article and approved the submitted version.

This study was made possible by the generous support of the BMGF funded Accelerated Varietal Improvement and Seed Systems for Legumes and Cereals in Africa (AVISA) Project Grant Number OPP1198373 and the American people through the United States Agency for International Development (USAID) Cooperative Agreement No. 7200AA 18CA00003 to the University of Georgia as management entity for U.S. Feed the Future Innovation Lab for Peanut (2018–2023). The funders had no role in experiments, data analyses and publication.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.