A population genomics approach to unlock the genetic potential of lablab (Lablab purpureus (L.) Sweet), an underutilized tropical forage crop

Abstract

Background

Lablab is one of the conventionally grown multi-purpose crops that originated in Africa. It is an annual or short-lived perennial forage legume which has versatile uses (as a vegetable and dry seeds, as food or feed, or as green manure) but is yet to receive adequate research attention and hence remains underexploited. To develop new and highly productive lablab varieties, using genomics-assisted selection, the present study aimed to identify quantitative trait loci associated with agronomically important traits in lablab and to assess the stability of these traits across two different agro-ecologies in Ethiopia. Here, we resequenced one hundred and forty-two lablab accessions, utilised whole genome genotyping approaches, and conducted multi-locational phenotyping over two years.

Results

The selected lablab accessions displayed significant agro-morphological variation in eight analysed traits, including plant height, total fresh weight, and total dry weight. Furthermore, the agronomic performance of the accessions was significantly different across locations and years, highlighting substantial genotype-by-environment interactions. The population genetic structure of the lablab accessions, based on ~ 500,000 informative single nucleotide polymorphisms (SNPs), revealed an independent domestication pattern for two-seeded and four-seeded lablab accessions. Finally, based on multi-environmental trial data, a genome-wide association study (GWAS) identified useful SNPs and k-mers for key traits, such as plant height and total dry weight.

Conclusions

The publicly available genomic tools and field evaluation data from this study will offer a valuable resource for plant breeders and researchers to inform a new cycle of lablab breeding. With the aid of these tools, the breeding cycle will be significantly reduced and livestock farmers will have access to improved lablab varieties in a shorter time-frame.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12864-024-11104-5.

Introduction

Globally, livestock production provides a prominent source of protein and income, supporting livelihoods, particularly among smallholder farmers of Sub-Saharan Africa (SSA) [1, 2]. In SSA, livestock production and productivity are severely restricted by the scarcity and low nutritional quality of feedstuffs, resulting in poor animal performance and leading to poverty and food/nutrition insecurity [3, 4]. Currently in SSA, feed availability, feedstuff production and utilization practices, forage conservation, improved forage variety development schemes, soil management, use of tools to quantify forage nutritional value, and other aspects of livestock systems are inadequate and need to be upgraded [1, 5, 6]. Most small-scale farmers who produce the majority of the meat and milk in the region rely on rangelands, harvested crop residues and industrial byproducts as their main source of livestock feed, resources that are challenged by their periodic availability and partial suitability for achieving maximum yield from the animals [7, 8]. This challenge is now exacerbated by climate change and dwindling common grazing lands since more land is being allotted for food crops and human settlement [4, 9, 10].

Continued research efforts, particularly focused on developing locally adapted underutilized forage crops known to farmers, could play a critical role in boosting animal performance, and lead to an ample supply of animal-sourced foods and ultimately ensuring food/nutritional security in the region [11]. Lablab (Lablab purpureus (L.) Sweet (also known as hyacinth bean) in the family Fabaceae, is one of a number of multifunctional crops that originated in Africa [12, 19] and is grown in the tropical and sub-tropical regions of Asia and Africa for fodder and food [13–15]. It is principally self-pollinated with a diploid chromosome number, where 2n = 2x = 22 [16]. It is an annual or short-lived perennial multipurpose legume which has versatile uses as food, feed for animals, or as green manure [17–19]. Interestingly, during the Covid-19 pandemic, it was noted that lablab extracts conferred health benefits against viruses such as influenza [20]. Despite all these applications, limited research attention has been given to lablab to date with no genetic improvement plan reported thus far.

Lablab was identified, a decade ago, by the African Orphan Crops Consortium (AOCC) as one of the 101 under-researched but highly nutritious African food/feed crops selected for genome sequencing [21]. A scaffold-level draft genome assembly was produced earlier [22] and, more recently, a chromosome-scale reference genome (~ 0.4 Gbp, with 11 chromosomes assigned as Lp numbers) was published [19]. Exploring the crop’s genetic potential by collecting appropriate germplasm and understanding its diversity status, using genomic tools in combination with phenotyping, is a key approach to fast-track breeding efforts and accelerate the crop’s full domestication [19, 23, 24]. Recently, some agronomic studies have shed light on seed maturity, yield and feed quality traits and intercropping capabilities of lablab [25–28] but there have been few studies on genetic variability and population dynamics [29–32]. For fast-tracking improvement plans, integrated approaches such as phenotyping, genotyping and phenotype-genotype association studies should be carried out and developing genomic tools is an integral part of this endeavor [33–35]. The availability of a high-quality lablab reference genome [19] now enables discovery of genome-wide SNP/Indel markers for targeted marker-assisted breeding.

In this study, a diverse set of lablab accessions was phenotypically evaluated for morphological and agronomic traits across three locations representing two agroecological settings (low and mid altitudes) in Ethiopia, for two consecutive years and genotyped using a whole genome sequencing approach. The main aim was to: (1) assess genome-wide diversity through phenotypic evaluations and a whole genome sequencing approach, (2) conduct a genome-wide association study and identify single nucleotide polymorphism (SNPs)/k-mers linked with important agronomic traits and (3) select high yielding and stable genotypes, across the environments, in order to initiate a new breeding program targeted at improving yield and feed quality traits of this crop. A range of state-of-the-art tools has been employed in this study to significantly advance knowledge of the genetic potential for the lablab crop.

Materials and methods

Description of the field experiment location and planting materials

In this study, a panel of 143 lablab accessions was assembled from collections conserved in the International Livestock Research Institute (ILRI) forage gene bank in Addis Ababa, Ethiopia, and used for field phenotyping and whole genome sequencing (Supplemental Table S1). The selection of these accessions, from a total collection of 340 accessions, was based on passport data and availability of seeds for phenotyping. These materials originated from eight different countries with most accessions having been donated to the genebank from Kenya and Australia. Seven accessions (6528, 6529, 6533, 6534, 6535, 6536 and 7278) in this panel were of unknown origin and were intentionally included to predict their origin based on their genotypic and phenotypic profile.

All of the accessions used in this study were obtained from the ILRI forage genebank, which is maintained as a global resource for public good. All the materials in the genebank were acquired either prior to the enforcement of the Convention on Biological Diversity (CBD) or, if acquired after this time, an understanding that the materials would be held ‘in trust’ and could be made available for any agricultural research, breeding and training purposes was set out under the terms and conditions of the agreement between ILRI and FAO in 1994. For further information see: https://genebank.ilri.org/gringlobal/search. The ILRI forage genebank is also responsible for characterizing the collection and for the preparation and identification of specimens, following FAO standards (10.4060/cc0023en), which are also available as public goods. The genebank uses reference samples as controls during regeneration and conservation processes to ensure the genetic integrity of all its materials. All available information on the accessions in the ILRI collection is recorded in an online platform, which holds information about Plant Genetic Resources for Food and Agriculture conserved in genebanks worldwide, Genesys, see: https://www.genesys-pgr.org/a/v2eKQXkOpY7. For the lablab accessions used in this study, information pertaining to identifier codes, voucher location and collector’s name is included in Supplemental Table S1.

The field phenotyping was conducted at three locations in different research stations in Ethiopia, namely Bishoftu geographically located at 8^o47’ N, 38^o59' E, and at an altitude of 1890 m above sea level (m.a.s.l.), with an Alfisol soil type, average annual rainfall (875mm), temperature (^oC) maximum (25), average (19), and minimum (11); Melkassa (located at 8^o24' N, 39^o21' E and at an altitude of 1550 m.a.s.l., with an Alfisol soil type, annual rainfall (875mm), temperature (^oC) maximum (23), average (21), and minimum (15)); and Mieso (9^o13' N, 40^o44' E, at an altitude of 1394 m.a.s.l., Alfisol soil type, annual rainfall (875mm), temperature (⁰C) maximum (21), average (18), and minimum (14) (Supplemental Figure S1). The day and night lengths (12h each) were the same across the study sites, which therefore did not directly impact flowering time. For each of the two study years (2020–2021), the trials were conducted under rainfed conditions during the months of July to November. The field phenotyping protocols comply with institutional, national, and international guidelines for experimental research on plants. The research adheres to the Convention on the Trade in Endangered Species of Wild Fauna and Flora (CITES).

The field experiment was laid out using a simple lattice design (13*13) with two replications in each location and conducted over the two years. To reduce the effect of soil heterogeneity, twenty-six accessions were duplicated within each replicate as controls. Some of these controls are released varieties from different countries (Supplemental Table 2). The total number of plots in each replicate was 169, partitioned into 13 incomplete blocks. The spacing between plants and rows was 50 cm and 75 cm, respectively. Land preparation, planting, weeding and other management practices were uniformly applied to all plots across locations. Harvesting of total above-ground biomass was carried out at different times, when 50% flowering had been achieved for each accession.

Table 2.

Genetic estimates of lablab accessions based on forage biomass traits evaluated at three different locations of Ethiopia in each of two years

Traits	min	max	mean	σg	σp	GCV%	PCV%	H (b)	GA	GAM
NoDF	52	194	97.2	233.4	341.3	15.7	19.0	0.68	26	26.8
PH	55	235	141.4	933.3	1995.9	21.4	31.2	0.47	43	30.1
FLW	11	467	188.6	6663.9	21,525.5	41	73.8	0.31	93.6	47.1
FSW	20	1043	391.5	37,524.5	93,668.9	48.3	76.2	0.40	252.6	62.9
TFWP	32	1478	583.2	74,763.7	175,706	45.5	69.8	0.43	367.4	61.2
DLW	5	105	43	286.5	1024.3	37.9	71.8	0.28	18.4	41.3
DSW	6	186	72.8	1328.6	5260.8	47.9	95.2	0.25	37.7	49.5
TDW	11	292	117.6	2840.5	8805.6	44	77.5	0.32	62.4	51.5

σg (genotypic variance), σp (phenotypic variance), GCV (genotypic coefficient of variation), PCV (phenotypic coefficient of variation), H (b) (heritability in a broad sense), GA (genetic advance), GAM (genetic advance as a per cent). NoDF (number of days to 50% flowering), PH (plant height), FLW (fresh leaf weight), FSW (fresh stem weight), TFWP (total fresh weight per plant), DLW (dry leaf weight), DSW (dry stem weight), and TDW (total dry weight

Phenotypic data collection

After 50% flowering, eight plants from the central row of each plot were used to record growth and forage biomass yield trait data. Plant height (PH, cm) was measured from the base at ground level to the tip of randomly selected plants. Fresh leaf weight plant⁻¹ (FLW, g), and fresh stem weight plant⁻¹ (FSW, g) were measured idependently after target plants were harvested. Total fresh weight plant⁻¹ (TFWP, g) was calculated as the sum of FLW and FSW. The dry leaf weight plant⁻¹ (DLW, g), dry stem weight plant⁻¹ (DSW, g), and the total dry weight per plant (TDW, g) were measured after oven-drying fresh material at 65 °C for 48h to a constant weight. Number of days to flowering (NoDF) was recorded when an inflorescence was observed in 50% of individuals per plot/accession. The numbers of seeds per pod (NSP) were also recorded, based on those plants for which pods were developing prior to the harvest point being reached. Accession 24778 failed to germinate sufficiently in all locations in both years, and hence it was excluded from phenotypic data analysis.

Phenotypic data analysis

The data were checked for normality using the Shapiro test [36]. The average values for all traits measured for each accession were calculated and normalized, before comparison of variance (Supplemental Figure S2), using the bestNormalize R package [37]. R statistical software [38] was used to assess the genetic variability. The model used was:

$$\mathit{Y}\mathit{i}\mathit{j}\mathit{l}=\mathit{\mu }+\mathit{\tau }\mathit{i}+\mathit{\gamma }\mathit{i}+\mathit{\rho }\mathit{l}(\mathit{j})+\mathit{\epsilon }\mathit{i}\mathit{j}\mathit{l}$$

where τ_i = treatment effect, γ_i = replicate effect, ρl(j) block within replicate effect, and εijl = random error. A significance level of 5% was used for ANOVA and multiple comparison tests. The additive main effects and multiplicative interaction (AMMI) analysis was used to account for the response of genotypes to changing environments using the metan R package v.1.18.0 [39]. Measures of mean variance, phenotypic variance, genotypic variance and coefficients of variation were used to calculate variance components, following the procedure described previously [40]. For cluster and principal component analysis, the optimum cluster number and membership of accessions within lablab clusters was carried out, using the FactomineR R-package and, for visualization, the Factoextra R package was used [41].

DNA isolation and sequencing

Leaf tissues from one-month-old seedlings of the lablab panel accessions were collected (Supplemental Table S1) and genomic DNA isolated, using the procedure described by the Qiagen DNeasy® Plant Mini kit (250) (Qiagen Inc., Valencia, CA). Before library preparation, DNA quality was checked on 1% agarose gels and DNA purity was checked using a Nanophotometer® spectrophotometer (IMPLEN, CA, The USA). DNA concentration was measured using a Qubit® DNA Assay Kit in a Qubit® 2.0 Fluorometer (Life Technologies, CA, USA). High-quality DNA with a minimum concentration of 50 ng/µl was used for whole genome sequencing (WGS). All the phenotypically evaluated accessions were sequenced by Novogene, using an Illumina high-throughput sequencer and a paired-end sequencing strategy after library construction.

DNA variant calling and annotation

Raw reads were trimmed and filtered using the Trimmomatic tool [42] to remove low quality and adaptor sequences ahead of mapping to the reference genome. Cleaned reads were mapped to the lablab reference genome [19] using the Burrows Wheller Aligner (BWA) [43]. Samtools was used to convert the aligned reads in SAM format into BAM format [44], followed by filtering out read duplications using Picard tools (http://broadinstitute.github.io/picard). Variant calling against the reference genome was carried out using the Genome Analysis Toolkit (GATK4 v.4.1) [45]. A merged VCF file, containing variants for all the genotyped accessions, was filtered using BCftools (v 1.8) [46]. The SNP quality filtering retained SNPs that are biallelic, polymorphic, with a read depth above 10 and below 300, and according to mapping quality (GQ > 20), minor allele frequency (MAF ≥ 0.05), and a missing rate below 1%. The variant effect predictor program, SnpEff (v 4.3) [47], was used to assess SNPs in coding sequences which may affect the corresponding gene products.

Sequence data analysis

Lablab genetic diversity estimation

Filtered genome-wide SNPs were used for population structure analysis (maximum likelihood estimation of individual ancestries) using the ADMIXTURE tool (v1.3.0) [48]. ADMIXTURE output results were systematically collated using the R pophelper program (v2.3.1) [49], which permits determination of similarities and differences in the ancestral make-up of each population. ADMIXTURE uses a cross-validation procedure by breaking the samples for each K value (the number of subpopulations that make up the total population) into five equally sized portions to train the model and estimate the allele frequencies, and ultimately predict ancestry assignments of accessions into most probable original populations. Ten-fold cross-validation 'C' scores were computed, ranging from 2 to 10, separately to determine the best fit model for ancestry estimation. A good value of K will exhibit a low cross-validation error compared to other K values. Furthermore, genomic principal component analysis (PCA) was performed to examine the inter-population distribution using the SNPRelate R package (v1.29.0) [50] and the output was plotted using the Plotly R software package [51]. Finally, phylogenetic trees were constructed using both identity by descent (IBD) and Maximum Likelihood Estimation (MLE) methods and visualized using iTOL [52].

Genome-Wide Association Study (GWAS)

For the Genome-Wide Association Study (GWAS), following the exclusion of those accessions that failed to generate enough reads from whole genome sequencing and potential “duplicates” identified during the diversity analysis, 93 accessions were selected from the panel. The exclusion of “duplicate” accessions, from GWAS analysis has minimized the false discovery rate (FDR). Best Linear Unbiased Estimator (BLUE) values for each of the measured traits were used for marker-trait association analysis. A marker-trait association was performed for each trait separately, using multi-locus GWAS algorithms, Fixed and random model Circulating Probability Unification (FarmCPU) [53], the Bayesian-information and Linkage-disequilibrium Iteratively Nested Keyway (BLINK) model [54], the Multiple-Locus Mixed Linear Model (MLMM) [55] were implemented in the Genomic Association and Prediction Integrated Tool (GAPIT) version 3 software package within the R environment [56]. The population structure was accounted for by including 3–5 principal components in the subsequent data analysis. The distribution of observed vs. expected − log10(p) values was visualized using Quantile–Quantile (Q–Q) plots to test the fitness of GWAS models for agronomic traits [55]. A threshold P-value of 0.001 (–log10(p) > 5) was used to declare significant SNPs for GWAS results. k-mer GWAS (kGWAS) analysis was carried out following a protocol described previously [57]; all the sequenced accessions (136) were included in kGWAS as this method is reference free and marker-trait associations based on this method are less impacted by any “duplicated” accessions. In brief, a 51-bp k-mer matrix from Illumina paired raw reads of 136 lablab accessions was created using Jellyfish (v. 2.1.4). Only k-mers present in at least two accessions were retained. Similarly, k-mers present in all but one accession were discarded. Each row in the k-mer matrix represented the sequence of a 51-bp unique k-mer from the entire collection in the first column followed by its presence (1) or absence (0) in each of the accessions in the remaining columns. A “header” file was created following the same order of the accessions in the matrix file. A tab separated file was created with 10⁵ randomly selected k-mers extracted from the matrix to account for population structure. kGWAS was run independently for each of the traits as described by others [57]. Associated k-mers were mapped to the lablab reference [19] to establish their genomic coordinates. kGWAS methods were sourced from https://github.com/wheatgenetics/owwc/tree/master/kGWAS and running scripts with descriptions from https://github.com/quirozcj/kmerGWAS_descriptions.

Results

Field evaluation of a diverse set of lablab accessions across locations and years

The combined analysis of results from the field evaluations indicated significant differences between lablab accessions and locations, for all traits within and between years. There were significant differences between years for leaf and stem fresh and dry weights (Table 1). In addition, the interaction between accession, location and year was significant for all traits indicating that the performance of genotypes was differentially affected by location and year. The mean performance for all traits per accession is presented in Supplemental Table S3. Accessions such as 18617, 21043 and 11631 flowered earliest while accessions 21045, 21083 and 24800 were among the late flowering group. The top biomass yielding accessions in terms of TDW were 6536, 14423 and 14439.

Table 1.

ANOVA summary from combined analysis of eight growth and forage biomass yield traits of lablab accessions evaluated across three locations in Ethiopia in each of two years

Source of variation	NoDF	PH	FLW	FSW	TFWP	DLW	DSW	TDW
Replication	17.67ns	232.9ns	185662ns	173230ns	717791ns	2066.43ns	2169.9ns	8422.5ns
Accession	808.1***	3862.6***	34,853***	168,718***	325,233***	1597.3	7918***	14,487***
Sites	27,806.5***	17,210.9***	211,653***	2,048,028***	3,464,989***	3703.4***	34,647***	50,376***
Year	45356ns	122,296	2,466,608***	13,587,224***	27,713,179***	120,752***	537,936***	1,181,882***
Accession*site	190***	1197***	1197ns	62,217***	92,667*	676ns	3714**	3070ns
Accession*site*year	103***	1113**	18,707***	49483ns	104,575*	806**	5342***	3580*
Error	107.9	1062.6	14,861	56,144	100,942	737.7	3932	5965

NoDF (number of days to 50% flowering), PH (plant height), FLW (fresh leaf weight), FSW (fresh stem weight), TFWP (total fresh weight per plant), DLW (dry leaf weight), DSW (dry stem weight), and TDW (total dry weight). Numerical values indicate mean squares, ns (non-significant)

^* p < 0.05 (5%), ** p < 0.01 (1%), and *** p < 0.01 (0.1%)

In this study, medium to high genetic variability estimation parameters (phenotypic coefficient of variation (PCV) and genotypic coefficient of variation (GCV)) were recorded for all traits, indicating significant genetic variability (Table 2). Generally, the PCV value was considerably greater than the corresponding GCV value for all measured traits. For example, the PCV value was almost two-fold compared to the corresponding GCV value for DLW, DSW and TDW, indicating a major contribution from the environment in the expression of these traits. PCV values ranged from 19% (NoDF) to 95% (DSW), whereas GCV varied from 15.7% (NoDF) to 48.3% (FSW). A moderate PCV value (19%) was observed for NoDF, whereas high PCV values (> 30%) were recorded for the rest of the traits (Table 2). NoDF scored the highest broad-sense heritability (0.68) while DSW the lowest (0.25). Furthermore, the analysis of expected genetic advance as a percentage of the mean (GAM) indicated that FSW could be improved by 62.9% whilst only a 26.8% alteration could be made for NoDF. TDW could also be improved by 51.5% whilst progress of 41.3 and 49.5% could be made in DLW and DSW, respectively.

Principal component and cluster analysis for forage biomass based agro-morphological traits

Principal component analysis (PCA) and cluster analysis were performed to produce a cluster that shows the approximate relationships among a set of accessions and to convert a set of observed traits of possibly correlated variables into a set of values of linearly uncorrelated standardized quantitative variables. In this analysis, divergence was observed based on growth and forage biomass traits which identified the grouping through the measured traits response.

Principal component analysis (PCA) identified two principal components with Eigen values of greater than one which contributed 83.13% of the cumulative explained variation from the first two components (Supplemental Table S4). The first principal component (PC1) accounted for 66.78% of the total explained variance, with significance loadings for NoDF (0.33), PH (0.46), FLW (0.86), FSW (0.95), TFWP (0.97), DLW (0.89), DSW (0.85) and TDW (0.96). All the measured traits in this trial showed a positive correltion with each other. Similarly, the second principal component (PC2) accounted for 16.34% of the total explained variance. A PCA biplot shows the degree of correlation among measured traits and a tight angle between vectors is an indicator of a high and positive correlation (Fig. 1). The strongest correlation was found between TDW and FSW. Likewise, DLW and TFWP were highly correlated. NoDF and PH were the furthest from the rest of the measured traits but still had a positive correlation with the other traits. Based on standardized data from two years of mean values of eight quantitative traits, the cluster analysis revealed four basic clusters with cluster I mainly composed of 2-seeded lablab accessions (Fig. 1B). All accessions identified to be high yielding (TDW/TFWP) were in cluster IV, except for accession 11615 (Supplemental Table S5).

Fig. 1.

Variable correlation plot showing the relationships among eight quantitative traits in 142 lablab accessions (A) and Cluster Analysis of phenotyped lablab accessions based on forage biomass traits (B)

Performance, yield and stability of lablab accessions

The AMMI analysis of variance indicated highly significant (P < 0.01) accession by environment interaction effects for all traits. Genotypic factors accounted for a larger proportion of the treatment mean of squares for PH, NoDF, and TDW. The environment was significant for TFWP, NoDF, DSW, and FSW but not for the other traits (Supplemental Table S6).

Additive main effect and multiplicative interaction stability analysis ranked the accessions based on AMMI stability variance index (ASI) score for TDW (Supplemental Table S7). Relatively low scores in ASI represent stable accessions for that particular trait across tested environments. Hence, in this study, accession 21048 scored the lowest ASI score indicating it was the most stable accession for TDW, whereas accession 18622 was ranked the least stable with the highest ASI score for this trait. The combination of yield and stability rankings (SSI) indicated that accessions such as 14423 have high yield with stable performance across the three locations, in contrast to 6529 which exhibited high mean TDW but showed lower stability across the locations. Therefore, the breeders’ aim of selecting high yielding and stable accessions across multiple environments could be met with priority accessions such as 14423, 14439 and 6536 (Supplemental Table S7).

Whole genome sequencing of lablab accessions

Although all phenotyped lablab accessions were genotyped, seven failed to generate enough sequence reads and hence were excluded from the diversity analysis. More than 2 billion reads were generated for the remaining 136 lablab accessions with an average of ca. 15 million reads/accession (Supplemental Table S1). The average coverage depth of all the BAM files was 7.7x. More than 20 million raw variants (SNPs: ca. 16 million; Indels: ca. 3 million) were detected when the trimmed reads were mapped against the lablab reference genome. Half a million SNPs passed the filtering threshold (Supplemental Figure S3) and were retained for measuring genetic diversity parameters and the association mapping study. Functional annotation using SnpEff revealed that 42% of the SNPs fall in the intergenic region of the lablab genome. Furthermore, 53% of the SNPs were missense, 0.7% were nonsense and the rest were silent polymorphisms. In terms of putative impact, ca. 5% of the variants were predicted to have low to high impact and most of the SNPs (95%) fell under the modifier category (Supplemental Figure S4).

Different numbers of SNPs were detected on different chromosomes, with the highest number of SNPs recorded for Lp03 (78,408) and the lowest for Lp08 (36,384) (Supplemental Table S5; Supplemental Figure S3). This is not unexpected as Lp03 is the largest chromosome in the reference genome [19]. Interestingly, the highest SNP density rate was recorded for Lp10 and the lowest for Lp07 (Supplemental Table S8). On average, a SNP was detected at ca. 800 bases for all chromosomes. Chromosomes Lp05, Lp07 and Lp11 scored the highest minor allele frequency (MAF, 0.24) and the lowest was recorded for Lp09 (Supplemental Table S8). A similar trend among the eleven lablab chromosomes was also observed for polymorphic information content (PIC, Supplemental Table S8). A rarefaction curve indicated that 95% unique SNPs were detected using less than half of the panel (ca. 60 samples) (Fig. 2).

Fig. 2.

Genic SNPs rarefaction curve per 100 replicates

Population structure and clustering among genotyped lablab accessions

We utilized a clustering approach based on the maximum likelihood estimation (ADMIXTURE software), using half a million filtered SNPs distributed over the 11 lablab chromosomes, to detect underlying ancestral populations of the 136 lablab accessions. The ADMIXTURE simulation demonstrated that a k value of 8 exhibited lowest cross validation error, inferring that the 136 analyzed accessions belong to eight underlying subpopulations with the highest likelihood (Supplemental Figure S5). This predicted population structure displayed membership of some accessions in more than one population, whereas others showed membership of only one population. For example, 21086, 21056, 21081 and 24796 grouped in a single cluster using k = 8. On the other hand, k = 2 with the highest cross validation error clustered the accessions into two clear distinct groups. The PCA analysis showed a similar trend where PC1 separated the panel into two groups (Fig. 3). A phylogenetic tree generated by hierarchical clustering provided corroborating evidence to support the ADMIXTURE analysis in that there was no clear segregation of accessions based on geographical origin (Supplemental Figure S6).

Fig. 3.

PCA plots of the first two principal components of 136 sequenced lablab accessions

Several accessions that were a member of a single cluster in the ADMIXTURE analysis also clustered similarly in the phylogenetic tree. Core hunter analysis with ca. half a million SNPs predicted that 27 accessions can represent the diversity among all the 136 accessions (Supplemental Table S9). Core Hunter analysis was further used to predict different numbers of core accessions that best represent the panel in this study (Supplemental Table S9). From the seven accessions of unknown origin, three (6529, 6533 and 6535) clustered together in the phylogenetic tree and showed single cluster membership in ADMIXTURE analysis, indicating that they are highly related, but the rest were clustered into different clusters. The accessions 6533 and 6535 also clustered together in the phylogenetic tree, based on agro-morphological traits.

Principal component analysis based on SNP marker set

The PCA revealed an interesting trend with two major clusters and a scattered set of materials, presumably landraces or wild materials but no clear separation of accessions based on region of origin (Fig. 3). The accessions in this panel were obtained from eight donor countries, but the PCA did not reveal ecotyping based on their origin. Nonetheless, each one of 127 accessions has a unique set of SNPs that can be utilized to develop molecular barcodes for precise identification in the future (Supplemental Figure S7).

Genome Wide Association Study (GWAS)

A GWAS was carried out to identify genetic variants (SNPs and k-mers) significantly associated with agro-morphological traits measured in the present study. In the SNP- based GWAS, using 93 lablab accessions, all traits showed significant associations higher than a − log10(p) value of 6 for at least one genomic location and with more than one GAPIT model (Fig. 4, Supplemental Table S10). Likewise, in the k-mer based GWAS, using 136 lablab accessions, strongly associated k-mers were identified for all traits, except for PH (Supplemental Table S11). Interestingly, nearly 200 k-mers were significantly associated with DLW (> − log10(p) value of 6) which was a significantly higher number of k-mers than recorded for other traits. Some k-mers identified were negatively associated with target traits, implying that these markers contribute negatively to the trait; for example, all significantly associated k-mers for NoDF were associated with early flowering, i.e., displayed negative correlation with this trait. Interestingly, the only k-mer significantly associated with FSW was negatively correlated with this trait (Supplemental Table S11).

Fig. 4.

GWAS Manhattan plot SNPs identifying QTLs associated with eight agro-morphological traits evaluated over two years for 93 diverse lablab accessions. A DLW (dry leaf weight), B DSW (dry stem weight) C FLW (fresh leaf weight), D FSW (fresh stem weight) E NoDF (number of days to 50% flowering), F PH (plant height), G TFWP (total fresh weight per plant), and H TDW (total dry weight)

Discussion

In this study, a panel of 142 lablab accessions, conserved by the ILRI genebank and donated by eight countries, was phenotyped across three different locations in Ethiopia in each of two years, revealing different levels of variability and heritability among agronomic performance traits. The same accessions were genotyped using a whole genome sequencing approach that generated high-density genome-wide markers which were used for diversity assessment and marker-trait association studies.

Field characterization of lablab accessions across locations and years

Early flowering is a key trait, particularly in lowland areas such as Meiso (one of the research stations in this study) where there is a short rainfall span and accessions such as 18600, 14429 and 24768 with medium NoDF and average forage yield can be selected for such localities. A number of plants, including legume species, are known to deploy drought escape mechanisms, i.e., completing their life cycle quickly in order to avoid water stress during the later growth stages [58, 59]; hence early maturing accessions identified in this study could be useful candidates for areas with short rainy periods. In the present study, the difference between PCV and their respective GCV values was relatively small for NoDF and PH, indicating that these traits were less influenced by the environment and potentially more amenable to improvement through genotypic selection. Furthermore, NoDF and PH showed the highest broad-sense heritability in the present study which implies that the behaviour of these traits was principally influenced by additive gene interactions and hence selection will be efficient for improvement of these traits. However, broad sense heritability (H) includes both additive and non-additive components of the variation and hence does not always provide a full indication of genetic gains that can be made through selection [60, 61]. Therefore, it is necessary to estimate the genetic advance as a percentage of the mean (GAM) to determine the potential progress. In this study, NoDF and PH scored 26.8% and 30.1% GAM respectively which indicated that progress could be achieved through selection of early or late maturing accessions in the next generation.

The additive main effects and multiplicative interaction (AMMI) models are the most suitable in studies where interaction between the environment and genotypes is significant such as in the present study [62]. Stable genotypes normally have low AMMI stability index (ASI) scores while unstable ones have high ASI scores [63]. However, stability in performance for yield or other traits does not necessarily warrant selection as in some scenarios the most stable genotypes do not always have the best performance and vice versa [64, 65]. AMMI analysis identified accessions 14423 and 14440 as two of the highest yielding accessions in terms of TDW with relatively low ASI scores, which implies that more of an additive genetic effect is the major driving factor (Supplemental Table S7). Subsequently, we have selected high-ranking accessions, based on TDW/TFWP (Supplemental Table S5) and these accessions are currently undergoing region-wide field evaluation tests, across East Africa, leading to the possibility of variety releases in the coming few years. Accession 6536, a wild accession, scored the second highest TDW and relatively good stability score. This accession and a few others performed well in terms of total dry weight (TDW), compared to released varieties and could be potential parental lines for a new breeding program aimed at improving yield traits (Supplemental Table S5). A number of field characterization studies have been conducted on lablab in Africa, Asia and USA [66–70], but direct comparisons with the present study are not possible due to differences in the accessions used and traits scored. Regardless, those evaluations, together with the present study, shed light on the global lablab genetic diversity and permit the identification of elite breeding lines.

Genomic tools for lablab improvement

Genomic tools are scarce among key tropical forage species such as lablab and, in the current study, a global panel of 136 lablab accessions was genotyped at the whole-genome level where, using the recently released reference genome, millions of SNPs were detected. On average a SNP was identified every 800 bases for all chromosomes (Supplemental Table S8) which is a much higher SNP density than that reported for lablab recently (one GBS-based SNP markers per ca. 100Kb) [19, 28]. Despite the high number of genome-wide SNP markers identified in the present study, the rate of unique SNPs identified decays with half of the genotyped accessions (Fig. 2) which implies possible duplications in our panel. Our findings, based on Core-Hunter analysis, that 27 accessions (ca. 20% of the total analysed) can represent the overall SNP diversity in the panel (Fig. 2) is similar to that reported previously, where a lablab core collection analysis revealed that 47 accessions (ca. 10%) represented a panel of nearly 500 lablab accessions [12]. Furthermore, the genomic tools developed in this study have facilitated the identification of unique SNP/s for 93% of genotyped accessions, which can be developed as highly valuable molecular barcodes in future breeding programs, in tracking the development of breeders’ materials, and in genebank management (Supplemental Figure S7).

Genetic diversity among lablab accessions

Germplasm identification and genetic diversity studies provide an opportunity for plant breeders to develop improved cultivars with farmer’s desirable traits such as biomass yield and with resilience against biotic and abiotic stresses [71]. A previous genetic diversity study on lablab, using a genotyping-by-sequencing approach, which included a significant number of the accessions in the panel used in the present study, predicted two main sub-populations composed of mainly 2-seeded and 4-seeded accessions [19]. In the present study, a similar output was found, where one cluster was composed of five accessions, of which two were 2-seeded (Supplemental Figure S6). On the other hand, two 2-seeded lablab accessions (14440, 21045), L. purpureus ssp. uncinatus, uncharacteristically clustered with 4-seeded accessions, in sharp contrast with what has been reported previously [19, 72]; this finding needs further investigation. However, phylogenetic analysis based on agro-morphological data clustered all 2-seeded accessions into cluster I (Fig. 1) with only two 4-seeded accessions (24783 and 21048) included as indicated previously [18]. Interestingly 24783, is a 4-seeded lablab accession which clustered with the 2-seeded accessions in both SNP- and morphological-based phylogenetic trees (Fig. 1), whereas 21048 was not successfully genotyped. Accession 21048 is classified as L. purpureus ssp. uncinatus which could partly explains its segregation from the rest of the 4-seeded accessions; Kongjaimun et al. [12] reported that some of the 4-seeded lablab accessions are closely related to wild 2-seeded accessions, which could explain the behaviour of accession 21048 in the present study.

Novel genomic resources to identify candidate loci in lablab

Genome wide association studies (GWAS) are common in the NGS era in order to identify linkage disequilibrium between phenotypic traits and genetic markers (SNPs/Indels) without any prior knowledge of population structure or kinship matrix of a panel [73]. Previously, a GWAS analysis in lablab has identified quantitative trait loci (QTL) for plant height (PH) in chromosomes Lp03 and Lp04 [19]; the present study has identified additional QTL on the same chromosomes and further loci on chromosomes Lp06, Lp07 and Lp08 (Fig. 4). PH is one of the key traits determining forage quality, yield, and management strategy (harvesting and/or grazing) [74, 75]. In the field evaluations, PH showed the highest broad-sense heritability (Table 2), which implies potential for improvement and the QTL identified for PH in this study can play a significant role in future selection efforts. Floral induction is triggered by environmental cues such as day length and temperature and genetic factors associated with this trait are key to determining biomass yield, feed quality and response to water stress [76–78]. In this study, two QTL, on chromosomes Lp01 and Lp09, were significantly associated with NoDF (Supplemental Table S10). A separate study [19] reported QTL for NoDF on a different chromosome (Lp03), suggesting that different regions of the genome impact this trait. Similarly, in legumes such as faba bean, chickpea, soybean, and pea, where significant QTL governing flowering time are located on multiple chromosomes [79–82].

When whole genome sequence data are available, such as in this study, k-mer-based GWAS has allowed identification of variation that cannot be covered with the more commonly used SNP-based GWAS [83–85]. Overall, k-mer-based analyses allow the detection of k-mers originating from regions that are absent from the reference genome, such as large insertions or deletions, and also help minimize false positive associations arising from errors during variant calling and/or genotyping steps [84]. In this study, k-mers significantly associated with all target traits except for PH were detected (Supplemental Table S11) and, to our knowledge, this is the first report of a k-mer-based association study for a tropical forage crop. These markers add impetus to the SNP-based marker-trait associations identified in this and previous studies, with the potential to greatly accelerate genome-assisted breeding in lablab. In summary, valuable genetic associations were identified for most key plant traits measured, with both k-mer and SNP based GWAS, which will enable fast-track selection of these traits in future breeding programs for lablab.

Conclusions

Lablab is an underexploited multipurpose crop, mainly used as animal feed and for soil conservation purposes in the tropics. Our work presents a comprehensive analysis of a diverse panel of lablab accessions, grown under different environmental conditions, leading to the identification of best-performing accessions for high fresh and dry forage biomass production, under low to medium precipitation environments. Furthermore, the genomic tools we have developed provide an unprecedented public resource for lablab genomic research and conservation efforts. This resource offers the potential to improve the overall efficiency and effectiveness of lablab breeding programs via marker-assisted selection (MAS) and/or genomic selection (GS) approaches. The methodologies employed here can be adapted to the genetic improvement of other underutilized tropical forage/food crops.

Abbreviations

AMMI

Additive main effects and multiplicative interaction

ANOVA

Analysis of variance

AOCC

African Orphan Crops Consortium

ASI

AMMI stability variance index

BLINK

Bayesian-information and Linkage-disequilibrium Iteratively Nested Keyway

BWA

Burrows Wheller Aligner

FarmCPU

Fixed and random model Circulating Probability Unification

FDR

False Discovery Rate

GAM

Genetic Advance as a Percentage of the Mean

GAPIT

Genomic Association and Prediction Integrated Tool

Gbp

Giga base pair

GCV

Genotypic Coefficient of Variation

GWAS

Genome-Wide Association Study

IBD

Identity by Descent

ILRI

International Livestock Research Institute

kGWAS

K-mer Genome-Wide Association Study

MAF

Minor Allele Frequency

MLE

Maximum Likelihood Estimation

MLMM

Multiple-Locus Mixed Linear Model

mm

Millimeter

NGS

Next-Generation Sequencing

^oC

Degree centigrade

PCA

Principal component analysis

PCV

Phenotypic Coefficient of Variation

Q–Q

Quantile–Quantile

QTL

Quantitative Trait Loci

SNPs

Single Nucleotide Polymorphisms

SSA

Sub-Saharan Africa

WGS

Whole Genome Sequencing

Authors’ contributions

TA, CD, MA and CSJ designed and supervised the project. Funding was acquired by TA, MA, CSJ and CD. LH collected leaf samples and extracted DNA. MA, LH, TA, MS and HE involved in phenotype data collection and supervision of the field trials. CJ involved in the supervision of the genotyping. TA, LH, QCJ and CJ participated in phenotypic and genotypic data summary. TA, LH, CD and CSJ wrote, proof-read, edited and revised the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.

Funding

The research was supported by the Royal Scoiety (RS) The Future Leaders – African Independent Research (FLAIR) Fellowships collaboration grant (FCG\R1\211019) and the CGIAR Research Program (CRP) on Livestock; Sustainable Animal Productivity for Livelihoods, Nutrition and Gender inclusion initiative (SAPLING), CGIAR research is supported by contributions to the CGIAR Trust Fund. CD gratefully acknowledges funding from UKRI for the Institute Strategic Programme grant (BBS/E/J/000PR9799).

Data availability

The raw sequence data generated and presented in this study are deposited in the European Nucleotide Archive (ENA) repository, accession number (PRJEB70881).

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.