IIIVmrMLM Provides New Insights into the Genetic Basis of the Agronomic Trait Variation in Chickpea

Abstract

Chickpea is a staple crop for many nations worldwide. Modeling genotype-by-environment interactions and assessing the genotype’s ability to contribute adaptive alleles are crucial for chickpea breeding. In this study, we evaluated 12 agronomically important traits of 159 accessions from the N.I. Vavilov All Russian Institute for Plant Genetic Resources collection. These included 145 landraces and 13 cultivars grown in different climatic conditions in Kuban (45°18′ N and 40°52′ E) in both 2016 and 2022, as well as in Astrakhan (46°06′ N and 48°04′ E) in 2022. Using the IIIVmrMLM model in multi-environmental mode, we identified 161 quantitative trait nucleotides (QTNs) with stable genetic effects across different environments. Furthermore, we have observed 254 QTN-by-environment interactions with distinct environment-specific effects. Notably, five of these interactions manifested large effects, with R² values exceeding 10%, while the highest R² value for stable QTNs was 4.7%. Within the protein-coding genes and their 1 Kb flanking regions, we have discerned 22 QTNs and 45 QTN-by-environment interactions, most likely tagging the candidate causal genes. The landraces obtained from the N.I Vavilov All Russian Institute for Plant Genetic Resources collection exhibit numerous favorable alleles at quantitative trait nucleotide loci, showing stable effects in the Kuban and Astrakhan regions. Additionally, they possessed a significantly higher number of Kuban-specific favorable alleles of the QTN-by-environment interaction loci compared to the Astrakhan-specific ones. The environment-specific alleles found at the QTN-by-environment interaction loci have the potential to enhance chickpea adaptation to specific climatic conditions.

1. Introduction

At present, GWAS (Genome-Wide Association Study) is considered the gold standard for detecting associations between genomic variants and traits [1]. A classical implementation of a single-trait GWAS tests each marker at a time for association with a phenotype [2]. The widespread usage of MLM (mixed linear) models improved the prediction of true associations by removing confounding effects introduced by population structure and accession relatedness [3,4]. The application of these models is hindered by the Bonferroni correction used to correct for multiple testing, proving overly restrictive in identifying certain associations with complex traits in crops [5]. Bonferroni correction leads to a strong overestimation of the type I error, thereby missing true effects, i.e., decreasing the power of the experiment as the number of hypotheses grows [6].

In addressing this issue, the multi-locus MLM models have been developed to test all markers within the frame of one linear model while simultaneously estimating all marker effects [7,8,9]. An important advantage of such models over single-locus GWAS is their ability to detect quantitative trait nucleotides (QTNs) with marginal effects, where the significance threshold set by the Bonferroni correction is too stringent.

Most GWAS models only consider additive marker effects. However, dominance, gene-by-gene, and gene-by-environment interactions also play a crucial role in shaping the genetic architecture of complex traits in plants. Current methods for detecting these interactions are computationally complex and only estimate the effects of allele substitution and allele interaction, considering the specific control of the polygenic background [10,11]. This leads to the inadequate control of the polygenic background and confounding in the estimation of the marker effect.

The IIIVmrMLM model [12,13] has been developed to address methodological challenges in detecting various interactions between alleles, genes, and environments, while also providing an unbiased estimation of their genetic effects. This multi-locus MLM model simultaneously estimates the effects of all genes and interactions, using a computationally less complex approach that involves calculating only three compressed estimates instead of a large number of variance components. Additionally, the IIIVmrMLM model uses the expectation maximization empirical Bayes algorithm to estimate all effects within one multi-locus model, and significant QTNs are further evaluated via likelihood ratio tests. This methodology theoretically ensures the accurate detection of loci and an unbiased estimation of their effects, making IIIVmrMLM a suitable choice for detecting associations between genes, traits, and environments.

As sessile organisms, plants demonstrate remarkable phenotypic plasticity [14]. Both genotype and environment contribute to the phenotypic variation in a trait, and these factors interact at times in complex and non-additive ways [15]. The combination of genes and the environment plays a crucial role in shaping the plant’s response to changes in the environment, particularly in terms of important agricultural traits. Identifying these interactions can help in developing plant varieties that are better equipped to withstand climate changes [16].

Chickpea, the second most extensively cultivated food legume, supplies important nutritional nitrogen and high-caliber protein for roughly 15% of the global population [17,18]. In West Asia and the Indian subcontinent, chickpea stands out as the most widely consumed legume. Its cultivation spans across 50 nations globally, as it has increasingly become a fundamental component of the Mediterranean diet. Currently, the proportion of grain legume crops, such as chickpeas, in EU agricultural regions is minimal, while chickpea production in Russia is on the rise [19,20]. Russia plays a significant role as a major global supplier of chickpeas, accounting for approximately 25% of the global chickpea trade prior to 2022. Furthermore, there has been a notable increase in crop breeding efforts, with six out of the 14 registered varieties in Russia being developed within the past five years.

The application of omics technologies in breeding has proven effective on several crops. However, progress in chickpea genomics has been relatively slow compared to other crop species such as cereals. A shift has occurred in the last decade through the large-scale characterization of germplasm and the construction of a pan-genome [21,22,23]. A combined analysis of the available phenotypic and genotypic data identified candidate markers for many agronomic important traits, including tolerance to abiotic and biotic stresses [21,22,23,24,25,26,27,28,29,30,31]. Breeding strategies based on genomic prediction to enhance crop productivity have been proposed [22,32,33].

Chickpea was often relegated to marginal lands where various abiotic stresses such as water deficits, extreme temperatures, short growing seasons, and poor soils contribute to limited yield potential [34]. For instance, drought decreases chickpea yield in the world by 50%, and losses caused by extreme temperature account for up to 20% [35]. In view of this, the cultivation of highly productive and climate change-resilient chickpea genotypes is essential given the evolving consumer demands, agricultural practices, and the need to adapt to a broader climatic range. The current elite high-yielding chickpea cultivars lack genetic and adaptive variation, highlighting the necessity to broaden the genetic base for continuous variety development. This entails exploring primitive landraces collected prior to the Green Revolution and the application of modern breeding methods to tap into the aforementioned significant additional source of genetic variation.

In the early 20th century, Nikolay Vavilov meticulously gathered chickpea land-races, which are currently preserved at the N.I. Vavilov All-Russian Institute of Plant Genetic Resources (VIR) in St. Petersburg, Russia. Previously, we interrogated these data to reveal marker–trait association using single-trait GWAS [31]. To gain a better understanding of the genetic factors behind variations in agronomic traits in chickpeas, we conducted GWAS using the IIIVmrMLM program. This program enables the detection of both quantitative trait nucleotides (QTNs) and quantitative environment interactions (QTN-by-environment interactions, QEIs).

2. Materials and Methods

2.1. Plant Growing and Phenotyping

A total of 159 chickpea genotypes were specifically chosen from the collection at the N.I. Vavilov All-Russian Institute of Plant Genetic Resources (VIR) in St. Petersburg, Russia (see Table S23). Within this dataset, 145 landraces and 13 elite cultivars were carefully included. These landraces, which were gathered by N.I. Vavilov during his expeditions in the 1920s–1930s, represent a valuable historical and genetic resource. To categorize the samples, six geographic regions were identified based on their geographical proximity: Mediterranean (MED), Lebanon (LEB), South of Russia (RUSS), Turkey (TUR), Uzbekistan (UZB), and India (IND).

Phenotyping of the chickpea accessions was conducted at two OSs of the VIR, the Kuban OS in 2016 and 2022 and the Astrakhan OS in 2022. The Kuban OS is located in the steppe zone of the Kuban–Priazovskaya lowlands, approximately 80 km from the Caucasus foothills. The soil at this location is predominantly black-rich, with a fertile layer depth of 140–150 cm and a slightly alkaline pH. The humus horizon is 130–170 cm thick, with humus content ranging from 3.6% to 4.6%. The climate is characterized as temperate continental, with hot summers, sub-optimal rainfall, and high fluctuation in climatic parameters. The sum of active temperatures above 10 °C ranges from 3200 to 3400 °C, and the average annual precipitation is 565 mm.

The Astrakhan OS is situated in the Caspian lowlands in the southern part of Astrakhan Oblast. The region experiences a continental climate, which is the driest in the European territory of the Russian Federation. It has substantial thermal resources, with a sum of active temperatures above 10 °C ranging from 3000 to 3500 °C. Annual precipitation varies from 180 to 200 mm, with the majority (70–75%) occurring in the warm season. The combination of low precipitation and high temperatures contributes to the dryness of both the air and soil. The predominant soils in the area are brown semi-desert soils with a humus content of 1.1%, light particle size distribution, low soil absorption capacity, and a neutral pH.

The duration of daylight likely had minimal impact on the plant development in the Kuban and Astrakhan regions, as both locations are situated at approximately the same latitude.

Both sites followed similar agronomic practices, with the exception that crops at the Astrakhan OS were irrigated seven times a day. A drip irrigation system was used at the recommended rate for irrigating vegetable crops, with an irrigation norm of 50.4 m³ per season, or 400 m³ per hectare. Sowing took place in late April 2022, with harvesting occurring in late July to early August. At both locations, the accessions were planted in a randomized complete block design with one replicate. Crop maintenance involved manual weeding (four times), mechanized row spacing treatment (two times), and the application of the pesticide “Stomp” (three times at a concentration of 4.5 L/ha). Urea (45% N) was used as fertilizer at a concentration of 800 g/L.

Throughout the vegetative period, we conducted measurements on 12 phenological and morphological traits (Table S1), including plant height (PH), the height of the first pod attachment (HFP), the number of primary branches (NPB), the number of secondary branches (NSB), plant dry weight with pods (PWwP), pod weight per plant (PoW), pod number per plant (PoNP), 100 seed weight (100SW), leaf size (LS), days from emergence to flowering start (DFst), flowering duration (DF), and days from emergence to full maturation (Dmat). Our analysis encompassed six plants for each accession.

2.2. DNA Sequencing and Variant Calling

The DNeasy Plant Mini Kit (Qiagen, Germantown, MD, USA) was used to extract DNA from collected leaves. DNA was sequenced at the BGI (Shenzhen, China) using the Illumina technology, generating paired-end reads of 150 bp. A total of 7700 Gbp of raw data comprising about 26 billon reads with an average of 25× coverage or about 37 Gbp per sample were generated. Reads were processed and aligned to the chickpea reference genome assembly ASM33114v1 with bwa-mem using default parameters [36]. NGSEP [37] version 4.0. was used to call variants. A total of 96,354,236 biallelic SNPs were further filtered with VCF tools [38] to retain SNPs with minor allele frequency (MAF) > 5% and genotype call-rate > 85%. A total of 171,038 SNPs passed all filters and remained for further analysis.

2.3. Genetic Data Analyses

The genetic structure in the dataset was evaluated using the ADMIXTURE software v.1.3.0 [39]. The analyses were performed for K values ranging from 2 to 7. The linkage disequilibrium (LD) decay was estimated using squared Pearson’s correlation coefficient (r2). The PopLDdecay [40] version 3.4.1 was run to calculate r² in a 500 kb window. The LD decay was calculated based on R² and the distance for each pair of SNPs using an R script in accordance with Hill–Weir approximation [41]. We applied the Mann–Whitney–Wilcoxon test [42] to make group comparisons. Two-way analysis of variance was carried out using the R function aov() from library stats.

2.4. GWAS

The genome-wide association analysis was performed using IIIVmrMLM program [13] run in Multi_env mode with parameters svpal = 0.01 and SearchRadius = 20. Suggested QTNs (SUG) were QTNs with LOD ≥ 3.0, significant QTNs (SIG) are QTNs with the Bonferroni corrected p-values calculated from LOD score using χ2 distribution. Candidate genes containing either QTNs or QEIs in gene bodies or within 1 kb flanking regions were annotated using The Pulse Crop Database https://www.pulsedb.org/Analysis/1869759 (accessed on 15 April 2024).

3. Results

3.1. Evaluation of Phenotypes

In 2022, 159 chickpea accessions were grown at two VIR (N.I. Vavilov All-Russian Institute for Plant Genetic Resources) outstations (OSs): the Astrakhan OS (46°06′ N, 48°04′ E, altitude 24 m) and the Kuban OS (45°18′ N, 40°52′ E, altitude 138.9 m). Additionally, the same chickpea accessions were also grown at the Kuban OS in 2016.

The hottest agricultural year across environments was at the Astrakhan OS in 2022 (Table S2). The sum of active temperatures above 10 °C was 3419. This exceeds the values at the Kuban OS in 2016 and 2022 by 346 °C and 525 °C, correspondingly (Tables S3 and S4). The wettest agricultural year across environments was at the Kuban OS in 2022, with annual precipitation equal to 744 mm, which exceeds the values at the same station in 2016 by 128 mm and at the Astrakhan OS by 547 mm.

In every year and at each location, all accessions were evaluated for 12 traits related to yield, vegetative growth and flowering time (Table S1, Figure 1a). In both regions, yield-related traits, i.e., plant dry weight with pods (PWwP), the weight of pods per plant (PoW), the number of pods per plant (PoNP), and a 100 seed weight (100SW) have the largest coefficient of variation (see Table S5). The variability of most other traits in the dataset is also large, ranging from 20 to 30%. The values for traits related to yield and vegetative growth, except for the height of the first pod attachment (HFP) and 100SW, are significantly different between environments (p value < 0.05). Specifically, a similar number of days was required for plants to start flowering in 2022, but not in 2016 (see Table S5). However, at the Astrakhan OS, the flowering phenophase lasted longer than at the Kuban OS. The period from emergence to full maturation was also much longer at the Kuban OS than at the Astrakhan OS (refer to Figure 1a).

In both Astrakhan and Kuban regions, the traits PoNP, PoW, and PWwP are correlated, as shown in Figure 1b and Table S6. However, these traits are not correlated across different environments, indicating a significant influence of the growing conditions on the phenotype. On the other hand, 100SW and HFP show correlation across different environments, suggesting that the growing conditions have minimal impact on these traits. In each environment, PoNP, PoW, and PWwP demonstrate a correlation with NSB, which is a major contributor to plant yield. Notably, in the data collected from Astrakhan and Kuban in 2016, both the time from emergence to flowering (DFst) and the flowering phase (DF) are significantly correlated with the time from germination to full maturity (Dmat). However, this correlation is not observed in accessions grown in Kuban in 2022, indicating a strong environmental effect on these traits (Table S6).

The effects of genotype, environment and genotype-by-environment interaction were statistically significant for traits related to yield and vegetative growth (Table S7). The estimates for gene-by-environment interaction were highly significant, suggesting that genotype performance should be assessed in each specific environment. In the case of flowering traits and LS, all plants belonging to the same accession exhibit the same trait value precluding the estimation of the genotype-by-environment interaction effects. However, the effects of genotype and environment are statistically significant for all these traits (Table S7).

3.2. Population Analysis

The genetic makeup of the dataset was determined using a group of 171,038 SNPs. In tune with previous findings, the principal component analysis did not show any distinct separation among the samples based on their geographical origin [31,43]. The lowest cross-validation error in the ADMIXTURE analysis occurred when the number of populations was set to three (K = 3). However, the errors at K = 4 and K = 5 were only slightly larger, suggesting that these three population splits are the most preferable (see Figure 2a,c,d). The ADMIXTURE analysis shows that accessions can be divided into six almost geographically isolated groups: Indian (IND), Turkish (TUR), Mediterranean (MED), Lebanese (LEB), Uzbek (UZB), and South Russian (RUS). Further examination reveals that the ADMIXTURE patterns of geographically adjacent populations (Indian and Uzbek, as well as Turkish and Mediterranean) are more similar than the patterns of populations located farther away. This likely reflects the history of chickpea migration after domestication [36].

LD decays fast in the dataset, reaching half of a maximum r² value at a distance of 50 kb (Figure 2b). Of note, the LD decay observed in this study was much faster than those detected for Desi and Kabuli chickpea cultivars in other datasets (340 kb and 330 kb correspondingly), as well as for the cultivated soybean (150 kb) [22,44].

3.3. Identification of QTNs and QEIs Associated with Phenotypic Traits

The IIIVmrMLM method allows us to calculate the QTNs and QEIs separately, thus dividing markers into groups, namely (a) markers with stable effects across different environments, and (b) markers associated with phenotypic effects in selected environments only.

We have identified 161 QTNs and 254 QEIs as the significant markers for 12 different phenotypic traits (refer to Tables S8 and S9). On average, QTNs accounted for approximately 19.6% of the variation across the different traits, while QEIs explained an average of 48.6% of the variation, ranging from 27.6% to 63% for different traits (see

Table 1. Percentage of trait variation explained by QTNs and QEIs. * Abbreviations of trait names are as shown in Table S1 and Table A1.

Trait Name *	Type	Total Variation Explained (%)	Marker Explaining the Largest Variation in a Trait	R2 (%)
DF	QTN	15.9	Ca5_21845257	1.9
	QEI	43.5	Ca2_495917	2.2
DFst	QTN	26.4	Ca7_22036380	4.2
	QEI	38.1	Ca7_18941640	6.0
Dmat	QTN	13.6	Ca3_17850594	2.5
	QEI	50.8	Ca4_22649405	5.0
LS	QTN	25.7	Ca8_12966424	3.3
	QEI	40.3	Ca2_27433383	17.2
NPB	QTN	19.0	Ca6_47226273	2.8
	QEI	48.8	Ca7_20175638	2.8
NSB	QTN	14.3	Ca6_31708450	1.5
	QEI	63.0	Ca4_29141628	14.8
PH	QTN	26.7	Ca4_22648344	4.6
	QEI	27.6	Ca4_6063638	8.0
HFP	QTN	24.4	Ca4_27164303	3.8
	QEI	48.3	Ca7_14064521	27.3
PWwP	QTN	15.2	Ca5_44207860	2.1
	QEI	55.2	Ca3_19996983	21.6
PoW	QTN	19.0	Ca4_25847872	3.1
	QEI	61.0	Ca5_21206336	11.4
PoNP	QTN	16.3	Ca4_43416762	2.2
	QEI	52.7	Ca3_27518251	5.4
100SW	QTN	18.2	Ca7_32823511	2.3
	QEI	54.2	Ca4_33626689	4.6

). This indicates that the independent quantitative effects of markers were generally lower than the effects of their interaction with the environment.

The QTN Ca4_22648344 linked to plant height explained 4.6% of the variation, while Ca7_14064521 associated with the HFP trait explained the largest percentage of variation at 27.3%. Likewise, four other QEIs, namely Ca2_27433383, Ca4_29141628, Ca5_21206336 and Ca3_19996983 associated with leaf size (LS), number of secondary branches (NSB), pod weight per plant (PoW) and plant weight with pods (PWwP), respectively, also explained more than 10% of the trait variation (Table 1).

Quantitative trait loci are primarily located on chromosomes 1, 4, 5, and 6 (Figure 3a). QTNs associated with traits PoW, NPB, DF, and Dmat, which showed the most variation between environments, were identified on chromosomes 4, 5, and 6 (Figure 3b,c). Additionally, QTNs linked to these traits were also found on chromosomes 1, 2, and 7. The most QEIs are found on chromosomes 1, 4, and 7. QEIs for DF and Dmat traits are mainly located on chromosomes 1 and 4, also on chromosome 7. Few QTNs for the Dmat trait are on chromosome 6 compared to other traits. QEIs for PoW are on all chromosomes except 8.

3.4. Known Genes around Predicted QTNs and QEIs

Within the protein-coding genes and their 1 Kb flanking regions, we found 22 QTNs and 45 QEIs, most likely tagging the candidate causal genes. Functional annotation was available for 13 QTN-harboring genes and 28 QEI-harboring genes (see

Table 2. QTNs located within protein-coding genes and their 1 Kb flanking regions. * Abbreviations of trait names are in Table S1 and Table A1; ** gene body, *** 5′—upstream, **** 3′—downstream.

Trait Name *	QTN	Effect (Add)	R2	Gene	QTN Position (bp)	Annotation	Function
100SW	Ca4_47959199	−0.1758	0.89	Ca_10794	GB **	BAG family molecular chaperone regulator 1	Chaperon
100SW	Ca5_8331723	−0.19	1.34	Ca_18706	GB	E3 ubiquitin-protein ligase MBR2	Flowering control, promotes degradation of the FT regulator [45]
DF	Ca1_27596774	−0.132	1.73	Ca_25069	864 bp, 5′ ***	carbon catabolite repressor protein 4 homolog 4-like	Regulator of circadian rhythms [46]
DF	Ca3_9196031	0.154	0.59	Ca_20953	824 bp, 3′ ****	probable xyloglucan glycosyltransferase 5	xyloglucan modification
DFst	Ca1_8374473	0.166	1.29	Ca_08073	303 bp, 3′ ****	ethylene-responsive transcription factor 1-like	Response to pathogens and salinity in plant [47]
DFst	Ca2_15613312	0.135	1.94	Ca_18543	440 bp, 3′	pentatricopeptide repeat-containing protein At2g33760
HFP	Ca4_8807893	0.055	2.19	Ca_08378	GB	transcription factor bHLH106	Salt and low temperature response [48]
HFP	Ca6_53308665	0.141	1.73	Ca_22925	790 bp, 3′	UDP-glycosyltransferase 89B2	Transfer of a sugar onto a lipophilic acceptor [49]
HFP	Ca7_14294458	−0.133	1.18	Ca_23043	GB	peptidyl-prolyl cis-trans isomerase CYP37, chloroplastic	Regulation of the electron transport chain [50]
LS	Ca5_7212528	0.177	2.59	Ca_21567	GB	AP-5 complex subunit mu	Vesicle transport regulator [51]
NPB	Ca5_27676347	0.172	2.02	Ca_08885	92 bp, 3′	iron-sulfur assembly protein IscA-like 1, mitochondrial	Assembly of mitochondrial iron-sulfur proteins [52]
NSB	Ca1_22298342	0.13	0.93	Ca_20631	GB	40S ribosomal protein S28-1-like
PoNP	Ca1_26015468	−0.121	1.42	Ca_18590	445 bp, 3′	triose phosphate/phosphate translocator, chloroplastic	Phosphate transport, light response [53]
PoW	Ca5_665190	0.144	1.05	Ca_23223	763 bp, 5′	ninja-family protein AFP2	Regulator of AREB/ABF transcription factors [54]

and

Table 3. QEIs located within protein-coding genes and their 1 Kb flanking regions. * Abbreviations of trait names are in Table S1 and Table A1; ** K16—Kuban in 2016, K22—Kuban in 2022, A—Astrakhan, *** GB—gene body, 3′—downstream, 5′—upstream.

Trait Name *	QEI	Add env1 (K16) **	Add env2 (K22)	ADD env3 (A)	R2	Gene	QTN Position	Annotation	Function
100SW	Ca2_15153528	0.096	−0.144	0.048	1.07	Ca_18572	28 bp, 3′	dof zinc finger protein DOF1.4-like	Abiotic stress tolerance [55]
100SW	Ca4_12413458	0.056	−0.1268	0.071	0.81	Ca_04464	GB ***	cyclic dof factor 1-like	Regulates a photoperiodic flowering response [56]
DF	Ca4_27283954	0.166	−0.048	−0.118	1.5	Ca_20463	695 bp, 3′	putative Ulp1 protease family catalytic domain-containing protein	SUMO protease [57]
DF	Ca7_37894908	0.06	−0.133	0.074	0.92	Ca_16382	GB	ubiquitin carboxyl-terminal hydrolase 2	Involved in the direct or indirect regulation of AUX/IAA proteins stability [58]
DFst	Ca3_22914338	−0.06	0.175	−0.113	1.76	Ca_06215	GB	pentatricopeptide repeat-containing protein At2g44880-like, partial	ABA hypersensitivity at germination, RNA editing [59]
DFst	Ca5_11805008	0.112	0.003	−0.115	0.95	Ca_17114	GB	cucumisin-like	Serine protease [60]
DFst	Ca7_25029782	0.113	0.035	−0.148	1.34	Ca_23595	GB	probable polygalacturonase	Polygalacturonases involved in cell separation in the final stages of pod shatter and in anther dehiscence [61]
Dmat	Ca1_26611572	0.147	−0.009	−0.138	1.43	Ca_18616	194 bp, 5′	WUSCHEL-related homeobox 9-like	Homeodomain transcription factor required for meristem growth and early development [62]
Dmat	Ca1_31837660	−0.098	0.009	0.089	0.62	Ca_26401	GB	cytokinin dehydrogenase 6-like	Catalyzes the oxidation of cytokinins [63]
Dmat	Ca3_8285781	−0.025	−0.137	0.162	1.62	Ca_24378	GB	homeobox-leucine zipper protein PROTODERMAL FACTOR 2-like	Regulator of shoot epidermal cell differentiation [64]
LS	Ca1_42237742	0.262	0.013	−0.275	6.31	Ca_22678	395 bp, 5′	transcription factor bHLH49-like	Involved in cell elongation regulation [65]
NPB	Ca4_25599443	0.117	0.011	−0.128	1.29	Ca_16586	GB	endochitinase A-like	Antifungal protection in crops [66]
NPB	Ca7_28494061	−0.091	−0.016	0.106	0.85	Ca_18936	183 bp, 3′	leucine-rich repeat extensin-like protein 4	Represent a link between the cell wall and plasma membrane [67]
NPB	Ca8_8247161	−0.017	0.123	−0.106	1.15	Ca_11459	GB	aspartic proteinase-like protein 2	Pathogen stress response [68]
PoNP	Ca1_3619635	−0.00	0.189	−0.189	2.34	Ca_00444	GB	protein MALE DISCOVERER 2-like	Involved in recognition female gametes after pollination [69]
PoNP	Ca3_1396781	−0.047	0.258	−0.212	3.75	Ca_19418	GB	protein DEHYDRATION-INDUCED 19 homolog 5-like	Involved in dehydration and salinity stress signaling pathways [70]
PoNP	Ca4_22873605	−0.131	−0.06	0.191	1.89	Ca_14464	518, bp 5′	putative disease resistance protein At3g14460
PoNP	Ca5_17861341	−0.028	−0.143	0.17	1.65	Ca_22848	GB	aspartyl protease family protein At5g10770-like	aspartyl protease
PoNP	Ca6_47911283	−0.151	0.056	0.095	1.15	Ca_23445	GB	alanine aminotransferase 2-like	alanine aminotransferase
PoNP	Ca7_20945484	−0.035	0.175	−0.139	1.68	Ca_14487	863 bp, 3′	Retrovirus-related Pol polyprotein from transposon TNT 1-94
PoW	Ca2_6214519	−0.249	0.024	0.225	3.56	Ca_20925	775 bp, 3′	fructose-bisphosphate aldolase 1, chloroplastic	Involved in photosynthesis [71]
PoW	Ca2_12581771	0.003	0.16	−0.163	1.64	Ca_18081	GB	ATP synthase subunit delta’, mitochondrial	[72]
PoW	Ca2_18976694	0.12	−0.014	−0.105	0.8	Ca_15955	GB	B3 domain-containing protein At5g42700-like	AP2/B3-like transcriptional factor [73]
PoW	Ca2_34943639	−0.067	0.2	−0.139	2.07	Ca_16876	GB	tRNA (guanine(26)-N(2))-dimethyltransferase	Posttranscriptional modification of tRNA [74]
PoW	Ca3_8932912	−0.351	0.047	0.304	6.83	Ca_25280	GB	alcohol dehydrogenase-like 6	alcohol dehydrogenase
PoW	Ca6_15593063	0.038	−0.222	0.184	2.66	Ca_05341	GB	protein EMBRYO DEFECTIVE 514	[75]
PWwP	Ca3_19996983	0.103	0.589	−0.692	21.59	Ca_09401	GB	proteasome subunit beta type-4-like
PWwP	Ca4_24437355	−0.218	−0.019	0.2367	2.69	Ca_20867	194 bp, 3′	cyclase-associated protein 1	Increases the rate of nucleotide exchange on actin [76]

).

3.5. Superior Genotypes for Key Traits

The development of high-yielding and early-maturing chickpea varieties is limited by a significant reduction in genetic and adaptive variation. Chickpea landraces provide a wide range of genetic variations that have not been thoroughly explored and utilized systematically by breeders [77,78].

To identify genotypes with high yield and early maturation among the VIR landraces we calculated two statistics, namely (1) the number of favorable alleles of the QTN and QEI loci controlling Dmat and yield-related traits (PoW, PoNP, PWwP and 100SW), and (2) the “trait improvement” (TI) score as the difference between the number of favorable alleles and the alleles negatively affecting the trait for each accession. Since QEI loci are environment-specific, the repertoire of the QEI alleles in samples was assessed for each environment separately.

The number of favorable alleles for the QTNs associated with the Dmat trait does not exceed four in the samples (Table S10). The highest TI score of 4 or 5 is calculated for 14 samples. The number of favorable alleles for the QTN loci associated with yield-related traits ranges from 17 to 26 (Table S11). Twenty two samples had a TI score higher than 10, and two of them have the highest TI score of 13.

In the Kuban region, there are more favorable alleles of the QEI loci for the Dmat trait and fewer alleles with negative effects compared to the Astrakhan OS (Tables S12–S14). The highest TI score for landraces grown in the Kuban region in 2016 and 2022 ranges from 4 to 9, with 26 accessions scoring in this interval at each of the outstations (Tables S12 and S13). In the Astrakhan region, only four landraces have positive TI scores, with a value not exceeding 3 (Table S14).

A similar situation was observed for the QEI loci associated with yield-related traits: in the Kuban region, 21 accessions in 2016 and 11 in 2022 showed the highest value of TI score in the interval from 6 to 16, with the number of favorable alleles ranging from 29 to 40 (Tables S15 and S16). However, in Astrakhan, a large number of favorable alleles (from 25 to 32) is counterbalanced by an equally large number of alleles characterized by a negative effect. As a result, only five samples produced a positive TI score value (Table S17).

Landraces VIR1171, VIR0603, VIR0620, and VIR0668 grown at the Kuban OS in 2016 are characterized by high TI scores for both QTN and QEI loci associated with the Dmat trait. However, we found only two such landraces, namely, VIR0620 and VIR0799 cultivated at the Kuban OS in 2022 (Table S18). In addition, the VIR0230, VIR0244, VIR0030 and VIR0042 landraces at the Kuban OS in 2016, VIR0637 at the Kuban OS in 2022, as well as VIR0855 at the Astrakhan OS show high TI values for the QTNs and QEIs related to productivity.

High TI score values for QEI loci associated with the Dmat and yield-related traits are observed in two landraces, namely VIR0241 and VIR0918 at the Kuban OS in 2016 and 2022, respectively (Table S19).

4. Discussion

In recent years, mixed linear models have been extensively utilized to predict genomic regions linked to crucial traits in chickpeas [22,24,25,31,79]. However, the majority of these models [3,4,80] merely address the additive effects of markers and fail to estimate dominance effects or gene-by-environment interactions. Importantly, the IIIVmrMLM model [12,13] utilized in this study enables the comprehensive evaluation of these effects.

Cultivars are evaluated based on their performance when grown in different environments [81,82]. Traits important for commercial agriculture, such as yield and maturity, often vary significantly between environments due to genotype-by-environment interactions. Therefore, it is essential to model these interactions and assess a genotype’s ability to provide adaptive alleles for the successful breeding of resilient and sustainable crop varieties [77].

In this study, 159 accessions from the VIR collection, including 145 landraces and 13 cultivars, were planted in three different environments. The first two environments were the Kuban outstation in 2016 and 2022, and the third was the Astrakhan OS in 2022. In 2022 at the Kuban outstation, daily temperatures were lower compared to 2016, and precipitation levels were higher. The Astrakhan OS experienced the hottest and driest agricultural conditions across all environments (see Tables S2–S4 for details).

The evaluation of 12 important agronomic traits (Table S1) revealed significant variation within a single environment and across different environments (Table S5, Figure 1a). The most pronounced variation across environmental gradients was observed for productivity and phenological traits, suggesting a genotype-by-environment interaction. To comprehend the genetic factors responsible for trait variation across different environments, we utilized the IIIVmrMLM program in Multi_env (multi-environment) mode.

We have confidently identified a total of 161 QTNs with stable genetic effects across various environments and 256 QEIs with environment-specific effects (Tables S8 and S9). Collectively, both QTNs and QEIs account for a significant proportion of the variation across traits (Table 1).

Twenty two QTNs and 45 QEIs are linked to protein-coding genes, likely identifying potential causal genes (see Table 2 and Table 3). The functions of many of these genes are known. For instance, Ca5_8331723, which is associated with 100SW, is located within the Ca_18706 gene for E3 ubiquitin-protein ligase MBR2. This gene promotes the degradation of the Flowering Locus T regulator in Arabidopsis [45] (Table 2). The QTNs Ca1_27596774 and Ca1_8374473, which impact phenological traits, are found in the flanking regions of Ca_25069 and Ca_08073 genes, respectively. These genes control circadian rhythm [46] as well as the response to pathogens and salinity [47]. The QTN Ca4_8807893, associated with the HFP trait, is located in the Ca_08378 gene, which encodes the transcription factors involved in responding to unfavorable environmental conditions [48]. The QTNs Ca6_53308665 and Ca1_26015468, associated with HFP and PoNP, respectively, are downstream of the Ca_22925 and Ca_18590 genes, which control sugar [49] and phosphate transport [53]. QTN Ca5_665190, associated with PoW, is upstream of the Ca_23223 gene implicated in hormone signaling [54].

Quantitative trait loci such as Ca7_37894908, Ca7_25029782 and Ca1_31837660 (see Table 3), associated with phenological traits, are located within genes that control AUX/IAA protein stability, cell separation during pod formation [58,61], and the oxidation of cytokinins [63], respectively. The QEIs Ca1_26611572 and Ca3_8285781, associated with the time from germination to full maturation (i.e., Dmat), map to the Ca_18616 and Ca_24378 genes. These genes encode a transcription factor participating in the regulation of meristem growth [62] and a homeobox-leucine zipper protein PROTODERMAL FACTOR 2-like (Table 3), respectively. Incidentally, PROTODERMAL FACTOR 2 regulates the differentiation of shoot epidermal cells in Arabidopsis [64].

The QEI Ca1_42237742, associated with leaf size, is upstream of the Ca_22678 gene involved in the regulation of cell elongation [65]. The QEI Ca7_28494061, associated with NPB, is located 183 bp downstream of the Ca_18936 gene, which encodes leucine-rich repeat extensin-like protein 4. This might coordinate processes in the cell wall, as leucine-rich extension receptors are known to do [67].

Ca1_3619635, which is located within the Ca_00444 gene involved in gamete recognition, is associated with the number of pods. The recognition of female gametes after pollination is crucial for successful seed formation [69]. Another QEI, Ca4_24437355, associated with the weight of plants with pods, is downstream of the Ca_20867 gene involved in actin dynamics [76].

According to the recent findings [83], only 4 QTNs and 5 QEIs were found to overlap with the QTNs identified using Super, FarmCPU, and Blink models of the GAPIT package (Table S20). One of these QTNs, Ca3_8285781, which is associated with NSB, falls into the Ca_24378 encoding PROTODERMAL FACTOR 2-like. The limited overlap between the results of these two analyses can be attributed to GAPIT identifying associated markers for each environment separately, while IIIVmrMLM in Multi_env mode predicts associations for the three environments simultaneously.

The comparison of GWAS hits with genomic regions from previous studies reveals that 48 QTNs and 88 QEIs intersect within the LD limits (50 kb) of QTNs identified in studies [21,22] (Tables S21 and S22). Although different studies assess slightly different traits, it is not surprising that many matched QTNs are associated with various traits. For instance, the study by Varshney et al. [21] measured days to 50% flowering instead of the Dmat, DF, and DFst traits from our study. Nevertheless, most of the associated traits, while different, measure the same characteristics of the plant, such as 100 seed weight, yield per plant, harvest index, and the number of primary and secondary branches characterizing productivity.

This study definitively established that VIR landraces possess multiple favorable alleles of the QTN loci with stable effects in varying climatic conditions of the Kuban and Astrakhan regions (Refer to Tables S10 and S11). Furthermore, a greater number of Kuban-specific favorable alleles of the QEI loci compared to the Astrakhan-specific ones (See Tables S12–S17) was observed. The impact of the favorable alleles of the QEI loci associated with the Dmat trait in the Astrakhan region is undeniably additive (see Figure 4), but it is adequately balanced by an equivalent or even larger number of negative alleles, as sown in Table S14. Despite these findings, it was irrefutable that the accessions in the Astrakhan OS matured at a faster rate than those in the Kuban region (Figure 1a and Table S5), suggesting that the explanation for this observation likely lies within the gene-by-gene interactions.

The analysis reveals negative correlations between maturation time (the Dmat trait) and several beneficial alleles for Astrakhan-specific QEI loci and loci specific for Kuban in 2016 (Figure 4). Conversely, a positive correlation is observed between productivity traits and the number of favorable alleles for most markers and traits, except for the PoW and 100SW traits measured at the Kuban OS in 2016. In these cases, the correlation was found to be statistically insignificant, with p-values equal to 0.66 and 0.41, respectively. The identified markers’ additive effect on most traits suggests that accessions carrying more favorable alleles are conducive to breeding through the pyramiding of loci.

Large-effect quantitative trait loci (QEIs) such as Ca2_27433383, Ca4_29141628, Ca5_21206336, and Ca3_19996983 (see Table 1) can be incorporated into marker-assisted selection programs, while markers with minor effects are more likely to be utilized in genomic selection (GS) in combination with markers with large effects. Previously GS models have been employed for the prediction of yield traits in chickpea [32,33]. In this context, the environment-specific alleles of the QEI loci (Table S9) are particularly significant, as breeders can utilize them to develop varieties with improved adaptation to specific climatic conditions.

5. Conclusions

The genetic and phenotypic variation present in chickpea landraces has yet to be comprehensively explored. This study aims to expand the genetic diversity of chickpea by identifying landraces with adaptive and favorable alleles that control maturity and yield-related traits, with a specific focus on the VIR landraces. The assessment of traits in varying environments in the Kuban and Astrakhan regions revealed significant variation across environmental gradients, indicating genotype-by-environment interactions. To thoroughly examine stable and environment-specific effects, the IIIVmrMLM model was employed. The results showed that the VIR landraces possess numerous favorable alleles of the QTN loci, with stable effects in all tested environments. Importantly, they demonstrate a greater abundance of Kuban-specific alleles of the QEI loci compared to the Astrakhan-specific ones. The annotation of the genetic repertoire of favorable alleles in landraces is fundamental and imperative as the first step towards their integration into modern breeding programs.