1. Introduction
White lupin (
Lupinus albus L.) is a Mediterranean grain legume that used to be a major crop in various regions of the Roman Empire because of its ability to adapt to and improve infertile soils [
1]. A recent surge of interest in its cultivation as a rain-fed food or feed crop in Europe is justified by its protein content that is close to 40% and other favourable quality traits of its seed [
2,
3,
4], which can respond to the increasing demand for healthy and nutritious plant-based food [
5] and high-protein feedstuff [
6]. In particular, the exploitation of white lupin seed as a component of functional, healthy, or vegan food is favoured by its good content of essential amino acids and several useful techno-functional properties [
7], the positive effects on human health that it can exert with respect to diabetes and glycaemia, hypertension, cardiovascular diseases, and obesity [
8], and 8–12% content of oil with excellent nutritional characteristics [
9]. The potential value of white lupin for feed protein production was confirmed by its greater crude protein yield per unit area compared with other cool-season grain legumes, such as pea (
Pisum sativum L.), faba bean (
Vicia faba L.), and narrow-leafed lupin (
Lupinus angustifolius L.), across climatically-contrasting, autumn-sown environments of southern Europe [
10]. However, insufficient yielding ability limits the spread of this crop crucially [
11].
White lupin adaptation to severe drought has paramount importance in Mediterranean-climate areas, where this stress occurs in coincidence with critical reproductive stages. Drought stress is predicted to increase throughout the Mediterranean basin and to expand northward and eastward into Europe, owing to lower rainfall and rising evapotranspiration caused by climate change [
12]. The available information on the extent of white lupin genetic variation for drought tolerance is modest and concerns mainly landrace material. A fairly narrow variation was observed in germplasm from Portugal [
13]. Variation for this trait was reported in Egyptian germplasm [
14], while a few accessions of Italian origins were found to be more drought tolerant than local germplasm in Egypt [
15]. Annicchiarico et al. [
16] reported large yield variation among accessions of a world landrace collection grown in a moderately favourable Mediterranean environment, along with the definite superiority of several landrace accessions over a set of control varieties. Variation for specific adaptation to severe drought, as indicated by a fairly large genotype × environment interaction (GEI) for grain yield across drought-prone and moisture-favourable managed environments, was reported by Annicchiarico et al. [
17] for a set of landraces having different geographic origins and similar phenology. Moderate GEI across these environments emerged also for individual genotypes sorted out of landrace accessions [
18]. Finally, large GEI was reported for a small set of breeding lines across subcontinental-climate and Mediterranean-climate sites of Italy [
19].
The tolerance of cool-season grain legumes to the terminal drought that features Mediterranean-climate regions is typically associated with an early phenology, but the exploitation of this drought stress-escape mechanism in autumn-sown environments may be limited by the lower winter hardiness of early-flowering germplasm [
20,
21]. When targeting such environments, it may be useful to assess and exploit the genetic variation for intrinsic drought tolerance obtained by partialling out the effect of phenology on genotype yield responses [
22]. As an alternative, more demanding avenue, one may dissect and exploit the variation for single traits that contribute to drought avoidance or drought resistance [
23,
24].
The scope for molecular marker-based selection has widened considerably after the development of next-generation sequencing techniques, such as genotyping-by-sequencing (GBS; [
25]), which allows genotyping large germplasm sets by thousands of single nucleotide polymorphism (SNP) markers at a relatively low cost. Książkiewicz et al. [
26] confirmed the ability of GBS to generate thousands of polymorphic SNP markers for white lupin genetic analyses and highlighted its value for a genome-wide association study (GWAS). Identifying and exploiting putative QTL (quantitative trait loci) hotspots of the genome that are associated with drought tolerance led to enhanced crop yield under drought in two other cool-season grain legumes, such as pea [
27] and chickpea [
28]. In the presence of a definitely polygenic trait genetic architecture revealed by a GWAS, which is likely to emerge for complex traits such as crop yield or drought tolerance, an alternative molecular breeding strategy is represented by genomic selection (GS). This strategy combines phenotyping and genotyping data of a genotype sample (training population) representing a target genetic base (reference population) into a statistical model for the prediction of breeding values in future plant selection [
29,
30]. Pioneer examples of GS for grain legume drought tolerance improvement were provided for pea [
22] and chickpea [
24], and a proof-of-concept study confirmed the ability of GS to identify drought-tolerant pea lines [
27]. Encouraging results for GS of white lupin yield emerged in Annicchiarico et al. [
18], who reported predictive ability values (as Pearson’s correlation between predicted and observed values based on intra-environment cross-validations) in the range of 0.47–0.76 for geographically-diversified landrace material evaluated in climatically-contrasting regions or across moisture-favourable and severely drought-prone managed environments. A second study on landrace germplasm reported high to moderate predictive ability (0.49–0.85) for a set of morphophysiological traits [
31].
White lupin studies on drought tolerance variation, GEI and genome-enabled predictive ability are scanty. In addition, their results relative to landrace germplasm are not necessarily meaningful to breeding programs, which select inbred lines whose variation for yielding ability and drought tolerance on the one hand and for molecular markers on the other may be smaller than those of geographically-diversified landrace material. This study focused on a reference population of inbred lines generated by crossing each of four elite sweet-seed cultivars or breeding lines with each of four elite landrace accessions of different geographical origins. Its main objectives were (a) to assess the phenotypic variation for adaptation to severe drought and the extent of GEI for grain yield, aerial biomass, harvest index, and the onset of flowering across drought-prone and moisture-favourable managed conditions, (b) to investigate the genetic architecture and the presence of highly relevant genomic regions for grain yield under severe drought and favourable conditions and intrinsic drought tolerance by a GWAS, and (c) to assess the ability of different GS models to predict genotype yield responses, drought tolerance, and the onset of flowering.
3. Discussion
The grain yield reduction of 61% arising from the imposed drought stress was severe but largely comparable, for example, with yield reductions in the range of 35–76% that were reported for narrow-leafed lupin across moisture-favourable and drought-prone environments of Australia [
34]. The observed extent of GEI for grain yield across moisture-contrasting environments was in substantial agreement with earlier studies on white lupin that were performed in managed environments on germplasm sets mainly composed of accessions [
17] or individual genotypes [
18] of landrace germplasm. These earlier studies revealed genetic correlation values for yield responses across managed environments that were somewhat lower than the current one (
rg = 0.76–0.77 vs. 0.84). The imposition of greater yield reduction under stress (nearly 80%) and the wider genetic variation of the tested germplasm in those studies may account for this minor difference. A reason for the moderately high genetic correlation across drought-prone and moisture-favourable environments that was observed in this study and the earlier ones could be a general impact of climatic adaptation, since the delay in fulfilling the vernalization requirement of later-flowering lines under the adopted late-winter sowing could disfavour a priori these lines in both managed environments [
17]. The occurrence of such a delay is supported by the fact that mean daily temperatures averaged 7.6 °C in the first two weeks following the sowing, while 6 °C would be needed for the vernalization of mid- to late-flowering genotypes in the same period [
35].
Correlation and regression results for the onset of flowering confirmed that drought stress escape by early flowering is an important mechanism for white lupin adaptation to environments with severe drought. This result was confirmed at the genomic level by the fact that the most important SNP for grain yield under stress coincided with that for the onset of flowering. The association of drought tolerance with early phenology emerged in earlier studies on white lupin [
17] and other grain legumes, such as narrow-leafed lupin [
36] and pea [
22]. The study on pea, which was performed on three sets of recombinant inbred lines under managed stress conditions similar to the current ones, revealed a nearly identical response pattern of line grain yield as a function of the onset of flowering when averaging its results across the three line sets, namely, an average yield loss of 0.033 t/ha per day of delayed flowering (compared with the current value of 0.036 t/ha) across a 20-day range of line onset of flowering that is identical to the current one. The only difference between the two studies was the greater
R2 for the regression of yield as a function of the onset of flowering in the pea study (which averaged 0.63, compared with the current value of 0.29), suggesting that drought stress escape had about two-fold greater impact on grain yield variation of pea lines compared with current lupin lines. The adjusted yield, which was indicative of intrinsic drought tolerance, had lower relative importance in the pea study (where it related to 37% of the total grain yield variation based on the average
R2 value) than in the current study (where it related to 71% of the total variation), but its reported broad-sense heritability was quite similar in the two studies (0.57 vs. 0.59). Importantly, the exploitation of the adjusted yield through marker-assisted selection or GS produced pea genotypes with distinctly improved drought tolerance in a following proof-of-concept study [
27]. Actually, stress escape by earlier flowering emerged as well as a plant plasticity mechanism in the present study, based on the modest anticipation of mean flowering date exhibited unanimously by all lupin genotypes (without GEI).
Although appealing for drought tolerance improvement, the exploitation of early onset of flowering is limited in many white lupin cropping environments, such as the autumn-sown ones in inland areas of the Iberian, Italian and Balkan peninsulas, because the early-flowering germplasm is more subjected than the late one to winter low-temperature stress and winter mortality [
21]. The ideal plant type in these environments ought to combine moderate lateness of flowering with high intrinsic drought tolerance. Indeed, the landrace accession La646, which possesses these characteristics [
17], exhibited an exceptionally wide adaptation pattern that made it the top-yielding one out of 121 landraces or modern cultivars evaluated across drought-prone or cold-prone autumn-sown environments of southern Europe [
16]. The progeny lines of this landrace clearly tended to inherit these characteristics (
Table 4), reinforcing the high value of this accession as a genetic resource for breeding programs. The heritability of intrinsic drought tolerance is also supported indirectly by the mean response of the progeny lines of the Greek landrace Gr56, an accession that exhibited high susceptibility to drought [
17]. The indications of heritable variation from parent genotypes to progeny lines and moderately high broad-sense heritability that emerged for complex traits, such as intrinsic drought tolerance and grain yield under severe drought, are encouraging for drought tolerance improvement of this crop.
Intrinsic drought tolerance may rely on different physiological mechanisms in white lupin, such as stomatal closure, greater stomatal conductance, or less reduced net photosynthesis under stress [
13,
37]. In addition, the ability to accumulate assimilates in the shoots upon drought stress intensification may contribute to plant survival and seed filling under stress through re-translocation [
13,
38]. This latter mechanism may have contributed to the greater increase of the harvest index under stress that represented a key feature of drought-tolerant lines in this study and emerged already for landrace germplasm [
17].
We observed a faster LD decay in our germplasm set compared with that reported earlier for a collection of cultivars and landraces [
39], possibly because of the generation by our crossing scheme of a high number of heterozygous loci differently combined depending on the specific geographically-contrasting parental lines. The occurrence of substantial variation among chromosomes for LD decay was already found in [
39]. Although challenged by the fast LD decay and the somewhat suboptimal genotype sample size, our study was able to reveal a few QTLs for grain yield under moisture-favourable or drought stress conditions while confirming the expected polygenic control of the crop yield traits, especially the adjusted yield (to which various physiological mechanisms with fairly limited individual impact may be expected to contribute). The polygenic control of all yield traits was confirmed by (a) the ability of the significant SNPs to account for only a minor portion of the phenotypic variation and (b) the definitely lower proportion of phenotypic variation that these SNPs could explain compared with that of GS models (which can also account for minor gene effects). An insufficient GWAS power to detect small-effect SNPs [
40] likely hindered our ability to identify QTLs with modest genetic effects for grain yield in drought or favourable conditions and the adjusted grain yield.
Interestingly, the significant SNPs for grain yield differed across managed environments in spite of the moderately high genetic correlation across environments. This result emphasized the partly different genetic control of the yield trait in the two cropping conditions. On the other hand, the moderately high genetic correlation for genotype yield response across the contrasting environments, and the only moderate decrease of GS predictive ability for grain yield passing from single-environment to cross-environment prediction scenarios (
Table 6), suggested that many small-effect loci for grain yield may be coincident between favourable and drought stress environments.
Our GWAS results suggested a polygenic control also for the onset of flowering, a result that agrees with recent unpublished results by Rychel et al. (personal communication). Some of the significant SNPs revealed by our study mapped on the same chromosomes of QTLs reported in previous GWAS and linkage mapping studies conducted under various vernalization conditions, including regions on chromosomes 1 [
41], 3 [
42], 13 [
26,
41,
42], and 20 [
42]. In contrast with earlier studies [
26,
41,
42], we found no QTL on chromosomes 2 and 16 for this trait, possibly because extremely late-flowering lines were absent from our germplasm set.
Our study revealed several genomic regions of potential interest for grain yield and onset of flowering selection by scanning a region as long as the mean chromosome distance at which LD dropped below 0.2 in both directions from each significant SNP (
Supplementary Table S1). Inter alia, Lalb_Chr13g0291541 that was associated with SNP Chr13_1653492 for the onset of flowering encodes a transcription factor of the C2H2 family [
43], which is known to play a role in flowering regulation [
44]. However, the polygenic control of all focus traits does support the exploitation of SNP information mainly by means of GS models, also in view of the high predictive ability exhibited by these models. In particular, our findings reinforced the high potential interest of GS for lupin grain yield improvement by confirming for a genetically-broad reference population of sweet-seed breeding lines the high predictive ability of GS found in landrace germplasm [
18]. That study reported predictive ability values for single-environment grain yield predictions in moisture-favourable or drought stress conditions in the range of 0.47–0.58, which are somewhat lower than the current range of 0.67–0.78 (
Table 7). Likewise, the cross-environment predictive ability values across moisture-favourable and drought-stress environments in [
18] (in the range of 0.42–0.51) are somewhat lower than the current values (in the range of 0.51–0.60;
Table 6). If the limited seed market size of white lupin supported the breeding of this crop for wide adaptation [
45], the moderate consistency of GS predictions across the moisture-contrasting environments could be exploited to build a comprehensive GS model trained on grain yield data from the contrasting environments. On the other hand, the selection for specific adaptation to drought-prone cropping environments could exploit (a) the GS model for grain yield under drought when targeting a mild-winter region or (b) the GS model for the adjusted yield (also featuring a substantial predictive ability), when targeting autumn-sown, relatively cold-prone environments by selection for intrinsic drought tolerance without affecting the phenology.
In conclusion, this study generated information that could support the phenotypic and genomic selection of white lupin for drought-prone or climatically-diversified target regions. It confirmed the importance of an early phenology for drought stress escape on the ground of phenotypic and GWAS results. However, it revealed the presence of heritable, polygenic genetic variation for intrinsic drought tolerance that could be exploited through phenotypic selection or, less expensively, through moderately reliable genomic predictions. It also indicated the feasibility of phenotypic or genomic selection for wide adaptation to moisture-contrasting target environments. The high predictive ability of the current GS models for drought tolerance will be verified by future proof-of-concept work aiming to assess actual genetic gains obtained from GS application to independent germplasm sets.