Common bean (Phaseolus vulgaris L.) is one of the most important legumes for human consumption and is cultivated worldwide, particularly in tropical and subtropical countries of Africa and America (Broughton et al., 2003). Cultivated common bean is organized in two major gene pools, Mesoamerican and Andean, which resulted from independent domestication events, generating landraces with diverse seed quality, nutritional traits, resistance to diseases, tolerance to abiotic stress, morphologic, and agronomic traits (Schmutz et al., 2014). Bean has been documented as a highly valuable food for the human diet, an accessible and relatively inexpensive source of proteins and micronutrients, such as iron and zinc (Beebe, 2012).

Micronutrient deficiencies are among the most common global nutritional problems, affecting half of the population of the world, especially women and children in developing countries (Bouis and Welch, 2010). Fe and Zn deficiencies are among the main common causes of micromineral malnutrition worldwide. In humans, Fe is essential for several cellular mechanisms (Steinbicker and Muckenthaler, 2013). Iron Deficiency Anemia (IDA), a pathology associated with scarce iron intake and absorption, affects a third of the population of the world and has been associated with several chronic diseases (Lopez et al., 2016). Zn has a key role in regulating growth and in the functioning of the immune system (Chasapis et al., 2012), and its deficiency can contribute to infections (Hambidge, 2000).

Wide phenotypic variability has been reported in bean in terms of Fe and Zn accumulation, with seed Fe concentration from 35 to 90 mg/kg and seed Zn concentration from 20 to 54 mg/kg (Beebe et al., 2000), these ranges are higher than those of other main staple crops, such as rice, wheat, and maize (Petry et al., 2015). For this reason, common bean is one of the crops targeted for biofortification. A cup of beans supplies 25% of Fe and 15% of Zn to the recommended daily allowance in the human diet, but breeding programs have the potential to develop genotypes with two to three times more Fe and Zn concentrations (Cichy et al., 2009).

Bouis (2003) reported that biofortification breeding has multiple positive effects, such as more efficiency in nutrients, fertilizer, and irrigation demands, and can increase disease resistance and yield. However, some difficulties have been described in bean breeding for micronutrient concentration, mainly a strong negative correlation between Fe and Zn concentration and yield. Source genotypes with high iron contents (>100 mg/kg) have been identified, but most of these lines have poor agronomic performance (Raatz, 2018). For this reason, a major breeding objective is overcoming the negative correlation between the micronutrient concentrations with yield components and finding the optimum balance between these traits.

While the phenotypic variability for Fe and Zn in common bean has been utilized in breeding programs with some success, the development of genomic tools will likely be helpful to reach the goals of biofortification (Beaver and Osorno, 2009). Currently, the genetic architecture of Fe and Zn accumulation in seed in common bean has been the object of several studies (Izquierdo et al., 2018), usually using Quantitative Trait Loci (QTL) analysis to find genomic regions related to the genetic control of these traits. Nevertheless, molecular markers have not been implemented in bean breeding programs focused on mineral concentration, as parental materials employed for populations are often distant from elite breeding materials, and the QTLs found with major effects are scarce and cannot be extrapolated to other populations (Raatz, 2018).

Another factor to consider in biofortification breeding is the accompanying agronomic traits, such as drought tolerance. Hummel et al. (2018) predicted a large part of current common bean growing areas to become unsuitable for cultivation by 2050, generating reductions in yield for current varieties and potentially affecting the nutritional quality, especially the iron levels of the crop, if current climate change trends continue. Drought is the main cause of yield loss worldwide, and its incidence and duration are expected to increase due to climate change (Lauer et al., 2012). Drought stress is a principal abiotic constraint that affects beans, causing losses close to 60% in production (Beebe et al., 2008).

The objective of this study was to investigate the correlation of genetic architecture of Fe and Zn accumulation and yield components in a common bean under irrigated and drought conditions using QTL and Genome-Wide Association Studies (GWAS) analysis in three biofortified bi-parental Mesoamerican populations and to identify genomic regions associated with yield components, micromineral accumulation, and their possible relationship.

The populations analyzed showed ample phenotypic variability in terms of average accumulation of micronutrients: SCR16 × SMC40 (53.7–108.3 mg/kg for Fe and 17.5–36.8 mg/kg for Zn), SMC33 × SCR16 (50.0–101.6 mg/kg for Fe and 16.9–36.8 mg/kg for Zn) and SMC44 × SCR9 (49.2–107.5 mg/kg for Fe and 18.7–42.4 mg/kg for Zn) (Supplementary Table 1). These ranges of micronutrient concentrations are wider than those reported for other populations: Andean biparental populations AND696 × G19833 (39–79 mg/kg for Fe and 16–29 mg/kg for Zn) (Cichy et al., 2009); G21242 × G21078 (28–95 mg/kg for Fe and 17–49 mg/kg for Zn) (Blair et al., 2011); Mesoamerican biparental population G14519 × G4825 (35–97 mg/kg for Fe and 17–49 mg/kg for Zn) (Blair et al., 2010); intergene pool biparental populations DOR364 × G19833 (40–84 mg/kg for Fe and 17–42 mg/kg for Zn) (Blair et al., 2009); Cerinza × G10022 (54–100 mg/kg for Fe and 23–38 mg/kg for Zn) (Blair and Izquierdo, 2012); an African collection (23.6–78.3 mg/kg for Fe and 19–56.1 mg/kg for Zn) (Tryphone and Nchimbi-Msolla, 2010); an Andean collection (54.5–99.3 mg/kg for Fe and 21.6–39.7 mg/kg for Zn) (Katuuramu et al., 2018); a European collection (38.4–93.7 mg/kg for Fe and 18.9–43.6 mg/kg for Zn) (Caproni et al., 2020); and a multiparent advanced generation intercross (MAGIC) population (37.9–87.6 mg/kg for Fe and 18.5–39 mg/kg for Zn) (Diaz et al., 2020). Common bean has significant biofortification potential due to the wide phenotypic variability in terms of Fe and Zn concentration, producing improved lines that can accumulate >100 mg/kg of Fe and >45 mg/kg of Zn. This high accumulation capacity of micronutrients exceeds that of other legume species, such as chickpea (36–91 mg/kg of Fe), pea (26–91 mg/kg of Fe), and mung bean (35–87 mg/kg of Fe), and only surpassed by lentil (37–176 mg/kg of Fe) (Shiva et al., 2016; Upadhyaya et al., 2016; Ma et al., 2017; Jha and Warkentin, 2020; Kumar and Pandey, 2020; Sab et al., 2020).

Breeding programs need fast, accurate, and cheap methods for screening large numbers of breeding lines and germplasm for genotypes with high micronutrients contents. EDXRF is an alternative to inductively coupled plasma-optical emission spectrometry (ICP-OES), the most common tool used for measuring microelements (Guild et al., 2017b). The benefits of this method compared to ICP lie in minimal sample preparation before analysis, less labor time, and cost required [∼$0.15 per sample to X-ray diffraction (XRD) method]. In addition, it is a method with a high level of prediction. Guild et al. (2017a) reported values of prediction models of 96% for Fe and 93% for Zn compared with ICP reference concentration, with an SE of prediction of 2.5 mg/kg of Fe and 1.5 mg/kg of Zn with a maximum coefficient of variation of 5% for Fe and 3.5% for Zn.

The genetic basis underlying the micronutrient concentration have been widely studied in common bean using QTL analysis, and GWAS (Gonzalez et al., 2017; Izquierdo et al., 2018; Gunjača et al., 2021). QTL mapping is an excellent strategy for gene mapping in this case because only a bi-parental population and a small set of markers are necessary a bi-parental population and a small set of markers, which are utilized to identify the genomic regions that segregate with a trait. However, this strategy is usually low in resolution since only two alleles per locus are analyzed, and genetic recombination is limited (Islam et al., 2016). Traditional QTL mapping is highly dependent on the genetic diversity of the two parents, and the effects of the detected QTLs can vary between populations. For this reason, the validation is necessary in other populations. GWAS analysis did not show any significant association. This lacks correlation to the type of population used for the biofortification analysis, as biparental populations are more suited for QTL mapping (Diaz et al., 2021). The limitations of QTL analysis can be overcome using GWAS, which can narrow down the candidate regions using natural populations. However, its use in biparental populations presents problems by low events recombination typical of this type of populations and low allelic richness required for GWAS. On the other hand, GWAS is susceptible to report false-positive associations, and its results require validation. The number of markers used in the GWAS highly affects its results (Han et al., 2018). In this study, 14 QTLs for SdFe were identified, of which 9 QTLs have been previously reported. Astudillo et al. (2013) reported two QTLs co-located respectively to SdFe3.2 and SdFe11.2 in an inter-gene pool population; the positive allele for high iron in that study came from the Andean parental (G19833). In this study, SdFe6.3 showed the highest percentage of variance explained and was the most significant QTL identified for Fe. This QTL was identified previously in an Andean biparental population (Cichy et al., 2009), in an intergene pool population (Astudillo et al., 2013), in an Andean collection (Katuuramu et al., 2018), and a Meta-QTL analysis using different populations (Izquierdo et al., 2018). Other QTLs, such as SdFe6.2, SdFe6.4, SdFe7.1, SdFe8.1, and SdFe11.1, have been identified in different populations belonging to the Andean and Mesoamerican gene-pool (Blair et al., 2010, 2011; Blair and Izquierdo, 2012; Cichy et al., 2014; Izquierdo et al., 2018; Katuuramu et al., 2018; Diaz et al., 2020). SdFe6.1 was only reported previously by Diaz et al. (2020) in a Mesoamerican Multi-parent Population and the positive allele came from a biofortified genotype MIB778. It was suggested that this allele in MIB778 may have been derived from a Phaseolus dumosus parent. There are no reports about SdFe3.1 but in this study, the QTL proved to be highly important for the analyzed biofortified populations.

For SdZn, 12 QTLs were identified, of which 6 have been reported previously. SdZn6.1 has been reported in several populations of both gene pools (Cichy et al., 2009; Blair et al., 2010; Blair and Izquierdo, 2012; Izquierdo et al., 2018; Katuuramu et al., 2018). SdZn3.1 and SdZn10.1 were only identified in an inter-gene pool bi-parental population (Astudillo et al., 2013) and an Andean Diversity Panel (Katuuramu et al., 2018). Conversely, SdZn2.3 and SdZn8.2 only were reported in a population belonging to the Mesoamerican gene pool (Freyre et al., 1998; Blair et al., 2009; Izquierdo et al., 2018; Diaz et al., 2020). Mapping certain QTL in different gene pools demonstrates the complex control over micronutrient accumulation in the common bean that can vary between the Andean and Mesoamerican genotypes. It has been reported that Andean genotypes showed higher Fe concentrations than Mesoamerican genotypes and that Mesoamerican lines presented higher Zn contents than Andean lines (Islam et al., 2002). The high degree of correspondence with other reports for both Fe and Zn lends credibility to the current results.

One of the main challenges that common bean breeders working on biofortification face is the negative correlation between iron accumulation and yield (Raatz, 2018). Genotypes with high iron accumulation have been associated with poor yield and lower yield components (fewer pods, fewer seeds, fewer seeds per pod, or poorly filled seed) (Beebe, 2020). This same tendency is reflected in the correlations presented in Figure 3 and Supplementary Table 3 and has been reported previously (Diaz et al., 2020). Out of 14 loci for high iron, 6 were associated with low values of yield components (SdFe6.2, SdFe6.4, SdFe7.1, SdFe9.1, SdFe11.1, and SdFe11.2). These loci could contribute as much as 39% of the total additive value for high iron. Similarly, out of 12 loci for high Zn, 6 were associated with negative effects on yield components (SdZn2.1, SdZn2.2, SdZn2.3, SdZn6.1, SdZn7.1, and SdZn8.2), summing up to 49% total additive effect for Zn concentration. Several of these “false” loci were reported by other authors, so this has been a problem in the genetic analysis of mineral concentration for several years. On the other hand, the QTLs SdFe3.2, SdFe5.1, SdFe6.1, SdFe6.3, SdFe7.2, SdFe8.1, and SdFe11.3 were not associated with loci related with yield components, making them attractive for breeding through marker-assisted selection (MAS) because their selection would not adversely affect the yield. Another advantage of these identified QTLs is that some of these loci have been found in populations belonging to both gene pools, making them useful for improving Andean or Mesoamerican genotypes. The sum of additive effects not associated with poor yield could contribute as much as 40 mg/kg Fe. Even though it is likely not possible to maintain the full additive effects of these loci through genetic recombination, ample variability exists without affecting yield. Similarly, there are QTLs for Zn that were not co-located with QTLs related to yield components (SdZn3.1, SdZn4.1, SdZn5.1, SdZn5.2, SdZn8.1, and SdZn10.1). These QTLs can be implemented in the breeding program, which theoretically could increase Zn seed concentration by 10 mg/kg. Nonetheless, if limits exist to raising Fe and/or Zn concentration with the QTL studied here, additional sources of genetic variability for high minerals are being explored. Beebe (2020) reported that interspecific progeny derived from Phaseolus species that evolved on alkaline soils might be more receptive to raising iron levels. Initial phenotypic results with these interspecific crosses are promising but have not been studied yet to determine their genetic makeup.

The observed negative associations of seed mineral concentration with PHI and Sd/Pd, and in two cases with 100SdW, probably reflect simple factors of concentration of minerals. This may be due to a dilution effect of Fe and Zn caused by high carbohydrate translocation to the grain, a typical characteristic of genotypes with high yield and PHI. This has also been reported in other crops, such as chickpea (Diapari et al., 2014). Conversely, if seed mass is reduced, then as the iron in the pods is translocated to the seed, the concentration in the seed will be increased arithmetically. Furthermore, where selection for high iron has received priority in breeding, this will have unconsciously limited yield potential. These results reveal many of the causes behind the persistent negative correlation of yield and high minerals that have been the greatest challenge in breeding biofortified beans.

A strong phenotypic correlation between iron and zinc concentrations was observed here and has been reported previously (Blair et al., 2009; Cichy et al., 2009). Izquierdo et al. (2018) reported loci that contributed to high concentrations of both elements. They identified a MetaQTL (MQTL) denominated “MQTL_Fe&Zn_6.2,” containing eight QTLs identified in previous studies, such as sources of favorable alleles for Fe and Zn in both gene pools. MQTL_Fe&Zn_6.2 displayed higher PVE (27%) compared with other MQTLs identified in the same study. However, in this work, only one locus common to SdFe and SdZn was found, and this was associated with low values of PdN. This suggests that the correlation of Fe and Zn concentration that has been observed here and elsewhere may be driven in part by poor pod formation and/or poor pod and grain filling. The hotspot 12 is the only locus in which QTLs controlling the accumulation of Fe and Zn were co-located on Pv06 (∼28–29 Mpb).

The processes of iron and zinc uptake, transport, metabolism, and storage in sinks of interest are regulated by complex genetic mechanisms, of which several genes have been identified as possible candidates regulating these processes (Connorton et al., 2017). MQTL_Fe&Zn_6.2 mentioned above is located with Hotspot 12 (SdFe6.4/SdZn6.1), which is near the gene model Phvul.006G196600 that codifies a basic helix-loop-helix DNA-binding protein (bHLH). This gene is homologous to bHLH105/ILR3, an Arabidopsis gene codifying a transcriptional activator of responses to Fe deficiency, that also downregulates the expression of ferritin genes, implicated in the control of Fe homeostasis (Samira et al., 2018; Tissot et al., 2019; Gao et al., 2020). FER-LIKE REGULATOR OF IRON UPTAKE (FER) is required for induction of iron mobilization genes in Arabidopsis thaliana (Bauer et al., 2007). Phvul.005G430500 is homologous to FER and is located near SdFe5.1 (∼36.8 Mpb). Phvul.011G211900 is located near SdFe11.3 and is homologous to FRD3, which has a function of root xylem loading of iron chelator and efficient uptake out of the xylem into sink cells in Arabidopsis (Green and Rogers, 2004). AtNRAMP6 is a gene codifying a multispecific vacuolar metal transporter and plays an important role in the growth of Arabidopsis in Fe-deficient conditions (Li et al., 2019). Phvul.005G182000 (corresponding to SdFe5.1) and Phvul.009G127900 (corresponding to SdFe9.1) are homologous to AtNRAMP6. Phvul.003G143800 is homologous of AHA2, a proton pump H+-ATPASE2 controlling acidification-reduction-transport of iron uptake in Arabidopsis (Campos, 2020). The Zrt and Irt-like protein (ZIP) is well characterized for its role in Zn and Fe transporter (Astudillo et al., 2013); Phvul.002G184200 (SdZn2.2), Phvul.008G259200 (SdZn8.2), Phvul.010G059200 (SdZn10.1), and Phvul.011G058500 (SdFe11.1) have been identified as possible candidate genes belonging to ZIP metal ion transporter family.

In terms of conventional breeding, successful cases have already been reported using these populations evaluated for obtaining varieties with high accumulation of Fe and acceptable yield in different conditions: BIO-101 in Colombia (83 mg/kg of Fe and 44 mg/kg of Zn),³ SMR100 in Nicaragua (89 mg/kg of Fe and 35 mg/kg of Zn),⁴ and INTA Bio-Apante in Nicaragua⁵ were biofortified lines that used RILs of SMC40 × SCR16 population as parental materials.

An effect of drought was observed only on yield components, while Fe and Zn accumulation was not affected by the drought stress. The positive correlation between yield components has previously been reported under drought conditions (Polania et al., 2016a,b). Regarding the invariable Fe and Zn contents under stress conditions, Smith et al. (2019) explained that seeds have an evolutionary tendency to ensure adequate resources for germination and present buffering effects to maintain similar nutrient concentrations in drought and irrigated conditions, but the explicit mechanisms are not clear.

Pod harvest index expresses the ratio between the biomass in seed and pod and is a useful indicator of the remobilization of photosynthates to seed (Assefa et al., 2013). However, until now, it is not clear yet which of these two components have the key effect on PHI, and if there is a possible genetic relationship that explains the different behavior of PHI in each genotype. In this study, a stronger phenotypic correlation was observed between PHI with Sd/Pd and SdN than with PdN in all populations. However, the phenotypic correlation of PHI with 100SdW was low or not significant in this study (Figure 3), similar to results reported by Diaz et al. (2020) and Berny Mier Y Teran et al. (2019). Berny Mier Y Teran et al. (2019) suggested that sink strength due to the number of seeds is higher than sink strength based on the seed size. This could indicate that a key factor that affects PHI is the number of seeds rather than seed weight itself. In the genetic analysis, most QTLs identified for the number of seeds per pod (Sd/Pd) were co-localized with QTLs identified for SdN, and Sd/Pd has a high positive correlation with PHI. Moreover, all QTLs of PdN having a positive additive effect were associated with the SMC allele, which was parental with high micronutrient concentration but poor pod filling.

The main QTL found in this study for PHI (PHI 7.2), with the highest PVE and additive effect for the trait (Table 2), has been reported previously in Mesoamerican populations (Diaz et al., 2018; Berny Mier Y Teran et al., 2019) and in an Andean population (Sedlar et al., 2020), and could be used for breeding in both gene-pools. PHI1.1 and PHI8.1 have been identified previously in a Mesoamerican MAGIC population (Diaz et al., 2020). PHI1.2, PHI5.1, and PHI7.3 have been described in previous studies in Mesoamerican bi-parental populations (Asfaw et al., 2012; Diaz et al., 2018; Berny Mier Y Teran et al., 2019). PHI5.2 was reported by Sedlar et al. (2020) in an Andean bi-parental population. These QTLs seem to be independent of micronutrients content loci and could be used for MAS and theoretically could enhance PHI values by >20%, which discovers these markers very promising for breeding for drought tolerance and/or high yield. Other QTLs for other yield related-traits have also been reported, Berny Mier Y Teran et al. (2019) reported a QTL for seeds per pod near to Sd/Pd1.1. Similarly, Sedlar et al. (2020) reported a QTL near to Sd/Pd2.1, the QTL with the highest PVE. Other QTLs, such as PdN2.2, PdN4.1, SdN2.1, and SdN6.1, were reported previously in other Mesoamerican bi-parental populations (Diaz et al., 2017). Mukeshimana et al. (2014) found a QTL near to PdN8.2, this QTL was present in multiple trials.

One issue of PHI that has been discussed little is its role as a domestication trait associated with greater partitioning to grain, and whether the selection of genotypes with better pod filling was linked to other traits related to the domestication of common beans, such as pod indehiscence and determinate growth habit. PHI5.2 and PHI8.1 were co-located in the same genomic regions that controlled pod fiber or dehiscence for common bean (Parker et al., 2020). In the same way, PHI1.2 and PHI1.3 are near to the fin locus, a key region that controlled determinacy and photoperiod sensitivity in common bean (Kwak et al., 2012). This could indicate that PHI reflects traits that were enhanced in domestication and that loci associated with PHI could be related to domestication.

Understanding the segregation variation between the populations. Heterozygosity information in the populations is presented (Supplementary Figure 5). The utility of the early generation’s higher variability could be implemented as sources as a pedigree follow-up. In consequence, the analysis of the use of F₃ and F₄ populations was mandatory previously for further selection cycles.

This study evaluated the phenotypic and genetic correlation between Fe/Zn accumulation and yield components using QTL analysis in three biofortified populations developed in the CIAT bean-breeding program. The results in this work validate the negative phenotypic correlation previously described between yield and Fe/Zn accumulation in phenotypic terms. In addition, we are identifying several loci controlling both traits. We hypothesize this strong correlation is a dilution effect of Fe and Zn caused by high carbohydrate translocation to grain. Conversely, loci associated with high minerals and low values of yield components should be avoided. On the other hand, we observed QTL for Fe/Zn segregating independently of the other QTL for yield components, possibly mediated by transporters with affinity to minerals. The selection of these regions will enable enhanced levels of Fe/Zn and should not affect yield performance. QTLs observed in this and previous studies suggest similar genetic control in both gene pools at some loci and these QTLs show a high potential to be implemented in a breeding program focused on biofortification.

The original contributions presented in the study are publicly available. This data can be found here: https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/LI31YG.

SB conceived the study and designed the populations. CC and MG carried out the breeding populations advancement. JP did the field trial evaluation. SD did the phenotypic data analysis and linkage analysis. BR supervised the molecular laboratory. SD and DA-S performed the genotypic data analysis. SD and SB wrote the manuscript. All authors read and approved the final version of the manuscript.

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

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