A COMMERCIAL BREEDING PERSPECTIVE OF MAIZE IMPROVEMENT FOR DROUGHT STRESS TOLERANCE

– As the most produced grain crop in the world, maize ( Zea mays ) is a cornerstone of the global agricultural economy. Technological innovations in molecular genetics, environmental characterization, and predictive breeding have continued to drive genetic gain in maize, even for target populations of environments with high heterogeneity of water availability. Environments prone to drought stress remain key targets where genetic gain must continue to maintain a resilient food supply. Here we review advances towards improving maize drought tolerance; the review focused on molecular and physiological mechanisms underpinning drought tolerance, and methodologies that improve prediction of the genotype by environment interactions under drought conditions.


2
Maize is the largest grain crop in the world with an estimated 1,133.89 million metric tons produced in the 2020/2021 season worldwide (USDA, 2022), representing enormous socioeconomic importance for humanity. Maize is a critical source of food, feed, energy and derivatives with hundreds of applications in the industry. In Brazil, maize is the second most cultivated crop, with its largest use for feed, ethanol production and export. However, increasing environmental instability around the world, including in Brazil, has caused significant losses in maize yield production and poses risks to the resiliency of agricultural systems (Cunha et al., 2019).

Drought events have increased in frequency
and intensity in several regions of the planet in recent decades, and drought is associated with the most serious global economic and social losses, affecting more people than any other natural disasters (Cunha et al., 2019). Because of this, selection of droughttolerant genotypes with improved water use efficiency and production in these adverse environments is one of the most effective ways to improve the global productivity and stability of maize (Paterniani et al., 2015). Investments in maize breeding for drought stress tolerance is becoming more critical as the global climate continues to change, with increased temperatures, lower precipitation rates, and irregular distributions of rainfall, especially in the tropical and subtropical regions (Cairns & Prasanna, 2018).
Crop growing regions consisting of environments where water limitation is ubiquitous but variable across planting dates, locations and/or years results in cyclic economic and social losses and represent a challenge to genetic progress for yield and other relevant traits achieved through plant breeding. If genotypes of variable levels of drought tolerance are grown across a set of test environments that vary in the occurrence, timing and intensity of water stress, significant changes in the genotype relative performance between environments are expected to occur. This phenomenon, known as genotypeby-environment (G×E) interaction, complicates the identification of superior genotypes, bringing a high level of uncertainty to the selection process (De la Vega & Chapman, 2006. Several analytical approaches have been proposed to estimate the relative size of the G×E interactions encountered in a genotype-environment system and to describe their nature, repeatability and predictability, allowing to accommodate their effects through appropriate selection strategies aimed at exploiting broad and/or specific adaptation in maize (Löffler et al., 2005;Cooper et al., 2021). Similar methods demonstrated for sunflower can be most useful to inform decisions (De la Vega & Chapman, 3 to drought-prone environments. The characterization of drought environments often includes the use of crop growth models, tools that consider all elements to calculate a water balance for the crop and that could be applied over the geographical region to characterize the target population of environments (TPE). Based on the soil water supply and demand, it is possible to quantify the frequencies of type and intensity of the drought environment, and thus provide clear target for selection to the breeders (Löffler et al., 2005;Harrison et al., 2014). Cooper and Messina (2021) show the link across methodologies starting from stability analyses to the use of gap analyses to enable the use of enviromics in plant breeding.
As a fundamentally biophysical and biochemical process, maize drought tolerance is a complex and dynamic process. Water scarcity affects several processes in plant growth and development, including interruption of cell expansion caused by decreased turgor, adjustments in the photosynthesis rate, stomata aperture and regulation of osmotic pressure -exchange of cellular gases, activation of antioxidative reaction mechanisms and accumulation of plant hormones (Cramer et al., 2011;Benešová et al., 2012). The combination of these processes convey drought tolerance to manifest as a quantitative trait controlled by many loci which are highly influenced by different environmental conditions, including temperature, precipitation, and soil status (Bänziger et al., 2004;Waraich et al., 2011). As such, achieving genetic gain for drought tolerance requires characterization of the target population of drought environments to ensure the environments in which selection decisions are made will enrich for adaptive traits, or that any biotechnological traits introduced are expected to impact the proper response. Here, we review relevant topics for improving drought tolerance in maize, starting with genetic basis via associated genes and quantitative trait loci (QTL); to hormonal, and metabolic pathways; to physiological determinants; to methods for environmental characterization and GxE interactions for selection decisions; and finally, to the integration of these findings and concepts in commercial breeding programs.

The genetic basis of drought tolerance in maize
Genomic regions associated with drought tolerance in maize have been identified using linkage mapping, often performed in biparental populations or association panels as part of genome-wide association studies (GWAS). These approaches have been contributing to the rapid technological advances in maize breeding for drought tolerance (Zhu et al., 2016;Liu & Qin, 2021). In this section, QTL and major genes associated with drought tolerance will be discussed. Additionally, genomic regions associated with secondary drought-related traits will be presented to illustrate the complex architecture of drought tolerance in maize. Finally, limitations and applications of these findings will be highlighted considering different approaches focusing on maize breeding for drought tolerance.
QTL associated with drought-related traits have been widely identified in maize, including survival rate at seedling stage (Liu et al., 2013;Mao et al., 2015;Wang et al., 2016c), leaf rolling (Gao et al., 2019), leaf firing (McNellie et al., 2018), anthesissilk interval (Wang et al., 2016a), flowering time (Wang et al., 2016a), and root traits (Li et al., 2018a).  A natural genetic variation in ZmVPP1 gene, which encodes a vacuolar-type H+ pyrophosphatase and is located close to the QTL peak, was highly associated with drought tolerance (Wang et al., 2016c). A 366-base pair (bp) insertion in the promoter region of the ZmVPP1 gene, containing three MYB cis elements, conferred drought-inducible expression in drought-tolerant genotypes (Wang et al., 2016c). In addition, transgenic maize with enhanced ZmVPP1 expression had improved drought tolerance (Wang et al., 2016c). Similarly, Mao et al. (2015) showed that an 82-bp miniature inverted-repeat transposable element (MITE) insertion in the promoter region of the ZmNAC111 gene is significantly associated with drought tolerance in maize seedlings. The overexpression of ZmNAC111 conveys drought tolerance in Arabidopsis and maize seedlings, improves water use efficiency, and enhances the expression of specific stress-associated genes (Mao et al., 2015).
Combining phenotypic and genotypic data, He et al. (2018) (He et al., 2018). The expression of Clade A PP2C-A (PP2C-A) gene ZmPP2C-A10, which encodes a phosphatase that interacts with ZmPYL and other ABA receptors, is negatively correlated with drought tolerance in maize seedlings (Xiang et al., 2017). The overexpression of ZmPP2C-A10 in maize plants repressed ABA responsive genes ZmPP2C-A9, ZmNCED3 and ZmABF2, which are associated with drought tolerance (Mao et al., 2015), indicating its involvement in the ABA signaling (Xiang et al., 2017). A deletion in the promoter region of the  (Zhang et al., 2016).
Most of these loci seem to be involved in accumulation of carbohydrates and ABA-derived metabolites under drought stress, which has been suggested as a drought response in maize (Seki et al., 2007;Mohammadkhani & Heidari, 2008 Second, there is strong interaction between QTL and environmental conditions. In addition, some of the QTL identified may be limited to a specific genetic background, usually those identified from biparental populations, due to marker association and low-resolution mapping (Campos et al., 2004).
Third, although GWAS takes advantage of ancestral recombination to improve mapping resolution (Huang et al., 2014) and accesses more than two alleles when compared to biparental mapping, there are minor alleles with limited power to detect. Fourth, loci identified by linkage mapping and GWAS must be tested in hybrids to verify their potential application for breeding. Once the effect of each locus is generally small, heterosis of grain yield and stress response could mask the effect of individual loci (Zhang et al., 2016). In this sense, these findings are applicable to improve specific secondary traits, such as leaf rolling, which might be selected using causal genes or strong QTL, not necessarily improving drought tolerance.
The difficulty in working with individual genes is exposed by Simmons et al. (2021) who report that only 1% of the genes evaluated during 20 years of transgenic research manifested significant phenotypes under drought conditions. Different approaches must be considered to accelerate genetic gain for drought tolerance in a maize breeding program, including genome-wide prediction and selection.
The clustered regularly interspaced short palindromic repeat (CRISPR)/CRISPR-associated nuclease protein (Cas) system has emerged as a valuable tool to unlock genetic knowledge and create novel products (Nuccio et al., 2021). In this context, Shi et al. (2017)   .

Transcriptomic and proteomic associations with plant drought response
The Photosynthesis becomes limited after frequent or long-term water deficits (Xin et al., 2018) and one important aspect for the plants to cope with drought stress is the recovery of photosynthesis (Wang et al., 2016c). Proteomic studies have revealed that nearly 20% of total drought-responsive proteins were involved in carbohydrate and energy metabolism in leaves to cope with drought stress (Wang et al., 2016b). In maize, carbohydrate metabolism becomes limited at seedling stage after frequent or longterm water deficits (Xin et al., 2018), while drought stress at kernel filling decreased the remobilization of photosynthetic assimilates causing a reduction in ear growth, kernel filling, and size, with more pronounced occurrence in a susceptible line compared to a tolerant line . In seedlings, the abundance of most proteins known to be involved in carbohydrate metabolism gradually decreased as drought stress progressed, including enzymes involved in glycolysis, tricarboxylic acid cycle, and pentose phosphate pathway (Xin et al., 2018).
Proteins of starch metabolism and proteins of sucrose biosynthesis were also found with significantly reduced abundance upon drought (Schulze et al., 2021).
Maize plants also respond to drought by an increase in aquaporin abundance, which is reverted upon rewatering of the plants. The increase in water channels may be interpreted to allow cells to take up more water into cells and thus overcome decreasing water potential in soils (Schulze et al., 2021). An obvious response of plants to drought would be to regulate the cellular water transport. In fact, aquaporins are among the most regulated transporters at the plasma membrane. Their abundance in the membrane is adjusted as well as their activity (Besserer et al., 2012).

Morpho-physiological determinants of Genotype by Management in drought environments
Yield in field crops is determined by the harvested kernel number per unit land area and average kernel weight. Both traits vary across genotypes and environments in maize. Kernel number is the primary trait responsible for most yield variations (Early et al., 1967;Otegui, 1995;Chapman & Edmeades, 1999) due to the sensitivity of reproductive physiology to drought (Westgate & Boyer, 1985 Kernel number has traditionally been modeled as a function of biomass accumulation at the reproductive structure bearing kernels (i.e., ears), and the reproductive efficiency by which this biomass is used for setting kernels (Fischer, 1985). This biomass accumulation in the ears depends on both total plant growth and biomass partitioning to reproductive structures (Echarte et al., 2004;Borrás & Vitantonio-Mazzini, 2018). Under drought stress conditions there is a general response of reduced plant growth. In maize there is substantial native genetic variation in yield tolerance to drought stress generating reductions in plant growth. Different trait combinations can explain genotype differential plant yield reductions under reduced plant growth by reduced water availability.
These consider the relationship between plant growth and biomass partitioning to the reproductive structure bearing kernels during the flowering period ( Figure   1A).
The first common effect of reduced water availability is limited canopy growth. Plant growth is an integrative response and is commonly well captured by crop simulation models through its effect on canopy leaf expansion and light capture, or water capture and use. There are known traits that affect genotype plant growth under stressful situations that can be related to differential radiation use efficiency (Lindquist et al., 2005) or water use efficiency (Reyes et al., 2015). Root anatomy and morphology-related traits are described as key components of maize drought tolerance. Large cortical cells have been associated with deeper rooting, improved stomatal conductance, and higher CO 2 assimilation (Chimungu et al., 2014). In this sense, drought tolerance may be associated with a reduction in the metabolic cost of soil exploration.

Maize recombinant inbred lines with few and long
lateral roots showed substantially deeper rooting, higher leaf relative water content, and improved stomatal conductance compared to lines with numerous short roots (Zhan et al., 2015).
Other traits commonly referenced to understand differential hybrid responses to water stress are related to genotypic differences in plant biomass partitioning during flowering ( Figure   1B). This is usually pronounced when comparing new vs. old genotypes. In maize, the reproductive structure where kernels are set is an axillary ear located at the middle of the plant. This structure is not dominant and has poor biomass allocation under conditions of reduced plant growth. Whenever plant growth is reduced by limited water or nutrients, ear growth is reduced not only because the entire plant is accumulating less biomass but also because the proportion of the total biomass that is effectively allocated at the ear level is further reduced ( Figure   1B). This non-constant biomass partitioning to the   Figure 1A describes the genotype by water environment interaction for yield, and Figure 1B (Messina et al., 2011;Cooper et al., 2014b;Messina et al., 2019). This is associated with silk appearance under conditions of water deficit and is referred as reproductive resilience in Figure 2. In contrast, in the absence of water deficit and ample nutrient availability, plant canopy size, leaf N concentration, and radiation use efficiency are major determinants of yield potential through radiation capture and transformation efficiency (Figure 2 (gray area, Figure 2) .  (Cooper et al., 2014a;Messina et al., 2015). Because the traits are affected by temperature (Shekoofa et al., 2016;Rotundo et al., 2019) and regulate growth and water use, they can lead to genotype by environment interactions that are not evident to predict. Modeling approaches are necessary to integrate this knowledge and inform selection decisions (e.g., Messina et al., 2011;Cooper et al., 2014bCooper et al., , 2020Cooper et al., , 2021.

Predictive breeding for maize drought tolerance
The decreasing costs of DNA sequencing and the availability of informative markers led to a major shift in the scale of quantitative genetics ( Washburn et al., 2020), allowing the implementation of genomic prediction and selection in breeding programs. In genomic prediction, thousands of markers are fitted simultaneously to estimate the genetic value of individuals to predict their phenotypes (Meuwissen et al., 2001). This strategy of predicting genomewide effects to obtain accurate breeding values for individuals allowed for the acceleration of genetic gain per breeding cycle in traits with low heritability and complex genetic architectures (Heffner et al., 2009;Voss-Fels et al., 2019), and offers the opportunity to reduce the breeding interval cycle to at least half the conventional time and produces lines that, in hybrid combinations, significantly increase grain yield performance over commercial checks (Crossa et al., 2017).
Whole genomic prediction (WGP) has been deployed in the development of drought-tolerant maize worldwide, including private and public seed companies (Cooper et al., 2014a(Cooper et al., , 2014bCrossa et al., 2014Crossa et al., , 2017Messina et al., 2020). This technology has been used to predict the performance of untested maize single-cross hybrids for drought tolerance in multi-environment trials, with genomic selection using additive and additive plus dominance models (Dias et al., 2018). Genomic prediction was also conducted Additional approaches to account for environmental variation include models that integrate dynamic crop physiological growth models into genomic prediction as a link function (Technow et al., 2015;Cooper et al., 2016;Messina et al., 2018Messina et al., , 2020, models that use environmental covariance structures (Jarquín et al., 2014;Heslot et al., 2015), and models that incorporate environmental indices (Li et al., 2018b). Black-box machine learned models have also been applied to predict G×E interactions (Montesinos-López et al., 2018;Washburn et al., 2020).
Generally, these approaches all recognize that the genetic state of the plant (represented by markers) are being integrated with respect to their spatiotemporal environment to generate a phenotype of interest. In this sense, nonlinear dynamic models represent the closest conceptual approximation of this system, and these dynamic models essentially represent data compression; that is, how a given genotype will respond in a wide range of temporally varying environments is compressed into marker effects or genotype effects regulating the genetic parameters of the model. This concept is particularly relevant for applications like breeding for drought tolerance where significant GxExM is anticipated . A proof-of-concept study for integrating Crop Growth Models with WGP (CGM-WGP) through Approximate Bayesian Computation allowed the incorporation of CGMs directly into the estimation of whole genome marker effects in WGP (Technow et al., 2015). The CGM-WGP methodology was applied to an empirical maize drought MET data set and positive prediction accuracy was achieved.
However, new areas for further research to improve prediction accuracy and to advance the CGM-WGP for a broader range of situations in plant breeding, were identified . Later, the CGM-WGP methodology was improved, and using both synthetic and experimental data from a maize drought breeding program, there were realized advantages in prediction accuracy for yield, in both the water limited and the not water limited environments relative to the reference method BayesA (Messina et al., 2018;Diepenbrock et al., 2021).  (Cooper et al., 2014a. In the US, the first year of commercialization of AQ maize hybrids was in 2011. Since then, the average area of the US corn-belt planted with DT maize hybrids grew quickly to over 20% of the total area. In drought prone areas in the western US cornbelt, the land allocated to DT maize reached 39 % or more (Figure 4) Over thousands of comparisons and environments in contrasting geographies, AQ maize yielded 37 g m -2 more than non-AQ maize when exposed to drought stress conditions. Yield improvement under drought increased with planting density to at least 6.9 pl m -2 , where the yield difference was 50 g m -2 (Gaffney et al., 2015). Messina et al. (2020) showed that dedicated efforts towards drought breeding and, consequently, the launch of AQ hybrids led to a genetic gain in yield rate of 1.0-1.6% yr -1 in recent years under drought stress, which is higher than 0.7 % yr -1 genetic gain prior reported (Cooper et al., 2014a).     are specific to each hybrid, that will optimize its performance under drought conditions. One of these factors is hybrid positioning based on maturity, which represents an alternative way to prevent or diminish losses caused by water scarcity. Early maturity and/or early flowering hybrids tend to be less affected or even scape from drought at late stages by completing lifecycle prior the occurrence of drought. A combination of different maturities and mechanisms of drought tolerance are crucial to mitigate the risks of cropping in drought prone areas.

Final considerations
Climate change and population growth are driving demand for increased maize productivity and resilience, motivating the technological innovations necessary to meet this demand in an ecologically sustainable manner. Improved drought tolerance is a key trait to achieve these goals, and the research and commercial community have leveraged a wide variety of tools to explain the genetic and molecular basis of drought tolerance and accelerate the selection of superior germplasm. With the use of these technologies, the plant breeding community continues to achieve genetic gain in drought environments without penalties in non-stress conditions and improve the productivity of maize production systems.