Reproductive resilience but not root architecture underpins yield improvement under drought in maize

Abstract Because plants capture water and nutrients through roots, it was proposed that changes in root systems architecture (RSA) might underpin the 3-fold increase in maize (Zea mays L.) grain yield over the last century. Here we show that both RSA and yield have changed with decades of maize breeding, but not the crop water uptake. Results from X-ray phenotyping in controlled environments showed that single cross (SX) hybrids have smaller root systems than double cross (DX) hybrids for root diameters between 2465 µm and 181µm (P<0.05). Soil water extraction measured under field conditions ranged between 2.6 mm d–1 and 2.9 mm d–1 but were not significantly different between SX and DX hybrids. Yield and yield components were higher for SX than DX hybrids across densities and irrigation (P<0.001). Taken together, the results suggest that changes in RSA were not the cause of increased water uptake but an adaptation to high-density stands used in modern agriculture. This adaptation may have contributed to shift in resource allocation to the ear and indirectly improved reproductive resilience. Advances in root physiology and phenotyping can create opportunities to maintain long-term genetic gain in maize, but a shift from ideotype to crop and production system thinking will be required.


Introduction
Changes in root system architecture (RSA) were implicated in the determination of long-term yield improvement of maize in the US corn belt. Hammer et al. (2009) proposed the hypothesis that long-term changes in RSA in maize resulted in deeper root systems, and increased water capture and yield response to plant population density. This hypothesis is consistent with the observation that canopy temperature decreased with increasing year of commercialization in a set of hybrids grown under water deficit (WD; Barker et al., 2005), and simulation of breeding strategies for improved drought tolerance (Messina et al., 2011). Soil water extraction measurements among single cross (SX) hybrids commercialized between 1963 and 2009 showed no correlation of water extraction, year, or commercialization with yield (Reyes et al. 2015). Because the older double cross (DX) maize hybrids used prior to the 1960s are genetically and phenotypically more diverse than SX hybrids, it is possible that the shift in RSA, water extraction, and yield occurred prior to the 1960s.
In the USA, long-term crop improvement in temperate maize has resulted from pedigree breeding combined with reciprocal recurrent selection to improve hybrid performance (Duvick, 2001), and the optimization of agronomic practices such as planting density (Duvick, 2005;Cooper et al., 2014b;Assefa et al., 2018). Long-term gains have been demonstrated in irrigated, drought, and rainfed conditions (Cooper et al., 2014a;Adee et al., 2016). Beginning in the 1920s to 1930s, breeders used DX hybrids to exploit hybrid vigor and economically produce sufficient high-quality seed for farmers to plant at scale (Duvick, 2001). This process was replaced by SX hybrids in the 1950s to 1960s, when more productive inbred lines resulted from the breeding efforts. The finding that genetic gain is greater at higher plant population densities (Duvick, 2005) suggests that breeding has led to maize genotypes with greater tolerance to stress. The greater stress tolerance may be attributed to increased resource capture and/or enhanced reproductive resilience, but the relative importance of these two factors is unknown.
As elucidated in Darwinian agriculture theory, a common plant adaptation to cultivated systems is the reduction of intraspecific competitive ability (Denison et al., 2003). Because of genetic segregation and fewer cycles of selection, higher intraspecific competition, emergence of stratified plant sizes, and lowyielding dominated plants (Daynard and Muldoon, 1983) are expected in DX hybrids, but to a much lower degree in SX hybrids. The more genetically and phenotypically uniform SX germplasm can produce deeper and more uniform root systems than DX germplasm, where small plants contribute root mass only in the top soil horizons. This population-emergent phenotype of RSA can influence patterns of water uptake in a manner consistent with the hypothesis proposed by Hammer et al. (2009). Simulations of plant populations that account for plant to plant variation in RSA show greater root length density in the top soil horizons in DX relative to SX hybrids (Fig. 1A, B). Taken together, theory, model simulations, and observations of maize performance under drought stress suggest that populations of SX hybrids capture water better than populations of DX hybrids because of lower intraspecific competition, but this has not been experimentally demonstrated.
The consequence of the root phenotypes on crop water status and drought tolerance depend on the capacity of the root system to supply the water demand established by the crop leaf area (van Oosterom et al., 2016) and reproductive resilience expressed in the germplasm (Messina et al., 2011). Simulations of breeding strategies suggest that expression and contribution of traits to drought tolerance are conditional on each other, and their evolution can vary with cycles of selection (Messina et al., 2011). Understanding past changes in RSA and reproductive resilience can help identify opportunities to hasten genetic gain in the future. The objectives of this study were (i) to characterize RSA and soil water extraction in DX and SX maize hybrids; (ii) to test for an association between long-term changes in water extraction and RSA as postulated by Hammer et al. (2009);and (iii) to test the hypothesis that long-term genetic gain in temperate maize was in part determined by an increased capacity for soil water extraction.

Characterizing RSA in controlled environments
Root phenomics for RSA were conducted on a sequence of ERA maize hybrids (Table 1)  A custom X-ray system developed by the PSC was used to image roots growing in polystyrene containers (Fig. 2). The expanded polystyrene containers are nearly transparent in the images at the X-ray energy used (25 kV, 800 µA). Once placed in the X-ray chamber, a computer-controlled positioner moved the plant vertically and horizontally in pre-determined steps to capture eighty 5 cm×5 cm highresolution X-ray images covering the entire 1 m deep root system. An X-ray imaging system is conceptually similar to a pin-hole camera-based system. The X-ray beam began as a point source and spread out as a cone beam. The exposure time of each X-ray image was ~400 ms. The optical resolution of the system was ~58 µms. The resulting images were grayscale images, with denser and thicker root tissue being a dark gray to black and very fine diameter root tissue being a light gray.
Each root system was imaged twice-once at the V6 developmental stage and once at the V8 stage. Images were analyzed using RhizoTraits, version 1, a custom software developed by the PSC to extract root traits from X-ray images. RhizoTraits is built off ImageJ (Schneider et al., 2012). Eighty high-resolution X-ray images were combined to create a composite image for analysis of the whole root system (7526×18 194 pixels,137 MP). A PSC proprietary stochastic-based segmentation algorithm was used to identify root tissue within the images. Quantitative root traits are extracted from the images and for this experiment included (i) total root length and (ii) root system width, at each of 40 transects separated by 25 mm (Table 2; Fig. 2). Analyses were conducted for five root diameter size classes (SCs; Table 2; Fig. 2).

Soil water uptake contrasting hybrid and plant populations
Previous experiments to evaluate effects of breeding on drought tolerance involved measuring soil moisture content in current plant populations, which may have induced plant to plant variation, and thus overestimated the role of population uniformity in water uptake (Reyes et al., 2015). A field experiment was conducted in Viluco, Chile using the sequence of ERA maize hybrids (Duvick, 2005;Cooper et al., 2020; Table 1) to test the effects of breeding era, plant population, and water stress on soil water uptake (Experiment 2). The experiment included four hybrids from the SX and DX hybrid breeding eras (Table 1), which were replicated eight times in a split-split-plot design, with density as the main plot treatment, irrigation as the subplot treatment, and hybrid as the sub-sub-plot treatment. Two irrigation levels were imposed: low (high) WD treatments received 408 (621) mm of water for both high (10 plants m -2 ) and low (3 plants m -2 ) planting density treatment levels using drip tapes installed 20 cm belowground. The experiment was planted in four-row plots on 7 November 2014 and it was harvested on 1 April 2015. Rows were 4.2 m long and spaced by 0.76 cm. For prior research (Reyes et al., 2015), trenches were excavated to verify the adequacy of four-row plots for studies on water extraction in the Viluco environment. Roots from plants growing in adjacent plots were not observed at the center of the four-row plots where water extraction measurements were made. Soil moisture content was monitored using Time Domain Reflectometry technology (TRIME-PICO IPH/T3, IMKO Micromodultechnik GmbH, Germany). Access tubes were installed in the center of the plot to 1 m where a rock riverbed was reached. In addition to soil moisture measurements, kernel number, ear length, and kernel area per ear were measured using photometric imaging (Cooper et al., 2014b); plant height, leaf number, and size of the ear leaf were measured for two plants per plot. Time to flowering was measured for 10 individual consecutive plants, based on daily observations. For analyses, plants that did not flower after 86 d were assigned a value of 87 d. Leaf area was estimated by length, width, and a 0.79 multiplier. The ratio between water use and leaf area per plant was used to calculate root system efficiency (van Oosterom et al., 2016). The total number of leaves and leaf area of the largest leaf were measured as estimators of plant leaf area (Soufizadeh et al., 2018). Flowering notes and the proportion of barren plants were recorded, based on observations for 10 plants per plot. Yield was measured using imaging methods calibrated for the location (Cooper et al., 2014b).

Statistical analyses
Total root length and plant height from Experiment 1 were modeled within a linear mixed-effects model framework with the objective to test cross type contrasts between DX and SX hybrids, with named hybrids (h) considered as samples taken from a broader population of the hybrid class (c), where L ijkl is total root length for SC k of plant l of cross type i, and hybrid h(i), at growth stage j. In this model, cross type (DX or SX, c i ), growth stage (g j ), SC (s k ), and all two-way and three-way interactions were considered as fixed effects; hybrid [δ h(i) ] served as random effect. The residual term is ∈ ijkl ~N(0, σ 2 ilk ), which means that for each level of cross type, growth stage, and size class, a unique variance component σ 2 ilk is fitted in the mixed model. For the plant height variable, a linear mixed-effects model was fitted, (2) where T ijl represents plant height for plant l of cross type i at growth stage j. Other notations are the same as described for total root length. Cross type, growth stage, and their interaction were considered fixed effects; genotype served as random effect. The residual term of this model is ∈ ijl ~N(0, σ 2 il ), where σ 2 il represents the residual variance term among plants for cross type i at growth stage j, allowing each level of cross type and growth stage to have a unique residual variance component.
To test for heterogeneity of plant to plant variation between SX and DX hybrids under different growth stages or root SCs, a likelihood ratio method was applied using two nested mixed-effects models. For the plant height trait, a cross type fixed effect and a hybrid (nested in cross type) random effect are included in both models.
(3) For the full model (M1), the residual variance parameter depends on the level of cross type, ∈ il ~N(0, σ 2 i ). For the reduced model (M2), a single residual variance parameter is used for all observations, ∈ il ~N(0, σ 2 ). The P-value of the likelihood ratio test was calculated as . A similar approach was implemented to test plant to plant variation among cross types for root-level traits. For a given growth stage, the full model (M1), where, cross type, SC, and its interaction were considered fixed effects. Hybrid was considered as random effect. For root-level trait models, the residual term contributes to plant to plant variation. In the full model (M1), the residual variance parameter depends on the specific level of Width of the root system at the defined transect depth for roots in the defined size class CountDensity (n mm -2 ) Number of roots crossing the plane of the defined transect. The area of the plane is given by the cross-sectional area of the container used and the measured width of the root system at the defined depth Root traits are derived from features extracted from images. cross type and root SC, ∈ ikl ~N[0, σ 2 ik (∈)]. In contrast, the reduced model (M2) considering the residual parameter depended only on the root class level, ∈ ikl ~N[0, σ 2 k (∈)]. The models to test plant to plant variation for traits that vary with depth (d) conditional on growth stage, M1 and M2, were extended to include the variable depth and interaction with cross type and SC as fixed effects: where hybrid was considered as a random effect. Just like the models of root-level traits, M1 has a residual variance parameter depending on cross type and root size ∈∈ ikml ~N[0, σ 2 ik (∈)]. For the reduced model, the residual variance parameter does not vary between cross types, ∈ ikml ~N[0, σ 2 (∈)]. To test the heterogeneity of plant to plant variation between SX and DX hybrids for two growth stages for a certain root SC, a similar method was applied with the underlying factor as root SC. In this case, SC was replaced with growth stage in the previous models.
The width of the root system was modeled as a function of depth using a non-linear mixed-effects model, with the underlying non-linear function as a gamma-Ricker function: (7) where width (w) is set as the response variable and depth (d) as the explanatory variable. For all three parameters a 1 , a 2 , and γ, growth stage (j), cross type (DX or SX, i), and their interaction were considered fixed effects, and hybrid and plant (l) as random effects. Taking γ as an example, the mixed effect model is ] . Parameters and fitted curves were estimated for each root SC (SC3, SC4, and SC5).
Reproductive and vegetative traits from Experiment 2 were analyzed within a generalized linear mixed-effects model framework to test for differences between cross type DX and SX. For the traits with continuous numeric values, a Gaussian model with identity link was used. For the traits with count or fraction values (e.g. proportion of barren plants), a binomial model with logistic link was used. In the generalized linear mixedeffects model, cross type, plant population, location, and their interactions were considered fixed effects, while field spatial factor defined as row and columns, and named hybrid nested in cross type (Table 1) were considered random effects. The proportion of barren plants was modeled as where c i is cross type (DX or SX) effect, b j is plant population effect, l k is irrigation treatment effect, (cb) ij , (cl) ik , (bl) jk , and (cbl) ijk are the two factor and three factor interaction effects between cross type, population, and irrigation, α x,l and β y,l are row and column effects at each irrigation treatment with α x,l ~N[0, σ 2 l (α)] and β y,l ~N[0, σ 2 l (β)], δ h(i) is the named hybrid random effect with δ h(i) ~N[0, σ 2 i (δ)], and σ 2 l (α), σ 2 l (β), and σ 2 i (δ) are variance parameters for the three random effects in the model. A generalized additive model with integrated smoothness was applied to analyze the effect of cross type, plant population, and total depth (800 mm or 1000 mm) on the temporal dynamics of soil water content for Experiment 2. The dependent variable y was total available soil water (mm) and the independent variable was days after planting (x), where hybrid cross (DX or SX) (c i ), plant population (p j ), total depth (d k ), and their interactions served as the parametric terms. The functions f 1 (x;c i ), f 2 (x;p j ), and f 3 (x;d k ), are the smoothing terms by cross type, plant population, and total depth, respectively. Cubic regression spline bases with dimensions of 20 were used to fit the smoothing function f 1 , f 2 , and f 3 . All the non-parametric smoothing terms estimated here were centered at 0. In this model, each parametric term represents the overall magnitude of a certain fixed effect, while each smoothing term represents the pattern of the curve under each specific level of the corresponding factor. All the linear mixed models and generalized linear mixed models were estimated using Asreml version 3 (Gilmour et al., 2009). The non-linear mixed-effect models were fitted using R package 'nlme' version 3.1-144 (Pinheiro et al., 2020). The generalized additive models were fitted using the R package 'mgcv' (Wood, 2011).

Root system architecture changed with long-term selection for yield in maize
A maize hybrid set spanning a century of breeding (ERA hybrids; Table 1) was used as a case study to test the hypothesis that water capture underpins crop improvement in maize. This set comprises hybrids commercialized since 1920 that were widely adopted by farmers of the time. The sequence starts with the open-pollinated Reid Yellow Dent and ends with AQUAmax® drought-tolerant germplasm (Duvick 2005;Cooper et al., 2020). Maize SX and DX hybrids were exposed to contrasting water treatments and plant population densities to determine genetic gain in water uptake and yield. Root architecture was measured using X-ray technology (Fig. 2). Consistent with theoretical predictions (Hammer et al., 2009; Fig. 1A, B), we show that older DX hybrids had significantly greater root length than modern SX hybrids (Fig. 3B). The largest effect of long-term selection manifested on the upper soil layers (Fig. 4). Modeling RSA using a gamma-Ricker function, we were able to establish that root systems of DX hybrids were generally wider than those of SX hybrids, and the difference was significant (P<0.05) for roots of diameters between 725-2465 µm and 181-616 µm (Fig. 4A, C).
Following the principles of Darwinian agriculture (Denison et al., 2003), the observed differences between DX and SX hybrids in total root length (Fig. 3) and width (Fig. 4) could have been caused by plant to plant variation in root size due to genetic segregation (Duvick, 2001) and intraspecific competition (Daynard and Muldoon, 1983). No significant differences were detected for plant height. In contrast, significant plant to plant variation was detected for root traits, which were measurable in total root length for SC2, SC3, and SC4, root system width for all SCs except SC2, and density for all SCs (Tables  2, 3). However, root system width is an indicator of the outer bound of root occupancy of a given volume of soil but not how effectively this volume is explored by the root system. The root length to width ratio (LWR) provides a metric to assess plausible changes from DX to SX hybrids in their capacity to explore occupied volumes. The root LWR calculated from total root length and width, and for SC3, SC4, and SC5 (Table 2) at V8, were 0.17, 0.08, and 0.04 cm cm -2 for DX hybrids, and 0.17, 0.08, and 0.03 cm cm -2 for SX hybrids. We show that neither the allometry between root SCs nor the efficiency by which roots explore an occupied volume have changed between the DX and SX breeding eras.

Water uptake remained unchanged over eras of maize breeding
Results from the root morphology and water uptake experiments indicate that selection improved root system efficiency but not total soil water capture. SX hybrids had a smaller RSA per plant when measured in growth chambers (Fig. 4), but they captured the same volume of water as the DX hybrids (Fig. 5) despite the similar leaf area (Table 4). While patterns of water use differed between hybrid types (P<0.05; Fig. 5E, F), the capacity to capture water from the soil, estimated by the change in water content between 18 d and 74 d after planting, was similar for SX and DX hybrids and plant populations (Fig. 5E, F). SX and DX hybrids used water at rates of 2.7 mm d -1 and 2.6 mm d -1 , and of 2.8 mm d -1 and 2.9 mm d -1 when grown under low and high plant population density, respectively (Fig.  5E, F). Differences between hybrid types occurred during the late grain-filling period possibly due to the capacity of modern hybrids to maintain their leaf area under stress (Duvick, 2001(Duvick, , 2005. Dividing the rate of water use by the average size of the ear leaf (Table 2), an estimator of plant canopy size (Soufizadeh et al., 2018), and by total root length, an estimator of root system size (van Oosterom et al., 2016), here we show that DX hybrids have lower root system efficiency than SX hybrids (0.00012 d m -1 versus 0.00016 d m -1 ).
The difference in soil moisture between low and high density plant populations was 17±2.8 for DX hybrids and 13±2.8 mm for SX hybrids when the soil moisture reached a minimum value. This result suggests that plant population is the main controlling factor of root occupancy and water capture, and that there was enough water in the soil column to quantify differences in soil water capture due to variation in RSA if differences were present (Fig. 5E, F). At low plant population density, between 17 mm and 20 mm of water was measurable in the soil. This water could have been utilized by the hybrid group with larger root systems or canopies. However, soil water content was not significantly different between the DX and SX groups when the soil moisture was at the minimum under high density (-0.8±3.8 mm; Fig. 5E, F). No differences were observed despite DX hybrids presumably having larger root systems based on the X-ray study (Figs 3, 4).

Yield improvement in maize was driven by enhanced reproductive resilience
In contrast to results shown for water capture, yield and yield components were significantly higher for SX than DX hybrids across treatments (P<0.001). Yield of DX hybrids decreased with increasing barrenness, and barrenness increased with increasing density and WD (Table 4; Fig. 5A, B). Yield of both DX and SX hybrids decreased with decreasing kernels per ear (Fig. 5C, D). Kernels per ear decreased with increasing anthesis-silking interval (ASI) [y=528(±48)-67.26(±19.03)×x, r=0.81]. The lower absolute correlation between kernels per ear and ASI relative to yield and ASI (|r|=0.81 versus |r|=0.58) is associated with compensation between kernel weight and number within fertile ears (Borrás et al., 2004;Messina et al., 2019). The high association between barrenness and ASI for DX hybrids, and the observation of scatter grains (Fig. 5B) indicates that protandry induced by WD (Hall et al., 1982;Fuad Hassan et al., 2008) was a major driver underpinning a reduction in kernels per ear. Because both SX and DX groups reached anthesis at the same time for an irrigation and planting density treatment (Table 4), and with the same soil water content (Fig. 5E, F), we can rule out that differences in stress were due to timing of reproductive stages and soil water content. The higher ASI and barrenness observed for DX than for SX hybrids indicates that protandry for DX hybrids was long enough to miss at least part of the pollination window (Messina et al., 2019). In addition, WD caused significant reductions in kernels per ear, which were larger in DX than in SX hybrids (Fig. 5D), which implies differences between hybrid types in tolerance to water deficit beyond those explained by protandry alone (Fig. 5C, D; Table 4). Significant differences between plant populations and hybrid types in yield and yield components indicate variation in stress tolerance unrelated to water capture (Table 4; Fig. 5C, D), such as sequential floret development (Oury et al., 2016), reduced sensitivity of silk elongation to drought (Fuad Hassan et al., 2008;Messina et al., 2019), maintenance of carbon metabolism (Zinselmeier et al., 1999;McLaughlin and Boyer, 2004), and resource allocation to the ear (Edmeades et al., 1993).

Yield gain from iterative genetic and agronomic optimization
We conclude that selection did not operate to increase water capture per plant and that the higher reproductive resilience in SX hybrids is not a consequence of improved water capture as postulated (Hammer et al., 2009;Messina et al., 2011) and reported before for a cohort of SX hybrids (Reyes et al., 2015). Instead, yield improvement since the commercialization of DX hybrids in maize is related to improved water capture due to higher planting rates that translate into higher aerial mass and yield. The differential response measured as ASI, kernels per ear, and yield between SX and DX hybrids when exposed to the same level of WD (Fig. 5E, F) provides unequivocal evidence that genetic improvement of yield precedes changes in RSA. We propose a non-dichotomous view, whereby selection for yield led to improvements in reproductive resilience, which in turn enabled changes in the structure of the plant community including plant density. Changes in agronomic practices such as plant population density could have led to changes in optimal root architecture, which further contributed to exposing genetic variation for RSA traits and genetic  gain for yield. RSA adapted to increasingly crowded stands by decreasing the root system angle, increasing the efficiency of water uptake, and increasing reproductive resilience through shifts in carbon allocation. Reduced metabolic costs associated with smaller root systems, such as the phenotypes with reduced branching (Lynch et al., 2014;Zhan et al., 2015), could contribute to carbon reallocation. The lower total root length, higher occupancy of small roots, equal or higher total plant leaf area, and constant water uptake suggest that SX hybrids have higher root system efficiencies measured on a per root length basis than DX hybrids. The observation that both SX and DX hybrids capture the same amount of soil water at low   Pandey et al., 2021). The reduction in RSA width, rather than being a cause of improved water capture as proposed (Hammer et al., 2009;Messina et al., 2011), is a contributor to improved root system efficiency and stress tolerance through shifts in carbon allocation to the ear and increased water capture through increased plant population. By decreasing reproductive failure under stress conditions, enhanced reproductive resilience increases yield stability  and can explain observations that selection operated to reduce genotype×environment interactions for yield in temperate maize (Gage et al., 2017). Results conform to regional analyses indicating yield improvement across levels of WD (Lobell et al., 2014). Drought is an important component of the target population of environments where yield is >2000 g m -2 , which is the yield level estimated by Cooper et al. (2020) at which water becomes non-limiting to yield. Hence, we propose that improved reproductive resilience, as shown here, underpins the reported yield gains in the vast majority of the US corn belt (Lobell et al., 2014). Because water capture has not changed between SX and DX hybrids, the results are also consistent with the reported changes in relative sensitivities to WD, and the observation that maize yield association with soil plantextractable water is more marked in the drier regions of the US corn belt (Lobell et al., 2020). The relative importance of soil water capture over reproductive resilience increases with increasing WD.
The feedback between genetics and agronomy, and the evidence that the impact of root phenomics and selection on yield within breeding programs have been slow (Tracy et al., 2020), brings into question the feasibility of ideotype breeding in maize for root systems as proposed before (Meister et al., 2014). However, improved phenotyping capabilities as shown here and elsewhere (Zhu et al., 2011;Kuijken et al., 2015;Atkinson et al., 2019), and recent advances in understanding root elongation, efficiency, and responsiveness to water content, soil compaction, and plant density (Lynch, 2013;Lynch et al., 2014;York et al., 2015;Zhan et al., 2015;Orosa-Puente et al., 2018;Pandey et al., 2021;Schneider et al., 2021) can help accelerate the impact of root biology on yield improvement. In contrast to previous cycles of selection, a germplasm improved for reproductive resilience will express changes in RSA as improvements in yield under WD. However, there are limitations on the speed at which one can integrate root phenotypes within breeding programs due to the sequential and iterative nature of co-selection and adaptation (Cooper et al., 2014 ). Exploring adjacent spaces in the adaptation landscape (Messina et al., 2011), whereby shifts in traits are tested as hypotheses within the genotype×management systems, can be a more productive approach to accelerate yield improvement. Using crop modeling and genomic prediction, it is possible to identify maize crosses that vary in tolerance to drought and plausible root elongation rate. This hypothesis could be tested within a breeding program using advanced root phenomics. With a clear definition of breeding objectives and precision phenomics, prediction methodologies that integrate quantitative genetics and agronomy models Cooper et al., 2020;Peng et al., 2020) offer a path to accelerate genetic gain for multiple outcomes. It may be possible to simultaneously improve yield, carbon sequestration in the soil (Kell et al., 2012), and yield stability under drought by rebalancing growth relationships between roots, shoots, and reproductive structures.