A non-irrigated research study was conducted in two locations for four years (i.e., eight site-years) in Oklahoma. The experiments were established at the Oklahoma State University Research Station in Stillwater (36°08’24.9’N, 9705’37.0”W) and the Cimarron Valley Research Station in Perkins (35°59’25.2”N, 97°02’41.2”W) during 2019-2020, 2020-2021, 2021-2022 and 2022-2023 wheat growing seasons. The soil type for the experiment in Stillwater was Port silt loam (Fine-silty, mixed, superactive, thermic Cumulic Haplustolls) (USDA/NRCS soil taxonomy). The soil type for the experiment carried out in Perkins was the Teller series (fine loamy, mixed, active, thermic Udic Argiustolls) and Konawa series (fine loamy, mixed, active, thermic Ultic Haplustalfs). Weather information was acquired daily from planting to harvest (i.e., from early October to the end of June) from automated weather stations operated by the Mesonet Oklahoma weather network, proximately located to the research sites ( Table 1 ).

The field trials in Perkins were performed under conventional tillage, implemented in the fall before wheat sowing, and under a no-tillage system in Stillwater. The experiments were sown using a Great Plains Not-till drill (3P605NT) with a seeding rate of two million seeds ha^-1. Plots were 10.6 m long and 1.5 m wide, with seven rows 0.19 m apart. At sowing, 47 kg ha^-1 of ammonium polyphosphate (10-34-0) was applied along the seed furrow in all plots. Diseases, insects, and weeds were chemically controlled as needed. Soil fertility was evaluated at the time of sowing and after harvesting for each site-year ( Table 2 ). At sowing, 15 soil cores were obtained at 0-6 inches and 6-12 inches depth across the field and combined to represent a composite sample using hand probes. Two weeks after harvest, soil samples were collected from each individual plot at the same depths (0-15 cm and 15-30 cm) using hand probes. Samples were analyzed at the Oklahoma State University’s Soil, Water, and Forage Analytical Laboratory (SWFAL) to analyze pH, N, P, K, Mg, Ca, and SO₄ ( Table 2 ).

The experiment was arranged in a randomized complete block design (RCBD) with a 2x4 factorial treatment structure (i.e., two N rates and four genotypes) and four replications. The N rates evaluated were zero N (0N) application, which represented the conditions with only the residual soil N pool, and 120 kg N ha^-1 (120N) of urea (46-0-0) broadcasted in the fall, which represented conditions with an appropriate fertilizer application for the region (Zhang et al., 2017). The four winter wheat genotypes evaluated were Iba with Plant Variety Protection Act (PVP, 201300135), released in 2012; Gallagher (Gal) (PI 667569; PV 201300134), released in 2012; Doublestop CL+ (Dob) (PVP 201400228), released in 2013; and Green Hammer (Grn) (PVP 201900171), released in 2018. These genotypes were selected due to their adaptability to the environments explored in this study, their similar yield levels, and differences in grain protein concentration (GPC). Iba represented a low protein genotype, Gal a medium protein genotype, and Dob and Grn high protein genotypes (De Oliveira Silva et al., 2023).

Aboveground biomass was collected from one linear meter of a middle row at anthesis (Z59) and physiological maturity (Z92) at four replications. In the first and second growing seasons (i.e., 2019-2020 and 2020-2021), aboveground biomass was collected by the average growth stage of all genotypes. In the following two growing seasons (i.e., 2021-2022 and 2022-2023), samples were collected based on the growth stage of each genotype. Plants were cut with an electric clipper as close to the soil surface as possible at each growth stage. Plants were partitioned into leaf, stem, and spike at anthesis and leaf, stem, grain, and chaff at physiological maturity. Then, samples were placed in a dryer set at 65°C for seven days. The dry weight of each plant organ was recorded, and plant samples were sent to the OSU Soil, Water and Forage Analytical Laboratory for total N analysis using a LECO CN Combustion Analyzer and an Elementa CN Combustion Analyzer. Grain yield was estimated from the aboveground biomass samples collected at physiological maturity.

Plant N uptake was calculated by multiplying the aboveground biomass of each plant organ by its N concentration (%) (Ciampitti and Vyn, 2013), as

Grain protein concentration (GPC) was calculated by multiplying the grain N concentration (%) by the conversion factor 5.7 (i.e., 5.7 units of protein per unit of N) (Sosulski and Imafidon, 1990), as

The Nitrogen Harvest Index (NHI, %) was estimated as the ratio of grain N uptake to whole plant N uptake at physiological maturity (Ciampitti and Vyn, 2012), as

Post-flowering N uptake (PostN, kg ha^-1) was calculated as the whole plant N uptake at physiological maturity minus whole plant N uptake at anthesis, as

Remobilized N (RemN, kg ha^-1) was calculated as whole plant N uptake at the anthesis stage minus stover N uptake (i.e., stem + leaf + chaff) at the physiological maturity stage, as

Nitrogen remobilized from individual vegetative organs (organ Rem N, kg ha^-1) was calculated as the difference between the N uptake of each vegetative organ (i.e., stem, leaf, and spike) at anthesis and N uptake in the same organs at maturity (Ortez et al., 2019), as

Nitrogen remobilized from all vegetative organs from anthesis to maturity (Sum RemN_Veg) was calculated as the sum of the remobilized N in the three vegetative plant organs (i.e., stem, leaf, and chaff) (Ortez et al., 2019), as

The contribution of N remobilization from each vegetative plant organ ( Contribution RemN organ ,   i.e., stem, leaf, or chaff) was estimated as relative to the total vegetative N remobilization (Ortez et al., 2019), as

The actual plant N concentration (N_a, %) was estimated by dividing the whole plant N uptake (N_a, %) by whole plant biomass at anthesis, as

Critical N concentration (N_c, %) was determined using the critical dilution curve model for wheat from Justes et al. (1994), where 5.35% was the whole plant N concentration, when biomass was between 1.5 to 12 Mg ha^-1 and -0.442 the dilution coefficient (i.e., rate of decrease in whole plant N concentration as the biomass increases), as

The Nitrogen Nutrition Index (NNI) (Equation 11) was estimated by dividing the actual N concentration of whole plant (N_a, %) at anthesis (Equation 9) by the critical N concentration (N_c, %) (Equation 10), as

Linear mixed models were applied for evaluating the treatment effect on the traits measured. For each trait at each location, we fitted the model i.e., we considered genotype, N rate, and their interactions as fixed effects, and blocks nested within each year as random effects. The models can be generally described as

where y_ijkl is the trait being studied observed for the ith genotype, jth N treatment, kth year and lth block, µ is the overall mean, τ i is the effect of the ith genotype, ρ j is the effect of the jth N treatment, ( τ ρ ) i j is the interaction between the ith genotype and jth N treatment, u k ∼ N ( 0 ,   σ u 2 ) is the random effect of the kth year, v k l ∼ N ( 0 ,   σ v 2 ) is the random effect of the lth block in the kth year, and ϵ is the residual. All three u, v, and ϵ are assumed iid and independent among each other.

For the remobilized N and its contribution to grain N accumulation, we fit the model

where y_ijklm is the trait being studied observed for the ith genotype, jth N treatment, kth year, lth block at the mth location, µ is the overall mean, τ i is the effect of the ith genotype, ρ j is the effect of the jth N treatment, ( τ ρ ) i j , ( τ γ ) i k , ( ρ γ ) j k are the two-way interactions, ( τ ρ γ ) i j k is the three way interaction, is the random effect of the lth year, v l m ∼ N ( 0 ,   σ v 2 ) is the random effect of the mth block in the lth year, and ϵ is the residual. All three u, v, and ϵ are assumed iid and independent among each other. We performed an ANOVA using the models outlined in Equation 12 for the major traits (i.e., yield, GPC, plant biomass, plant N uptake). Next, we compared mean differences between treatments (i.e., combinations of N treatment and genotype) within locations using the Tukey adjustments and a significance level of 0.05.

To describe the relationship between grain yield vs plant N uptake at maturity, total biomass vs N uptake at maturity, and grain N accumulation vs N uptake at maturity, we compared simple linear regression (i.e., intercept and slope) to quadratic, and cubic. All models were fitted using R software. Linear mixed models were fitted using the R packages “lme4” (Bates et al., 2015) and “lmerTest” (Kuznetsova et al., 2017), and multiple comparisons were done with the “multcomp” package in R (Hothorn et al., 2008).

This study evaluated the ability of genotypes to produce grain yield while minimizing the trade-off with GPC by evaluating four winter wheat genotypes (selected for their known similar yield levels and varying GPC), two N rates (0N and 120N), two locations, and four growing seasons (from 2019-2020 to 2022-2023) resulting in a total of 256 observations. The results were reported for each site across years (i.e., individual sites were analyzed separately) independently.

The grain yield values ranged from 769 to 8005 kg ha^-1 with an estimated standard deviation of 1246 kg ha^-1, and GPC values ranged from 8 to 16% with a standard deviation of 1.5% points across all sources of variation evaluated (i.e., genotypes, N rate, and site-years). As expected, the genotypes tested in this study showed similar yield levels across N rates in both sites ( Figure 1A ) and different GPC at the high N rate in both sites ( Figure 1B ). Iba had the lowest GPC, Gal had a medium GPC, and Dob and Grn with higher GPC values. The high protein genotypes Dob and Grn had greater mean GPCs of 11.4% and 11.6%, respectively, than the low protein genotype Iba, with a mean GPC of 10.3% at the 120N in both sites. They also had greater GPC than the medium protein genotype Gal, with a mean GPC equal 10.6% in Stillwater. A quadratic model better illustrated the relationship between grain yield and total N accumulation at maturity (slope = NUtE) than a linear model ( Supplementary Table S1 ). The model showed a quadratic rise towards a maximum yield value, indicating that increases in total N accumulation at maturity resulted in greater yield up to a certain level, after which yield gains diminished as N accumulation continued to rise ( Figure 1C , Supplementary Figure S1 ). Thus, there was an overall positive and quadratic relationship between grain yield and total N accumulation at maturity, with 87% (for Perkins) and 89% (for Stillwater) of the variation in grain yield due to N rate, genotypes, and growing seasons explained by differences in total N accumulation at maturity ( Figure 1C ). In both sites, the association between grain yield and total N accumulation was investigated to quantify the N Utilization Efficiency (NUtE) of genotypes differing in GPC. This analysis further revealed that increases in yield via increases in NUtE are possible in high grain yield and N accumulation scenarios. Furthermore, in Stillwater, 70% of each high-protein genotypes Dob and Grn data points fell below the trendline, while 54% of the low-protein genotype Iba fell at or above the trendline, specially under low N uptake levels, indicating lower NUtE for high protein genotypes. In Perkins, 54% of each high-protein genotype Grn and Dob datapoints were below the trendline, while 66% of the low-protein genotype Iba were at or above the regression line, indicating a greater NUtE for Iba in comparison to the high-protein genotypes ( Supplementary Figure S2 ). Our results suggest that the high NUtE of the low protein genotype Iba may have penalized its GPC, and that the high protein genotypes Dob and Grn seemed to use different pathways to achieve GPC. The higher protein from the genotype Dob did not come from greater N uptake, while it could have for Grn.

We evaluated whether the high protein genotypes showed increased N and biomass accumulation compared to the low protein genotypes, and whether their accumulation dynamics would change when they were evaluated under low or high N conditions. Nitrogen rate increased biomass and N uptake of all genotypes, but the accumulation patterns between genotypes remained consistent under both low and high N conditions ( Supplementary Tables S4, S5 ). A strong positive association was observed between total N accumulation and biomass accumulation at maturity for both sites ( Figure 2 ; Supplementary Figure S3 ). A linear model better illustrated the relationship between total N and biomass accumulation at maturity ( Supplementary Table S2 ). At both sites, increases in total biomass were related to a corresponding increase in total N accumulation at maturity, irrespective of genotype GPC levels. Based on the r² of this relationship, 83% (Perkins) and 77% (Stillwater) of the variation in N accumulation due to N rate, genotypes, and growing seasons were explained by changes in biomass at physiological maturity. Hence, the stoichiometry of whole plant N uptake and biomass did not change for genotypes with high and low GPC.

There was a strong and positive association between grain N accumulation and total plant N at maturity in both sites (p<0.01), with approximately 96 and 95% of the variation in grain N accumulation across all sources of variation (i.e., N rate, genotypes, and growing seasons) explained by differences in total N accumulation at maturity in both sites ( Figure 3 , Supplementary Figure S4 ). This strong relationship demonstrates that genotypes with higher total N accumulation tended to partition a more significant proportion of their acquired N towards grain and that some genotypic variation may occur. For instance, in Perkins, 50% of the datapoints for the low protein genotype Iba were below the trendline while the 72% of the datapoints for the high protein genotype Dob were at or above the trendline, indicating a greater NHI for Dob as compared to Iba ( Figure 3 ; Supplementary Figure S5 ). Based on the slope of this relationship, the average NHI was 63% for Perkins and 67% for Stillwater. The consistency of these strong relationships in both sites (R²>0.90 and p<0.05) indicated that the greater NHI observed in the high-protein genotype is relatively stable and less influenced by environmental factors or N availability levels. Therefore, the strong partitioning capability of some high-protein genotypes enables them to maintain high GPC levels while achieving competitive yields. This supports the hypothesis that high-protein genotypes may exhibit increased N partitioning to the grain, irrespective of N availability.

High and low protein genotypes did not change their PostN uptake and N remobilization dynamics when tested under low and high N rate ( Table 3 ). While there were no statistical differences between genotypes and their interaction with the N rate, the 120N treatment increased the N remobilized from individual vegetative organs in both sites (p<0.01) ( Table 3 ; Supplementary Table S7 ). Thus, genotypes with different GPC levels had similar N remobilization from individual vegetative organs and the sum of all vegetative organs. When comparing the RemN of plant organs averaged over genotypes and growing seasons, the stem remobilized more N from anthesis to maturity than the chaff and was not statistically different from leaf in both sites ( Figure 4 ).

There was a positive relationship between the proportional contribution of RemN_leaf to the Sum RemN_Veg (%) and NNI, indicating that under favorable N conditions (higher NNI), a more significant proportion of the Sum RemN_Veg was contributed by the leaves relative to other organs ( Figure 5A ; Supplementary Figure S6 ). The high-protein genotypes (Grn and Dob) exhibited a higher proportional contribution of RemN_leaf compared to the low-protein genotype (Iba), as evidenced by the 84% of data points above the trend line for high-protein genotypes. This pattern was consistent across both environments, but slightly more evident at the Perkins site, where the high-protein genotype Grn showed greater leaf N uptake at anthesis than the medium-low protein genotypes Gal and Iba ( Supplementary Table S5 ). This suggests that genotypes with higher GPC tend to allocate a significant fraction of their Sum RemN_Veg from leaves to support N accumulation during grain filling period (hypothesis 2). Contrarily, the proportional contribution of RemN_stem to Sum RemN_Veg exhibited a very weak and negative correlation with NNI (Perkins: r² = 0.02, Stillwater: r² = 0.03) at both sites ( Figure 5B ; Supplementary Figure S7 ). This implies that as N availability improved, the relative contribution of RemN_stem decreased, perhaps, due to a shift in N allocation towards other organs, possibly leaf, for optimizing grain protein formation. Notably, the high protein genotypes Grn and Dob exhibited a more pronounced decrease in contribution RemN_stem as NNI increased compared to low and medium protein genotypes, as evidenced by their data points above the trend line. This finding suggests potential strategy high protein genotypes employ to prioritize grain N accumulation over vegetative biomass. Similarly, examining the relationship between the contribution of RemN_chaff to Sum RemN_Veg and NNI, a weak and negative correlation was observed in both Perkins and Stillwater sites (Perkins: r² = 0.13, Stillwater: r² = 0.10) ( Figure 5C ; Supplementary Figure S8 ). This negative association indicates that the chaff contributed a lower proportion of the Sum RemN_Veg under more favorable N conditions. Notably, while low and medium protein genotypes (Iba and Gal) showed a more pronounced reduction in the proportional contribution of RemN_chaff, the high protein genotypes (Grn and Dob) exhibited a distinct pattern. The high-protein genotypes exhibited a distinct pattern in N remobilization under varying NNI conditions. As NNI increased, these genotypes showed a lower proportional contribution of RemN_chaff to total remobilized N, with their data points falling below the trend line, especially at higher NNI values. Simultaneously, they demonstrated a higher proportional contribution of RemN_leaves. This pattern indicates that under improved N status, high-protein genotypes prioritized N remobilization from leaves over chaff, potentially as a strategy to enhance GPC while maintaining efficient overall N use.

The relationship between GPC and yield in wheat has been extensively studied due to its economic and end-use importance. Multiple studies have been conducted to explore the factors contributing to their negative relationship (Sieling and Kage, 2021; Tanin et al., 2022; Donaire et al., 2023), but further discussion is needed on the N uptake dynamics of genotypes with similar yield levels and different grain protein levels in dryland environments. Generally, the negative relationship between GPC and grain yield arises from competition for energy between C and N compounds in the plant, and N dilution in the grain as grain yield increases (Acreche and Slafer, 2009). The latter occurs when yield gains occur under limited N conditions (Munier‐Jolain and Salon, 2005). As demonstrated in our research and previous studies, genotypes differ in their magnitude of trade-off between GPC and yield (Cox et al., 1986; Bogard et al., 2010). Thus, understanding the physiological traits that enable genotypes to maintain or improve yield with minimal impact on GPC is relevant for the hard red wheat industry.

Wheat genotypes can perform differently under low and high N conditions. Cormier et al. (2016) found that certain wheat genotypes exhibited better tolerance to N deficiency, maintaining higher biomass and grain yield under low N conditions. However, in our study, genotypes did not differ in any of the traits evaluated under low N. This discrepancy between our findings could be attributed to several factors, including the severity of N stress in our experiment, which may have been more extreme than in other studies, potentially masking genetic variation. Gaju et al. (2014) observed that under severe N stress, relationships between N uptake, remobilization, and yield components changed compared to conditions of mild N stress or high N availability. Furthermore, genotype and environment interactions play a crucial role in the expression of N use efficiency (NUE) traits. Teng et al. (2022) reported that genetic factors accounted for a more significant portion of the variation in GPC in wheat. Gaju et al. (2011, 2014) observed that wheat genotypes differed in their ability to maintain GPC under varying N supply, with some genotypes maintaining higher grain N concentrations through more efficient N remobilization from vegetative tissues. Our study corroborates with previous findings regarding varied genotypic responses under high N conditions, particularly regarding yield and GPC. This is consistent with Bogard et al. (2010), who reported that high N conditions allowed for continued N uptake during grain filling, resulting in increased grain protein accumulation for some genotypes. Barraclough et al. (2010) found that high-yielding wheat genotypes often had lower GPC, particularly under high N supply. Regarding the differences in grain N sources under low and high N rates, Kichey et al. (2007) observed that under high N conditions, some genotypes showed increased post-anthesis N uptake, which contributed more significantly to grain N accumulation compared to N remobilization from vegetative tissues. This observation does not align with our results, where N remobilization increased as compared to post-anthesis N uptake under high N conditions.

The greater NUtE exhibited by low protein genotype Iba aligns with the concept of a physiological tradeoff between NUtE and GPC which is similar to findings of previous studies (Ivić et al., 2021; Teng et al., 2022; Thompson et al., 2022). Future research efforts should focus on identifying physiological traits and management strategies to optimize both NUtE and GPC to maintain or increase yields, reduce environmental nitrogen loss, and achieve desirable protein levels while considering economic factors.

Variations in GPC among genotypes may be driven by differences in other traits or N partitioning rather than total biomass. Dier et al. (2022) suggested that greater GPC in some genotypes may originate from their ability to partition N in the grain more efficiently. The subtle difference observed in Grn, however, suggests that some high-protein genotypes may indeed utilize increased N uptake as a strategy to enhance GPC. This finding underscores the complexity of the mechanisms underlying GPC variations and highlights the importance of considering both general trends and trait-specific strategies when breeding for improved grain quality.

Plants with greater total N accumulation at maturity tend to allocate more N to grain, as consistently observed in wheat across various studies (Barraclough et al., 2010; De Oliveira Silva et al., 2020). The relationship between grain N uptake and total N uptake, as reflected in the NHI observed in our study, aligns with and extends previous research findings. Consistent with studies by Montemurro et al. (2006) and Rahimizadeh et al. (2010), we found that NHI plays a critical role in influencing GPC across diverse wheat genotypes and environments. Our results revealed that high-protein genotypes (Grn and Dob) demonstrated higher NHI compared to the low-protein genotype (Iba), supporting our hypothesis that high-protein genotypes exhibit increased N partitioning to the grain relative to low-protein genotypes. A higher NHI indicates that more of the absorbed N is being utilized for grain protein production, potentially reducing the amount of excess nitrogen left in crop residues or lost to the environment (Kirika, 2021). This efficient use of N can contribute to reducing environmental nitrogen pollution, a major concern in agricultural systems.

While numerous studies have investigated management strategies and traits that could help in enhancing both GPC and grain yield without substituting one for the other (Dhillon et al., 2020; Lollato et al., 2021; Kartseva et al., 2023), no one has investigated the different remobilization strategies of genotypes with high and low protein under zero and high N conditions nor have they detailed the organ-specific contributions to N remobilization in modern wheat varieties. Across all genotypes, RemN_leaf and RemN_stem emerged as the primary contributors to grain N, followed by RemN_chaff. This aligns with previous research showing stems and leaves as significant sources of RemN during grain-filling (Masoni et al., 2007; Egle et al., 2015). Interestingly, the effect of the N rate on RemN_chaff varied depending on the environment, suggesting that environmental or management factors may influence its contribution (Dordas, 2009). Thus, the lack of significant differences among genotypes for PostN and RemN traits suggests they responded similarly to environmental conditions, underscoring the importance of N management in optimizing grain yield and GPC. These findings indicate that N availability, rather than genotypic differences, is the primary driver of these processes across all genotypes studied.

Consistently with previous research (Kichey et al., 2007), leaf N remobilization plays a critical role during grain filling to sustain plant N status and attain protein levels. Similar findings from studies by Barraclough et al. (2010) and Gaju et al. (2011, 2014) also posited that in wheat, remobilizing N from the leaves during grain filling determines the ultimate grain N concentration. Conversely, the negative correlation between RemN_stem contribution and NNI at both sites, suggesting a shift in nitrogen allocation as availability increases, consistent with findings by Barraclough et al. (2010). High-protein genotypes exhibited a more pronounced decrease in RemN_stem contribution with increasing NNI, suggesting a distinct physiological strategy that may favor grain protein accumulation over vegetative growth (Triboi and Triboi-Blondel, 2002). Similarly, RemN_chaff decreased as NNI increased, with high-protein genotypes showing lower RemN_chaff but higher NNI values. These results support our hypothesis that N remobilization patterns differ between high and low protein genotypes, aligning with Pask et al. (2012) and Gaju et al. (2014) proposition that high-GPC genotypes may have evolved efficient N remobilization mechanisms.

The enhanced leaf N remobilization observed in high-protein genotypes under increased NNI is particularly noteworthy, corroborating the importance of leaf senescence and N remobilization in determining final GPC (Masclaux-Daubresse et al., 2010). These differential N partitioning strategies, especially the prioritization of leaf N remobilization in high-protein genotypes, likely contribute to their ability to maintain elevated GPC while responding to N availability conditions (Cormier et al., 2016). These findings have important implications for breeding programs aimed at developing wheat varieties with improved nitrogen use efficiency and high GPC. However, future research should consider how these N partitioning strategies may be modulated under different environmental stresses, particularly in the context of climate change and increasing water scarcity in wheat-growing regions.