Rainfed field experiments were conducted in Kansas, United States, near Belleville (39.81°N, 97.67°W; 471 m; moderately well-drained Crete silt loam) and near Hutchinson (37.93°N, 98.03°W; 468 m; well-drained Ost loam) during the winter wheat seasons of 2017–2018 and 2018–2019. Each environment will be referred to as Bel18, Bel19, Hut18, and Hut19. Winter wheat was sown under conventional tillage after a summer fallow using a Great Plains 606 no-till drill (7 rows spaced at 19 cm) with plot dimensions of 1.3 m × 9.1 m. Seeds were treated with 6.9-g a.i. ha^–1 thiamethoxam, 1.4-g a.i. ha^–1 mefenoxam, and 8.9-g a.i. ha^–1 difenoconazole to avoid early-season diseases and insects. Composite soil samples (i.e., 15 individual soil cores) were collected at sowing from the 0–15- and 15–60-cm depth to quantify initial soil nutrient status (Supplementary Table 1). Weeds were controlled and insect pressure was not observed across the study.

Treatments were arranged in a complete factorial structure established in a split-plot design with four replications. Whole plots were assigned to six management intensities, and sub-plots were assigned to four winter wheat genotypes. Treatment combinations represented stepwise increases in management intensity from a baseline reflecting the level of technology adoption of an average producer in the region and will, hereafter, be referred to as “farmer practice” (FP), “enhanced fertility” (EF), “ecological intensification” (EI), “increased foliar protection” (IFP), “water-limited yield” (Yw), and “increased plant productivity” (IPP) (Table 1).

The FP consisted of a seeding rate of 2.7 million seeds ha^–1 plus an N application at Zadoks GS23-25 with a rate reflecting a yield goal of the 10-year wheat grain yield average in the county where the experiment was located (∼2.4–2.8 Mg ha^–1). The first increase in intensity was the enhanced fertility (EF) treatment, which included 112 kg ha^–1 micro essentials (MESZ; 13-kg N ha^–1, 45-kg P ha^–1, 11-kg S ha^–1, and 1-kg Zn ha^–1) placed in a furrow with the seed, and an increased N rate for a 6.7 Mg ha^–1 yield goal applied at Zadoks GS23–25 in the spring (Table 1). The fertilizer treatments aimed at increasing tiller and biomass production. The N rate in this treatment was selected so that N was not a limiting factor based on the long-term wheat yield potential of ∼5.2 Mg ha^–1 (Lollato et al., 2017). The next step was ecological intensification (EI), which consisted of EF plus one fungicide application (fluxapyroxad-26 g ha^–1, pyraclostrobin-171 g ha^–1, propiconazole-107 g ha^–1) at Zadoks GS55. Increased foliar protection (IFP) was the next step, consisting of EI plus the same fungicide product and the rate applied at Zadoks GS31. The aim of these fungicide applications was to protect the green canopy cover of the crop (i.e., source) during the different stages of development. The water-limited yield potential (Yw) treatment consisted of IFP plus micronutrients (81-g S ha^–1, 90-g Zn ha^–1, 67-g Mn ha^–1, and 2-g B ha^–1) applied at Zadoks GS31. Finally, the increased plant productivity (IPP) treatment was designed to explore whether a high resource availability scenario allowed for reduced crop density; thus, the seeding rate was 1.1 million seeds ha^–1, reflecting the low seeding rates used by progressive growers in the region (Lollato et al., 2019b).

Wheat genotypes were selected based on their adoption by growers, adaptation to the region, and contrasting traits of interest for intensive management as well as performances in regional trials. The genotypes tested and their percent of the seeded area in central Kansas during 2020–2021 were WB4303 (<1%), WB4458 (2.2%), WB-Grainfield (5.5%), and Zenda (7.8%) (USDA-NASS, 2020). Information about traits of interest of each genotype within the context of management intensification is provided in Table 1.

The nitrogen rate was determined considering the soil NO₃-N measured at sowing, potential N released from the organic matter, and a 40 kg ha^–1-applied N per a Mg ha^–1 grain yield goal (Leikam et al., 2003). Due to the residual soil NO₃-N carry over from the previous growing season and estimated N release from organic matter, the N rate varied across environments (Supplementary Table 2). A pressurized CO₂ backpack sprayer with a three-nozzle boom was used to apply the N as urea ammonium nitrate (UAN, 28-0-0) with a streamer nozzle (SJ3-03-VP), and foliar fungicide and micronutrients using a flat-fan nozzle (XR11002) with a constant volume of 140 L ha^–1. Treatment application dates are provided in Supplementary Table 2.

Crop density was recorded in two linear meters per plot, 3–4 weeks after sowing. Percent green canopy cover was measured approximately at bi-weekly intervals from heading (Zadoks GS55) until maturity (Zadoks GS 95) from downward-facing digital photographs from an area of about 1 m² processed using Canopeo (Patrignani and Ochsner, 2015). Aboveground biomass was sampled from a one-linear row-meter area (∼0.19 m²) from one of the center rows of each plot the same day of wheat harvest. Samples were dried at 65°C until constant weight and dry aboveground biomass were measured. The spikes were counted and separated from the stover prior to threshing to remove the chaff from the kernels. Grain weight was measured after threshing. The grain weight divided by the total aboveground biomass weight (including stover, chaff, and grain) determined the harvest index (HI). A 1,000-kernel weight was determined from a random kernel sub-sample. The ratio between total grain weight and 1,000 kernel weight determined kernels m^–2; and the ratio between kernels m^–2 by spikes m^–2 determined kernels per spike. The number of productive tillers per plant was calculated as the ratio of spikes m^–2 and plants m^–2. Plots were trimmed prior to harvest to avoid edge effects, and wheat was harvested from a ∼13-m² area using a small-plot Massey Ferguson 8XP combine. Grain moisture was measured at harvest, and grain yield was corrected for 135-g kg^–1 water content. Grain protein concentration was measured using near-infrared spectroscopy.

Weather data, including precipitation, reference evapotranspiration (ETo), and maximum and minimum temperatures, were collected from a station pertaining to the Kansas Mesonet (Patrignani et al., 2020) located ∼50 m from the experiments. Plant available water at sowing was estimated using non-growing season precipitation and the soil’s available water-holding capacity (Lollato et al., 2016). At each environment, the weather variables were averaged (Tmax, Tmin) or accumulated (precipitation, ETo) for the entire growing season, as well as separated into four distinct phases: fall (the period between sowing and December 31); winter (January 1 to March 31), critical period [20 days prior to anthesis through 10 days afterward (Fischer, 1985)], and grain filling (10 days after anthesis through harvest). This sub-division intended to reflect (i) the conditions surrounding sowing that affect crop establishment and fall tiller initiation; (ii) the dormant period that can affect tillering and winterkill; and (iii) the yield determination period in the spring, similar to previous reports in the region (e.g., Lollato and Edwards, 2015).

Pearson’s correlation analysis was performed in R using the “corrplot” package (Wei et al., 2017) to determine the degree of linear association between the weather variables at the different periods and the measured crop variables. Because the data only derived from four environments, we relaxed the assumptions of p-values for this specific analysis to 0.15, while, for all other analyses, effects were significant at α = 0.05. ANOVA was performed using “lmerTest” in R software version 3.4.0 (Kuznetsova et al., 2017). Management, genotype, environment, and their interactions were fixed effects, while block nested within environment and management intensity nested within the block were random effects (the latter accounted for the split-plot design).

We used the stability method (Eberhart and Russell, 1966) to further understand the genotypic and management effects on grain yield and on productive tillers per plant (the latter to quantify the impact of crop density). This method consists of a linear regression of trait expression of each genotype (or management) versus an environmental index calculated as the mean trait expression of all genotypes at each environment minus the overall mean trait expression across all environments. Each management-by-environment combination was considered an environment (n = 24) for the genotype analyses (e.g., Ferrante et al., 2017; Lollato et al., 2021), and each genotype-by-environment combination was considered an environment (n = 16) for the management analyses (e.g., Raun et al., 1993). The slope (α) indicates whether the genotype has broad adaptability (α = 1) or adaptability specific to low (α < 1) or high- (α > 1) trait-expression environments, and is associated with phenotypic plasticity (Sadras and Richards, 2014). The intercept (β) is an estimate of the trait expression across environments; and a model goodness of fit index (i.e., R²) quantifies stability.

The modulators of yield in response to management were quantified as the relationships between yield components and yield using linear regression (e.g., de Oliveira Silva et al., 2020b). Differences in grain yield between the FP and each management for each genotype were calculated and regressed for: (i) all environment and management practices by wheat genotype combinations (n = 96), (ii) on average of each management intensity (n = 24; 6 managements × 4 environments), and (iii) on average for each genotype (n = 24; 6 managements × 4 genotypes). To understand the drivers of yield improvements in response to each step within the management intensification practices evaluated, we explored the relationships between the responsiveness of yield and the responsiveness of each yield component using linear regression (Slafer et al., 2014). Responsiveness was calculated as the ratio of each trait in a given management intensity over the same trait measured in the preceding management intensity so that we could quantify the effects of each management addition (e.g., responsiveness calculated as EF over FP associated with changes resulting from improved fertility).

Finally, we evaluated the green canopy cover data and the cumulative radiation intercepted during grain filling to better interpret the effects of fungicide and of crop density on grain yield in terms of source limitation. First, we calculated the linear slope of canopy cover dynamics between heading and maturity to detect whether the presence of foliar fungicides delayed canopy senescence, which would be indicated by a less-negative slope. This comparison was made between treatments EF and EI to isolate the effect of a single fungicide application at Zadoks GS55. Second, green canopy cover values at anthesis and their association with grain yield were compared for the Yw and IPP treatments to detect whether grain yield limitation from lower crop density could be explained by reduced green canopy cover. Finally, for the selected treatments above, cumulative radiation intercepted between anthesis and harvest maturity was calculated as the product between daily solar radiation and percent green canopy cover (Purcell, 2000). Daily values of green canopy cover were estimated for days between measurements using linear interpolation between consecutive measurements (Lollato and Edwards, 2015).

Growing season total precipitation ranged from 297 to 823 mm, and seasonal ETo ranged from 637 to 801 mm (Figure 1). Environments in 2017–2018 had a cold and dry fall, winter, and early spring, and a hot and dry late spring and early summer. Environments in 2018–2019 had warm and moist fall and cool and moist late spring and early summer, increasing disease pressure (Hollandbeck et al., 2019). Above-normal May and June temperatures in 2017–2018 (average temperatures between 23 and 27°C vs. 15–23°C in 2018–2019) accelerated and shortened the reproductive crop development (duration of grain fill, ranging from 27 to 29 days in 2017–2018 and from 33 to 52 days in 2018–2019; Figure 1), consequently decreasing the yield potential of the crop. The contrasting environments resulted in growing season length ranging from 239 to 288 days.

Table 2 shows the correlations between weather variables during specific crop developmental stages and yield components. Productive tillers plant^–1 related negatively with fall Tmin and positively with Tmin during the critical period. Harvest index related positively to winter Tmin. Spikes m^–2 related negatively to Tmin and precipitation during the winter. The negative relation between winter Tmin and spikes m^–2 or productive tillers plant^–1 reflects a delayed incorporation of the N fertilizer into the root zone until late spring in these environments, reducing the formation of spring tillers. Kernels spike^–1 related positively to precipitation and water supply during the season, fall and grain-filling precipitation, and duration of the grain-filling period; and negatively to Tmax (growing season, and at each stage evaluated), and Tmin during grain filling. Kernel weight associated positively with winter Tmin and precipitation, as well as critical period precipitation.

Across all sources of variation, mean grain yield ranged from 2.3 to 7.2 Mg ha^–1 (Figure 2). Environmental mean yield (across management and genotypes) ranged from 3.3 Mg ha^–1 in Hut18 to 5.6 Mg ha^–1 in Bel19, with overall greater yields in 2019 (5.43 Mg ha^–1) as compared to 2018 (4.28 Mg ha^–1). Mean grain yield for the genotypes was highest for WB4303 (5.11 Mg ha^–1), followed by Zenda (4.96 Mg ha^–1), WB-Grainfield (4.72 Mg ha^–1), and WB4458 (4.58 Mg ha^–1) (Figure 2A). Mean yield for the different management intensities was 3.96, 4.46, 5.34, 5.11, 5.34, and 4.82 for FP, EF, EI, IFP, Yw, and IPP, respectively (Figure 2B).

There were significant G × E and M × E interactions for grain yield, but no three-way interaction (Supplementary Table 3). General trends as related to the G × E interaction were: (i) WB4303 was in the highest yielding group at all environments; (ii) Zenda was in the highest yielding group in three out of four environments; and (iii) WB4458 yielded well in dryer conditions (i.e., Hut18) but yielded poorly at the higher-yielding environments (Bel19) (Table 3). General trends as related to M × E interaction were: (i) the FP yielded similarly to other treatments only in one environment (Bel18); (ii) EF yielded higher from FP in three environments; (iii) increases in grain yield from foliar protection (i.e., EI) only occurred in environments with greater rainfall (i.e., Bel19 and Hut19); (iv) the addition of the early fungicide (i.e., IFP) did not increase yields compared to a single fungicide application later in the season; (v) wheat grain yield benefited from all the management practices combined (i.e., Yw) only in one environment (i.e., Hut19); and (vi) reducing crop density under an otherwise highly managed system had no effect on grain yield except in one environment (i.e., Hut19) (Table 3).

Further exploration of the significant interactions through the adaptability and stability indices suggested that wheat genotypes varied in stability and adaptability across the different yield environments (Figure 2A). The wheat genotype WB4458 had the lowest α (0.78 ± 0.11), suggesting that this genotype was the least adapted to high-yielding environments and was unstable with a high variation about the fitted line (R² = 0.67). Due to their α equal to one (1.12 ± 0.09, and 0.96 ± 0.09), the wheat genotypes WB4303 and WB-Grainfield showed broad adaptability and greater stability (R² > 0.83), while Zenda was adapted to high-yielding environments (α = 1.18 ± 0.09). Management practices also showed environmental-specific adaptability, with EF showing greater yields in low-yielding environments (α = 0.72 ± 0.14), Yw showing adaptability to high-yielding environments (α = 1.45 ± 0.15), and the remaining management intensities showing broad adaptability (Figure 2B). Yield stability improved from the FP to the Yw treatments (R² ranging from 0.67 to 0.87, Figure 2B).

With the exception of 1,000 kernel weight and grain protein, the yield components followed the yield analysis and were not affected by the three-way interaction, mostly reflecting G × E and M × E interactions (Supplementary Table 3). Briefly, in terms of crop density, the IPP treatment had fewer plants m^–2 (149–163) as compared to other treatments (223–266 plants m^–2) as expected (Table 4), which resulted in more productive tillers per plant (3.18–4.97 vs. 2.16–4.22) (Table 5). Management intensification tended to increase aboveground biomass as compared to the FP (magnitude: 18–100%), while the latter usually resulted in the greatest HI—with exception of Bel19 –although the magnitude of change was not large (16–46%) (Supplementary Table 4). The magnitude in the differences in spikes m^–2 due to management and genotype was similar (38–72%) as those compared to changes in kernels spike^–1 (39–64%) (Supplementary Table 5). The results of kernels m^–2 reflected those for grain yield (Supplementary Table 6), while 1,000 kernel weight and grain protein were impacted by a G × E × M interaction (Supplementary Table 7).

Different genotypes had different tillering abilities and adaptation to tillering environments, which were mostly modulated by reduced crop density (Figure 2C). Zenda had the highest tillering ability across environments (mean: 3.81 productive tillers per plant) with even greater tillering expression in high-tillering environments (α = 1.18 ± 0.19), which was followed by WB-Grainfield, WB4458, and WB4303 (3.50, 2.97, and 2.75 productive tillers per plant) (Figure 2C). While WB-Grainfield and WB4458 had wide adaptability of productive tillers per plant, the ability of WB4303 to produce tillers decreased in reference to the other genotypes as tillering environment increased (α = 0.66 ± 0.13). Reduced crop density (IPP) allowed for the greatest expression and maintenance of tillers (mean of 4.43 productive tillers plant^–1), which increased at α = 1.71 ± 0.36 with the environmental index for tillering production (Figure 2D). The lowest tillering production and response to tillering environment occurred at the FP (mean of 2.29 productive tillers plant^–1, α = 0.46 ± 0.20).

Across E, M, and G, aboveground biomass at maturity explained 77% of the variation in yield, showing a positive relationship (Figure 3A). Although significant, a negative relationship of HI only explained 8% of the variation in yield (Figure 3D). Across environments, differences in grain yield were dependent on differences in biomass accumulation (Figure 3B) and independent of differences in HI (Figure 3E). Following the same trend, differences in biomass accumulation among the different wheat genotypes under different managements were also strongly related to differences in grain yield (Figure 3C) as compared to HI (Figure 3F). Increasing management intensity (the difference of each management practice to FP) significantly increased biomass accumulation, which increased yield across environments (Figure 3B, insert). Likewise, increased management intensity increased the responsiveness of biomass accumulation for wheat genotypes, which increased grain yield (Figure 3C, insert). Meanwhile, increased management intensity had limited effect on HI across environments or across genotypes (Figures 3E,F, inserts).

Kernels m^–2 had greater importance in increasing grain yield as compared to kernel weight (Figure 4). Across E, M, and G, a positive relationship of kernels m^–2 explained 78% of the variation in grain yield (Figure 4A), while no relationship between kernel weight and yield occurred (Figure 4D). Averaged across genotypes, increasing management intensity increased grain yield through differences in kernels m^–2 (Figure 4B), and yield responses to management practices were associated with increases in kernels m^–2 (Figure 4B, insert). Similarly, averaged across management practices, wheat genotypes that had greater kernels m^–2 also had greater grain yield (Figure 4C), and yield responses were dependent on the genotype’s kernels m^–2 responsiveness (Figure 4C, insert). Following a different trend, increases in grain yield were independent of kernel weight for both management practices and wheat genotypes (Figures 4D–F); however, increases in kernel weight due to management were associated with increased grain yield within environment (Figure 4E, insert). Differences in kernel weight within each genotype were not associated with increases in grain yield (Figure 4F, insert).

Spikes m^–2 and kernels spike^–1 both had a positive effect on grain yield (Figure 5). Across G, E, and M, a positive relationship of spikes m^–2 and of kernels spike^–1 explained 19 and 39% of the variation in yield, respectively (Figure 5A). Averaged across either management practices or wheat genotypes, grain yield differences were dependent on differences in spikes m^–2 (Figures 5B,C). Likewise, wheat genotype responsiveness to spikes m^–2 resulted in positive differences in grain yield (Figure 5C, insert). Interestingly, management practices resulting in greater number of kernels spike^–1 also significantly affected yield (Figure 5E), but there were no differences across genotypes (Figure 5F). Likewise, the responsiveness of kernels spike^–1 to management practices affected grain yield, with no differences among genotypes (Figure 5F, inserts).

Each stepwise increase in management intensity modulated different yield components (Figure 6). In the first step (i.e., addition of enhanced fertility to the FP), the responsiveness of yield ranged from 0.85 to 2.22 (mean: 1.23 ± 0.03) and was positively linked to the responsiveness of the productive tillers plant^–1 (range: 0.48–4.28, mean: 1.49 ± 0.06), biomass (range: 0.50–4.26, mean: 1.40 ± 0.06), spikes m^–2 (range: 0.53–2.75, mean: 1.40 ± 0.04), and kernels m^–2 (range: 0.39–4.12, mean: 1.44 ± 0.06) (Figure 6, first row). We also note that yield responsiveness was positively associated with grain protein responsiveness (range: 0.93–1.52, mean: 1.11 ± 0.01) when fertility drove yield increase. When one fungicide application was added to the EF, yield responsiveness ranged from 0.77 to 1.82 (mean: 1.20 ± 0.02) and associated positively with responsiveness of biomass (range: 0.61–1.86, mean: 1.14 ± 0.03), spikes m^–2 (range: 0.55–1.68, mean: 1.06 ± 0.02), and kernel weight (range: 0.79–1.58, mean: 1.11 ± 0.02) (Figure 6, second row). The addition of an early fungicide application to the EI had very weak relationships of yield responsiveness (range: 0.65–1.36, mean: 1.02 ± 0.01) to the responsiveness of biomass (range: 0.58–1.44, mean:0.97 ± 0.02) and HI (range: 0.70–1.84, mean: 1.08 ± 0.03) (Figure 6, third row). Likewise, the addition of micronutrients to the IFP treatment suggested that responsiveness of biomass (range: 0.77–1.67, mean: 1.12 ± 0.06) and of HI (range: 0.62–1.43, mean: 0.96 ± 0.03) associated with responsiveness of yield (range: 0.81–1.53, mean: 1.06 ± 0.01) (Figure 6, fourth row). Finally, when crop density was reduced from the Yw, responsiveness in yield (range: 0.61–1.21, mean: 0.91 ± 0.01) was positively related to responsiveness of biomass (range: 0.45–1.53, mean: 0.89 ± 0.03), of HI (range: 0.54–2.20, mean: 1.08 ± 0.03), and of kernel weight (range: 0.75–1.47, mean: 1.05 ± 0.02), and negatively related to responsiveness of plants m^–2 (range: 0.32–2.75, mean: 0.80 ± 0.06) and protein (range: 0.72–1.10, mean: 1.00 ± 0.01) (Figure 6, fifth row).

The slope of green canopy cover dynamics following fungicide application, as well as the cumulative radiation intercepted during the grain filling period, was positively associated with grain yield for the selected treatments that allowed for a direct comparison between fungicide and non-fungicide application (EF versus EI) (Figures 7A,B). Likewise, the difference between slopes of these treatments or intercepted radiation was positively related to grain yield difference (Figures 7A,B, insert). For each individual slope, intercept, and regression fit, please refer to Supplementary Table 8. Following a similar trend, green canopy cover values measured at anthesis, and the cumulative radiation intercepted after anthesis for the Yw and IPP treatments, related positively with grain yield (Figures 7C,D), as did their differences (Figures 7C,D, insert), providing empirical evidence for the reason behind decreased yields from reduced crop density in an otherwise well high-input system.

Our results align well with previous literature reporting that, across all sources of variation, wheat grain yield relates closely to aboveground biomass and kernels m^–2, and is relatively independent of harvest index and kernel weight (Slafer et al., 2014; Ferrante et al., 2017; de Oliveira Silva et al., 2020b). However, an original contribution of our research is the detailed yield responsiveness analysis and its relation to yield component responsiveness for each individual step in management intensification (Figure 6). To our knowledge, this has not been previously attempted in the existing literature of wheat response to management intensification. From this analysis, it was clear that the yield responsiveness was greater for added fertility (EF) and one fungicide application (EI) (mean responsiveness of 1.20–1.23) as compared to the remaining practices (mean responsiveness of 0.91–1.06). The added fertility drove improvements in yield mostly through greater number of productive tillers plant^–1, biomass, spikes m^–2, and kernels m^–2, while the added fungicide modulated yield through biomass, spikes m^–2, and kernel weight (Figure 6). Interestingly, the reduced crop density mostly decreased yield (responsiveness: 0.91) through reductions in biomass (responsiveness: 0.89), although there was some compensation through increased in harvest index (responsiveness: 1.08). The remaining practices only slightly modulated biomass and harvest index, having little effect on yield.

The modulation of yield through kernels m^–2 driven by the added fertilizer (EF) is justified as both in-furrow P fertilizer, and N fertilizer increases tiller initiation (Spiertz and De Vos, 1983; Rodríguez et al., 1999), and N fertilizer can reduce floret abortion (Ferrante et al., 2010; González et al., 2011). Tiller production determines the potential spikes m^–2, and floret development determines the potential kernels spike^–1. Both yield components interact with environmental conditions to determine kernels m^–2, which were highly positively related to yield (Figure 4). Thus, N availability has to meet the requirements for both of these processes during the growing season as untimely N deficiency can result in floret abortion and reduce kernels m^–2, potentially reducing yield. Nitrogen rates offer an opportunity for increased yields (Lollato et al., 2021), especially in favorable seasons where the crop can capitalize on a greater yield potential (Cruppe et al., 2017; Lollato et al., 2019a). Expected N uptake based on yield potential can serve as a guide for managing N rates in the season (Leikam et al., 2003); and, for wheat, a recent synthesis of global literature has suggested that N uptake ranges from ∼20 to 400 kg N ha^–1 (de Oliveira Silva et al., 2020a). Thus, matching N availability with the time when the potential number of kernels m^–2 is determined (i.e., early stem elongation) results in yield increases as grain number is the dominant driver of yield (Borrás et al., 2004; Slafer et al., 2014). We also note that this developmental stage coincides with the greatest N uptake rate by the crop, which increases under intensive management (de Oliveira Silva et al., 2021).

Kernels m^–2 and kernel weight are affected by complex interactions among many environmental factors in the late reproductive stages. Our results support available literature that suggests that kernels m^–2 is a coarse regulator of wheat yield as compared to kernel weight (Borrás et al., 2004; Slafer et al., 2014), which is justified as each individual kernel has a narrow range in size (Sadras, 2007); thus, greater increases in grain yield come from filling more kernels (Borrás et al., 2004). We note, however, that increases in kernel weight through management associated positively with increases in yield (insert, Figure 4E), in particular through the application of foliar fungicides (Figure 6). These findings agree with previous reports of highly managed wheat in the U.S. Great Plains (Lollato and Edwards, 2015; Jaenisch et al., 2019; Cruppe et al., 2021) suggesting that kernel weight might, in some conditions, partially explain increases in yield for wheat.

Foliar diseases can occur prior to anthesis and last throughout the grain-filling period, coinciding with a period of significant demand for photosynthesized resources by the developing grain (i.e., a very strong sink; Fischer, 1985). These foliar diseases decrease the green leaf area of the plant (Schierenbeck et al., 2019), reducing radiation interception and radiation use efficiency (Schierenbeck et al., 2016), and ultimately decreasing the source of assimilates to the developing sink. This mismatch between a reduced assimilate supply (i.e., source) during a period with large demand can cause kernel abortion and reduce yield (Ferrante et al., 2010; González et al., 2011). Foliar fungicides can also increase kernel weight under severe disease infestations, which can reflect increases in grain yield (Cruppe et al., 2021), although this increase is environment specific (Lynch et al., 2017). Wheat kernel weight is sensitive to environmental stresses (e.g., heat or drought) between booting to anthesis when carpel (which will turn into the external grain structures) growth increases rapidly (Calderini et al., 2001), and from anthesis to maturity during kernel weight determination (Bergkamp et al., 2018). Foliar diseases during these developmental stages can reduce kernel weight, which could reduce yield (Figure 4E, insert; Figure 6). Similarly, increases in kernel weight associate with kernel-filling rate, and foliar diseases can reduce the rate of fill due competition for assimilates (Simmons et al., 1982).

Foliar fungicides maintain the yield potential at time of application by protecting the upper canopy and spikes, which supply a large portion of the carbohydrates that determine yield (Rawson et al., 1983) and can increase kernels m^–2 (Brinkman et al., 2014). The prolonged green leaf area maintained through fungicides also allows for longer duration of active photosynthesis, ultimately increasing N uptake (de Oliveira Silva et al., 2021) and grain yield (Joshi et al., 2019; Nehe et al., 2020). This was shown in the current research as a more negative slope of the green canopy cover dynamics and a lower cumulative radiation interception after anthesis in the treatments not receiving foliar fungicides (Figure 7). The positive relationship between the slope of canopy cover and grain yield also suggests that treatments not receiving foliar fungicides were, at least, to some extent, source limited, which was also evidenced by the greater grain protein concentration of treatments receiving foliar fungicides (data not shown). Further evidence for this source limitation is shown in the inset of Figures 4E, 6, in which increases in kernel weight through management associated positively with yield increases. However, we note that large reductions in the green leaf area or radiation intercepted were needed to cause modest reductions in yield (Figure 7), likely because wheat is mostly sink-limited and very efficient in translocating stem reserves to the developing kernels (Borrás et al., 2004). Even though foliar fungicides applied around anthesis have increased wheat yield and reduced the yield gap in the region (Thompson et al., 2014; Jaenisch et al., 2019), producers may be reluctant to apply it consistently due to high environmental unpredictability (Couëdel et al., 2021) and inconsistencies in yield response (Cruppe et al., 2021).

The evaluation of a reduced crop density under an otherwise highly managed system (IPP) suggested that yield responsiveness was negatively related to responsiveness in plants m^–2 (Figure 6), reflected on the overall yield reduction of IPP as compared to Yw (4.82 vs. 5.39 Mg ha^–1; Supplementary Table 3). This aligns with findings suggesting that crop density is an important determinant of the yield gap in rainfed wheat (Tokatlidis, 2014). Furthermore, it seems like the opportunity to reduce crop density in dryland conditions for winter wheat might not be as evident as that for irrigated spring wheat in low latitudes (Fischer et al., 2019), likely due to the unpredictability of conditions for tillering in the fall, which is dependent on many environmental variables (Tokatlidis, 2014). Nonetheless, we showed that there was a large genotypic component of tillering plasticity (Figure 2C) that might be further explored in this region. Tillering allows wheat plants to compensate for a low crop density, with greater opportunities in higher-yielding environments (Bastos et al., 2020), which was shown in this study with the IPP producing more tillers than other treatments. Tillering plasticity regulates the ability of a given genotype to tiller in different environments, which also interacts with crop density. Thus, a wheat variety with high-tillering potential and tillering plasticity (e.g., Zenda, Figure 2) has the ability to produce more productive tillers at reduced density (Figure 2C) and modulate yield through harvest index and kernel weight (Figure 6). On the other hand, a variety with low-tillering potential and plasticity (e.g., WB4303, Figure 2) is reliant on higher crop densities to attain desirable yields because individual plants are inefficient in using available resources (Tokatlidis, 2017). Evidence for other cereals suggests that high phenotypic plasticity of tillering can result in increased panicle weight under low-seeding rates (Kikuchi et al., 2017). Thus, selecting wheat genotypes for increased tillering capacity through conventional breeding could help reduce the risk associated with low-crop density (Fischer et al., 2019), which aligns with the early concept (Fasoulas, 1973) and more recent developments (Tokatlidis et al., 2006; Fasoulas, 2013) of selecting per-plant yield under nil competition.

Wheat genotypes responded to the environment differently but not to management practices or to the interaction of management and environment. Thus, our findings suggest that wheat genotypes have to be adapted to specific reoccurring environmental conditions or broadly adaptable and have other desirable agronomic traits, such as high-yield potential (Ferrante et al., 2017), disease resistance (Serrago et al., 2011), heat or drought stress tolerance (Bergkamp et al., 2018), to match those commonly experienced in the environment where the genotype is grown. While the lack of G × E × M in our data might result from the limited number of observations (i.e., four environments), previous research in the region also only found weak evidence for G × E × M in response to management intensification (p = 0.14; de Oliveira Silva et al., 2020b).

The wheat genotype WB4303 was better adapted to higher-yielding environments and responded to increased environmental index by producing more kernels m^–2, which was highly correlated with increases in grain yield (Figure 4). These findings agree with those for other growing regions where modern genotypes were more adapted to higher-yielding environments and led to the hypothesis that the growers use older genotypes in their lowest-yielding soils and modern genotypes in their highest-yielding soils (Ferrante et al., 2017). While we did not test this hypothesis in Kansas, our findings suggest that this could be a promising strategy as the older genotype WB4458 was more adapted to lower-yielding environments, although further research is needed on this topic. For producers, selecting newer released genotypes might offer opportunities to capitalize on their ability to capture greater yields in higher-yielding environments (Slafer and Andrade, 1993; Perronne et al., 2017; de Oliveira Silva et al., 2020b) despite the challenge of finding information on new genotypes coupled with their limited life span (Perronne et al., 2017).