This study was conducted in 2018–2020 at an experimental station (36°13′N, 114°35′W) in Anyang County, Henan Province, China. The site has a typical semi-humid climate with an annual average air temperature of 14.2°C and precipitation of 553.8 mm (1973–2017) (Li et al., 2020). In 2018, 2019, and 2020, annual growing season precipitation was 569.3, 454.1, and 317.5 mm, respectively, and the respective mean growing season temperatures were 22.6, 22.6, and 21.6°C (Figure 1). The soil type is classified as Inceptisols (USDA Soil Taxonomy). The topsoil (0–20 cm) of the experimental field soil is a sandy loam, with a pH (H₂O) of 8.03, organic matter of 13.8 g kg^–1, total nitrogen (N) of 0.9 g kg^–1, Olsen phosphorus (P) of 17.9 mg kg^–1 and available potassium (K) of 168.4 mg kg^–1 at the beginning of the experiment.

The experiment was a split-plot design with three replicates. The same layout was maintained in the three consecutive years. The main plot factors were seven N application rates (0, 60, 120, 180, 240, 300, and 360 kg N ha^–1). The subplot factors were two cotton cultivars (Jimian 228 and Lumian 28). Jimian 228 and Lumian 28 were used for their middle and mid-early maturity characteristics, with sprout-to-harvest times of approximately 130 and 125 days, respectively, and they are the main high-yielding cultivars grown in areas of the Yellow River Valley. Each plot was 28.8 m² in area, 3.2 m wide and 9 m long, and included four rows (north-south orientation). Adjacent plots were separated by a 15-cm-high ridge to avoid contamination caused by N dynamics between the treatments. The planting density was 6.0 plants m^–2 in rows spaced 80 cm apart.

Planting of cotton seeds took place on 28, 30, and 21 April in 2018, 2019, and 2020, respectively. All plots were overseeded and then manually thinned to the desired density at the squaring stage. Cotton plants absorbed about 58% of fertilizer N at 18–20 days after first bloom (Yang et al., 2013), thus allocating more fertilizer N to the later growth stages is of importance to enhance nutrient use efficiency and cotton yield (Yang et al., 2011; Geng et al., 2015). As such, two fifths of the N fertilizer (urea, 46% N) was evenly broadcasted and incorporated into soil at sowing, while the remaining N was applied with irrigation at the early flowering stage to avoid N losses (e.g., N volatilization). Flood irrigation was adopted according to the local farmers’ practice. In addition, P and K were applied as basal fertilizers at 90 kg P₂O₅ ha^–1 and 120 kg K₂O ha^–1 in the form of triple superphosphate (44% P₂O₅) and potassium sulfate (50% K₂O), respectively. Other field management practices, such as pest and weed control, were conducted following local practices and remained uniform and normal for all treatments. After harvest, cotton straws for each plot were wholly removed from the field, leaving the soil bare till sowing the next year.

Due to variation in maturity of the two cultivars, phenological events (plant emergence, squaring, flowering, and boll opening) were not synchronized but had little difference (Supplementary Table 1). In this study, therefore, sampling started only when 50% of observed plants in each plot reached flowering using the middle maturity cultivar Jimian 228 as a reference. Plant samples were collected at 0, 15, 30, 45, and 60 days after flowering (DAF) in 2018; 0, 17, 44, 60, and 81 DAF in 2019; and 35 and 60 DAF in 2020. Note that two samplings in 2020 were made to validate critical fruits dilution curve. For each sampling, three plants were randomly selected from the middle rows in each plot to avoid border effects. The samples were partitioned into leaves (including petioles), stems, squares, flowers and bolls. All samples were placed in an oven at 120°C for 30 min and then further dried at 80°C until a constant weight was reached. The dry mass (DM) of each plant organ was determined. Dried plant samples were finally ground to a fine powder, after which [N] was determined using the micro-Kjeldahl distillation method (Bao, 2005). Shoot N uptake was obtained from the sum of N in each organ, which was estimated from the product of DM and [N]. Six petiole samples from the fully expanded functional leaves (i.e., the 4th leaf from the apex) on the main stem were collected at each plant sampling date per plot in 2019. The petioles were placed into an ice box immediately after removal and transported to the laboratory for the measurement of nitrate-N (NO₃-N) concentration using Reflectoquant^® (RQ flex^@ 10 plus, Merck KGaA, Darmstadt, Germany).

We presented cotton yield and yield components in 2018 and 2019. The seedcotton was handpicked from the entire area of each plot on 12 September and 30 October in 2018 and 27 October in 2019. In 2019, only once cotton harvest was made, as the number of opening bolls was lower as a result of bad weather in September (heavy precipitation and low temperature, Figure 1). In both years, we also considered large bolls that opened only after sun drying when determining the final yield. The seedcotton was weighed after sun-drying. Prior to harvest, ten plants were randomly tagged in each plot to record observations on the number of open bolls per plant. For each harvest, all open bolls of the ten plants were collected in each plot, and dried to determine boll weight (seedcotton weight per boll). Lint percentage was calculated as the ratio of lint weight after ginning to seedcotton weight before ginning. The seedcotton yield was multiplied by the lint percentage to estimate the lint yield. In this study, lint yield was used as a proxy for cotton yield, as cotton production is mainly conducted to obtain cotton fibers. Shoot DM was determined as the seedcotton plus stalk yield.

Shoot DM, fruits DM, [N] and petiole NO₃-N concentrations for each sampling date, along with lint yield, RY and yield components at harvest were subjected to linear mixed effects models with N rate, cultivar and year set as fixed effects and N rate nested within replicate treated as a random effect. In the analysis of petiole NO₃-N concentration, year was excluded because data were from a single year. Where significant, differences between the means were compared using Tukey’s honestly significant difference (HSD) test at P < 0.05.

Non-linear regression analyses were used to examine the relationship between stems, leaves and fruits DM and stems, leaves and fruits [N]. The differences in two N dilution curves between cultivars or years were tested using analysis of covariance after data were log-transformed. We also used simple linear regression to investigate the relationships of NNI_sh, N rate, RY, boll density, boll weight, and petiole NO₃-N concentration with the NNI_f. All statistical analyses were performed in R version 3.6.2 (R Core Team, 2019), with the “nlme” package for the linear mixed effects model (Pinheiro et al., 2020) and the “basicTrendline” package for non-linear and linear regression (Mei and Yu, 2018).

Overall, N fertilization affected fruits and shoot DM in 80 and 95% of the sampling dates across cultivar-years, respectively. Fruits and shoot DM increased over the growing season, with values ranging from 0.10 to 8.52 t ha^–1 for fruits and from 1.36 to 15.70 t ha^–1 for shoot across sampling dates and cultivar-years (Supplementary Tables 2, 3).

In general, fruits and shoot [N] were significantly affected by N fertilization in 45 and 55% of sampling dates across cultivar-years, respectively. As expected, N concentrations decreased during the growing season. Across sampling dates and cultivar-years, fruits [N] ranged from 1.55 to 3.71%, while shoot [N] ranged from 1.38 to 2.77% (Supplementary Tables 2, 3).

Lint yield and yield components are shown in Table 1. Lint yield and yield components were considerably affected by the N rate and cultivar, but not by their interactions, indicating consistency of the N effect over cultivars and vice versa. Lint yield, boll density and boll weight changed little between N fertilizer rates except when no N was included, and no significant increase above 60 kg N ha^–1 was observed. Jimian 228 showed a higher lint yield and lint percentage than Lumian 28, with a significant difference between the 2 years. The averaged greatest lint yield and lint percentage was 1725 kg ha^–1 and 40.2% for Jimian 228, and 1655 kg ha^–1 and 38.8% for Lumian 28 across 2 years, respectively. Similar results were also observed for RY (Supplementary Figure 1).

Fruits [N]_c showed a gradually declining trend and the decrease was more rapid for the lower N fertilizer rates corresponding to the lower DM values (Figure 2). Fruits [N]_c decreased from a maximum of 3.21% to a minimum of 1.86% while the corresponding fruits DM increased from a minimum of 0.33 t ha^–1 to a maximum of 7.80 t ha^–1 across cultivar-years. Similar results were also observed for stems and leaves [N]_c (Supplementary Figure 2).

The decline in fruits [N]_c was closely related to fruits DM regardless of the climatic conditions of the year or cultivar, which was described by a negative power function (Figure 2). The N dilution curves between the two cultivars were not significantly different (P > 0.05). Therefore, a unique dilution curve across cultivar-years was fitted as follows: [N]_c = 2.49 × DM^–0.12 (R² = 0.649, P < 0.0001). However, the critical N dilution curves for stems and leaves were not well determined by a negative power function in some cultivar-years cases (Supplementary Figure 2). The explanatory powers of the stems and leaves N dilution curve model (R²) were also weaker (Figure 2 and Supplementary Figure 2). Moreover, fruits specific N dilution followed a similar trend as that in the case of the entire plant level, but it was a notably lower than the general curve for C3 crops (Figure 3).

The datasets in 2020 were used to test the accuracy of the critical fruits N dilution curve determined with the 2018 and 2019 datasets (Figure 4 and Supplementary Table 4). The fruits N dilution curve could effectively discriminate between N limiting and N non-limiting situations within the range for which it was established. Most data points from N limiting points were located under the curve, while those from N non-limiting points were close to and above the curve.

From the dilution curve developed for fruits DM, the NNI_f was calculated for all sampling dates. The NNI_f changed little between sampling dates. Across sampling dates, N rates and years, NNI_f values ranged from 0.822 to 1.033 (average 0.994) for Jimian 228, and 0.838–0.998 (average 0.943) for Lumian 28, respectively (Figures 5A,B). In both cultivars, moreover, N rates positively affected the NNI_f at most sampling dates (Figures 5A,B), and the highest NNI_f value was recorded with the highest N application. The NNI_f was not affected (P > 0.05) by cultivar on average (Supplementary Table 5).

The NNI_f values were highly related to the NNI_sh for both cultivars in most cases, with the exception of Jimian 228 at the fourth sampling after the flowering period (Figure 6), indicating the strong synchrony between fruits and shoot N status. When the NNI_sh was equal to one, the average NNI_f was 0.968 and 0.963 for Jimian 228 and Lumian 28 across sampling dates and years, respectively.

Across the 2-year period, the relationship between RY vs. NNI_f was described by a simple linear model under two cultivars at most sampling dates, except for 0 and 17 DAF for Jimian 228 (Figure 7A). However, RY could not reach 1 at an NNI_f of approximately 1. When the RY was 0.95, the NNI_f reached 1.143 and 1.063 for Jimian 228 and Lumian 28, respectively. Moreover, the responses of yield components to the NNI_f showed that boll density was positively correlated with the NNI_f in both cultivars (Figure 7B), although this pattern did not repeat in terms of boll weight vs. NNI_f (Figure 7C). On average, when the NNI_f = 1.0, the boll density reached 69.3 and 73.2 bolls m^–2 for Jimian 228 and Lumian 28, respectively.

The nitrogen fertilization effect (P = 0.0004) was more pronounced than the cultivar effect on the petiole NO₃-N concentration (P = 0.0141) (Supplementary Table 6). N fertilization increased the NO₃-N concentration but no significant increase beyond 120 kg ha^–1 was recorded. On average, the petiole NO₃-N concentration was 4,143 and 4,449 mg L^–1 for Jimian 228 and Lumian 28, respectively. Moreover, the petiole NO₃-N concentration of Lumian 28 was consistently greater than that of Jimian 228 during any growth stage and N application rate (Figure 8A and Supplementary Table 6).

Overall, there was a positive relationship between petiole NO₃-N concentration and the NNI_f under the two cotton cultivars, although it was not significant at 17 DAF for Jimian 228 and 60 DAF for Lumian 28 (Figure 8B). This relationship indicated that the NO₃-N concentration in the petiole was a good predictor of the NNI_f and consequently could be an alternative approach to evaluate fruits N status.

In the present study, significant effects on cotton shoot DM, fruits DM and [N] were observed in response to the N application rate. The shoot and fruits DM observed here were 8.88–15.7 and 4.02–8.48 t ha^–1 for Jimian 228, 6.9–12.43 t ha^–1 and 4.34–7.72 t ha^–1 for Lumian 28, respectively, which are comparable to those reported by previous studies, depending on the treatments applied and environmental conditions (Bange and Milroy, 2004; Zhang et al., 2008; Dai et al., 2015). For example, in a similar agro-ecological region (near our study site), aboveground DM was 12.7 t ha^–1 and fruits DM was 8.0 t ha^–1 at a high plant density of 10.5 plants m^–2 (Dai et al., 2015). Bange and Milroy (2004) also reported final shoot dry matter up to 18.7 t ha^–1 varying significantly between diverse cotton genotypes in New South Wales of Australia. In addition, significant effects of the N fertilizer rate on DMs in shoot and fruits were more pronounced than on the respective [N] in the same sampling date, suggesting that fruits and shoot DMs were more responsive to N fertilizer than the respective [N].

However, a limited cotton yield response to the N fertilizer rate was observed, with a significant cultivar effect, indicating that the effects of the cultivar were far greater than those of N fertilizer. Little or no yield response to fertilizer N application was also reported by Dong et al. (2012) in the Yellow River Delta of China (near our study site) and Stamatiadis et al. (2016) in central Greece. The former observed a yield response to the N fertilizer rate only at low plant density (3 plants m^–2). Contrary to our results, Li et al. (2017) found a linear increase in N concentration with increasing rates of N fertilizer despite the leveling-off or decline of yield. This conflict is likely derived from the difference between the critical soil N level or tested cultivar sensitivity to N fertilizer (Stamatiadis et al., 2016).

Fruits [N] gradually decreased with increasing fruits DM, due to the N dilution effect resulting from an increase in biomass accumulation that was greater than the N uptake (Lemaire et al., 2007). The N dilution effect has also been widely reported in several crop species, including winter wheat (Justes et al., 1994), maize (Ziadi et al., 2008b) and rice (Ata-Ul-Karim et al., 2013). In this study, the mean value of the [N]_c in fruits with DM of 1 t ha^–1 (2.49), estimated by the coefficient (a) of the N dilution curves, was much lower than that in the shoot biomass (greater than 5.0) (Xue et al., 2007; Zhang et al., 2008). This discrepancy may be related to the growth stage (Zhao et al., 2016a). The development stage when cotton fruits was equal to 1 t ha^–1 is later than that when shoot biomass was 1 t ha^–1. The dilution coefficient (b), however, was similar to that observed previously (Xue et al., 2007; Zhang et al., 2008), suggesting that there was a similar dilution process between fruits and the shoot as a result of the plants’ indeterminate growth habit. However, fruits N is more dependent on physiological N mobilization from stems and leaves, while only N added at fruit formation is directly mobilized to fruits (see below).

The critical N dilution curve for cotton fruits, in our study, had a notably lower dilution of N with increasing dry mass than a universal curve for C3 crops reported by Lemaire and Gastal (1997). The limited N dilution of fruits is partly explained by the indeterminate nature. Cotton plants continuously produce new organs, including fruits, simultaneously, which differs in many respects from C3 crops. Continued production of new fruits in cotton crops maintains a relatively high N content throughout the crop growth, which was attributed to the high translocation of N from stems and leaves to fruits. In other crops, relatively limited dilution was also observed in sunflower (Debaeke et al., 2012) and sweet pepper (Rodríguez et al., 2020).

In addition, fruits-based critical N curves showed small variations between the tested cotton cultivars differing in maturity. A recent study reported that the changes in plant canopy architecture may be a potential interpretation of variations in the N dilution curve (Ciampitti et al., 2021). Therefore, plant architecture, rather than precocity, needs to be considered with a focus on the genotypic effects on the N dilution curve in future studies.

Although the critical N curve developed for cotton fruits could effectively differentiate between N limiting and N non-limiting situations within the range for which it was established, note that this study was conducted in a single field site. Therefore, further study with additional sites will be necessary since the determination of [N] with shoot biomass might be affected by sites as reported for rice (Ata-Ul-Karim et al., 2013) and forage grasses (Agnusdei et al., 2010).

The critical N dilution curve allows for the determination of the NNI, which quantifies the intensity of both N deficiency and N luxury consumption for a given crop (Lemaire and Gastal, 1997). Values of the NNI close to 1 indicate that the crop is not limited by the N supply, while values lower or higher than 1 indicate N deficiency or N luxury consumption, respectively. The NNI_f changed little with cotton growth after the flowering period, indicating a stable N requirement in fruits after the flowering period.

The positive relationship between the NNI_f and NNI_sh under the two cultivars at most sampling dates suggested that the NNI in shoot biomass could be easily estimated using the NNI in fruits. These results were aligned with similar recent findings in which a close relationship between the NNI in vines or tubers and the NNI in total biomass was observed for potato (Giletto et al., 2020). Therefore, N nutrition sufficiency can be estimated with derived NNI_f values from fruits based on N dilution curve by only sampling for cotton fruits biomass, which would represent a fast and cost-effective option.

The linear relationship of RY with the NNI_f indicated that the RY was nearly 95% when the NNI_f was approximately one, and it decreased with a decreasing NNI_f below 1. Therefore, our results demonstrated that the NNI_f adequately identified the N status in cotton cultivars and had potential as a plant diagnostic tool to estimate the N status of cotton crops.

In this study, the NNI_f was used to predict the yield response to N fertilization and to develop strategies to manage crop N nutrition to match the N supply with crop demand after the flowering period. As a result, attention should be paid to the time lag if we use it to guide N fertilizer application in cotton production, especially at the early stage of cotton growth.

Although the NNI can be used to guide N management in crop production, the measurement is a laborious process that greatly limits the practical application of these diagnostic methods. Therefore, a rapid measurement method is needed. A good correlation between petiole NO₃-N concentration and the NNI_f was observed in our study, suggesting that the NNI_f could be estimated with petiole NO₃-N concentration. Moreover, the use of petiole NO₃-N has the advantages of being a more rapid and simple measurement than determinations of shoot DM and [N] (Wu et al., 2007). Therefore, this indirect method could potentially provide an alternative for estimating the NNI, avoiding time-consuming and labor-intensive shoot biomass sampling, and could characterize crops and environments in situations where the NNI cannot be measured directly.

Future research efforts should attempt to investigate whether other N diagnostic tools (e.g., chlorophyll meters, active crop sensors, and remote sensing) can be used to estimate the N status across different cultivars and environmental conditions (Liu et al., 2020; Shankar et al., 2020), which could complement our results.