The experimental materials were collected from 419 rice landraces in Guangxi, including 330 indica rice, 78 japonica rice, and 11 other rice. Three biological replicates were set up for the NUE experiment, with 12 plants in each replicate. To break dormancy, the rice seeds were soaked in water for 24 hours at an ambient temperature of 28°C. Then, they were put on a damp cotton cloth at the same temperature for 24 hours to allow germination. The germinated seeds were sown in 96-well culture boxes and cultured with the normal nitrogen-level nutrient solution (1 mM NH₄NO₃). The nutrient solution was a thousand-fold dilution of Yoshida culture solution B (Yoshida, 1976). The seedlings were all placed in an incubator, with an ambient temperature of 28°C, humidity of 80%, light intensity of 40%, and light/dark cycle of 13/11 h. The seedlings were cultured for 20 d.

The phenotype parameters included plant height, plant fresh weight, plant dry weight, root length, root number, root branch number, root volume, number of root nodes, root fractal dimension, average root diameter, and root surface area. Plant height was measured from the base of the rice to the leaf tip. Plant dry weight was obtained by weighing the dried samples. The number of root nodes is the sum of the root number, root branch number, and the number of points at which the roots cross. The root fractal dimension is a direct indicator of root development. A higher root fractal dimension indicates a more developed root system, while a relatively small root fractal dimension suggests a weaker root meristem ability. The root data was obtained by scanning and analyzing the roots with the LA-S root scanning system (WSeen, Hangzhou, China).

Specific-locus amplified fragment sequencing was performed on an Illumina Hiseq 2500 system. The clean reads were clustered using BLAT software to obtain polymorphic SLAF tags. Then, the BWA software was used to align the polymorphic SLAF tag sequence to the Nipponbare reference genome (http://rice.uga.edu/). GATK and SAM toolkits were used to analyze SNP calling. A total of 208,993 SNPs were obtained based on a minor allele frequency (MAF) > 0.05 and a deletion rate < 0.5 (Yang et al., 2018).

The genetic relationship between samples was calculated using the Centcred_IBS module of Tassel. The population structure was analyzed by ADMIXTURE software. We conducted GWAS using the TASSEL software on 208,993 SNPs genotypes and seedling phenotype data. The MLM used a (Q+K) model, where Q was the population structure and K was the kinship coefficient. The SNPs with p < 2.39×10^–7 were considered to have a significant association. Manhattan and Q-Q plots were generated in the R environment.

At the same time, GEMMA, another commonly used software for GWAS, was used to conduct association analysis on 208,993 SNPs genotypes and seedling phenotype data. The SNPs with p < 1×10^–6 were considered to have significant correlations. Manhattan and Q-Q plots were also generated in the R environment.

The Nipponbare was used as the reference sequences (http://rice.uga.edu/). The candidate regions were selected based on LD decay of 419 rice landraces, with 150-kb intervals upstream and downstream of the SNPs showing a significant correlation.

The materials for extracting RNA were the variety with the smaller RN: C117, and the variety with the larger RN: C347. The total RNA was extracted with an RNA extraction kit following kit instructions. For reverse transcription, 1 μg of RNA was used as the template for reverse transcription to synthesize cDNA. A 20 μl PCR reaction system was prepared following the kit instructions. The system consisted of 1X RT buff, 1 mM d NTPs, 0.5 μM oligo-dT primer, and 0.5 U RNase inhibitor. The 20 μl RT-qPCR system was prepared with 2xUniversal Blue SYBR Green qPCR Master Mix (Servicebio, Wuhan, China), plus 0.4 μl forward/reverse primers, 2 μl product cDNA, and 7.2 μl Nuclease-Free Water. Actin was used as the internal reference gene (Gaur et al., 2012). All primers used for RT-qPCR are in Supplementary Table 1

The RT-PCR reaction was carried out on the BIO-RAD T100 Thermal Cycler PCR instrument (Bio-Rad, California, USA), and the RT-qPCR reaction was carried out on the BIO-RAD CFX 96 Touch fluorescence quantitative PCR imaging system (Bio-Rad, California, USA).

RT-qPCR data were analyzed using the 2^{^ΔΔCt} method (Thomas and Kenneth, 2008). Statistical analysis and plotting were performed using Origin 2022b and GraphPad Prism 8 software.

The scope of research on NUE covers the entire process of nitrogen uptake, transport, assimilation, redistribution, and signal transduction in plants, which are essential for the improvement of NUE. In this study, 11 traits relating to N uptake, transport, and assimilation were investigated, and the genes associated with these traits were identified under unoptimized N conditions.

The seedlings of 419 rice landraces had significant differences in traits such as PH, PFW, and PDW after 20 days of hydroponic cultivation ( Table 2 ). The data distribution of PH and RFD were less diverged, with a small coefficient of variation, indicating small genetic variations. The data distribution of PFW, PDW, RSA, ARD, and RL was more diverged and exhibited a higher divergence, indicating greater genetic variabilities. The data distribution of RBN, RN, NRN, and RL was the most divergent among all traits, with the largest coefficient of variation, showing a high degree of genetic variabilities.

We used Origin 2022b software to analyze the correlation among the 11 different traits and found that the PFW and PDW were highly positively correlated. Additionally, RSA and RV were also highly positively correlated; RBN was significantly positively correlated with RN, NRN, and RL, while NRN was significantly positively correlated with RBN and RL. On the other hand, ARD was significantly negatively correlated with PH, as well as multiple traits. Overall, the correlations among the root traits were significant ( Figure 1 ). To reflect NUE, we identified PH PH, PFW, PDW, RSA, RBN, RFD, RN, NRN, ARD, RV, RL under normal nitrogen level (Figure 2). After 20 days of cultivation, 419 rice landraces showed differences in 11 traits.C51 has the largest PFW, RFD, RV, PDW and RSA. The C93 has the largest ARD. C339 has the largest PH. C349 has the largest RBN, RN, RL and NRN.

Among the candidate genes associated with significant SNP from the MLM model, eight were known NUE-related genes, including OsAMT2.3, OsGS1, OsNR2, OsNPF7.4, OsPTR9, OsNRT1.1B, OsNRT2.3, and OsNRT2.2. OsAMT2.3 encodes an ammonium transporter protein. When nitrogen input is increased, the number of grains per panicle, thousand-grain weight, percentage of nitrogen in biomass, and protein content in grains between nitrogen-efficient and nitrogen-inefficient varieties show significant differences; the gene expression of OsAMT2.3 was also different in the flag leaves at the filling stage (Gaur et al., 2012). In this study, OsAMT2.3-associated SNPs were detected in RSA and RV, suggesting that OsAMT2.3 may also affect these two parameters. OsGS1 encodes a cytoplasmic glutamine synthetase. OsGS1 is expressed in all rice tissues and is highly expressed in leaves. When rice uses ammonium as the main nitrogen source, OsGS1 plays an important role in coordinating the entire metabolic network and can affect the normal growth and filling of rice (Tabuchi et al., 2005; Kusano et al., 2011). OsGS1-associated SNPs were detected in RSA. OsGS1 is also expressed in roots and is important for root growth. OsNR2 encodes nitrate reductase, which exhibits a high enzyme activity and is known for its sensitivity to hypochlorite as well as its ability to absorb high amounts of nitrate. OsNR2 promotes the expression of OsNRT1.1B in indica rice 9311, which in turn enhances the expression of OsNRT1.1B (Gao et al., 2019). OsNR2-associated SNPs were detected in RL, RBN, and RN. OsNR2 is expressed in the vascular tissue of rice shoots and roots, as well as the elongation zone of young roots, but OsNRT1.1B was not found in the linkage intervals of the SNP significantly associated with these three traits. It is possible that the interaction between OsNR2 and OsNRT1.1B was not obvious in the rice landraces included in this study. OsNPF7.4 encodes a nitrate transporter. Knockout OsNPF7.4 can increase seedling biomass, tiller number, seed number, and yield per plant, and excessive nitrate in plants with high OsNPF7.4 expression may lead to amino acid accumulation in leaf sheaths, thereby inhibiting the seedling biomass (Weiting et al., 2019). OsNPF7.4-associated SNPs were detected in PDW. Since the high expression of OsNPF7.4 expression could inhibit seedling biomass, it may also affect seedling PDW. OsPTR9 encodes a peptide transporter that localizes to the plasma membrane. The expression of OsPTR9 is regulated by exogenous nitrogen and circadian rhythm. The overexpression of OsPTR9 can increase ammonium uptake, promote the formation of lateral roots, and increase yield; the downregulation of OsPTR9 has the opposite phenotypic effect (Fang et al., 2013). OsPTR9-associated SNPs were detected in PDW. The positive effect of OsPTR9 on lateral root formation may increase PDW. OsNRT1.1B encodes a nitrate transporter, which is mainly expressed in root hairs, epidermis, and vascular tissues, and it is also highly expressed in epidermal cells and stele cells adjacent to the xylem in roots. OsNRT1.1B affects the NUE of rice by regulating the microorganisms with nitrogen transformation ability in roots (Hu et al., 2015; Zhang et al., 2019). OsNRT1.1B-associated SNPs were detected in RSA, PDW, RV, and PFW. Since OsNRT1.1B is expressed in roots and regulates root microbiome, root surface area and volume can affect the contact area between roots and microorganisms; thereby, OsNRT1.1B may be involved in the regulation of RSA and RV. OsNRT2.2 encodes a high-affinity nitrate transporter, which is up-regulated by nitrate and inhibited by NH₄ ⁺ and high temperature; its expression level is increased by light or exogenous sugar treatment (Feng et al., 2011). OsNRT2.3 is localized on the plasma membrane and is mainly expressed in the parenchyma cells of the root xylem. It is responsible for nitrate loading in the root and transport to the aerial part and has no effect on the uptake of nitrate in the root (Tang et al., 2012). Both OsNRT2.2 and OsNRT2.3-associated SNPs were detected in RV, so both OsNRT2.2 and OsNRT2.3 may have a certain effect on RV.

The key to plant development is the uptake of micronutrients and macronutrients by the root system. The use of these resources, which are often unevenly distributed in the soil, is optimized as root architecture responds to nutrient availability (Kemo et al., 2017; Hans et al., 2019). Previous GWAS studies have successfully elucidated the adaptive mechanism of root structure to nutrients (Bouain et al., 2018). In a previous study, 96 varieties of Arabidopsis thaliana were screened under nitrogen treatment, and the GWAS results on seven root traits showed that only one-third of the genes were associated with the same trait (average lateral root length) under two different nitrogen concentrations (Miriam et al., 2017). In this study, we analyzed the RSA, RL, RBN, RN, PDW, PH, RV, PFW, RFD, NRN, and ARD of 419 rice landraces in a nutrient solution. The phenotypic data showed a large genetic diversity and a large correlation between different traits. Through gene expression analysis, we found that the expression of LOC_Os05g51700 and LOC_Os05g51710 in C347 was significantly lower than that in C117, while the expression of LOC_Os05g51740, LOC_Os05g51780, LOC_05g51960, LOC_Os05g51970, and LOC_Os10g33210 in C347 was significantly higher than that in C117. Since LOC_Os10g33210 was found in RL, NRN, and RN, it is likely to participate in the regulation of these traits. Although there are expression differences among LOC_Os05g51700, LOC_Os05g51710, LOC_Os05g51960, and LOC_Os05g51970, the functions of the encoded proteins are unknown, and they belong to hypothetical proteins. LOC_Os10g33210 has been identified before (Jie et al., 2010; Leran et al., 2014). An intuitive tool called CRISPR-adapted Functional Redundancy Checker has been proposed to facilitate functional genomics in rice (Hong et al., 2020). LOC _ Os10g33210 was found to have protein sequences and expression patterns similar to several nitrate transporter proteins (https://cafri-rice.khu.ac.kr/inspector). LOC_Os05g51780 encodes a zinc finger protein, which belongs to a family of zinc finger transcription factors that can transmit signals. Some studies have found that zinc finger transcription factors are involved in nitrogen assimilation; under a condition with sufficient water, zinc finger transcription factors could up-regulate target genes OsNR1.2 and OsPrx24, which can improve nitrogen assimilation and promote the opening of stomata. Under osmotic stress, zinc finger transcription factors could down-regulate the expression of OsNR1.2 and OsPrx24, leading to inhibited nitrogen assimilation and stomatal closure, which enhances the drought tolerance of rice. LOC_Os05g51780 was also associated with RSA, RFD, PDW, PFW, and RV. In LOC_Os05g51690 and LOC_Os05g51830, although there was no significant difference in the expression of these two genes in C117 and C347, they were associated with RSA, RFD, PDW, PFW, and RV. LOC_Os05g51690 is involved in the response to the lack of macronutrients, and it affects rice growth. Through alternative splicing, LOC_Os05g51690 produces two transcripts with the same 5’ end, NRRa and NRRb, which encode 308aa and 223aa proteins, respectively. NRRa has one more CCT domain at the C-terminus than NRRb. NRRa and NRRb can regulate the structure of rice roots to improve macronutrient absorption; they also play a negative regulatory role in root growth and regulate the heading timing of rice (Yu-Man et al., 2012; Yuman et al., 2013). LOC_Os05g51830 is located in the nucleus and expressed throughout the life cycle of rice. It can regulate rice seed germination under abiotic stress conditions. The overexpression of LOC_Os05g51830 reduced the responses to abscisic acid (ABA), salt stress, and osmotic stress during seed germination and delayed seed germination (Zhao et al., 2014).

When analyzing the group structure, we divided the 419 rice landraces into five groups, while other studies divided the materials into six groups. The reason for the different groupings may be that the criteria used to filter SNPs were different. The GWAS results from the MLM model and GEMMA model were obviously different, and the MLM model generated much more significant SNPs than the GEMMA model. The different results of the two models may be caused by the strict SNP filtering in the GEMMA model. More than 200,000 SNPs were used in the MLM model, while less than 50,000 SNPs were used in the GEMMA model. On the other hand, the alleles associated with phenotypic diversity might have occurred in a low frequency, which made them hard to be detected by GWAS (Korte and Farlow, 2013).