Reduced N fertilization had a significant impact on eco-physiological, developmental and agronomic traits
A number of eco-physiological traits classically used to evaluate maize growth and productivity at key stages of plant development were measured to evaluate the impact of reduced N. In the 29 hybrids, a two-way ANOVA analysis (P < 0.05 after FDR correction) showed that in the LN condition, a significant decrease was observed in 17 out of the 25 eco-physiological traits (Table 1). An increase was observed in only one of them (Sen80) when plants were grown under reduced N fertilization input. These differences were more or less marked depending both on the plant’s developmental stage and on the measured trait, with a generally greater impact of reduced N fertilization at flowering. On average, at this stage of plant development, a 35% and 30% decrease was observed for QN_Flo and NNI, respectively. In addition, the rate of leaf senescence (Sen80) increased by 6% at the latest stages of plant development. When N fertilization was reduced, there was a lower than the 20% expected yet significant decrease in kernel production (KW) at maturity (6%) and in total plant biomass production (8–10%), both at the vegetative stage (TDW_4DL) and at maturity (TDW) (Additional Fig. 1).
Table 1
Average variations of eco-physiological, developmental and yield-related traits in the 29 hybrids grown under non-limiting (HN) and reduced (LN) nitrogen fertilization conditions. Variations were calculated using the mean of 3 replicates corresponding to three different blocks. Each replicate is composed of a pool of 8 individual plants. ns = not significant. Details of the analysis and statistics are presented in Additional Table 2.
Trait
|
Trait
|
Variation (%)
|
|
abbreviation
|
|
Foliar surface (1 to 8 FDL)
|
Surf_i
|
-9.67
|
Foliar surface (12 to 14 FDL)
|
Surf_o
|
-9.69
|
Total foliar surface (FDL)
|
Surf_tot
|
-10.05
|
Leaf emergence vigour at 10 EqD
|
EV
|
-2.99
|
Leaf emergence rate (VL)
|
LER_V
|
ns
|
Leaf emergence rate (FDL)
|
LER_D
|
ns
|
Senescence at 40 EqD
|
Sen40
|
-5.99
|
Senescence at 80 EqD
|
Sen80
|
6.27
|
Ear leaf rank
|
Rk
|
ns
|
Male flowering date
|
FloM
|
ns
|
Female flowering date
|
FloF
|
ns
|
Total nitrogen at 4FDL
|
QN_4FDL
|
-13.45
|
Total carbon at 4FDL
|
QC_4FDL
|
-8.77
|
Nitrogen nutrition index at 4FDL
|
NNI_4FDL
|
-4.09
|
Total nitrogen at flowering
|
QN_Flo
|
-34.72
|
Total carbon at flowering
|
QC_Flo
|
-8.74
|
Nitrogen Nutrition Index at flowering
|
NNI_Flo
|
-30.52
|
Relative growth rate (until 11 VL)
|
Rgr_Veg
|
ns
|
Relative growth rate (until flowering)
|
Rgr_Flo
|
ns
|
Thousand kernel weight
|
TKW
|
-1.87
|
Kernel number per ear
|
KN
|
-4.42
|
Total kernel dry weight
|
KW
|
-6.33
|
Total dry weight at 4FDL
|
TDW_4FDL
|
-9.73
|
Total dry weight at harvest
|
TDW
|
-8.49
|
Kernel nitrogen percent
|
K%N
|
-12.95
|
Correlation studies were then performed to identify positive or negative relationships between eco-physiological, developmental and agronomic traits. When the LS-means of all these traits were used, we found that most of the correlations were significant (P < 0.05 after FDR correction), irrespective of the level of N fertilization. The correlation coefficients are shown for the HN and LN conditions in Figs. 1A and 1B, respectively, and their values are presented in Additional Table 2d. All the traits exhibiting a positive or negative correlation were interconnected directly or indirectly and could be grouped into three main clusters, irrespective of the level of N fertilization. One of these three clusters contained most of the measured traits, including the kernel and shoot-biomass yield components measured at harvest (KN, KW, TDW, K%N) together with the traits related to the leaf surface (Surf_o and Surf_tot) and to status of plants C and N at flowering (QN_Flo, NNI_Flo, QC_Flo,). All these traits were the most strongly and positively correlated. In this cluster, only the kernel N content (K%N) was negatively correlated to KW in both HN and LN conditions.
Two other small clusters of traits were also identified, including those related to the plant physiological status during vegetative growth at 4DL and to leaf senescence. Within these two clusters, negative correlations were observed between Rgr_Veg and the other traits measured at the vegetative stage of plant development (TDW_4FDL, QN_4FDL, GC_4FDL), and a positive correlation was observed between the latter three traits. In both HN and LN conditions, the second minor cluster encompassed the two traits used to monitor leaf senescence (Sen40 and Sen80) and Surf_i, which were negatively correlated.
To further assess the robustness of the measured eco-physiological and agronomic traits, correlation studies were performed with the activity of key marker enzymes involved in primary C and N metabolism (Additional Table 4). The most interesting result was a positive correlation between total leaf GS activity and the two main yield components represented by KW and TDW when N was reduced (Additional Table 4d). For this enzyme, both the level of N fertilization and its interaction with the genotypic effect were highly significant, with 65% variation for the former (two-way ANOVA, corrected P < 0.05, Additional Table 4c).
As most of the correlations between the different phenotypic traits were similar in both HN and LN conditions, the identification of N-responsive metabolic and protein markers and their network integration was further performed by grouping the results obtained with the two N fertilization conditions.
Strong genotypic variability for tolerance to N-limiting conditions
A PCA was conducted using as variables the different phenotypic and physiological traits measured in the 29 hybrids that were significantly different between the HN and LN conditions (Fig. 2). Along the first two components of the PCA (PC1 and PC2), which accounted for 27% and 19% of the variance respectively, a clear separation between HN and LN could be observed for most hybrids along PC1. The Euclidian distance (Ed) between the two N fertilization conditions in the PC1 x PC2 plan depended on the hybrids and roughly defined their responsiveness to N fertilization. For example, B84 (Ed = 6.5) exhibited the greatest difference between HN and LN. This difference was the lowest for F618 (Ed = 1.3), whereas an intermediate N responsiveness (Ed spanning from 2 to 5) was obtained for FR19 (Ed = 4.1).
Identification of N-responsive metabolite and protein markers
PCAs were performed to obtain an overview of the metabolite (Additional Fig. 2) and protein (Additional Fig. 3) composition in the 29 hybrids, together with a visual representation of their distribution in the HN and LN conditions. The score plots show that the two levels of N fertilization were clearly separated along the first principal component (PC1), accounting for 13 and 18% of total variability for metabolites and proteins, respectively. The 29 hybrids were separated along the PC2 axis, which reflects the genetic variability for their metabolite and protein contents. Moreover, the distribution of the 29 hybrids along the PC2 axis was not always similar in the HN and LN conditions, thus indicating that a genotype/level of N fertilization interaction probably occurred in some of them.
The comparison of each score plot with its corresponding loading plot helped identify a set of proteins and metabolites present in higher amounts under LN and HN fertilization conditions on the negative or positive sides of the PC1 axis. For the metabolites (Additional Fig. 2), several carbohydrates, an organic acid and amino acid glutamate were more abundant in the LN condition. Several other amino acids and choline were more abundant in the HN condition. Higher amounts of precursors of lignin biosynthesis such as quinic acids and derivates, flavonoids, coumaric acid ester and several benzoxazinoids were detected in the HN condition.
A two-way ANOVA (N fertilization and genotype, corrected p-value < 0.05, Additional Table 3b) confirmed the trends observed in the PCA presented in Additional Fig. 2. In the LN condition, among the 1,800 metabolites or metabolic signatures exhibiting changes in their level of accumulation (both according to the genotype and to the level of N fertilization), 658 were significantly up-regulated and 1,142 were significantly down-regulated. Higher amounts of starch, sucrose, inositol, malate, trans-aconitate, glutamate, HBOA-glucoside, DIM2BOA-glucoside and three flavonoids were detected in LN, whereas eight free amino acids, trigonelline, choline, HMBOA-glucoside, DIMBOA-glucoside and 13 flavonoids including three caffeoyl-quinate and two N-acyl amines were less abundant. Among the 40 metabolites exhibiting a change in their level of accumulation according to the level of N fertilization yet not affected by genotype, 20 were present in higher amounts (including rhamnose) and 20 in lower amounts (including tyrosine).
The proteome PCA loadings plot (Additional Fig. 3) showed that most of the proteins involved in the control of biosynthesis and homeostasis were more abundant in the LN condition. In contrast, proteins involved in the control of cellular respiration were more abundant in the HN condition.
The two-way ANOVA confirmed the results of the PCA (corrected p-value < 0.05, Additional Table 5b). Among the 2,846 identified proteins, 873 exhibited differences in their level of accumulation, according to both the genotype and the level of N fertilization. The level of accumulation of 140 proteins was modified only according to the N fertilization regime. Among these 873 proteins, 412 were present in higher amounts in LN while the others were less abundant. These 873 proteins were then classified according to the Mercator functional categories as shown in Additional Table 5d. For the proteins that were up-regulated in the LN condition, the three most represented functional categories were biosynthesis (65 proteins), homeostasis (45 proteins) and photosynthesis (35 proteins). For the proteins that were down-regulated in the LN condition, the three most represented functional categories were photosynthesis (42 proteins), redox homeostasis (38 proteins), and amino acid metabolism (37 proteins). Among the 140 N-responsive proteins, the two predominant functional categories up-regulated in LN were proteins involved in the control of their biosynthesis (30 proteins) and in photosynthesis (9 proteins). Eighteen proteins involved in photosynthesis and seven proteins involved in solute transport were down-regulated in LN, representing the two main functional categories that were altered when N fertilization was reduced.
Identification of coregulations between metabolites, proteins, eco-physiological and agronomic traits
To highlight coregulations between eco-physiological, developmental and yield-related traits, metabolites and proteins under non-limiting and reduced fertilization, a multiblock sparse Partial-Least-Squares Discriminant Analysis (sPLS-DA) was first performed using their LSmeans values (Additional Tables 2c, 3c and 5c). Such integrated analysis of omics-based and phenotypic data helped select 100 metabolites and proteins that discriminated the two fertilization regimes and covaried across the three sets of traits (Figs. 3A to 3F). The first latent variable (LV1) separated the two N fertilization regimes as expected (Figs. 3A, 3C and 3E). The loading values of the variables selected on LV1 are presented in Additional Table 6. On the positive side of LV1, NNI_Flo, QN_Flo, Surf_tot, NNI_4FDL, Surf_o and QN_4FDL were higher in HN, covaried positively with choline, with four LC-QTOF-MS metabolic signatures (M391T738, M390T739, M389T739, M389T949), and for proteins with a glycerophosphodiester phosphodiesterase, a phosphocholine phosphatase, a nucleoside diphosphate kinase, a putative glucose-6-phosphate 1-epimerase, a peptidyl-prolyl isomerase, a copper/zinc superoxide dismutase, a chloroplastic zeaxanthin epoxidase and a phosphosugar phosphatase. On the negative side of LV1, Sen80 was higher in LN, covaried positively with M347T265, M174T271, rhamnose, M279T237 and M511T963, and with a component eIF-iso4G of eIF-iso4F unwinding complex, a component eIF3b of eIF3 mRNA-to-PIC binding complex and a glycine-rich RNA-binding ABA-inducible protein. A separation of genotypes could be observed along LV2. On the positive side of LV2, FloF and FloM, were delayed in a several hybrids, including B97, B84, B104, B73 and F618. These two developmental traits covaried positively with a peroxidase 70, AC210204.3_FGP002 and a component PsaD of PS-I complex. On the negative side of LV2, Sen40 occurred later in hybrids D09, FV252 and D06. This marker of leaf senescence covaried positively with a range of LC-QTOF-MS metabolic signatures, including M191T338, M370T342, M371T337 and M210T343, and for proteins with a lipoamide-containing component H-protein of glycine cleavage system, an uncharacterized chloroplastic methyltransferase and a 2,3-bisphosphoglycerate-independent phosphoglycerate mutase.
To focus on the strongest coregulations involved in the responses to the two N fertilization regimes, the variables exhibiting an absolute loading value higher than 0.1 on LV1 were selected in the three data sets. A correlation network between these variables was then constructed if the Spearman correlation values between the variables were above 0.75 or below − 0.75. Before reconstructing the network, redundancies within the selected LC-QTOF-MS signatures of the metabolite data set were discarded. The selected LC-QTOF-MS variables were clustered manually according to their chromatographic profiles and putative chemical formulas. For each cluster, only the predominant metabolic signatures were selected. A tentative annotation of these metabolite signatures led to the putative annotation of three metabolites: dehydroascorbate, a caffeoyl-glucose and a UDP-hexose (Additional Table 3e). The resulting network presented in Fig. 4 contained two phenotypic traits NNI_Flo and QN_Flo, 16 metabolites and 31 proteins. Metabolite signatures and proteins directly correlated with NNI_Flo and QN_Flo are listed in Table 2.
Within this network, five proteins were involved in carbohydrate metabolism, three in lipid metabolism, three in protein biosynthesis and three in redox homeostasis. NNI_Flo was negatively correlated to M347T265_dehydroascorbate, a glycine-rich RNA-binding ABA-inducible protein, and component eIF-iso4G of eIF-iso4F unwinding complex. NNI_Flo was positively correlated to M398T711. NNI_Flo and QN_Flo were both positively correlated to chloroplastic glycerophosphodiester phosphodiesterase GDPD1. QN_Flo was also positively correlated to a nucleoside diphosphate kinase and a 4-hydroxy-4-methyl-2-oxoglutarate aldolase. The three metabolite nodes exhibiting the highest number of connections were M347T265_dehydroascorbate, choline and M389T739 (an unidentified N-containing compound), with at least 12 connections each. The three protein nodes exhibiting the highest number of connections were chloroplastic glycerophosphodiester phosphodiesterase GDPD1, the glycine-rich RNA-binding ABA-inducible protein and a peptidyl-prolyl isomerase, with at least 26 connections each.
Table 2
Metabolite signatures and proteins directly correlated to NNI_Flo and to QN_Flo in the correlation network presented in Fig. 4.
Eco-physiological trait
|
Metabolome or proteome variable
|
Annotation of metabolites and proteins
|
Spearman correlation coefficient
|
p-value
|
QN_Flo
|
GRMZM2G005771_P01
|
Putative 4-hydroxy-4-methyl-2-oxoglutarate aldolase 2
|
0.7639
|
3.06E-12
|
QN_Flo
|
GRMZM2G018820_P01
|
Glycerophosphodiester phosphodiesterase GDPD1, chloroplastic
|
0.7899
|
1.70E-13
|
QN_Flo
|
GRMZM2G134797_P02
|
Nucleoside diphosphate kinase
|
0.7735
|
1.11E-12
|
QN_Flo
|
GRMZM2G165901_P01
|
Glycine-rich RNA-binding, abscisic acid-inducible protein
|
-0.7777
|
6.93E-13
|
NNI_Flo
|
M347T265
|
Dehydroascorbate
|
-0.7622
|
3.65E-12
|
NNI_Flo
|
M398T711
|
Not annotated
|
0.7567
|
6.41E-12
|
NNI_Flo
|
GRMZM2G018820_P01
|
Glycerophosphodiester phosphodiesterase GDPD1, chloroplastic
|
0.7580
|
5.58E-12
|
NNI_Flo
|
GRMZM2G157061_P01
|
Component eIF-iso4G of eIF-iso4F unwinding complex
|
-0.7660
|
2.45E-12
|
NNI_Flo
|
GRMZM2G165901_P01
|
Glycine-rich RNA-binding, abscisic acid-inducible protein
|
-0.7957
|
8.46E-14
|
Metabolites and proteins linked to N-susceptibility
Following the identification of N-responsive metabolites and proteins in the 29 hybrids using multiblock sparse PLS-DA and network analyses, a complementary approach was conducted using the susceptibility of the hybrids to reduced N fertilization. The aim was to identify biochemical markers representative of a maize ideotype exhibiting better agronomic performance in the LN condition. When the susceptibility to LN was calculated using QN_4FDL, QC_4FDL, QN_Flo, QC_Flo, TKW, KN, KW, TDW_4FDL and TDW LS-means data, an HCA identified four distinct groups of hybrids (Fig. 5). The first group (cluster C1) was characterized by low susceptibility to reduced N fertilization at the early stage of plant development (4DL), including shoot biomass production (TDW_4FDL). All these hybrids were less susceptible to the LN condition, not only in terms of their C and N contents at flowering but also for all yield-related traits, especially TDW at maturity. In contrast, the fourth group (cluster C4) contained hybrids that were more susceptible to reduced N fertilization at 4DL and less susceptible at maturity, especially regarding the traits related to kernel and total biomass production. The other two clusters C2 and C3 were mostly characterized by low susceptibility in the LN condition, mostly for their C and N contents at flowering. N-susceptibility for all other traits was high, except for the yield-related traits in C3.
A one-way ANOVA was then performed on the susceptibilities to reduced N fertilization of the metabolites and proteins exhibiting significant changes following the two-way ANOVA previously performed on their contents (Additional Tables S3b and S5b). The aim was to determine which metabolites and proteins differentiated the hybrids grouped in cluster C4, compared to all other hybrids (Additional Fig. 4). For metabolites in this cluster, susceptibility to the LN condition was characterized by an increase in metabolic signatures M519T1858, M385T279 and M468T733 and a decrease in M343T742 (P < 0.01). M519T1858 was tentatively identified as hypatulin B, and M343T742 as a dimethoxybenzoic acid glucoside. For proteins, susceptibility to the LN condition was characterized by an increase in GRMZM2G018074, a regulatory protein (GCN4) of RIN4, and a decrease in GRMZM5G886257, a cytosolic NADP-dependent malic enzyme.