For RNA extraction we used 10 grains from the same ear of a flint and a dent inbred line grown in the Limagrain research station (Mons, France) and harvested at 15 and 20 DAP. The inbred lines L3 (flint) and L5 (dent) were chosen from a previous study where 13 flint and dent inbred lines were analyzed for their protein, starch, and lipid contents in relation to endosperm vitreousness (Gayral et al., 2015, 2016). Two biological repetitions were carried out with grains from different ears. The grains were hand-dissected. In a first time, the middle third of the grain was isolated with a scalpel blade, then an endosperm piece of tissue was extracted from the central endosperm (which will develop in the floury region) and the peripheral endosperm (which will develop in the vitreous region) using a small cookie cutter (Ø = 0.9 mm), and immediately poured into liquid nitrogen before grinding. Total RNA extraction for microarray and RT-PCR (reverse-transcription PCR) analysis was performed using the RNeasy Plant Mini Kit (Qiagen, Les Ulis, France).

Experiments were designed with the Genomics Networks team of the Institute of Plant Sciences (IPS2). For each array, the raw data comprised the logarithm of median intensities of feature pixels at wavelengths of 635 nm (red) and 532 nm (green). For each array, a global intensity-dependent normalization using the loess procedure (Yang et al., 2002) was performed to correct the dye bias. The differential analysis was based on averaging the log-ratios over the duplicate probes and over the technical replicates. Hence, the number of available data for each gene equals the number of biological replicates and is used to calculate the moderated t-test (Smyth, 2004). Under the null hypothesis, no evidence that the specific variances vary between probes was highlighted by Limma analysis, and consequently the moderated t-statistic was assumed to follow a standard normal distribution. To control the false discovery rate (FDR), adjusted p-values found using the optimized FDR approach described by Storey and Tibshirani (2003) were calculated. We considered the probes with an adjusted p-value ≤ 0.05 to be differentially expressed. Analysis was performed using R software. The function SqueezeVar of the library Limma was used to smooth the specific variances by computing empirical Bayesian posterior means. The library kerfdr (kernel-based local fdr) was used to calculate the adjusted p-values.

Microarray data from this article were deposited in the international repository GEO, Gene Expression Omnibus (Edgar et al., 2002; http://www.ncbi.nlm.nih.gov/geo/), accession no. GSE88993. All steps of the experiment, from growth conditions to bioinformatic and statistical analyses, were detailed in CATdb (Complete Arabidopsis Transcriptome database; Gagnot et al., 2008; http://tools.ips2.u-psud.fr/CATdb/; Project: 4 plex_Maize_2015_02) according to the “Minimum Information About a Microarray Experiment” standards.

Genes were annotated by sequence homology with A. thaliana or O. sativa according to MaizeGDB (Maize Genetics and Genomics Database; http://www.maizegdb.org; Sen et al., 2009).

Identification of GO (gene ontology) categories significantly enriched (p ≤ 0.05) was based on “plant GO slim” and done using online AgriGO software (http://bioinfo.cau.edu.cn/agriGO; Du et al., 2010).

Total RNA (4 μg) was reverse-transcribed using SuperScript III and PolyT primer (Invitrogen, ThermoFisher Scientific, Courtaboeuf, France). qPCR (qualitative PCR) quantification was performed with CFX Connect (Bio-Rad, Marnes-la-Coquette, France) and Power SYBR Green PCR Master Mix (Applied Biosystems, ThermoFisher Scientific, Courtaboeuf, France). The relative gene expression level was normalized to ubiquitin (GRMZM2G109977), and was defined as ΔCq = 2^∧−(Cq_Gene − Cq_Ubi). The data represent the mean of two technical replicates for two biological repetitions (± SD). All primers used in the study are listed in Table S1.

For metabolite extraction, we used five grains (30 DAP) from the same ear. Three biological repetitions were carried out from different ears. Grains were hand-dissected, and samples were freeze-dried and ground in liquid nitrogen. Extraction was conducted on 20 mg of sample with 500 μL of CH₃OH containing 0.2 mM norleucine and 0.4 mM adonitol as an internal standard. After 15 min of stirring, 250 μL of CHCl₃ and 500 μL of water were added. After stirring and centrifugation, the aqueous phase was collected and used for metabolite quantification. Aliquots (100 μL) were dried in a SpeedVac concentrator (Savant, ThermoFisher Scientific, Courtaboeuf, France) and derivatized for 2 h at 37°C in 50 μL 20 mg.mL⁻¹ methoxyamine in pyridine, then for 30 min at 37°C after the addition of 50 μL of BSTFA (N, O-bis (trimethylsilyl) trifluoroacetamide). Metabolites were analyzed by GC-FID/MS (gas chromatography with flame ionization detection and mass spectrometry) using Clarus SQ8C (Perkin Elmer, Villebon sur Yvette, France) equipped with a DB-5MS capillary column (30 m × 0.25 mm × 0.25 μm; Agilent Technologies, Les Ulis, France), and identified according to mass spectra and retention time compared with standard solution. Metabolites were normalized and quantified according to adonitol.

For quantification of free amino acids, aliquots (200 μL) were dried in a SpeedVac concentrator, and derivatized in 20 μL H₂O/CH₃CH₂OH/N(CH₂CH₃)₃/PITC (phenyl isothiocyanate) (1/7/1/1) for 10 min at room temperature. Samples were then dried and solubilized in 100 μL 2 mM-Na₂HPO₄ pH 7.4, with 5% of CH₃CN. Free amino acids were analyzed by reversed phase HPLC (reversed-phase high-performance liquid chromatography) using a Pico-Tag column (3.9 × 300 mm; Waters, Guyancourt, France) on an Alliance HT apparatus equipped with a UV detector (254 nm; Waters, Guyancourt, France). Elution (1 mL.min⁻¹) was performed at 40°C using a linear gradient from 100% eluent A (70 mM-CH₃ÂČCOONa pH 6.45 with 2.5% CH₃CN) to 100% eluent B (CH₃CN/CH₃OH/H₂O (45/15/40). The quantity of amino acid was normalized to the norleucine internal standard and quantified in regard to the calibration curve obtained by injection of standard amino acids (20,088, Pierce, ThermoFisher Scientific, Courtaboeuf, France) derivatized under the same conditions.

Maize grain development starts with the double fertilization of the female gamete and the central cells, which leads to the development of the embryo and the endosperm respectively. Upon 10 DAP, protein and starch begin to accumulate in the endosperm. To ascertain whether transcriptome modifications are related to vitreousness, we analyzed the RNAs extracted from central and peripheral endosperm that will become at maturity the floury and the vitreous areas, respectively. From here on, we will refer to these regions as maize endosperm center (MEC) and maize endosperm periphery (MEP). MEC and MEP transcriptomes were analyzed on a flint and a dent inbred line harvested at 15 and 20 DAP. By considering only genes up-regulated in MEC, i.e., 5,518, or in MEP, i.e., 5,167, at least in a developing stage and in a genetic background (flint or dent), gene ontology (GO) analysis revealed enrichment in “metabolic process,” “cellular process,” “cellular metabolic process,” and “biosynthetic process” in MEP, whereas in MEC we notably observed enrichments in “response to stress,” “response to stimulus,” and “response to abiotic stimulus” (Figure 1A). These results suggest that MEC and MEP follow their own development program, especially with respect to the stronger stress responses in MEC.

From 15 to 20 DAP developing stages, at least 9369 genes were found to be differentially expressed when MEC and/or MEP of flint and/or dent inbred lines were compared. Considering genes significantly regulated in MEC and MEP, we did not observe strong GO enrichment pattern differences (Figure S1A). Enriched GO categories are mainly related to metabolic and biosynthetic processing (Figure S1A), in agreement with the massive synthesis of storage compounds. However, we only observed GO enrichment of “response to stress” in MEC, whereas “generation of precursor metabolites and energy” and “multicellular organismal process” were enriched in MEP. Therefore, MEC and MEP follow a close developmental process, and differences in stress responses observed between MEC and MEP persist throughout grain filling.

To further characterize the relationships between developmental stress responses and maize vitreousness, we focused on genes that showed significant differences in expression levels between MEC and MEP in flint and dent endosperm at 15 and 20 DAP: 634 genes were up-regulated in MEP at 15 and 20 DAP in flint and dent inbred lines, and 525 genes were annotated, while in MEC, 776 genes were up-regulated and 602 were annotated. Differentially expressed genes involved in stress management are shown in Figure 1B. In MEP, the expression of genes involved in oxido-reduction (including responses to oxidative burst), ER-associated degradation (ERAD) and abiotic/biotic stress responses was clearly activated. In MEC, we also observed the expression of genes involved in vesicular trafficking, PCD, autophagy, hypoxia, and regulation of transcription (Figure 1B); these data suggested a higher stress response in this area at 15 and 20 DAP.

To further investigate the temporal modifications linked to vitreousness, we focused on genes showing significant differences in MEP and/or MEC in both flint and dent lines between 15 and 20 DAP. 378 genes were significantly up-regulated from 15 to 20 DAP in MEP and/or MEC in the two inbred lines, and 258 of these genes were annotated. On the other hand, 348 genes were found to be significantly down-regulated (from 15 to 20 DAP) in MEP and/or MEC in the two inbred lines, and of these genes, 280 were annotated. If we consider the genes involved in stress responses, their expression was mainly up-regulated at 20 DAP (Figure S1B). In fact, almost the same genes were found to be up-regulated in MEP and MEC, with the exception of those implicated in ERAD, essentially up-regulated in MEC. Between 15 and 20 DAP, PCD genes were found to be up-regulated in both MEC and MEP (Figure S1B), confirming that PCD increases throughout endosperm development as previously shown (Young and Gallie, 2000a,b).

In regard to the genes encoding enzymes involved in core starch biosynthesis, ADP-glucose pyrophosphorylase, starch synthases, branching, and debranching enzymes, we did not observe any coherent differential expression. Indeed, certain genes were found to be up-regulated in MEP, while their homologs were found to be up-regulated in MEC (Figure S2). These findings are probably due to the fact that starch deposition follows gradients from the upper to the basal part of the endosperm (Dochlert, 1990) as well as from the periphery to the center (Gayral et al., 2016). This fact could also be related to the regulation of starch biosynthesis at the post-translational level through phosphorylations and formation of multienzyme complexes (Hennen-Bierwagen et al., 2009). In fact, starch biosynthesis starts with ADP-glucose a product provided by the sucrose synthase (SUS) pathway. Indeed, ectopic overexpression of a potato SUS in maize induces an increase in ADP-glucose and starch contents (Li et al., 2013). Three sus genes are strongly expressed in maize endosperm: sus1 (GRMZM2G152908), sus2 (GRMZM2G318780), and sus-sh1 (GRMZM2G089713; Figure S2). Previous studies suggest that SUS1 and SUS-SH1 are involved in starch and cell wall polysaccharide synthesis, respectively (Chourey et al., 1998). In contrast with invertases, SUS is an energy-conserving pathway for sucrose breakdown since it uses one PPi molecule and produces one UTP molecule. UTP can be used to phosphorylate fructose- and subsequent entry of fructose-6-phosphate in the glycolysis pathway (Geigenberger, 2014). Sus1 (GRMZM2G152908) and sus2 (GRMZM2G318780) are up-regulated in MEP (Figure S2), and display the highest differential expression among all genes that discriminate developing MEP from developing MEC. This is consistent (i) with the localization of the intense corresponding activity at the periphery of developing maize endosperm (Wittich and Vreugdenhil, 1998) (ii) the formation of SUS1-SUS2 hetero-oligomers (Duncan et al., 2006) and (iii) the up-regulation of sus1 in hypoxic conditions (Zeng et al., 1998). Any differential expression is observed for sus-sh1 (GRMZM2G089713) that encodes SUS-SH1 (Figure S2) in agreement with the fact that the expression of this gene is not up-regulated by hypoxia but by anoxia, a huge depletion of oxygen (Zeng et al., 1998). However, up-regulation of sus1 and sus2 in MEP is contradictory to (i) slightly lower starch contents in the vitreous than in the floury endosperm (Gayral et al., 2015) and (ii), as shown below, to up-regulation of genes markers of hypoxia in MEC. Altogether these results suggest that the management of hypoxia differs between MEP and MEC.

In contrast with starch deposition, protein deposition is significantly higher in the periphery of the endosperm (Gayral et al., 2016) than in its longitudinal axis, i.e., from the upper to the lower part of the endosperm (Dochlert, 1990). Furthermore, expression of zein genes is essentially regulated at the transcriptional level (Zhang et al., 2015; Qiao et al., 2016). Indeed, transcriptomic analysis revealed stronger zein expression in MEP at 15 and 20 DAP, especially for α and δ-zeins (Figure 2). These results agree with those of mRNA in-situ hybridization experiments, which revealed stronger α-zein expression at the periphery of developing endosperm (Woo et al., 2001). Furthermore, we observed a global up-regulation of genes specifying zeins at 20 DAP (Figure 2), in accordance with epigenetic regulation of α-zein expression in endosperm at 15–23 DAP (Locatelli et al., 2009). These results are consistent with the protein decrease observed previously from the sub-aleurone layer to the endosperm center (Gayral et al., 2016).

Down-regulation of genes specifying zeins in MEC is associated with up-regulation of genes involved in stress responses (Figure 1). When checked, chaperones, protein degradation (ERAD), vesicular trafficking, autophagy, and the transcription process are up-regulated in MEC (Figure 1B). All these mechanisms are generally induced in response to ER stress, leading to the corresponding signaling cascade named the unfolded protein response (UPR) and suggest stronger UPR in MEC than in MEP. In plants, two UPR pathways have been described. The first UPR pathway involves bZIP17/bZIP28, i.e., two membrane-associated TFs that are exported from ER to Golgi apparatus en route to the nucleus; the second pathway involves IRE1, i.e. a dual-protein kinase, and the bZIP60 TF. When unfolded or misfolded proteins accumulate in the ER, IRE1 is activated and the bzip60 mRNA is spliced (Howell, 2013). The spliced form of bzip60 leads to mRNA encoding a TF with a canonical nuclear targeting signal (Li et al., 2012; Wang B. et al., 2012). In addition to promoting UPR genes, bZIP60 up-regulates its own expression. In fact, microarray analysis showed a higher expression of bzip60 in MEC than in MEP for both flint and dent maize and at both 15 and 20 DAP (Table S2). To characterize the UPR level in MEC and MEP, we followed bzip60 splicing during endosperm development by quantifying the unspliced and spliced forms of this TF. RT-PCR analyses revealed spliced mRNA in both MEC and MEP at 15 and 20 DAP for both flint and dent maize (Figure 3A), attesting to the fact that UPR took place in developing endosperm. In flint, and especially in dent, inbred lines, qRT-PCR analyses globally showed a higher expression level at 20 DAP (Figure 3). These results reflect the increase in stress during endosperm development, as shown in Figure S1. In terms of total bzip60 and unspliced bzip60 expression levels, a higher level was observed in MEC, confirmed by microarray analysis (Table S2). When compared, splicing of bzip60 is more pronounced in MEC than in MEP, confirming a stronger UPR in this region. This difference was found to be more acute at 20 DAP. It is interesting to note that ER stress was previously associated with higher content of triacylglycerols (TAG) in maize endosperm of the floury2 mutant (Shank et al., 2001), in accordance with higher amounts of these storage lipids in the floury regions of conventional maize (Gayral et al., 2015). The higher TAG deposition in MEC relative to that in MEP is also related to major changes in the proteins that cover and stabilize oil droplets. Indeed, while an oleosin-like protein is highly up-regulated in MEP (GRMZM2G480954), an ortholog of Arabidopsis lipid droplet-associated proteins (LDAP) (GRMZM2G150367) is one of the most up-regulated proteins in MEC. These LDAPs are associated with stress responses and are mostly expressed in vegetative tissues (Laibach et al., 2015). Therefore, ER stress and UPR could be responsible for TAG accumulation in the floury regions of maize endosperm. Taken together, these results reveal a permanent ER stress in the endosperm, with a stronger UPR in MEC than in MEP.

UPR evolution is associated with spatiotemporal up-regulation of autophagy and PCD (Figure 1B and Figure S1), suggesting a possible relationship between these processes, especially in MEC. Such a relationship controls the pro-life and pro-death evolution of cells and is commonly encountered in all eukaryotes (Suzuki et al., 2017). Indeed, it is worth noting that the gene specifying BAX inhibitor-1, an ER-resident transmembrane protein that attenuates PCD owing to ER-stress in Arabidopsis (Watanabe and Lam, 2008), is up-regulated in MEC (Table S2). In the same way, the gene specifying Bcl-2-associated athanogene 1 (BAG1) is up-regulated in MEC (Table S2). In Arabidopsis, BAG1 acts as a co-chaperone in various processes related to environmental stress and PCD (Nawkar et al., 2016). The DAD1 (defender against cell death) gene, also localized in the ER, is another regulator of cell death. dad1 is well expressed in both MEC and MEP, with a slight tendency to be more up-regulated in MEC (Table S2). DAD1 is a glycosyltransferase that is known to mediate, in Arabidopsis, N-glycosylation of nascent proteins and to suppress cell death induced by abiotic stresses (Danon et al., 2004). From 15 DAP, expression of the genes specifying these proteins, in maize endosperm, as well as their up-regulation in MEC, confirm the early PCD observed in this area. Nine metacaspases have been identified in Arabidopsis, AtMC1 to AtMC9, and their homologs, i.e., ZmMC1 to ZmMC9, were found in maize (Minina et al., 2013). Though these proteins differ from their animal counterparts, plant metacaspases are markers of PCD (Lam and del Pozo, 2000; Tsiatsiani et al., 2011; Fagundes et al., 2015). Microarray analyses revealed that two type II metacaspases are differentially expressed in maize endosperms: mc7 is up-regulated in MEC, while mc9 is up-regulated in MEP (Table S2), suggesting different PCD pathways in these two areas. Indeed, in Norway spruce, a close relationship was established between a type II metacaspase, autophagy and the subsequent PCD pathway (Minina et al., 2013).

Autophagy is a ubiquitous complex process allowing the degradation of cell organelles and the recycling of these components. Many genes encoding autophagy-related proteins (ATG proteins) have been identified (Li et al., 2015), most of them being expressed during endosperm development, suggesting that the autophagic recycling machinery is fully active (Table S2). Remarkably, although differential expression of these genes between MEC and MEP is not systemically evidenced, some genes are up-regulated in MEP as atg4b, while atg6a and atg8c are up-regulated in MEC. In fact, most atg8 genes, i.e., atg8b, atg8c, atg8d, and atg8e tend to be up-regulated in MEC (Table S2). ATG8 proteins are essential in the autophagic process since their lipidation by phosphatidylethanolamine is the actual primum movens of the autophagic process. ATG12, which participates in ATG8 lipidation, is also essential for autophagy (Suzuki et al., 2017), and consequently atg12a is up-regulated in MEC (Table S2). Expression of all atg genes confirms that an autophagic process is implicated in maize endosperm development (Chung et al., 2009), and suggests that a stronger autophagy takes place in MEC. These results agree with spatiotemporal relationships between UPR, autophagy and PCD in this endosperm area and show that UPR corresponds with the spatiotemporal development of PCD in maize endosperm.

The first step in the TCA cycle is the production of citrate catalyzed by citrate synthase (CTS). The gene specifying CTS is up-regulated in MEC in dent maize. Subsequently, aconitase (ACO) and isocitrate dehydrogenase (IDH) produce α-ketoglutarate using citrate. In flint and dent inbred lines, the results revealed the up-regulation of aco and idh expression in MEP (Figure 5). During the TCA cycle, α-ketoglutarate is decarboxylated by the 2-oxoglutarate dehydrogenase complex (OGDC), leading to succinyl CoA and NADH, the succinyl CoA being subsequently hydrolyzed to succinate by succinyl-CoA synthase (SCS). The ogdc gene does not display significant differential expression levels in MEP and MEC, while the scs gene is up-regulated in the MEC of dent endosperm (Figure 5). Succinate dehydrogenase complex (SDH, also known as complex II of the mitochondrial electron transport chain) catalyzes the oxidation of succinate to fumarate, which is then hydrated, via fumarase (FUM), leading to the production of malate. sdh and fum genes are both clearly up-regulated in MEP (Figure 5). All these results support the hypothesis that maintaining a functional TCA cycle is probably a key step in the formation of vitreous endosperm. Moreover, the up-regulation of sdh increases both the net output of NADH and FADH as substrates for the electron transport chain and the direct input of electrons into this chain via SDH itself. These results support the idea of a metabolic switch when MEC and MEP are compared, where the latter maintains a significant respiratory metabolism.

Under hypoxic conditions, α-ketoglutarate is also used by glutamate synthase (GOGAT) and glutamate decarboxylase (GAD), which respectively produce glutamate and GABA via the GABA shunt pathway, leading to accumulation of glutamate and GABA (António et al., 2016). The genes specifying GOGAT and GAD are up-regulated in MEC (Figure 5), which concurs with the higher amount of glutamate in this tissue area (Figure 4). Hypoxia leads to a pH decrease (Felle, 2005) since protons are consumed during GABA production. The up-regulation of gad in MEC also supports stronger hypoxia in this area. The GABA shunt leads ultimately to succinate production via γ-aminobutyric acid trans-aminase (GABA-T) and succinic semi-aldehyde dehydrogenase (SSADH). The genes specifying these enzymes are clearly up-regulated in MEP (Figure 5). These results suggest that in MEP, the GABA shunt leads to the production of succinate and NADH production necessary to maintain the TCA cycle and respiration, respectively. These differences in the regulation of GABA shunt enzymes testify in favor of differences in hypoxia tolerance between MEP and MEC.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.