Transcriptomics at Maize Embryo/Endosperm Interfaces Identifies a Transcriptionally Distinct Endosperm Subdomain Adjacent to the Embryo Scutellum

A population of endosperm cells adjacent to the embryo scutellum shows transcriptomic enrichment in transport functions and is influenced by embryo development.

Abstract

Seeds are complex biological systems comprising three genetically distinct tissues nested one inside another (embryo, endosperm, and maternal tissues). However, the complexity of the kernel makes it difficult to understand intercompartment interactions without access to spatially accurate information. Here, we took advantage of the large size of the maize (Zea mays) kernel to characterize genome-wide expression profiles of tissues at different embryo/endosperm interfaces. Our analysis identifies specific transcriptomic signatures in two interface tissues compared with whole seed compartments: the scutellar aleurone layer and the newly named endosperm adjacent to scutellum (EAS). The EAS, which appears around 9 d after pollination and persists for around 11 d, is confined to one to three endosperm cell layers adjacent to the embryonic scutellum. Its transcriptome is enriched in genes encoding transporters. The absence of the embryo in an embryo specific mutant can alter the expression pattern of EAS marker genes. The detection of cell death in some EAS cells together with an accumulation of crushed cell walls suggests that the EAS is a dynamic zone from which cell layers in contact with the embryo are regularly eliminated and to which additional endosperm cells are recruited as the embryo grows.

INTRODUCTION

Cereal grains are not only essential for plant propagation but are also high-value products that represent an important source of calories and proteins for human nutrition and animal feed as well as a coveted resource for bio-sourced industries. In maize (Zea mays), the accumulation of oil in the embryo and of starch and protein in the endosperm requires the development of adapted structures and the coordinated regulation and distribution of nutrient flow from the mother plant. The development of the embryo, which will form the future plant, and the endosperm, which will nourish the embryo during germination, occurs in three main phases (Lopes and Larkins, 1993; Berger, 1999; Dumas and Rogowsky, 2008). During the first 2 weeks of early maize seed development, embryo and endosperm cells differentiate into populations forming distinct tissues and organs (Randolph, 1936; Leroux et al., 2014), including two storage organs, the scutellum of the embryo and the starchy endosperm (early development phase). These two zygotic compartments then start to accumulate large quantities of storage compounds during the following 2 to 3 weeks (filling phase), while the surrounding maternal tissues provide or transport the necessary nutrient supplies (Porter et al., 1987; Wu and Messing, 2014). During the final 4 weeks (maturation phase), the kernel dehydrates and enters into quiescence prior to dispersal (Vernoud et al., 2005; Sabelli and Larkins, 2009; Sreenivasulu and Wobus, 2013). These three phases are determined by distinct genetic programs and characterized by distinct anatomical and cytological features. Spatially, the maize kernel is organized like Russian dolls, the embryo being enclosed within the endosperm, which is itself surrounded by the pericarp (maternal tissues).

A closer look at the highly differentiated structure displayed by the maize embryo shows that 4 d after pollination (DAP), two distinct parts can be distinguished: an apical embryo proper and a basal suspensor that will degenerate at the end of early development (Giuliani et al., 2002; Doll et al., 2017). At around 8 DAP, the embryo proper generates, at the abaxial side, a shield-shaped organ, the above-mentioned scutellum. The shoot apical meristem develops on the adaxial side. Marking the apical pole of the future embryonic axis, the shoot apical meristem will produce several embryonic leaves over time. The root apical meristem differentiates within the embryo body, defining the basal pole of the embryonic axis. Shoot and root meristems will be surrounded by the protective coleoptile and coleorhiza, respectively (Randolph, 1936; Bommert and Werr, 2001; Vernoud et al., 2005).

The surrounding endosperm, which occupies 70% of the kernel volume at the end of early development (Sabelli and Larkins, 2009; Leroux et al., 2014; Rousseau et al., 2015; Zhan et al., 2017), has been described as differentiating only four main cell types. The basal endosperm transfer layer (BETL) and the aleurone layer (AL) are two peripheral cell types in contact with maternal tissues. The embryo-surrounding region (ESR) is formed of small densely cytoplasmic cells encircling the young embryo. Lastly, the starchy endosperm (SE) corresponds to the central region of the endosperm, which subsequently accumulates huge amounts of storage compounds before undergoing progressive programmed cell death. The developing endosperm is surrounded by maternal tissues: the nutritive nucellus, which degenerates as the endosperm expands, and the protective pericarp, which comprises the pedicel at the basal pole (Olsen, 2001; Berger, 2003; Sabelli and Larkins, 2009; Zhan et al., 2017).

The parallel growth and profound developmental transformations of the three kernel compartments highlight the need for constant coordination, which likely requires a complex intercompartmental dialogue (Nowack et al., 2010; Ingram and Gutierrez-Marcos, 2015; Widiez et al., 2017). Since maternal tissues, endosperm, and embryo are symplastically isolated, their apoplastic interfaces represent essential zones for this dialogue (Diboll and Larson, 1966; Van Lammeren, 1987; Widiez et al., 2017). A good example to illustrate the importance and specialization of interfaces is carbon transport. Sugars must be transported from the maternal tissues to the embryo for growth and fatty acid accumulation, passing through the endosperm, which needs to retain part of the carbon for its own growth as well as the biosynthesis of starch and storage proteins (Sabelli and Larkins, 2009; Chourey and Hueros, 2017). In maize, nutrients are unloaded from open ends of the phloem vessels into the placento-chalazal zone of the maternal pedicel (Porter et al., 1987; Bezrutczyk et al., 2018). At the base of the endosperm, the BETL cells form dramatic cell wall ingrowths, thus increasing the exchange surface (Kiesselbach and Walker, 1952; Davis et al., 1990). BETL cells express a specific set of genes, including Miniature1, encoding a cell wall invertase, which cleaves sucrose into hexoses (Lowe and Nelson, 1946; Miller and Chourey, 1992; Cheng et al., 1996; Kang et al., 2009). These are taken up by the sugar transporter SWEET4c (Sugars Will Eventually be Exported Transporter), which has been demonstrated to be the key transporter of sugar at the pedicel/endosperm interface, since the defects in seed filling of the corresponding sweet4c mutant lead to a miniature kernel phenotype (Sosso et al., 2015). The remaining endosperm interface with maternal tissues (initially the nucellus and later on the pericarp) is the AL, which is not known to contribute to nutrient exchange during seed development (Gontarek and Becraft, 2017).

The interface between the endosperm and the embryo is also developmentally dynamic. At 3 to 6 DAP, the embryo is totally surrounded by ESR-type cells. As the embryo expands, it emerges from the ESR, which consequently becomes restricted to the zone surrounding the basal part (suspensor) of the embryo and ultimately disappears together with the suspensor at the end of the early development phase (Opsahl-Ferstad et al., 1997; Giuliani et al., 2002). From 8 to 9 DAP, the upper part (embryo proper) forms two new interfaces: (1) at the adaxial side, the embryo is enclosed by a single cell layer, which is called the scutellar aleurone layer (SAL) in barley (Hordeum vulgare; Jestin et al., 2008); and (2) at the abaxial side, the embryo is brought into direct contact with central SE cells (Van Lammeren, 1987). This interface is constantly moving due to the growth of the scutellum inside the endosperm. On the embryo side of this interface, the epidermis of the scutellum has a distinct morphology and gene expression pattern (Ingram et al., 2000; Bommert and Werr, 2001). The dynamics of the endosperm/embryo interface, and the processes that occur there, remain largely undescribed.

At many intercompartmental interfaces, such as the BETL, the ESR, and the AL, the cells constitute readily identifiable tissues with distinctive and often striking cell morphologies and with defined organizations and established functions (except for the ESR; for review, see Doll et al., 2017). In many cases, specific sets of genes are expressed in these tissues, as revealed by the identification and characterization of marker genes, for example of Maternally expressed gene1, Myb-related protein1, and Betl1, Betl2, Betl3, and Betl4 in the BETL (Hueros et al., 1999a, 1999b; Cai et al., 2002; Gómez et al., 2002; Gutiérrez-Marcos et al., 2004), Viviparous1 in the AL (Suzuki et al., 2003), and Esr1 to Esr3 in the ESR (Opsahl-Ferstad et al., 1997).

Genome-wide gene expression studies at numerous developmental stages of whole kernels and/or hand-dissected endosperm and embryo (Downs et al., 2013; Lu et al., 2013; Chen et al., 2014; Li et al., 2014; Qu et al., 2016; Meng et al., 2018) have been complemented by a recent transcriptomic analysis of laser-capture microdissected cell types and subcompartments of 8-DAP kernels (Zhan et al., 2015). However, even the latter study did not address specifically the transcriptomic profiles of the embryo/endosperm interfaces and did not answer the question of whether the endosperm at the scutellum/endosperm interface is composed of cells with specific transcriptional identities.

In this study, we took advantage of the large size of the maize kernel to characterize the genome-wide gene expression profile at embryo/endosperm interfaces at 13 DAP. RNA-seq profiling revealed that endosperm cells in close contact with the embryo scutellum have a distinct transcriptional signature, allowing us to define an endosperm zone we named the EAS for endosperm adjacent to scutellum, which is specialized in nutrient transport based on Gene Ontology (GO) enrichment analysis. In situ hybridization shows that the EAS is confined to one to three endosperm cell layers adjacent to the scutellum, whereas kinetic analyses show that the EAS is present when the scutellum emerges at around 9 DAP and persists throughout embryo growth, up to ∼20 DAP. The detection of cell death in the EAS together with the impaired expression of EAS marker genes in an embryo specific mutant suggest that the EAS is a developmentally dynamic interface influenced by the presence of the neighboring growing embryo.

RESULTS

RNA-Seq Profiling of 13-DAP Maize Kernel Compartments and Embryo/Endosperm Interfaces

To obtain the gene expression patterns of embryo/endosperm interfaces in maize kernels, six (sub)compartments were hand-dissected for transcriptomic analysis (Figure 1; Supplemental Figure 1). The three whole compartments were the maternal tissues excluding the pedicel, which were labeled pericarp (Per), the whole endosperm (End), and the whole embryo (Emb; Figure 1). The subcompartments corresponding to three distinct embryo/endosperm interfaces were the SAL (the single endosperm cell layer at the adaxial side of the embryo), the apical scutellum (AS; corresponding to the embryo tip composed uniquely of scutellum tissues without the embryo axis), and a new region that we named the EAS, corresponding to several layers of endosperm cells in close contact with the scutellum at the abaxial side of the embryo (Figure 1; Supplemental Figure 1). The tissues were collected from kernels of inbred line B73 (used to establish the maize reference genome) at 13 DAP (embryo size of ∼2.5 mm), the earliest developmental stage at which hand dissection of these embryo/endosperm interfaces was feasible, and also the transition from early development to the filling phase.

Figure 1.

Scheme Representing the Six (Sub)compartments Hand-Dissected for Transcriptomics Analysis at Maize Embryo/Endosperm Interfaces.

Ab, abaxial; Ad, adaxial.

For each of the six samples, four biological replicates, each composed of a pool of dissected tissues from two different plants, were produced (Supplemental Table 1). A total of 24 RNA-seq libraries were constructed and sequenced in paired-end mode using Illumina HiSeq2000 technology. The resulting reads (on average 62 million pairs per sample) were checked for quality, cleaned, and mapped to the current version of the B73 maize reference genome (AGP v4). On average, 95.8% ± 0.4% of the pairs were mapped, and on average, 78.3% ± 5.3% corresponded to annotated genes (Supplemental Figure 2A). Pairs that mapped to multiple genes (10.2% ± 5.3%) or to no gene (5.2% ± 1.1%), as well as ambiguous hits (1.5% ± 0.6%), were filtered (Supplemental Figure 2A). A gene was considered to be not expressed if it gives rise to less than one read per million. At least 25,000 genes were found to be expressed per replicate, with the largest number found in the SAL (∼30,000 genes expressed; Supplemental Figure 2B). The results generated for each replicate are available in Supplemental Data Set 1. Venn diagrams were generated to visualize overlaps between the sets of genes expressed in the three whole compartments (Per, Emb, and End) and between the sets of genes expressed in the End and the two endosperm subcompartments (EAS and SAL; Figures 2A and 2B).

Figure 2.

Validation of the RNA-Seq Approach.

(A) and (B) Venn diagrams. For each fraction, the number of genes expressed is indicated: for End, Emb, and Per (A) and for End, EAS, and SAL (B). The total number of genes expressed for all three compartments analyzed are indicated below each Venn diagram.

(C) PCA of the 24 RNA samples consisting of four biological replicates of Per, AS, Emb, End, EAS, and SAL.

(D) to (G) Graphs represent the expression levels (read counts were normalized using the trimmed mean of M-value method) in the different samples of the two embryo-specific genes ZmLec1 and ZmNac124 (D), the two endosperm-specific genes O2 and ZmZou (E), the two aleurone-specific genes Al9 and Zm00001d024120 (F), and the three Esr genes Esr1, Esr2, and Esr3 (G). Gray and black y scale numbering in (F) are for Zm00001d024120 and Al9 expression levels, respectively, and in (G) are for ESr1 and Esr3 (gray) and Esr2 (black).

In order to assess the relationships between the different samples, a principal component analysis (PCA) was performed (Figure 2C). As expected, biological replicates grouped together, indicating experimental reproducibility. The PCA also revealed distinct sample populations corresponding to each (sub)compartment, with the exception of the AS and Emb samples, which were partially superimposed (Figure 2C). Interestingly, the two endosperm interfaces SAL and EAS formed groups that were distinct both from each other and from the whole endosperm samples. The EAS was more similar to the whole endosperm than to the SAL, indicating a more similar transcriptomic landscape (Figure 2C).

To explore potential contamination between tissues during the dissection process, the expression profiles of previously identified marker genes with tissue-specific expression patterns were investigated (Figures 2D to 2G). Leafy cotyledon1 (ZmLec1, Zm00001d017898) and Nam/Ataf/Cuc124 (ZmNac124, Zm00001d046126; named ZmNac6 by Zimmermann and Werr [2005]), two embryo-specific genes, were specifically expressed in the embryo samples in our data set (Figure 2D). As expected, ZmLec1 was more strongly expressed in the Emb than in the AS sample (Zhang et al., 2002). Absence of ZmNac124 expression in the AS was consistent with the strong and specific in situ hybridization signal for this gene in the basal part of the embryonic axis (Zimmermann and Werr, 2005). The two endosperm-specific genes ZmZhoupi/Opaque11 (ZmZou/O11, Zm00001d003677) and Opaque2 (O2, Zm00001d018971; Schmidt et al., 1990; Grimault et al., 2015; Feng et al., 2018) were found to be strongly expressed in the End and EAS and weakly in the SAL sample (Figure 2E). The weak expression in the Per sample was unexpected but consistent with other transcriptomics data (Sekhon et al., 2011) and could also reflect possible contamination of the Per samples with the aleurone layer, since the aleurone layer has a tendency to stick to the pericarp (see Discussion). In addition, the preferential expression of Aleurone9 (Al9, Zm00001d012572) and Zm00001d024120 genes in the aleurone (Gómez et al., 2009; Li, 2014; Zhan et al., 2015) was reflected by a stronger signal in SAL compared with End (Figure 2F). Al9 and Zm00001d024120 also showed a signal in the pericarp samples, again indicating a possible contamination of the Per samples by SAL (Figure 2F).

The expression patterns of ESR marker genes (Esr1, Esr2, and Esr3) were also evaluated in our samples. At 13 DAP, the ESR comprises a small endosperm region situated at the base of embryo, around the suspensor (Opsahl-Ferstad et al., 1997). We observed elevated expression of ESR markers in the SAL and to a lesser extent in the EAS (Figure 2G). Previous in situ hybridizations of Esr1 transcripts showed that Esr1 expression is restricted to the ESR and absent from the EAS and most if not all of the SAL at both 12 and 14 DAP (Opsahl-Ferstad et al., 1997). However, the basal part of the SAL is in direct contact with the ESR (Opsahl-Ferstad et al., 1997), and the published data do not exclude the possibility that the Esr1 signal might extend to the SAL in this basal part. The apparent elevated expression of ESR marker genes in our SAL transcriptomes may thus reveal contamination with adjacent ESR cells during dissection and/or expression in the basal part of the SAL.

In order to compare our full transcriptomic data set with published RNA-seq data, we used a unique, spatially resolved maize kernel transcriptome (Zhan et al., 2015). Although different (sub)compartments and developmental stages (8 versus 13 DAP) were used, we retreated both RNA-seq raw data sets using the same bioinformatic pipeline and the same genome version (see Methods) in order to increase comparability. We then performed a PCA on joint data sets. The first principal component (PC1) carries 43.7% of the variance and clearly separates the two data sets (Supplemental Figure 3A). It may reflect a batch effect, a combination of the biological effect of the age of sampling (8 versus 13 DAP) and technical differences between the two transcriptomes (growing environment, library preparation, etc.). The next components group together samples from the two data sets and still carry a relatively high fraction of the variance (26.9 and 9.7% for PC2 and PC3, respectively). When PC2 was plotted against PC3, 13-DAP Emb is most similar to 8-DAP Emb samples among the 8-DAP samples (Supplemental Figure 3B), indicating that although important differences exist between these two data sets, these two embryo samples share some similarities in their transcriptomic profiles. Likewise, the 13-DAP AS is most similar to the 8-DAP Emb samples among 8-DAP samples (Supplemental Figure 3B). The 13-DAP SAL groups most closely to the two 8-DAP samples BETL and ESR. Interestingly, the 13-DAP EAS samples form an independent group that is closer to the two 8-DAP SE samples (which are the central starchy endosperm [CSE] and the conducting zone [CZ]) among the 8-DAP samples (Supplemental Figure 3B).

In summary, we have generated RNA-seq profiles from 13-DAP maize kernel compartments and embryo/endosperm interfaces. We have made this data available to the community in a user-friendly format via the eFP Browser (http://bar.utoronto.ca/efp_maize/cgi-bin/efpWeb.cgi?dataSource=Maize_Kernel; see Supplemental Figure 4 for examples).

Preferentially Expressed Genes and Biological Processes Associated with Specific Maize Kernel (Sub)compartments

Differential expression analyses were performed between the six (sub)compartments by comparing expression levels between pairs of tissues using a likelihood ratio test with P values adjusted by the Benjamini-Hochberg procedure to control false discovery rates (see Methods). Genes with both adjusted P values lower than 0.05 and an expression difference of fourfold or greater [log₂(fold change) ≥ 2] were classed as differentially expressed genes (DEGs; Supplemental Table 2). The full lists of DEGs for the 15 intertissue comparisons performed are available in Supplemental Data Set 2.

To identify the biological processes associated with the DEGs, a GO analysis was performed. Due to the limited resources available, a new genome-wide annotation of all predicted proteins was performed and linked to GO terms (see Methods). In a first instance, GO terms enriched in the two zygotic compartments Emb and End were identified by analyzing DEGs upregulated in each compartment compared with the two other main compartments (Table 1). The top 10 GO terms enriched in the DEGs upregulated in the embryo relative to endosperm and pericarp showed a significant enrichment in GO terms related to the cell cycle, DNA organization, and cytoskeleton organization, consistent with the extensive developmental and mitotic activity within the embryo at this stage (Table 1). In contrast, the GO terms enriched in the DEGs upregulated in the endosperm relative to embryo and pericarp were linked to metabolic functions such as nutrient reservoir activity and starch biosynthetic process (Table 1). These enrichments were consistent with the fact that the endosperm is a nutrient storage compartment where starch and reserve proteins are synthesized (Nelson and Pan, 1995; Zheng and Wang, 2015).

Table 1. Top 10 GO Terms (Sorted by Increasing P Value) Enriched in the DEGs Upregulated in One Main Compartment Compared with the Two Others.

GO Term	Level	DEGs/Total	Enrichment	P
DEGs Emb versus (End and Per): 1,601 of 29,845 genes
GO:0010369 chromocenter (C6)	(C6)	8/13	11.47	2.11E-09
GO:0042555 MCM complex	(C3)	9/18	9.32	5.65E-08
GO:0003777 microtubule motor activity	(F9)	24/144	3.11	1.92E-07
GO:0007018 microtubule-based movement	(P4)	24/144	3.11	1.92E-07
GO:0006928 movement of cell or subcellular component	(P3)	24/145	3.09	2.20E-07
GO:0098687 chromosomal region	(C5)	13/50	4.85	2.34E-07
GO:0008092 cytoskeletal protein binding	(F4)	42/348	2.25	3.35E-07
GO:0003774 motor activity	(F8)	24/149	3.00	3.76E-07
GO:0031492 nucleosomal DNA binding	(F5)	7/16	8.15	5.89E-07
GO:0000786 nucleosome	(C4)	19/105	3.37	6.85E-07
DEGs End versus (Emb and Per): 818 of 29,845 genes
GO:0045735 nutrient reservoir activity	(F2)	11/47	8.54	3.59E-09
GO:0019252 starch biosynthetic process	(P8)	7/27	9.46	4.30E-07
GO:0019863 IgE binding	(F5)	3/4	27.36	5.60E-07
GO:0019865 Ig binding	(F4)	3/4	27.36	5.60E-07
GO:0004866 endopeptidase inhibitor activity	(F6)	9/55	5.97	2.17E-06
GO:0010466 negative regulation of peptidase activity	(P7)	9/55	5.97	2.17E-06
GO:0010951 negative regulation of endopeptidase activity	(P8)	9/55	5.97	2.17E-06
GO:0030414 peptidase inhibitor activity	(F5)	9/55	5.97	2.17E-06
GO:0052548 regulation of endopeptidase activity	(P7)	9/55	5.97	2.17E-06
GO:0061135 endopeptidase regulator activity	(F5)	9/55	5.97	2.17E-06

Emb, embryo; End, endosperm; Per, pericarp. Level indicates minimal depth of the GO term in the GO tree, where P = biological process, F = molecular function, and C = cellular component. DEGs/Total indicates the number of genes associated with the GO term in the DEGs list/the number of GO term-annotated genes expressed in at least one sample. Enrichment is defined in Methods.

Enrichment for Putative Transporters at the Endosperm/Embryo Interface

Focusing in on the embryo/endosperm interfaces, DEGs between the three subcompartments (AS, SAL, and EAS) and their whole compartments of origin were identified (Supplemental Table 2). A total of 682 genes were found to be differentially expressed between AS and Emb according to the above criteria. Among them, 82 were more strongly and 600 were more weakly expressed in AS compared with Emb samples (Supplemental Table 2). As expected, ZmNac124, which is expressed in the coleorhiza (Figures 2D, 3C, and 3D; Zimmermann and Werr, 2005), was found among the genes showing reduced expression in the apical scutellum. Only the GO term DNA binding transcription factor activity was found to be significantly enriched in our analysis in the comparison of AS versus Emb (Table 2).

Figure 3.

In Situ Hybridization on 13-DAP Maize Kernel Probes.

Probes detecting GFP (negative control; [A] and [B]), Zmnac124 (positive control; [C] and [D]), Sweet14a ([E] and [F]), Sweet15a ([G] and [H]), Umamit_eas1 ([I] and [J]), Pepb11 ([K] and [L]), Zm00001d017285 ([M] and [N]), and Scl_eas1 ([O] and [P]) are shown. Arrows indicate main in situ hybridization signals. emb, embryo; end, endosperm; ped, pedicel; per, pericarp. Bars = 500 µm in (A), (C), (E), (G), (I), (J), (K), (M), and (O) and 1000 µm in (B), (D), (F), (H), (L), (N), and (P).

Table 2. All GO Terms from F3 (Molecular Function at Level 3) Significantly Enriched in the DEGs Upregulated in a Subcompartment Compared with its Compartment of Origin.

GO Term	Level	DEGs/Total	Enrichment	P
DEGs AS versus Emb: 82 of 29,845 genes
GO:0003700 DNA binding transcription factor activity	(F3)	8/743	3.91	0.000202
DEGs EAS versus End: 485 of 2,9845 genes
GO:0022857 transmembrane transporter activity	(F3)	26/1,111	1.44	0.0256
DEGs SAL versus End: 1,995 of 29,845 genes
GO:0008289 lipid binding	(F3)	24/183	1.96	0.000529
GO:0003700 DNA binding transcription factor activity	(F3)	70/743	1.41	0.00158
GO:0022857 transmembrane transporter activity	(F3)	97/1,111	1.31	0.00305
GO:0005319 lipid transporter activity	(F3)	4/30	1.99	0.0468

Emb, embryo; End, endosperm. Level indicates minimal depth of the GO term in the GO tree, where F = molecular function. DEGs/Total indicates the number of genes associated with the GO term in the DEGs list/the number of GO term-annotated genes expressed in at least one sample. Enrichment is defined in Methods.

The comparison between the EAS and the End revealed 1498 DEGs with 485 genes showing stronger expression in the EAS than in the End and 1013 genes with the inverse profile (Supplemental Table 2). Among the genes more strongly expressed in the EAS, our GO analysis revealed a significant enrichment in only one GO term (GO analysis on molecular function terms at the F3 level): transmembrane transporter activity (Table 2), which suggests a stronger expression of transporter-encoding genes in the EAS compared with End.

Finally, 2975 genes were found to be differentially expressed between SAL and End, 1995 corresponding to genes more strongly expressed in the SAL and 980 corresponding to genes with lower expression levels in the SAL (Supplemental Table 2). Interestingly, in the first group, our GO analysis revealed an enrichment in two (out of four) GO terms related to transport (Table 2).

A closer look at gene families encoding transporters among DEGs confirmed the overrepresentation seen in the GO analysis and revealed differences between the SAL and EAS. Among the genes that were at least eight times more strongly expressed compared with End, 8.45% (45/532) of the genes enriched in the SAL and 16.04% (34/212) of the genes enriched in the EAS have at least one ortholog in rice (Oryza sativa) or in Arabidopsis (Arabidopsis thaliana) that encodes a putative transporter (Table 3). In the SAL, transcripts of genes encoding MATEs (Multi-antimicrobial extrusion proteins), which have been implicated in a diverse array of functions (for review, see Upadhyay et al., 2019), and ABC (ATP binding cassette) transporters were found to be the most strongly enriched, whereas in the EAS, genes encoding transporters from the MtN21/UMAMIT (Usually Multiple Acids Move In And Out Transporter), MtN3/SWEET, and ABC transporter families were the most represented. When looking at the putative molecules transported, a large number of genes encoding putative amino acid transporters were found to show stronger expression in the EAS than in the End samples, although genes encoding transporters for various other molecules, including sugars, heavy metals, phosphate, inorganic ions, or nucleotides, also showed stronger expression (Table 3). Regarding the comparison of SAL versus End, transporters mainly annotated as involved in amino acid and inorganic ions transport were identified (Table 3). In summary, our work shows that both SAL and EAS cells strongly express putative transporter-encoding genes, suggesting that these cells are characterized by an elevated transmembrane transport of various molecules and potentially mediate nutrient repartitioning around the embryo. However, each tissue preferentially expresses different classes of transporters, with MtN21/UMAMIT and MtN3/SWEET transporters involved in amino acid and/or sugar transport, respectively, more likely to be enriched in the EAS.

Table 3. Number of Genes Encoding Putative Transporters in the DEGs Upregulated in the SAL or in the EAS Compared with the Endosperm per Family and per Molecules Putatively Transported.

Transporter Family	Ratio SAL/End > 8	Ratio EAS/End > 8
MtN21/UMAMIT	1	5
MtN3/SWEET	0	3
AAP	1	2
MATE	7	1
ABC	3	4
GDU	1	2
VIT	0	2
Phosphate transporters	0	2
Other	32	13
Total No.	45	34
Percentage in the gene list	8.45%	16.04%
Molecules Putatively Transported	Ratio SAL/End > 8	Ratio EAS/End > 8
Amino acids and/or auxin	7	12
Nucleotides	1	1
Heavy metal	3	3
Sugar	0	4
Phosphate	0	2
Other inorganic ions	5	2

Analysis was done base on orthology to rice and Arabidopsis (see Methods).

The EAS Is Restricted to One to Three Endosperm Cell Layers Adjacent to the Scutellum

The SAL has both cellular and biochemical characteristics of the aleurone, making it inherently different from other endosperm tissues (Zheng and Wang, 2014; Gontarek and Becraft, 2017). In contrast, EAS cells have not been reported to have distinct features that allow them to be distinguished cytologically from SE cells, which constitute the majority of the volume of the endosperm (Van Lammeren, 1987). However, our transcriptomic analysis suggests that these cells deploy a specific genetic program. In order to (1) confirm EAS expression specificity and (2) provide a more precise spatial resolution to define and characterize this new region, in situ hybridizations were performed with a set of six genes more than 10-fold enriched in the EAS transcriptome compared with the End transcriptome (Supplemental Table 3; see Supplemental Figure 4 for two examples of the eFP browser pattern). Three of these genes encode putative transporters, namely Sweet14a (Zm00001d007365) and Sweet15a (Zm00001d050577), encoding putative sugar transporters of the SWEET family, and Zm00001d009063, called Umamit_eas1, encoding a putative amino acid transporter belonging to the UMAMIT family (Müller et al., 2015; Sosso et al., 2015). The three remaining genes were Phosphatidylethanolamine binding protein11 (Pebp11, Zm00001d037439), a Serine carboxypeptidase-like (Zm00001d014983 or Scl_eas1), and Zm00001d017285, a gene with no name and unknown function (Supplemental Table 3). The negative control chosen for in situ hybridizations was an antisense probe generated against a GFP-encoding open reading frame. The positive control was ZmNac124, which is specifically expressed in the Emb compartment in our transcriptome (Figure 2D; Supplemental Table 3) and which had previously been shown by in situ hybridization to be expressed in specific embryonic tissues (Zimmermann and Werr, 2005). In situ hybridizations were performed on 13-DAP kernels, the same stage as used for the transcriptome analysis. The four probes detecting Sweet15a (Figures 3G and 3H), Pepb11 (Figures 3K and 3L), Zm00001d017285 (Figures 3M and 3N), and Scl_eas1 (Figures 3O and 3P) gave a strong signal restricted to a few layers of endosperm cells immediately adjacent to the scutellum, with little or no expression detected elsewhere in the kernel. For the probe directed against Sweet14a, the signal was strong in the EAS but was also present, albeit more weakly, in other kernel tissues, especially in the embryo and aleurone (Figures 3E and 3F). The probe against Umamit_eas1 gave a weaker signal restricted to the apical part of the EAS region, consistent with the lower expression levels of this gene in our transcriptome data (Supplemental Table 3). However, the signal for Umamit_eas1 was specific to these EAS cells (Figures 3I and 3J). These results confirmed that EAS cells have a specific transcriptional program and that this program (and thus the EAS) is restricted to one to three layers of endosperm cells adjacent to the scutellum.

The EAS Is a Dynamic Region Reflecting the Period of Strong Embryo Growth

To evaluate the dynamics of gene expression in the EAS during kernel development, in situ hybridizations were performed on kernels at different developmental stages (9, 11, 14, 17, and 20 DAP; Figure 4; Supplemental Figure 5). The four probes giving strong and EAS-specific signal at 13 DAP (Sweet15a, Pepb11, Zm00001d017285, and Scl_eas1) were used (Figure 4). In 9-DAP kernels, the probes for Pepb11 and Scl_eas1 showed no signal, whereas those for Sweet15a and Zm00001d017285 gave a strong signal in the endosperm cells adjacent to the apical part of the embryo (Figure 4; Supplemental Figure 5). This signal was restricted to a few layers of cells in the vicinity of the nascent scutellum. At this stage, the basal part of the embryo was still surrounded by ESR cells, and no signal was detected in this region. At 11 DAP, all four probes tested gave a very strong signal in the layers of endosperm cells adjacent to the scutellum. At 14 and 17 DAP, the signal was still detected and restricted to the cell layers in close contact with the embryo (Figure 4; Supplemental Figure 5). Finally, at 20 DAP, the signal decreased for all four probes, with a total disappearance for Sweet15a. Together, these results revealed that the EAS transcriptomic region was restricted to a defined time window. Its onset at 9 DAP was concomitant with the formation of the scutellum, marking a switch in embryo/endosperm interactions from an ESR/embryo to an EAS/scutellum interface. Its decline occurred around 20 DAP, when rapid embryo growth comes to an end.

Figure 4.

In Situ Hybridization of Four Probes Detecting EAS Marker Genes (Sweet15a, Pepb11, Zm00001d017285, and Scl_eas1) on Kernel Sections at Different Developmental Stages.

Probe detecting GFP was used as a negative control. Images are zoomed from Supplemental Figure 5. For each image, the name of the probe is indicated at the top and the stage on the left. Arrows indicate main in situ hybridization signals. emb, embryo; end, endosperm; nu, nucellus; ped, pedicel; per, pericarp. Bars = 200 µm for 9-DAP kernels and 500 µm for the other stages.

EAS Cells Originate from the SE and Undergo Cell Death

Despite the preferential or specific expression of EAS marker genes, and consistent with their SE-like morphology, EAS cells also showed some transcriptomic characteristics of the SE, such as a strong expression of genes encoding ZEIN storage proteins (Supplemental Figure 6). The presence of Zein transcripts in the EAS region is supported by in situ hybridization data (Woo et al., 2001). In order to perform a more global comparison, we asked to which samples from the Zhan et al. (2015) data set (at 8 DAP) our EAS transcriptome was most similar, using PCA (Supplemental Figure 3). Interestingly, on PC3, EAS at 13 DAP was most similar to two specific SE subregions at 8 DAP: the CSE and the CZ (Supplemental Figure 3B). As EAS cells, both CZ and CSE have no striking morphological characteristics differentiating them from the SE, strengthening the idea that EAS originate from the SE.

To address the question of EAS cell fate in proximity to the scutellum, sagittal sections of the EAS/scutellum interface were both hybridized with an EAS-specific probe (against Sweet15a transcripts) and stained with calcofluor to reveal cell walls (Figures 5A and 5B). The accumulation of cell wall material occurred at the endosperm interface with the scutellum, which may result from the compaction of crushed endosperm cells. Interestingly, in situ hybridization signal for the EAS marker genes was found in the first uncrushed cell layer (Figure 5B). An appealing model is that EAS cells are actually SE cells that are forced into juxtaposition with the scutellum because of the invasive growth of the embryo into the SE during kernel development (Figure 6), suggesting that the EAS program may not be fixed within a static group of cells but instead may be triggered as SE cells enter into contact with the scutellum.

Figure 5.

Crushed Cell Walls and Cell Death Occur in the EAS.

(A) and (B) Calcofluor staining of cell walls of 13-DAP maize kernel sections (A) together with in situ hybridization with Sweet15a antisense probes (B) on sagittal sections. Solid white arrows indicate the accumulation of crushed cell walls, and empty black arrows indicate in situ hybridization signal.

(C) and (D) TUNEL labeling of 15-DAP kernels. Fluorescein labeling of the TUNEL-positive nuclei is shown in green and propidium iodide counterstaining in purple. Arrows indicate the nucleus stained by TUNEL in the EAS.

emb, embryo; end, endosperm. Bars = 200 µm in (A) and (B), 500 µm in (C), and 100 µm in (D).

Figure 6.

Scheme Summarizing the EAS Dynamic.

Three different consecutive times points (t0, t1, and t2) are represented. Embryo scutellum invades (representing by arrows) the surrounding SE cells, which enter in cell death (yellow stars). The endosperm cell layers in contact with the embryo scutellum are regularly eliminated, resulting in an accumulation of crushed cell walls. Additional endosperm cells are thus recruited as EAS, as the embryo grows. Three cells are labeled by a cross pattern to illustrates this dynamic. Emb, embryo scutellum; End, endosperm.

If this model is correct, EAS cells would be likely to be successively eliminated as they come into contact with the embryo. Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assays were performed on 15-DAP kernels to visualize DNA degradation as a potential indicator of the presence of dying cells. However, it should be noted that TUNEL signals are neither a fully reliable indicator of all forms of cell death nor diagnostic of specific cell death programs (Charriaut-Marlangue and Ben-Ari, 1995; Labat-Moleur et al., 1998). In addition to the principal component and coleoptile regions, both of which had previously been shown to give strong TUNEL signals (Giuliani et al., 2002; Kladnik et al., 2004), we also observed a clear TUNEL-positive signal in some EAS cells in close contact with the scutellum (Figures 5C and 5D). This result is consistent with the possibility that a form of cell death occurs at this interface.

To clarify whether transcriptional activation of EAS-specific genes is linked to the initiation of known cell death programs, the expression levels of genes associated with programmed cell death in plants were analyzed (Supplemental Figure 7; Fagundes et al., 2015; Arora et al., 2017). Surprisingly, orthologs of none of the previously identified programmed cell death-associated genes were found to be particularly upregulated in the EAS compared with other samples. In addition, no enrichment of GO terms associated with programmed cell death was found in the DEGs strongly expressed in the EAS relative to the End samples. Similar results were obtained when comparing genes strongly expressed in the EAS relative to the Emb, which remains alive (the GO term programmed cell death [GO:0012501] was slightly enriched [ratio of 1.56], but in a not statistically significant manner [P = 0.098]). These data suggested that either only a small proportion of EAS cells undergo cell death or that crushing of EAS cells does not trigger a classical programmed cell death program. A parallel could be drawn with accidental cell death defined in animals, in which cells die as a result of their immediate structural breakdown due to physicochemical, physical, or mechanical cues (Galluzzi et al., 2015).

Impaired Expression of Some EAS Marker Genes in emb Mutants

To test to what extent the proximity of the embryo/scutellum was required for EAS gene expression, the embryo specific mutation emb8522 was used in an R-scm-2 genetic background enhancing the early embryo-deficient phenotype (Sosso et al., 2012). In this background, the recessive emb8522 mutation produced vestigial embryos composed of a small heap of cells. Nevertheless, a cavity corresponding to the size of a normal embryo was generated that was only very partially occupied by the aborting embryo (Heckel et al., 1999; Sosso et al., 2012). Self-fertilization of heterozygous plants carrying the emb8522 mutation was performed, and in situ hybridizations were performed on 13-DAP sibling kernels with either phenotypically wild-type or mutant embryos to visualize the transcripts of four EAS marker genes (Figure 7). Similar EAS-specific expression patterns were observed in R-scm-2 kernels with embryos (Figure 7) to those observed in B73 kernels (Figures 3 and 4) for all genes tested, indicating a conservation of EAS cell identity in this genetic background. In emb kernels, the probes detecting Zm00001d017285 and Sweet15a still showed a signal in the EAS region but with an altered distribution (Figure 7). In emb kernels, Zm00001d017285 expression was found to be restricted to the apical part of the embryo cavity and Sweet15a expression expanded to the SAL, suggesting an inhibitory role of the normal embryo on Sweet15a expression in the SAL tissue. Interestingly, the two other EAS marker genes tested showed either only very weak expression (Scl_eas1) or no expression (Pepb11) in emb kernels, indicating a promoting effect of the normal embryo on the expression of these two genes (Figure 7).

Figure 7.

In Situ Hybridization with Several Probes Marking the EAS on 13-DAP Maize Kernel Sections of the R-scm-2 Genetic Background.

Probe detecting GFP was used as a negative control. Kernels come from a self-pollination of a mother plant heterozygous for the emb8522 mutation. The top row (Rscm2 +emb) corresponds to kernels with embryo (emb8522 +/− or +/+), and the bottom row (Rscm2 –emb) corresponds to kernels without embryo (emb8522 −/−). Arrows indicate the main in situ hybridization signal. emb, embryo; emb cav, embryo cavity containing an aborted embryo; end, endosperm; per, pericarp. Bars = 1000 µm.

DISCUSSION

Transcriptomes at Embryo/Endosperm Interfaces

As in other flowering plants, seed development in maize is governed by specific temporal and spatial genetic programs, distinguishing early development, filling, and maturation on the one hand and embryo, endosperm, and pericarp on the other (Downs et al., 2013; Lu et al., 2013; Chen et al., 2014; Li et al., 2014; Qu et al., 2016; Meng et al., 2018). Recently, a transcriptome analysis of the nucellus (including the fertilized embryo sac) increased the temporal resolution and allowed unprecedented access to information regarding the genetic control of early seed development (Yi et al., 2019). The most detailed spatial analysis to date used laser-capture microdissection on 8-DAP kernels (Zhan et al., 2015) to reveal the expression of specific populations of genes in the maternal tissues, the embryo, and the main endosperm cell types, namely ESR, BETL, AL, and SE (which was subdivided into CSE and CZ). Although providing an extremely valuable resource, these studies did not address the question of whether specific transcriptional domains exist at embryo/endosperm interfaces.

The endosperm and embryo are complex compartments with several morphologically and functionally distinct domains (Olsen, 2004; Sabelli and Larkins, 2009). Because they undergo complex and coordinated developmental programs, the interfaces between the embryo and the endosperm represent important, and constantly changing, zones of exchange, both in term of nutrition and communication (Nowack et al., 2010; Ingram and Gutierrez-Marcos, 2015; Widiez et al., 2017). In order to understand these interactions, two subdomains of the endosperm and one subdomain of the embryo were hand-dissected: the SAL at the adaxial side of the embryo; the SE in close contact with the abaxial side of the embryo (EAS); and the scutellum of the embryo (AS). Kernels at 13 DAP were chosen for our analysis because at this stage the embryo has emerged from the ESR and is establishing new interactions with endosperm. From a practical point of view, 13 DAP is also the earliest stage allowing reliable hand dissection of the chosen interfaces.

Contamination with neighboring tissues is an important issue in any dissection experiment. For example, in Arabidopsis, an extremely valuable and globally very reliable resource generated by laser-capture microdissection (Le et al., 2010; Belmonte et al., 2013) was recently shown to contain some inter-compartment contamination, which caused problems for the investigation of parental contributions to the transcriptomes of early embryos and endosperms (Schon and Nodine, 2017). In our study, precautions were taken to limit intertissue contamination by (1) washing each sample before RNA extraction (see Methods) and (2) generating four biological replicates for each tissue. Marker gene analysis confirmed the conformity of the samples, with the exception of a potential minor contamination of pericarp by the AL, suggested by the apparent expression of both the Al9 and Zm00001d024120 aleurone marker genes and the endosperm marker genes ZmZou/O11 and O2 in the pericarp sample (Figures 2E and 2F). This could have been caused by the tendency of the AL to stick either to the SE or to the pericarp. In addition, residual ESR tissues at 13 DAP might have contaminated both our SAL and EAS samples (Figure 2G).

The EAS, an Endosperm Subdomain Likely Involved in Carbon and Nitrogen Fluxes from the Endosperm to the Embryo

Transcriptomic profiling of the two endosperm interfaces with the embryo (SAL and EAS) revealed specific transcriptional signatures. While this could have been expected for the cytologically distinct SAL, it was rather unexpected for the cell layers adjacent to the abaxial side of the embryo, which do not present any obvious cytological differences from other SE cells (Van Lammeren, 1987). Based on the observed enrichment of hundreds of transcripts in these cell layers, they represent a novel subdomain of the endosperm, which we named the EAS.

GO analysis revealed a significant enrichment in the GO category transmembrane transporter activity for both the SAL and EAS and additionally for lipid transporter activity for SAL (Table 2). A closer look at DEGs for both EAS and SAL shows the presence of different transporter gene families (Table 3). Interestingly, many Umamits and Sweets, thought to transport amino acids/auxin and sugars, respectively, were found to be enriched in the EAS. UMAMITs and SWEETs are considered to be bidirectional transporters, although they tend to act as exporters when located at the plasma membrane, exporting nutrients down concentration gradients generated by sinks in adjacent tissues (Chen et al., 2012; Müller et al., 2015). Two nonexclusive hypotheses could explain the elevated expression of transporter-encoding genes in the EAS: either these cells actively take up nutrients that arrive from the BETL via the SE and then export them into the apoplastic space surrounding the growing embryo, or they are simply involved in recycling nutrients from dying endosperm cells that are crushed by the growing embryo.

With regard to nutrient uptake on the embryo side, one might expect the expression of genes encoding nutrient importers at the surface of the scutellum in order to take up apoplastic metabolites. However, in our AS transcriptome, we were not able to detect differentially expressed importer-encoding genes with respect to the entire Emb. While this could suggest that the regulation of importer activity does not occur at the transcriptional level, it seems more likely that our transcriptomic comparison AS versus Emb was not well designed for the identification of such genes, since the whole embryo is mainly composed of scutellum tissues.

In the future, a more detailed comparison of the gene expression profiles of the BETL (import) and the EAS (export) regions could be informative. The BETL is an interface specialized in nutrient transfer from maternal phloem terminals to the endosperm (Chourey and Hueros, 2017). The hexose transporter SWEET4c is preferentially expressed in the BETL, and loss of function of Sweet4c results in the production of a shriveled endosperm, illustrating the critical importance of hexose transport in the BETL for normal endosperm growth (Sosso et al., 2015). Interestingly, the Sweet4c gene is also found in the DEGs, showing strong expression in the EAS compared with the endosperm as a whole, possibly suggesting commonalities between BETL and EAS function. EAS-specific knockdown of Sweet4c might be one strategy to test this hypothesis and to address the question of possible redundancy with Sweet14a and Sweet15a, also enriched in the EAS. Nonetheless, notable differences exist between the EAS and the BETL. First, BETL cells have structural features including dramatic cell wall ingrowths that make them unique in the endosperm (Leroux et al., 2014; Chourey and Hueros, 2017). In contrast, EAS cells cannot be morphologically differentiated from the SE (Van Lammeren, 1987). Second, the BETL represents a static interface, contrary to the EAS, which is displaced as the embryo scutellum expands (Figure 6) during the most rapid growth phase of the embryo (Chen et al., 2014).

The EAS Is a Developmentally Dynamic Interface

The detection of DNA fragmentation, a characteristic of cell death, in EAS cells (Figures 5C and 5D) together with an accumulation of cell wall material in this zone (Figures 5A and 5B) suggested that endosperm cells are eliminated as the embryo grows. An important question is whether this involves a genetically controlled cell-autonomous death or a more atypical and passive cell death process caused by embryo growth. In the Arabidopsis seed, where most of the endosperm degenerates during seed development, the expression of developmental cell death marker genes such as PLANT ASPARTIC PROTEASE A3 or BIFUNCTIONAL NUCLEASE1 has been detected at the embryo interface (Olvera-Carrillo et al., 2015; Fourquin et al., 2016). In maize, less is known about the molecular actors involved in developmental cell death. To the best of our knowledge, cell death marker genes have not been comprehensively identified in maize. Nevertheless, a survey of putative cell death marker genes derived from comparisons with other plant systems showed their expression in EAS cells, but without any significant enrichment compared with other compartments (Supplemental Figure 7; Supplemental Data Set 2). Although cell death in the EAS could be triggered by the activation of unknown cell death-associated genes, a more likely explanation for our observations could be a dilution of the transcriptional signal in the EAS transcriptome, making it undetectable. This is supported by TUNEL staining, which revealed a very localized signal limited to a few cells at the immediate interface with the embryo (Figures 5C and 5D). In addition, previous cell death staining with Evans blue did not reveal any massive cell death in the EAS, further supporting the hypothesis of very localized cell death events (Young and Gallie, 2000).

The precise spatial organization of cell death and transporter expression remains unclear, but the expression of transporters might allow the recycling of nutrients from the cells before they die. As these cells are SE in origin, they could already have initiated nutrient storage at 13 DAP, as illustrated by substantial expression of Zein genes (Supplemental Figure 6). Nutrient recycling could be an advantageous way for the plant to efficiently reuse stored nutrients. Interestingly, in Arabidopsis, the STP13 gene, coding for a sugar transporter, is upregulated in several cell death contexts and the expression of many transporters increase during organ senescence, suggesting a function in nutrient recycling from dying cells (Norholm et al., 2006; van der Graaff et al., 2006; Zhang et al., 2014). However, the precise role of transporters in nutrient recycling remains poorly understood in plants.

The Importance of the Embryo for the Expression of EAS Marker Genes

Since the EAS is a mobile interface, forming adjacent to the expanding scutellum, we asked whether the presence/absence of the embryo influences the activation of EAS marker genes (Figure 7). In emb8522 mutant kernels, which produce a seemingly empty, but normally sized, embryo cavity containing an aborted embryo (Heckel et al., 1999; Sosso et al., 2012), the expression of different EAS marker genes was affected in different ways. The Sweet15a gene was still expressed in EAS cells but also became strongly expressed at the opposite embryo/endosperm interface (SAL). Based on the precedent of the Sweet4c transporter gene, which is induced by sugar (Sosso et al., 2015), it is possible that a similar induction could occur in the case of Sweet15a. The absence of a normal embryo could lead to a build-up of sucrose in the embryo cavity of emb8522 mutants, leading to such an induction. In contrast, the expression domain of the Zm00001d017285 marker gene is reduced in emb8522 mutants, with expression becoming restricted to the apical part of the EAS. Finally, the expression of Pepb11 and Scl_eas1 is dramatically reduced in emb8522 mutants compared with phenotypically wild-type kernels. Our results suggest that EAS-specific gene expression could be a result of several independent factors, some of which could originate from the endosperm and others from the embryo. The mechanisms involved in embryo cavity formation remain elusive, although a recent study showed that the SHOHAI1 protein is required in the endosperm for the formation of the embryo cavity (Mimura et al., 2018).

Interestingly, the expression of both Pepb11 and Scl_eas1 initiates relatively late in the EAS, whereas the expression of Sweet15a and Zm00001d017285 initiates before 9 DAP. This suggests the presence of at least two transcriptional programs in the EAS: one initiating early and weakly influenced by the embryo, and a second activated later and more strongly embryo-dependent. The generation of comparable transcriptomes at earlier developmental stages could help us identify the key signals activating gene expression in the EAS and potentially pinpoint transcription factors regulating gene expression in this tissue. In parallel, phenotypic analysis of loss-of-function mutants of genes enriched in the EAS is needed to further elucidate the biological role of this novel endosperm subdomain.

METHODS

Plant Material and Plant Growth Conditions

The maize (Zea mays) A188 and B73 inbred lines were cultivated in the S2 greenhouse with a 16-h illumination period (100 W/m²) at 24/19°C (day/night) and without control of the relative humidity, as described previously (Rousseau et al., 2015; Gilles et al., 2017). The A188 inbred line depicted in Supplemental Figure 1 was cultivated in a growth chamber as described previously (Doll et al., 2019). The emb8522 mutant in the R-scm-2 background (Sosso et al., 2012) and the B73 plants used for in situ hybridization were grown in a field plot located at the École Normale Supérieure, de Lyon, France.

Isolation of Maize Kernel Compartments

Kernel (sub)compartments of the B73 inbred line were hand-dissected and quickly washed with Dulbecco’s phosphate-buffered saline solution (HyClone, SH30378.02) before freezing them in liquid nitrogen. For each (sub)compartment, four independent biological replicates were produced (Supplemental Table 1). For each biological replicate, the material comes from two independent, 13-d-old maize ears [i.e., eight different ears were used for each (sub)compartment]. Within each biological replicate, tissues from 4 to 84 kernels were pooled depending on the size of the considered (sub)compartment (Supplemental Table 1).

RNA Extraction and RNA-Seq

Total RNAs were extracted with TRIzol reagent, treated with DNase using the Qiagen RNase-Free DNase Set, and purified using Qiagen RNeasy columns according to the supplier’s instructions. RNA-seq libraries were constructed according to the TruSeq_RNA_SamplePrep_v2_Guide_15026495_C protocol (Illumina). Sequencing was performed with an Illumina HiSeq2000 at the Institut de Génomique-Centre National de Séquençage. The RNA-seq samples were sequenced in paired-end mode with a sizing of 260 bp and a read length of 2 × 100 bases. Six samples were pooled on each lane of a HiSeq2000 (Illumina) and tagged with individual bar-coded adapters, giving ∼62 million pairs per sample. All steps of the experiment, from growth conditions to bioinformatics analyses, were managed in the CATdb database (Gagnot et al., 2008; http://tools.ips2.u-psud.fr/CATdb/) with project identifier NGS2014_21_SeedCom, according to the minimum information about a high-throughput sequencing experiment standard (http://fged.org/projects/minseqe/).

RNA-Seq Read Processing and Gene Expression Analysis

RNA-seq reads from all samples were processed using the same pipeline from trimming to counts of transcript abundance as follows. Read quality control was performed using the FastQC (S. Andrew, http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). The raw data (fastq files) were trimmed using fastx Toolkit version 0.0.13 (http://hannonlab.cshl.edu/fastx_toolkit/) for Phred quality score > 20, read length > 30 bases, and ribosomal sequences were removed with the sortMeRNA tool (Kopylova et al., 2012).

The genomic mapper TopHat2 (Langmead and Salzberg, 2012) was used to align read pairs against the maize B73 genome sequence (AGP v4; Jiao et al., 2017) using the gene annotation version 4.32 provided as a GFF file (Wang et al., 2016). The abundance of each isoform was calculated with the tool HTSeq-count (Anders et al., 2015) that counts only paired-end reads for which paired-end reads map unambiguously one gene, thus removing multiple hits (default option union). The genome sequence and annotation file used was retrieved from the Gramene database (http://www.gramene.org/, release 51, in September 2016; Gupta et al., 2016).

Choices for the differential analysis were made based on Rigaill et al. (2018). To increase the detection power by limiting the number of statistical tests (Bourgon et al., 2010), we performed an independent filtering by discarding genes that did not have at least one read after a count per million analysis in at least one-half of the samples. Library size was normalized using the method trimmed mean of M-values, and count distribution was modeled with a negative binomial generalized linear. Dispersion was estimated by the edgeR package (version 1.12.0; McCarthy et al., 2012) in the statistical software R (version 2.15.0; R Development Core Team, 2005). Pairwise expression differences were performed using likelihood ratio test, and P values were adjusted using the Benjamini-Hochberg procedure to control false discovery rate (Benjamini and Hochberg, 1995). A gene was declared to have a differential expression if its adjusted P value was lower than 0.05. The FPKM value (fragments per kilobase of transcript per million mapped reads) is used to estimate and compare gene expressions in eFP Browser. This normalization is based on the number of paired-end reads that mapped each gene, taking into account the gene length and the library size. This RNA-seq read processing method was used for all analyses presented in this article except for the comparison of our RNA-seq with published RNA-seq, for which the data were processed as described below.

Comparison of our RNA-Seq Data Set with a Published RNA-Seq Data Set

For the comparison of our data set with previously published RNA-seq data (Zhan et al., 2015), the raw RNA-seq reads published by Zhan et al. (2015) were retrieved at the National Center for Biotechnology Information Sequence Read Archive (Leinonen et al., 2011) from Bioproject PRJNA265095 (runs SRR1633457 to SRR1633478). That represents 53 million pairs of length 2 × 100 bases for 22 samples. The reads from the two data sets were processed using the same pipeline: quality control was performed using FastQC version 0.11.7 (S. Andrew, http://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Sequencing adapters were clipped using cutadapt v1.16 (Martin, 2011), sequencer artifacts were removed using fastx Toolkit version 0.0.14 (http://hannonlab.cshl.edu/fastx_toolkit/), and custom Perl scripts were applied to trim regions of reads having an average Phred quality score (Ewing and Green, 1998) lower than 28 bases over a sliding window of 4 bases. We noticed that some samples retrieved from the Sequence Read Archive exhibited a high rRNA content. We built a maize rRNA database by comparing sequences from Silva (Quast et al., 2013) and RFAM (Kalvari et al., 2018) with the maize B73 genome sequence; we then used this custom database to filter the RNA-seq reads with sortMeRNA version 2.1b (Kopylova et al., 2012). Reads shorter than 25 bases at the end of this processing, or with no mate, were discarded.

The genomic mapper hisat2 v2.2.0 (Kim et al., 2015) was used to align read pairs against the maize B73 genome sequence (AGP v4; Jiao et al., 2017) using the gene annotation version 4.40 provided as a GFF file (Wang et al., 2016). A first mapping pass was performed with the complete set of read pairs to discover unannotated splicing sites before the per-sample mapping, with options -k 10-no-discordant -no-softclip and allowing introns of length 40 to 150,000 bp. Mapped reads were counted by gene (not distinguishing isoforms) using FeatureCounts (Liao et al., 2014). The genome sequence and annotation file used was retrieved from the Gramene database (http://www.gramene.org/, release 51, in September 2016; Gupta et al., 2016).

Normalization, differential analysis, and PCAs were performed with DESeq2 (Love et al., 2014) under R version 3.6.2 (R Development Core Team, 2005). The PCAs were done using the 1000 genes with the highest variance, after applying the variance stabilization transformation described by Anders and Huber (2010) and implemented in DESeq2 v1.24.0. In parallel, FPKM values and confidence intervals were estimated using Cufflinks version 2.2.1 (Roberts et al., 2011) with options -frag-bias-correct-multi-read-correct-max-multiread-fraction 1.

Venn Diagrams

For each compartment/subcompartment, the mean expression of the four samples was calculated. If the value of the normalized read counts was equal or superior to one, the gene was considered as expressed. Venn diagrams were drawn using tools available at http://bioinformatics.psb.ugent.be/webtools/Venn/.

Functional Annotation of the Maize Transcriptome and GO Term Enrichment Analysis

The maize B73 genome sequence v4 (Jiao et al., 2017) and the gene annotation v4.40 were used to predict transcript sequences using the gffread script from the Cufflinks package v2.2.1 (Trapnell et al., 2013). In each isoform sequence, the putative open reading frames were identified using TransDecoder (Haas et al., 2013; https://github.com/TransDecoder/TransDecoder/wiki), and the amino acid sequence was predicted. From 46,272 genes, 138,270 transcripts were predicted, leading to 149,699 amino acid sequences.

The predicted protein sequences were annotated for functional domains with InterProScan v5.27-66.0 (Jones et al., 2014) using databases Pfam v31.0 (Punta et al., 2012) and Panther 12.0 (Mi et al., 2013). They were also compared with UniProtKB protein database version 2017_12 (UniProt Consortium, 2019). The complete Swiss-Prot database of curated proteins was used, (containing 41,689 plant sequences and 514,699 nonplant sequences) but only the plant subset of the noncurated database TrEMBL (containing 5,979,810 sequences). The comparison was performed using WU-BlastP v2.0MP (Altschul et al., 1990) with parameters W=3 Q=7 R=2 matrix=BLOSUM80 B=200 V=200 E=1e-6 hitdist=60 hspsepqmax=30 hspsepsmax=30 sump postsw. BLAST output was filtered using custom Perl scripts to keep only matches with log₁₀(e-value) no lower than 75% of the best log₁₀(e-value). GO terms (Ashburner et al., 2000) associated with matched proteins were retrieved from AmiGO (Carbon et al., 2009) with all their ancestors in the GO graph, using the SQL interface. For each maize protein, we kept the GO terms associated with all its matched proteins or at least with five matched proteins. For each maize gene, we merged the GO terms of all its isoforms.

For subsets of genes selected based on their expression pattern, we used our GO annotation to perform an enrichment analysis. The enrichment of a gene subset in a specific GO term is defined as follows:

A hypergeometric test (R version 3.2.3; R Development Core Team, 2005) was applied to assess the significance of enrichment/depletion of each subset (Pavlidis et al., 2004; Falcon and Gentleman, 2007). Custom Perl scripts using GraphViz (Ellson et al., 2001; https://graphviz.gitlab.io/) were used to browse the GO graph and identify enrichments or depletions that were both statistically significant and biologically relevant. Only genes with at least one match on UniProt and only GO terms with at least one gene in the subset were considered for all those statistical tests.

Analysis of Gene Categories and Orthology

Analysis of orthology to rice (Oryza sativa) and Arabidopsis (Arabidopsis thaliana; Table 3) was based on Maize GDB annotations (https://www.maizegdb.org/; Andorf et al., 2016). The Zein genes were selected based on a previous gene list (Chen et al., 2014, 2017) and on Gramene database annotations (http://www.gramene.org/; Gupta et al., 2016). The list of cell death-associated genes was based on previously published lists (Fagundes et al., 2015; Arora et al., 2017). Heat maps were drawn with the online Heatmapper tool (http://www2.heatmapper.ca/; Babicki et al., 2016).

Kernel Fixation and in Situ Hybridization

Kernels were fixed in 4% (w/v) paraformaldehyde (pH 7 adjusted with H₂SO₄) for 2 h under vacuum. For increased fixation efficiency, the two upper corners of the kernels were cut and vacuum was broken every 15 min. Kernels were dehydrated and included with Paraplast according to the protocol described by Jackson (1991). Sections of 10 to 15 µm were cut with an HM355S microtome and attached on Adhesion Slides Superfrost Ultra plus (ThermoFisher Scientific). RNA probes were amplified from genomic DNA or cDNA (Supplemental Table 4) and labeled by digoxigenin (DIG) using the T7 reverse transcriptase kit of Promega, according to company instructions. RNA probes were then hydrolyzed in carbonate buffer (120 mM Na₂CO₃ and 80 mM NaHCO₃) at 60°C for various times depending on the probe length (Supplemental Table 4) in order to obtain RNA fragments between 200 and 300 nucleotides (Jackson, 1991).

For the prehybridization of the sections, the protocol described by Jackson (1991) was followed with some slight changes: pronase was replaced by proteinase K (1 µg/mL; ThermoFisher Scientific) in its buffer (100 mM Tris and 50 mM EDTA, pH 8), and formaldehyde was replaced by paraformaldehyde as described above. For each slide, 1 μL of RNA probe was diluted in 74 μL of DIG Easy Hyb buffer (Roche), denatured for 3 min at 80°C, and dropped on a section that was immediately covered by a cover slip. Hybridization was performed overnight at 50°C in a hermetically closed box. Initial posthybridization treatments were performed using gentle shaking as follows: 0.1× SSC buffer (from stock solution 20× SSC [3 M NaCl and 300 mM trisodium citrate, adjusted to pH 7 with HCl]) and 0.5% (v/v) SDS for 30 min at 50°C to remove the cover slips. Two baths were used of 1.5 h in 2× SSC buffer mixed with 50% formamide at 50°C and followed by 5 min in Tris-buffered saline (TBS) buffer (400 mM NaCl and 0.1 mM Tris-HCl, pH 7.5) at room temperature. Slides were then incubated in 0.5% (w/v) blocking reagent solution (Roche) for 1 h, followed by 30 min in TBS buffer with 1% (w/v) BSA and 0.3% (v/v) Triton X-100. Probe immunodetection was performed in a wet chamber with 500 μL per slide of 0.225 units/mL anti-DIG antibodies (Anti-Digoxigenin-AP, Fab fragments; Sigma-Aldrich) diluted in TBS with 1% (w/v) BSA and 0.3% (v/v) Triton X-100. After 1.5 h of incubation, slides were washed three times for 20 min in TBS buffer with 1% (w/v) BSA and 0.3% (v/v) Triton X-100 and equilibrated in buffer 5 (100 mM Tris-HCl, pH 9.5, 100 mM NaCl, and 50 mM MgCl₂). Revelation was performed overnight in darkness in a buffer with 0.5 g/L nitroblue tetrazolium and 0.2 g/L 5-bromo-4-chloro-3-indolyl phosphate. Slides were finally washed four times in water to stop the reaction and were optionally stained with calcofluor (fluorescent brightener 28; Sigma-Aldrich) and mounted in entellan (VWR). Photographs were taken either with a VHX900F digital microscope (Keyence) or for magnification with an AxioImager 2 microscope (Zeiss).

Terminal Deoxynucleotidyl Transferase dUTP Nick End Labeling Assay

Kernels at 15 DAP were fixed in paraformaldehyde, included in Paraplast, and sectioned as described above. Paraplast was removed by successive baths in xylene (2 × 5 min), and samples were then rehydrated through the following ethanol series: ethanol 100% (5 min), ethanol 95% (v/v; 3 min), ethanol 70% (v/v; 3 min), ethanol 50% (v/v; 3 min), NaCl 0.85% (w/v) in water (5 min), and Dulbecco’s PBS solution (5 min). Sections were then permeabilized using proteinase K (1 µg/mL; ThermoFisher Scientific) for 10 min at 37°C and fixed again in paraformaldehyde. Sections were washed in PBS, and terminal deoxynucleotidyl transferase dUTP nick end labeling was performed with the ApoAlert DNA Fragmentation Assay Kit (Takara) according to the manufacturer’s instructions. Sections were then counterstained with propidium iodide (1 µg/mL in PBS) for 15 min in darkness before being washed three times for 5 min in water. Slides were mounted in Anti-fade Vectashield (Vector Laboratories). The fluorescein-dUTP incorporated at the free 3′ hydroxyl ends of fragmented DNA was excited at 520 nm and propidium iodide at 620 nm. Images were taken on a spinning-disk microscope, with a CSU22 confocal head (Yokogawa) and an Ixon897 EMCCD camera (Andor), on a DMI4000 microscope (Leica).

Accession Numbers

RNA-seq raw data were deposited in the international repository Gene Expression Omnibus (Edgar et al., 2002; http://www.ncbi.nlm.nih.gov/geo) under project identifier GSE110060. RNA-seq data as FPKM values are available via the eFP Browser engine (http://bar.utoronto.ca/efp_maize/cgi-bin/efpWeb.cgi?dataSource=Maize_Kernel), which paints the expression data onto images representing the samples used to generate the RNA-seq data. Custom codes and scripts are available at http://flower.ens-lyon.fr/maize/seedcom/.

Supplemental Data

• Supplemental Figure 1. Illustration of hand-dissected maize kernel compartments and sub-compartments.

• Supplemental Figure 2. Proportion of mapped reads and expressed genes.

• Supplemental Figure 3. Relationships between transcriptomic data-sets at 13 DAP (this study), and at 8 DAP (Zhan et al., 2015) assessed by PCA analysis.

• Supplemental Figure 4. Example of eFP Browser views

• Supplemental Figure 5. Whole kernel views of the in situ hybridizations presented in Figure 4.

• Supplemental Figure 6. Heat map of Zein precursor gene expression.

• Supplemental Figure 7. Heat maps for genes potentially involved in programmed cell death.

• Supplemental Table 1. Number of kernels used for each of the four biological replicates.

• Supplemental Table 2. Number of genes differentially expressed between a sub compartment and its compartment of origin.

• Supplemental Table 3. Mean expression values and gene IDs of genes selected for in situ hybridization.

• Supplemental Table 4. Primers used in this study and conditions for RNA probes synthesis.

• Supplemental Data Set 1. Number of normalized read counts per gene annotated in the AGP v4 version of the B73 maize genome.

• Supplemental Data Set 2. Pairwise comparison of gene expression levels between the tissues.

DIVE Curated Terms

The following phenotypic, genotypic, and functional terms are of significance to the work described in this paper:

• Esr1 Gramene: Zm00001d027820

• Esr1 Araport: Zm00001d027820

• STP13 Gramene: At5g26340

• STP13 Araport: At5g26340

• PCA Gramene: Principal component analysis

• PCA Araport: Principal component analysis

• Umamit_eas1 Gramene: Zm00001d009063

• Umamit_eas1 Araport: Zm00001d009063

• Pepb11 Gramene: Zm00001d037439

• Pepb11 Araport: Zm00001d037439

• Scl_eas1 Gramene: Zm00001d014983

• Scl_eas1 Araport: Zm00001d014983

• Sweet14a Gramene: Zm00001d007365

• Sweet14a Araport: Zm00001d007365

• Sweet15a Gramene: Zm00001d050577

• Sweet15a Araport: Zm00001d050577

• ZmNac124 Gramene: Zm00001d046126

• ZmNac124 Araport: Zm00001d046126