Dynamic Expression of Imprinted Genes Associates with Maternally Controlled Nutrient Allocation during Maize Endosperm Development

Mo17 Genome Assembly, Single Nucleotide Polymorphism Identification, and RNA-Seq Data Processing

We performed reference-guided assembly of the Mo17 genome based on the B73 genome (release 5b.60, http://maizesequence.org) using 650-Gb Illumina short reads obtained from the Joint Genome Institute. The resulting Mo17 genome assembly comprised 2,058,527,894 bases containing a total of 117,847,390-bp short gaps. Thus, the coverage of the Mo17 genome was ∼94.3%, producing a similar genome size to the 2,066,432,971 bases in the B73 reference. The identification of single nucleotide polymorphisms (SNPs) and insertions and deletions (INDELs) was performed using the SHORE pipeline (Schneeberger et al., 2011), which identified 6,557,611 SNPs, 157,994 insertions, and 191,549 deletions between the B73 and Mo17 genomes. In the working gene set (WGS), 38,557 protein-coding genes contained at least one SNP (see Supplemental Figure 1 online). The 90-nucleotide pair-end RNA-Seq reads from the 12 libraries were mapped to the B73 and Mo17 genomes using Tophat2 (Kim et al., 2013). The B73-unique, Mo17-unique, and B73/Mo17 common reads were distinguished based on SNP information. After excluding genes with few reads in the 12 samples, 24,984 SNP-containing genes expressed (at least six reads from combined 12 samples) were retained to measure allelic expression.

Because of short gaps in the Mo17 genome assembly that resulted from insufficient read coverage or high levels of variation between Mo17 and B73, ∼10 to 15% of the RNA-Seq reads from the 12 samples of the hybrid kernel and endosperm could not be mapped to the Mo17 genome, but they could be mapped to the B73 genome. This issue potentially led to biased reporting of higher expression for many B73 alleles, which might influence the identification of the imprinted genes. Thus, we normalized the expression levels of the Mo17 and B73 alleles according to the gap size (see Supplemental Figure 2A and Supplemental Methods1 online). Additionally, genes containing gaps in their coding regions longer than 100 bp were not included in further analyses because such long gaps might influence the accuracy of the allelic expression measurements (see Supplemental Figure 2B online). The expression information for each gene per sample included the count of reads common to the B73 and Mo17 alleles and of reads unique to either B73 or Mo17 alleles, as distinguished by SNP information. Whereas the B73/Mo17 common reads exhibited a 1:1 ratio, the B73- and Mo17-unique reads with different ratios were used to measure allelic expression (see Supplemental Figure 3 online). Specifically, the fraction of maternal reads, including the B73-unique reads in the B73×Mo17 cross and the Mo17-unique reads in the Mo17×B73 cross, represents the expression of maternal alleles and vice versa for paternal expression. The allelic expression inferred from allele-specific reads was proportional to total gene expression, indicating the ratio of allele-specific reads properly reflected the ratio of the maternal and paternal transcripts in the RNA samples (see Supplemental Figure 4 online).

Postfertilization Activation of the Paternal Transcriptome

Upon double fertilization, the paternal genome inherited from the sperm cell is activated in the embryo and the endosperm. With support of informative SNPs and high-throughput sequencing, we examined the process of paternal genome activation in 3- and 5-DAP maize hybrid kernels. To minimize artifacts from ambiguous mapping, genes identified with more than five paternally derived reads at 0 DAP were excluded from further analysis. By plotting the paternal versus maternal expression for all retained genes, we observed a gradual activation of the paternal genome in 3- and 5-DAP kernels in terms of both the number of activated genes and their expression levels (Figure 1A). The number of genes with activated paternal alleles (>10 paternal reads from reciprocal crosses) increased from 0 at 0 DAP to 941 at 3 DAP and then to 4063 at 5 DAP. The corresponding fraction of paternal reads to total reads for the activated paternal alleles rose from 0 at 0 DAP to 0.0012 at 3 DAP and then to 0.0081 at 5 DAP. In 7-DAP endosperm, 11,027 genes containing SNPs were expressed; of these, 8128 (73.71%) exhibited the expected 2m:1p ratio (χ² test, P value > 0.001), whereas 2899 (26.29%) genes exhibited allele-biased expression (χ² test, P value ≤ 0.001). Similar proportions were observed in 10-DAP endosperm, with 72.75% (7691) of the genes exhibiting the expected ratio and 27.25% (2882) exhibiting an allelic bias. These results suggest that complete transcriptional activation of the paternal genome was achieved by 7 DAP in the endosperm. This observation is consistent with a previous report (Grimanelli et al., 2005). However, identification of paternal transcripts in the complex structure of the maize kernel indicated that at least a small fraction (<5%) of the paternal alleles might already be activated by 3 DAP. Although these genes had substantially fewer paternal transcripts than maternal transcripts at 3 and 5 DAP, we still cannot conclude an earlier activation of maternal compared with paternal alleles during maize seed development because ∼97.5% of the SNP-containing genes detected with at least 10 reads in the endosperm stages could also be detected in the 0-DAP kernel stage.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Paternal Genome Activation in the Maize Kernels and Endosperms.

(A) Full activation of the paternal genome is likely achieved by 7 DAP. The expression level of paternal alleles (y axis) and maternal alleles (x axis) are represented by the log₂-transformed sum of the paternally and maternally derived reads in the reciprocal crosses, respectively. The color scale in blue (low), yellow (medium), and red (high) represents the relative density of the genes. The solid diagonal line represents the expected 2m:1p ratio.

(B) Numbers of genes with transcriptionally activated paternal alleles in the reciprocal crosses at 3 and 5 DAP.

We observed higher numbers and transcript abundances of genes with activated paternal alleles in the Mo17×B73 cross than in the B73×Mo17 cross at 3 and 5 DAP (Figure 2B; see Supplemental Figure 5 online). Namely, at 3 DAP, 1194 (8.3%) genes with activated paternal alleles (more than five paternal reads) were identified in the Mo17×B73 compared with 874 (6.4%) in B73×Mo17 cross, respectively, and only 351 (20.4%) of these were commonly activated in the two reciprocal crosses (Figure 1B); at 5 DAP, 4295 (25.3%) genes were paternally activated in the Mo17×B73 compared with 3478 (20.3%) in B73×Mo17 crosses, and, surprisingly, only 2561 (49.1%) of these genes were commonly activated (Figure 1B), while in 7-, 10-, and 15-DAP endosperm, the proportion of commonly activated genes accounted for 74.3, 71.3, and 70.9%, respectively. To determine whether the activated paternal alleles in the two reciprocal crosses were functionally different, we compared the GO enrichments of the paternally activated genes in the 3- and 5-DAP kernels of the Mo17×B73 and B73×Mo17 crosses. Overall, their enriched GO terms were very similar, without any GO class demonstrating significant differences (see Supplemental Figure 6 online). This similarity suggested the early-activated genes in both crosses might belong to the same functional pathways, even though their paternal alleles were not simultaneously activated at the same stage.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Computational Identification of Imprinted Genes in the Maize Endosperm.

(A) Ratio-based cutoff to identify MEGs and PEGs. Spots clustered at the top right corners have >90% maternal reads (MEGs), whereas spots clustered in the bottom left corners have <30% of the maternal reads (PEGs). Imprinted genes in Waters et al. (2011), Zhang et al. (2011), and identified in this study are represented in a circle, triangle, square, and cross, respectively.

(B) Numbers and average expression levels of imprinted genes in 7-, 10-, and 15-DAP endosperm.

(C) Hierarchical clustering of the 290 imprinted genes. Parent-of-origin expression patterns of imprinted genes were exhibited in the 0-, 3-, and 5-DAP kernels and 7-, 10-, and 15-DAP endosperm. The color scale from green (low) to red (high) represents the ratio of maternal or paternal reads, m represents maternal allele, and p represents paternal allele.

(D) Heat map of 30 imprinted genes expressed after fertilization. Their imprinting status was traced back to as early as 3 DAP, and 25 imprinted gens are persistent to 15 DAP. The color scale in blue (low), white (medium) and red (high) represents the relative expression level of maternal or paternal alleles, m represents maternal allele, and p represents paternal allele.

Computational Identification of Imprinted Genes in the Maize Hybrid Endosperm

Because most of the genes expressed in endosperm samples were also expressed in the whole kernel, we decided to identify imprinted genes among the 11,027, 10,573, and 7777 SNP-containing genes (more than five allele-specific reads in both reciprocal crosses) identified in the 7-, 10-, and 15-DAP endosperm samples, respectively (see Supplemental Data Set 1 online). We first performed a χ² test to determine whether the ratio of maternally to paternally derived reads significantly deviated from the normal ratio of 2m:1p in each cross. The genes exhibiting parent-specific patterns in both reciprocal crosses were selected as candidate imprinted genes. At a significance level of P = 0.001, 284, 606, and 190 genes were determined to have allele-biased, parent-of-origin-specific expression patterns at 7, 10, and 15 DAP, respectively (see Supplemental Figure 7A online). In total, 300 PEGs and 499 MEGs were identified from the three developing endosperm stages by a χ² test (see Supplemental Figure 7B online). Among the 499 MEGs, 418 (83.8%) were identified at 10 DAP only, and only 15 (3.0%) were identified at all three time points (see Supplemental Figure 7C online). Among the 300 PEGs, 213, 130, and 163 were observed at 7, 10, and 15 DAP, respectively. Of these, 75 (25%) were present at all three time points (see Supplemental Figure 7C online).

We further narrowed the list of candidate imprinted genes using more stringent criteria to precisely reflect the parent-of-origin expression status of the allele-specific genes; namely, 90% of the maternal reads and 70% of the paternal reads of the total SNP-associated reads (with a minimum 40 reads) that mapped to a gene in both reciprocal crosses were used to define the MEGs and PEGs, respectively (Figure 2A). The ratio-based cutoff shortened the list to 290 imprinted genes identified at three developmental stages. These included 36, 184, and 15 MEGs and 80, 50, and 48 PEGs at 7, 10, and 15 DAP, respectively (Figure 2B). Among these imprinted genes, 38 were considered to be imprinted at all three stages, 51 were considered to be imprinted at two stages, and the remaining 206 were considered to be imprinted at only one stage, with the majority of this last category imprinted at 10 DAP. In addition, whereas the average expression level of the PEGs gradually decreased from 7 to 15 DAP, the MEGs reached their highest expression level at 10 DAP (Figure 2B). Subsequently, we traced the 290 imprinted genes identified in endosperm back to the early 3- and 5-DAP kernel stages to examine their early imprinting status (Figure 2C). Although we cannot completely exclude the influence of maternal tissues, especially for the MEGs, the genes that were not expressed at 0 DAP are likely to exhibit postfertilization imprinting behavior within the endosperm. We only found 11 MEGs and 19 PEGs that were not expressed in 0-DAP kernels (Figure 2D). Therefore, these genes were considered to be newly activated after fertilization. The imprinted expression patterns of these genes were evidently established at either 3 or 5 DAP, and all were maintained as such at 7 and 10 DAP. Interestingly, some of the same genes exhibited a loss of their parent-specific, imprinted expression pattern subsequently at 15 DAP (Figure 2D).

A number of MEGs and PEGs we identified were found to have functional counterparts in other plants. Among the 11 MEGs, an APETALA2/B3-like transcription factor gene (GRMZM2G423393) (Figure 2D) is highly similar to NGATHA1 (NGA1) in Arabidopsis, which redundantly together with three other NGA genes (NGA2 to NGA4) direct style development in the gynoecium through the YUCCA (YUC)–mediated auxin synthesis pathway (Alvarez et al., 2009). Additionally, overexpression of the Brassica rapa gene Br-NGA1 in Arabidopsis reduces lateral organ growth by inhibiting cell proliferation (Kwon et al., 2011). Of the 19 PEGs, 14 were paternally expressed as early as 3 DAP, and most (13 of 14) of their imprinted expression persisted until 15 DAP (Figure 2D). Among these genes, GRMZM2G472096 encodes a MADS box transcription factor gene that is highly similar to AGAMOUS-LIKE61, which was shown to be required for proper development of the central cell and seed in Arabidopsis (Steffen et al., 2008). Two additional PEGs, a zinc finger gene (AC191534.3_FG003) and a homeobox gene (GRMZM2G047104) are similar to Arabidopsis VARIANT IN METHYLATION1 (VIM1) and SAWADEE HOMEODOMAIN HOMOLOG1 (SHH1), respectively; both were reported to be associated with epigenetic regulation. In Arabidopsis, vim1 mutations produced DNA hypomethylation of the centromere and subsequent decondensation of heterochromatin (Woo et al., 2008), whereas SHH1 was shown to be required for de novo and maintenance methylation in the RNA-directed DNA methylation pathway (Law et al., 2011). Although the functional roles of these early expressed genes are largely unknown in maize, identification of these MEGs and PEGs suggests an important role for transcription and epigenetic reprogramming in control of early endosperm development.

Experimental Validation of Imprinted Genes

Two previous RNA-Seq studies identified 179 and 100 imprinted genes (223 genes in total) in 10- and 14-DAP endosperm samples (independently staged material), respectively, from reciprocal B73 and Mo17 crosses. However, only 56 of the imprinted genes were common (Waters et al., 2011; Zhang et al., 2011). We compared the 290 imprinted genes identified in our study with the 223 imprinted genes reported by Waters et al. (2011) and Zhang et al. (2011). Interestingly, only 81 of the imprinted genes identified in our study were also identified in one of the two previous studies (Figure 3A).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Experimental Validation of Previously Reported and Newly Identified Imprinted Genes by CAPS Assay.

(A) Venn diagram analysis of imprinted genes. The imprinted genes identified at 7, 10, and 15 DAP in this study and the imprinted genes previously reported by Waters et al. (2011) and Zhang et al. (2011) are represented in red, blue, green, purple, and orange.

(B) Validation of the status of 15 imprinted genes. Group I, seven genes previously identified by either Waters et al. (2011) or Zhang et al. (2011) also found in our data. Two genes (AC208539.3_FG002 and GRMZM2G449489) did not show any polymorphisms between parental alleles using NlaIII and BglII restriction enzymes, respectively, as reported by Zhang et al. (2011), but when using HindIII instead, GRMZM2G449489 exhibited parent-specific expression patterns; Group II, eight imprinted genes only identified in our data. BB, B73 × B73; MM, Mo17 × Mo17.

We selected two groups of genes for experimental validation using RT-PCR experiments followed by CAPS assays. The first group included seven genes previously identified by either Waters et al. (2011) or Zhang et al. (2011) that existed in our data. Based on this analysis, five of the seven genes, including three MEGs (GRMZM2G150134, GRMZM2G160687, and GRMZM2G370991) and two PEGs (GRMZM2G127160 and GRMZM2G028366), exhibited parent-specific imprinted expression patterns consistent with the previous reports (Figure 3B). The other two genes (GRMZM2G449489 and AC208539.3_FG002) did not exhibit any polymorphisms between the parental alleles when using BglII and NlaIII restriction enzymes, as reported by Zhang et al. (2011). Instead, when using the enzyme HindIII for GRMZM2G449489 in the CAPS assay, we observed a clear paternally biased expression pattern (Figure 3B). The second group included eight imprinted genes that were only identified in this study. All eight genes exhibited imprinting patterns by CAPS, consistent with the results from the RNA-Seq data (Figure 3B).

A Large Number of Both Nonimprinted Allele-Specific Genes and Inbred Line–Dependent Imprinted Genes Are Expressed in Endosperm

In addition to the imprinted genes that have allele-specific expression dependent on parent of origin, two more types of allele-specifically expressed genes were also investigated: nonimprinted, allele-specifically expressed genes and inbred line–dependent imprinted genes. Nonimprinted, allele-specific expression is widespread and believed to be responsible for quantitative variations in the phenotypic traits of hybrid offspring (Guo et al., 2008). In the three staged samples of endosperm, 1688 B73 allele-specific and 1130 Mo17 allele-specific genes were detected using a χ² test that exhibited expression dosages deviating from the 2m:1p ratio (P ≤ 0.001). Among these, 303 B73 genes and 128 Mo17 genes exhibited a non-parent-specific, monoallelic expression pattern under our stringent ratio-based criterion (≥90% reads in both reciprocal crosses; Figure 4A). Three genes (GRMZM2G446999, GRMZM2G000361, and AC199068.2_FG017) of this type were experimentally validated by CAPS experiments (Figure 4B).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Nonimprinted, Allele Specifically Expressed Genes and Inbred Line–Dependent Imprinted Genes.

(A) Numbers of inbred line–dependent imprinted genes and nonimprinted allele specifically expressed genes.

(B) CAPS validation of four genes showing allele-specific expression or inbred line–dependent imprinting patterns. BB, B73 × B73; MM, Mo17 × Mo17.

The second type of gene, previously reported in Arabidopsis, rice, and maize, is an inbred line–dependent imprinted gene that exhibits allele-specific expression in one direction of each reciprocal pair of crosses but biallelic expression in the other direction (Gehring et al., 2011; Hsieh et al., 2011; Waters et al., 2011; Wolff et al., 2011). Using the same threshold for identifying imprinted genes but only considering a single direction of allele-specific expression, we identified 377 inbred line–dependent MEGs (ILMEGs) and 129 inbred line–dependent PEGs (ILPEGs) in the B73×Mo17 cross and 292 ILMEGs and 235 ILPEGs in the Mo17×B73 cross (Figure 4A). We used CAPS to experimentally validate the inbred line–dependent imprinting pattern for GRMZM2G147226 (encoding a Epsin NH2-Terminal Homology /Vps27p/Hrs/Stam family protein) (Figure 4B). Overall, these data indicate that our RNA-Seq approach was sufficiently robust to identify multiple allele-specific expression patterns in maize endosperm.

Endosperm-Expressed MEGs and PEGs Encode Distinct Functional Groups

We performed Blast2GO analysis (Conesa et al., 2005; Conesa and Götz, 2008; Götz et al., 2008; Götz et al., 2011) to examine the functional distribution of the 290 imprinted genes identified in our study. Only four GO categories exhibited significantly higher functional enrichments compared with the background gene set (P ≤ 0.01). These groups included “regulation of gene expression by genetic imprinting,” “endosperm development,” “response to hormone stimulus (auxin response),” and “DNA binding (sequence-specific DNA binding) (see Supplemental Figure 8A online). These functional categories indicated involvement of the imprinted genes in hormone signaling pathways and transcriptional regulation of endosperm development. When the 194 MEGs and 96 PEGs were examined separately, we found only two categories and one category, respectively, exhibiting significant enrichments for MEGs and PEGs (P ≤ 0.01) (see Supplemental Table 1 and Supplemental Figures 8B and 8C online). The first category of MEGs was the “response to hormone stimulus” group, which included 13 genes, six of which were involved in the auxin-mediated transcriptional response. The second was the “cytoplasmic membrane-bounded vesicle” category, which included 26 genes encoding a variety of enzymes, transporter proteins, and cell wall formation proteins located in the cytoplasmic membrane, suggesting these MEGs might be involved in intercellular nutrient transport and signal transduction (see Supplemental Table 1 and Supplemental Figure 8B online). The only one enriched among the 29 PEGs was the “binding” category. Close inspection of the 29 PEGs revealed this category represented various forms of molecular interactions, such as DNA binding, RNA binding, protein binding, and ATP binding (see Supplemental Table 1 and Supplemental Figure 8C online). Although the exact functions of the MEGs and PEGs remain largely unknown, the two groups of imprinted genes enriched in distinct functional pathways might play specific roles in regulating endosperm development: Whereas the MEGs maybe involved in nutrient transport and hormone signaling, the PEGs are likely to be involved in the regulation of gene transcription.

To further determine whether the functions of inbred line–dependent imprinted genes were different from those of imprinted genes, we applied Blast2GO analysis separately on the 669 ILMEGs and 364 ILPEGs. We observed five GO categories of ILMEGs exhibiting significantly higher enrichment compared with the background (P ≤ 0.01) (see Supplemental Table 2 and Supplemental Figures 8D and 8E online). The “cellular lipid metabolic process” category included 39 ILMEGs that encode for numerous enzymes bound to the cell membrane. The second category was “inorganic anion trans-membrane transporter activity,” which included eight genes belonging to transporter protein families: an ATP-Binding Cassette transporter, a sulfate transporter, and several phosphate transporter proteins. The third category was the “lactate metabolic process” category, and the other two categories were the “cellular response to extracellular stimulus” and the “regulation of response to stress” categories. These findings are in contrast with the MEG categories, which suggested involvement in “response to organic substance” and “response to endogenous stimulus” (see Supplemental Table 2 and Supplemental Figure 8D online). Thus, the allele-specific expression of ILMEGs may be mainly involved in the response to environmental cues during endosperm development, similar to the hypothesis put forth by Guo et al. (2004). By contrast, we found that only nine of the 364 ILPEGs had significant enrichment in two categories: Four genes were enriched in the “histone acetyltransferase activity” category, and five genes were enriched in the “membrane lipid biosynthesis” category (see Supplemental Table 2 and Supplemental Figure 8E online). The four genes with histone acetylation activity included three members of the histone acetyltransferase family that are similar to Arabidopsis HISTONE ACETYLTRANSFERASE1 (HAC1) and HAC12. Both must be recruited to specific promoters to activate the transcription initiation of target genes by acetylating the N terminus of histone H3 or H4. The fourth gene is a homolog of RELATIVE OF EARLY FLOWERING6 (REF6), also known as JUMONJI DOMAIN-CONTAINING PROTEIN12 (JMJ12). The Arabidopsis REF6 gene has been shown to function as a histone H3 Lys 27 demethylase that facilitates removal of Histone H3 lysine 27 trimethylation (H3K27me3) during the gene transcription initiation process (Lu et al., 2011). Therefore, our data suggest that allelic expression may control specific cellular or biochemical processes that are essential for endosperm or kernel development.

Distinct Patterns of MEG and PEG Imprinting and Related Gene Functions Are Associated with Specific Stages of Endosperm Development

We found that MEGs and PEGs exhibited distinct patterns of gene imprinting during endosperm development, with PEGs predominantly detected at 7 DAP and MEGs predominantly detected at 10 DAP (Figure 2). To determine dynamic patterns of gene imprinting during development, we identified and tracked the patterns of imprinting and levels of expression for individual MEGs and PEGs. Remarkably, of the 194 MEGs identified in the three stages of endosperm development (Figures 2B and 2C), 150 (78.5%) were shown to be expressed maternally in both crosses uniquely at 10 DAP (Figure 5A; hereafter referred to as 10-DAP-specific). By contrast, the 7–DAP endosperm contained the highest number of PEGs (80 out of 96; Figures 2B and 2C), 26 of which were shown to be paternally expressed only at this developmental stage in both crosses (Figure 5A; 7-DAP–specific). Furthermore, we found the occurrence of 7-DAP–specific PEGs were mainly (22 of 26) due to the significantly deviated parental allele ratio, whereas the 10-DAP–specific MEGs could be primarily attributed to the sharply increased expression levels at the respective stages (Figures 5A and 5B). For example, among the latter group, 112 (74.6%) were expressed at least threefold higher at 10 DAP than at 7 and 15 DAP (Figure 5B).

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Development-Dependent, Dynamic Expression of Imprinted Genes.

(A) Imprinting status of 7-DAP–specific PEGs and 10-DAP–specific MEGs. The 26 PEGs were detected at 7 DAP but not at 10 and 15 DAP mainly because the parental allele expression ratios of most PEGs (22/26) significantly deviated from 2m:1p only at 7 DAP but not at 10 and 15 DAP, whereas 150 MEGs were only detected in 10-DAP endosperm due to their much higher expression levels compared with 7- and 15-DAP endosperms. RPKM: Reads per kilobase per million.

(B) Average expression levels of 7-DAP–specific PEGs and 10-DAP–specific MEGs at the three endosperm developmental stages according to RNA-Seq data.

(C) Validation of three 7-DAP–specific PEGs in developmental kernels and endosperms. BB, B73 × B73; MM, Mo17 × Mo17.

(D) Quantitative RT-PCR analysis of nine 10-DAP–specific MEGs.GRMZM2G058032 was used as a control, with expression level increased from 7 to 15 DAP in reciprocal crosses, and the maize thioredoxin gene (gene ID GRMZM2G066612) was used as an internal reference gene. The results indicated that the 10-DAP–specific MEGs were more abundantly expressed at 10 DAP compared with 7 and 15 DAP.

To validate the transient imprinting patterns of the 7-DAP–specific PEGs and 10-DAP–specific MEGs, we performed CAPS and quantitative RT-PCR experiments on three 7-DAP–specific PEGs and nine 10-DAP–specific MEGs (including one contrast gene, GRMZM2G058032, with an expression level that increased from 7 DAP to 10 and 15 DAP), respectively. All the three PEGs (GRMZM2G057436, GRMZM2G092101, and GRMZM2G088793) exhibited paternally biased expression patterns at 7 DAP in both reciprocal crosses, although their maternal alleles were weakly detected, but at later developmental stages, both parental alleles were strongly expressed in Mo17×B73 cross for GRMZM2G057436 and in reciprocal crosses for GRMZM2G092101 and GRMZM2G088793 (Figure 5C). All nine MEGs tested showed a predominant pattern of expression at 10 DAP with very low to nondetectable levels of expression at 7 or 15 DAP in both reciprocal crosses (Figure 5D). GO analysis of the 150 MEGs confirmed the observed enrichment in MEGs for the “response to organic substance,” “cytoplasmic membrane-bounded vesicles,” and “response to endogenous stimulus” categories were mostly due to the contribution of the 10-DAP–specific MEGs (see Supplemental Figure 9 online). The stage of maize endosperm development beginning around 10 DAP is likely critical for accumulation of nutrients prior to starch and zein storage protein synthesis, and it is likely coincident with the abrupt increase of auxin levels, as described for comparable stages of kernel development (Lur and Setter, 1993; Sabelli and Larkins, 2009). An iodine potassium iodide staining assay indicated the accumulation of starch granules began in 10-DAP endosperm under our specific growth conditions in Arizona (Figure 6A). Previous studies suggested the 10-DAP–specific MEGs identified in our study were likely associated with nutrient synthesis and transport mediated by auxin signaling pathways. For instance, pin-formed1 (pin1; GRMZM2G098643), which encodes an auxin efflux transporter highly expressed in the maize BETL and embryo-surrounding region (ESR) (Forestan et al., 2010), was identified as a 10-DAP–specific MEG (see Supplemental Data Sets 2 and 3 online). Expression of pin1 is coincident with a rapid auxin increase, which precedes starch and zein protein accumulation (Lur and Setter, 1993; Sabelli and Larkins, 2009). Additionally, among the 10-DAP–specific MEGs, two were identified to encode auxin response factors belonging to the auxin/indole-3-acetic acid family (GRMZM2G317900 and GRMZM2G066219). Taken together, these data suggest auxin-mediated nutrient uptake from maternal tissues to the endosperm is under control of the maternal genome.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Maternally Biased Expression of Auxin-Related Genes and Paternally Biased Expression of Epigenetic-Related Genes.

(A) Iodine potassium iodide staining analysis of the 7-, 10-, and 15-DAP endosperms from reciprocal crosses. The results indicated that starch accumulation began at 10 DAP. Micropylar region of the kernel is located in the lower left-hand side of each panel.

(B) Maternally biased expression of auxin-related genes and paternally biased expression of epigenetic-related genes. Top panel: 11 of 82 auxin-related genes exhibit maternally biased expression at 10 DAP with a χ² test ≤ 0.05 and a parental expression ratio of over 4m:1p in reciprocal crosses. Bottom panel: Eight of 111 epigenetic-related genes exhibit paternally biased expression at 15 DAP with a χ² test ≤ 0.05 and a parental expression ratio of less than 1m:1p in reciprocal crosses.

To address this hypothesis using available data and further understand the nature of imprinted gene functions, we selected 82 SNP-containing genes that are potentially involved in auxin biosynthesis, transport, response, and ubiquitination pathways (auxin-related) based on Maize Genetics and Genomics Database annotations. We used a less stringent set of criteria, namely χ² (P ≤ 0.05) with 4m:1p and 1m:1p ratio cutoffs, to examine whether the expression dosages of their parental alleles deviated from the expected 2m:1p ratio. We then plotted the maternal and paternal expression ratios separately for genes expressed at 7, 10, and 15 DAP (Figure 6B). At 7 DAP, 80 genes were centered on the diagonal line of 2m:1p ratio, except for two genes exhibiting paternal bias. At 10 DAP, 11 genes exhibited maternally biased expression, including four that had been classified as 10-DAP–specific MEGs; only one 10-DAP gene exhibited paternal bias (Figure 6B). At 15 DAP, the number of maternally biased auxin-related genes decreased to two genes (Figure 6B). This indicates a preponderance of auxin-related genes among maternally biased genes expressed at 10 DAP.

Also, because we detected certain PEG functions to be associated with epigenetic regulation during endosperm development, we performed a similar analysis of the 111 SNP-containing epigenetic-related genes collected from ChromDB (Gendler et al., 2008). We found that five, three, and eight genes exhibited paternally biased expression at 7, 10, and 15 DAP, respectively, whereas one, four, and one genes exhibited maternally biased expression at the same three stages (Figure 6B). Interestingly, in contrast with the 10-DAP–prevalent pattern of maternally biased auxin-related genes, the paternally biased epigenetic-related genes were instead more prevalent in the 7- and 15-DAP endosperm. Together, these results suggest that the maternally biased, auxin-related genes and the paternally biased, epigenetic-related genes have coordinated and yet complementary expression patterns during maize endosperm development.