Conserved noncoding sequences and de novo Mutator insertion alleles are imprinted in maize

Abstract

Genomic imprinting is an epigenetic phenomenon in which differential allele expression occurs in a parent-of-origin-dependent manner. Imprinting in plants is tightly linked to transposable elements (TEs), and it has been hypothesized that genomic imprinting may be a consequence of demethylation of TEs. Here, we performed high-throughput sequencing of ribonucleic acids from four maize (Zea mays) endosperms that segregated newly silenced Mutator (Mu) transposons and identified 110 paternally expressed imprinted genes (PEGs) and 139 maternally expressed imprinted genes (MEGs). Additionally, two potentially novel paternally suppressed MEGs are associated with de novo Mu insertions. In addition, we find evidence for parent-of-origin effects on expression of 407 conserved noncoding sequences (CNSs) in maize endosperm. The imprinted CNSs are largely localized within genic regions and near genes, but the imprinting status of the CNSs are largely independent of their associated genes. Both imprinted CNSs and PEGs have been subject to relaxed selection. However, our data suggest that although MEGs were already subject to a higher mutation rate prior to their being imprinted, imprinting may be the cause of the relaxed selection of PEGs. In addition, although DNA methylation is lower in the maternal alleles of both the maternally and paternally expressed CNSs (mat and pat CNSs), the difference between the two alleles in H3K27me3 levels was only observed in pat CNSs. Together, our findings point to the importance of both transposons and CNSs in genomic imprinting in maize.

Newly silenced transposable elements create two potentially novel imprinted genes, and conserved noncoding sequences show genomic imprinting.

Introduction

Genomic imprinting involves parent-of-origin differences in the expression of the maternal and paternal alleles of a gene. Imprinting is observed in both mammals and flowering plants (Law and Jacobsen, 2010). In mammals, imprinting occurs in the placenta as well as the embryos (Wang et al., 2013), whereas in plants it is primarily restricted to the endosperm, the triploid tissue that supplies nutrients and energy to the developing embryo and seed germination. The endosperm (2 maternal:1 paternal) arises as a result of the double fertilization of the central cell (2n) with one of the two sperm cells (n), whereas the second sperm cell (n) fertilizes the egg cell (n) to give rise to the diploid embryo (1 maternal:1 paternal) (Li and Berger, 2012; Rodrigues and Zilberman, 2015).

Genome-scale surveys of hybrid endosperms and other tissues derived from genetically distinct parents have identified dozens to hundreds of imprinted genes in Arabidopsis (Arabidopsis thaliana), rice (Oryza sativa), sorghum (Sorghum bicolor), maize (Zea mays), and wheat (Triticum aestivum) (Hsieh et al., 2011; Zhang et al., 2011; Rodrigues et al., 2013; Waters et al., 2013; Xin et al., 2013; Zhang et al., 2016; Chen et al., 2018; Yang et al., 2018). This research has demonstrated that the imprinted expression states are under epigenetic control that involves DNA methylation and histone modification that act as key regulators of plant imprinting (Kohler et al., 2012; Zhang et al., 2014; Rodrigues and Zilberman, 2015). Site specific hypomethylation of the maternally inherited DNA in endosperms has been observed for many species (Lauria et al., 2004; Gehring et al., 2009; Waters et al., 2011; Zhang et al., 2014). In Arabidopsis, this hypomethylation is introduced by demeter (DME), a 5-methylcytosine DNA glycosylase that is specially expressed in the central cell of female gametophyte as well as in the vegetative cell of the male gametophyte before fertilization (Choi et al., 2002; Gehring et al., 2006; Schoft et al., 2011; Ibarra et al., 2012). DME demethylates a large number of silenced transposable elements (TEs) and other repeats in the central cells (Gehring et al., 2009; Hsieh et al., 2009; Zemach et al., 2010; Ibarra et al., 2012), which can result in maternally specific expression of genes in the endosperm when the TEs or repeats are located near those genes. For instance, the imprinted expression of the flowering wageningen (FWA) gene in Arabidopsis depends on maternal-specific DNA hypomethylation in the promoter region, which contains two sine retroelement repeats (Kinoshita et al., 2004; Fujimoto et al., 2008). In addition to DME, recent research has revealed that that the allelic DNA methylation is linked to allelic gene expression of some imprinted genes that are dependent on another DNA demethylase repressor of silencing 1, which also can removes methylation in the central cell resulting in imprinting of some genes in maize (Xu et al., 2022). In addition to DNA methylation, trimethylation of histone H3 lysine 27 (H3K27me3), catalyzed by polycomb-repressive complex 2 (PRC2), is the only histone modification known to specifically associate with imprinting in plants (Wolff et al., 2011). Examples include silencing of the maternal alleles of pheres1 (PHE1) and the paternal alleles of MEDEA (MEA), which depends on fertilization independent seed (FIS), a putative H3K27 methyltransferase that is required for H3K27me3 and transcriptional repression of target genes (Kohler et al., 2005; Baroux et al., 2006). There is also evidence that DNA methylation and H3K27me3 are mutually exclusive and alternative repressive marks that may compensate for each other in the suppression of a subset of TEs (Weinhofer et al., 2010; Deleris et al., 2012).

TEs serve as major sites for both DNA methylation and histone modification and are closely associated with a substantial number of imprinted genes (Gehring et al., 2009; Wolff et al., 2011; Pignatta et al., 2014). DNA methylation of a TE located near a gene can be removed or reduced by DME in the central cell, leading to the specific expression of the maternal allele of that gene in the endosperm. TE insertions can also interfere with expression of imprinted genes. In maize, Mutator (Mu) insertions within the promoter of the Mez1 gene disrupted imprinted expression of this gene (Haun et al., 2009). Given that TEs are dynamic and polymorphic between different species, it is not surprising that imprinting of genes is poorly conserved between species, and many imprinted genes have experienced relaxed selection (Luo et al., 2011; Waters et al., 2011; Waters et al., 2013; Qiu et al., 2014; Hatorangan et al., 2016; Chen et al., 2018). Numerous imprinted genes in Arabidopsis originated from gene duplication events, and have experienced accelerated sequence evolution after duplication, with some potentially having undergone neofunctionalization (Spillane et al., 2007; Miyake et al., 2009; Qiu et al., 2014). A genome-scale comparison of imprinting between maize and rice identified only a small number of syntenic genes that exhibit conserved imprinting in both species, and these conserved imprinted genes showed a higher ratio of nonsynonymous to synonymous substitution, suggesting either positive selection in favor of new functions or relaxed selection (Waters et al., 2013; Tuteja et al., 2019). Ultimately, the extant data suggests that although imprinting can have important functions with respect to plant fitness, the primary and most often proximal role of imprinting appears to be related to tissue specific amelioration of the negative consequences of epigenetic silencing of TEs (Batista and Kohler, 2020).

Although there is strong evidence showing that the imprinting in plants is closely associated with demethylation of TEs (Suzuki et al., 2007; Pignatta et al., 2014), there is no evidence for the creation of a novel imprinted gene by a newly silenced transposon. Here, we used maize as a model system to study the epigenetic effects of TEs on parent-of-origin expression. Maize is an excellent model to investigate this phenomenon because there are many well characterized naturally active or silenced transposons in this species. The line we used here, which has been introgressed into the reference background B73 for at least eight generations, contains a large number of newly silenced Mu insertions (Zhang et al., 2020a). Mu elements preferentially insert into the 5’ end of genes, particularly 5’ untranslated regions (UTRs) and promoter regions (Liu et al., 2009; Zhang et al., 2020a), making them ideal models to determine the causes and the consequences of insertions on allele specific expression of genes. Through high throughput ribonucleic acid (RNA) sequencing (RNA-seq) of endosperm with segregating epigenetically silenced Mu insertions, we identified two potentially novel paternally suppressed maternally expressed imprinted genes (MEGs), demonstrating the ability of transposons to create novel imprinted genes.

Although imprinted long noncoding RNAs (ncRNAs) have been observed in rice and maize (Luo et al., 2011; Zhang et al., 2011), there is limited research on imprinting of expressed regulatory elements, such as conserved noncoding sequences (CNSs). CNSs are islands of noncoding sequence showing a low level of divergence between related species (Freeling and Subramaniam, 2009; Turco et al., 2013). Although the specific functions of CNSs remain unclear, many CNSs are likely to function as cis-regulatory elements involved in regulating gene transcription and chromatin structure (Kaplinsky et al., 2002; Vavouri et al., 2007; Van de Velde et al., 2016). Although most of them do not appear to be expressed, like many expressed genes, CNSs are often associated with CHH (H = A, T, or C) islands, regions of high CHH methylation near active genes, which mark the boundaries between relatively open and closed chromatin in the maize genome (Li et al., 2015). In this study, we discovered 233 maternally expressed CNSs (mat CNSs) and 174 paternally expressed CNSs (pat CNSs). Consistent with the observations of imprinted genes, imprinted CNSs have undergone rapid evolution relative to non-imprinted CNSs. In addition, DNA methylation is reduced but H3K27me3 is increased in the maternal alleles of the pat CNSs, and the increase in H3K27me3 in the maternal alleles is dramatically larger when these pat CNSs are in the flanking regions of paternally expressed imprinted genes (PEGs). Overall, our results provide insights into roles of both transposons and CNSs in genomic imprinting.

Results

Identification of imprinted genes in the hybrid endosperms

To investigate the potential effects of TEs on genomic imprinting, we reciprocally crossed a Mu silenced line in the B73 background with Mo17 (Figure 1, B73; Les28/-;Muk/- × Mo17 and Mo17 × B73;Les28/-;Muk/-). The Mu silenced line we used here was derived from a Mu active line crossed with the homozygous Muk/Muk, a dominant locus that can heritably silence Mu activity and that results in stably methylated Mu elements (Figure 1) (Slotkin et al., 2003, 2005). Both the Mu active line and Muk have been introgressed into B73 for at least eight generations. The Mu line has the advantage of carrying the Les28 Mu insertion allele (Hu et al., 1998; Zhang et al., 2020a), which only shows a dominant mutant phenotype (lesions in the leaves) in the presence of Mu activity, making it possible to monitor Mu activity in these plants. Because all new insertions are in the B73 background, all the Mu insertions are stabilized and hemizygous in the B73 genome. Given that previous studies identified a tendency for Mu insertions to target the 5’ ends of genes, particularly 5’ UTRs of genes (Liu et al., 2009; Zhang et al., 2020a), we looked for evidence of novel MEGs and PEGs due to differences in the male and female alleles caused by genetic and epigenetic differences introduced by the newly silenced Mu insertions (Figure 1B).

Figure 1.

Genetic strategy to develop F1 hybrid endosperms to detect imprinting introduced by Mu transposable elements. A and B, The reciprocal crosses. B73;Les28/- is a line containing many active Mu transposons. Mu killer (Muk) is a dominant locus that can heritably silence Mu activity.

RNA was isolated from four endosperms that were derived from a single reciprocal cross 14-day after pollination (DAP). Of these four endosperms, two sibling endosperms (11-5 and 11-6) are from the cross B73;Les28/-;Muk/- × Mo17, and the other two (11R-5 and 11R-6) are from the exactly reciprocal cross Mo17 × B73;Les28/-;Muk/-. Because all these endosperms were picked from two ears generated from a single reciprocal cross, they are all siblings of each other, and the only variables are the segregating Mu insertions and the direction of the cross. Deep sequencing of this RNA was performed to identify imprinted genes. A total of 246 million read pairs (61.6 million read pairs per sample) were generated to analyze allelic expression patterns of genes and CNSs. The allele specific expression values were assessed using the ∼9.5 million single nucleotide polymorphisms (SNPs) between B73 and Mo17 (Bukowski et al., 2018). Only transcripts that contained at least one SNP and 10 reads that could be assigned to either allele were retained for further analysis, resulting in 14,614–14,948 genes that were used for the four comparisons (Supplemental Figure S1A, 11-5 versus 11R-5, 11-5 versus 11R-6, 11-6 versus 11R-5, and 11-6 versus 11R-6). Of these SNP containing genes, we compared the observed ratio of female to male allele transcript levels with the expected 2m:1p (2 maternal:1 paternal) ratio. Using these parameters, we detected a total of 340 MEGs and 750 PEGs that exhibit parent-of-origin allele biased expression (Supplemental Figure S1A, P ≤ 0.001, χ² test). Following previously established cutoffs (Waters et al., 2013), a total of 139 nonredundant strong MEGs and 110 strong PEGs were defined in the four comparisons (Supplemental Figures S1A, S1B, and Supplemental Table S1). These samples (11-5, 11-6, 11R-5, and 11R-6) are in the same genetic backgrounds (hybrids of B73 and Mo17) except the segregated Mu insertions, and thus each pair of samples can be treated as replicates. Out of the 139 MEGs, 124 (89.2%) were detected as MEGs in at least two comparisons, and 77 (55.4%) were detected as MEGs in all four comparisons. Of the 110 PEGs, 95 (86.4%) were detected as PEGs in at least two comparisons, and 77 (70%) were detected as PEGs in all four comparisons (Supplemental Figure S1C; Supplemental Table S1). These imprinted genes were also compared with the publicly available data that were reanalyzed following the same pipeline as is used here (Waters et al., 2011; Dong et al., 2017). Out of the 139 MEGs and 110 PEGs, 64 (46%) and 91 (83%), respectively, were detected as imprinted genes in either of the two previous studies (Supplemental Figure S2). These data provide a measure of the replicability of our RNA-seq data and our subsequent imprinting analysis pipeline.

Identification of potentially novel MEGs and PEGs introduced by newly silenced Mu insertions

In order to investigate the effects of the newly silenced Mu transposons on the allele-specific expression of genes, DNA was isolated from the same endosperms to identify the Mu insertions using a Miseq-based Mu element profiling method (Zhang et al., 2020b). A total of 269 Mu insertions were identified in the four samples (Supplemental Figure S3; Supplemental Table S2). These insertions include not only the newly silenced Mu elements that we introduced, but also 23 deeply silenced Mu elements that are fixed in the B73 (10 unique insertions and 10 shared insertions with Mo17) and/or the Mo17 (three unique insertions and 10 shared insertions with B73) backgrounds (Supplemental Table S2). Of the 269 Mu insertions, 196 (73%) insertions are located in genic regions, including 1-kb upstream and downstream regions of genes, as well as in gene bodies. Consistent with previous results, the majority of the genic insertions are in 1-kb upstream regions and in 5′-UTRs (Supplemental Figure S3). We categorized the Mu insertions with the allele-specific expression of their flanking genes into seven different groups based on their effects on these genes (Figure 2). Next, we separated these genic insertions into four different scenarios to investigate the effects of Mu insertions on gene expression, as shown in Figure 2A. In all of these four scenarios, we compared instances in which insertions exist in both the female and male alleles. Therefore, we assume that they have the same genetic effects, and thus, that the effects are epigenetic. The instances in which insertions are absent were used as controls to determine the effects of the insertion on the same direction of crossing. As an additional negative control, we reanalyzed the publicly available imprinting data for genes without Mu insertions, following the same bioinformatics pipeline used for our data (Waters et al., 2011; Dong et al., 2017). Among the 196 genic insertions, 27 are in both the female B73 allele and its reciprocal male B73 allele but are absent in the sibling endosperms in both directions and are also absent in both the B73 and the Mo17 backgrounds (scenario a in Figure 2A). Because they include the controls for effects of the insertion on both directions of crossing within our sample set, these are the most informative insertions. Of them, the imprinting status of 11 (40.7%) were not determined due to the low or no allelic reads, and 14 (51.9%) insertions do not have dramatic effects on gene expression in either the female or the male allele, leaving 27 instances in which the Mu insertion appeared to have an effect on expression in at least one sample (Figure 2B). One of these 27 most informative Mu insertions clearly represses gene expression only in the male, resulting in a paternally suppressed MEG (Zm00001d016585, Figure 2C; Supplemental Figure S4A; Supplemental Table S3). Analysis of the additional available data sets provides additional evidence that in the absence of Mu insertions this gene is not imprinted. Twenty-one Mu insertions are present in all four of the endosperms (scenario b in Figure 2A). Of these, 12 are present in either the B73 or Mo17 reference genomes, and thus were excluded from further analysis (Figure 2B). We interpret the remaining nine as de novo insertions that are present by chance in all four endosperms. In this scenario, sibling endosperms can be treated as biological replicates, and publicly available data can serve as negative controls. We find that one Mu insertion strongly represses gene expression only in the male B73 allele. This results in a paternally suppressed MEG Zm00001d040629 (Figure 2C; Supplemental Figure S4B; Supplemental Table S3). As shown in Supplemental Table S3, when the Mu insertion is present in the female B73 allele of Zm00001d040629, the ratio of the allelic expression values of the female B73 allele to the male Mo17 allele in the two sibling endosperms are on average 2.66, which is similar to the expected 2m:1p ratio and is similar to previously published expression values. In contrast, when the Mu insertion is present in the male B73 allele of Zm00001d040629, the expression of the male B73 allele is completely repressed in both of the two sibling endosperms of the reciprocal cross (Figure 2C; Supplemental Figure S4B). In scenarios c and d, both directions of the reciprocal cross have Mu insertions in at least one endosperm. In both of the two scenarios, one direction lacks Mu insertions in either female or male, and thus can be used as controls for effects of the insertions in the same direction (Figure 2A). In both scenarios c and d, no potentially novel imprinted genes were identified. In summary, two potentially novel MEGs are associated with de novo Mu insertions. Importantly, neither of these two genes show evidence in imprinting in previous datasets (Waters et al., 2011; Dong et al., 2017).

Figure 2.

Potentially novel imprinted genes introduced by silenced Mu insertions. A, Different scenarios to identify potentially novel imprinted genes. In each scenario (a, b, c, and d), the top four boxes with black borders represent four endosperms sequenced in our study. Two top left black-bordered boxes are two siblings from the cross B73;Les28/-;Muk/- × Mo17, and two top right black-bordered boxes are two siblings from the reciprocal cross Mo17 × B73;Les28/-;Muk/-. The two bottom black-bordered boxes indicate the published imprinting data (left, B73 × Mo17, and right, Mo17 × B73) without Mu insertions being used as controls (Waters et al., 2011; Dong et al., 2017). B, Number of cases observed in different scenarios of (A). C, Two potentially novel MEGs Zm00001d016585 and Zm00001d040629 introduced by Mu insertions. Dong and Waters represent the publicly available data from Dong et al. (2017) and Waters et al. (2011).

Imprinted CNSs are largely localized within genic regions and near genes

Although the function of most CNSs is unknown, a number of them have been demonstrated to function as cis-regulatory elements involved in regulating transcription of genes (Kaplinsky et al., 2002; Vavouri et al., 2007; Van de Velde et al., 2016). Although imprinted ncRNAs have been observed (Zhang et al., 2011), the conservation status of expressed ncRNAs has not been assessed. To identify imprinted CNSs, we downloaded 432,936 previously identified CNSs between maize and sorghum (Song et al., 2021), compared the sequences of these CNSs in maize with the sorghum genome sequence, and kept 362,162 (83.7%) that we independently determined have substantial similarities between maize and sorghum. Next, we examined the allele specific expression of these 362,162 CNSs in our RNA-seq data sets. In maize endosperms, we find that 41,286 (11.4%) of CNSs are expressed higher than 1 fragments per kilobase of transcript per million mapped reads (FPKM), with the average expression value being largely lower than the average expression level of genes (Supplemental Figure S5). We used the same cutoff used for determination of the imprinted genes to identify imprinted CNSs. A total of 233 strong mat CNSs and 174 strong pat CNSs were identified in our comparisons (Supplemental Figure S6; Supplemental Table S4). We then compared the distribution of these mat and pat CNSs with that of expressed nonimprinted CNSs. Here, expressed nonimprinted CNSs are defined as those CNSs with at least 1 SNP and at least 10 allelic RNA-seq reads, which is the same cutoff used for the determination of the imprinted CNSs except for evidence of allele specific expression. This left 23,041–25,060 (6%–7%) across four comparisons (11-5 versus 11R-5, 11-5 versus 11R-6, 11-6 versus 11R-5, and 11-6 versus 11R-6) of the total CNSs as expressed nonimprinted CNSs. The genomic distribution of the imprinted CNSs was compared with that of genes and TEs. Consistent with the fact that they are expressed, both mat and pat CNSs are largely localized within genic regions and near genes. Of the 233 mat and 174 pat CNSs, 217 (93%) and 168 (97%) are within genes or within 2-kb upstream or downstream regions of genes (Figure 3A). We next examined the relationship between imprinted CNSs and their corresponding genes. Because transcripts of CNSs in UTRs (untranslated regions) of genes are hard to be distinguished with the transcripts of the co-localized genes (Supplemental Figure S7), we focused on imprinted CNSs in gene introns and within 2-kb upstream and downstream of genes, and excluded flanking genes for which imprinting status could not be determined due to a lack of sufficient polymorphism. Our data show that 24%–56% of these imprinted CNSs are located in or near imprinted genes (Figure 3, B–D). For example, out of the 81 mat CNSs in gene introns, only 22 (27%) are in MEGs (Figure 3B, top panel). In contrast, we observed a much higher proportion (56%, or 30 out of the 54) of pat CNSs in gene introns that are in PEGs (Figure 3B, bottom panel). This suggests that mat and pat CNSs in gene introns have distinct relationships with the genes with which they are associated.

Figure 3.

The imprinting status of the majority of the imprinted CNSs is independent of the imprinting status of their co-localized or flanking genes. A, Genomic distribution of the imprinted CNSs compared with nonimprinted CNSs. B, Imprinting status of the imprinted CNSs in gene introns and their corresponding genes. C, Imprinting status of the imprinted CNSs within 2-kb upstream of genes and their flanking genes. D, Imprinting status of the imprinted CNSs within 2-kb downstream of genes and their flanking genes. In (B), (C), and (D), the left pie charts show the percentages of the imprinted CNSs located in and not in introns of genes, and 2-kb upstream and downstream of genes. The right bar plots show the imprinting status of the genes harboring the imprinted CNSs. Exp nonimprinted CNSs, expressed nonimprinted CNSs. Exp nonimprinted genes, expressed nonimprinted genes.

Finally, we investigated the potential effects of TEs on these imprinted CNSs. Of the 233 mat CNSs, 103 (44.2%) are positioned within 2 kb of a TE, which is significantly lower than the percentage of the nonimprinted CNSs that are within 2 kb of TEs (Supplemental Figure S8A; P < 0.0001, χ² test). The most enriched TEs near the CNSs are DNA transposons, particularly Helitron elements (Supplemental Figure S8B). A similar pattern with respect to TEs was observed for the pat CNSs.

Imprinted CNSs and genes exhibit variation in length, exon number, expression level, and tissue specificity

Gene length, exon number, and expression level are primary features of genes. We examined the lengths of both imprinted CNSs and imprinted genes. Overall, imprinted CNSs are significantly longer than expressed nonimprinted CNSs (Figure 4A). Given that the imprinting status of the flanking genes of these imprinted CNSs are different, we separated the imprinted CNSs as those in or flanking MEGs or PEGs and those in or flanking non-imprinted genes. When mat CNSs were separated as those in or flanking MEGs and those in or flanking nonimprinted genes, neither exhibit significant differences in length from expressed nonimprinted CNSs. This is the same pattern for pat CNSs in or flanking nonimprinted genes. In contrast, pat CNSs in or flanking PEGs are significantly longer than expressed nonimprinted CNSs and mat CNSs (Supplemental Figure S9A).

Figure 4.

Imprinted CNSs and genes exhibit variation in the lengths, exon numbers, expression levels, and tissue specificities. A, CNS lengths. B, Complimentary DNA (cDNA) lengths of genes. C, Exon numbers of genes. D, Expression values of CNSs in the endosperms 14 days after pollination in this study. E, Expression values of genes in the endosperms 14 days after pollination in this study. F, Entropy across 24 maize tissues. Tissue specificity was determined by calculating Shannon entropy. Expressed nonimprinted genes and CNSs were following the same standard as the imprinted genes and CNSs with at least 1 SNP and 10 allelic reads. The numbers above the arrows indicate P-values between each comparison. The statistics analysis was conducted by Student’s t test. ^*P ≤ 0.05; ^**P ≤ 0.0005. P ≤ 0.0005 was considered as significant under the Bonferroni correction. The bottom and top boundaries of the boxes are the first and third quartiles, and the bold lines within individual boxes are the medians, which are referred to as the second quartiles. The ends of the whiskers (the dotted lines) represent the minimum values and maximum values of the data. Exp nonimprinted CNSs, expressed nonimprinted CNSs. Exp nonimprinted genes, expressed nonimprinted genes.

To compare MEGs and PEGs to expressed nonimprinted genes, we restricted our comparison to nonimprinted genes that, like our imprinted genes, have at least 1 SNP and at least 10 allelic RNA-seq reads. Interestingly, the average transcript of PEGs is significantly longer than expressed nonimprinted transcripts. This is not the case for MEG transcripts, the average length of which is significantly shorter than that of expressed nonimprinted genes (Figure 4B). PEGs also contain a significantly higher number of exons than both MEGs and nonimprinted genes. In contrast, MEGs harbor a significantly lower number of exons than expressed nonimprinted genes (Figure 4C).

To shed light on the function of these imprinted CNSs and genes, we next investigated their expression values in our RNA-seq data sets. On average, in 14 DAP endosperms, pat CNSs are expressed at a significantly lower level than expressed nonimprinted CNSs (Figure 4D). This is likely because of the lower expression of the subset of pat CNSs that are in or within 2-kb upstream or downstream of nonimprinted genes (Supplemental Figure S9B). MEGs are expressed at lower levels than both expressed nonimprinted genes and PEGs (Figure 4E). We also determined the specificity of expression of the imprinted genes compared with nonimprinted genes. We analyzed the tissue specificity of the 139 MEGs and 110 PEGs measured by Shannon entropy across 24 maize tissues using publicly available RNA-seq data (Sekhon et al., 2013; Eveland et al., 2014). Higher entropy values indicate constitutive expression, and lower levels of entropy reflect more tissue specificity (Schug et al., 2005; Makarevitch et al., 2013). The average entropy values of MEGs are significantly lower than those of nonimprinted genes and of expressed nonimprinted genes (Figure 4F). In contrast, PEGs have entropy values that are similar to expressed nonimprinted genes. These data suggest that MEGs are more tissue specifically expressed than expressed nonimprinted genes.

PEGs have been subject to relaxed selection and MEGs were already subject to a higher mutation rate prior to their being imprinted

As a rule, imprinted genes show a low degree of conservation, both with respect to whether they are imprinted in any given species, and with respect to purifying selection (Luo et al., 2011; Waters et al., 2013). We investigated the syntenic relationships and the conservation of imprinting of these imprinted genes with sorghum and rice. Of the 139 MEGs, 104 (74.8%) have syntenic homologs (syntelogs) in sorghum, which is significantly lower than the percentage (89.1%, 98 out of 110) of PEGs with syntelogs (Supplemental Table S5). There were several instances in which imprinting could be determined and that was retained in both of the two maize homoeologs that were generated by the most recent whole-genome duplication, roughly 12 million years ago (Swigonova et al., 2004). Of the 41 MEGs with maize homoeologs, only one pair of homoeologs were imprinted in the same way. The same was true of eight (four pairs) of the 44 PEGs with homoeologs (Supplemental Table S5). Finally, there was one pair of homoeologs that showed a reversal of imprinting, with one homoeolog being a MEG and one being a PEG (Supplemental Table S6). This pair of genes was previously described as maize homologs of Arabidopsis fertilization-independent endosperm1 (FIE1), a MEG in that species that is part of the PRC2 complex (Danilevskaya et al., 2003).

We next accessed the imprinting status of the sorghum syntelogs of imprinted maize genes using publicly available data in sorghum (Zhang et al., 2016). Of these syntelogs, 19 (18%) maize MEGs and 7 (7%) maize PEGs also show imprinting in sorghum. We performed a similar analysis of imprinted genes in rice. There are 83 maize MEGs and 74 maize PEGs that have syntelogs in rice whose imprinting status can be ascertained in that species (Yuan et al., 2017). Of these, only 2 maize MEGs and 17 maize PEGs show imprinting in rice as well. Overall, these data suggest that imprinting of specific genes has limited conservation both within polyploid genomes and between species.

To determine the frequency of mutation and the strength of selection on the imprinted genes, we compared the rates of sequence substitution of imprinted genes with those of nonimprinted genes measured by the nucleotide divergence between maize and sorghum. On average, the rate of nonsynonymous substitution (Ka) of both MEGs and PEGs is significantly higher than that of expressed nonimprinted genes (Figure 5A). The rate of synonymous substitution (Ks) of MEGs is significantly higher than that of expressed nonimprinted genes and PEGs, suggesting that the overall mutation rate for MEGs is higher than that for other genes. No significant difference was observed with respect to Ks between PEGs and expressed nonimprinted genes. In contrast, ω (Ka/Ks) is significantly higher for PEGs than for expressed nonimprinted genes, suggesting that PEGs have undergone weaker purifying selection than expressed nonimprinted genes (Figure 5A). To shed light on the cause of the higher Ks of MEGs and the relaxed selection on PEGs, we compared the sequence divergence of the sorghum syntelogs of MEGs and PEGs relative to rice with that of the sorghum syntelogs of nonimprinted genes relative to rice. Interestingly, Ks is significantly higher for the sorghum syntelogs of MEGs than for the syntelogs of other classes of genes even though the sorghum syntelogs are not imprinted (Supplemental Figure S10), suggesting that MEGs in maize were already subject to a higher mutation rate prior to their being imprinted. In contrast, sorghum syntelogs of PEGs do not show a significant difference in ω relative to sorghum syntelogs of other genes (Supplemental Figure S10), suggesting that imprinting may be the cause of the relaxed selection of PEGs in maize or at least that the relaxed selection of PEGs in maize occurred after their divergence from sorghum.

Figure 5.

PEGs have been subject to relaxed selection. A, Imprinted genes. B, Imprinted CNSs. Expressed nonimprinted genes and CNSs were following the same standard as the imprinted genes and CNSs with at least 1 SNP and 10 allelic reads. The evolutionary distances (Ka, Ks, and K) were calculated by pairwise comparison between maize and sorghum. The numbers above the arrows indicate P-values between each comparison. The statistical analysis was conducted using the Student’s t test. ^*P ≤ 0.05; ^**P ≤ 0.0005. P ≤ 0.0005 was considered as significant under the Bonferroni correction. Ka, nonsynonymous substitution; Ks, synonymous substitution; ω, Ka/Ks. The bottom and top boundaries of the boxes are the first and third quartiles, and the bold lines within individual boxes are the medians, which are referred to as the second quartiles. The ends of the whiskers (the dotted lines) represent the minimum values and maximum values of the data. Exp nonimprinted CNSs, expressed nonimprinted CNSs. Exp nonimprinted genes, expressed nonimprinted genes. MEGs, maternally expressed genes.

Next, we examined the nucleotide divergence (K) of the 233 mat and 174 pat CNSs when comparing maize and sorghum. As shown in Figure 5B, the sequence divergence of both mat and pat CNSs is significantly higher than those of nonimprinted CNSs, but no such differences were observed between the imprinted CNSs and expressed nonimprinted CNSs (Figure 5B). This suggests that the relaxed selection of imprinted CNSs is likely because they are expressed rather than because they are imprinted.

DNA methylation is lower and H3K27me3 is higher in the maternal alleles of pat CNSs

Genomic imprinting is an epigenetic phenomenon that involves both DNA cytosine methylation and H3K27me3 (Rodrigues and Zilberman, 2015). We wanted to determine whether these epigenetic marks are also associated with the imprinting of CNSs. In order to do this, we examined the allele-specific levels of DNA methylation and H3K27me3 of the imprinted genes and CNSs using the publicly available DNA and histone methylation data in F1 B73/Mo17 hybrid endosperms (Zhang et al., 2014), with SNPs being used to map bisulfite and ChIP reads to each of the two genomes. Consistent with previous results, the overall levels of CG, CHG, and CHH methylation are higher for PEGs than those for MEGs. In addition, the level of DNA methylation in all three cytosine contexts is higher in the paternal alleles than the maternal alleles of PEGs in the 2-kb upstream and downstream regions of the genes, as well as in gene bodies. In contrast, methylation levels of both the maternal and paternal alleles of MEGs are at similar levels (Supplemental Figure S11) (Zhang et al., 2014).

Overall, a similar pattern was observed for the imprinted CNSs. The methylation levels of CG, CHG, and CHH of the maternal and paternal alleles of expressed nonimprinted CNSs are at similar levels. In contrast, both CG and CHG methylation are lower in the maternal alleles than in the paternal alleles of both the mat and pat CNSs in both 5′-upstream and 3′-downstream regions and in CNS bodies (Figure 6, A and B). The difference between the two alleles is similar for both mat CNSs in or flanking MEGs and those in or flanking nonimprinted genes. Interestingly, the methylation differences between the two alleles in both 5′-upstream and 3′-downstream regions of pat CNSs that are in PEGs or within 2-kb flanking regions of PEGs are obviously larger than those pat CNSs that are in or flanking nonimprinted genes at the same regions (Figure 6A). No dramatic difference in CHH methylation was observed between the maternal and paternal alleles of either the mat or pat CNSs (Figure 6C). Overall, our data suggest that levels of CG and CHG methylation of the paternal and maternal alleles are distinct for both mat and pat CNSs.

Figure 6.

CG and CHG (H = A, T, or C) DNA methylation are lower in the maternal alleles of both the maternally and paternally expressed CNSs. A, CG methylation. B, CHG methylation. C, CHH methylation. Mat and pat CNSs were classified into two categories depending on the imprinting status of their co-localized (overlapping) or flanking (within 2-kb upstream and downstream) genes. CNS bodies were divided into 10 bins, and the 2-kb up and downstream regions were separated into 20 bins. Start, 5′-end of CNS; End, 3′-end of CNS. Exp nonimprinted CNSs, expressed nonimprinted CNSs.

To further explore the effects of epigenetic marks on imprinted CNSs, using previously published data sets, we examined allele-specific H3K27me3 enrichment (Zhang et al., 2014). As shown in Figure 7, overall, the H3K27me3 level is higher in the pat CNSs than that in the mat CNSs and in nonimprinted CNSs. The H3K27me3 enrichment of the maternal and paternal alleles of nonimprinted CNSs are at similar levels (Figure 7, A and B). This is also the case for mat CNSs that show no obvious difference with respect to H3K27me3 enrichment between the two alleles (Figure 7C). In contrast, H3K27me3 is enriched at the maternal alleles relative to the paternal alleles for pat CNSs in or flanking both PEGs and nonimprinted genes (Figure 7, A and D). Interestingly, the difference of H3K27me3 enrichment between the two alleles of pat CNSs is obviously larger when their co-localized or flanking genes are PEGs (Figure 7D). Together, these data show that DNA methylation is reduced, but H3K27me3 is increased at the maternal alleles of the pat CNSs, suggesting that DNA methylation and H3K27me3 are alternative repressive marks that may compensate for each other in the suppression of these imprinted CNSs.

Figure 7.

H3K27me3 is higher in the maternal alleles of the paternally expressed CNSs. A, H3K27me3 levels between two parental alleles of mat CNSs. The bottom and top boundaries of the boxes are the first and third quartiles, and the bold lines within individual boxes are the medians, which are referred to as the second quartiles. The ends of the whiskers (the dotted lines) represent the minimum values and maximum values of the data. B, H3K27me3 levels between two parental alleles of pat CNSs. C, H3K27me3 levels between two parental alleles of nonimprinted CNSs. D, Ratio of H3K27me3 levels of the maternal allele to paternal allele. The blue dashed line indicates the expected ratio of 2:1. Exp nonimprinted CNSs, expressed nonimprinted CNSs.CNSs)

Discussion

De novo creation of potentially novel imprinted genes by newly silenced Mu transposon insertions

Genomic imprinting involves two major epigenetic marks, DNA methylation and H3K27me3. The major sites of these epigenetic marks within imprinted genes are TEs, most of which are silenced in order to suppress their ability to replicate (Gehring et al., 2009; Wolff et al., 2011; Pignatta et al., 2014). Genome-wide analysis has demonstrated that maternally inherited TEs are extensively demethylated, and it has been proposed that imprinting of genes may be a by-product of this process (Gehring et al., 2009; Hsieh et al., 2009). Given this, we hypothesized that the introduction of silenced transposons near or within genes could cause the differential expression of the maternal and paternal alleles, and thus create novel imprinted genes in maize endosperm. The Mu transposons we used are an ideal model to test this hypothesis because (1) they are currently the most active transposons in maize, and this activity can be easily and reliably silenced using Muk (Slotkin et al., 2003, 2005); (2) they prefer to insert into the 5′-ends of genes and thus can potentially affect gene expression; and (3) they have been introgressed into the reference line B73, which allows relatively straightforward bioinformatics analysis of crosses between B73 and Mo17. By taking advantage of segregating silenced Mu insertions and the availability of additionally public data from B73/Mo17 hybrids that lacked Mu insertions, we detected two high confidence paternally suppressed MEGs that were introduced by the silenced Mu insertions (Figure 2). Although this is a limited number, this is an important result given that we were only able to examine the effects of a total of 37 Mu insertions for which the proper controls were available. It is worth noting that due to the absence of DNA methylation and H3K27me3 data of these specific F1 endosperms, the genetic and epigenetic effects of these Mu insertions on the allele-specific expression of the two potentially novel imprinted genes could not be distinguished. Here genetic effects refer to the physical knockdown or knockout of the gene expression by Mu insertions or instances in which Mu transposons provide regulatory sites that serve as repressors or enhancers (Lisch, 2013; Anderson and Springer, 2018). Epigenetic effects refer to the changes in expression due to the level of DNA methylation and/or histone modification introduced by the Mu insertions. These epigenetic effects could be caused by the Mu insertions per se or by regulatory sites that Mu elements introduce. However, in instances in which insertions exist in both the female and male alleles (Figure 2), we assume that they have the same genetic effects, and thus, that the effects are likely to be epigenetic. If methylation of Mu insertions in the female alleles is removed by DME in the central cell during female gametophyte development before fertilization (Choi et al., 2002; Gehring et al., 2009; Pignatta et al., 2014), this could potentially result in MEGs. Notably, if we remove the 24 Mu insertions whose effects on expression could not be determined, the majority (33, 89%) of the remaining 37 Mu insertions we identified do not substantially affect the expression of the inserted or flanking genes (Figure 2).

A substantial number of imprinted CNSs are expressed independently from their co-localized or flanking genes

In addition to the novel imprinted genes introduced by Mu insertions, we also detected 407 imprinted CNSs. The 432,936 CNSs we examined are noncoding sequences that are conserved between maize and sorghum, which diverged from a common ancestor roughly 12 million years ago (Swigonova et al., 2004; Song et al., 2021). Because this amount of time is not sufficient to result in a complete loss of sequence similarity in the absence of selection, evidence of conservation was combined with additional evidence of function, including chromatin accessibility and histone marks consistent with active chromatin (Song et al., 2021). The functions of most CNSs remain unknown, but several CNSs are known to function as cis-regulatory regions (Kaplinsky et al., 2002; Vavouri et al., 2007; Van de Velde et al., 2016). Our data show that 24%–56% of the imprinted CNSs are located within or near imprinted genes (Figure 3, B–D). This suggests that a substantial number of the imprinted CNSs are expressed independently of their associated genes. The function of this differential imprinting, if there is one, is unclear. In contrast to most imprinted CNSs, 56% pat CNSs in gene introns are associated with PEGs, suggesting either these pat CNSs exhibit cis-regulatory effects on the corresponding PEGs, or that they are simply regulated in the same way that is the gene in which they are located (Figure 3B).

Possible factors that shape the distinct patterns of MEGs and PEGs, and mat and pat CNSs

Gene length and exon number are important factors that shape the transcription and function of genes. Longer genes and gene with more exons are more likely to exhibit alternative splicing, leading to potentially new functions (Grishkevich and Yanai, 2014). In humans, longer genes tend to be associated with structures functioning early in life, such as blood vessels, cervix uteri, and brains, while smaller genes are more likely to be expressed through the organisms’ life, such as in the pancreas, skin, vagina, and testis (Lopes et al., 2021). This is consistent with what we observed. In plants, imprinting is primarily restricted to the endosperm, a tissue that functions early in life to supply nutrients and energy to the developing embryo. PEGs tend to be longer, and MEGs tend to be shorter than expressed nonimprinted genes (Figure 4B). In addition, PEGs have more exons, and MEGs have fewer exons than do expressed nonimprinted genes (Figure 4C). Overall, the expression values of most PEGs were higher than those of MEGs (Figure 4E). The kinship or parental conflict theory proposes that paternal expression of genes promotes nutrient acquisition in the offspring and maternal expression of genes restricts nutrient allocation to any one offspring. Certainly, this is true of some imprinted genes (Haig, 2013; Rodrigues and Zilberman, 2015). However, it is clear that most imprinted genes are not directly involved in the parental conflict. Many MEGs are likely to be genes whose methylation has been purged to ameliorate the negative effects of methylated TEs on gene expression. Alternatively, or in addition, the hypomethylation of TEs in the endosperm could be a similar phenomenon to hypomethylation of TEs observed in pollen vegetative cells, where it is a mechanism to permit TE expression in order to generate small RNAs that can be transmitted to the adjacent gametes in order to enforce their silencing (Slotkin et al., 2009; Rodrigues and Zilberman, 2015; Batista and Kohler, 2020). Imprinting of genes may be a byproduct of this process.

Many imprinted genes appear to evolve rapidly, with limited conservation among different species (Luo et al., 2011; Waters et al., 2013; Tuteja et al., 2019). The available evidence, both here and in previous publications, suggests that most imprinting is not well conserved and likely has more to do with amelioration of the effects of TEs than the specific function of any given gene. Indeed, our analysis of de novo imprinting due to Mu insertions suggests that imprinting of gene expression is relatively easy to introduce (and likely rapidly purged in natural populations). In this study, we also observed a significantly higher level of ω (Ka/Ks) for PEGs than for nonimprinted genes (Figure 5), suggesting these PEGs have undergone relaxed selection. This relaxed selection is likely caused by the imprinting status of the PEGs given that relaxed selection of the sorghum syntelogs of the PEGs is not observed (Supplemental Figure S10). However, we cannot rule out the possibility that these PEGs had been subject to relaxed selection after their split with sorghum syntelogs but prior to their being imprinted given that maize and sorghum have been diverged from each other for approximately 12 million years (Swigonova et al., 2004) and we do not know when these PEGs became imprinted. In contrast, the higher level of Ks of MEGs is independent of their imprinting status given that Ks is higher in both MEGs and their nonimprinted sorghum syntelogs (Supplemental Figure S10). This suggests that rather than imprinting causing an increase in Ks, genes with a higher Ks are more likely to be MEG imprinted.

Previous studies revealed that recombination facilitates the generation of single-nucleotide mutation (Lercher and Hurst, 2002; Gaut et al., 2007), and distinct chromatin types and duplication status are important determinants shaping the evolutionary divergence (Jordan et al., 2004; Yang and Gaut, 2011; Du et al., 2012; Zhao et al., 2017; Zhao et al., 2021). To determine the factors that may have shaped the variation of MEGs and PEGs in length, exon number, expression level, tissue specificity, rate of sequence substitution, and epigenetic status of the two alleles, we investigated the recombination rates, genomic locations (genes in chromosomal arms versus in pericentromeric regions), duplication status after the whole-genome duplication event (duplicates versus singletons), and numbers of genes with syntelogs in sorghum of these MEGs and PEGs (Gaut and Doebley, 1997; Liu et al., 2009). No significant differences with respect to the recombination rates, genomic locations, and duplication status were detected between MEGs and PEGs (Supplemental Table S5), suggesting that these three genomic features are not the primary determinants shaping the divergence of MEGs and PEGs. However, MEGs are more likely to be have syntelogs with sorghum than PEGs (Supplemental Table S5). In general, maize genes without syntenic homologs are more likely to be relatively dispensable with lower conservation and expression, which may in part explain why this is true of PEGs.

To adapt to the changes of imprinted genes, regulatory elements such as CNSs might be expected to evolve coordinately with the genes they regulated. The coevolution of imprinted CNSs and genes may affect the evolutionary trajectory of both. Indeed, we find that pat CNSs and PEGs both exhibit evidence of increased rates of evolutionary change, and pat CNSs are more likely to be associated with PEGs than mat CNSs are with MEGs. The mechanisms underlying the relationships between these imprinted CNSs and imprinted genes remain unclear. Future experiments to test the functional role of these imprinted CNSs and imprinted genes, particularly in seed development, are necessary.

H3K27me3 is likely spread from PEGs to pat CNSs or vice versa

DNA demethylation caused by DME glycosylase in the central cell of female gametophyte results in differential methylation of the maternal and paternal alleles of many genes in endosperm (Choi et al., 2002; Ibarra et al., 2012; Zhang et al., 2014; Dong et al., 2018). We observed lower levels of CG and CHG methylation of the maternal alleles at both mat and pat CNSs (Figure 6, A and B). However, we find a higher level of H3K27me3 enrichment at the maternal alleles only at the pat CNSs (Figure 7). These data are consistent with the hypothesis that DNA demethylation of the maternal alleles of the pat CNSs recruits targeting by the FIS PcG complex in order to confer the observed suppression (Weinhofer et al., 2010; Deleris et al., 2012). PRC2 activity is generally anti-associated with DNA methylation, and likely functions after DNA demethylation (Rodrigues and Zilberman, 2015; Batista and Kohler, 2020). This phenomenon has been observed at the maternal alleles of PEGs in both Arabidopsis and maize, suggesting that maternally repressed PEGs require both maternal-specific demethylation and H3K27me3 enrichment (Wolff et al., 2011; Zhang et al., 2014). Silencing of the maternal allele of a PEG gene (PHE1) in Arabidopsis, for instance, requires demethylation of the maternal allele and depends on the PRC2 complex (Kohler et al., 2005; Makarevich et al., 2008).

H3K27me3 spreading is mediated by PRC2 from the nucleation region to distal regions in the flowering locus C gene body through like heterochromatin protein 1-mediated looping (Lovkvist et al., 2021). H3K27me3 may spread to neighboring regions via its “write and read” mechanism and form large domains of H3K27me3 (Reinberg and Vales, 2018; Yu et al., 2019). Our data show that H3K27me3 is largely enriched at the maternal alleles relative to the paternal alleles for pat CNSs in or flanking PEGs (Figure 7D). We hypothesize that either H3K27me3 is spread from PEGs to their flanking pat CNSs or H3K27me3 is initiated from the CNSs and spreads into the PEGs. Further experimental evidence is required to test these hypotheses.

Although hundreds of imprinted genes have been discovered in multiple species, the origin and biological roles of the vast majority of imprinted genes remain unknown. We provide evidence that transposons can create potentially novel imprinted genes by insertions into the proximity of genes. In addition, we provide further insights into the understanding of imprinted CNSs and imprinted genes at the transcription, nucleotide divergence, and epigenetic levels.

Materials and methods

Genetic material construction and tissue collection

A maize (Z. mays) Mu active line (B73;Les28/-) was crossed with the homozygous Muk/Muk (B73;Muk/Muk) to obtain the Mu silenced line (B73;Les28/-;Muk/-) (Figure 1). Plants derived from this cross carried multiple, heterozygous segregating, silenced Mu insertions, as well as heterozygous insertions that were fixed in the B73 background. A plant from this Mu silenced line (B73;Les28/-;Muk/-) was reciprocally crossed with a Mo17 plant. Two endosperms from the ear of the cross B73;Les28/-;Muk/- × Mo17, and two endosperms from the ear of the exactly reciprocal cross Mo17 × B73;Les28/-;Muk/- were collected 14 DAP. Because all the Mu insertions, with the exception of those fixed in the B73 background, were segregating, each seed had a 50% chance of carrying a given insertion from its B73 parent. Total messenger RNA was extracted using the QIAGEN RNeasy Plant Mini Kit according to the manufacturer’s instructions. Library preparation and sequencing on Illumina platform HiSeq 4000 were performed by Beijing Genomics Institute. DNA was also extracted from the same endosperms to construct libraries to identify Mu insertions following previously described methods (Zhang et al., 2020b).

Identification of imprinted genes and CNSs

RNA-seq reads from four endosperm samples were mapped to the B73 reference v4 genome and the SNP replaced Mo17 genome sequences using Hisat2 (Kim et al., 2015). The SNP replaced Mo17 genome sequences were generated by taking the B73 reference sequences and replacing the nucleotides where a SNP was present between the two inbreds. All the SNPs used here are from the maize Hapmap3 project (Bukowski et al., 2018). Only reads with perfect and unique matches were retained and used for generating relative values for gene expression (FPKM). To access allele-specific expression values of each gene, the mapping results were incorporated with the SNPs from Hapmap3 using SAMtools (Li et al., 2009), and only SNPs detected in our RNA-seq data were used to determine allele-specific expression. Genes that contain at least 10 reads that could be assigned to a particular allele were used to perform χ² test (relative to the expected 2 maternal:1 paternal) (Supplemental Figure S1A). Genes exhibiting a significant difference (χ² test, P < 0.001) from the 2m:1p ratio in favor of the male or female allele in both B73 × Mo17 and Mo17 × B73 were retained. Strong MEGs and PEGs were defined with a higher stringency, requiring that at least 90% of RNA reads were derived from the maternal alleles (MEGs), or at least 70% of RNA reads from the paternal alleles (PEGs), in both directions of the reciprocal cross. In addition, the raw data from public research were downloaded and reanalyzed following the same bioinformatics pipeline as was used for analysis of our data (Waters et al., 2011; Dong et al., 2017).

The CNSs between maize and sorghum were obtained from published research (Supplemental Figure S6) (Song et al., 2021). We extracted the sequences of these CNSs in maize, and searched against the sorghum genome sequence using BLAST. CNSs with synteny and high similarity between maize and sorghum were kept for subsequent analyses. We used a similar pipeline as the identification of the imprinted genes to detect the imprinted expressed CNSs: (1) at least 10 allelic reads to a parental allele of the CNSs were used to perform the χ² test; (2) CNSs exhibiting a significant difference (χ² test, P < 0.05) from the 2m:1p ratio in both directions were retained; (3) at least 90% of RNA reads were derived from the maternal alleles, or at least 70% of RNA reads from the paternal alleles in both directions of the reciprocal cross were defined as strong mat and pat CNSs.

Identification of Mu insertions

We followed similar bioinformatics pipelines to identify Mu insertions in our samples (Williams-Carrier et al., 2010; Zhang et al., 2020b). The adapter (GAGATAATTGCCATTAT) and Mu end (TGGTCGACGGCCCGGGCTGCT, terminal of the Mu elements) sequences were trimmed from the raw reads with the cutadapt software using the default parameters except with a maximum error rate 0.15 (-e 0.15). Next, the trimmed reads were mapped to the B73 reference v4 genome by Bowtie2 (Langmead and Salzberg, 2012). The reads with partial Mu end sequences and partial flanking sequences of the Mu elements were used to define the insertion site, which was further confirmed by the target site duplication (TSD). Because when Mu element inserts into a new locus, it generates a 9-bp TSD that does not belong to the Mu end sequences. The presence of the TSD that is not present in the reference genomes provides strong evidence for a legitimate de novo insertion. The 9-bp TSD on both sides of the junction reads was extracted from the mapping file. The first base of the TSD coordinate was defined as the Mu insertion site.

Estimation of evolutionary distance

The analysis of evolutionary distance followed our previously published research (Zhao et al., 2015; Zhao et al., 2017). Homologous nucleotide sequences were aligned using the MUSCLE program (Edgar, 2004) or ClustalW (Thompson et al., 1994) using default parameters with manual inspection. For imprinted genes, pair-wise alignments between maize and sorghum, and between sorghum and rice were made to calculate Ka and Ks using the yn00 module under the PAML software (Yang, 2007). For imprinted CNSs, pairwise alignments between maize and sorghum were made to calculate nucleotide divergence K using the baseml module in the PAML software.

Analysis of DNA methylation and H3K27me3 ChIP-seq data

The raw reads of the publicly available DNA methylation and H3K27me3 in the F1 B73/Mo17 hybrid endosperms were downloaded (Zhang et al., 2014). Similar bioinformatics pipelines were used to access the allele-specific DNA methylation and H3K27me3 levels as described in (Zhang et al., 2014; Liu et al., 2021). The bisulfite sequencing reads were aligned to the B73 reference v4 genome using Bismark (Krueger and Andrews, 2011). Using the SNPs detected by Hapmap3, the methylation level of each allele was determined. Given that treatment of DNA with bisulfite converts cytosine residues to uracil, but leaves 5-methylcytosine residues unaffected, SNPs of C to T and G to A (B73 to Mo17) were excluded from the analysis when considering the B73 allele, and SNPs of T to C and A to G (B73 to Mo17) were excluded from the analysis when considering the Mo17 allele. Cytosine sites covered by at least five reads with at least one SNP were retained for downstream analysis. Fisher’s exact test was performed to assess the significance of the allele biased DNA methylation. The P-value generated by Fisher’s exact test was further corrected with the Q-value method (Storey and Tibshirani, 2003), and the FDR (false discovery rate) value <0.01 was considered as statistically significant.

ChIP-seq reads were mapped to the B73 reference v4 genome and the SNP replaced Mo17 genome by Bowtie2 (Langmead and Salzberg, 2012). Given the two copies of the maternal allele of maize endosperm, the abundance of the paternal allele was doubled when comparing the H3K27me3 levels between the two parental alleles.

Statistical analyses

Comparisons of lengths, exon numbers, expression values, entropies, and sequence divergences between imprinted and nonimprinted genes, and between imprinted and nonimprinted CNSs were conducted by Student’s t-test using the SAS software. The significance of different ratios between the paternal and maternal alleles was evaluated by χ² test or Fisher’s exact test. The Bonferroni correction was further performed to correct the P-values. P < 0.05 was considered as significant, and P < 0.0005 was considered as significant under the Bonferroni correction. All the information of the statistical tests and significance has been indicated in the legends of the figures.

Accession numbers

The RNA-seq data has been deposited in the National Center for Biotechnology Information Sequence Read Archive under associate code PRJNA556108 and accession number GSE146647 (Zhang et al., 2020a). Sequence data from this article can be found in MaizeGDB, https://www.maizegdb.org/.

Supplemental data

Supplemental Figure S1. Identification of imprinted genes.

Supplemental Figure S2. Imprinted genes identified in our data set largely overlap with published data.

Supplemental Figure S3. Genomic distribution of the Mu insertions.

Supplemental Figure S4. Read distribution of the two potentially novel imprinted genes introduced by silenced Mu insertions.

Supplemental Figure S5. Expression levels of genes and CNSs.

Supplemental Figure S6. Identification of imprinted conserved CNSs.

Supplemental Figure S7. pat CNSs exhibit the same imprinted status with their corresponding genes in UTRs.

Supplemental Figure S8. Distribution of transposons in the proximity of CNSs.

Supplemental Figure S9. Comparisons of lengths and expression values of imprinted CNSs in or flanking imprinted and non-imprinted genes.

Supplemental Figure S10. Evolutionary distance of sorghum syntelogs relative to rice.

Supplemental Figure S11. DNA methylation is lower in the maternal alleles of the PEGs.

Supplemental Table S1. Detailed list of the imprinted genes.

Supplemental Table S2. Summary of the Mu insertions identified in this study.

Supplemental Table S3. Detailed list of genes with both female and male B73 alleles in the reciprocal cross containing de novo Mu insertions.

Supplemental Table S4. Detailed list of the imprinted CNSs.

Supplemental Table S5. Comparisons of genomic features between MEGs and PEGs.

Supplemental Table S6. List of the maize duplicated genes with both copies imprinted.

 

M.Z. and D.L. designed the research. M.Z. prepared the research materials for transcriptomic and Mutator insertion sequencing. T.L., L.Y., C.E.S., and M.Z. analyzed the data. T.L. and M.Z. wrote the original draft, and D.L. reviewed and edited the manuscript.

The authors responsible for distribution of materials integral to the findings presented in this article in accordance with the policy described in the Instructions for Authors (https://academic.oup.com/plphys/pages/general-instructions) are: Damon Lisch (dlisch@purdue.edu) and Meixia Zhao (meixiazhao@ufl.edu).