Dynamic patterns of the translatome in a hybrid triplet show translational fractionation of the maize subgenomes

Wncho Zhu, Siji Chen, Tifu Zhng, Ji Qin, Zi Luo, Hn Zho, Yirong Zhng, Lin Li,*

a National Key Laboratory of Crop Genetic Improvement, Huazhong Agricultural University, Wuhan 430070, Hubei, China

b Jiangsu Provincial Key Laboratory of Agrobiology, Institute of Germplasm Resources and Biotechnology, Jiangsu Academy of Agricultural Sciences, Nanjing 210014, Jiangsu, China

c Key Laboratory of Crop Heterosis and Utilization (MOE) and State Key Laboratory for Agrobiotechnology, Key Laboratory of Crop Genomics and Genetic Improvement (MOA),Beijing Key Laboratory of Crop Genetic Improvement, China Agricultural University, Beijing 100193, China

Keywords:Heterosis Translatome Ribosome-sequencing (Ribo-seq)Subgenomes Maize

ABSTRACT Heterosis, the phenomenon in which hybrids outperform their parents, has been utilized in maize (Zea mays L.)for over 100 years.To provide a more complete understanding of heterosis,we collected a comprehensive transcriptome and translatome dataset on seedling leaves for B73,Mo17,and their F1 hybrid,which provided a dynamic landscape of transcriptomic and translatomic variation in maize.Although additivity accounted for a large proportion of variation at two omics-levels, an elevated nonadditive effect was observed in the translatome, especially in the translated subgenome maize1 genes, and the genes that switched from additivity in the transcriptome to nonadditivity in the translatome were significantly enriched in the subgenome maize1.Many genes with allele-specific expression and translation show dramatic regulatory switches between the transcriptome and translatome, and partial genes with allele-specific translation underlying regulatory mechanism also exhibited subgenome bias.Interestingly,we found the translated isoforms show different expression patterns compared with transcriptome and more genes changed their dominant isoforms during the genetic flow from parents to the hybrid at the translatome level.The translated genes with switched dominant isoforms significantly biased to the subgenome maize2 while genes with conserved dominant isoforms significantly enriched in subgenome maize1.Together, the dynamic changed patterns in translatome across hybrid and parental lines show translational fractionation of the maize subgenomes, which may be associated with heterosis in maize and provides a potential theoretical basis for breeding.

1.Introduction

Heterosis, also known as hybrid vigor, refers to a phenomenon in which the phenotypic performance of the F1offspring significantly deviates from that of the parents [1-3].A substantial number of studies have been conducted to understand the genetic basis of heterosis[4-9],and many genetic models-dominance,overdominance, epistasis, and dosage effect etc.have been proposed to explain heterosis[10-15].Although these models are not mutually exclusive,no one model can completely explain heterosis,probably due to limitations in detection power [16].With the advent of next-generation sequencing and mass spectrometry (MS), a substantial number of‘‘omic”studies have been performed to investigate the molecular mechanisms of heterosis [4,17-27].Some important loci associated with heterosis have been identified and many genes show complementary expression patterns in the hybrid [16,23,28,29].All these endeavors contribute to our understanding on heterosis.

The transcriptome is a vital link between DNA sequence and phenotype[30],while the proteome translated by RNAs is directly linked to phenotype.However, proteomic variation is not usually in concert with transcriptomic variation because the latter can be buffered by the multifaceted regulatory mechanisms at the protein level[31-34].Thus,monitoring protein levels in hybrids is vital for understanding heterosis.It is worth noting that current proteomic techniques (MS) usually detect no more than 10,000 highly expressed proteins [34,35].Given the large number of expressed transcripts, other state-of-the-art high-resolution techniques are required for the detection and accurate quantification of the proteome.A new method called ribosome profiling (Ribo-seq) has been developed,which is based on deep sequencing of mRNA fragments wrapped in ribosomes and makes it possible to quantify the abundance of whole genome proteins [36].The development of ribosome profiling provides us an unprecedented chance to monitor the constant proteomic variation underlying heterosis in maize.

The maize genome was derived from two ancient allogenomes that were subjected to dramatic genomic breaks and fusion before evolving into the current diploid genome [37].The maize genome can be classified into syntenic and non-syntenic regions based on their syntenic relationships to those of ancient genomes.The syntenic regions were further grouped into two distinct subgenomes:maize1 and maize2 [38].The two subgenomes have undergone fractionation in terms of both gene content and expression [38].Subgenome maize1 is thought to be the dominant subgenome, as it has retained more genes that are highly expressed [39].While non-syntenic genes have been shown to contribute to heterosis at the transcriptome level [23], the dynamic translatomic landscape of the two subgenomes and non-syntenic genes in an F1hybrid and its parental lines and its potential roles in heterosis at the translatome (proteome) level remain largely unknown.

Here, we generated high-quality ribosome profiling data with two biological replicates for the leaves of two parental lines (B73 and Mo17)and their F1at the V4 plant stage.Combining these data with analyses of transcriptomic data,we observed that the effect of nonadditivity was elevated during translation, and a significant subgenome bias was shown in switched genes from additivity(transcriptome)to nonadditivity(translatome).A dramatic regulatory switch occurred in the process from transcriptome to translatome, and genes with allele-specific translation underlying regulatory mechanism exhibited subgenome bias.The translated genes were likely to vary their dominant isoforms from the parents to hybrid, and the genes with switched and conserved dominant isoforms across the parents and their F1also exhibit significant subgenome bias.Furthermore, protein MS verified that the genes corresponding to detectable proteins show similar patterns to translatome analyses in some aspects.Overall, this study shows translational fractionation of the maize subgenomes in dynamic changed patterns between transcriptome and translatome across the hybrid triplet,which is likely as one potential mechanism associated to heterosis at the translatome level.

2.Material and methods

2.1.Plant materials

Seeds of maize(Zea maysL.)inbred lines B73,Mo17,and their F1(Mo17 × B73) were planted in a greenhouse under conditions of 30 °C for 16 h of light and 25 °C for 8 h of darkness.The collected V4-stage(the period when the fourth leaf is fully expanded)leaves of the three genotypes were then used for total RNA extraction,ribosome profiling, and sequencing.Samples for two biologically replicates of three genotypes (independent samplings) were used for the further next generation sequencing at the transcriptome and translatome levels.

2.2.RNA-seq and Ribo-seq preparation

RNA was extracted from frozen tissue by using TRIzol reagent(Invitrogen) according to the manufacturer’s instructions.RNA-seq libraries were prepared using an Illumina Standard mRNA-seq library preparation kit according to the manufacturer’s protocol.RNA-seq libraries were sequenced by paired-end 150 on an Illumina HiSeq X Ten platform.The maize polysome complexes were extracted from about 5 g of ground leaf tissue using PE buffer(44 mmol L-1Tris-HCl, pH 7.5, 175 mmol L-1KCl, 13 mmol L-1MgCl2, 100 μg mL-1cycloheximide, 15 mmol L-12-mercaptoethanol, 1% Triton X-100, 10 units mL-1DNase I).Then,RNase I (10 units μg-1RNA; New England Biolabs) was added to these polysome pellets to digest for 1 h at room temperature.The resultant monosomes were enriched by a MicroSpinS-400 column.Ribosome-protected fragments were extracted based on the instructions for the miRNeasy RNA isolation kit (Qiagen cat.no.217004, http://www.qiagen.com/) with some modifications.After purification, RNA samples were used to construct libraries according to the procedure of construction of miRNA library.Successfully constructed libraries were sequenced on an Illumina HiSeq X Ten platform.

2.3.Analysis of raw sequencing data

After removing adaptors and performing quality controls, the rRNA reference sequence was downloaded from NCBI by searching with the keywords ‘ribosomal DNA’.Ribo-seq reads were then aligned against the rRNA reference sequence by Bowtie (version 1.1.2) to filter out reads derived from rRNAs [40].To remove the effect of sequence length, we randomly truncated original mRNA sequence reads of ~300 nt into short reads with similar readlength distribution(ranging from 18 to 35 nt)to the Ribo-seq data.There is high density of variants between B73 and Mo17 [41,42].The variant calling has been performed by Zhou et al and the results of final variant has been produced [27].The mRNA-seq and Ribo-seq reads were mapped to the B73 reference genome(AGPv4) using STAR [43] with default parameters, and the Mo17 variants(SNPs are the major)were considered to remove mapping bias.

2.4.Calculation of transcriptional and translational abundance, and feature characterization for translatome data

Transcriptional abundance was measured by fragments per kilobase of exon model per million mapped reads (FPKM) by Cufflinks v2.2.1 [44] with parameters -p 6 -u -G.We only counted mRNA-seq reads located in exons to calculate transcriptional abundance.Translation abundance of each gene was calculated by Cufflinks v2.2.1 with parameters -p 6 -u -G.We counted Ribo-seq reads that were located in CDSs to calculate translational abundance.We defined the standard for gene expression as FPKM ≥1.Translational efficiency was calculated as FPKM (translational level)/FPKM (transcriptional level) for genes with the FPKM ≥1 at both the transcriptome and translatome levels.

We obtained the uniquely mapped reads based on the tag(NH:i:1)in the bam file from the Ribo-seq data.The 3-nt periodicity,Asite position calibration and meta-gene read depth were analyzed according to a previous method [45].

2.5.Identification of differentially expressed genes (DEGs) and singleparent expression (SPE) genes

We used the htseq-count and edgeR to identify DEGs (Pvalue < 0.05) [46,47].According to the log2fold change of the expression abundance between the two parents, DEGs between B73 and Mo17 were classified as DE2+ (|log2FPKMB73/FPKMMo17|≥1, equal and more than 2-fold difference between the two parents),DE2-4(1≤|log2FPKMB73/FPKMMo17|<2,equal and more than 2-fold difference but<4-fold difference between the two parents),DE4-8 (2≤|log2FPKMB73/FPKMMo17|<3, equal and more than 2-fold difference but<8-fold difference between the two parents),and DE8+ (|log2FPKMB73/FPKMMo17| ≥3, equal and more than 8-fold difference between the two parents).Genes with singleparent expression (SPE) were defined as follows:FPKMB73≥1 and FPKMMo17< 0.1, FPKMMo17≥1 and FPKMB73< 0.1.

2.6.Classification of gene expression pattern

To classify the pattern of gene expression related to heterosis,we identified 14 categories of gene expression variation across the three genotypes (Table S1).All DEGs with significant differences in the expression levels of the three genotypes were classified into additivity and nonadditivity.When the difference between the expression level of the F1and the mid-parent value was not significant, expression was defined as mid-parent (MP,additive).The remaining DEGs were nonadditivity genes that and were further classified as below low-parent (BLP), low-parent(LP), partially dominant (PD), high-parent (HP), and above highparent (AHP).

Moreover, the dominance/additivity (D/A) value between the hybrid and mid-parent expression levels were calculated for each differentially expressed gene between B73 and Mo17 as follows:

2.7.Characterization of allele-specific expression (ASE) patterns for transcriptome and translatome

To estimate allele-specific gene expression between the three genotypes,we used STAR[43]to map F1reads to the B73(AGPv4)reference genome that modified using the Mo17 variant.According to the different single-nucleotide polymorphism (SNP) position information existing in B73 and Mo17, the read in the hybrid can be accurately assigned to one of the alleles from the two parents.Since a read is likely to span multiple SNPs, it was possible to obtain ‘‘contradictory reads” that mapped on both the B73 allele and the Mo17 allele.We selected genes with at least 30 allelespecific reads and less than 10% contradictory reads to infer the ratio of expression of each allele in the heterozygote [27].

To classify gene expression patterns into different regulatory models, we defined the following:

Cis-only:PBP = FBP

Cis+transopposite direction:FBPPBP& FBP≤0.5

Cis+transsame direction:FBP>PBP&PBP≥0.5 or FBP

Trans-only:FBP = 0.5

Unexpected bias:FBP > PBP & PBP < 0.5 & FBP greater than 0.5 or FBP < PBP & FBP < 0.5 & PBP greater than 0.5

2.8.Characterization of alternative splicing (AS) and alternative translation (AT) events

To classify splicing events at the transcriptome and translatome levels,we first imitated the short sequence data structure of Riboseq by randomly truncating the RNA-seq reads into short reads to avoid the influence of sequence length.Then, we classified the splicing events with the ASTALAVISTA tool[48]based on the transcript model and splice junctions.The splicing events were divided into five categories:intron retention(IR),exon skipping(ES),alternative acceptor (AA), alternative donor (AD), and other complex patterns (OP) [49].

2.9.Calculation of Shannon entropy of gene expression and definition of dominant isoform

To analyze the specificity of transcript expression for each gene locus, we selected the expressed genes and transcripts (FPKM≥1).For each expressed gene, we defined the following [50]:

The proportion of expression of transcriptiout of the sum of all expressed transcripts:

The genes with Shannon entropy value(SEVs)lower than 15%of the whole gene set were considered as the ones with a dominant expressed isoform.For each gene considered to have a dominant isoform, the isoform with the highest expression was considered the dominant isoform for that gene.

2.10.Protein preparation and iTRAQ-labeling Mass Spectrometry

The process of Mass Spectrometry was in the Method S1.

3.Results

3.1.High-quality translatome profiling provides an accurate translation landscape of maize inbred lines B73, Mo17, and their F1 hybrid

Young leaf samples of the V4 developmental stage from two biological replicates of each genotype were collected and at least six different individuals per replicate were bulked for standard ribosome separation followed by Ribo-seq, transcriptome profiling by RNA-seq, and proteome profiling by Mass Spectrometry(Fig.1A).A total of 105.9 million, 115.8 million and 93.9 million reads of ribosome profiling were obtained for B73, Mo17 and their F1, respectively (Tables S2, S3).The translatome profiling in a classical hybrid triplet enables us to survey the translatomic dynamics at a high-resolution scale in maize (Fig.1B).The length of ribosome-imprinted RNA fragments (RPFs) ranged from 24 to 32 nt for all three genotypes with 28 nt being the dominant size(Fig.1C).After stringent sequence quality filtering and mapping,~68% of the Ribo-seq reads were mapped to genic or flanking regions.Moreover, most (greater than80%) of the sequencing reads for all three genotypes mapped to the coding sequence(CDS) region, while only a small proportion of reads (~17%) were aligned to UTRs (Figs.1D, S2).During translation, the ribosome recognizes three nucleotides for each translational cycle, a feature referred to as three-nucleotide periodicity [51].We scanned the first 50 nt from the CDS start site and the last 40 nt before the stop codon and observed an obvious three-nucleotide periodicity in all three genotypes (Figs.1E, S1 and S2).A metagene analysis,in which many gene profiles are aligned and then averaged, was used to reveal global differences in ribosome density surrounding the start and stop codon sites, and the peaks were shown(Figs.1F, S2).The translatome data demonstrated similar features to previous studies in different organisms and matched the notion that translation initiation and termination may restrict translation rate [45,52].Additionally, the translatome datasets with high correlations (r= 0.86-0.87, Fig.S3A) between the two biological replicates of each genotype correlated to the transcriptome (r= 0.76-0.83, Fig.S3B).As expected, the translatome exhibited better consistency with the proteome than transcriptome (Fig.S3C, D).All these results indicate the high quality of the translatome dataset collected in this study, which thus provides an unprecedented opportunity to survey the potential molecular mechanisms associated with heterosis at a highresolution translatomic scale in maize.

Fig.1.High-quality translatome profiling provides a dynamic translatomic landscape across B73,Mo17,and their F1 hybrid.(A)V4-stage seedlings of a classical maize hybrid triplet and the multi-omics experimental design based on the leaf samples from the three genotypes.MS, Mass Spectrometry; Ribo-seq, ribosome profiling).(B) Flowchart summarizing the analyses conducted in our study.DE,differential expression;SPE,single-parent expression.(C)Length distribution of ribosome-imprinted fragments(RPFs).(D)Percentages of RPF reads located in coding sequence(CDS),5′ UTR,and 3′ UTR in B73(blue),F1(red),and Mo17(green)seedlings.(E)Three-nucleotide periodicity within the first 50 nt and last 40 nt of CDSs of B73.Red, blue, and green represent the first, second, and third nucleotide of the codon, respectively.(F) Meta-gene analysis of RPF density along CDS start and stop codon regions.

3.2.The subgenome maize1 genes are more detectable across B73,Mo17, and their hybrid in translatome

To fully profile the dynamic landscape of genetic flow from DNA to proteome, the corresponding transcriptome data with two biological replicates of the three genotypes were used as the comparison.To remove the effect of sequence length, we randomly truncated original mRNA sequence reads of ~ 300 nt into short reads with similar read-length distribution (ranging from 18 to 35 nt) to the Ribo-seq data.At the transcriptomic level, a total of 37,409 isoforms from 19,381 genes, 40,431 isoforms from 20,996 genes, and 36,947 isoforms from 19,836 genes were identified in B73, F1, and Mo17, respectively (Fig.2A).At the translatome level,25,063 isoforms from 17,820 genes, 25,574 isoforms from 17,962 genes, and 25,136 isoforms from 17,836 genes were detected in B73,F1, and Mo17,respectively(Fig.2A).As expected,the number of transcripts expressed in the transcriptome was generally higher than that in the translatome (Fig.2A), and the simultaneously expressed genes (17,722 with FPKM ≥1) in all three genotypes at the transcriptome level also outnumbered that at the translatome level(15,637 with FPKM ≥1)(Fig.S4).This is in accordance with previous study [53].The hybrid expressed more genes and more isoforms in the transcriptome than either parent (Fig.2A),which may be caused by SPE and indicates a greater diversity of gene expression in the hybrid.However, a similar number of isoforms from a similar number of genes were translated across F1,B73,and Mo17(Fig.2A,B).Furthermore,we identified many genes that were transcribed but not translated in all three genotypes,especially in the F1(Fig.2A,B).And the remaining translated genes were significantly enriched in subgenome maize1 and significantly depleted among the non-syntenic genes,i.e.,those that do not have synteny with sorghum(Fig.2C)[38,39].Additionally, more subgenome maize1 genes (~60%) were likely to be translated at higher levels than maize2 genes in all three genotypes (Fig.2D), consistent with previous findings [54].All these results suggest that genes in maize1 subgenome are likely to be highly translated and more detectable.

Fig.2.Genes with detectable transcription and translation show asymmetric subgenome enrichment.(A) Number of transcribed and translated isoforms and genes across B73,F1,and Mo17.Error bar shows the standard error.(B)Average number of isoforms per gene that were transcribed and translated across B73,F1,and Mo17.(C)Proportion of untranslated genes and translated genes in F1 that are subgenome maize1,subgenome maize2,and non-syntenic genes.The significance of enrichment was tested by Chisquare tests.Control groups were defined as corresponding genes in the gene sets selected 1000 times randomly from expressed genes in the transcriptome with equal gene numbers to the test group.**, P <0.01; ***,P <0.001.(D) The distribution of the ratios between expression levels of subgenomes maize1 and maize2 genes.The ratios were calculated using maize1-maize2 gene pairs along the ancient sorghum(Sorghum bicolor)chromosomes,and the fold change ≥2 represents the significant difference between maize1 and maize2.

3.3.Elevated nonadditive effect in translatome and significant subgenome bias in switched genes from additivity (transcriptome) to nonadditivity (translatome)

To explore the gene expression modes in the hybrid and its parental lines at both the transcriptome and translatome levels, we identified DEGs including SPEs in the transcriptome (4198) and translatome (1462) datasets, which showed dramatic expression variation across all three genotypes(Fig.3A).In both comparisons,<40% of the DEGs showed nonadditive expression in the hybrids,but the proportion of nonadditivity was higher in the translatome than in the transcriptome (36.4% vs.23.4%) (Fig.3B).This result suggests an elevated nonadditivity effect in translatome and indicates that nonadditive effects might play important roles in heterosis.To further quantify the changes in gene-action modes between the transcriptome and translatome, the distribution of D/A (dominance/additivity) for the DEGs between two parents was plotted(Fig.S5A, B).This analysis suggested that the D/A distribution in the translatome was significantly different (Student’st-test,Pvalue = 4.77E-61)from that in the transcriptome and also showed an elevated nonadditivity effect in translatome.Moreover, we identified the subgenome maize1, subgenome maize2 and nonsyntenic genes in theses DEGs and the D/A distribution of three types of genes were compared.The translatome showed higher elevated nonadditivity effect, especially in the translated maize1 genes (Fig.3C, D).To examine these differences in detail, all of the DEGs were further classified as above-high-parent genes(AHP), high-parent genes (HP), mid-parent genes (MP), partially dominant genes (PD), low-parent genes (LP), and below-lowparent genes (BLP) on the basis of their expression in the F1(Table S1) [19,22,27].Genes with the MP pattern, i.e., showing an additive effect,accounted for the majority of genes across different expression-level variation scales in the transcriptome (Fig.3E).Genes with HP and LP patterns,indicative of dominance,accounted for most of the remainder (Fig.3E).In the translatome, although additive genes (MP genes) still represented the largest proportion of genes, the proportions of LP genes were significantly (Student’st-test,P-value=8.5E-03)increased,suggestive of an increased nonadditivity effect contributing to heterosis at the translatome level(Fig.3F).During the genetic flow from transcriptome to translatome, there were a substantial number of genes (225) with switched modes (From additivity to nonadditivity and vice versa,Fig.3G).More importantly, genes that switched from additivity in the transcriptome to nonadditivity in the translatome were significantly enriched in subgenome maize1 (Fig.3H).These results showed that the maize1 genes play a dominant role in elevated nonadditivity and may be associated with the formation of heterosis at the translatome level.

Fig.3.Dynamic transcriptomic and translatomic patterns across B73, maize, and their F1 hybrid.(A) Expression-level variation of DEGs in the B73, F1, and Mo17 at the transcriptome (left) and translatome (right) levels.(B) Proportions of DEG expression patterns (additive and nonadditive) at the transcriptome (left) and translatome (right)levels.Distributions of dominance/additivity [D/A] values at the transcriptome (C) and translatome levels (D).For each DEG in the parents, the [D/A] value was calculated as(F1 - MP)/(HP - MP), where F1, MP, and HP indicate expression of the F1, mid-parent (average of high and low), and high parent, respectively.[D/A] value of - 1 and 1 indicate the same level of expression as the low parent and high parent, while a value of 0 indicates mid-parent expression.The [D/A] values are shown for all differentially expressed genes including subgenome maize1 genes, subgenome maize2 genes and non-syntenic genes.*, P < 0.05.(E) and (F).Proportions of DEGs and SPE genes showing different expression patterns in the F1 (BLP, below low-parent; LP, low-parent; PD, partially dominant; MP, mid-parent; HP, high-parent; AHP, above high-parent) at the transcriptome(E) and translatome (F) levels.(G) Number of genes with expression pattern changes from the transcriptome to translatome.(H) Subgenome enrichment for genes with expression patterns that switched from additivity in the transcriptome to non-additivity in the translatome.The significance of enrichment was tested by Chi-square tests.Control groups were defined as corresponding genes in the gene sets selected 1000 times randomly from expressed genes in the additive plus non-additive gene sets in the translatome with equal gene numbers to the test group.*, P < 0.05.

To verify the accuracy of above analyses based on the translatome, we also performed an MS experiment with the mixture of three biological replicates of V4-stage leaves from the three genotypes.Almost all of the proteins were represented in our Ribo-seq dataset and showed a relatively high correlation with those obtained by ribosome profiling (Fig.S3C).Compared with the two parents,the DEGs represented in the F1proteome showed a divergent expression pattern,as was also seen in the translatome(Figs.S6A, 3A-right).Among the genes detected in the proteomes of all three genotypes,the subgenome maize1 genes showed a significant enrichment (Fig.S6B), which is also consistent with the translatome in Fig.2C.As found in the translatome datasets, dramatic gene-action mode switch from transcriptome to proteome were also observed(Fig.S6C).Moreover,genes that switched from additivity in the transcriptome to nonadditivity in the proteome were likely to be enriched in subgenome maize1 (P= 0.051)(Fig.S6D).Taken together, all the proteome profiling results confirm the similar patterns in the translatome, and suggest that the translated genes have undergone an extensive translation selection during the formation of heterosis.

3.4.Dramatic regulatory switches between the transcriptome and the translatome

To dissect the regulatory mechanism associated with the dramatic expression variation observed in the classical hybrid triplet,we focused on the identification of ASE genes.In total 8255 genes and 8167 genes were identified as ASE genes in the transcriptome and translatome,respectively.The ASE genes in both the transcriptome and the translatome displayed asymmetric parental ratios,similar to the results of a previous study[27].Based on the values of allelic expression ratios in hybrid with parental proportions,the regulatory mechanisms of the ASE genes that overlapped with the DEGs could be classified as ‘‘cis-only”, ‘‘trans-only”, ‘‘cis + trans:opposite direction”, ‘‘cis + trans:same direction” and ‘‘unexpected bias” in both transcriptome and translatome (Fig.4A, B).In the transcriptome,the‘‘cis+trans”genes accounted for the largest proportion while‘‘cis-only”genes accounted for the second proportion and‘‘trans-only”genes were the least prevalent(Fig.4C).However,the proportion of ‘‘trans-only” genes increased dramatically(greater than3-fold) in the translatome, while that of ‘‘cis + trans opp.dir.” genes dropped markedly and ‘‘cis-only” kept conserved(Fig.4C), suggestive of the importance of the regulatory of ‘‘trans only” in translatome and more genes were likely to change their allelic expression ratio in F1compared to in parents.All of these results demonstrate that variations in expression pattern from transcriptome to translatome were accompanied by alterations in gene regulation mechanisms.Importantly, we found the syntenic genes (subgenomes maize1 and maize2 genes) enriched in the genes with expected regulatory mechanisms including the ‘‘cisonly”,‘‘trans-only”and‘‘cis+trans”groups at the translatome level.Especially, the subgenome maize2 genes showed significant enrichment while the non-syntenic genes showed a significant depletion (Fig.4D), suggesting the diversity of regulatory mechanisms of ASE genes in maize2.

3.5.Extensive differences between alternative splicing and alternative translation and subgenome bias shown in different translation patterns of isoforms

AS is an important mechanism for increasing the diversity of isoforms from one precursor mRNA by selecting varied splice sites during transcript processing[55].Accordingly,AT is the translation of mRNA isoforms resulting from AS events.To profile the AS events and compare AS variation to AT, we used the randomly truncated RNA-seq reads that similar to the read-length distribution (ranging from 18 to 35 nt) of the Ribo-seq data to remove the effect of sequence length from the comparisons.Five different types of AS and AT events including IR, ES, AA, AD, and OP were identified for both the transcriptome and translatome (Fig.S7A).Overall, the translatome is dramatically different from that of the transcriptome although it is derived from the transcriptome.First,the translatome exhibited much fewer AT events (13,121) than their counterpart AS events (54,630) across all the different AS/AT types and for all three genotypes (Figs.S7B, C).Second, the divergence between the hybrid and the two parents was more obvious for AS than for AT(Figs.S7B,C).Third,a shift in the proportion of changes of different types of AS and AT events were observed between transcription and translation (Fig.S7D).The switch from transcriptome to translatome reveals a genome-wide balance of the dramatic transcriptomic variation in the translatome of the hybrid.

SEVs were employed to determine whether each gene expresses all of its isoforms evenly(the value is larger)or has a biased/dominant isoform(the value is close to 0)[50].Notably,more genes are more likely to express all of their isoforms at the transcriptome level than at the translatome level across all three genotypes(Fig.S8).At the transcriptome level, the SEVs were significantly higher (P< 1.0E-10) in the hybrids than in the two parents, suggesting that the hybrids tend to evenly transcribe all isoforms(Fig.5A).However, a dramatic difference was not observed at the translatome level, which suggests that the genes show similar translation patterns of the isoforms, with either even expression or dominant isoforms(Fig.5A).Such results confirmed the complementation mechanisms at the transcriptome level,and suggest the existence of other mechanisms conferring heterosis at the translatome level.

4.Discussion

Heterosis has been recognized and widely utilized for more than 100 years [56].Many studies have been conducted using genetics, genomics [16,28,29,57,58], transcriptomics[17,19,22,24,27,59], and proteomics [60-63] to dissect the molecular mechanisms underlying heterosis.However, the understanding of heterosis at the translatome level is still limited.Here, we conducted a comprehensive translatomic analysis across a classical hybrid triplet in maize, which provides a genome-wide landscape of the translatomic variation among B73, Mo17, and their F1hybrid, profiles the detailed variation between F1and the parents at the translatome level,and deepens our understanding of heterosis in maize.

4.1.The translatome provides a new perspective to explore the heterosis mechanism

According to the central dogma of molecular biology, genetic information is transferred from genome to transcriptome to proteome, in a process that is under complex regulation.The transcriptome has been widely reported to be a key contributor to phenotypic variation [64]; nevertheless, the correlation between transcriptome and proteome is low [65].Although the proteome is most directly linked to function, the throughput of detecting the proteome by MS is relatively low [66].As an alternative, the translation of the transcriptome into the proteome can be instantly monitored by Ribo-seq [36].Ribo-seq provides an unprecedented chance to profile the translatome and further infer the proteome.In our study, we collected high-quality Ribo-seq data across B73,Mo17, and their F1hybrid.We also collected proteomic MS data,which confirmed that Ribo-seq quantified the proteomic variation more accurately than RNA-seq (Fig.S3C, D).These high-quality translatome data as well as their counterpart transcriptome data accurately monitored the dynamic genetic flow and dramatic variation from transcriptome to translatome between the F1and the parents.

Fig.4.Analysis of biased allelic expression patterns at both transcriptome and translatome levels.(A)and(B)Scatterplot showing the expression ratio of the B73 allele in F1(y-axis)or parents(x-axis)for DEGs at the transcriptome(A)and translatome(B)levels.Different colors represent different patterns of ASE(see Methods).(C)The proportion of DEGs with each type of regulatory mechanisms at the transcriptome (left)and translatome(right).(D)Subgenome enrichment for genes with all regulatory mechanisms except the‘‘unexpected bias”at the translatome level.The significance of enrichment was tested by Chi-square tests.Control groups were defined as corresponding genes in the gene sets selected 1000 times randomly from expressed genes in the translatome with equal gene numbers to the test group.*, P< 0.05; ***,P< 0.001.

4.2.Additivity is prevalent in both the transcriptome and translatome while translatome shows elevated nonadditivity in maize

Additive and nonadditive effects play key roles in heterosis formation.In rice, the transcriptome of shoot of four-leaf stage seedlings of two rice subspecies and their hybrids were profiled and most(~65%)of the DEGs showed a nonadditive expression pattern[67].InBrassica, HP expression-level dominance (the nonadditivity) plays a critical role in heterosis associated to phytohormone response, and nonadditive genes account for about 72.6% of total DEGs (4946 out of 6816) in the other study [68,69].Nevertheless,one different and key observation in our work is that additivity is prevalent in both the transcriptome and translatome.Although there is a dramatic phenotypic change of the F1relative to the parents, it is observed that more than 75% of the DEGs between two parents exhibit additive expression patterns at the transcriptome level in this study, and the dominance of additive effect is consistent to previous studies in maize(Fig.3B)[19,27,70].Here,we also observed that most (63.6%)exhibited additive expression patterns at the translatome level (Fig.3B).This would suggest a low possibility to adjust the expression level for the parents which have a single allele per locus, but a high possibility for the hybrid which has two different alleles per locus.All these results are in accordance with the ‘‘Goldilocks hypothesis”, suggesting that additive expression is advantageous for both transcriptome and translatome [71].

Although most genes exhibited additive expression in the F1, a substantial number of genes exhibited nonadditive expression in both the transcriptome and translatome, consistent with previous studies[22,27].Of these nonadditive genes,most showed complete dominance at both the transcriptome and translatome levels.In the transcriptome,most SPE genes were expressed in the F1hybrid(Fig.S5A).This is consistent with ample previous studies, indicative of a pervasive complementation mechanism associated with heterosis [22-26].However, it is worth noting that more SPE genes,which exhibited low-parent expression,were not translated in the F1hybrid (Fig.S5B).Most intriguingly, more genes were likely to exhibit dominant patterns at the translatome level than at the transcriptome level, suggesting an elevated effect of nonadditivity in the formation of heterosis at the translatome level.Accordingly, the nonadditive effect may be further elevated at the proteome level.More than 50%of differentially expressed proteins in maize seeds exhibit nonadditive pattern in the hybrids.Consistently,most(74)of detectable proteins showed nonadditive expression patterns while much fewer proteins (32)showed additive expression patterns in Sorghum.The elevated effect of nonadditivity in both translatome and proteome suggests the enhanced effect of nonadditivity in post-transcription [72,73].

Fig.5.The genes with different mode of switched dominant isoforms show subgenome enrichment.(A)Shannon entropy values of expressed isoforms for each gene in B73,F1, and Mo17.(B) Number of genes with dominant isoforms that switched between B73, Mo17, and their F1 progeny.(C) Subgenome enrichment for genes with dominant isoforms that switched between F1 and the parents in the translatome.(D)Subgenome enrichment for genes with dominant isoforms that switched between the two parents in the translatome.(E)Number of genes that translated multiple isoforms but have the same dominant isoform in all three genotypes.(F)Subgenome enrichment for genes that translated multiple isoforms but have the same dominant isoform in all three genotypes.The significance of enrichment was tested by Chi-square tests.Control groups were defined as corresponding genes in the gene sets selected 1000 times randomly from expressed genes in the translatome(C,D and F)with equal gene numbers to the test group.*, P< 0.05; **,P< 0.01; ***,P < 0.001.

4.3.Dramatic instraspecific translatomic landscape demonstrates selectively utilization of best alleles in hybrids

One of the interesting observations was the dramatic translatomic landscape across the F1and the parents, which to our knowledge is the first study to show the dramatic intraspecific variation at the translatome level.This turbulent translatomic variation was evident not only in the changed number of translated vs.transcribed gene loci but also in the differences between translated and transcribed isoforms.In the translatome, more genes likely switch their dominant isoforms than in the transcriptome.Importantly, more genes switch the dominant isoforms between the F1and two parents than between the two parents.This observation indicates that the best allele is more likely to be selectively utilized in hybrids for a given environment, leading to higher efficiency of protein accumulation in the hybrids [74,75].

4.4.Asymmetric translational fractionation of the maize subgenomes may be associated to heterosis

Hybrid breeding has been widely applied in crop breeding or animal breeding for a long time, and the heterosis contribute to the improvement of quality or yield.As so far, the complicated mechanism of heterosis has not been explained clearly and completely.Although different studies have proposed different hypotheses associated with heterosis,there lacks a unified mechanism that can explain all of the heterosis phenomena [76].In our study,any single hypothesis can only account for a certain proportion of the genes at the transcriptome and translatome levels.It seems that multiple mechanisms simultaneously contribute to heterosis at both the transcriptome and translatome levels.

Maize has been reported to be an ancient allotetraploid species that evolved from the ancient fusion of two distinct genomes and dramatic chromosome fragmentation and fusion.The current maize genome can be classified into two subgenomes (maize1 and maize2), and all of the annotated genes have been classified as either syntenic or non-syntenic based on the syntenic relationships between maize and sorghum[39,77].An obvious asymmetric evolutionary fate of the two maize subgenomes has been observed,with maize1 as the dominant subgenome[38].It is also worth noting that nonsyntenic genes have been evidenced to contribute to heterosis in maize [22-25].Although the maize subgenomes are reported to function unevenly, the role of functional fractionation of two subgenomes in heterosis has been largely unknown.In our study,we found the subgenome maize1 genes are more detectable and the differentially translated maize1 genes show higher elevated nonadditivity effect.A dramatic switch of gene action mode (additivity vs.nonadditivity) was observed between the transcriptome and translatome,and an elevated translation differentiation between the two subgenomes was frequently observed(however in translatome or proteome).The genes with different biased allele usage underlying expected regulatory mechanisms are enriched in subgenome maize2.Moreover, the genes with switched dominant isoforms across the F1and parents show subgenome maize2 enrichment while the genes with conserved dominant isoforms show subgenome maize1 enrichment.Taken together,significant translational fractionations of two maize subgenomes were consistently observed for the effects of additivity,dominance, biased allele usage and dominant isoform usage in hybrids, indicating that asymmetric subgenome translational selection might be an important and uniform factor contributing to heterosis in maize.

Data availability

The sequencing data for this project have been deposited in the NCBI Sequence Read Archive with a BioProject number:PRJNA576092.

CRediT authorship contribution statement

Lin Li, Yirong Zhang, and Han Zhaodesigned and supervised this study.Wanchao Zhu,Sijia Chen,Tifu Zhang,and Jia Qiancollected all the data.Wanchao Zhu, Sijia Chen, Zi Luo, Tifu Zhang,and Jia Qianperformed the data analysis.Lin Li, Wanchao Zhu,Sijia Chen, and Tifu Zhangprepared the manuscript.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was supported by the National Natural Science Foundation of China(31771798),the National Key Research and Development Program of China (2016YFD0100800), the Competition Fund of the National Key Laboratory of Crop Genetic Improvement,and Huazhong Agricultural University Scientific & Technological Self-Innovation Foundation (2015RC016).

Appendix A.Supplementary data

Supplementary data for this article can be found online at https://doi.org/10.1016/j.cj.2021.02.002.