Plant materials

The RIL122 from the F288 x F271 RIL progeny was crossed with F271 to produce F1 seeds. Then, 31 BC1 lines from the F1 plants x F271 progeny were backcrossed with F271. Genotyping of 31 BC2 lines (25 plants sown per BC1 progeny) was performed using a 50K single-nucleotide polymorphism chip from TraitGenetics. After quality controls based on missing data, heterozygosity, and minor allele frequency, 12,758 markers were retained for genotyping and two BC2 lines (30–16 and 31–04) were selected (S1 Table). Genotyping of BC2S1 lines (30–16 progeny referred to as 1^F271 and 1^F288; 31–04 progeny referred to as 2^F271 and 2^F288) was confirmed by High Resolution Melting (S2 Table). The selected BC2S2 lines were then grown in the field at Mauguio (Hérault, France) in 2015 with two row replicates. Irrigation was applied during summer to prevent water stress. For RNA-seq analysis, the bottom third of upstream-ear internodes (without nodes) of three representative plants in each of the two replicates were harvested two days before silking at tasseling stage. All samples were immediately frozen in liquid nitrogen and stored at -80°C. Because of the presence of borer on the stem in two plants, one of the two 1^F288 samples was removed from the downstream analysis. For cell wall investigations, ears were removed by hand from plants at silage stage the day before and/or the day of harvest. Representative samples of 1 kg chopped material per plot were collected, dried in a ventilated oven 72 h at 55°C and ground with a hammer mill gondard to pass through a 1-mm screen. Genotyping of BC2S2 lines was further assessed by PCR (S2 Table and S1 Fig).

Cell wall composition and degradability assessment

Dried samples from plants without ears were dried at 50°C for at least 14 h to exclude any moisture, before scanning through the Antaris II near-infrared reflectance analyzer (ThermoFisher). Briefly, for each sample, the measurement consisted of an average of 16 scans done while rotating the cups over the 10,000 cm-1–4,000 cm-1 range with a resolution of 8 cm-1, resulting in absorbance data every 4 cm-1. The resulting spectra were loaded into the TQ analyst software (v.9.4.45; ThermoFisher) and smoothed with a running average in a window size of 9 data points. Their first derivate was computed, normalized and used to estimate the cell wall-related traits of each sample using a formula that was previously calibrated for plants without ears [16]. A two-way ANOVA analysis was then conducted using R software (v.3.4.1; http://cran.at.r-project.org) as described [44].

Pangenome construction

Oriented reads (sizing 150) of both F288 and F271 1.5 Mb-genomic sequences which included the QTL6.05 peak were first simulated using wgsim (v.0.3.0, depth = 20). Mapping of these simulated reads and AGPv3.22 real reads to one parental DNA sequence was then performed using Bowtie2 (v2.2.3, sensitive parameter). Then 2,322 representative transcript assemblies (RTAs) [34] that were lower than 45% coverage and 80% identity to the AGPv3.22 or absent in the AGPv3.22 sequence were mapped to both genomic sequences using gmap (v.2011-08-15, default parameter). Genomic sequences aligned with at least one of the previous sequences were then hard masked with N using maskfasta (v.2.25.0) with default parameters (S2 Fig).

RNA-seq datasets

The internodes were crushed in liquid nitrogen using an Ika Mill crusher (IKA, Staufen-Im-Breisgau, Germany) prior RNA extraction. Total RNA was isolated using the TRIzol Reagent (Invitrogen) and purified with the Qiagen RNeasy Plant Mini kit and RNase-free DNase set according to the suppliers’ instructions. Total RNA integrity was evaluated using the Agilent 2100 bioanalyzer according to the Agilent technologies (Waldbroon, Germany). Individual Illumina TruSeq stranded RNA-seq libraries were prepared for each of the genotypes according to the supplier’s instructions. Libraries were sequenced in four lanes on the Illumina HiSeq 2000 platform at the Centre National de Séquençage (CNS, Evry, France) to generate paired-end stranded reads (sizing 260). For each library, RNA-Seq pre-processing was performed including adapters removing and quality control assessment with fastx toolkit (http://hannonlab.cshl.edu/fastx_toolkit/). The fastq datasets were trimmed for Phred Quality Score (Qscore >20, read length >30 bases), and free of ribosome sequences with tool sortMeRNA [45]. All steps of the experiment, from growth conditions to bioinformatic analyses, were managed in CATdb database [46] according to the international standard MINSEQE ‘minimum information about a high-throughput sequencing experiment’.

RNA-seq reads (35 million to 55 million quality-filtered RNA-seq reads per RNA-seq library) were then mapped to panB using TopHat2 (v. 2.0.14) without gene annotation file with minimum intron length set to 5 bp and maximum intron length to 60 kb, with default settings for other parameters. Next, transcript prediction was performed using Cufflinks (v.2.2.1, default parameters). Using bedtools getfasta (v.2.25.0, default parameters), coding sequences were then extracted leading to the CuffpanB dataset. To quantify expression levels, RNA-seq reads were mapped to panB using TopHat2 (v. 2.0.14) with a gene annotation file and the parameters described above. Expression quantification was determined using HTseq-count (v.0.6.0) with stranded orientation and ‘union’ option. The RNA-Seq project was submitted into the international repository NCBI-Gene Expression Omnibus [47] project ID GSE115157.

Gene annotation

Genes were predicted using an improved and maize dedicated version of the TriAnnot pipeline (v.5.2p04) [48,49] as described below.

Differential expression analysis

RNA-seq data were normalized using TMM implemented in the edgeR package (v 3.18.1) [62,63] in R. Gene-wise RNA-seq counts were then analyzed using a negative binomial generalized linear model with a stepwise procedure for increasing the power to detect QTL6.05ⁱ-specific differentially expressed (DE) genes between NILs^F288 (i.e. both 1^F288 and 2^F88) and NILs^F271 (i.e. both 1^F271 and 2^F271) as follows:

In step I, the differential expression was assessed for the two pairs of NILs using a negative binomial generalized linear model with allele and NIL main effects, and an allele × NIL interaction. A Benjamini-Hochberg adjusted P-value of 0.05 was used as the cut-off criterion to identify genes with significant allele × NIL interactions.

In step II, the differential expression was assessed for each pair of NILs (e.g. 1^F288 vs 1^F271) using a negative binomial generalized linear model with allele main effect for each gene. A Benjamini-Hochberg adjusted P-value of 0.05 was used as the cut-off criterion to identify genes with significant allele effect for both NILs 1 and 2.

In step III, to identify DE genes that were not a result of NIL effects but were due to the QTL6.05ⁱ region only, step II DE genes were further dissected by making the intersection between the step I genes with no significant differential expression for NIL effects (i.e. no significant differential expression for both NIL and allele × NIL interaction) and the step II DE genes for both NILs 1 and 2.

In the final step, the intersection between the step III DE genes for NILs 1 and the step III DE genes for NILs 2 delivered DE genes for which the differential expression between the NILs^F288 and the NILs^F271 was declared QTL6.05ⁱ-specific.

Hierarchical clustering of gene expression profile

Hierarchical clustering analyses of the set of common QTL6.05ⁱ DE genes were performed using Euclidean distances and unweighted pair group averages as the aggregation method with the R function hclust.

Functional annotation of common QTL6.05ⁱ genes

MapMan [64] was first used to identify homologous genes in Arabidopsis. GO term enrichment analysis based on the GO classification was then performed by comparing the relative occurrence of a GO term into each DE list (downregulated or upregulated gene list) to its relative occurrence in the genome (reference list) by a hypergeometric test with the R function phyper. A Benjamini-Hochberg adjusted P-value of 0.05 was used as the cut-off criterion.

Correlation analyses

For correlation analysis between gene expression levels and cell wall-related traits, Pearson correlation coefficients (PCCs) were calculated between each DE gene and each trait over the two pairs of NILs using the R cor function. The Student’s t-distribution was used to test the significance of the PCC as follows: t = r/√[(1-r²)/(n-2)] with a level of significance set at 0.05 and n = 4, absolute PCCs higher than 0.95 were significant. The expression pattern of the set of DE genes correlating positively or negatively with at least one cell wall-related trait in both pairs of NILs were visualized using the R cim function from MixOmics (v. 6.3.1) [65].

Gene co-expression network analysis was performed to group DE genes into connected components based on absolute PCCs higher than 0.995 (level of significance of 0.01) and visualized using Cytoscape (v. 3.3.0) [66]. A node corresponded to a DE gene; an edge was determined by the similarity between expression profiles of paired genes calculated with PCCs. Connected modules were identified using NetworkAnalyzer (v. 2.7) [67].

Promoter motif analysis

The presence of conserved motifs in the 5’ regions of the co-expressed genes was determined using the preferentially located motif (PLM) method [68], with minor modifications. Briefly, the dataset was set up using 1,000 bases upstream the transcription start site and 5’ untranslated region (UTR) from the maize AGPv3 plus [69] and the final GFF of the 1.5 Mb-genomic sequences from F271 and F288. Promoters with a 5’ UTR smaller than 10 bases were excluded in order to avoid false promoter regions. The final promoter dataset consisted of 119 promoter sequences, which were divided into two regions. First, the [−1,000, −300] region was used to learn the distribution model using a simple linear regression and determine a 95% confidence interval. Second, non-evenly distributed motifs (i.e. those exhibiting a peak above the confidence interval) were searched within the [−300, UTR] region, leading to the identification of 33 PLMs among 419 motifs tested coming from PLACE [70] and AGRIS [71].

Quantitative RT-PCR analysis of candidate gene expression

Primer pairs were designed from the F271 and F288 1.5 Mb-genomic sequences at the QTL6.05 interval (S2 Table) using Primer-BLAST [72]. Quantitative RT (qRT)-PCR reactions were performed using the Bio-Rad CFX384 touch (Bio-Rad, France) and the SYBR Premix Ex Taq (Tli RNAseH Plus) (Takara, Ozyme, France) following the supplier’s instructions. A dilution series of the pooled cDNAs and a melting analysis were used to validate the primer pairs. Each sample was performed in duplicate. Gene expression was normalized with the mean of the two reference genes ZmGRP2 (GRMZM2G080603) and ZmUBC30 (GRMZM2G102471) coding for Glycine-rich RNA-binding protein 2 and Ubiquitin-conjugating enzyme E2, respectively. Differential expression for each gene was tested with a two-way ANOVA as described [44].

Allelic variation at QTL6.05ⁱ leads to higher cell wall degradability

The RIL122 from the F288 x F271 RIL progeny carried F288 alleles at QTL6.05, although F271 alleles were already fixed at roughly more than half of the genome (Fig 1A) [21]. Therefore, this RIL was backcrossed with the parental inbred line F271 for three generations to fasten the introgression of F288 alleles at QTL6.05 into the background of F271. With the help of deep DNA genotyping, two BC2 lines (30–16 and 31–04) were selected that (1) showed a tremendous increase of F271 alleles fixed throughout the genome (93% and 94.3% for 30–16 and 31–04, respectively) compared to the RIL122 (53.8%) and (2) remained heterozygous at QTL6.05ⁱ with a relatively similar introgressed genomic region size, ranging from 18 to 20 Mb (Fig 1A and S1 Table). According to QTL detection [22], the QTL6.05ⁱ region included the QTL6.05 as well as a cluster of four QTL for cell wall component and degradability (referred to as QTL6.07) with ‘minor’ effects relative to QTL6.05 (R² values ranging between 8 and 12% whatever the trait; Fig 2A). The two selected BC2 lines were then selfed twice to fix the introgressed genomic region at homozygous state and to multiply the progenies of 30–16 (1^F271 and 1^F288) and 31–04 (2^F271 and 2^F288) lines.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Selection and characterization of NILs with an introgressed genomic region at QTL positions for cell wall degradability.

(A) Experimental workflow used for introgression of F288 alleles at an 18–20 Mb region encompassing the QTL6.05 (referred to as the introgressed QTL6.05 region or QTL6.05ⁱ) into the background of F271. SNP: single nucleotide polymorphism. (B) The selected BC2S2 NILs 1 (referred to as 1^F271 and 1^F288) and NILs 2 (referred to as 2^F271 and 2^F288) were grown in the field under well-watered conditions for DNA and mRNA sampling and phenotyping. (C) Comparison of the cell wall degradability IVNDFD performance of NILs introgressed with F288 alleles at QTL6.05ⁱ (blue) and their respective recipient lines with F271 alleles (red). Open circles represent the means of two biological replicates.

https://doi.org/10.1371/journal.pone.0227011.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Structure of the introgressed genomic QTL6.05ⁱ region.

(A) Large structural variation between B73, F271 and F288 in the introgressed genomic QTL6.05ⁱ region. Dark and light purple boxes represent clusters of QTL in bins 6.05 and 6.07, respectively, as previously defined [22]. The grey dotted line insertion represents a zoom of the genomic region that included the 1.5 Mb-genomic sequences from F271 and F288. Orange and green lines indicate sense and antisense DNA strands, respectively, of at least 910 bp and sharing at least 98% of identity; arrowed boxes represent genes and pseudogenes. (B) Number of annotated genes supported by RNA-seq predicted genes in the targeted QTL6.05 locus. (C) Gene content distribution in B73, F288 and F271 in the targeted QTL6.05 locus.

https://doi.org/10.1371/journal.pone.0227011.g002

To assess the cell wall properties of the two pairs of NILs grown in the field (Fig 1B), near-infrared spectroscopy data were then collected as was done in the past for QTL6.05 detection [12,22]. The introgressed F288 alleles at QTL6.05ⁱ conferred on the NILs 1^F288 and 2^F288 (NILs^F288) a higher cell wall degradability (as estimated by IVNDFD) than the F271 alleles on the NILs 1^F271 and 2^F271 (NILs^F271) (Fig 1B and 1C). Although the two pairs of NILs had similar lignin content, the NILs^F288 exhibited lower cellulose content in the cell wall as compared to the NILs^F271 (S4 Table). The NILs^F288 also tended to have higher hemicellulose content in the cell wall as compared to the NILs^F271 (S4 Table). Overall, these results evidenced that substitution of F271 alleles with those from F288 at the QTL6.05ⁱ region led to cell wall composition variation and confirmed the presence of the detected QTL.

Aligning RNA-seq reads to a pan-genomic reference genome: a tool to assess the number and type of genes predicted at the targeted QTL6.05 locus

RNA-seq datasets from internodes of the selected NILs were produced at tasseling stage in which transcript patterns of secondary cell wall cellulose synthase genes and lignin-related genes were shown to be highly abundant [73,74]. To take into account any genetic polymorphism, particularly putative present/absent genes, between F271 and F288 at the targeted QTL6.05 locus, a pan-genomic sequence (panB) was assembled through an in lab pipeline and used as a reference sequence for RNA-seq read mapping (Fig 2A and S2 Fig). This pan-genomic reference sequence included (1) the B73 transcript reference sequences (AGPv3), (2) 2,322 RTAs absent in B73 [34] and (3) genomic sequences specific to either F271 or F288 at the QTL6.05 locus which were kindly provided by the French Plant Genomic Resources Center (CNRGV) (Fig 2A and S2 Fig). The pipeline evidenced total lengths of 40.7 kb and 13.1 kb that were specific to F271 (3.1% of the F271 genomic region) and F288 (0.9% of the F288 genomic region), respectively.

To achieve better mapping of the RNA-seq reads to the panB reference sequence, gene prediction at the targeted QTL6.05 locus was then performed using (1) the TriAnnot pipeline [48] through de novo and similarity-based methods and (2) the transcript prediction based on the mapping of the RNA-seq dataset. About 80% of the predicted genes through the TriAnnot pipeline were supported by the mapped RNA-seq dataset (Fig 2B), highlighting the quality of the genomic sequences of each two lines at the QTL6.05 locus and the annotation. Of the 51 genes present in B73 at the targeted QTL6.05 locus, 39 (76%) were found in both F271 and F288 (Fig 2C and S3 Table). Three genes (genes 23, 41 and 42) were found only in F288 and B73; one gene (gene 43a) provided by the RNA-seq dataset was found only in F288 and F271; one gene (gene 12) was found only in F271. Overall, four genes (genes 12, 23, 41 and 42) were present in only one of the two F271 and F288 parental inbred lines at the targeted QTL6.05 locus.

Mapping of the RNA-seq dataset to the panB reference sequence yielded 32,446 genes, including 1,198 RTAs. It is noteworthy that RNA-seq counts from the F271- and F288-specific genomic sequences present in the panB reference sequence were identified (including counts for genes 10, 11, 12, 13 and 21), highlighting the relevancy of the pan-genomic approach. Two (genes 23 and 41) of the four genes present in only one of the two F271 and F288 parental inbred lines at the QTL6.05 locus did not exhibited measurable expression in all the experimental samples. In contrast, the gene 12 was expressed in all the experimental samples, although it was found only in F271 at the QTL6.05 locus (S3 Table). The gene 42 was consistently expressed at a very low level in the NILs 1^F288 and 2^F288, but not measurable in the NILs 1^F271 and 2^F271. Hence, this gene represents a case of single pattern expression (SPE) [75]. We designated an observed pattern as SPE_F288 if genes were expressed in the NILs^F288 (i.e. in both 1^F288 and 2^F288), but not in the NILs^F271 (i.e. in both 1^F271 and 2^F271). In contrast, we referred to genes that were expressed in the NILs^F271, but not in the NILs^F288 as SPE_F271.

Allelic variation at QTL6.05ⁱ leads to transcriptional variation

To test the hypothesis that the substitution of F271 alleles with those of F288 at QTL6.05ⁱ led to transcriptional variation, RNA-seq counts were analyzed using a stepwise procedure to detect DE genes upon the NILs^F288 relative to the NILs^F271 that depend only on the QTL6.05ⁱ region. As shown in Fig 3A, this stepwise statistical analysis identified 241 and 2,794 DE genes for the NILs 1 and the NILs 2, respectively. Since our goal was to identify DE genes that were due to the introgressed QTL region, we ensured the robustness of the DE gene dataset by keeping only the DE genes that were common to both NILs. This selective approach delivered 125 DE genes, which are referred to as common QTL6.05ⁱ DE genes (Fig 3A and S5 Table). As shown in Table 1, these included one RTA (cov45_id80_joint_Locus_58585), which highlighted once again the relevancy of the pan-genomic approach, and 18 genes located in QTL6.05ⁱ, of which five were located in the targeted QTL6.05 locus: (1) a δ-1-pyrroline-5-carboxylate synthase (gene 03; ZmP5CS1), (2) an aldo-keto reductase (AKR) (gene 06a; hereafter named ZmAKR4C9 according to [76]), (3) a FK506-binding protein (gene 10; hereafter named ZmFKB20-1 according to [77]), (4) one gene (gene 14a) with homology to Arabidopsis ESMERALDA1 (ESMD1), a putative O-fucosyltransferase [78], hereafter named ZmESMD1, and (5) a 40 S ribosomal protein S23 (gene 36; ZmRPS23).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Transcriptomic adjustments in internodes of NILs introgressed at QTL6.05ⁱ.

(A) Venn diagram of upregulated or downregulated genes in the NILs^F288 relative to the NILs^F271. (B) Hierarchical clustering of DE genes in the NILs^F288 relative to the NILs^F271. The indicated scale is the natural log value of the normalized level of gene expression. (C) GO Biological Process analysis of DE genes in the NILs^F288 relative to the NILs^F271. Only Biological Processes with a Benjamini and Hochberg adjusted P-value below 0.05 are represented. Down_List: analyzed downregulated gene list; Up_List: analyzed upregulated gene list; Ref_List: reference list; ACC biosynthetic process: 1-aminocyclopropane-1-carboxylate biosynthetic process; Cell wall macrom. metabolic process: Cell wall macromolecule metabolic process; JA metabolic process: jasmonic acid metabolic process; Reg. hydrogen peroxide met. Process: Regulation of hydrogen peroxide metabolic process; SA-mediated signaling pathway: salicylic acid-mediated signaling pathway.

https://doi.org/10.1371/journal.pone.0227011.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Comprehensive list of the 125 common QTL6.05ⁱ DE genes.

https://doi.org/10.1371/journal.pone.0227011.t001

This prompted us to compare our common QTL6.05ⁱ DE gene dataset to the deregulated transcriptome of four RILs derived from the cross of F288 x F271 [38]. Interestingly, five of the 125 common QTL6.05ⁱ DE genes exhibited DE patterns between F271 and the set of four RILs (Table 1): (1) ZmFKB20-1 described above, (2) an universal stress protein (USP) (GRMZM2G009719), (3) the maize J3 ortholog (GRMZM2G028218; ZmHSP40) encoding an HSP40 known to function in meristem size control in Arabidopsis [79], (4) one gene (GRMZM2G041714) with homology to Arabidopsis hypoxia-induced gene domain 3, and (5) GRMZM2G169382, an ortholog of Arabidopsis group VII ethylene response factor (ERF-VII) ERF71/HYPOXIA RESPONSIVE ERF2 (ZmERF71/HRE2) that may supervise plant intracellular reactive oxygen species (ROS) homeostasis [80]. It is noteworthy that the gene 42 with unknown function did not belong to the common QTL6.05ⁱ DE gene dataset, which is not in favor of a direct role of this gene in cell wall-related trait variation. Importantly, none of the common QTL6.05ⁱ DE genes were located in the QTL6.07 confidence interval.

Of the common QTL6.05ⁱ DE genes, 52 and 73 were found to be downregulated and upregulated in the NILs^F288 relative to the NILs^F271, respectively (Fig 3A and 3B, and Table 1). Interestingly, two of the 52 downregulated genes showed SPE_F271 patterns, while seven of the 73 upregulated genes showed SPE_F288 patterns (Table 1). Additionally, GO analysis revealed that the downregulated genes had significant enrichment in the “1-aminocyclopropane-1-carboxylate (ACC) biosynthetic process”, which is the first committed and in most instances the rate-limiting step in ethylene biosynthesis (Fig 3C and S5 Table). For the upregulated genes, enriched GO terms associated with biological processes were related to oxidation-reduction, regulation of hydrogen peroxide metabolism, response to hypoxia, jasmonic acid (JA) metabolism, salicylic acid (SA)-mediated signaling pathway, protein folding, response to heat, cell wall macromolecule metabolism and xylem development (Fig 3C and S5 Table).

Almost two-thirds of the common QTL6.05ⁱ DE genes are associated with the cell wall-related traits

To test the ability of the 125 common QTL6.05ⁱ DE genes to be associated with the cell wall-related traits, linear correlation between gene expression levels and traits was determined by calculating PCCs between the expression level of each of the common QTL6.05ⁱ DE genes and each cell wall-related trait. We found 79 DE genes significantly associated with at least one of the traits, referred to below as the correlating DE gene set (Table 1 and S6 Table). This included the cov45_id80_joint_Locus_58585 and the five DE genes (ZmP5CS1, ZmAKR4C9, ZmFKB20-1, ZmESMD1 and ZmRPS23) located in the QTL6.05 locus (Table 1).

The number of correlating DE genes ranged from 15 for one trait up to 20 for three traits (Fig 4A and S6 Table). A subset of three upregulated genes (AC209215.4_FG004, GRMZM2G085964 and GRMZM2G142802) were associated with six traits (S6 Table). Among them, one gene (AC209215.4_FG004) with homology to Arabidopsis voltage dependent anion channel 1 showed SPE_F288 pattern (Table 1). Interestingly, the subset of three upregulated genes also included a transcription factor (TF) with homology with Arabidopsis ERF72/RELATED TO APETALA2.3 (RAP2.3) (GRMZM2G085964; hereafter named ZmERF72/RAP2.3) linked to the hypoxia response [81]. The third upregulated gene (GRMZM2G142802) was annotated as a stem-specific protein TSJT1 and related to enriched GO terms “SA-mediated signaling pathway”, “regulation of hydrogen peroxide metabolic process” and “hypoxia” (Table 1 and S5 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Identification of DE genes associated with cell wall-related traits.

(A) Number of common QTL6.05ⁱ DE genes correlating with at least one of the cell wall-related traits. (B) Gene expression levels of the 79 correlating DE genes in NILs introgressed into QTL6.05ⁱ. Columns represent gene expression profiles, and rows represent cell wall-related traits as follows: cell wall content (black), degradability (orange), sugar content (pink), lignin content (green) and p-hydroxycinnamic acids content (purple). IVNDFD: in vitro neutral detergent fiber digestibility; Hcell/NDF: hemicellulose content in % NDF; IVDMD: in vitro dry matter digestibility; EsterFA: esterified ferulic acid content in mg g^-1 NDF; LK/NDF: lignin klason in % NDF; NDF: neutral detergent fiber; ADL/NDF: acid detergent lignin in % NDF; Cell/NDF: cellulose content in % NDF; pCA: p-coumaric acid content mg g^-1 NDF; EtherFA: etherified ferulic acid content in mg g^-1 NDF.

https://doi.org/10.1371/journal.pone.0227011.g004

The expression levels of the 79 correlating DE genes and their correlation with the cell wall-traits were visualized in a clustered map, revealing opposite correlation of individual DE genes with multiple traits (Fig 4B). It is worth noting that upregulated genes (including ZmAKR4C9, ZmESMD1 and ZmRPS23 located in the QTL6.05 locus) showed positive correlations with cell wall degradability and hemicellulose content, and negative correlations with cellulose content (Fig 4B). In contrast, downregulated genes (including ZmP5CS1 and ZmFKB20-1 located in the QTL6.05 locus) showed negative correlations with cell wall degradability and hemicellulose content, while exhibiting positive correlations with cellulose content (Fig 4B).

Toward a co-expression network of genes associated with QTL6.05ⁱ

To get more insights into the putative coregulation of the 125 common QTL6.05ⁱ DE genes, a network analysis of co-expression relationships among the common QTL6.05ⁱ DE genes was performed using PCCs between the gene expression levels. The resulting gene co-expression network incorporated 116 DE genes and 329 edges (Fig 5 and S7 Table). Seven connected modules (designated M1-M7) were determined and significant correlations with cell wall degradability and sugar content were visualized in the gene co-expression network for each DE gene (Table 1). As shown in Fig 5, modules M1 (77 DE genes and 283 edges), M4 (4 DE genes and 3 edges) and M7 (2 DE genes and 1 edge) brought together the majority of the correlating DE genes since 58% (45/77), 100% (4/4) and 100% (2/2) of their nodes, respectively, were correlated with at least two of the three cell wall-related traits. It is also worth noting that 70% (54/77) of the genes in module M1 correlated with cell wall degradability (Table 1). In contrast, DE genes within modules M2 (24 DE genes and 34 edges), M3 (5 DE genes and 6 edges), M5 (2 DE genes and 1 edge) and M6 (2 DE genes and 1 edge) tended to be almost not correlated with cell wall-related traits (Table 1 and Fig 5). Interestingly, up to two-thirds of the genes in module M1 that were associated with at least one of the three cell wall-related traits were related to the enriched biological processes we identified. One noteworthy gene of module M1 that negatively correlated with cell wall degradability and hemicellulose content was an ACC synthase (ACS) gene (GRMZM5G894619; hereafter named ZmACS7) involved in the ethylene biosynthetic pathway (Fig 5). Also related to hormone responses were the two ERF-VIIs (ZmERF71/HRE2 and ZmERF72/RAP2.3) that showed positive correlation with the three cell wall-related traits (Fig 5). Another class of genes correlated with cell wall degradability are heat shock proteins (HSPs) known to affect protein folding and response to heat (Fig 5). Two positively correlating genes (GRMZM2G012631 and GRMZM2G069651) with homology to Arabidopsis HSP90 were connected to a limited set of other genes, including another maize HSP90 ortholog (GRMZM2G024668) and ZmHSP40.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Co-expression network of common QTL6.05ⁱ DE genes.

Downregulated genes in the NILs^F288 relative to the NILs^F271 are represented by diamonds, and upregulated genes by circles. Colored diamonds indicate genes with expression levels negatively correlating with cell wall degradability (IVNDFD) and/or hemicellulose content (Hcell/NDF), and/or positively correlating with cellulose content (Cell/NDF). Colored circles indicate genes with expression levels positively correlating with IVNDFD and/or Hcell/NDF, and/or negatively correlating with Cell/NDF. Node colors represent correlation with IVNDFD (yellow), IVNDFD and Hcell/NDF (blue); IVNDFD, Hcell/NDF and Cell/NDF (red); Hcell/NDF and Cell/NDF (orange). The “i” present in the nodes indicates DE genes located in the QTL6.05ⁱ region; numbered and larger nodes indicate genes located in the QTL6.05 locus (gene 03: ZmP5CS1; gene 06: ZmAKR4C9; gene 10: ZmFKB20-1; gene 14a: ZmESMD1; gene 36: ZmRPS23). The letters “A”, “D”, “E”, “H”, “N” and “U” present in the nodes indicate genes encoding ACC synthase, DREB TF, ERF-VII TF, HSP, NAC TF and universal stress protein, respectively. Blue stars indicate DE genes previously identified [38]; yellow stars indicate DE genes with SPE patterns co-expressed with ZmRPS23 (gene 36). Thick borders indicate gene with the GCCGCC motif in their promoter. Edges colored blue connect co-expressed genes with a PCC less than -0.995, red edges represent a PCC bigger than 0.995. Modules as determined by NetworkAnalyzer are designated M1-M7.

https://doi.org/10.1371/journal.pone.0227011.g005

Of the 116 DE genes incorporated in the gene co-expression network, 16 (almost 14% of the DE genes incorporated in the gene co-expression network) located in QTL6.05ⁱ, including the fives genes (ZmP5CS1, ZmAKR4C9, ZmFKB20-1, ZmESMD1 and ZmRPS23) located in the QTL6.05 locus (Table 1 and Fig 5). Furthermore, we observed that 13 of these 16 DE genes were in module M1 (including ZmP5CS1, ZmAKR4C9 and ZmRPS23), while three were in module M4 (including ZmFKB20-1 and ZmESMD1). All these 16 DE genes were correlating with at least two of the three cell wall-related traits. Of note, ZmRPS23 in module M1 had high connectivity with 13 DE genes (including ZmERF71/HRE2 and eight of the nine DE genes displaying SPE_patterns) that were all correlated with cell wall degradability and sugar content (Fig 5). Moreover, seven neighbors of ZmRPS23 were connected with ZmAKR4C9, while four neighbors of ZmRPS23 were connected with ZmP5CS1. It is also worth noting that ZmAKR4C9 had high connectivity with eight DE genes (including ZmERF2/RAP2.3) that were all correlated with cell wall degradability and sugar content (Fig 5). Additionally, three of the four DE genes of module M4 were located in QTL6.05ⁱ, namely ZmFKB20-1, ZmESMD1 and one gene (GRMZM5G869403) annotated as exocyst subunit EXO70 family protein B1 (ZmEXO70B1).

Transcription factors associated with QTL6.05ⁱ

Genes having the same TF binding sites are assumed to be functional partners since they are regulated by the same TFs. Therefore, a PLM study [68] was carried out to determine whether there were any over-represented TF binding sites in the promoter of the common QTL6.05ⁱ DE gene dataset. Among the promoters of the common QTL6.05ⁱ DE genes, we identified the CATGTG motif (S8 Table) which is bound by NAC TFs [82]. Interestingly, there was one gene (GRMZM2G465835) in module M1 (Fig 5) located in QTL6.05ⁱ and showing homology to Arabidopsis NAC090, which is a negative regulator of SA-mediated leaf senescence [83]. Also present in the promoters of the common QTL6.05ⁱ DE genes were (1) the TGACG motif which is involved in transcriptional activation of several genes by auxin and/or SA [84] and (2) the GCCGCC motif which is the core site of GCC-box required for the binding of AP2/ERF TF family and for the regulation of JA-responsive gene expression [85,86]. It is noteworthy that in module M1 the two ERF-VIIs (ZmERF71/HRE2 and ZmERF72/RAP2.3) were central in the sense that together they covered 73% (19/26) of the DE genes (including ZmAKR4C9 and ZmRPS23 located in the QTL6.05 locus) that were correlated with cell wall degradability and sugar content. Additionally, a member (GRMZM2G368838) of the DREB (dehydration-responsive-element-binding) subfamily A-4 of AP2/ERF TF family was also present in module M1 (Fig 5). One noteworthy neighbors of GRMZM2G368838 was ZmACS7 (Fig 5). Finally, several F271 or F288-specific PLMs were identified in the promoter of ZmAKR4C9, ZmFKBP20-1, ZmESD1 and ZmRPS23 (S3 Fig). Although, the biological importance of these genotype-specific motifs will require further investigation, these results, together with those of Fig 5, support ZmAKR4C9, ZmFKBP20-1, ZmESD1 and ZmRPS23 as candidate genes for the targeted QTL6.05 locus. Therefore, to confirm the differential expression, qRT-PCR experiments were performed based on the same samples that had been used for the initial RNA-seq analysis. Transcript levels of ZmAKR4C9 were not quantified because it was not possible to design gene-specific primers of sufficient quality for qRT-PCR. The other three candidate genes were significantly affected by the substitution of F271 alleles with that of F288 at QTL6.05ⁱ with opposite effects: (1) downregulation of ZmFKBP20-1 in the NILs^F288 relative to the NILs^F271; (2) upregulation of ZmESD1 and ZmRPS23 in the NILs^F288 relative to the NILs^F271 (S9 Table). These results were in agreement with those obtained by RNA-seq (Table 1 and S5 Table), thus validating our differential expression analysis.

Expanding maize genetic and genomic resources to shed light on the mechanisms whereby the QTL6.05 locus influences cell wall degradability

We implemented a systems biology approach that combined genetic, genomic, transcriptomic and phenotypic data to begin answering the mechanisms whereby the major QTL6.05 locus influences cell wall degradability. The use of the NILs that we developed using the F271 inbred line as the recurrent parent and the RIL122 carrying F288 alleles at QTL6.05 as donor enabled us to show that substitution of F271 alleles with those of F288 at QTL6.05ⁱ led to an increase of four percentage points of IVNDFD, as previously reported for QTL6.05 in QTL mapping experiments [21,22]. Consequently, the results obtained with the selected NILs most likely reflected the QTL6.05 rather the QTL6.07, which had ‘minor’ effects [21,22]. Consistently, none of the 125 common QTL6.05ⁱ DE genes were located in the QTL6.07 support interval. Additionally, similar to what we observed for cell wall degradability, we found that hemicellulose content tended to be higher in the NILs^F288 compared to the NILs^F271. In contrast, cellulose content in the NILs^F288 was significantly lower relative to the NILs^F271. These data are comparable to the previous characterizations of the F288 x F271 RIL progeny [12], as well as those of a forage maize doubled haploid population [23], an association panel [87] and six mapping populations derived from European elite maize germplasm [25]. Therefore, these results, together with those of cell wall degradability, showed that the NILs developed in this study provided the suitable biological tools to understand the underlying mechanisms and genes that control cell wall-related trait variation at QTL6.05.

Gene prediction at the QTL6.05 locus from both F271 and F288 inbred lines highlighted the three genes 23, 41 and 42 that are present in F288 but not in F271, and the gene 12 that is present in F271 and not in F288. It is noteworthy that the gene 12 was expressed in the two pairs of NILs, which might reflect relocated single-copy gene in F288. In contrast, we evidenced 10 genes displaying SPE patterns, including the gene 42. Hence, such SPE pattern was consistent with the absence of the gene 42 in F271 at the QTL6.05 locus interval and suggests that this gene is completely absent from the genome of F271. It was not possible to determine the fraction of genes displaying presence/absence variation among the nine other identified SPE patterns because of the lack of genomic data from both F271 and F288. Considering that low proportions of presence/absence variations had been observed among genes displaying SPE patterns [37,75,88,89], we anticipate that these nine genes (two-thirds of which have unknown function) represent cases of genotype-dependent differential gene expression. Alleles from the unfavorable NILs for cell wall degradability were activated in the favorable NILs for cell wall degradability presumably via interactions with QTL6.05ⁱ-specific regulatory factors from the favorable NILs. Consistently, the group of co-expressed genes that were located in QTL6.05ⁱ included the maize NAC090 ortholog involved in SA-mediated signaling pathway [83], which is one of the enriched biological processes observed here. Moreover, the nine SPE patterns were correlated with cell wall-related traits. This observation confirms the expectation that SPE can affect the phenotype variation directly [35,37] and suggests a biological role of these SPE patterns in the variation of cell wall degradability controlled by the QTL6.05 locus.

Identification of hub genes associated with cell wall-related trait variation at QTL6.05ⁱ

Additionally, the systems biology approach employed in this study revealed biological processes occurring in response to allelic variation at QTL6.05ⁱ. First, a set of processes related to oxidation-reduction and regulation of hydrogen peroxide metabolism was included in the gene co-expression network. Two noteworthy genes are a ferulate 5-hydroxylase (GRMZM2G100158) and a cinnamyl alcohol dehydrogenase (GRMZM2G118610) involved in lignin biosynthesis [90]. In addition, two plant cysteine oxidases (AC203989.4_FG001 and GRMZM2G113355; PCOs), one ACC oxidase (GRMZM2G126732) involved in the ethylene biosynthetic pathway, one respiratory burst oxidase homolog gene (GRMZM2G300965 or ZmRbohB) encoding a NADPH-oxidase and ZmAKR4C9 were also observed in the gene co-expression network. These results, together with those described previously [38], support the idea that a tight and complex regulation, a long way upstream of the monolignol biosynthetic pathway, is driven by the F288 alleles at QTL6.05ⁱ.

Second, genes involved in hormone responses are co-expressed in the network. Notably, ZmACS7, the maize PCO ortholog AC203989.4_FG001 and the two ERF-VIIs, ZmERF71/HRE2 and ZmERF72/RAP2.3, were co-expressed in module M1 and associated with cell wall-related trait variation. PCOs were shown to oxidize the penultimate cysteine of ERF-VIIs by using oxygen as co-substrate, thereby controlling the lifetime of these proteins in Arabidopsis [91,92]. A perspective is emerging in which a diversified set of mechanisms can influence ERF-VII expression in order to specify their functions in a wider network of physiological pathways, in particular those activated by hormones [81,93]. In that respect, a target of ERF-VIIs, namely hypoxia responsive universal stress protein 1, has been shown to coordinate oxygen sensing by PCO/ERF74 (also known as RAP2.12) with H₂O₂ production by NADPH oxidases in Arabidopsis [94]. Consistently, module M1 also included one USP ortholog already described [38] and one maize NADPH oxidase ortholog ZmRbohB. Co-expression of these genes in module M1, together with the identification of the TGACG and GCCGCC motifs (included in the maize PCO ortholog AC203989.4_FG001 and the USP gene) and the recent observation that methyl jasmonate and SA are able to modify the cell wall structure in Brachypodium [95], thus opens the exciting possibility that ERF-VIIs might play a role in the mechanisms whereby the QTL6.05 influences cell wall degradability.

Third, genes involved in protein folding and response to heat are co-expressed in the network, including ZmHSP40 already described [38] and three maize HSP90 orthologs. HSP40 proteins can initiate a conserved chaperone assembly line that mediates conformational changes required for the activity of many native proteins, including HSP70 and HSP90. In plants, HSP90 chaperone pathway plays crucial role in development, and known clients include auxin and JA receptors [96,97]. Furthermore, transcripts of the Arabidopsis HSP90 orthologs of GRMZM2G024668 and GRMZM2G069651 were reported to move in root-to-shoot direction [98], which may indicate that they function in shoot. Overall, these results also implicate protein folding in the mechanisms underlying the QTL6.05 effects.

Identification of four candidate genes for the targeted QTL6.05 locus

Finally, combining meta-analysis of gene co-expression network, phenotypic NIL data and specific TF binding sites within promoters revealed ZmFKB20-1, ZmESMD1, ZmAKR4C9 and ZmRPS23 as promising candidate genes for the targeted QTL6.05 locus. The first two genes belonged to module M4. ZmFKB20-1 which was one of only two DE genes already identified in the targeted QTL6.05 support interval [38] belongs to the FK506-binding protein family, which contains up to 30 paralogous genes in the maize genome [77,99]. Although historically linked to immunosuppression and proline bond rotation, the physiological importance of FK506-binding proteins extends general protein folding to diverse biological processes, including hormone signaling, DNA transcription and protein trafficking. In module M4, ZmFKBP20-1 was co-expressed with a limited set of other genes, including ZmEXO70B1 involved in cell polarity and morphogenesis [100] and ZmESMD1 involved in cell adhesion through as yet unidentified cell wall modifications in Arabidopsis [78]. Together, these results indicate that ZmFKBP20-1 may function in cell wall-related trait variation and highlight the potential importance of cell adhesion in the mechanisms underlying the QTL6.05 effects.

The two other candidate genes ZmAKR4C9 and ZmRPS23 belonged to module M1. They both had high connectivity with other DE genes in the gene co-expression network, including the two ERF-VIIs and eight of the nine DE genes displaying SPE_patterns, which were found to be associated with cell wall-related traits. Comparison of multiple sequence alignments suggests that ZmAKR4C9 may be involved in detoxification mechanisms similar to those of AtAKR4C9 [101]. Consistently, gene expression of tomato AKR4C9 was shown to be induced by hydrogen peroxide and plant hormones such as SA and JA [102], which coincide with the enriched biological processes we observed here. In contrast to ZmAKR4C9, ZmRPS23 belongs to the ribosomal protein S12/S23 family that functions in translation fidelity [103]. Interestingly, several studies identified RPS23 as an enzyme substrate for the oxygenase OFD1 in fission yeast [104]. When little oxygen is around, OFD1 cannot hydroxylate RPS23. This traps OFD1 away from a TF called SRE1. As a result, unassembled RPS23 regulates SRE1 signaling by sequestering OFD1 in an oxygen-dependent manner, thereby coupling hypoxic gene expression to rates of ribosomal synthesis. Therefore, our findings raise the intriguing possibility that unassembled ZmRPS23 regulates an as yet undetermined TF-mediated signaling pathway in which ERF-VIIs may be involved.