Weeds are the major problem in agricultural production worldwide because they cause high crop yield losses and economic damage. Italian ryegrass (Lolium multiflorum
Lam.; LOLMU) is the most common annual weed found in fields in temperate climates [1
]. LOLMU is a C3
annual grass that reproduces using seeds and is self-incompatible, allowing genetic diversity to evolve and adapt to a wide range of environments [2
]. This grass grows vigorously, is highly competitive with crops, and generally is cultivated as pasture, favoring its high density [5
]. LOLMU also interferes in wheat and corn, reducing crop yields by 60% and 48%, respectively [4
In Brazilian agricultural fields in the past two decades, LOLMU has generally been controlled with glyphosate, a 5-enolpyruvylshikimate 3-phosphate synthase (EPSPS) inhibitor [8
]. However, the repetitive application of glyphosate has selected glyphosate-resistant (GR) LOLMU populations. The first case of GR LOLMU in Brazil was reported in 2003, and it is now present over approximately 3.5 million hectares [6
]. Moreover, it is estimated that the cost to control GR LOLMU is about 150% higher than to control glyphosate-sensitive (GS) plants, resulting in a significant economic impact [8
Herbicide resistance is the result of weed evolution and the genetic variability of plants present in several environments. It can occur in two ways, involving either herbicide target-site resistance (TSR) or non-target-site resistance mechanisms (NTSR) [10
]. TSR mechanisms include alterations at the herbicide’s target enzyme, preventing herbicide–enzyme binding or conferring its overexpression and increase in activity [12
]. NTSR describes any other mechanism not categorized as TSR, such as differential uptake and translocation, vacuolar sequestration, metabolic resistance by enhanced herbicide degradation, or protection against oxidative damage [14
]. Despite the high importance of GR weeds to agriculture, their molecular mechanisms are unknown in any species. The high complexity of NTSR mechanisms represents diverse ways that weeds have evolved to deal with the stresses caused by herbicides. It is probable that many of them are still to be understood. A new NTSR mechanism could include the ability to prevent or reduce the amount of herbicide entering the cell in resistant plants.
Transcriptome analysis (RNA-seq) through next-generation sequencing (NGS) is a powerful method for studying global gene expression in organisms [15
]. This method has been applied successfully in weed science and provided breakthrough data for molecular studies, such as on mechanisms of herbicide resistance [17
]. Transcriptome studies produce quantitative and qualitative data [22
], and the assembling of a transcriptome can be performed without the need for a reference genome (de novo
assembly). It represents a huge advantage for studies in non-model plants such as LOLMU, which does not have a reference genome available. Likewise, global differential gene expression analysis allows comparisons between GR and GS biotypes with and without herbicide treatment, paving the way to the identification of the most responsive genes involved in the molecular herbicide-resistance mechanisms.
The comparisons of global patterns of gene expression between GR and GS biotypes have the potential to provide the first step towards the identification of the molecular mechanisms of glyphosate resistance in LOLMU. The identification of molecular mechanisms is critical to understanding weed biology, and consequently, to the development of new management strategies to manage herbicide-resistant weeds. In this way, the present transcriptome study aimed to analyze the global gene expression between GR and GS biotypes in response to glyphosate treatment, in order to identify highly responsive genes in the GR biotype and to provide a list of candidate target genes that might be conferring glyphosate resistance in LOLMU.
The present study is the first transcriptome analysis reported for the L. multiflorum species, and providing a large amount of high-quality data. The molecular data provided here will support future general molecular studies in this species, as well as serving as a dataset for further studies towards understanding the glyphosate-resistance mechanisms in LOLMU. The 87,433 annotated contigs in the LOLMU transcriptome contained the most responsive molecular changes in the GR and GS biotypes in response to glyphosate treatment. The genes that will lead us to the mechanisms of resistance to glyphosate in LOLMU are most likely represented among the differentially expressed contigs.
The high proportion of non-annotated contigs’ functions demonstrate the necessity for the characterization of weed genomes—in this particular case, for L. multiflorum
or other closely related species. The characterization of the LOLMU transcriptome will support molecular studies in the weed science field, such as weediness and the evolution of herbicide-resistance mechanisms [25
]. Since there is no available LOLMU genome and knowing that weed genomics is essential to the study of weed biology, and understanding of weed biology is critical for weed management [25
], the present LOLMU transcriptome will serve as a useful dataset for this type of study. The differential gene expression in both GR and GS biotypes in response to glyphosate treatment will also help to identify candidate genes and mechanisms involved in the evolution of glyphosate resistance.
The non-occurrence of nucleotide substitution at Thr 102 and Pro 106, the absence of a significant increase in EPSPS gene expression in GR biotype, and the presence of a single coding-gene sequence in both GR and GS biotypes provided evidence that the glyphosate resistance in the studied GR LOLMU biotype is due to a non-target-site resistance (NTSR) mechanism. Another study using the same GR and GS LOLMU biotypes investigated alterations in EPSPS coding sequences through Sanger sequencing and also concluded that there were no alterations at the 102 and 106 codons [26
]. We also verified the EPSPS expression by qRT-PCR, and both methodologies, RNA-seq and qRT-PCR, indicated relatively higher responses in GS than in GR in response to glyphosate treatment (Table S1; Figure S4
The gene ontology assignments performed for the up- and down-regulated genes in GR as a response to glyphosate treatment demonstrated a significant proportion of up-regulated genes classified into the transmembrane transport (Figure 5
). These genes also represented the highest proportion of those annotated in the cellular component. On the other hand, genes related to photosynthesis processes into the chloroplasts were the most abundant down-regulated genes (Figure 5
). The great number of highly glyphosate-induced genes related to cell membrane processes in the GR biotype indicated that the responses to glyphosate action stimulate the plant defenses to create a barrier to reduce the amount of or prevent entirely the glyphosate entering the cell. Plant cells are surrounded by highly dynamic cell walls and plasma membrane, performing significant roles in plant development [27
]. The cell wall and membrane are the first layer of protection against abiotic stresses [27
], which might include herbicide [28
]. The detection of potential stressors to the cell by receptors triggers several coordinated signaling events that result in the production of protective metabolites, cell-wall and membrane remodeling, or even cell death [27
Glyphosate directly inhibits the enzyme EPSPS, which results in an interruption of the shikimic acid pathway and consequent disruption of aromatic amino acid biosynthesis [29
]. Interruption of the shikimic acid pathway will trigger an accumulation of shikimic acid [30
]. A previous study in Conyza canadensis
reported a linear correlation of shikimic acid content with the amount of glyphosate transported into the cells [31
]. In the present study, the low accumulation of shikimic acid in the GR biotype is strong evidence of the reduced amount of glyphosate entering the cells and reaching its target, the EPSPS enzyme (Figure 2
). Hence, another study with the same biotypes evaluated high rates of glyphosate (0–11,520 g ae ha−1
) found that increasing the glyphosate rate did not increase the shikimic acid accumulation at 48 h after treatment [26
]. Additionally, the significant number of down-regulated genes involved in the photosynthesis process in the GR biotype indicated the occurrence of signaling for plant metabolism slowdown as a coping mechanism against glyphosate action. This will ameliorate the plant’s defenses against the glyphosate movement and damage. The reduction in plant metabolism is the initial response to abiotic stress [32
The reduced number of up- and down-regulated genes in the GR biotype (146) indicated that glyphosate treatment did not induce significant molecular changes when compared to the GS biotype, which had over a thousand DEGs (1683) (Figure 4
). The low number of DEGs in the GR biotype is further evidence of the reduced amount of glyphosate entering the cell, because glyphosate action causes great disturbances in plant physiology and metabolic processes [33
]. Because of this, it was expected that a great number of DEGs would be observed for the GR biotype. Recent transcriptome studies seeking glyphosate-resistance mechanisms in Conyza bonariensis
] and Echinochloa colona
] found about 4500 DEGs in response to glyphosate in GR biotypes and similar results in GS. After glyphosate reaches and binds to EPSPS, it interrupts the shikimic acid pathway and the biosynthesis of aromatic amino acids (phenylalanine, tyrosine, and tryptophan) [30
]. The inhibition of the shikimic acid pathway leads to an accumulation of shikimic acid, reducing power (NADPH+H), production of reactive oxygen species (ROS), lipidic peroxidation, and membrane disintegration, ultimately leading to cell death [30
]. Therefore, glyphosate action results in wide perturbation of the plant’s metabolic system [35
]. In the present study, the alterations on gene expression in the GR biotype were low. Still, the results enabled the selection of a narrow and effective candidate gene list for involvement in glyphosate resistance mechanisms.
Among the differentially expressed genes from the RNA-Seq analysis, we selected the 21 most responsive to glyphosate treatment in GR in comparison to GS, 14 being up-regulated and 7 down-regulated (Table 2
). In the up-regulated candidate gene list, two groups of well-known genes established to be involved in herbicide conjugation and transport were the most induced by glyphosate, glycosyltransferase (GTs) and ABC transporters [23
]. In general, it is well accepted that herbicide NTSR, usually metabolism and degradation, follows a four-phase process to protect the plant against irreversible herbicide damage and death: first—oxidation, second—conjugation, third—transport, and fourth—degradation, detoxification, and protection [10
]. On the candidate target gene list reported in the present study, there are three upregulated GTs and one ABC transporter. GTs are in the cytoplasm and conjugate lipophilic molecules, such as herbicides, directly or to several substrates, which results in a polar product favoring its transport or pumping into vacuoles by ABC transporters [10
The candidate gene list also includes additional highly induced genes that most likely act to prevent the glyphosate molecule of transposing through the cell wall and plasmatic membrane (Table 2
). From the 14 upregulated genes, nine are related somehow to membrane processes. Once this type of mechanism is proven, the prevention of herbicide entering the cell might be a new phase added to the four phases of NTSR mechanisms. Recently, an article reported that the aldo-keto reductase metabolizes glyphosate and confers resistance in E. colona
]. We investigated the transcription levels of the aldo-keto reductase in the LOLMU transcriptome and found no difference in expression between GR and GS biotypes with or without glyphosate treatment.
Two up-regulated genes were annotated as ubiquitin (Table 2
), which is a type of protein that exists freely or conjugated to another protein. In general, ubiquitin is involved in protein posttranslational modification or degradation via the proteasome. However, it is also involved with the activation of the protein kinases and cell signaling [39
]. Proteins and regulators from several processes are targeted by ubiquitin for further degradation at the proteasome, allowing the cells to maintain cellular responses to environmental changes, such as abiotic stresses [39
]. The role of ubiquitin during abiotic stress, such as herbicide exposure, involves controlling the protein load in the cell, which will affect many cellular activities, including signaling and gene expression [39
]. As glyphosate inhibits the biosynthesis of amino acids, a possible hypothesis that still needs to be determined for increasing ubiquitin expression is that it works to increase protease activities in order to release free amino acids upon herbicide treatment [41
]. Another hypothesis might be the involvement of ubiquitination on the degradation of toxic proteins produced after glyphosate action.
In addition to GTs, ABC transporters, and those genes related with activities on the cell membrane, assuming that their functions prevent glyphosate entering into the cell, the participation and function of the other glyphosate-resistance candidate genes reported in Table 2
on molecular responses to glyphosate action are yet to be determined (up-regulated genes related to oxidoreductase activity (Fe2OG dioxygenase), gibberellin oxidase; and down-regulated genes related to oxidoreductase process (DAO-domain protein), cell-wall organization (glycine-rich cell wall structural protein), RNA polymerase, senescence-associated protein, transcription factor, uncharacterized protein, and vacuolar protein-sorting-associated).
In the present study, two biotypes were studied. Future studies should increase the number of populations of LOLMU to clarify the representativeness of this mechanism in the field. Further genomic data and complete characterization of the gene ontology, as well as functional genomics of the L. multiflorum or related species, will be helpful to validate the mechanisms of glyphosate resistance suggested by the target gene list. In the meantime, the present data will drive further studies on functional genomics of this narrow group of 21 target genes towards underlying the mechanisms of glyphosate resistance. Techniques such as genome editing approaches, e.g., CRISPR/Cas9 systems, could be used to knock out the candidate genes for further phenotyping evaluations and validation of the mechanism. Labeled-glyphosate studies could also be performed to evaluate glyphosate movement at the plant level, as well as at the cellular level using protoplasts (cells without a cell wall) and cell culture approaches (cell with cell wall).
4. Material and Methods
4.1. Glyphosate Dose–Response and Whole-Plant Shikimic-Acid Bioassay
Two biotypes of LOLMU, GR (SVA04) and GS (SVA02), originating from São Valentin-RS/Brazil (27.35° S, 54.28° W) were vegetatively multiplied (separating tillers), transplanted to individual pots, and grown in a greenhouse. Glyphosate dose–response experiments were performed, following the official criteria to determine the dose required to cause 50% of growth reduction (GR50
) in both GR and GS biotypes [42
]. The Roundup Original 360 SL (Monsanto - Brazil) was applied on 4–6-leaf-stage plants at 0, 180, 360, 720, 1440, 2880, 5660, and 11,520 g ae ha−1
using a CO2
backpack sprayer delivering 120 L ha−1
The quantification of the shikimic acid content (SAC) was performed according to Singh and Shaner and Perez-Jones et al. [43
] with previously described modifications [45
]. The top three leaves of GR and GS biotype plants were harvested after glyphosate treatment and immediately stored at −80 °C. The time-points used for SAC determination were 0, 24, 48, 96, and 192 h after treatment with 2160 g ae ha−1
of glyphosate. Fresh weight samples of 0.25 g were harvested from leaves, chopped and homogenized in 5 mL of 1.25 N HCl solution, and frozen at −80 °C. Samples were kept at room temperature (22 °C) for approximately 15 min, then incubated at 37 °C for 45 min. Subsequently, 125 µL per technical sample (total of five technical samples) was collected and mixed with the reaction buffer (0.25% (w/v
) periodic acid and sodium(meta)periodate solution) and incubated at 37 °C for 30 min. This reaction allowed the oxidation of the shikimic acid. After incubation, an aliquot of 1000 µL of 0.6 N NaOH/0.22 M Na2
was added to the sample. After that, the shikimic acid was measured spectrophotometrically at 380 nm using a cuvette, and the SAC was determined using a standard curve in µg.g−1
fresh weight (µg.g−1
FW). The results were expressed as percentage of SAC in relation to the control.
4.2. RNA-Seq Experimental Design and RNA Extraction
The RNA-seq experimental design included six biological replicates each of the GR and GS biotypes, three with glyphosate treatment (2160 g ae. ha−1
) and three without treatment, giving a total of 12 plants. Three plants of each GR and GS biotypes were treated with glyphosate at 2160 g ae ha−1
, according to the method described above. At 24 h after treatment, the second and third leaves (from the apex) from all treated and non-treated plants were harvested and immediately frozen in liquid nitrogen and stored at −80 °C. Each plant formed an individual sample (Figure 6
The Trizol reagent (Invitrogen, Carlsbad, Calif, USA) was used for the RNA extraction following the company’s recommendation. The residual genomic DNA was removed with DNase I (Invitrogen). The final experimental design comprised 12 RNA samples: 3 GS untreated (GS t0), 3 GR untreated (GR t0), 3 GS treated (GS t1), and 3 GR treated (GR t1) (Figure 6
4.3. cDNA Library Construction and Illumina Sequencing
The cDNA library preparation and Illumina sequencing were performed at the Laboratory of Functional Genomics Applied to Agriculture and Agri-Energy, University of São Paulo (USP), Piracicaba, Brazil. The RIN (RNA integrity number) values and concentration of each sample were examined in the Agilent 2100 Bioanalyzer (Agilent Technologies, USA). A reference cDNA library was constructed using 500 ng of total RNA samples and, the mRNA was enriched and purified according to the Illumina TruSeq Stranded mRNA Sample Preparation Kit (Illumina, San Diego, CA) following the manufacturer’s LT protocol to break it into short fragments with incubating mix for 8 min at 94 °C. The first-strand cDNA was synthesized by adding the Superscript II reverse transcriptase (Invitrogen), followed by thermal cycle incubation at 25 °C for 10 min, 42 °C for 15 min, and 70 °C for 15 min. The cDNA plate (CDP) barcode was removed, and the second-strand synthesis proceeded after addition of the master mix at 16 °C for 60 min. Subsequently, end-repair was performed to remove the 3′ overhangs and proceed immediately to ligate adapters at the 5′ and 3′ ends of each strand in the DNA fragment, which was important for library amplification during cluster formation. The DNA fragments were enriched in a preheated thermal cycler using 1 cycle at 98 °C for 30 s, 15 cycles at 98 °C for 10 s, 60 °C for 30 s, and 72 °C for 30 s, followed by one cycle at 72 °C for 5 min. Finally, the 12 libraries were sequenced using the HiSeq Flow Cell v4, with the Illumina HiSeq 2500, producing 125 bp paired-end reads (2×).
4.4. De Novo Transcriptome Assembly and Functional Annotation
Preprocessing was performed using the raw data to remove low-quality reads (ambiguous sequence ‘N’ or very short sequences), adaptors, and contamination. The preprocessing was performed with the Fastqscreen tool [46
], followed by the checking of sequence quality using the FastQC (https://www.bioinformatics.babraham.ac.uk/
). All raw reads were submitted to the preprocessing to trimming and quality filtering (option
: LEADING:3, TRAILING:3, SLIDING WINDOW:4:15), and to remove the adapter sequences (option
: ILLUMINACLIP 2:40:15) using the Trimmomatic [47
] which were used to reconstruct a full-length transcriptome.
The clean reads were assembled de novo
using Trinity [48
] with the default settings except for the K-mer value (25 mer), and transcripts under 300 bp discarded during assembly. The transcriptome assembly, used for differential gene expression, was performed for all 12 libraries (all samples and treatments), producing a single de novo
transcriptome, which is recommended as an essential step for differential expression analysis [48
]. For EPSPS sequences analysis, a single de novo
transcriptome was assembled for both GR and GS biotypes. Trinity was executed using the default settings, and at the end of the assembly, statistics (e.g., N50, L50, CG%) were calculated using the accessory script “TrinityStats.pl
”. The BioPython package [51
] was used to do additional assembly analysis. After assembling, the transcripts were aligned to the UniProt-trEMBL [52
] database using Diamond [53
Annotation was performed using the Trinotate pipeline, using the Pfam [50
] UniProt-SwissProt database to identify the protein families, and SignalP [54
] and TMHMM [55
] to identify transmembrane proteins and peptides, respectively. The prediction of rRNA transcripts was performed using the RNAmmer [56
]. A BLAST with a significance threshold of E-value ˂10−5
was used to compare all assembled unigenes with the non-redundant proteins from Swiss-Prot, TrEMBL, CDD, Pfam, and KOG databases.
Gene ontology (GO) terms were used to evaluate the functional categories of the best BLASTX hits from the non-redundant protein database with the BLAST2GO software with an E-value threshold of 10−5, grouping by molecular function, biological process, and cellular component. The unigenes were subjected to clusters of orthologous groups for eukaryotic complete genomes (KOG) classification to evaluate the integrity of the transcriptome library and the effectiveness of the annotation process.
4.5. Differential Gene Expression Analysis and Candidate Target Selection
In the differential expression analysis, we assumed and termed each contig as a “gene.” The RNA-Seq data were normalized, and the gene expression determined through transcripts reads per million mapped reads (TPM ≥ 2). For each replicate, an estimation of gene expression was made using the Kallisto method [57
] implemented in the Trinity accessory, which generated the expression matrix. The differential gene expression was assessed in the edgeR mode by processing the expression matrix, GO binning, and enrichment [58
Differentially expressed genes (DEGs) were contrasted within each biotype (t1 vs. t0) using the false discovery rate (FDR) and p
-value threshold set of ≤0.05, and then the lists of all DEGs were exported for each comparison. MA plots were produced within DESeq2, and the DEG lists were filtered to remove genes with log2 fold-change values (log2 FC) less than 2 and higher than −2. In this case, DEGs with a log2 fold-change ≥2 (log2 FC ≥2) were considered up-regulated, while ≤−2 (log2 FC ≤−2) were considered down-regulated. A Venn diagram was produced using all DEGs from the GR and GS biotypes in response to glyphosate treatment [59
]. The up- and down-regulated genes for GR biotype were categorized according to GO functions using the methods described above.
4.6. RNA-Seq Dataset Validation through qRT-PCR
The same RNA samples as were used for RNA-sequencing were used for quantitative reverse transcriptase polymerase chain reaction (qRT-PCR) analysis. The RNA integrity was assessed in agarose gel electrophoresis at 1% (p/v), while the concentration and purity were measured in a NanoDropTM 2000 spectrophotometer (Thermo Scientific). The RNA was converted to cDNA using the SuperScriptTM First-Strand Synthesis System Kit according to the manufacturer’s methodology.
The qRT-PCR analysis was carried out in a Light Cycler 480 Instrument II (96)TM (Roche Applied ScienceTM), using three biological replicates of cDNA. Amplification was performed with 6.25 μL of SYBR Green I Master (Roche Applied Science), 0.5 μM of primers (10 mM), 1 μL of cDNA (0.2 μg), and 4.25 μL ultrapure water, giving a final volume of 12 μL. The qRT-PCR parameters were denaturation cycle at 95 °C for 5 min, followed by 45 cycles at 95 °C for 20 s, 60 °C for 15 s, and 72 °C for 20 s, which was followed by a dissociation curve with denaturation at 95 °C for 5 s, cooling at 70 °C for 1 min, and gradual warming at 0.11 °C to 95 °C and cooling at 40 °C for 30 s. The amplification was verified by the presence of a single peak on the qRT-PCR melting curve and a single band with the expected size in the 2% of agarose gel electrophoresis.
We evaluated the stability of four candidate reference genes previously reported as 18s ribosomal protein (18s) [60
], glyceraldehyde 3-phosphate dehydrogenase (GAPDH) [61
], alpha-tubulin 5 (TUA5) [62
], and eukaryotic elongation factor 1 alpha (eEF1As) [63
] (Table S2
). The stability of the expression of candidate reference genes was evaluated via qRT-PCR for all cDNA treatments (GR and GS with and without application of glyphosate). The cycle threshold (Ct) values were analyzed by RefFinder software [64
], indicating that the 18s and eEF1As were the most stable reference genes across the treatments. Therefore, the average of the results for both 18s and eEF1As was used to normalize the qRT-PCR data. The relative expression was calculated using the delta–delta ct method RQ = 2−(∆∆CT)
]. A total of 13 genes from different levels of expression were randomly selected from the general DEG list to get a range of expression, and qRT-PCR was performed. The primers for the 13 evaluated genes are shown in Table S3
. The Pearson model (r) was used to correlate the qRT-PCR and RNA-seq results. The results of expression between GR and GS biotypes were compared by F-test at p
≤ 0.05 [65
4.7. EPSPS Transcript Sequence Analysis
The sequence of the single-copy EPSPS gene of L. multiflorum
was obtained from GenBank (DQ153168.2) and mapped into the individual transcriptomes from the GR and GS biotypes to identify their respective contigs. The mapping was performed using BLASTn command-line. Additionally, the EPSPS sequences of Arabidopsis thaliana
(GenBank CAA29828.1) and Zea mays
(GenBank AF349754) were used for comparisons. Trinity-assembled EPSPS sequences for LOLMU GR and GS biotypes, and those from L. multiflorum, A. thaliana
, and Z. mays
from GenBank were converted to amino acids. Next, all sequences were aligned into BioEdit (http://www.mbio.ncsu.edu/BioEdit/
) using the ClustalW multiple alignment functions with default settings. The amino acid substitutions at threonine 102 and proline 106 were evaluated because these are commons coding regions conferring resistance to glyphosate in weeds [12
]. The transcription level of the EPSPS contigs was also analyzed to verify the changes in expression in both GR and GS after glyphosate treatment, as well as the number of copies.
4.8. Selection of Differentially Expressed Candidate Genes
The DEGs were selected with a focus on the GR biotype responses to glyphosate treatment after comparisons with the GS biotype. The expression levels of genes with log2 FC ≥2, ≤−2, FDR, and p-values of ≤0.05 were selected for the GR biotype in response to glyphosate treatment, i.e., GR t1 vs. t0. The log2 FC results from all selected DEGs from GR were contrasted with those from GS ((GRt1/GRt0)/(GSt1/GSt0)). DEGs that presented a significant difference by F-test (p ≤ 0.05) between GR and GS biotypes were selected.