C. canadensis accessions were collected from middle Tennessee, USA [24]. The collected seeds were grown in potting media at the University of Tennessee, Knoxville, TN, USA, under a 16/8 h of light/dark photoperiod at 25 °C. Glyphosate-resistant plants were screened by spraying glyphosate (Roundup WeatherMax; 0.84 kg acid equivalent ha⁻¹; Monsanto, St. Louis, MO, USA). Glyphosate-susceptible (TNS) or -resistant (TNR) C. canadensis of the Tennessee biotype were watered and fertilized with Osmocote slow-release fertilizer. For glyphosate treatment, three plants from each of the six TNR or TNS biological replicates were grown in 1 L pots until rosette leaves reached a diameter of 6 to 8 cm (approximately 3-month-old plants). Rosette-stage plants were treated by spraying glyphosate. For the non-glyphosate treated (water-sprayed) TNS and TNR plants, three plants from each of the seven TNR, or eight TNS biological replicates, were used as experiment control samples. The aboveground biomass of the glyphosate-sprayed and water-sprayed TNS and TNR was collected 24 h after treatment, and quickly frozen in liquid nitrogen. Samples were stored at −80 °C before RNA extraction.

The frozen leaves were ground by mortar and pestle. Total RNA was extracted using TriReagent following the manufacturer’s protocol (Invitrogen, ThermoFisher Scientific, Carlsbad, CA, USA). Extracted RNA was characterized for purity and concentration with a Nanodrop spectrophotometer (ThermoFisher Scientific). The RNA quality was assessed again by RNA chip analysis on an Agilent 2100 Bioanalyzer system (Agilent, Santa Santa Clara, CA, USA). All RNA samples were treated with DNase I (Invitrogen, ThermoFisher Scientific) to remove residual genomic DNA.

The Illumina libraries were prepared by using an RNA-Seq prep kit (Illumina Inc., San Diego, CA USA) according to the manufacturer’s protocol. A total of 27 libraries were constructed from total RNA from 7 TNR (TNRC) water-sprayed leaves, as well as those from glyphosate-sprayed 6 TNR (TNRG), water-sprayed 8 TNS (TNSC), and glyphosate-sprayed 6 TNS (TNSG) leaves. Each library was ligated with a unique index barcode. All cDNA libraries were sequenced by using a HiSeq2000 (Illumina Inc.) at Oak Ridge National Laboratory. The generated sequence reads were single-ended and 100 bp in length.

The read quality control (QC) of raw reads was performed by MultiQC [25]. The adaptor sequence and low-quality reads were removed by Sickle (https://github.com/najoshi/sickle; accessed on 10 March 2020). The trimmed sequences were subjected to de novo transcriptome analysis using Trinity software [26]. The whole transcriptome reference sequence was assembled from TNRC, TNRG, TNSC, and TNSG raw reads by the Trinity assembly function. The assembled transcript contigs were functionally annotated from five different databases (nr, nt, Pfam, Uniprot, and eggnog) by the Trinotate function. NCBI accession identities were then fetched by efetch distributed with BlastX [27].

The qualified reads were mapped and annotated against the whole transcriptome reference and protein annotations by the Trinity quantification script, using default parameters. Under this script, Bowtie2 was performed to align reads, and RSEM (RNA-Seq by Expectation-Maximization) was used for the quantification with maximum likelihood estimates [28,29]. The read matrices of gene-level and transcript-isoform-level data were generated by our own script for downstream analysis.

Principal component analysis (PCA) was performed and plotted by the prcomp and ggplot packages in R [30].

The gene-level transcript read count matrix from the RSEM output was submitted to edgeR for DEG analysis [31]. A total of 24 transcript quantities of comparable combinations between glyphosate-treated TNR or TNS biotypes, and water-sprayed TNR or TNS, were subjected to DEG analysis. In addition, water-sprayed TNR and TNS were analyzed following the same process. The genes were filtered using the false discovery rate (FDR) criterion for each analysis. Statistically, the transcripts of up- or downregulation over 4 folds (2 of log2) with significant FDR (<0.05) were taken as differentially expressed genes.

Total RNA samples were obtained by the RNA-extraction protocol as described above. One microgram of total RNA was reverse transcribed by the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific). One μL of three-times diluted cDNA was added to 15 μL of 1 × PowerUp SYBR Green Master Mix, as well as a pair of primers, as listed in Table S1 (Applied Biosystems, Waltham, MA, USA). After preheating at 50 °C for 2 min, and predenaturation at 95 °C for 10 min, two-step protocol sequencing at 95 °C for 15 s, and at 60 °C for 30 s, were performed. The real-time PCR reaction was operated by the QuantiStudio™ 6 Flex real-time PCR system (Thermo Fisher Scientific). The relative gene expression change was determined by the 2^−ΔΔCt equation [32]. Statistical analysis was performed with the relative fold change of three technical replicates from three different horseweed plants by a Student’s t-test.

To find the transcript contigs of the ABC transporters in whole transcript assemblies, we downloaded the transcript sequence from the reference [21]. We aligned the sequences to a fasta file of the whole transcript assembly using Minimap2 [33]. Other transcript contigs of interest were also aligned with transcript sequence information from the NCBI database by the same procedure [27]. To verify the gene identities of those not assigned a protein function, we translated the open reading frame in the transcript contigs using the amino acid sequence translation tool at the ExPaSy website (https://web.expasy.org/translate/; accessed on 11 May 2020). The longest open reading frame was submitted to the blastp suite in NCBI (https://blast.ncbi.nlm.nih.gov/Blast.cgi; accessed on 11 May 2020) to identify the similar protein sequence in the protein database.

The correlation efficiency was calculated using the Pearson method in R. All plots were generated by ggplot2 in R [30].

On the basis of the annotation file information, the NCBI protein accession number was matched for each gene. GO terms were assigned using the accession numbers provided by DAVID [34], with 24 significantly overrepresented GO terms based upon Arabidopsis thaliana’s definition (p < 0.05).

Twenty-seven cDNA libraries were generated from TNR and TNS leaves 24 h after glyphosate or water treatment, and then submitted for high-throughput Illumina sequencing (Figure S1). A total of 100.2 million reads were sequenced (Table 1). The sequenced data were adapter-trimmed and quality-filtered. Through read quality control, we excluded three raw sequencing replicates because of their low recovery rates. The remaining high-quality reads from 24 libraries (6 TNRC; 5 TNRG; 8 TNSC; 5 TNSG) were used for de novo transcriptome analysis and downstream DEG analysis.

As the first step for de novo transcriptome analysis, whole transcripts were assembled to be used for the reference transcript by Trinity. A total of 77,072 gene-level contigs were assembled (105,354 contigs were assembled at isoform level). The N50 of the assembled transcript contigs was 1780 base pair (bp), with an average contig length of 1039 bp (Table 1). The genes predicted among the assembled transcript contigs were annotated by a blast search against a protein database. Of these contigs, 32,493 (63,249 isoform contigs) were functionally annotated and uniquely assigned to an NCBI protein accession number by the BlastX database (E-value < 0.05; Table S2). The generated whole transcript assembly and its annotation was used as a reference for further DEG analysis. The raw reads and the fasta file of all assembled transcript contigs were submitted at the publicly open database (sequence read archive number: PRJNA764068).

To conduct the DEG analysis in TNR and TNS with respect to glyphosate treatment, we aligned each group of reads against the whole transcript assembly by the Bowtie2 option in the Trinity software. The aligned reads were quantified to each transcript by RSEM. Real read counts were compiled for four different treatment groups through this process. PCA with four different group alignments showed that most of the variance (91%) among the gene-level read count was on the x-axis (Figure 1a). Glyphosate-treated groups were separated from water-sprayed TNRC and TNSC. Along the y-axis (14% of the variance), the gene expression across the different biotypes of TNR and TNS is clearly separated. The PCA test with the transcript isoform-level read counts also showed similar clustering patterns to that of the gene-level read counts (Figure 1a). This result indicates that many genes were differentially expressed concerning glyphosate treatment and the different biotypes, even within the same genotype. The gene expression of randomly selected DEG from RNA-seq data were validated by RT-qPCR with the contig specific primers (Table S3). For the following analysis, we used gene-level read counts.

Both the TNR and TNS biotypes had many DEGs at 24 h after glyphosate treatment. In the TNR biotype, a total of 23,371 genes were differentially expressed depending upon glyphosate treatment (FDR < 0.05; Table S4); 3640 genes were upregulated (>4 fold); and 3852 genes were downregulated (<−4 fold) in TNRG compared to TNRC. The top 20 of the most differentially expressed genes by glyphosate treatment are listed in Table 2 and Table 3.

In the TNS biotype, a total of 23,663 genes were differentially expressed depending on glyphosate treatment (Table S5). Among them, 3715 genes were upregulated (>4 fold), and 2411 genes were downregulated (<−4) in TNSG (FDR < 0.05). The top 20 of the most differentially expressed genes by glyphosate treatment are listed in Table 4 and Table 5. When we compared the fold change (FC) of each gene in the analysis of TNRG vs. TNRC and TNSG vs. TNSC, the FC value was correlated between each analysis (r = 0.93; Figure 1b). Therefore, the glyphosate response of most expressed genes was very consistent between TNR and TNS.

On the basis of the GO term of each protein accession number, we predicted the biological processes of differentially expressed genes. The proportion of gene counts in DEG involved in the same GO term in TNRG vs. TNRC, and TNSG vs. TNSC, was different (Figure 1c). Interestingly, the biological process of transport functions, such as drug transmembrane transport and transmembrane transport, were predominantly up- or downregulated in the TNRG group compared to TNSG, indicating that the transmembrane transport is a typical biological process in TNR after glyphosate treatment. The genes relating to the transcription process were also expressed differentially in TNRG. A larger number of genes relating to signal transduction, protein phosphorylation, and the oxidation-reduction process were down- and upregulated in TNRG. In contrast, the gene relating to protein phosphorylatioin and the oxidation-reduction process was expressed more in TNSG, but the proportion in DEGs was relatively smaller than in TNR.

When we assessed the correlation of DEGs between TNRG vs. TNRC, and TNSG vs. TNSC, the relationships were largely sorted into three zones (Figure 2a). Eight modes of gene count distribution in each treatment were included in the three zones (Figure 2b). The majority of DEGs were sorted in zone 1, with a similar reponse in the TNR and the TNS (Figure 2a). The genes whose read counts were distributed, as shown in the (i) and (ii) modes, were involved in this zone, accompanying similar gene regulation patterns in TNR and TNS after glyphosate treatment (Figure 2b). The genes in these modes may not be specifically regulated in both biotypes by different glyphosate responses. On the other hand, the genes in zone 2 and zone 3 (six types of expression modes) represented the overexpressed genes in TNRG vs. TNRC, but not in TNSG vs. TNSC, and those in TNSG vs. TNSC, but not in TNRG vs. TNRC, repectively. We focused on these groups because these group genes have a higher potential to act specifically on the glyphosate response in glyphosate-resistant plants. In addition, the DEG results in the present study analyzed by edgeR did not include the absolutely expressed genes in the glyphosate-treated samples (0 counts in non-treated samples), but these absolutely expressed genes were included in zones 2 and 3 as well. We could not detect the significant gene showing its read count distribution as (iv) and (viii) that the glyphosate response was reversed in TNR and TNS. In contrast, the present DEG analysis observed the genes in the (iii), (v), (vi), and (vii) modes that should be expressed differentially in different biotypes under glyphosate treatment. To find the dominant functional gene unique to glyphosate treatment, we first analyzed absolutely expressed genes in either biotype by glyphosate treatment, and then collected the genes, including these four modes (iii, v, vi, and vii).

A total of 517 genes were upregulated in TNRG (FC > 4), while the corresponding read counts in TNRC, TNSG, and TNSC were near zero (mode vi). However, many transcripts in this mode were expressed with small read counts (less than an average of 20 read counts) so we filtered out these genes. A total of 33 genes remained, of which 15 genes had corresponding protein accession numbers (Table S6). Their read count matrix and protein ids are listed in Table 6. Trinity_DN2344_c1_g1 had the highest read counts in this mode, but its protein function is not identified even though its translated protein had a real open-reading-frame sequence (Table 6). Interestingly, 10 genes in this group were predicted as coding for receptors or transporters on the membrane structure, wherein several were lectin-domain-containing or leucine-rich repeated kinase. The real read count profiles of representative receptor kinases are shown in Figure 3, which is representative of the distribution data as shown in mode vi.

Reversely, 26 genes were significantly downregulated in TNRG (FC > 4) but not in TNSG. However, protein id was assigned for only 13 genes (Table 7 and Figure 4). The read count profile of these genes is shown as mode iii, which was sorted in zone 2 (Figure 2b). In this group, we could not find a conserved functional protein family.

We also found that 12 genes were expressed in TNS but not in TNR after glyphosate treatment (mode v in Figure 2b). Six genes among these had a functional protein id (Figure S2). While these genes were coincidentally expressed in TNRG, their read counts were small enough to ignore. We were unable to find absolutely downregulated genes in TNSG (mode vii in Figure 2b). Taken together, absolutely expressed genes existed in both biotypes depending on glyphosate treatment.

Even though the majority of the genomic sequences of TNR and TNS are identical, precise sequence variation still exists. The sequence variation, which included the promoter and coding regions, was identified in the whole genome sequencing of another ecotype of C. canadensis [23,24]. In a DEG study of the type presented here—between TNR and TNS—it is important to consider the expression of candidate resistance genes under glyphosate-treated and untreated conditions. First, we tried to find differentially expressed genes in TNR and TNS under normal conditions by analyzing the DEGs in TNRC and TNSC. In the results, a total of 10,629 genes were differentially expressed between the two biotypes in normal conditions (FDR < 0.05); 640 genes were expressed higher (FC > 4), and 960 genes were expressed lower (FC < −4) in TNRC compared to TNSC (Tables S6 and S7). When we filtered out low read counts (<20 read average), 127 and 181 genes remained in the up- and downregulated groups, respectively (Tables S7 and S8). Among these, we were able to predict the protein function of 202 genes. Two CYP 450 proteins, an ABC transporter C family member, and seven transmembrane protein kinase genes, including those encoding leucine-rich- and lectin-domain-containing receptor proteins were detected in this list.

To examine the glyphosate response of these 308 differentially expressed genes between TNRC vs. TNSC, we assessed FC in the DEG results from TNRG vs. TNRC, and TNSG vs. TNSC. A total of 140 genes were coincidently expressed in both the DEG analysis results, whose FCs were correlated in TNR and TNS (r = 0.72; Figure 5a). Of these genes, the top 10 of the most differently expressed in TNR are listed in Table 8. TRINITY_DN4487_c0_g1 had the highest expression in TNRC compared to TNSC. However, its gene expression was not altered by glyphosate treatment in both biotypes (Table 8). Interestingly, TRINITY_DN535_c0_gl, a CYP450 protein-encoding gene was originally expressed 5.1 (log2)-fold higher in TNR than TNS. This transcript expression was boosted in TNR after glyphosate treatment, but not in TNS (Table 8 and Figure 5b). TRINITY_DN9714_c0_g1, a COMT-type methyltransferase, was also highly induced in TNR compared to TNS in the glyphosate treatment groups (Figure 5b). Therefore, of the genes originally with higher expression in the TNR compared to TNS, several genes were more highly induced in TNR by glyphosate treatment than in TNS. Even though we observed many genes that were expressed only in TNS, most of them were annotated as uncharacterized genes (Table S7).

Regarding the well-known NTSR candidates, we compiled read counts from the DEG analysis result of the ABC transporters, CYP450, GST, and GTs, which are regarded as important NTSR gene families in preventing herbicide damage in weeds.

A total of 134 unique ABC transporter genes (296 at the transcript isoform-level) were assigned in the present TNR and TNS transcriptomes. Of these transcripts, 79 ABC transporter transcripts were differentially expressed in TNR after glyphosate treatment (FDR < 0.05, FC > 4 or <−4; Table S9). In TNS, 90 ABC transporter transcripts were differentially expressed depending on the glyphosate treatment. (FDR < 0.05, FC > 4 or <−4). The FC of ABC transporter gene expression was correlated in both biotypes (r = 0.97, Figure 6a), indicating that most ABC transporters of TNR and TNS were similarly regulated in response to glyphosate.

To compare the previously proposed 17 ABC transporter transcripts with the present transcriptome, each of the reported ABC transporter sequences were aligned independently to the present whole transcript reference [21]. Fifteen transcript sequences were collected from the published data. Of those 15, 12 were identified in our transcriptome (Table S10). However, the M1, M4, and P3 contigs were missing in the present transcriptome. In addition, M3, P4, and P5 were aligned together on a long transcript contig of Trinity DN77_c0_g1 in the whole transcript reference, indicating that these three contigs may be isoforms or fragments of an ABC transporter gene. As identified in Peng and Hereward’s independent studies [13,21], M10 and M11 read counts in TNRG and TNSG were higher than other ABC transporter genes (Figure 6b). In the present study, M3 and M7 transcripts were also upregulated with higher read counts (over 2000 reads) under glyphosate treatment. In contrast, M5 and P2 were downregulated severely in the TNR after glyphosate treatment.

A total of 273 gene contigs were annotated as CYP450 transcripts in the whole transcript reference (455 contigs at transcript isoform-level). Among them, 163 genes were expressed differentially in TNR and TNS after glyphosate treatment (Table S11). The FC of the CYP450 genes in TNRG vs. TNRC was correlated with that of TNSG vs. TNSC (r = 0.95), meaning that most genes in the CYP450 family were regulated similarly in response to glyphosate in TNR and TNS, as observed in ABC transporter gene expression. As we observed in Figure 5b, one CYP450 annotation gene (TRINITY_DN535_c0_gl) was expressed with a higher read count in TNR depending on glyphosate.

Nine GST and 16 GT genes were differentially expressed in both biotypes by glyphosate treatment. The responsive gene number in these families shows a drastic reduction when compared to the previous NTSG target. As previously described, NTSG genes, and the FC and read count distribution of GST and GT genes were similar in both biotypes after glyphosate treatment (Tables S12 and S13).

Therefore, four representative NTSG gene family members were differentially expressed depending on glyphosate, but most FC of their expression did not differ in both TNR and TNS in the present analysis.