Brachypodium sylvaticum (Huds.) P. Beauv. (slender falsebrome; Poaceae), with an estimated genome size of 470 Mb and 17 chromosomes (Foote et al., 2004), is a perennial bunchgrass native to Europe, Asia, and North Africa and is closely related to the bioenergy feedstock model grass B. distachyon (L.) P. Beauv. (Wolny et al., 2011), which has a sequenced genome of 272 Mb and five chromosomes. In its native range, B. sylvaticum occurs in habitats ranging from forest understory to open meadows and tolerates conditions from full shade to full sun (Holten, 1980; Long, 1989; Aarrestad, 2000; Kirby and Thomas, 2000). In the United States, B. sylvaticum is invasive and listed as a noxious weed covering the west coast of California, Oregon, and Washington (Oregon Department of Agriculture, 2009; Washington State Department of Agriculture, 2009; Lionakis Meyer and Effenberger, 2010). It is also expanding into the eastern United States where it has been reported in Missouri and Virginia (Roy, 2010). In Oregon, B. sylvaticum forms thick monocultures in open forests at elevations from nearly sea level to approximately 1200 m. It threatens the endangered Oregon Willamette Valley oak savanna ecosystem by replacing the native flora and reduces habitat for rare butterflies (Kaye and Blakeley-Smith, 2006; Severns and Warren, 2008). False brome is shade tolerant (Murchie and Horton, 1997; Holmes et al., 2010), which makes it a particularly dangerous invasive threat to undisturbed habitats (Martin et al., 2009).
False brome was first introduced into Oregon via plant introduction studies in the early part of the twentieth century, and later identified and collected from the wild in 1939 (Chambers, 1966; Kaye and Blakeley-Smith, 2006). Field trials for exotic grasses were performed by the United States Department of Agriculture (USDA) in Corvallis at the Oregon State University facilities, and B. sylvaticum was widely planted to “improve range” throughout the western United States (Hull, 1974). Oregon State University herbarium records indicate that two separate experimental gardens were established in Eugene and Corvallis, Oregon. Genetic profiling with microsatellite markers confirm these introductions were independent and that they probably consisted of the same set of multiple accessions from the native range in Europe that had been collected by the USDA Division of Plant Introduction (Rosenthal et al., 2008). Accessions in each of these two plantings have crossed, and the invasive plants that are now spreading across Oregon forests are recombinant products of hybridization. Brachypodium sylvaticum has thus become increasingly common in the past 15 yr (Chambers, 1966; Kaye and Blakeley-Smith, 2006; Rosenthal et al., 2008). The introductions in Corvallis and Eugene retain unique marker signatures, but Bayesian cluster analyses indicate that similar sets of native accessions from Western Europe contributed to the hybrid genotypes that are spreading from each introduction location (Rosenthal et al., 2008).
Despite the economic and environmental impact of B. sylvaticum, there remains a lack of adequate genomic resources for the study of the invasive populations. To thoroughly investigate genetic differences between invasive and native populations, a reference transcriptome specific for B. sylvaticum is needed to precisely align, map, and interrogate gene sequences. The major aim of this study was to assemble, annotate, and characterize a high-quality reference transcriptome that will enable researchers to assess gene expression levels, conduct comparative analyses, and identify putative single-nucleotide polymorphism (SNP) and simple sequence repeat (SSR) sequence sites in the genomes of B. sylvaticum populations for developing genetic markers to be used in future genotyping, identification, and genetic tracking studies.
Over the past several years, next-generation sequencing (NGS) has emerged as a low-cost, large-scale, fast, and accurate approach for de novo transcriptome sequencing (reviewed in Ward et al., 2012). Moreover, the tremendous depth of coverage generated by the Illumina sequencing platform in particular enables marker and gene discovery, comparative genomics, and gene expression analysis in nonmodel organisms (Wang et al., 2010; Huang et al., 2012; Nicolai et al., 2012; Varshney et al., 2012; Wang et al., 2012; Zhao et al., 2012). We conducted RNA-Seq transcriptome assemblies on B. sylvaticum plants originating west of Corvallis, Oregon (hereafter referred to as Corvallis, or Brasy-Cor), Spain (Brasy-Esp), and Greece (Brasy-Gre). We generated >36 Gb of B. sylvaticum transcriptome and assembled the sequences into 120,091, 104,095, and 177,095 transcript contigs with an average length of 1652, 1728, and 1566 bp for samples from Corvallis, Spain, and Greece, respectively. The cDNA sequence files are provided in fasta format for each population in Appendices S1 (Brasy-Cor), S2 (Brasy-Esp), and S3 (Brasy-Gre). We estimate that these transcriptomes represent >90% of the B. sylvaticum gene space. Furthermore, we identified SNP and SSR sequences that could be used to design genetic markers for use in future population studies. Along with the substantial genomic resources available in a close congener (B. distachyon; Mur et al., 2011) providing reference and comparative material, this transcriptome assembly will be useful on a broad scale as a greatly needed resource for ecologists and geneticists conducting research on native and invasive populations of B. sylvaticum, and on the role of climate change and adaptation toward successful invasiveness.
Statistics of data sets from the sequencing and de novo transcriptome assemblies of the Brachypodium sylvaticum samples from Corvallis (Brasy-Cor), Spain (Brasy-Esp), and Greece (Brasy-Gre) and their comparison with the transcriptome of B. distachyon Bd21 sequenced genome.
METHODS AND RESULTS
Transcriptome sequencing and de novo assembly—Populations from Spain and Greece were selected to investigate correlations between the Oregon, USA samples and European progenitors used in the USDA field trials. Our samples were drawn from populations in Corvallis, Oregon (population OR-C1; Rosenthal et al., 2008; GPS coordinates: 44°39′35″N, 124°45′ 41″W), Avila, Spain (population SAV, USDA accession PI 318962; GPS coordinates: 40°39′27″N, 5°18′38″W), and Thessaloniki, Greece (population GRE, USDA accession PI 206546; GPS coordinates: 40°37′48″N, 22°57′36″E). The OR-C1 plants were selected from field-collected seed, and the GRE and SAV seed samples were collected and maintained by the USDA Plant Germplasm division in Pullman, Washington (USA). All plants were grown in a common greenhouse garden at the Portland State University campus in Portland, Oregon, under 12 hours light at 25°C and 12 hours dark at 15°C. At 60 wk, leaf tissue was collected from two individuals per population and ground in liquid nitrogen. Total cellular RNA was extracted using a modified protocol described elsewhere (Fox et al., 2009). In brief, RNA was extracted using RNA Plant Reagent (Life Technologies, Grand Island, New York, USA) and treated with RNase-free Turbo DNase (Life Technologies). Concentration, integrity, and extent of contamination by ribosomal RNA were assessed using an ND-1000 spectrophotometer (Thermo Fisher Scientific, Waltham, Massachusetts, USA) and Bioanalyzer 2100 (Agilent Technologies, Santa Clara, California, USA). Samples were prepared using the TruSeq RNA Sample Preparation Kit (v2) and sequenced on the Illumina HiSeq 2000 instrument (Illumina, San Diego, California, USA). We generated >358 million 101-bp paired-end reads with a fragment size of ∼325 bp. This represents an overall total of 36.19 Gb of 101-bp paired-end B. sylvaticum transcriptome sequence. The reads from each sample were indexed for use in preliminary expression analyses and represent between 47 and 79 million reads per sample (Table 1). A total of 12.1 Gb paired-end sequences were used to assemble the Corvallis reference transcriptome, and 10.9 and 12.8 Gb were used for the Greece and Spain references, respectively.
The raw Illumina reads were processed for quality and parsed for index sequences and pairs using custom Perl scripts prior to assembly. The metrics used to assess transcriptome assembly quality included the overall number (coverage) of contigs, the average length of contigs, and the diversity of contigs (the estimated number of discrete loci assembled). We compared de novo assemblies using Velvet/Oases and Trinity algorithms, two published software programs built specifically to assemble de novo transcriptomes from short-read sequence data (Grabherr et al., 2011; Schulz et al., 2012). After evaluating the performance of the Velvet/Oases and Trinity algorithms, we conducted our final assembly using Velvet/Oases ( Appendix S4 (fox_et_al_appendix_s04.pdf)). Our assembly generated 120,091, 104,095, and 177,682 contigs for the Corvallis, Spain, and Greece assemblies (Table 1; Appendices S1, S2, and S3). In all three B. sylvaticum assemblies, ∼96% of assembled contigs were longer than 250 bp, with the longest contig for each of the three assemblies being greater than 15,000 bp. The average contig length was 1652, 1728, and 1566 bp for samples from Corvallis, Spain, and Greece, respectively, while the average length of the reference B. distachyon Bd21 gene models (v1.2) is 1086 bp. The median for each assembly, while lower, was near the average, indicating that the assemblies were not composed of an over-representative set of small contigs. Furthermore, the frequency distribution of contig scaffold sizes shows the majority of the lengths near the median and the overall frequency distribution similar to that observed in B. distachyon (Fig. 1A). However, when we compared the number of assembled contigs with the average length, we observed many more contigs in our de novo assemblies than the number of gene models in B. distachyon, and the average contig lengths in B. sylvaticum were greater than that of B. distachyon as well (Fig. 1B). The greater number of contigs in the B. sylvaticum assemblies could result from several factors, including gene splicing isoforms, gene paralogs, and mis-assemblies. As discussed below, the vast majority of assembled contigs are well-annotated, indicating that the inflated number of contigs are not due to mis-assemblies but more likely due to transcript isoforms and/or paralogous loci.
BLAST comparisons of Brachypodium sylvaticum transcriptomes against various databases. Shown are the number and percentage of contigs from B. sylvaticum samples from Corvallis (Brasy-Cor), Spain (Brasy-Esp), and Greece (Brasy-Gre) that hit a gene from the respective database. BLASTx comparisons were made to the GenBank nonredundant peptide database (nr), while BLASTn nucleotide comparisons were made against the B. distachyon, Oryza sativa, and Sorghum bicolor cDNA databases (E-value threshold cutoff of 1e-10-5).
Homology-based transcriptome annotation—We examined the number of homologous loci identified through BLAST to approximate the diversity and coverage of the assembled gene loci (Mount, 2007). We annotated genes based on BLAST similarity (E-value threshold cutoff of 1e-10-5) to sequences available in the GenBank protein database and directly to transcripts identified in the sequenced monocot genomes. We used our transcript scaffolds as queries in BLASTx to search against the National Center for Biotechnology Information (NCBI) nonredundant protein database (Table 2), which resulted in >80% correlation between the contigs of each B. sylvaticum assembly and GenBank peptide sequences.
Similar trends were observed when we compared the B. sylvaticum nucleotide sequences directly against other monocot nucleotide sequences using BLASTn (E-value threshold cutoff of 1e-10-5 and percent identity >90%). The B. distachyon Bd21 (v1.2), Oryza sativa (Japonica, MSU6), and Sorghum bicolor (v1.4) transcriptome databases were obtained from Gramene BioMart (Spooner et al., 2012). In the Corvallis assembly, 103,752 (86.4%) contigs hit B. distachyon genes, 96,375 (80.3%) contigs hit rice genes, and 93,751 (78.1%) contigs hit S. bicolor genes, with similar results for the Spain and Greece assemblies (Table 2). We further examined the number of homologous loci identified through direct comparisons between B. sylvaticum and B. distachyon (Fig. 2). We found that the Corvallis assembly uncovered 28,791 (92.8%) of the 31,029 B. distachyon v1.2 transcripts, while the Spain and Greece assemblies hit 91.3% and 94.1% of the B. distachyon transcripts, respectively (Fig. 2A). When we compared the B. distachyon genes hit by all three B. sylvaticum assemblies, we found that ∼96.8% (30,054) of the B. distachyon genes were hit by a contig from at least one of the three B. sylvaticum transcriptomes. In addition, 27,187 (87.6%) B. distachyon genes were commonly hit by all three B. sylvaticum transcriptome assemblies.
Further, when we extended our comparisons with other sequenced grass genomes, we found that the Corvallis contigs hit 44,538 (∼67%) of 66,338 rice transcripts and 24,999 (∼84%) of 29,448 sorghum transcripts with similar results for the other two B. sylvaticum assemblies. Overall, ∼90% of all the B. sylvaticum contigs were assigned a homology-based annotation. While the properties of any transcriptome are uniquely associated with the spatial, temporal, and environmental factors present at the precise time of tissue sampling, these results indicate that we have sequenced the great majority of B. sylvaticum gene loci and have assembled three quality reference transcriptomes for B. sylvaticum.
Functional characterization of the transcriptome—All the B. sylvaticum contigs were translated into peptides by querying the longest predicted open reading frame (ORF) using the ORFPredictor tool (Min et al., 2005) and were functionally characterized using InterProScan version 4.8 (Quevillon et al., 2005; Hunter et al., 2012). Nearly half of the translated ORFs were assigned InterPro and Gene Ontology (GO) annotations ( Appendix S5 (fox_et_al_appendix_s05.xlsx)), which is consistent with other published annotated genomes. The functional annotations were enriched by performing Blast2GO analysis by adopting a stringent BLASTx search (E-value ≤le-20 and percent identity ≥90%) against the NCBI GenBank nonredundant protein database (Conesa and Gotz, 2008), and the resulting best hits with GO annotations were used to project GO assignments to B. sylvaticum contigs (Götz et al., 2008; Barrell et al., 2009). Enrichment resulted in assigning GO annotations to 69,628 Corvallis, 61,015 Spain, and 107,700 Greece contig assemblies ( Appendix S6 (fox_et_al_appendix_s06.xlsx)).
Gene expression analysis—We conducted relative gene expression analyses to assess the utility of these reference transcriptomes for future differential gene expression studies. We mapped B. sylvaticum Illumina reads to B. distachyon transcripts and calculated the reads per kilobase per million reads (RPKM) values. We then mapped these RPKM values to B. distachyon loci on all five chromosomes and graphically depict the RPKM values averaged over 0.5 megabase intervals (Fig. 2B). When we compared the RPKM values among the three B. sylvaticum plants, we found very high Pearson's correlation coefficients between gene expression data sets ( Appendix S7 (fox_et_al_appendix_s07.pdf)). Although the three B. sylvaticum plants do not exhibit significant expression differences over our mapped intervals, we do observe a uniform distribution of contigs mapping to the B. distachyon genome (Fig. 2B; B. distachyon v1.0 genome sequence from http://mips.helmholtz-muenchen.de/plant/brachypodium/). To further demonstrate the utility of pairing the B. sylvaticum transcriptomes and their close congener to investigate questions regarding gene expression, we used the BrachyCyc pathway tool ( http://pathway.gramene.org/gramene/brachycyc.shtml), which contains biochemical pathways consisting of over 7000 B. distachyon genes coding for enzymes. Using the BrachyCyc pathway tool, we mapped RPKM values to B. distachyon metabolic pathways to examine gene expression profiles of homologous genes ( Appendix S7 (fox_et_al_appendix_s07.pdf)). These results demonstrate the potential of the B. sylvaticum transcriptomes for use in future studies investigating differential gene expression and metabolomics, as well as for making comparisons with B. distachyon.
Genetic variation—To quantify the number of SNP sites across the three transcriptomes, we mapped the reads from each B. sylvaticum sample to the B. distachyon genome v1.0 using Bowtie version 0.12.8 (Langmead et al., 2009). We then used custom Perl scripts to identify nucleotide differences in positions with at least eight aligned reads and 75% of those aligned reads confirming the SNP (Kimbrel, unpublished). Using these criteria, we identified 394,654 putative SNPs among the three B. sylvaticum genotypes (Fig. 3A; Appendix S8 (fox_et_al_appendix_s08.xlsx)). Of these, 157,835 SNPs were in sites common to all three genotypes. Although the total number of SNPs was similar among the genotypes, we observed more SNPs unique to the Greece sample (66,963) when compared to Corvallis (39,936) and Spain (40,027). To address the biological relevance of these SNPs and their potential role to be studied in the future for relevance to invasive phenotypes, we predicted the potential effects of the variants and identified a diverse set of consequences (McLaren et al., 2010). Notably, we identified more than 230,000 downstream variants, more than 92,000 missense variants, and 234 stop codons introduced (Tables 3,4). We observed only slight variation when we mapped the SNP densities of each B. sylvaticum genotype onto B. distachyon chromosomes. Generally, the B. sylvaticum SNP density mirrors the B. distachyon gene density and centromeric region (Fig. 3B). We also constructed a maximum parsimony tree based on concatenated variant sites where at least five reads from each B. sylvaticum transcriptome aligned with Bowtie (Fig. 3C). Of the 157,835 SNPs, only 628 loci were polymorphic within at least one of the three B. sylvaticum samples. These 628 positions were concatenated to generate the maximum likelihood tree depicting relative relationships among the three B. sylvaticum genotypes. While this SNP analysis shows the utility of the B. sylvaticum transcriptomes for genotyping studies, much work needs to be done to fully elucidate the relationships of the various native and invasive populations.
We mined the assembled B. sylvaticum contigs for SSRs using Perl code from the Simple Sequence Repeat Identification Tool (SSRIT; Temnykh et al., 2001; http://www.gramene.org/db/markers/ssrtool), looking for di-, tri-, tetra-, penta-, and hexanucleotide SSRs with a minimum of nine, six, six, five, and five repeat units, respectively (Table 5; Appendix S9 (fox_et_al_appendix_s09.xlsx)). In total, we identified 23,535 SSRs in Corvallis, 20,303 in Spain, and 32,847 in Greece transcriptome sequences (Table 5). These SSRs were identified in 18,281 contigs from Corvallis, 15,975 from Spain, and 25,567 from Greece. To test whether SSRs showing polymorphism in our computational analysis can be used to develop potential genetic markers, we conducted a test PCR amplification of a sample SSR site from the three sequenced genotypes as well as additional B. sylvaticum plants from Oregon, USA, and Europe ( Appendix S10 (fox_et_al_appendix_s10.pdf)).
Brachypodium sylvaticum resources and data download—All sequence, annotations, and data files are available from the project website ( http://jaiswallab.cgrb.oregonstate.edu/genomics/brasy). SNPs and contig alignments to B. distachyon are also available from the B. distachyon Bd21 genome browsers available from BrachyBase ( http://www.brachypodium.org/) and Gramene ( http://www.gramene.org/Brachypodium_distachyon/) databases. The raw sequence data are available from the NCBI Sequence Read Archive (SRA; accession SRA062855).
The list of noxious invasive plants identified by the Oregon Department of Agriculture (2009) includes species such as kudzu, goatgrass, knapweed, and others that exact high economic and ecological costs in regions where they have been introduced. Ecologists make distinctions among species that are introduced (able to persist), naturalized (establishes self-sustaining populations, but is not a dominant component of the vegetation), and invasive (is able to dominate habitats to the exclusion of native species). False brome meets the “invasive” criteria but has not yet become notorious because its distribution is still restricted compared to other invasive plants. Our ultimate goal is to use B. sylvaticum as a model for studying adaptation and invasiveness and for the general study of grasses. Therefore, we consider it a necessary first step to establish baseline resources for B. sylvaticum by generating de novo transcriptomes from multiple genotypes, use them to study gene expression and regulation, and identify functional nucleotide polymorphisms to develop new sets of genetic markers for future population-wide screening. The results of our de novo assemblies produced a relatively large number of long, reconstructed transcripts, as demonstrated by average contig lengths. Overall, we were able to assign homology-based annotations to ∼90% of B. sylvaticum contigs, and more than 50% of the translated sequences were functionally annotated by assigning InterPro and Gene Ontology annotations. More than 96% of B. distachyon Bd21 gene loci were associated with B. sylvaticum contigs, thereby demonstrating diversity and broad coverage in our transcriptomic data. When compared to B. distachyon, we discovered ∼390,000 SNPs, of which ∼158,000 SNPs were common to the three B. sylvaticum samples. Based on the SNP calls, we identified the number of SNPs with consequences to the gene and transcripts including those altering the potential intron splicing sites and translated protein sequences. These resources, when paired with well-established B. distachyon genomic data, will be useful in the future characterization of B. sylvaticum invasiveness.
SNP consequence predictions. After aligning all of the 394,654 SNPs from the Brachypodium sylvaticum samples from Corvallis (Brasy-Cor), Spain (Brasy-Esp), and Greece (Brasy-Gre) on the B. distachyon (Bd21) genome v1.0, we predicted the potential effects of SNP variant loci on the Bd21 genes. We identified a diverse set of consequences for the various gene loci.
Characterization of the number of transitions and transversions that were predicted from our SNP analysis, showing a much higher prevalence of transitions over transversions as expected. The SNPs were identified in the Brachypodium sylvaticum samples from Corvallis (Brasy-Cor), Spain (Brasy-Esp), and Greece (Brasy-Gre).
Summary of SSR sites identified in the transcriptomes of Brachypodium sylvaticum samples from Corvallis (Brasy-Cor), Spain (Brasy-Esp), and Greece (Brasy-Gre). We identified di-, tri-, tetra-, penta-, and hexanucleotide SSRs with a minimum of nine, six, six, five, and five repeat units, respectively. More than 20,000 SSRs were identified in the B. sylvaticum transcriptomes; trinucleotide repeats were the largest class of SSRs.
 The authors thank members of the Center for Genome Research and Biocomputing at Oregon State University for sequencing and computational support, Tanya Cheeke for suggestions on the manuscript, and Cathy Gresham and Fiona McCarthy for InterProScan analysis. Thanks to Sharon Wei and Doreen Ware from Gramene database, and Henry Priest and Todd Mockler from BrachyBase for help with the integration of data sets in the respective databases. Funded by laboratory startup funds provided to PJ.