Introduction

Plant biomass has recently been promoted as a source of renewable feedstock for the conversion to liquid transportation fuels [13, 15, 22, 29]. Plant cell walls can be biochemically or thermochemically deconstructed into the primary subcomponents (cellulose, hemicellulose, and lignin) necessary for this conversion [14, 19, 20, 26]. The carbohydrate fractions are used as feedstocks for sugar and ultimately ethanol production, and lignins are typically separated and used in combustion processes to fuel the reactions. The resistance of lignin, an amorphic polymer, to separate from the carbohydrate fractions during the deconstruction phase has made lignin a target for overcoming recalcitrance [12].

Lignin, a complex polyphenolic polymer, is one of the most abundant polymers on earth. Lignin content of the cell wall influences the cell rigidity, drought tolerance, and insect and disease resistance [7]. The biochemical pathway for lignin biosynthesis is fairly well characterized and involves approximately 12–15 enzyme-regulated steps controlling the conversion of single aldehyde to syringyl and guaiacyl precursors [5, 37]. Lignin content varies across the tissue types and organs of a plant with developmental age and environmental interactions [32]. These responses are genetically controlled and heritability for lignin is moderately high [20]. The rate limiting/critical steps in lignin formation are not yet fully determined though several studies have used reverse genetic approaches and expression analysis to modify and/or characterize lignin composition in transgenic plant materials [9, 10, 23, 24].

Lignin and other cell wall traits display a pattern of continuous phenotypic distribution rather than discrete, Mendelian distribution. Such traits are typically polygenic in nature and are influenced by the environment in which they occur. Genetic mapping can be used to compare the inheritance pattern of a trait and establish the chromosomal regions associated with such phenotypes. These chromosomal intervals may encompass one or more genes responsible for the trait and are known as quantitative trait loci (QTLs).

After the identification of QTL intervals, filtering the list of genes down to a subset of likely candidates is a difficult task. The length of the QTL intervals may be in mega base pairs (Mbp) and include hundreds of genes. One approach to reducing the number of candidate genes is to conduct further experiments using larger numbers of segregating progeny to reduce the QTL interval. Then, classical methods such as positional cloning [25, 27] and insertional mutagenesis [3, 30] can be used to identify influential genes. A complementary approach would be to use the bioinformatics tools and genome information to assign genes in the QTL interval to bins of higher probability than other candidates. This gives a smaller number of candidate genes that can be verified using transgenesis.

The recent availability of several draft and fully sequenced plant genomes have shed light on the evolutionary history of genome structure, and the role whole-genome duplication events have played in determining genome structure and gene family evolution. It is becoming apparent that nearly all plant genomes have experienced at least one whole-genome duplication event [18, 39]. These events have influenced gene family evolution and created opportunities for paralogous genes to experience neo-functionalization and/or sub-functionalization within all gene families [8, 28, 36].

The Populus genome contains three whole-genome duplication events [35]. The most recent, the Salicoid duplication, is found only in members of the Salicoid family and is present in approximately 8,000 paralogous gene pairs. The second duplication event, shared by Populus and Arabidopsis, is found in 3,500 paralogous gene pairs in Populus. In addition, the molecular clock in Populus is ticking a rate that is six times slower than in Arabidopsis, creating a duplicated molecular preservation of the ancestral genome within the extant Populus genome [35]. Together these genomic features can complicate genome assembly, annotation as well as map-based cloning of individual gene(s) responsible for specific phenotypes.

We use a combination of traditional QTL mapping, comparative intragenomic analysis, estimates of gene divergence, and differential expression evidence to identify regions of the Populus genome that contain genes controlling lignin content in shoots and roots and demonstrate that this combinational approach can be used to filter a candidate gene list to a substantially smaller subset of genes within a fixed confidence interval.

Materials and Methods

Description of QTLs

An F2 inbred interspecific hybrid poplar family was used to create a comprehensive genetic map containing 848 markers based on 293 segregating progeny as described by Yin et al. [40]. The overall observed genetic length was 1,927.6 cM. Phenotypic data was collected for all progeny using pyMBMS to obtain estimates of root lignin, root S/G ratio, stem lignin content, and stem S/G ratio [11, 32, 33]. MapQTL 5.0 was used to detect the underlying QTLs [38]. The establishment of genetic map, phenotying of lignin content, and S/G ratio of mapping individuals have been described by Yin et al. (in review).

Assigning Physical Position to the SSR Markers

Populus genome sequence, gene models, and functional categories for genes were downloaded from the JGI Populus Genome portal (http://genome.jgi-psf.org/Poptr1_1/Poptr1_1.home.html). The SSR primer resource, available at the International Poplar Genome Consortium website (http://www.ornl.gov/sci/ipgc/ssr_resource.htm; [34, 41]), was used to obtain the sequence information for the SSR primers and predicted SSR length. The physical position of each interval in the Populus genome was assigned based on BLAST results of the SSR primer nucleotide sequence against the genomic sequence. Additionally, the number of base pairs between the start of the left primer and end of right primer (according to BLAST result) had to be equivalent to the predicted length of the SSR marker. Perl script was used to automate this process. In total, 210 markers were successfully assigned physical position in the genome.

Assigning Physical Position to the QTLs

Assigning QTLs to physical positions in the genome was challenging as linear relationship between the physical and genetic maps vary by position within the genome due to non-homogeneous distribution of chiasma across the genome. Thus, SSR markers flanking the QTLs were initially identified based on genetic positions. The relationship between genetic and physical distance for each QTL was then obtained by the ratio of physical distance between the markers and the genetic distance between them. This relationship was used to obtain the physical coordinates of the QTL in the genome by subtracting the difference between the ends of the QTL and the flanking marker.

Identification of Duplicated Genes Corresponding to the QTL Interval

Around 8,000 pairs of paralogous genes of similar age (excluding tandem duplications) were identified in the Populus genome. All genes in each QTL interval were identified based on the position of genes on each chromosome/linkage group (LG). The duplicated interval and corresponding duplicated gene information were then identified. Next, percent identity between a gene and its paralog was calculated using BLAST to align each pair. Finally, the best match Arabidopsis genes were identified by reciprocal blasting BLAST of the Arabidopsis gene set (TAIR Version 9) and Populus gene set to identify the top pair in each case.

Data Mining of Microarray Expression Profiles of Genes and Duplicated Genes

Populus balsamifera Affymetrix microarray datasets containing developmental tissue series (GSE13990 series) in GEO database at NCBI were used to examine the transcriptome level attributes of roots and differentiating xylem. We used this dataset to identify differences in gene expression between root and stem. The 50,848 probe sets with genome match correspond to 40,236 unique JGI Populus trichocarpa gene models. Cross-hybridizing and redundant probes for gene models as well as probes for alternatively spliced version of genes were eliminated in the analysis.

Identification of Differentially Expressed Genes

We used the RankProd package [16] to analyze the expression array data to identify differentially expressed genes. RankProd utilizes a rank product non-parametric method [6] to identify up- or down-regulated genes under differential conditions, e.g., two treatments, two tissue types, etc. The false discovery rate (FDR) value obtained was based on 10,000 random permutations [16]. The genes that had FDR values less than or equal to 0.10 were considered as differentially expressed.

Results

QTL Intervals

The QTL intervals for lignin and S/G ratio are located on seven linkage groups in the Populus genome (Fig. 1). The genetic position of each QTL interval is shown in Table 1. The QTL intervals for root lignin content were observed on LG II, LG VI, LG X, and LG XIV; for stem lignin content on LG II, LG VI, and LG X; for root S/G ratio on LG III, LG VI, LG X, and LG XIII; and for stem S/G ratio on LG II, LG III, LG VI, LG XIV, and LG VIII. These QTL intervals generally do not overlap, except on LG VI where QTL intervals for root and stem lignin content and root and stem S/G ratio co-localize and on LG X where QTL intervals for root and stem lignin content co-localize (Fig.1). The length of the QTL intervals ranged from 0.4 to 11 Mbp (Table 2); the majority of the QTL intervals were less than 2 Mbp in length. Correspondingly, the number of genes in the QTL intervals ranged from 44 to 1,501. The total number of genes in all the intervals was 4,530 (Supplementary Table 1). As there were two regions of overlapping QTLs, some genes were common to those QTLs, and the number of unique genes from all the QTL intervals was 3,788.

Fig. 1
figure 1

Nineteen QTL interval intervals distributed on seven linkage groups (LG II, LG III, LG VI, LG VIII, LG X, LG XIII, and LG XIV) in Populus. RL root lignin QTL intervals, RSG root S/G ratio, SL stem lignin content, SSG stem S/G ratio

Full size image
Table 1 Location of QTL intervals based on genetic map
Full size table
Table 2 Details of QTL in terms of physical distances
Full size table

Duplication in Populus Genome and Duplicated Regions in QTL Interval

The Populus genome has undergone a recent genome-wide duplication event that has resulted in a conserved linear order of most of the genes within the duplicated chromosomal segments. QTLs on LG II have duplicated intervals on LG V; QTLs on LG III have duplicated intervals on LG V and scaffold_29; QTLs on LG VI have duplicated intervals on LG XVI and LG XVIII; QTLs on LG X have duplicated intervals on LG VIII (Fig. 2). Some intervals had higher numbers of genes conserved in the duplicated region as compared to the others. Across all intervals, on average, more than 53% of genes had retained a paralog in the duplicated interval and ranged from 25% to 80% (Table 3).

Fig. 2
figure 2

QTL interval and display of duplication in genes in the interval for a root lignin content-1, b root S/G ratio-1, c stem lignin content-3, and d stem S/G ratio. Each blue line represents a gene and its paralog in the duplicated region

Full size image
Table 3 Details of duplication and differences in expression
Full size table

Comparison of Expression of Genes that Lie in the QTL Interval and Their Paralogs

Based on microarray evidence, 13 out of 19 QTL intervals were tissue specific, i.e., the QTL intervals corresponding to lignin content or S/G ratio were unique for either root or stem. Four of these QTL intervals, root lignin content (RL-1), stem lignin content (SL-3), root S/G ratio (RSG-1), and stem S/G ratio (SSG-5), were selected for a detailed analysis because they occurred in paralogous regions and contained differential tissue data from microarray experiments.

In order to use the above data to filter the candidate gene list within each of the selected QTL intervals, three alternative approaches were used to integrate duplication information and differential expression of paralogs (Fig. 3). First, filtering was based on non-duplicated genes within a QTL and those which have higher expression in tissue related to the QTL. Second, differential microarray results were used to identify genes within the interval with expression evidence in the identified tissue whose paralogous genes did not display expression in the corresponding tissue. For example, we identified genes, present in QTL interval for stem lignin content, that show higher expression in xylem relative to root as compared to gene expression of paralogs. In addition, the genes within the QTL interval that have a predicted role in cell wall biosynthesis (e.g., PAL, 4CL, etc.) were promoted to the candidate gene list.

Fig. 3
figure 3

Process of filtering genes. SL-3 stem lignin QTL interval

Full size image

Genes in the QTL Interval

Root Lignin Content

The total number of genes in the RL-1 interval on LG II was 138. Out of these, 94 had paralogs and 44 genes did not retain paralogs in the duplicated interval (Table 3). Two of these non-duplicated genes with higher expression in root were calmodulin (eugene3.00020820) and a signal transduction response regulator gene (gw1.II.42.1; Table 4). Five duplicated genes with <90% similarity and higher expression in root were replication factor (estExt_fgenesh4_pg.C_LG_II0728), an auxin response factor (fgenesh4_pg.C_LG_II000830), a glycosyl hydrolase hydrolyzing o-glycosyl compound (fgenesh4_pg.C_LG_II000867), a zinc finger transcription factor (grail3.0003072701), and an exoribonuclease (gw1.II.1849.1; Table 5). Four duplicated genes with >90% similarity and higher expression in root were sulfate transporter (eugene3.00020855), alcohol dehydrogenase (fgenesh4_pg.C_LG_II000742), a NAC domain protein (grail3.0003068301), and a nodulin-like protein (gw1.II.1386.1; Table 6). In total, the number of genes within the interval was filtered down from 138 equally likely candidates to 11 with supportive duplication and/or expression evidence.

Table 4 List of genes that did not have a paralogous gene model in duplicated region and that showed higher expression in the tissue related to the QTL
Full size table
Table 5 List of genes that have a paralogous gene model; and the % identity is less than 90%, and that showed higher expression in the tissue related to the QTL
Full size table
Table 6 List of genes that have a paralogous gene model; and the % identity is greater than 90%, and that showed higher expression in the tissue related to the QTL
Full size table

Stem Lignin Content

The total number of genes in the SL-3 interval on LG VI was 247. Out of these, 171 had paralogs and 76 genes did not retain paralogs in the duplicated interval (Table 3). Eight of these non-duplicated genes with higher expression in xylem were hypothetical protein (estExt_fgenesh4_pm.C_LG_VI0468), proteins with no known function and unique to Populus (eugene3.00061181, fgenesh4_pg.C_LG_VI001243), a nucleic acid binding (eugene3.00061373), protein associated with CCR4 transcription complex (grail3.0030015901), a plastocyanin-like domain-containing protein (gw1.VI.2580.1), a ribosomal protein (gw1.VI.2649.1), and a peptidyl-prolyl cis–trans isomerase, cyclophilin type protein (gw1.VI.847.1; Table 4). Six duplicated genes with <90% similarity and higher expression in xylem were kinesin protein involved in microtubule-based movement (estExt_fgenesh4_pm.C_LG_VI0481), a hypothetical protein (estExt_fgenesh4_pm.C_LG_VI0500), a senescence associated protein (estExt_Genewise1_v1.C_LG_VI2154) and unknown proteins (grail3.0030003201, grail3.0030006902), and a nodulin-like protein (gw1.VI.781.1; Table 5). Three duplicated genes with >90% similarity and higher expression in xylem were protein with hydrolase activity (estExt_fgenesh4_pg.C_LG_VI1102), an expressed protein with no known function (eugene3.00061209), and a UDP-d-glucuronate 4-epimerase involved in nucleotide sugar interconversion pathway (eugene3.00061339; Table 6). In total, the number of genes within the interval was filtered down from 247 equally likely candidates to 17 with supportive duplication and/or expression evidence.

Root S/G ratio

The total number of genes in the RSG-1 interval on LG III was 278. Out of these, 170 had paralogs and 108 genes did not retain paralogs in the duplicated interval (Table 3). Eight of these non-duplicated genes with higher expression in root were oligopeptide transporter (estExt_fgenesh4_pg.C_LG_III0677), a RNA helicase protein (estExt_Genewise1_v1.C_LG_III1770, grail3.0018015801), a peroxidase (eugene3.00030584), a formin-like protein (eugene3.00030600), an expressed protein (fgenesh4_pg.C_LG_III000886), a DNA J protein (gw1.III.1044.1), and a transporter-like protein (gw1.III.2608.1; Table 4). Four duplicated genes with <90% similarity and higher expression in root were calcium transporting ATPase (fgenesh4_pg.C_LG_III000669), a glucosyl transferase (grail3.0018007101), a tetratricopeptide-containing protein (grail3.0018018701), and a proline-rich protein (gw1.III.1613.1; Table 5). One duplicated gene with >90% similarity and higher expression in root was WRKY family transcription factor (fgenesh4_pg.C_LG_III000900; Table 6). In total, the number of genes within the interval was filtered down from 278 equally likely candidates to 13 with supportive duplication and/or expression evidence.

Stem S/G ratio

The total number of genes in the SSG-5 interval on LG VIII was 226. Out of these, 155 had paralogs and 71 genes did not retain paralogs in the duplicated interval (Table 3). 2Six of these non-duplicated genes with higher expression in xylem were unknown protein (eugene3.00080178), a protein unique to Populus (eugene3.00080195, eugene3.00080273), an ankyrin repeat family involved in transmembrane transport (eugene3.00080203), a cytochrome P450 protein (fgenesh4_pg.C_LG_VIII000212), and a vacuolar protein (gw1.VIII.950.1; Table 4). Ten duplicated genes with <90% similarity and higher expression in xylem were germin-like protein (eugene3.00080177), an acyl-CoA-binding family protein (eugene3.00080251), a GATA zinc finger protein (eugene3.00080330), unknown proteins (fgenesh4_pg.C_LG_VIII000250, fgenesh4_pg.C_LG_VIII000264, fgenesh4_pm.C_LG_VIII000069), a pumilio-family RNA-binding protein (fgenesh4_pm.C_LG_VIII000111), a chorismate synthase (grail3.0049006403), a pectate lyase-like protein (gw1.VIII.1321.1), and an acyl-CoA-binding family protein (gw1.VIII.1497.1; Table 5). Seven duplicated gene with >90% similarity and higher expression in xylem were expressed protein (estExt_fgenesh4_pg.C_LG_VIII0179), a glucosyl transferase also annotated as cellulose synthase-like (estExt_fgenesh4_pm.C_LG_VIII0087), a vacuolar protein (eugene3.00080299), pleckstrin homology domain-containing protein (eugene3.00080329), a CCAAT-box-binding transcription factor (grail3.0049010802), a photoreceptor-interacting protein (gw1.VIII.1083.1), and an exostosin family protein also annotated as glucoside transferase 47 (gw1.VIII.2327.1; Table 6). In total, the number of genes within the interval was filtered down from 226 equally likely candidates to 23 with supportive duplication and/or expression evidence.

Discussion

Whole-genome duplication events, followed by extensive genome reorganization, chromosomal rearrangements, and gene loss, have been widespread during the evolution of plants [31]. As a consequence of duplication, paralogs created in the genome may have one of several possible fates, including non-functionalization, neo-functionalization, and sub-functionalization [18]. The most acknowledged mode is non-functionalization where one of the copies loses function or is silenced, resulting in a pseudogene [1]. The process of neo-functionalization, where one ancestral copy retains its function and the other is free to accumulate mutations, results in acquisition of novel function. During the process of sub-functionalization the sister copies, i.e. paralogs, show different but overlapping functions [18]. Here, some duplicated genes show differential expression among organs within a single plant. In recent allopolyploidization in cotton some genes were silenced in one organ with respect to another. Similar outcomes were detected in artificial allopolyploidization [2]. In Arabidopsis there is evidence of sub-functionalization where clusters of duplicated genes show evidence of concerted divergence in their expression in an organ-specific expression [4].

The Populus genome has undergone multiple genome-wide duplications [35]. The Salicoid duplication currently contains around 8,000 pairs of genes that are syntenous across mega-base regions of the genome. As a result, almost every segment in the Populus genome has a parallel paralogous interval somewhere else in the genome. Yet, QTL intervals for many stem and root lignin and S/G ratio phenotypes are present in only one position (Fig. 1). This suggests that different sets of genes to root and stem QTLs providing an opportunity to leverage the segmental duplication information. Along with gene expression of paralogous gene intervals, higher likelihood values can be assigned to genes or gene sets that are functionally related to the measured phenotype.

This expansive gene declaration results in each QTL interval having hundreds to thousands of genes. Our filtering approach led to a reduced set of genes, most not previously reported to play a direct role in monolignol biosynthesis. These filtered genes included regulatory proteins that may have roles in cell wall formation, vascular transport, and unknown function. For example, in the root lignin interval, a NAC domain transcription factor (grail3.0003068301) is present, and NAC domain transcription factors have been implicated as key regulator of secondary cell wall synthesis in Arabidopsis [43]. A signal transduction response regulator (gw1.II.42.1) was also identified in this interval and is an ideal candidate for further transgenic work as is kinesin (estExt_fgenesh4_pm.C_LG_VI0481), which is involved in the oriented deposition of cellulose microfibrils in Arabidopsis [42]. In the root S/G ratio interval two proteins seem very promising. One is glycosyl hydrolase (fgenesh4_pg.C_LG_II000867) and the other is UDP-d-glucuronate 4-epimerase (eugene3.00061339) involved in nucleotide sugar interconversion pathways. In the stem S/G content interval a cytochrome P450 (fgenesh4_pg.C_LG_VIII000212) is a good candidate for further experimental work. Exostosin gene (gw1.VIII.2327.1) has also been shown to be more highly co-regulated with cellulose synthase genes in Arabidopsis [21].

The filtered gene set provides a feasible opportunity to determine gene function via functional genetics work. That is, based on the computational approach described above, a set of 15–20 candidate genes can be used in RNAi knockdown experiments, mutant complementation experiments, and in association genetics studies correlating single nucleotide polymorphisms frequency and measured phenotypes. An integrated approach that combines QTL mapping with fine-scale mapping using association mapping would require investigating SNPs associated with trait using SNP arrays [17]. Multiple SNPs need to be assayed per gene as linkage disequilibrium decays rapidly in Populus. The filtration strategy discussed in this paper can be used to select candidate genes to assay SNPs and uncover the underlying DNA polymorphism associated with lignin content and lignin S/G ratio.

The unique approach of filtering for genes based on duplication evidence and expression data of paralogs has its limitations. The assembly of the Populus genome is still in a draft state and has numerous captured gaps where the length of the missing fragment is known and non-captured gaps where the length of the missing fragment is not known. Moreover, the lack of microarray datasets for Populus compared to Arabidopsis is also a limiting factor. As more microarray datasets become available, more robust statistical analyses will be feasible. The design of Affymetrix microarray adds to the challenge. Due to the overlapping nature of the Affymetrix probe sets it is frequently difficult to distinguish paralogs. Due to the lack of the microarray datasets, we based our analysis on microarray datasets from P. balsamifera on Affymetrix chips, whereas the QTL intervals were obtained from P. trichocarpa. Future studies of this nature should use the expression data from individuals with extreme phenotypes in the population used to detect QTL.

Conclusions

This paper provides a computational approach for integrating QTLs with expression data and Populus genome duplication information to assign higher likelihood values to candidate genes with greater precision than other. The analytical approach was successful in identifying both genes of suspected cell wall biosynthetic function as well as genes of putative cell wall biosynthetic function. Genes of unknown or putative functions would most likely not have been examined without such an approach. In total, the list of genes in QTL intervals was reduced from hundreds or thousands of genes to 15–20 genes. These results provide a roadmap for future experimental work attempting to discover cell wall recalcitrance genes and the ultimate utility of plant biomass as an energy feedstock.