Homeolog loss and expression changes in natural populations of the recently and repeatedly formed allotetraploid Tragopogon mirus (Asteraceae)

Background Although polyploidy has long been recognized as a major force in the evolution of plants, most of what we know about the genetic consequences of polyploidy comes from the study of crops and model systems. Furthermore, although many polyploid species have formed repeatedly, patterns of genome evolution and gene expression are largely unknown for natural polyploid populations of independent origin. We therefore examined patterns of loss and expression in duplicate gene pairs (homeologs) in multiple individuals from seven natural populations of independent origin of Tragopogon mirus (Asteraceae), an allopolyploid that formed repeatedly within the last 80 years from the diploids T. dubius and T. porrifolius. Results Using cDNA-AFLPs, we found differential band patterns that could be attributable to gene silencing, novel expression, and/or maternal/paternal effects between T. mirus and its diploid parents. Subsequent cleaved amplified polymorphic sequence (CAPS) analyses of genomic DNA and cDNA revealed that 20 of the 30 genes identified through cDNA-AFLP analysis showed additivity, whereas nine of the 30 exhibited the loss of one parental homeolog in at least one individual. Homeolog loss (versus loss of a restriction site) was confirmed via sequencing. The remaining gene (ADENINE-DNA GLYCOSYLASE) showed ambiguous patterns in T. mirus because of polymorphism in the diploid parent T. dubius. Most (63.6%) of the homeolog loss events were of the T. dubius parental copy. Two genes, NUCLEAR RIBOSOMAL DNA and GLYCERALDEHYDE-3-PHOSPHATE DEHYDROGENASE, showed differential expression of the parental homeologs, with the T. dubius copy silenced in some individuals of T. mirus. Conclusions Genomic and cDNA CAPS analyses indicated that plants representing multiple populations of this young natural allopolyploid have experienced frequent and preferential elimination of homeologous loci. Comparable analyses of synthetic F1 hybrids showed only additivity. These results suggest that loss of homeologs and changes in gene expression are not the immediate result of hybridization, but are processes that occur following polyploidization, occurring during the early (<40) generations of the young polyploid. Both T. mirus and a second recently formed allopolyploid, T. miscellus, exhibit more homeolog losses than gene silencing events. Furthermore, both allotetraploids undergo biased loss of homeologs contributed by their shared diploid parent, T. dubius. Further studies are required to assess whether the results for the 30 genes so far examined are representative of the entire genome.

Several outcomes for duplicated genes are possible at the genomic and transcriptional levels. First, both members of a duplicate gene pair may retain their original function. Second, one copy of a duplicate gene pair may retain the original function, but the other copy may become lost or silenced [3,13,14,[23][24][25][26]. Third, duplicate genes may partition the original gene function (subfunctionalization), with one copy active, for example, in one tissue and the other copy active in another tissue [25,[27][28][29][30][31]. Fourth, one copy may retain the original function, while the other develops a new function (neofunctionalization) [32][33][34][35][36][37][38].
One of the best systems for the study of naturally occurring polyploids is provided by the genus Tragopogon (Asteraceae). Tragopogon comprises ca. 100 to 150 species distributed throughout Europe, temperate Asia, and North Africa [74][75][76]. Three diploid species (T. dubius, T. porrifolius, and T. pratensis) were introduced from Europe into the Palouse region of eastern Washington and adjacent Idaho, USA, in the early 1900s [69,70]. The introduction of these three diploid species brought them into close contact, and as a result, two allotetraploid species (T. mirus and T. miscellus) formed [69]. First collected in 1949 [69], these recently formed polyploids are less than 80 years old. Morphological, cytological, flavonoid, isozymic, and DNA evidence confirmed the ancestries of these two allotetraploids [77][78][79][80][81][82][83]. Multiple lines of evidence suggest that T. miscellus has formed recurrently, possibly as many as 21 times, including reciprocal formation, and T. mirus has formed repeatedly perhaps 13 times (but not reciprocally) [70,84,85]. Therefore, T. mirus and T. miscellus afford unique opportunities for the investigation of recent and recurrent polyploid evolution. In fact, nearly every population of these species may have formed independently (V. Symonds et al., unpublished data).
Tate et al. [26,86] and Buggs et al. [87] studied genomic changes and expression differences of homeologs within natural populations of Tragopogon miscellus, as well as in synthetic F 1 hybrids and first-generation polyploids formed from the diploid parents T. dubius and T. pratensis. Most of the genes analyzed show additivity in T. miscellus at both the genomic (seven out of 23) and cDNA levels (12 out of 17). However, loss of one parental homeolog was observed at several loci (27 out of 46 homeologs), as were several examples of gene silencing (nine out of 34 homeologs). Both homeolog losses and silencing patterns vary among individuals in natural polyploid populations of independent origin [26,87]. Changes were also detected in rDNA content [71] and expression [72] in populations of T. miscellus. Although T. miscellus has fewer rDNA repeats of T. dubius than of T. pratensis [71], apparently due to concerted evolution, most of the rDNA expression derives from the T. dubius repeats [72]. The same pattern of rDNA expression has been observed in populations of T. mirus compared to its parents [71,72]; T. mirus has fewer repeats of T. dubius than of T. porrifolius [71], but most of the rRNA is produced by the T. dubius copies [72]. Although homeolog loss events and expression changes were observed in natural populations of T. miscellus, no such changes were observed in comparable analyses of F 1 hybrids between the diploid parents, T. dubius and T. pratensis [26,87], or in first-generation synthetic lines [87].
In this study we extend our examination of gene loss and differential expression to the polyploid T. mirus. In nature, T. mirus has formed repeatedly, but only when T. dubius is the paternal parent and T. porrifolius is the maternal parent [69,82]. However, T. mirus can be produced synthetically in both directions with about equal frequency [88]. Tragopogon mirus provides an opportunity to compare expression differences at the genomic and transcriptional levels with the results obtained for T. miscellus [26,87]. Our main objectives were to: 1) investigate the genomic changes and expression differences of parental homeologs in T. mirus relative to its diploid parents, 2) determine the identity of the genes that exhibit those changes, and 3) assess whether individuals within and among recurrently formed natural populations of T. mirus show similar patterns of genome evolution and gene expression.

cDNA-AFLP polymorphism and identification of putatively differentially expressed genes
We used cDNA-AFLPs [26,89,90]as a first step toward identifying genes with putative differential expression in the allotetraploid T. mirus, relative to its diploid parents (T. dubius and T. porrifolius). From our initial screen with 37 primer pairs, 1,440 fragments were produced, and of these, 504 were monomorphic (35.0%), and 936 were polymorphic (65.0%) among the three species. Novel cDNA-AFLP bands in the polyploid plants comprised 0.4% (6 fragments) of all fragments, fragments in the polyploids of maternal origin constituted 5.0% (72 fragments) of all fragments, while fragments having a paternal origin in the polyploids made up 3.5% (51 fragments) of all fragments. From this initial screening, we selected for further study 21 of the 37 primer sets, which produced an average of 50 different fragments per primer pair. From the remaining 16 primer sets, we obtained an average of 24 different fragments, but these were too short (below 250 bp) for further analysis. We then conducted an analysis on an expanded sample of the Pullman-1 population and its progenitors (10 individuals of T. mirus, 10 individuals of T. porrifolius, and 6 individuals of T. dubius) to obtain a larger set of potentially informative fragments. From the 21 primer pairs, 1,056 fragments were produced, and of these, 375 were monomorphic (35.5%), and 681 were polymorphic (64.5%) ( Table 1). Novel cDNA-AFLP fragments in the polyploids comprised 0.6% (6 fragments) of all fragments. Shared fragments with a maternal or paternal origin in the polyploids represented 6.3% (67 fragments) or 4.6% (49 fragments) of the total fragments, respectively.
At the Palouse site, 251 fragments were scored, and of these, 79 were monomorphic (31.5%), and 172 were polymorphic (68.5%) ( Table 1). Fragments with a maternal or paternal origin in the polyploids made up 6.8% (17 fragments) and 5.2% (13 fragments) of all fragments, respectively. No novel cDNA-AFLP bands were detected in polyploid plants from Palouse. When we compared 20 individuals of T. mirus from the Pullman-1, Pullman-2, and Palouse populations, we observed very similar patterns in the Pullman-1 and Pullman-2 populations. However, individuals of T. mirus from Palouse have more complex patterns than individuals of T. mirus from Pullman-1 and Pullman-2. Individuals 2602-1 and 2602-3 from Palouse shared an AFLP pattern, whereas 2602-2 and 2602-4 showed a different pattern. There are at least three genotypes among the five individuals of T. mirus from Palouse.
With 125 variable fragments (>350 bp) identified from cDNA-AFLP analyses, we then searched for fragment identity based on sequence similarity using BLAST searches and identified 33 putative genes in T. mirus ( Table 2). Further comparison with the Arabidopsis genome indicated that these genes are involved in various cellular processes, such as carbohydrate metabolism, signal transduction, protein transport and degradation, and cell division (Table 2). However, we could not reliably identify the remainder of the fragments because of their short length (~150 bp).

Rapid loss of parental homeologs
The genes, enzymes, and sizes of digested genomic and cDNA amplifications for CAPS analysis of T. mirus and its parents are listed in Additional file 1.
For the genomic CAPS analysis, 20 of 30 genes showed additivity, with both parental copies maintained in all allopolyploid individuals ( Figure 1). Nine of 30 genes showed that at least one allotetraploid individual was missing one parental homeolog ( Figure 2, Table 3).
To determine whether these losses were due to true homeolog loss or simply loss of a restriction site (due to sequence polymorphism), we sequenced the PCR fragments of all genes exhibiting putative losses. Sequencing revealed that all individuals exhibiting apparent loss events have only one parental homeolog, confirming that these inferred homeolog losses are not due to restriction site divergence and loss of a CAPS marker.
Two genes exhibited homeolog losses of one parental copy or the other in at least one individual, whereas seven genes showed loss of only the T. dubius homeolog.  Table 3). For the putative Nucleic acid binding homolog, eight T. mirus plants from the Pullman-2 (2678-3 and 2678-11), Oakesdale (2673-4), and Arizona (1747-1, 1747-2, 1747-3, 1747-6, 1747-9) populations showed loss of the T. porrifolius band, while one Palouse T. mirus individual (2602-25) lost the T. dubius band. In contrast, preferential loss of the T. dubius parental homeolog was observed in several individuals for seven genes (putatively identified as Myosin heavy chain CLASS xI, LRR protein, Prenyltransferase, NADP/FAD oxidoreductase, Tetratricopeptide repeat protein, RNA binding, and Glyceraldehyde-3-phosphate dehydrogenase). In addition, variation was observed among populations; losses of the T. dubius homeolog occurred at more loci in the Pullman-1, Pullman-2, and Palouse populations than in the Oakesdale, Rosalia, Tekoa, and Arizona populations. For example, individuals from the Palouse population showed loss of the T. dubius homeolog for four genes, while individuals from the Oakesdale population exhibited gene loss for only one gene (Table 3). Therefore, the Pullman-1, Pullman-2, and Palouse populations of T. mirus show higher levels of, and greater variation in, homeolog loss than do populations from Oakesdale, Rosalia, Tekoa, and Arizona. However, one putative gene (Adenine-DNA glycosylase) was polymorphic in both T. mirus and T. dubius ( Figure 3). Most individuals of T. dubius have a single copy of this gene, but five T. dubius individuals have an extra copy that corresponds to the PCR amplicon produced in T. porrifolius. Also, six T. dubius individuals only have the "T. porrifolius" type ( Figure 3). This polymorphism observed in T. dubius can affect interpretation of the expression patterns of T. mirus, making it hard to distinguish loss from polymorphism. As a result of this polymorphism, this gene was not employed in our analyses of loss events.
cDNA CAPS analysis was performed for 15 of 30 genes. The remaining 15 genes analyzed above for genomics CAPS could not be amplified (see Materials and Methods). Eleven of the genes included in the cDNA analyses showed additivity, whereas four genes (putative homologs of Thioredoxin M-type 1, Myosin heavy chain CLASS XI, Nuclear ribosomal DNA, and Glyceraldehyde-3-phosphate dehydrogenase) showed expression differences in some polyploid individuals relative to the diploid parents. However, the apparent expression differences from the Thioredoxin M-type 1 and Myosin heavy chain CLASS XI result from genomic losses (see above; Figure Table 4). In summary, 27 of 40 individuals sampled of T. mirus showed loss of at least one homeolog, and 12 individuals exhibited true loss of expression of one parental homeolog (Tables 3, 4).

Diploid F 1 hybrids are additive of their parental genomes
Genomic CAPS analyses for six synthetic F 1 hybrids from two independent crosses between T. dubius and T. porrifolius were also performed for the same 30 genes surveyed in natural populations of T. mirus. Significantly, all genomic CAPS analyses of F 1 hybrids exhibited additivity of the parental homeologs. cDNA CAPS analysis for all 15 genes investigated in natural populations of T. mirus showed that both parental homeologs were expressed in all F 1 hybrids examined (Additional file 4).

cDNA-AFLP variation in populations of T. mirus
As cDNA-AFLPs reveal potentially differentially expressed genes, the results can provide useful initial information on the genetics of polyploids, especially those that lack developed genomic resources, such as Tragopogon. Thus, cDNA-AFLPs provide numerous candidate genes relatively quickly and inexpensively. However, cDNA-AFLP analysis must be followed by other approaches, such as CAPS analysis, because cDNA-AFLP fragment differences may result from true expression differences, sequence polymorphism, or gene or homeolog loss [26,91].
From cDNA-AFLPs, we identified 33 putative genes that were not additive of the parental bands in T. mirus; most of these (21 out of 33) exhibited maternal banding patterns (Table 2). However, subsequent analysis of 23 of these genes using genomic and cDNA CAPS analyses showed that only four of these genes exhibited an expression difference; two genes, THIOREDOXIN M-TYPE 1 and MYOSIN HEAVY CHAIN CLASS XI, showed homeolog losses, while NUCLEAR RIBOSOMAL DNA and GLYCERALDEHYDE-3-PHOSPHATE DEHY-DROGENASE exhibited true silencing ( Figure 4, Table  4). However, most of the genes exhibiting apparent maternal, paternal, or novel banding patterns in cDNA-AFLPs actually showed additive patterns in genomic and cDNA CAPS analyses, indicating that the cDNA-AFLP fragment differences observed in T. mirus may be derived from sequence polymorphism and are not indicative of homeolog loss or silencing.

Rapid genomic changes in T. mirus
Analyses of CAPS markers provide evidence of rapid, frequent, and preferential elimination of homeologous loci and changes in gene expression in allotetraploid individuals of T. mirus. Most of the genes examined showed genomic CAPS patterns that are additive of the parental genes, but nine genes of the 30 examined showed homeolog loss, fewer than observed for T. miscellus, in which 16 loci of the 23 examined showed homeolog loss [26,86,87].
In T. mirus, the T. dubius parental homeolog has been lost at more loci (12.1%; 7 out of 58 homeologs) than the T. porrifolius homeolog (6.9%; 4 out of 58 homeologs). Likewise, in T. miscellus, T. dubius parental homeolog has been lost at more loci (32.6%; 15 out of 46 homeologs) than the T. pratensis homeolog (26.1%; 12 out of 46 homeologs) [26,86,87]. However, the T. porrifolius homeolog has actually been lost from more individuals of T. mirus (25) than has the T. dubius homeolog (10), in contrast to data for T. miscellus [26,87]. This difference between T. mirus and T. miscellus results from the extensive loss of the T. porrifolius homeolog for NADP/FAD OXIDOREDUCTASE. Furthermore, each homeolog absence in a polyploid should not necessary be viewed as a unique loss, as a single loss may subsequently be transmitted throughout a population.
The preferential loss of T. dubius homeologs also agrees with the biased rDNA homogenization of T. dubius repeats to the other diploid parental type in both T. miscellus and T. mirus [71]. That is, the number of T. dubius rDNA units has typically been greatly reduced in the genomes of the allotetraploids T. mirus and T. miscellus relative to the other diploid parent (either T. porrifolius or T. pratensis) due to apparent concerted evolution [71,72]. The single exception is the Palouse population of T. mirus, in which individuals have a relatively high number of T. dubius rDNA repeats relative to T. porrifolius repeats (compared to the diploid parental plants from the Palouse) [71], illustrating variation in rDNA repeat composition among populations of T. mirus of independent origin. In contrast, allopolyploids in Arabidopsis and Brassica apparently have not undergone loss or concerted evolution of rDNA units [92,93]. However, rapid concerted evolution of rDNA units in just a few generations [94,95] has occurred in some synthetic hybrids and allotetraploids, including synthetic hybrids between maize and Tripsacum [96], somatic hybrids of Medicago sativa [97], synthetic Nicotiana allotetraploids [94,98], and synthetic allotetraploid Arabidopsis suecica [99].
Despite being fewer in number, the rDNA units of T. dubius origin dominate rDNA transcription in most populations of T. mirus [72]. rDNA gene reduction by concerted evolution in allopolyploids can therefore be countered by high levels of expression controlled by epigenetic regulation. We obtained similar results for rDNA expression here in our cDNA CAPS study. Based on visual comparison of banding intensity, there is higher rDNA expression of T. dubius-origin units than of T. porrifolius-origin units in all plants from all populations examined, except for individuals from the Palouse population (Additional file 3). Interestingly, for the Palouse population, cDNA and genomic CAPS indicate the silencing of the T. dubius rDNA unit in one plant of T. mirus. Hence, these results further highlight the importance of populational surveys, by indicating some stochasticity for rDNA expression in the young polyploid T. mirus.
Some synthetic and natural allopolyploids show remarkable genomic restructuring (e.g., Arabidopsis suecica [56], Brassica napus [51,54,55], Nicotiana lines [94,95,100], Primula [101], and wheat [48]). For example, synthetic Brassica napus allopolyploids exhibit many chromosomal translocations and transposition events during the S 2 to S 5 generations, based on RFLP analysis of synthetic lines [51,55]. Genome evolution in the natural polyploids Tragopogon mirus and T. miscellus appears most similar to the results obtained for these synthetic Brassica allopolyploids. Homeolog loss appears frequent in both systems. Recent cytogenetic studies using FISH and GISH indicate that both T. mirus and T. miscellus show evidence of rapid genomic rearrangement, including translocations and inversions [102]. Genetic changes observed in synthetic Brassica napus as well as natural populations of Tragopogon mirus and T. miscullus may be related, in part, to chromosomal rearrangements in these polyploids [51,55,102].
Plants of the synthetic and naturally occurring allopolyploid Arabidopsis suecica also exhibit chromosomal rearrangement [56], as well as many changes in gene expression [40,41]. Genes from one parent, A. thaliana, have often been silenced epigenetically by DNA methylation [41]. Through microarray analysis, Wang et al. [40] showed that approximately 65% of nonadditively expressed genes in the synthetic allotetraploids were repressed, and more than 94% of them matched the genes that are highly expressed in one parent, A. thaliana. Tragopogon allopolyploids have undergone many losses of homeologous loci, often eliminating copies of one parent, T. dubius, but fewer instances of gene silencing. Additional studies of Tragopogon are needed to similarly examine gene expression on a genomic-level scale. F 1 diploid hybrids and early synthetic allotetraploids (S 1 to S 3 generation) between Aegilops sharonensis and Triticum monococcum ssp. aegilopoides showed both gene loss and silencing by DNA methylation [1,47,48,[103][104][105]. However, such immediate changes have not been detected in Tragopogon. Genomic and cDNA CAPS data for synthetic F 1 hybrids between T. dubius and T. porrifolius showed additivity rather than gene loss or silencing (Additional file 4). Similarly, no homeolog loss was observed in synthetic F 1 hybrids between T. dubius and T. pratensis [26], or in newly produced (S 0 ) or first-generation (S 1 ) synthetic T. miscellus [87]. Therefore, in contrast to wheat, both T. mirus and T. miscellus exhibit genome evolution, not immediately following hybridization or allopolyploidization, but apparently shortly thereafter, given that the species are probably less than 80 years old (or 40 generations; these plants are biennials) [69,70]. Also, these results suggest that loss of homeologs and gene expression changes, while still rapid in evolutionary time, may be slightly more gradual in Tragopogon, occurring over several generations, but further studies are required to assess the speed and magnitude with which genomic changes have occurred in Tragopogon.
A major question centers on the mechanisms responsible for the loss of homeologs in these young allopolyploids. Ownbey [69] observed the formation of complex multivalents during meiosis in both T. mirus and T. miscellus shortly after their formation and in F 1 hybrids between T. dubius and T. porrifolius, and between T. dubius and T. pratensis. Furthermore, multivalents have also been observed in synthetic T. mirus and T. miscellus [88,106]. In addition, rare patterns observed in analysis of allozyme variation in Tragopogon are consistent with non-homologous recombination [84]. Non-homologous recombination could provide a mechanism of homeolog loss in T. mirus and T. miscellus, as in Brassica [55]. Recent cytogenetic data provide additional insights into potential mechanisms for gene loss in Tragopogon. GISH studies have revealed that chromosomal rearrangements and other changes may be common in natural populations of T. mirus and T. miscellus [102]. Intergenomic translocations, inversions, as well as apparent monosomy and reciprocal trisomy occur in fertile individuals of both polyploids [102]. Such rearrangements provide a potential mechanism for the homeolog losses observed in both T. mirus and T. miscellus.

Genomic changes versus differential expression in T. mirus
For most of the genes examined here, homeolog losses appear to be responsible for the cDNA-AFLP fragment differences observed in individuals of T. mirus relative to its diploid progenitors. However, in two genes (putatively Nuclear ribosomal DNA and Glyceraldehyde-3phosphate dehydrogenase), we found true expression differences in the allopolyploid relative to its parents. Nuclear ribosomal DNA encodes ribosomal RNA, and Glyceraldehyde-3-phosphate dehydrogenase encodes an enzyme that participates in multiple processes, including transcription activation, initiation of apoptosis, and ER to Golgi vesicle shuttling [107], so both of these genes are crucial for cell function. The pattern of Nuclear ribosomal DNA from genomic and cDNA CAPS analyses is consistent with Matyášek et al.'s [72] study: all individuals from the Palouse population showed additivity with genomic CAPS and in the Southern blot rDNA study of Matyášek et al. [72]. However, in one individual from the Palouse population, the T. dubius Nuclear ribosomal DNA homeolog was completely silenced in both our cDNA CAPS analysis and in the rDNA transcript study of Matyášek et al. [72].
Most of the T. mirus individuals examined here, except those from Tekoa and Arizona, show no expression of the T. dubius homeolog for Glyceraldehyde-3phosphate dehydrogenase (Figure 4). Genomic CAPS data indicate homeolog loss in two individuals, but additivity of parental homeologs in the remaining individuals. cDNA CAPS analyses therefore exhibit gene silencing in 13 individuals (Figure 4). Recent studies have shown that rapid epigenetic gene silencing following allopolypoid formation can be reversed by chemical demethylation in allopolyploid Arabidopsis suecica [41,43,44,108]. Therefore, silencing of these two genes might result from epigenetic phenomena such as DNA methylation or histone acetylation [109,110].
In this study, gene silencing of T. dubius homeologs occurred in only two cases. In addition, expression studies of T. miscellus [26,87] showed that seven out of 17 genes exhibited silencing of the T. dubius homeolog, while for two other genes, the T. pratensis homeolog was silenced. These biased patterns in T. mirus and T. miscellus indicate that T. dubius homeologs might be more susceptible to silencing than the alternative parental homeologs.
When we compare gene silencing with gene loss with respect to the number of individuals examined, the previous studies of T. miscellus [26,87] show that silencing events are slightly more frequent than homeolog loss, while in this study of T. mirus, homeolog losses are slightly more frequent than silencing events. These expression patterns result from biased expression of only a few genes. For example, 14 out of 20 T. miscellus individuals have silencing events in Leucine-rich repeat transmembrane protein kinase, and 11 out of 40 T. mirus individuals have silencing events in glyceraldehyde-3-phosphate dehydrogenase. Therefore, comparing gene loss with silencing events in T. mirus and T. miscellus seems to be affected by specific genes, with a few of the genes examined here especially prone to silencing.
However, when we consider the number of genes examined in this study of T. mirus and in the previous T. miscellus studies [26,86,87], homeolog losses in T. mirus (18.97%; 11 out of 58 homeologs) and in T. miscellus (58.7%; 27 out of 46 homeologs) are more frequent than silencing events (6.7%; two out of 30 homeologs in T. mirus; 26.4%; nine out of 34 homeologs in T. miscellus). Nevertheless, we investigated only a small portion of the genome, so further studies are required to assess whether the results for the~30 genes so far examined are representative of the entire genome.
Genomic CAPS analysis of a putative Adenine DNA glycosylase gene showed polymorphism in the populations of the diploid parent T. dubius surveyed here (Figure 3). Although previous studies in Tragopogon diploids using allozymes and other markers indicated that genetic variation within populations is quite low [83,84], T. dubius is the most genetically variable of the three diploids; allozyme variation within populations of T. porrifolius and T. pratensis was limited or absent. Therefore, polymorphism among T. dubius individuals for one of the genes analyzed here is not surprising. A recent survey of Tragopogon diploids and polyploids from the Palouse using microsatellite markers similarly revealed low levels of genetic variation within populations of T. dubius, but none within either T. pratensis or T. porrifolius (V. Symonds et al., unpublished data).

Conclusions
Recently formed Tragopogon allotetraploids (<80 years; 40 generations for these biennial plants) exhibit various consequences of genome evolution and gene expression following polyploidy. In this study, using cDNA-AFLPs, we found differential banding patterns, possibly attributable to gene silencing, novel expression, and/or maternal/paternal effects between T. mirus and its diploid parents. Most of the banding patterns subsequently investigated with genomic and cDNA CAPS analyses revealed additivity. Most of the differences observed in T. mirus result from homeolog loss, rather than gene silencing; the latter was detected only infrequently (in two genes in some individuals). Genomic and cDNA CAPS analyses indicated that plants of T. mirus have experienced frequent and preferential elimination of the T. dubius homeolog, whereas comparable analyses of synthetic F 1 hybrids between the parents (T. dubius × T. porrifolius) of T. mirus showed only additivity.
These same results were also obtained for the recently and repeatedly formed allotetraploid Tragopogon miscellus [26,87]. Both T. mirus and T. miscellus undergo biased loss of homeologs contributed by their shared diploid parent, T. dubius. Furthermore, both allotetraploids exhibit more homeolog losses than gene silencing in terms of the number of genes undergoing change. Taken together, our results suggest that in Tragopogon loss of homeologs and gene silencing are not immediate consequences of hybridization or polyploidization, but are processes that occur fol-lowing polyploidization, occurring over a relatively small number of generations. These results further support the idea of polyploidy as a dynamic evolutionary process (reviewed in 117), with abundant and rapid genomic changes occurring within a short time period following polyploidization. Further studies of homeolog loss, nonadditive expression patterns, and subfunctionalization of homeologs are needed to explore the roles of genetic and epigenetic phenomena in the evolution of allotetraploid Tragopogon species.

Plant materials
For populations Pullman-1 and -2, Palouse, and Rosalia, seeds were collected from natural populations and grown in the greenhouse at Washington State University (Pullman, WA, USA) and allowed to self-fertilize. Seeds from these greenhouse-grown plants were collected, germinated, and grown under controlled conditions in the greenhouse at the University of Florida (UF; Gainesville, FL, USA). Seeds from the Oakesdale, Tekoa, and Arizona populations were collected and then grown at UF without a round of selfing. Each population of T. mirus is inferred to be of separate origin (V. Symonds et al., unpublished data) and was analyzed, along with samples of the diploid progenitors from each location (Table 5; [70,83,84]). However, only the paternal parent, T. dubius, was investigated for populations from Oakesdale, Rosalia, Tekoa, and Arizona because T. porrifolius was not found at those sites.

cDNA-AFLP display and identification of polymorphic fragments
Here, following Tate et al. [26], we initially employed cDNA-amplified fragment length polymorphisms (cDNA-AFLPs) to identify potentially differentially expressed genes [111]. This approach has proven to be useful in systems without well-developed genetic resources [1,26,43,91,[112][113][114]. However, the weakness of this approach is that fragment differences observed on a cDNA-AFLP gel may result from true expression differences, sequence polymorphisms, or gene or homeolog loss [26,40]. Due to this limitation, we subsequently examined the expression patterns of genes isolated from cDNA-AFLPs using genomic and cDNA CAPS analysis [26,115]. This approach can determine whether apparent expression differences observed at the transcriptional level result from genomic changes such as gene loss or from true differences in expression. Putative homeolog losses were further tested via DNA sequencing.
Leaf segments less than 30 cm in length were collected from young plants six weeks after germination. Due to heavy latex in the leaf tissue, we extracted total RNA from leaf tissue using the method of Kim et al. [116], which combines a CTAB extraction protocol [117] and subsequent use of the RNeasy Plant Mini Kit (Qiagen, Valencia, CA, USA). cDNA was synthesized from total RNA using SuperScript Double-Stranded cDNA Synthesis Kit (Invitrogen, Carlsbad, CA, USA), and the cDNA-AFLP technique was performed as previously described [43], except that we replaced isotope-based signal detection with silver staining [118].
To investigate the utility of cDNA-AFLPs in T. mirus, we conducted an initial survey of the Pullman-1 population with 37 primer combinations. Following the success of this survey, expanded cDNA-AFLP analyses were conducted on 5-10 individuals each from the Pullman-1, Pullman-2, and Palouse populations of T. mirus and from the diploid progenitors that occurred with the tetraploid populations. We employed the same methods, using the primer sets that were the most variable in our initial screen (21 for Pullman-1 and four each for Pullman-2 and Palouse; Additional file 5). We analyzed 56 individuals: 20 of T. mirus, 16 of T. dubius, and 20 of T. porrifolius from the Pullman-1, Pullman-2, and Palouse populations, respectively ( Table 5). The expressed bands on cDNA-AFLP gels were scored as monomorphic (present in all individuals) or polymorphic (present in at least one individual/absent in at least one individual) ( Table 1).
From the expanded cDNA-AFLP work, 125 variable fragments exhibiting novel, maternal/paternal, or other polymorphic patterns were identified from the Pullman-1, Pullman-2, and Palouse populations. To determine the putative identity of these fragments, we excised and sequenced fragments over 250 bp in size, as described in Lee and Chen [43], Wang et al. [91], and Tate et al. [26]. The polymorphic bands were cut from the polyacrylamide gels, and the fragments were re-amplified using the same set of selective amplification primers and cloned using a Topo TA Cloning Kit (Invitrogen). Sequencing was performed with the CEQ DTCS-Quick Start Kit (Beckman Coulter, Fullerton, CA, USA). To identify the sequences obtained above, we used BLAST searches against the NCBI database http://www.ncbi.nlm.nih.gov, and the sequence identity was rechecked against the Compositae Genome Project database http://compgenomics.ucdavis.edu. Identified sequences have been deposited in the EMBL/GenBank database under accession nos (Additional file 6).

CAPS analyses
To determine if cDNA-AFLP fragment polymorphisms resulted from genomic changes or expression differences, we conducted both genomic and cDNA CAPS analyses. In CAPS analyses, amplified PCR products are digested with diagnostic restriction enzymes that distinguish the diploid parental sequences, and the fragments are separated by agarose gel electrophoresis.
For this study we included populations of T. dubius and T. mirus from Oakesdale, Tekoa, and Rosalia, Washington, and from Arizona, in addition to the Pullman-1, Pullman-2, and Palouse populations. Therefore, 100 individuals were examined for genomic CAPS analyses ( Table 5) (Table 5). We also analyzed 6 F 1 hybrid plants using both genomic and cDNA CAPS to determine whether genomic changes and expression differences appear in the first-generation hybrids.
For CAPS analyses, we used 30 primer sets: 23 primer sets were designed based on the sequences that were variable in the cDNA-AFLP analysis of the Pullman-1, Pullman-2, and Palouse populations, and 7 primer sets were from studies of T. miscellus [26,87]. To design primer sets, we first BLASTed our fragments against Lactuca or Helianthus ESTs and then used the "hit" ESTs for primer design because the ESTs are likely longer than the isolated fragments, with a greater chance that the expressed sequence spans the introns in genomic DNA. With those primer sets, we amplified fragments from T. dubius and T. porrifolius and then confirmed their sequences. From those sequences, we redesigned the primer sets to be more specific for analysis of Tragopogon CAPS. All primers were designed using the web interface program, Primer3 (v. 0.4.0; http://frodo.wi.mit. edu/). Primer sequence information is given in Additional file 6.
For each of the 30 gene regions, we aligned DNA sequences from the diploid parents using Sequencher v. 4.1.4 (Gene Codes Corporation, Ann Arbor, MI, USA) and identified diagnostic restriction sites between the species.

Genomic CAPS analyses
We isolated genomic DNA from 100 individuals of the allotetraploid and its two progenitors using a modified CTAB protocol [117]. Genomic fragments were amplified in a 25 μl volume with 20 ng template, 5× buffer, 1.5 mM MgCl 2 , 0.2 mM dNTPs, 0.1 mM each primer, and 0.5 unit Taq polymerase (Promega, Madison, WI, USA). Thermal cycling conditions were as follows: 95°C for 2 min, followed by 35 cycles of 95°C for 30 sec, 54-56°C for 30 sec, 72°C for 1 min 20 sec, and a final 7min extension at 72°C. Products were separated on a 1.5% agarose gel, stained with ethidium bromide, and visualized using a UV transilluminator. Genomic digests were performed in a 12 μl volume, containing 1× buffer, 3 μl PCR product, 2 units of restriction enzyme (New England Biolabs, Ipswich, MA, USA), and 100 μg/ml bovine serum albumin (when required), and incubated at the appropriate temperature for 9 hr. Digested products were separated on 2-4% agarose gels, stained with ethidium bromide or SyberGold (Invitrogen), and visualized using a transilluminator.
To determine whether putative homeolog losses observed in genomic CAPS analyses were due to true homeolog loss or simply a loss of a restriction site (due to sequence polymorphism in one parental fragment in the polyploid), we sequenced the initial PCR product. A homeolog loss would yield only one parental diploid sequence, whereas loss of a restriction site in one parent would still yield both parental DNA sequences.

cDNA CAPS analyses
We isolated total RNAs from 96 of the individuals analyzed for genomic CAPS (see above) using the method of Kim et al. [116] and the RNeasy Plant Mini Kit (Qiagen). The first-strand cDNA was synthesized with 5 μg total RNA using Superscript II reverse transcriptase (Invitrogen) with a poly-T (T17) primer. Using the same primer sets as for 15 of the loci in the genomic CAPS analyses, RT-PCR was carried out using 50 ng of template from the first-strand cDNA in a 25 μl volume with 20 ng template, 5× buffer, 1.5 mM MgCl 2 , 0.2 mM dNTPs, 0.1 mM each primer, and 0.5 unit Taq polymerase (Promega). The remaining 15 genes could not be amplified from these cDNAs; the reasons for this are unclear, but perhaps these genes were not expressed in the leaf tissue sampled for cDNA CAPS. For amplification of fragments and digestion of RT-PCR products, we employed basically the same approach as described for the genomic CAPS analyses. In addition, for NUCLEAR RIBOSOMAL DNA, the relative PCR band intensities of the two homeologs were measured using KODAK 1D Image Analysis Software (Kodak, Rochester, NY, USA).