Microarray analysis reveals differential gene expression in hybrid sunflower species

Abstract

This paper describes the creation of a cDNA microarray for annual sunflowers and its use to elucidate patterns of gene expression in Helianthus annuus, Helianthus petiolaris, and the homoploid hybrid species Helianthus deserticola. The array comprises 3743 ESTs (expressed sequence tags) representing approximately 2897 unique genes. It has an average clone/EST identity rate of 91%, is applicable across species boundaries within the annual sunflowers, and shows patterns of gene expression that are highly reproducible according to real-time RT–PCR (reverse transcription–polymerase chain reaction) results. Overall, 12.8% of genes on the array showed statistically significant differential expression across the three species. Helianthus deserticola displayed transgressive, or extreme, expression for 58 genes, with roughly equal numbers exhibiting up- or down-regulation relative to both parental species. Transport-related proteins were strongly over-represented among the transgressively expressed genes, which makes functional sense given the extreme desert floor habitat of H. deserticola. The potential adaptive value of differential gene expression was evaluated for five genes in two populations of early generation (BC₂) hybrids between the parental species grown in the H. deserticola habitat. One gene (a G protein-coupled receptor) had a significant association with fitness and maps close to a QTL controlling traits that may be adaptive in the desert habitat.

Introduction

Interspecific hybridization is frequent in many groups of plants (Ellstrand et al. 1996; Arnold 1997) and animals (Hubbs 1955; Grant & Grant 1992) and may contribute to biodiversity at both the subspecific level, through introgression, and the specific level, through the formation of polyploid and homoploid hybrid species (Anderson 1948; Grant 1981). Both types of hybrid species must achieve a stable, fertile genotype and become reproductively isolated from the parental species. Allopolyploid species fulfil these requirements via genome doubling, which facilitates homologous pairing at meiosis and also provides a reproductive barrier with parental species of different ploidy. Homoploid (or diploid) hybrid species, in contrast, have the same ploidal level as the parental species. These species gain a stable, fertile, genotype through the sorting of parental sterility factors, and depend on both ecological and karyotypic divergence to generate reproductive isolation between the neospecies and the progenitors (Grant 1981).

Homoploid hybrid species are difficult to document rigorously, and only a handful of cases are known in the literature (Dowling & Secor 1997; Rieseberg 1997; Coyne & Orr 2004; Seehausen 2004; Gross & Rieseberg 2005). Their apparent rarity notwithstanding, homoploid hybrid species represent powerful study organisms for the exploration of speciation and species divergence. Homoploid hybrid speciation appears to be facilitated by rapid ecological divergence (Buerkle et al. 2000), and the hybrid species frequently inhabit extreme environments relative to the parental species (Rieseberg et al. 2003; Gross & Rieseberg 2005). At the same time, early generation hybrids exist in close proximity with parental species and may backcross with them, making homoploid hybrid speciation an example of speciation with gene flow. This interaction between selection and gene flow is of broad relevance to studies of speciation. Most importantly, homoploid hybrid species are uniquely amenable to true experimental approaches because it is possible to compare hybrid species directly to parental species and synthetic hybrids (Hodges et al. 1996; Rieseberg et al. 1996; Brochmann et al. 2000; Rieseberg 2000; Greig et al. 2002; Lexer et al. 2003a, b; Gross et al. 2004; Ludwig et al. 2004; Gross & Rieseberg 2005).

The annual sunflowers (genus Helianthus) represent an excellent study system for homoploid hybrid speciation because three distinct species have resulted from hybridization between the same two parents, Helianthus annuus and Helianthus petiolaris. Both parental species occur throughout the central and western United States, but H. annuus typically inhabits heavy, clay-based soils, while H. petiolaris inhabits sandier soils (Heiser et al. 1969). The hybrid species (Helianthus anomalus, Helianthus deserticola, and Helianthus paradoxus) are ecologically and karyotypically divergent compared to each other and the two parental species (Rieseberg et al. 1990, 1991; Rieseberg 1991).

Helianthus deserticola, the focus of the present study, is a xerophytic species restricted to sandy soils of the desert floor in western Nevada, west-central Utah, and the border of Utah and Arizona (Rogers et al. 1982). The extreme habitat of H. deserticola is matched by the extreme or transgressive phenotypic and ecophysiological traits that characterize the species. Greenhouse and field experiments have shown that H. deserticola consistently exceeds its two parental species in either a positive or a negative direction for a variety of traits (Rosenthal et al. 2002), several of which have been shown to be adaptive in an arid habitat (Rosenthal et al. 2002; Gross et al. 2004). Thus, H. deserticola likely originated via habitat-mediated selection acting on transgressive traits in early generation hybrids between H. annuus and H. petiolaris.

Reproductive isolation of H. deserticola has been enhanced by rapid karyotypic evolution; a minimum of three chromosomal fusions and four breakages are required to derive the H. deserticola genome from its progenitor species (Lai et al. 2005). Past research indicates that such chromosomal rearrangements can accumulate rapidly in Helianthus hybrids (Rieseberg et al. 1995, 1996; Burke et al. 2004), so early hybrids likely experienced drastic changes in genome composition and patterning, which could modify genetic and epigenetic controls of gene expression (Hegarty & Hiscock 2005).

Numerous studies of gene expression in hybrid species have been conducted in recent years, but these have focused mostly on polyploid hybrids (reviewed in Soltis et al. 2003). While some factors influencing gene expression are shared by homoploid and polyploid hybrids, others differ. For example, the evolution of duplicated genes within a polyploid genome is a major theme in the study of allopolyploid speciation and divergence (Wendel 2000; Adams et al. 2003; Adams & Wendel 2004; Blanc & Wolfe 2004), but not in homoploid hybrid species. In contrast to gene duplication events, many of the genetic and epigenetic changes that are thought to alter gene expression in allopolyploids are characteristic of hybridization rather than polyploidy (Osborn et al. 2003; Soltis et al. 2004; Hegarty & Hiscock 2005), and are thus good candidates for contributing to gene expression changes in homoploid hybrid species (Kashkush et al. 2002; Madlung et al. 2002; Salmon et al. 2005). Genetic changes include loss of DNA fragments and chromosomal rearrangements (Liu et al. 2001; Ozkan et al. 2001; Shaked et al. 2001; Madlung et al. 2005); the latter may be caused by increased transposon activity, which is frequently associated with hybridization (Weil & Wessler 1993; Madlung et al. 2005). Transposon activity may also alter patterns of gene expression directly or lead to sequence amplification or gene duplication (Martienssen et al. 1989; Jin & Bennetzen 1994). Finally, epigenetic changes in gene regulation are common in synthesized allopolyploids, usually in the form of DNA methylation (Shaked et al. 2001; Madlung et al. 2002; Wang et al. 2004).

In sum, both of the factors that have been major components of the evolutionary history of H. deserticola, ecological selection and karyotypic divergence, might result in modified gene expression in the hybrid species. So how does gene expression in H. deserticola compare with gene expression in the parental species? This question is especially interesting because transgressive traits in hybrid species may have several different causes. It has been shown that transgressive segregation in early generation hybrids is likely caused by the ‘stacking’ of alleles with the same effect from two divergent genomes (Rieseberg et al. 1999, 2003). Alternatively, as with allopolyploids, some component of transgressive segregation could result from novel gene expression brought on by chromosomal restructuring, transposon activity, and other consequences of hybridization (Hegarty & Hiscock 2005). Both of these scenarios would result in an immediate change in the phenotypes of early hybrids, although the first scenario would require very little change in gene expression compared to the second. Finally, it is probable that ecological selection after speciation would result in the gradual accumulation of mutations that would eventually cause differential gene expression in the hybrid species compared to the parental species.

Here, we report the first use of microarrays to analyse patterns of gene expression in a homoploid hybrid species and its progenitors. We describe the development of an EST-based microarray for sunflower, and its use to examine patterns of gene expression in H. annuus, H. petiolaris, and H. deserticola. We specifically ask whether the hybrid species, H. deserticola, exhibits transgressive levels of gene expression and if differentially expressed genes are associated with karyotypic differences and/or QTLs (quantitative trait loci) underlying phenotypic differences. Also, to assess whether the gene expression changes may contribute to ecological divergence, the fitness effects of a subset of differentially expressed genes were tested in the extreme desert habitat of H. deserticola.

Materials and methods

Plant materials

Seeds of the parental species, Helianthus annuus and Helianthus petiolaris, and the hybrid species Helianthus deserticola, were collected from their typical natural habitat: H. annuus (Rieseberg1312) located 2.2 miles (3.52 km) west of route 6 on the road to Little Sahara Recreation Area, Juab County, Utah; H. petiolaris (Rieseberg1325) located 0.5 miles (0.8 km) east of exit 95 (route 20) off I-15, Iron County, Utah; and H. deserticola (Rieseberg1320) located about 1 mile (1.6 km) south of Toquerville exit off I-15 on frontage road, Washington County, Utah. Seeds were germinated following (Welch & Rieseberg 2002), and germinating seeds were grown in a growth room at 22–25 °C with a 9-h light cycle. Experiments were conducted 4–6 weeks after germination when seedlings were 5–8 cm high and had 5–8 true leaves. The entire seedlings, including roots, shoots and leaves were harvested, quick frozen in liquid nitrogen and stored at −80 °C until RNA was extracted. To minimize individual differences that are prominent in natural populations compared to laboratory strains, three individuals were pooled as one biological sample and the microarray experiments were conducted using a loop design, in which RNAs from each species were directly compared with each other rather than indirectly by reference to a common standard (Fig. 1). Four biological replicates were employed for each comparison.

Fig. 1.

Loop design employed for microarray experiments. Twelve competitive hybridizations were performed using total RNA from three species.

Expressed sequence tag (EST) cDNA clones

The sunflower cDNA array used in this study was derived from cDNA libraries of three species: H. annuus, Helianthus paradoxus and Helianthus argophyllus, which were generated by the Compositae Genome Project (http://compgenomics.ucdavis.edu/) in collaboration with Celera AgGen. Information on tissue treatments and library construction may be obtained from http://cgpdb.ucdavis.edu/Library_Construction/. A total of 44 061 ESTs from the H. annuus cDNA library and 23 127 ESTs from the H. paradoxus and H. argophyllus cDNA libraries were sequenced. Together, they represent approximately 18 031 unique genes (http://cgpdb.ucdavis.edu/).

Array generation

A total of 3743 cDNA clones, representing several sources/kinds of cDNAs, were placed on the array. First, using tag information from the H. annuus cDNA library, 700, 1300 and 700 cDNA clones were chosen from environmentally stressed roots, environmentally stressed shoots and environmentally stressed flowers, respectively. Second, 275 cDNAs with known genetic map positions in H. annuus (Z. Lai et al., in press) were also arrayed, so that effects of chromosomal rearrangements on gene expression variation could be monitored. Third, 768 cDNA bacterial clones were selected from salt- and drought-stressed subtractive cDNA libraries of H. paradoxus and H. argophyllus, respectively. All cDNA bacterial clones were rearrayed using a Biomek FX liquid handling robot (Beckman-Coulter) with barrier tips to prevent clone cross-contamination. Selected clones were inoculated into 96-well plates with LB/Ampicillin and incubated for 16 h at 37 °C. Following overnight growth, 10 μL of culture suspension was transferred into 96-well plates containing 90 μL of Milli-Q water per well. Diluted culture plates were heated at 95 °C for 10 min in a thermal cycler to lyse the cells and release the plasmid clones. Bacterial clones in culture were then used directly as the templates to amplify cDNA clone inserts.

The polymerase chain reaction (PCR) products were generated from the H. annuus cDNA library using the following M13 primer pairs: M13-F, 5′-TGTAAAACGACGGCCAGT-3′; M13-R, 5′-CAGGAAACAGCTATGACC-3′ and from H. paradoxus and H. argophyllus cDNA libraries using the following modified M13 primer pairs: M13-F, 5′-GTTTTCCCAGTCACGACGTT-3′; M13-R, 5′-GGAAACAGCTATGACCATGA-3′. PCR amplification was performed in 96-well format with a hot start of 94 °C for 3 min followed by 40 cycles of 94 °C for 45 s, 56 °C for 1 min, and 72 °C for 1 min with a final extension of 10 min at 72 °C. PCRs were run in 100 μL total volume with 10 μL diluted culture as templates. Following amplification, products were purified using the Multiscreen-PCR 96-well purification system (Millipore). The concentration of each cleaned PCR product was then quantified in 96-well format using a SpectraMax 190 microplate spectrophotometer (Molecular Devices Corporation) and 5 μL of each was run on an agarose gel for quality control before arraying. In addition, more than 120 PCR products from the three libraries were sequenced on an ABI 3730 (Applied Biosystems) to verify cDNA clone identities.

PCR products were dried down and resuspended in 10 μL of spotting solution [3X standard saline citrate (SSC) supplemented with 1.5 M betaine] to achieve a final concentration of approximately 150 ng/μL. Note that the effects of three different spotting solutions: 3X SSC, 3X SSC with 1.5 M betaine, and 50% DMSO were tested, but that 3X SSC with 1.5 M betaine produced the most uniform spots and relatively equivalent signal intensities (data not shown). Products were then arrayed from 384-well microarray plates onto CMT-GAPS II microarray slides (Aminosilane coated slides, Corning). A more detailed description of the printing protocol can be found at http://dgrc.cgb.indiana.edu/microarrays/protocols.html. The sunflower array was printed with a GeneMachines Omingrid 300 arrayer (Genomic Solutions) using 48 silicon pins (Parallel Synthesis Technologies). Each PCR product was deposited in triplicate side by side. The control genes, including sunflower actin, Drosophila α-cop, Drosophila actin, Escherichia coli LacZ and spotting solution were printed on each sub-array to monitor printing quality. Following printing, slides were allowed to dry at room temperature overnight and the spotted DNA was immobilized by baking at 85 °C for 3 h. The slides were allowed to cool, washed in a 5X SSC 0.1% SDS solution at 55 °C with vigorous agitation for 5 min, and rinsed twice in MilliQ water at room temperature for 1 min to remove all SDS. The slides were submerged in MilliQ water at 95 °C for 4 min, rinsed in MilliQ water at room temperature for 1 min, and quickly dried by centrifugation. The slides were then placed in the original slide containers and stored in the dark at room temperature.

RNA extraction and preparation of fluorescent-labelled probes

Three whole seedlings were pooled to make a single sample, and four independent replicate RNA extractions were prepared from each species and used for probe synthesis. Total RNA was extracted using the TRIzol reagent (Invitrogen)/RNeasy (QIAGEN) approach. Briefly, seedlings were homogenized in the presence of liquid nitrogen, then dissolved in TRIzol with a ratio of TRIzol to tissue at 1 mL/100 mg. After the addition of chloroform (1/5 TRIzol volume) and centrifugation for phase separation, the aqueous phase containing RNA was removed, the volume was recorded, and 0.53X volumes of 100% ethanol was added. The mixture was then applied to an RNeasy mini column (QIAGEN) and purified RNA was obtained following the manufacturer’s instructions. Note that to achieve a high yield of RNA a large volume of mix was applied to a single column. Probes were made using the indirect amino allyl labelling method. Approximately 15–20 μg of total RNA was reverse transcribed to first-strand cDNA in the presence of amino-modified nucleotides. This product was purified, labelled with fluorescent Alexa Fluor 555 (similar to Amersham Cy3) or Alexa Fluor 647 (similar to Amersham Cy5) dye (Molecular Probes), and purified again according to standard protocols from the SuperScript indirect cDNA labelling system (Invitrogen). To avoid dye bias, two of four cDNA samples from each species were labelled with Alexa Fluor 555 and the other two with Alexa Fluor 647.

Hybridization, washing and scanning

After reverse transcription, labelling, and purification, 2 μL of the labelled probes were run on an agarose gel and analysed by a Typhoon 9200 Variable Mode Imager (Amersham Biosciences) to ascertain dye incorporation and size range of synthesized cDNAs. Probes were then combined for each desired comparison (e.g. H. annuus labelled with Alexa Fluor 555 and H. petiolaris labelled with Alexa Fluor 647, etc.) and dried to 5.6 μL with a Speed-Vac (Thermo Savant). Next, 2 μL human COT-1 DNA (1 mg/mL) and 2 μL poly d(A) (1 mg/mL) were added along with 20μL 100% ultra formamide, 10 μL 20X SSPE, and 0.4 μL 10% SDS to make the final concentration of 50% formamide, 5X SSPE and 0.1% SDS in a total volume of 40 μL. Probes were heated at 95 °C for 4 min, centrifuged 1 min at room temperature, and employed for hybridization.

Slides were incubated in prehybridization buffer containing 50% formamide, 5X SSPE, 0.1% SDS and 0.5% I-Block (w/v) at 42 °C for 1 h during probe preparation. Following the prehybridization treatment, slides were washed twice with 0.1X SSC for 5 min, MilliQ water for 5 min, and dried by centrifugation. Probe was applied to the prehybridized microarray slides and covered with cover slips (LifterSlips, Erie Scientific), and slides were placed into hybridization chambers (Corning). Hybridization chambers were wrapped in aluminium foil to protect them from light and submerged in a water bath at 42 °C for 16–20 h. After incubation, the microarray slides were washed in 0.5X SSC 0.01% SDS wash solution at 55 °C to remove cover slips. The arrays were then washed in 0.5X SSC, 0.01% SDS at 55 °C for 5 min, 0.5X SSC at 55 °C for 5 min, 0.1X SSC at room temperature for 5 min, and 0.01X SSC at room temperature for 2 min. The slides were dried by centrifugation after washing.

Slides were scanned immediately after drying using an Axon GenePix 4200A scanner (Axon Instruments). Photo-Multiplier Tube (PMT) settings were adjusted manually to achieve good signal intensities for the majority of spots and to minimize the number of spots on the array with saturated signal values. To normalize the two channels with respect to signal intensity, PMT settings were adjusted so that the signal ratio of Alexa Fluor 647/Alexa Fluor 555 across the whole array was close to one.

Data collection, normalization and analysis

Spot intensities were quantified using Axon GENEPIX PRO 5.1 image analysis software (Axon Instruments) and channel ratios were determined by the median-of-ratio method. The data sets were filtered for spots flagged as ‘Bad’ or ‘Not Found’ by GENEPIX PRO 5.1. Raw data from GENEPIX were then imported into Bioconductor R and analysed using the LIMMA (Linear Models for Microarray Data) library (Smyth 2004; software manual available from http://bioinf.wehi.edu.au/limma/). Background-subtracted signal intensities from un-flagged spots were used for within-array print-tip loess normalization and linear model fitting.

Differential gene expression was established using moderated t-statistics with empirical Bayes (EB) shrinkage of the standard errors. After EB analysis in LIMMA, false discovery rate (FDR) was carried out using the method of Benjamini & Hochberg (1995) to adjust P values for multiple testing. Genes were ranked according to B-statistics (essentially the log-odds of differential expression) using a threshold of B = 3 (Smyth 2004). Statistical significance was determined for each individual test using a threshold of B = 3, which corresponds to a P value of 0.05 (Smyth 2004). GENETRAFFIC (Iobion) software was used for cluster analysis of differentially expressed genes.

Real-time RT–PCR

The RNAs extracted for the microarrays were also used for the real-time RT–PCR to validate the microarray results. Total RNAs were treated with RNase-free DNase I (QIAGEN) to eliminate genomic DNA contamination. First strand cDNA was synthesized from 1 μg DNase-treated total RNA using the SuperScript III Platinum two-step qRT–PCR kit with SYBR green (Invitrogen). After reverse transcription, the reactions were heat terminated and 1 μL E. coli RNase H was added to remove the RNA template from the cDNA: RNA hybrid molecule. The reactions were diluted fourfold with water and used as templates for real-time RT–PCR.

Primers for real-time RT–PCR (Table 1) were designed using OLIGOPERFECT software, available at www.invitrogen.com/oligos. Amplicon lengths were between 100 and 250 bp. Real-time RT–PCR was carried out using a Bio-Rad iCycler (Bio-Rad) in 25-μL total volume with 12.5 μL Platinum SYBP Green qPCR SuperMix-UDG, 1 μL SYBR green reference dye, 2.5 μL fourfold diluted first strand cDNA, and in the presence of 200 nM gene-specific forward and 200 nM reverse primer. Each reaction was run in triplicate. A two-step cycling programme was employed: 50 °C for 2 min hold, 95 °C for 2 min hold, 45 or 50 cycles of 95 °C for 15 s and 60 °C for 30 s. Melting curve analysis followed after PCR amplification. Standard curves were generated by the ICYCLER software with a cDNA mix from the three species as a template and a corresponding gene-specific forward and reverse primer. The data were analysed using the ICYCLER IQ optical system software version 3.0a. The cycle number at which the PCR product crossed the set threshold was reported, and differential gene expression was therefore assessed by the difference in mean cycle number among the three sunflower species (Appendix).

Table 1.

Primers used for real-time RT–PCR and DHPLC analysis

EST name	Forward primer (5′–3′)	Reverse primer (5′–3′)	Amplicon length (bp)
Primers for real-time RT–PCR:
QHN24G07	CAATAGCGCTGCAAAAACAA	TAAGGGATGCTGACACCACA	135
QHB19I06	TGCTTCGCGTCTAATGTCTG	CCATTTTAACCGTCGCTTGT	106
QHB30N12	CGTATGATCTTCTTCACGGC	CTCCGCGTCTTCTTCAGATA	268
QHA3e09	CAAGAAGAGGACGACCTTGA	TCAACAGCCTCTTCATTTCC	185
QHB17O16	TGACAGCCGAATCAACTCAG	TCGGTGATCGTTTGATCGTA	198
QHB33B18	TACCGATGAGTTTCGGATGT	ATAAACGCGTCCTGTATTGC	251
QHB22O21	TGGCCCACATCTTCTTTCTC	CAGCCATACACTCTCCAGCA	242
QHE17B21	AAGACGAGTTGGGCTTTTCT	TTGGTAACCGACCCTTCATA	221
QHN24G03	GCAGAGGCTCTTACCTGTGG	GGATGCTGACACCACACTTG	241
Primers for DHPLC:
QHB33B18	TACCGATGAGTTTCGGATGT	ATAAACGCGTCCTGTATTGC	251
QHB19I06	ATTCTACGCCCTTCGAGC	CATTAGACGCGAAGCAGG	356
QHB21B01	ACTTCACAATTAAGTGCGCC	CCAACTCTGCCTATTGCTGT	374
QHB29A11	GATGACCAAGAACAAATCGG	GACAGGTCATCCACCAAAAG	395
QHB30N12	CGTATGATCTTCTTCACGGC	CTCCGCGTCTTCTTCAGATA	268

Differential gene expression and fitness in the wild

Our candidate gene analysis focused on five differentially expressed genes that were judged to be biologically interesting on the basis of predicted function (transcription factors, signalling cascades, transporters, etc.) and their close proximity to QTLs controlling morphological and ecophysiological traits. Primers were designed based on the sequences of the five genes (Table 1), and PCR products were amplified by employing a standard ‘touchdown’ cycling programme with a final annealing temperature of 54 °C. Genotyping was conducted by denaturing high-performance liquid chromatography (DHPLC) on an automated WAVE nucleic acid fragment analyser (Transgenomic), which allows for efficient detection of length and sequence polymorphisms (Xiao & Oefner 2001). Alleles for the five genes could be discriminated on the basis of length and were therefore assayed under nondenaturing conditions (50 °C).

Cosegregation between candidate gene polymorphisms and fitness were tested in reciprocal second-generation backcross hybrids between the parental species (H. annuus and H. petiolaris) grown in the H. deserticola environment. The synthetic hybrid populations (hereafter BC₂Ann and BC₂Pet) were used previously to document selective pressures and the genetic architecture underlying adaptive traits in the desert environment (Gross et al. 2004; Gross & Rieseberg 2005). Sample size (N) was 231 for BC₂Ann and 218 for BC₂Pet. The fitness surrogate (head number) was square root transformed to improve normality, and tested against candidate gene polymorphisms (all segregating alleles) in each population using one-way ANOVAs. The resulting P values were adjusted for the number of candidate genes tested (Table 2).

Table 2.

Candidate genes tested for fitness associations in BC₂Ann and BC₂Pet populations grown in the Helianthus deserticola habitat

Sequence name	Arabidopsis hit	Description	Potential function	LG	Pop.	d.f.	F	Fitness effect
QHB33B18	At2g40140	CCCH-type zinc finger protein-related	Signalling/Gene regulation	LG03	BC2Ann BC2Pet	4 5	0.635 1.729	NS NS
QHB19I06	At4g17880	bHLH protein family	Signalling/Gene regulation	LG05	BC2Ann BC2Pet	7 8	1.610 0.995	NS NS
QHB21B01	At3g13050	transporter - related	Transport	LG10	BC2Ann BC2Pet	5 —	1.321 —	NS NA
QHB29A11	At1g53190	zinc finger (C3HC4-type RING finger) protein family	Signalling/Gene regulation	LG12	BC2Ann BC2Pet	3 2	1.198 0.329	NS NS
QHB30N12	At5g65280	G protein-coupled receptor-related protein	Signalling/Gene regulation	LG14	BC2Ann BC2Pet	3 4	4.122 1.483	P = 0.007* NS

Sequence names, corresponding Arabidopsis hit gene name, description of gene function, potential function in sunflower, and linkage group location (LG) are given for all five genes. Degrees of freedom (DF), F value, and fitness effect in both the BC₂Ann and BC₂Pet populations based on one-way ANOVAs are given for each gene.

^*P < 0.05 after correction for multiple tests.

NS, not significant after correction for multiple tests.

NA, data not available for BC₂Pet population.

Results

Array development

A 3743 abiotic stress cDNA array was developed for sunflower that included ESTs from environmentally stressed tissues, ESTs from salt- and drought-subtracted libraries, and ESTs of known genetic map locations (Z. Lai et al., unpublished). A cDNA array was viewed as preferable to an oligo array both because of cost considerations and the need to make cross-species comparisons. To reduce cross-contamination, which has been a serious problem for cDNA arrays (e.g. Halgren et al. 2001), bacterial clones in culture were employed as templates for PCR. This approach has also been shown to be more cost-efficient and less labour intensive than amplification from plasmid DNA (Hegde et al. 2000).

PCR products were amplified using the universal primers in the cloning vectors. The size range of the majority cDNA inserts from Helianthus annuus was between 500 bp and 1500 bp, whereas the size of PCR products from Helianthus paradoxus and Helianthus argophyllus were shorter, ranging from 300 bp to 500 bp. Approximately 91% of the PCR products appeared as single bands, 3% as multiple bands, and 6% failed to amplify well. Assessment of clone identity in H. annuus from sequence of > 80 randomly chosen PCR products, revealed that 95% matched the expected identities, compared to only 75% of 40 PCR products sequenced for H. paradoxus and H. argophyllus. This translates to an overall match between PCR product and expected identity in the sunflower array of 91%.

Gene annotation

Sequences of the 3743 ESTs were downloaded from the Compositae Genome Project database (http://cgpdb.ucdavis.edu/), trimmed, assembled by CAP3, and aligned using BLAST against Arabidopsis using an e-value cut-off 1e-10. Results indicate that approximately 2897 unique genes were printed on the array, of which 1696 have apparent homologues in Arabidopsis. These were classified according to the Gene Ontology (GO) annotation search and functional categorization protocol in www.arabidopsis.org/ (Fig. 2). Genes spotted on the array have a wide range of functions including protein metabolism, transport, transcription, signal transduction, cell organization and biogenesis, and developmental processes.

Fig. 2.

Distribution of the five largest groups of genes according to functional category. The bar graph shows the proportion of genes in each category on the entire array, genes with significant differential expression in the comparison of Helianthus annuus vs. Helianthus petiolaris, and the genes with significant transgressive expression in Helianthus deserticola compared to the two parental species. Absolute number of genes in each category is noted above the bar. Because only five groups of genes are shown for simplicity, percentages will not total to 100.

Cross-species hybridization

A major concern of microarray-based studies involving multiple species is whether variation in spot intensities can be confidently attributed to changes in gene expression rather than sequence. Fortunately, the Helianthus species employed in both array development and cross-species hybridizations are very closely related [< 1 million years (Myr) divergence]. As a consequence, probes from the three species employed to make the array (H. annuus, H. paradoxus and H. argophyllus) showed equivalent signal intensities, regardless of the derivation of the cDNA. Likewise, no consistent differences in signal intensities were observed in hybridizations involving the focal species of the study, H. annuus, Helianthus deserticola, and Helianthus petiolaris, suggesting that the array may be widely applied across Helianthus. The consistency of the microarray and real-time RT–PCR results (below) also confirm that the differences were due to gene expression and not DNA sequence divergence.

Patterns of gene expression in the homoploid hybrid and parental species

Our experimental design and analyses enabled us to identify many small but reproducible differences among the three species (Fig. 3). This is critical because even small differences have the potential to be biologically important (Gibson 2002). Comparison of H. annuus to H. petiolaris revealed significant differential expression for 206 unique genes, 108 genes exhibiting greater expression in H. annuus and 98 in H. petiolaris (Fig. 4). Annotation of the genes differentially expressed between H. annuus and H. petiolaris according to biological process revealed that 12% encode proteins that function in transport, significantly more than would be predicted given that only 8% of the total genes on the array are annotated as transport-related proteins (χ² = 6.97; P = 0.008) (Fig. 2). As predicted from their evolutionary relationship, fewer genes showed significant differential expression in comparisons of H. deserticola and its parental species: 151 for comparisons with H. annuus and 174 for comparisons with H. petiolaris (Fig. 4). Across all three species, 370 unique genes representing 12.8% of total unique genes on the array showed significant differential expression.

Fig. 3.

Volcano plot of significance against relative expression differences from one microarray comparison between Helianthus annuus and Helianthus petiolaris. Each dot represents one of the 3743 genes that has been filtered and had detectable expression in either species. The X-axis displays Log₂- transformed signal intensity differences between H. annuus and H. petiolaris; the Y-axis is the log-odds calculated according to moderated t-statistic test for differential expression between H. annuus and H. petiolaris. The horizontal dashed line and vertical lines represent significance threshold log-odds = 3 and 2 fold expression differences, respectively. All spots above the horizontal dashed line are genes that were identified as showing significant differential expression between the two species.

Fig. 4.

Venn diagrams of genes with significant differential expression in cross-species microarray comparisons. The Venn diagrams show the number of overlapping and nonoverlapping up- and down-regulated genes in each of the three species relative to the two other species. The first two Venn diagrams show gene expression in Helianthus deserticola (des) compared to Helianthus annuus (ann) and Helianthus petiolaris (pet), the second two show gene expression in H. annuus compared to H. petiolaris and H. deserticola, and the final two show gene expression in H. petiolaris compared to H. annuus and H. deserticola.

Further analyses revealed that expression in H. deserticola was transgressive for 58 genes, significantly exceeding the expression levels of both parental species in either a positive (32 genes) or negative (26 genes) direction (Table S1, Supplementary material). There is a diversity of genes with transgressive expression in H. deserticola according to annotation by biological process (Fig. 2) when compared to the unique gene distribution in the whole array (Fig. 2). Again, transporters were significantly over-represented, with 16% showing transgressive expression in H. deserticola compared to 8% on the entire array (χ² = 12.18, P = 0.0005). H. deserticola displayed H. petiolaris-like expression patterns for another 99 unique genes, whereas H. deserticola displayed H. annuus-like expression patterns for 117 unique genes. The abundance of function categories observed for genes with H. petiolaris- and H. annuus-like expressions did not differ greatly from the genes on the entire array (data not shown).

Real-time RT–PCR

Real-time RT–PCR was performed to validate the microarray analysis. Nine cDNA ESTs that did or did not show differential expression in the microarray analysis were arbitrarily selected for real-time RT–PCR. The identity of the PCR products was confirmed by sequencing, and primers were designed according to OLIGOPERFECT software.

Comparison of the results from real-time RT–PCR with those from microarrays analyses revealed roughly similar patterns or tendencies of expression (Fig. 5). For example, a bHLH family protein (QHB19I06), a G protein-coupled receptor (QHB30N12), a CCCH-type zinc finger protein (QHB33B18), and an expressed protein (QHB22O21) displayed the same expression pattern in both real-time RT–PCR and microarray assays: highest expression in H. annuus, and lowest expression in H. deserticola. Similarly, a translation initiation factor (QHA3E09), an acetyl-CoA-related carboxylase (QHB17O16), and an aminomethyltransferase (QHE17B21) showed the same expression pattern in either assay: highest expression in H. deserticola, no expression difference between H. annuus and H. deserticola. Lipid transfer proteins (QHN24G07 and QHN24G03) were more variable in expression between the two assays. While H. annuus had the lowest expression in both assays for both genes, expression in H. deserticola and H. petiolaris varied between assays, perhaps due to polymorphism for a splicing site (Lee & Roy 2004). Overall, the magnitude of expression changes assessed by real-time RT–PCR was usually greater than those assessed by microarray (data not shown).

Fig. 5.

Gene expression differences as measured by microarray vs. real-time RT–PCR. For both the microarray and real-time RT–PCR, the species with the highest level of expression was set at 100%, and the expression levels of the other species were calculated relative to this standard. Error bars are based on the standard deviation from the mean. See Appendix for details.

Associations with karyotypic differences, QTLs, and fitness

The expression levels of the 275 mapped cDNAs (Z. Lai et al., in press) were analysed to survey the effect of chromosomal rearrangements on gene expression. Rearrangement may cause changes in gene expression directly (Hegarty & Hiscock 2005) or allow differential gene expression to be maintained by reducing interspecific gene flow on rearranged chromosomes (Noor et al. 2001; Rieseberg 2001). There are 40 unique, mapped genes that showed differential expression patterns in the three species. We first asked whether differentially expressed genes were more likely to be located on rearranged than collinear chromosomes. Although there was a trend in this direction (25 vs. 15) it was not significant after correcting for the smaller number of collinear than rearranged chromosomes (χ² = 1.38; P = 0.2404). We also asked whether there was an excess of differentially expressed genes in linkage groups that carried inversions, since inversions would be most effective in reducing interspecific gene flow. Again, there was a trend in this direction (29 vs. 11), but it was only marginally significant after correcting for expectations (χ² = 2.67; P = 0.1022).

To assess whether any of the differentially expressed genes might underlie phenotypic variation in Helianthus, we compared the locations of the 40 mapped and differentially expressed genes with QTL positions in an F₃ mapping population of wild × domesticated H. annuus (Burke et al. 2002) and a BC₂ (BC₂Pet) population of H. annuus × H. petiolaris (Lexer et al. 2005). Five occur coincident with QTLs that correspond loosely with predicted functions: (i) QHB33B18, a CCCH-type zinc finger protein, maps close to QTLs for plant height in the F₃ population, and ligule and phyllary length in the BC₂Pet population; (ii) QHB19I06, which belongs to the bHLH protein family, occurs within the confidence interval of a ligule length QTL in BC₂Pet; (iii) QHB21B01, a transporter-related protein, maps coincident with potassium and manganese uptake in the BC₂Pet population; (iv) QHB29A11, a zinc finger (C3HC4-type RING finger) family protein maps to a QTL for head number in the F₃ population; (v) QHB30N12, G protein-coupled receptor-related protein, maps to a chromosomal segment containing QTLs for ligule number and branch height in BC₂Pet. Of course, we recognize that an apparent match in position and function between candidate genes and QTLs may be spurious since there are likely to be numerous genes within each one-LOD QTL region.

Analysis of possible fitness effects of these five genes in the natural habitat of H. deserticola detected a single candidate gene–fitness association: QHB30N12 and only in the BC₂Ann population (Table 2), indicating that the fitness effect of this chromosomal region varies with genetic background. QHB30N12 is up-regulated in H. annuus compared with H. petiolaris and H. deserticola, and has a hit in the Arabidopsis genome (e-value ≤ 1e–10) with a G protein-coupled receptor related protein (At5g65280). G protein-coupled receptors are important in cell-signalling cascades and gene regulation.

Discussion

This research programme was designed to elucidate the changes in gene expression associated with homoploid hybrid speciation and ecological differentiation. To this end, we developed the first microarray for sunflowers, used this array in a series of cross-species hybridizations, and compared gene expression in the homoploid hybrid species Helianthus deserticola to gene expression in the two parental species Helianthus annuus and Helianthus petiolaris. Gene expression results were confirmed using real-time RT–PCR, with strongly consistent results. The potential fitness effects of differentially expressed candidate genes were assayed in the habitat of H. deserticola to assess their potential roles in ecological divergence among species. Below, we discuss our results in the context of developing genomic resources for nonmodel organisms and current research regarding gene expression and hybrid speciation.

The sunflower cDNA array

Microarrays may be classified into two groups based on the nature of the DNA fixed to solid support; oligonucleotide microarrays and cDNA microarrays (e.g. Aharoni & Vorst 2001). For oligonucleotide microarrays, also called ‘gene chips’, synthetic oligos are arrayed on a glass substrate through direct synthesis or spotting. Oligo arrays have the advantage of greater gene density and more uniform hybridization because of similar sequence lengths, and more specific hybridization, which enables differentiation between duplicated or differentially spliced genes (Lee & Roy 2004). The oligo approach also obviates concerns about potential mismatches between cDNA clones and EST sequences (e.g. Knight 2001). However, synthetic oligo arrays are expensive to produce and less sensitive than cDNA arrays, particularly in hybridizations involving distantly related species (Lee et al. 2004). Moreover, the approach requires extensive genomic resources that are typically only available for organisms such as Arabidopsis and rice with whole genome sequences.

In the alternative cDNA approach, PCR-amplified inserts of partially sequenced cDNA clones are spotted onto arrays (e.g. Richmond & Somerville 2000). Despite concerns about mismatches between cDNA clones and EST sequences, the approach has been widely applied to nonmodel organisms in which cDNA and/or EST resources, but not fully sequenced genomes, are available. Given these considerations, we chose to generate a cDNA array for sunflower. The composition of cDNAs on the array reflected our primary interest, which was to dissect the genetic causes and consequences of homoploid hybrid speciation in the annual Helianthus. Both ecological divergence and karyotypic evolution may contribute to gene expression changes in hybrid sunflower species; the sunflower microarray captures both features by including ESTs present in environmentally stressed tissues and ESTs with known genetic map locations (Z. Lai et al., in press).

The main concern associated with cDNA arrays is potential mismatch between cDNA clones and EST sequences, which may as be high as 35%. Sequencing of > 120 PCR products that were ultimately spotted on the sunflower array suggests a clone/EST identity of approximately 91% in sunflower, which compares favourably to previously published EST-based arrays (Hegde et al. 2000; Ma et al. 2001) and cDNA clone studies (e.g. Halgren et al. 2001).

Another concern that is not specific to cDNA microarrays is the accuracy of cross-species comparisons. However, cross-species hybridization with cDNA microarrays have been successful in species with divergence times as great as 65 million years (e.g. Renn et al. 2004; Taji et al. 2004). Given divergence times of < 1 Myr in the sunflower species studied here, cross-species hybridizations were not expected to pose a problem in our studies, and indeed, consistent expression profiles were obtained for all three species. However, we cannot rule out that some apparent changes in gene expression result from differences in gene number among species. Control hybridizations using genomic DNA are underway to determine whether significant differences in expression observed in this study might result from interspecific variation in sequence or copy number.

A final question concerns the overall accuracy of the microarray results, which we tested by real-time RT–PCR. The real-time RT–PCR results were consistent with those from the microarray thereby validating the latter (Fig. 5), although the magnitude of expression change revealed by real-time RT–PCR was usually considerable larger that suggested by microarray analysis. Thus, while microarrays provide a high-throughput means of expression profiling, the precision may not be as great as that afforded by traditional small-scale molecular techniques such as Northern blot or RT–PCR.

Patterns and mechanisms of gene expression changes

Low sequence divergence among sunflower species suggests that many phenotypic differences may result from quantitative differences in gene expression, as is the case in human evolution (Enard et al. 2002), rather than structural changes in gene product. It is no surprise therefore that a large number of genes (370 or 12.8%) varied significantly in gene expression among the three species. While some of these differences may represent interpopulational rather than interspecific differences, most variation in sunflowers occurs within rather than among populations (Gross et al. 2003). Thus, it seems likely that many of the detected differences represent interspecific expression polymorphisms.

As noted in the Introduction, one of the major differences between polyploid and homoploid hybrid speciation is the presence or absence of duplicated genes in the hybrid. As a consequence, silencing of duplicated genes is a major focus of gene expression studies in allopolyploid species, and there is a consistent pattern of a of transcript deactivation as compared to transcript activation in allopolyploids (Kashkush et al. 2002; Soltis et al. 2004; Wang et al. 2004). This contrasts with our study in which roughly equal numbers of genes were up- (32) and down-regulated (26) in the homoploid hybrid species relative to both parental species.

Despite differences in the relative number of genes up- and down-regulated in H. deserticola compared to allopolyploids, other genetic and epigenetic changes that contribute to changes in gene expression in allopolyploids occur at the point of hybridization rather than at genome doubling, and are thus likely to be common to both forms of speciation (Shaked et al. 2001; Kashkush et al. 2002; Madlung et al. 2002; Osborn et al. 2003; Soltis et al. 2004; Wang et al. 2004; Hegarty & Hiscock 2005; Salmon et al. 2005). For example, the extensive chromosomal rearrangements found in the three homoploid hybrid species are characteristic of the effects of transposable elements (Weil & Wessler 1993; Madlung et al. 2005), which could also have a strong effect on gene expression (Martienssen et al. 1989; Jin & Bennetzen 1994). However, the relationship between hybridization, transposable element activity, chromosomal rearrangements, and gene expression in Helianthus remains to be explored. A trend towards greater expression differentiation in rearranged chromosomes was found in the present study, but sample sizes were too small to make any definitive conclusions.

Although we do not yet understand the mechanisms underlying transgressive gene expression in H. deserticola, the very occurrence of transgressive gene expression is interesting because it has been shown that transgressive traits in hybrid species might be caused by the ‘stacking’ of alleles with the same effect from two divergent genomes (Rieseberg et al. 1999, 2003), which would not require a change in the level of gene expression. Our results suggest that some of the divergence between H. deserticola and its parental species might instead be the result of novel gene expression. The source of these changes in gene expression might be chromosomal restructuring, transposon activity, and other products of hybridization (as for allopolyploids), or ecological selection acting over time to promote the gradual accumulation of mutations. The timing and mechanisms of changes in gene expression will be explored in future research by incorporating an F₁ hybrid, synthetic hybrid lineages, and the other ancient hybrid species (Helianthus anomalus and Helianthus paradoxus) into microarray comparisons.

Gene expression and ecological divergence of H. deserticola

A series of greenhouse and field experiments have shown that transgressive traits that characterize H. deserticola are key to its ecological divergence from parental species and survival in the extreme desert floor habitat (Rosenthal et al. 2002; Rieseberg et al. 2003; Gross et al. 2004). The species is negatively transgressive for most traits, whether measured in the greenhouse (Rosenthal et al. 2002) or field (Gross et al. 2004). In contrast, levels of gene expression show a more balanced pattern; 26 genes show negative transgression for gene expression and 32 genes show positive transgression for gene expression. Thus, the negative transgression that dominates the H. deserticola phenotype is not simply a product of reduced gene expression. Another puzzle is that while H. deserticola closely resembles H. petiolaris and differs greatly from H. annuus at the phenotypic level, H. deserticola shows an equivalent divergence from both taxa in terms of the number of genes with differential expression.

Transgressive gene expression in the hybrid species was spread across a broad range of functional classes, implying that changes in gene expression may contribute to many aspects of H. deserticola’s organismal biology. Perhaps the most striking pattern is the abundance of transport-related genes that were expressed to transgressive levels in H. deserticola compared with the parental species; 8% of all genes on the array are transport related, but 16% of the genes with transgressive expression in H. deserticola are transport related. These transport-related genes may be important in the extremely arid environment of the desert floor, functioning as both osmotic sensors and ionic regulators to prevent desiccation (Zhu 2002; Jang et al. 2004).

Finally, ESTs with differential expression across the three species represent candidate genes for adaptive differentiation in Helianthus, and one such gene, QHB30N12 (which resembles a G protein-coupled receptor), was shown to be associated with a fitness-related trait (head number) in the natural habitat of H. deserticola. This gene also maps coincident with QTLs for ligule (ray flower) number and height of first branch, phenotypes that may be of adaptive significance in arid environments. Petal/ligule size and number typically is reduced in desert plants, presumably to reduce water loss (Galen 1999; Herrera 2005), and H. deserticola is negatively transgressive for this trait (Rosenthal et al. 2002). Likewise, early branching is associated with rapid flowering in Helianthus, an advantageous trait in stressful conditions or short growing season (Aronson et al. 1993). Both H. deserticola and H. petiolaris branch earlier than H. annuus (Rosenthal et al. 2002), and early flowering was favoured by selection in the BC₂Ann population when grown in the field (Gross et al. 2004). Thus, if either or both of these traits are controlled by QHB30N12, it is possible that variation at the locus would have an effect on fitness in the H. deserticola environment. Of course, there are many genes underlying this QTL, so it is possible that another unknown gene is responsible for the differences in phenotype and fitness associated with this genomic region. Extensive future research is necessary to establish a direct connection between QHB30N12 and the traits associated with the QTLs, document expression of this gene in wild populations of H. deserticola and its progenitors, and determine its role in ecological divergence among the three species of sunflower. Nonetheless, our results suggest a promising avenue of research for further understanding the origin of Helianthus species.

Supplementary Material

suppl. Supplementary material.

The supplementary material is available from http://www.blackwellpublishing.com/products/journals/suppmat/MEC/MEC2775/MEC2775sm.htm

Table S1 Genes with transgressive expression in Helianthus deserticola

Appendix

Real-time RT–PCR analysis

The species with the lowest average PCR cycle was taken as the highest relative gene expression (the 100% expression reference), and the value of relative gene expression has been calculated by eqn 1. The mean value of gene relative expression examined by real-time RT–PCR was calculated from three biological replicates, the last one QHN24G03 from four technical replicates.

$$j_{(\text{real}-\text{time}\text{RT}-\text{PCR})}=100·2^{\mathrm{\Delta }{\text{C}}_{\text{T}}}$$ (eqn 1)

ΔC_T = C_T (the minimum cycles of gene i among three species) − C_T (gene i in species j), where the C_T is the cycle number at which the PCR product crosses a set threshold.

Microarray analysis

The signal intensities of all spots from both channels were taken as the measurement of differential gene expression for the microarray analysis. The value of relative gene expression examined by a microarray where the relative expression value of gene i in species j is calculated by the eqn 4. For each microarry slide, the LIMMA normalized values of M and A for gene i in species j were used to calculate the normalized R and G values through the simultaneous eqn 5. For gene i, by adding signal intensities of R and G together and taking the species with the maximum signal intensities among three species as the highest of expression (the 100% expression reference), comparisons can therefore be made within three species. The mean value of relative gene expression examined by the microarray was calculated from four biological replicates.

$$M={log}_2\left(\frac{R}{G}\right)$$ (eqn 2)

where R represents the Red channel intensity value and G represents the Green channel intensity value.

$$A={log}_2\sqrt{R·G}$$ (eqn 3)

where R represents the Red channel intensity value and G represents the Green channel intensity value.

$$j_{(\text{microarray})}=100·2^{{log}_2\alpha -{log}_2\beta }$$ (eqn 4)

where α represents signal intensity of gene i in species j and β represents the maximum signal intensity of gene i among three species.

$$R=2^{\frac{2A+M}{2}}\text{and}G=2^{\frac{2A-M}{2}}$$ (eqn 5)

where M is described in eqn 2 and A is described in eqn 3.