Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
Apomixis, a natural form of asexual seed production in plants, has evolved independently in various taxa, and represents an important potential technology for agriculture. The switch to apomixis is based on de-regulation of developmental pathways originally leading to sexual seed formation. Hybridization and polyploidy, both typical characteristics of asexual plants and animals, are mechanisms that could trigger de-regulation. Here we show that up-regulation of alleles in apomeiotic ovules is mediated by genomic duplication, heterochrony and the residual effects of ancient hybridization in diploid apomicts of the Boechera holboellii complex. Using SuperSAGE, we have identified over 4000 differentially expressed mRNA tags between micro-dissected ovules from two diploid sexual (Boechera stricta and B. holboellii) and two diploid apomictic (Boechera divaricarpa) accessions. Pairwise sequence comparisons between tags enabled identification of allelic variants of the same loci. Up-regulated candidate apomeiosis alleles consistently have more than three related alleles, thus demonstrating transcription from duplicated loci. A further 543 alleles were heterochronically expressed between sexual and apomeiotic ovules at developmental stages 2-II to 2-IV. Intriguingly, 69 B. holboellii specific alleles were preferentially up-regulated in apomeiotic ovules, thus showing a remnant‘parent of origin’ effect stemming from the Pleistocene origin of the hybrid B. divaricarpa from taxa related to B. holboellii and B. stricta. These data implicate polyploid gene dosage in the expression of asexual seed formation, and support hypotheses of de-regulation of the sexual pathway. The observed ‘parent of origin’ effect suggests that the genomic memory of hybridization has somehow been maintained after hundreds, if not thousands, of asexual generations.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
The origin and persistence of asexual reproduction, despite mutation accumulation and limited genetic variability, remain one of the most challenging phenomena in evolutionary biology (Bell, 1982). A comparison of asexual animals and plants demonstrates that, despite widely differing reproductive mechanisms, a number of characteristics are frequently found in common, the most obvious of which include hybrid origin and polyploidy (Richards, 2003). Thus, while asexuality across these large phylogenetic distances has been attained independently, the natural selective forces that have shaped its evolution have led convergently to similar variation at the nuclear level.
Asexual animals (parthenogenetic) and plants (apomictic) have arisen repeatedly from sexual ancestors, and it remains unclear how this switch in reproductive mode occurs at the molecular genetic level, or how polyploidy and hybridity contribute to asexual lineage origin and its subsequent evolution in the absence of normal meiosis. For example, polyploidy could both initiate and maintain asexuality through gene dosage effects (Grimanelli et al., 2001), and may additionally provide lineage stability by buffering deleterious mutations in gametes and somatic cells (Richards, 2003). Moreover, hybridization could change regulatory gene expression patterns to induce asexuality (Carman, 1997), or may have positive fitness effects through fixed heterosis in asexual lineages (Kearney, 2005). Clearly both hybridization and polyploidy could induce the global regulatory changes needed to initiate asexuality (Comai et al., 2003; Osborn et al., 2003), but disentangling their relative contributions has been difficult.
Among plants, gametophytic apomixis is a naturally occurring form of asexual seed formation producing progeny that are genetically identical to the maternal genotype from meiotically unreduced embryo sacs (Koltunow and Grossniklaus, 2003). The majority of gametophytic apomictic species are found in the Asteraceae, Rosaceae and Poaceae, where they have arisen independently and recurrently (Grossniklaus et al., 2001). Nevertheless, polyploidy, facultative apomixis (both sexual and apomictic seed production within one individual), and faster development of the apomeiotic ovule relative to the sexual one (Savidan, 2007) are traits that are shared among many of these taxa. Compared to parthenogenesis in animals, apomixis has been the subject of deeper analyses into its developmental mechanisms, because of its potential importance as an enabling technology for agriculture. Harnessing apomixis would greatly facilitate and accelerate the ability of plant breeders to fix and faithfully propagate genetic heterozygosity and associated hybrid vigor in crop plants (Spillane et al., 2004).
Although exceptions do exist, three independent developmental steps must be acquired by an amphimictic (i.e. sexual) plant to produce seeds apomictically: formation of an unreduced embryo sac, e.g. through meiotically unreduced megaspore formation (apomeiosis), development of an embryo from an unfertilized and unreduced egg cell (parthenogenesis), and formation of functional endosperm, e.g. fertilization of the binucleate central cell (pseudogamy). The apomeiotically derived embryo thus receives its entire genome through the female line (although paternal apomixis does occur in cypress; Pichot et al., 2001). As these components are under separate genetic control, it is highly unlikely that all three could evolve in unison in a sexual ancestor through randomly occurring mutations, as expression of any single step would decrease the fitness of its sexual carrier (but see van Dijk and Vijverberg, 2005). An alternative hypothesis is that apomixis arises through de-regulation of the sexual developmental pathway (Koltunow, 1993; Grossniklaus, 2001), which would be manifested at multiple loci simultaneously. In wild apomictic taxa, this coordinated de-regulation could be influenced by global regulatory changes resulting from hybridization and/or polyploidy (Grossniklaus, 2001). Both naturally occurring and induced mutants showing each component separately have been identified (Curtis and Grossniklaus, 2007; Ravi et al., 2008), implying that many taxa have the potential to express apomixis-like traits, in addition to corroborating the hypothesized independent regulation of each component.
Apomictic Boechera are characterized by Taraxacum-type diplospory, whereby the megaspore mother cell undergoes meiosis I without completing the reductional phase (apomeiosis), followed by meiosis II, leading to a nucleus that has the same ploidy as the mother plant (Böcher, 1951; Naumova et al., 2001). As with many asexual taxa, microsporogenesis is typically disturbed in apomictic individuals, and variations in levels of normally reduced, non-reduced and aneuploid pollen exist within and between accessions (Böcher, 1951; Dobešet al., 2004b; Schranz et al., 2005; Sharbel et al., 2005; Voigt et al., 2007). Analyses of meiosis in pollen cells have furthermore demonstrated variable degrees of chromosomal synapsis (univalent and multivalent) in apomictic accessions (Böcher, 1951), and both psuedogamous and autonomous endosperm formation have also been identified (Naumova et al., 2001; Voigt et al., 2007).
Here we have attempted to elucidate the first step in apomixis, apomeiosis, by means of a deep transcriptomic comparison between micro-dissected live sexual and apomeiotic ovules. Importantly, we have exploited the rare phenomenon of diploid gametophytic apomixis (see also Siena et al., 2008) in the Boechera holboellii complex (Böcher, 1951) in order to overcome the confounding effects of polyploidy. Our data show that, in comparison to sexual ovule formation, apomeiosis is characterized by shifts in gene expression (heterochrony), ‘parent of origin’ effects stemming from ancestral hybridization, and allele expression from duplicated loci. In screening the transcriptomes of both sexual and apomictic ovules using combined SuperSAGE and 454 sequencing analyses, we have laid the groundwork for independent validation experiments using additional Boechera accessions and tissues.
Using flow cytometric seed screening (Matzk et al., 2000) to assess reproductive mode, we chose two diploid sexual (B. stricta and B. holboellii) and two highly expressive diploid apomictic individuals (B. divaricarpa3 and B. divaricarpa8) from which 10 live ovules each were micro-dissected for linear cDNA amplification and SuperSAGE analysis (Figure 1 and Tables 1 and 2). Based on the flow cytometric seed screen (Matzk et al., 2000), the two chosen sexual individuals exclusively comprised 2C embryo:3C endosperm, which, as expected, reflected an embryo composition of C maternal (Cm) genomes + C paternal (Cp) genomes = 2C genomes, and an endosperm composition of 2Cm + Cp = 3C. In contrast, the two diploid apomicts were both characterized by >95% apomeiotically derived embryos (unfertilized 2Cm) and >95% fertilized endosperm (e.g. 4Cm + 2Cp = 6C; Table 2).
Table 1. Boechera accessions used in SuperSAGE and RT-PCR analyses
aSpecies identifications were based upon silique orientation and trichome morphology.
Gold Creek, Colorado
Bandy Ranch, Missoula County, Montana
Birch Creek, Ravalli County, Montana
Charlies Gulch, Ravalli County, Montana
Bandy Ranch, Missoula County, Montana
Birch Creek, Ravalli County, Montana
Table 2. Flow cytometric seed screen results showing the number of seeds for each Boechera accession characterized by the embryo:endosperm ploidy values
Embryo:endosperm ploidy value
aThe sexuality of this accession (SAD4) was also confirmed by crosses and isozyme analyses.
B. strictaa (B06-485)
B. holboellii (105.6.7)
B. divaricarpa3 (300.6.1)
B. divaricarpa8 (205.3.4)
SuperSAGE analysis of the four ovule-specific cDNA libraries yielded a total of 452 516 sequenced mRNA tags (26 bp), 339 888 of which were found in two or more copies in all libraries together, and, from these, 43 289 different tag sequences were identified (Table 3). An additional 112 628 tags occurred as singletons.
Table 3. Number of shared (above diagonal) and differentially expresseda SuperSAGE tags (below diagonal) between Boechera ovule-specific libraries, in addition to the total number of sequenced and singleton tags per library
aP < 0.05 (Audic and Claverie, 1997); for each relative comparison (up/down-regulated), ‘up-regulated’ refers to the first number in the column (e.g. in holboellii versus stricta, 923 and 2014 tags were up- and down-regulated, respectively, in holboellii).
Tags sequenced per library
Number of singletons
Both sexual flower-specific and apomictic flower-specific cDNA libraries were generated using 454 (FLX, 454 Life Sciences, Roche, http://www.roche.com) technology. The sexual flower-specific library had a total of 12 365 737 high-quality sequenced base pairs (mean sequence length 409 bp, median sequence length 244 bp), and a Newbler assembly (454 Life Sciences) generated 30 237 assembled contigs. The apomictic flower-specific library had a total of 14 622 601 high-quality sequenced base pairs (mean sequence length 414 bp, median sequence length 246 bp), and Newbler assembly generated 35 362 assembled contigs.
SuperSAGE tags are frequently allele-specific as they are derived from the most 3′NlaIII restriction site of the 3′ UTR of each mRNA molecule (Matsumura et al., 2006). Any two tags sampled at random from the four Boechera SuperSAGE libraries will probably originate from different genes, and are thus not expected to share sequence homology. This was supported by pairwise comparisons of all tag sequence variants obtained from Boechera (n =155 917), in addition to those from an in silico Arabidopsis SuperSAGE analysis (n =26 197), all of which demonstrated no overall deviation from the expected result [Figure 2(a)]. Nonetheless, pairwise sequence comparisons of all Boechera and Arabidopsis SuperSAGE tags showed differences of 50- to >105-fold in the observed versus expected variation for tag pairs with eight or fewer base pair differences, implying that the majority of these represent allelic variants of the 3′ UTR regions at the same or duplicated loci [Figure 2(b)].
Sex- and apomeiosis-specific tags, and heterochrony
Our goal was to identify transcriptomal differences between the sexual and apomictic ovules. For each SuperSAGE tag, we thus tested for significant differences in copy number between each of the four libraries using the method described by Audic and Claverie (1997), and three classes of differentially expressed tags were categorized through comparison of the four libraries (Table 4).
Table 4. Criteria based on corrected numbers of SuperSAGE tags per library which were used to differentiate allele classes
aLetters refer to each SuperSAGE library: a, B. holboellii; b, B. divaricarpa3; c, B. divaricarpa8; d, B. stricta.
bTag numbers are considered not equal (≠) if P ≤0.05 (Audic and Claverie, 1997); otherwise tag numbers are considered almost equal (≈).
Apomeiosis- or sex-specific (I)
(a & d) = 0 & (b & c) > 0 OR (a & d) > 0 & (b & c) = 0
Differentially expressedb (II)
(a, b, c, d) > 0; (a ≈ d & b ≈ c) & (a ≠ b & a ≠ c & d ≠ b & d ≠ c)
Species-specific (trans-regulated) (III)
(a > 0 & d = 0) & (b ≠ a & c ≠ a) OR (a = 0 & d > 0) & (b ≠ d & c ≠ d)
The first class comprised 1879 sex-specific and 2395 apomeiosis-specific tags that were found exclusively in both sexual or both apomictic libraries (Table 4). Of the apomeiosis-specific tags, 83 (3.5%) were significantly differentially expressed between the two apomicts, while 47 (2.5%) of the sex-specific tags were significantly differentially expressed between the two sexual libraries (P <0.05) (Audic and Claverie, 1997). Thus, most apomeiosis- and sex-specific tags were consistently expressed between the two apomictic or sexual libraries, and, in addition, expression of the sex-specific tags was conserved even when compared between two species (B. stricta and B. holboellii).
The presence and absence of tags (i.e. alleles) between SuperSAGE libraries could be explained by allelic differences between the accessions or variation in developmental timing between the accessions and/or tissues stages analysed here, or, alternatively, heterochronic allele expression could characterize the sexual and apomeiotic ovules. In the case of allelic variation for any sex- or apomeiosis-specific tag, corresponding tags of high sequence homology (i.e. alleles) should be found in the pools of tags sequenced from other libraries. Alternatively, heterochrony would be shown if identical alleles were found in both sexual and apomictic genomic DNA, and if they were expressed only within sexual or apomictic ovules at the developmental stage sampled here.
We tested for allelic differences by first searching for homologous SuperSAGE tags that differed by one nucleotide from the sex- and apomeiosis-specific tags, and identified an additional 899 related tags. We then calculated the transcriptional profile for each family of tags (original tag plus its homologues) by adding the SuperSAGE tag copy numbers together for each family, and, after doing so, 1809 (96%) and 1712 (71%) of the families remained as sex- and apomeiosis-specific, respectively. Thus, most of the alleles that are specifically expressed either during sexual or apomeiotic egg formation do not have a corresponding allele (as defined by a 1 bp difference) expressed in the opposite reproductive mode.
In order to examine whether expression of the apomeiotic- and sex-specific tags was temporally shifted (heterochrony), we performed a sequence homology search against two normalized cDNA libraries from pooled flower stages 1–12 (Smyth et al., 1990) of three diploid sexual plants (two B. holboellii and one B. stricta) and three apomictic plants (two diploid and one triploid B. divaricarpa) using 454 (FLX) technology. Of the apomeiotic-specific tags, 180 (10%) and 177 (10%) blasted perfectly (100% homology) to the apomixis- and sex-specific 454 libraries, respectively (Figure 3). In contrast, 366 (20%) and 329 (18%) of the sex-specific tags blasted perfectly to the apomixis- and sex-specific 454 libraries, respectively (Figure 3). We have thus identified 177 apomeiosis-specific and 366 sex-specific tags that are found in the ‘opposite’ cDNA libraries that contain summed transcripts expressed over many developmental stages. In other words, 543 alleles that are common to both sexual and apomictic genomes are turned on or off with respect to apomeiosis or sex at this stage of ovule development, and, when turned on, show similar expression patterns between the two sexual or two apomictic ovules.
The goal of this study was to screen the ovule transcriptomes for differentially expressed mRNAs whose expression patterns will subsequently be examined using microarrays for various Boechera accessions and multiple developmental stages and tissues. We nonetheless attempted to confirm our hypothesis of heterochronic expression using RT-PCR on genes corresponding to four sex- and four apomeiosis-specific SuperSAGE tags at two ovule stages (Figure 1, Table 1 and Table S1). Compared to stage 1, the four genes corresponding to the sex-specific SuperSAGE tags demonstrated an overall shift to over-expression in sexual ovules in stage 2 (Table S1 and Figure S1). An overall change in expression between ovule stages 1 and 2 for the four genes corresponding to the apomeiosis-specific tags was not apparent (Table S1 and Figure S1). Although the weak correlation between our SuperSAGE and RT-PCR data sets may reflect the general problem of comparing transcript abundance across platforms (Wang, 2007), we suspect that (i) variable allele concentrations (i.e. mRNA dosage) relative to RT-PCR detection threshold levels (Grimanelli et al., 2005) in micro-dissected samples, (ii) the use of gene- rather than allele-specific primers in the RT-PCR analysis, and (iii) the compounded influence of widespread duplications in the apomictic genome (Corral et al., 2009) on allele dosage, also contributed to the inconsistencies.
Differentially expressed tags
A second class was characterized by tags that were found in all four libraries, but that were differentially expressed between sex and apomeiotic ovules, and expressed consistently within each reproductive mode (Table 4). We identified 59 tags that were expressed in all four libraries and were not different in copy number when the two sexual or two apomictic individuals were compared, but were significantly differentially expressed between the apomictic and sexual individuals in all comparisons [Figure 4(a)]. Of these, 39 and 20 tags were down- and up-regulated, respectively, and furthermore showed similar levels in both apomeiotic ovules [Figure 4(a)].
‘Parent of origin’ effects
The final category of apomeiosis candidate alleles was characterized by tags that were found in one sexual library only (either B. stricta or B. holboellii) and in both apomictic libraries (Table 4). These tags support hypotheses of apomixis origin via hybridization (Carman, 1997), as B. stricta and a taxon related to B. holboellii have been shown phylogenetically to be progenitors of the hybrid apomictic B. divaricarpa (Dobešet al., 2004a,b).
Regulatory changes in mRNA expression can be inferred by measuring cis- and trans- effects in expression patterns in both parents and hybrids, a comparison which is appropriate when the parental species have undergone a long period of reproductive isolation (Landry et al., 2005), as is the case with Boechera. With regard to mRNA tags that were found in only one of the two sexual individuals, 13 581 and 7412 tags were identified as either B. holboellii- or B. stricta-specific, respectively. Of these, 440 B. holboellii- and 525 B. stricta-specific tags were significantly differentially expressed in both apomictic individuals [Figure 4(b,c)], and hence are probably under trans control (i.e. ‘parent of origin’ effect). Interestingly, the species-specific alleles exhibit different patterns of expression in the apomicts, with 16% (n =69) and 83% (n =366) of the B. holboellii-specific alleles being up- and down-regulated, respectively, while virtually all (99%, n =519) of the B. stricta-specific alleles are down-regulated [Figure 4(b,c)]. These data suggest that apomeiosis is characterized by asymmetrical up-regulation of B. holboellii alleles in the hybrid genome. The relatively scattered pattern of down-regulated tags in both distributions [Figure 4(b,c)] is suggestive of the number of mechanisms by which a gene can be down-regulated (Vaucheret, 2006), in addition to variability in the rate of post-transcriptional regulation (Garneau et al., 2007).
Gene copy number variation
Gene duplication was estimated by calculating allele family size in the four SuperSAGE libraries. For each SuperSAGE tag, a network of tags (i.e. allele family) from its respective library was computed such that each member of a network was related to at least one another network member by 1 bp. This analysis provides a conservative estimate of allele family size, as insertion/deletion variation is not considered, nor are allelic variants with differences >1 bp, although there is evidence for allelic variation characterized by up to 8 bp differences in observed tag pairs [Figure 2(b)]. The number of members per allele family was calculated and plotted for each SuperSAGE library [Figure 5(a)], and demonstrates a trend of increasing frequencies of large allele families (more than nine alleles per family) in the apomictic libraries [Figure 5(a)]. Furthermore, up-regulated tags belonging to allele classes II and III (Table 4) show a trend of belonging to larger allele families compared to down-regulated tags [Figure 5(b)]. Thus, it appears that many of the differentially expressed tags that are up-regulated in the apomeiotic compared with the sexual ovules may have been transcribed from duplicated loci in both apomicts. The elevated values for the apomictic libraries (Figure 5) are not consistent with a larger number of sequenced SuperSAGE tags, as the divarcarpa3 library had the least number of tags sequenced (Table 3).
Gene ontology analysis
Using Fisher’s exact test (Conesa and Gotz, 2008), we analysed Boechera 454 sequence groups corresponding to each SuperSAGE tag class (Table 4), and found no significant enrichment for particular gene ontology (GO) classes in any comparison (data not shown). Eighteen level 3 GO (biological process) terms associated with reproduction were nonetheless identified from the Boechera cDNAs corresponding to differentially expressed B. stricta- and B. holboellii-specific SuperSAGE tags (Figure S2 and Table S2).
The data presented here were generated from live micro-dissected ovules in the developmental stage during which apomeiosis is hypothesized to be expressed, thus minimizing the effects of mRNA degradation through tissue fixation (Goldsworthy et al., 1999) and non-specificity of transcriptomic profiles due to analysis of mixed tissues. The effects of mRNA degradation could conceivably distort statistical comparisons of SuperSAGE tag frequencies between libraries, given that the majority of tags identified here had relatively low copy numbers (<100) (Figure 6). A combination of deep transcriptomic profiling of the ovules using SuperSAGE (Matsumura et al., 2006), in conjunction with sequencing of the sexual and apomictic flower transcriptomes using 454 technology, has enabled us to identify over 4000 differentially expressed mRNAs between sexual and apomeiotic ovules at this single stage of development. Finally, use of diploid apomictic B. divaricarpa accessions that show high levels of apomeiosis (Table 2) has reduced the confounding effects of polyploidy.
The relatively low levels of sequence overlap (<50%; data not shown) between the SuperSAGE (3′-biased) and 454 cDNA libraries may be reflective of 5′ bias in the 454 sequencing (Weber et al., 2007), although 3′ bias resulting from incomplete (i.e. not full-length) cDNA synthesis has also been demonstrated (Bainbridge et al., 2006; Emrich et al., 2007). As cDNA preparation for the 454 sequencing here entailed linear amplification, 3′ bias in our cDNA libraries is more likely to be the case. Hence an alternative explanation for the low overlap between our SuperSAGE and 454 libraries could be that ovule-specific transcripts may have been under-represented or absent from the pooled flower-specific cDNA libraries. Furthermore, as 3′ UTR regions demonstrate allelic variability (Andolfatto, 2005; Eveland et al., 2008) that may involve factors that were not extensively analysed here, including insertion/deletion mutations as well as single nucleotide variation (Figure 3) at single or duplicated loci (Figure 5), our assessment of homology between the SuperSAGE and 454 datasets is probably an under-estimation.
The pairwise comparisons of all SuperSAGE tags, both in the four Boechera libraries and in the in silico Arabidopsis SuperSAGE experiment, demonstrate that allelic variation has been captured for many of the sampled mRNAs from the sexual and apomeiotic ovules (Figure 2). SuperSAGE tags are derived from the 3′ UTR of each mRNA molecule (Matsumura et al., 2006), which have been shown to be allele-specific in plants (Eveland et al., 2008). Although they are not translated into proteins, 3′ UTRs are involved in the regulation of mRNA transcript numbers, and have been shown to be under measurable levels of selection pressure in humans (Chen and Rajewsky, 2006). In Drosophila, 3′ UTRs are frequently under higher levels of selection compared to synonymous polymorphisms in codon regions (Andolfatto, 2005), a probable result of functional constraints on specific sequence motifs involved in down-regulation of gene expression through translation inhibition and rapid targeted decay of mRNAs (Shyu et al., 2008).
Thus the elevated numbers of observed tags that share high levels of sequence homology [Figure 2(b)] are reflective of multiple alleles at the same or duplicated loci. The observed numbers of homologous (i.e. allelic) Boechera tag pairs were consistently higher than those from the in silico Arabidopsis SuperSAGE analysis [Figure 2(b)], which could reflect the different sequencing technologies used to generate the data (GS20 versus Sanger). However, it is more likely that the differences can be explained by elevated levels of homozygosity in Arabidopsis (Marais et al., 2004) compared to Boechera, whose complex history of glacial isolation, hybridization and gene flow has led to elevated heterozygosity (Roy, 1995; Dobešet al., 2004a,b).
Transcriptomal differences between sexual and apomeiotic ovules
Remarkably, the differentially expressed profiles of many hundreds of alleles representing three classes (Table 4) are consistently similar between the apomeiotic ovules that were independently collected and prepared from two asexual lineages (Figures 4 and 6). In contrast, the global expression profiles of the two sexual Boechera are less similar (Figure 6), and probably demonstrate species-specific differences at this stage of development. While expression differences between the apomictic libraries may have resulted from experimental noise, some could also reflect the dynamics of apomixis evolution from sexual ancestors; for example, whether the two B. divaricarpa accessions are from one or multiple asexual lineages (i.e. single or repeated origin from sexual ancestors). In the former case, the differences in the transcriptomal patterns could reflect independent mutation accumulation since the last shared asexual ancestor between them. In the latter, the independent origins of the two asexual lineages from sexual ancestors must have led convergently to highly similar global gene expression patterns at this stage of ovule development (Figure 6).
The identification of 543 genes that exhibit a developmental timing shift in expression between the sexual and apomeiotic ovules implicates sexual pathway de-regulation in apomeiosis expression (Koltunow, 1993; Grossniklaus, 2001). This is probably a conservative estimate of the number of genes that actually show heterochronic shifts, as we have sampled one developmental stage and could thus only consider the presence and absence of mRNA tags (i.e. genes turned on and off). We have recently completed a second SuperSAGE analysis (over two million sequenced tags) of sexual and apomeiotic ovules sampled across four developmental stages so that we can additionally test for heterochrony in genes that exhibit regulatory changes over time.
One of the most intriguing results was the class of apomeiosis candidate alleles (class III) that displayed the ancient ‘parent of origin’ effect [Table 4 and Figure 4(b,c)]. Boechera stricta- or B. holboellii-specific mRNAs that are differentially expressed in the B. divaricarpa (hybrid apomictic) nucleus could be explained by heterochronic shifts in gene expression, ‘genomic shock’-like responses (McClintock, 1984) in hybrid genomes that may lead to post-transcriptional mRNA degradation, or changes to major regulatory genes that have cascade effects on transcription networks (Osborn et al., 2003). It is striking that the unfertilized apomeiotic ovule is characterized by up-regulation of B. holboellii alleles [Figure 4(c)], and furthermore that this pattern exists several hundred (if not thousands) of generations after the original hybridization event (Dobešet al., 2004a). Given that maternal allele expression frequently characterizes early embryo development (Grimanelli et al., 2005; Springer and Stupar, 2007), B. holboellii (or a closely related taxon) was probably the maternal parent in the origin of the asexual (hybrid) lineages analysed here.
We have not performed reciprocal crosses in order to test the hypothesized ‘parent of origin’ effect (Springer and Stupar, 2007), nor would this be straightforward given the variable pollen and egg cell ploidy and their influence on offspring ploidy in sexual and apomictic accessions (Schranz et al., 2005; Voigt et al., 2007). Differential allelic expression is known to occur in homoeologous genes in hybrid plants (Adams et al., 2003; Adams and Wendel, 2005; Schlueter et al., 2006; Udall et al., 2006; Adams, 2007), and thus the ‘parent of origin’ effect could also reflect species-specific differences (i.e. between B. holboellii and B. stricta, or related ancestral lineages involved in the original hybridization event) with regard to tissue specificity or response to environmental conditions (Guo et al., 2004), rather than being the underlying factor in the switch from sexual reproduction to apomixis.
Genomic duplications in apomicts
Duplications in the apomictic genome could provide polyploid dosages of alleles required for apomeiosis, and this would explain the rare evolutionary stability of diploid apomixis in Boechera. Apomictic genomes, unconstrained by selection pressure to maintain stable meiosis, are typically structurally variable compared to sexual genomes (Roche et al., 2001). In apomictic Boechera, this is supported by variable chromosomal synapsis in meiosis I (Böcher, 1951), as well as allele sequence divergence and genomic duplication (Kantama et al., 2007; Corral et al., 2009).
If one assumes that SuperSAGE allele families (i.e. groups of tags where any two members differ by 1 bp) comprising more than two members represent alleles transcribed from duplicated loci in diploid Boechera, evidence for the highest number of genomic duplications whereby all duplicated alleles are expressed is shown for both apomeiotic SuperSAGE libraries [Figure 5(a)]. Furthermore, up-regulated candidate apomeiosis alleles in classes II and III (Table 4) show a trend for larger allele family sizes relative to down-regulated alleles [Figure 5(b)]. Taken together, it appears that many of the differentially expressed tags that are up-regulated in apomeiotic ovules have been transcribed from loci that have undergone duplication in the apomictic genome.
Evolution of apomeiosis in Boechera
We have provided evidence that heterochrony, gene duplication, and, interestingly, ancient ‘parent of origin’ effects characterize the transcriptomic signature of apomeiosis in Boechera. Apparently allelic up-regulation is important for apomeiosis, and gene duplication is the mechanism that mediates transcript elevation in the apomictic genome. Alternatively, down-regulation of alleles expressed during sexual seed development may also be implicated in apomeiosis, although we cannot make any conclusions regarding 3′ UTR motifs and their involvement in post-transcriptional regulatory differences between sexual and apomeiotic ovules due to limited sequence coverage (26 bp) by the SuperSAGE data.
We have attempted to overcome the difficulties of polyploidy by using diploid apomictic accessions, but the analyses performed here still leave us with a similar conundrum when considering the effects of duplicated gene dosage versus hybridization on global gene expression patterns leading to apomeiosis. Both the ‘parent of origin’ effect described here and previous chromosomal analyses (Kantama et al., 2007) show that diploid apomictic Boechera are hybrid, and analyses of seed formation demonstrate that diploid and triploid apomictic Boechera generally produce unreduced gametes (Dobešet al., 2004b; Schranz et al., 2005; Voigt et al., 2007). Hence hybridization probably preceded polyploidization during the evolution of apomixis, as the latter would have been made possible by the unreduced gametes generated by diploid apomicts. The selective pressure to maintain homologous chromosome pairing during meiosis is, by definition, relaxed in asexual organisms, and this is reflected in chromosomal rearrangements (i.e. chromosomal heteromorphy) that typically characterize asexual genomes (Birky, 1996).
One possibility is that apomeiosis was first induced through hybridization, with subsequent establishment of asexuality leading to meiotic perturbations and the accumulation of chromosomal rearrangements (including duplications). Alternatively, the potential for expressing apomeiosis could be an older characteristic of the genus (i.e. pre-Pleistocene), as there is evidence for unreduced gamete formation in other Boechera species (Dobešet al., 2006). Hybridization is nonetheless a major factor that has played a significant role in the evolution of Boechera (Koch et al., 2003; Dobešet al., 2004a,b, 2006, 2007; Schranz et al., 2005), and hence inter-breeding between members of this genus could have led convergently to the expression of apomeiosis independently in different apomictic lineages. Analyses of apomeiosis expression in other apomictic Boechera are required in order to differentiate between these scenarios. Finally, the identification of differentially expressed species-specific mRNAs representing many genes involved in reproduction (Table S2) is consistent with the ‘hybridization-derived floral asynchrony’ hypothesis of Carman (1997).
The question remains whether one or more linkage groups are involved in duplications in the apomictic genome, as apomixis has been shown to be inherited in a Mendelian pattern in some taxa (Richards, 2003). The B chromosomes of apomictic Boechera (Sharbel et al., 2004) might harbour duplicated regions, although their DNA sequence homology with B. stricta (Sharbel et al., 2004; Kantama et al., 2007) is not consistent with the ‘parent of origin’ effect involving B. holboellii alleles [Figure 4(c)]. Furthermore, B chromosomes are not found in all apomictic accessions (Sharbel et al., 2005), and hence duplication could also entail autosomal regions. Phylogenetic analyses have shown B. holboellii and B. divaricarpa to have paraphyletic Pleistocene origins (Dobešet al., 2004a), and hence the ‘parent of origin’ effect may stem from taxa other than B. holboellii. Nevertheless, the ovule transcriptomes of apomictic Boechera lineages continue to show residual expression patterns reflective of hybridization after many generations of asexuality. Our analyses of diploid apomeiotic ovules have thus implicated both hybridization and gene duplication in the switch from sexual to apomictic seed formation, and have revealed unexpected levels of variation in the asexual genome.
Sample selection and ovule micro-dissection
We used a flow cytometric seed screen (Matzk et al., 2000) to analyse reproductive variability in four Boechera accessions (Table 1). Early flower and embryo sac development was compared to that of Arabidopsis (Smyth et al., 1990; Schneitz et al., 1995), and we selected ovules at megasporogenesis (between stages 2-II and 2-IV; Schneitz et al., 1995; differentiated megaspore mother cell, inner and outer integument initiated) in order to examine changes in gene expression associated with apomeiosis (Figure 1). The gynoecia of sexual and apomictic Boechera (Table 1) were dissected from non-pollinated flowers at the megasporogenesis stage in 0.55 m sterile mannitol solution, at a standardized time (between 8 am and 9 am) over multiple days. Micro-dissections were performed in a sterile laminar air flow cabinet using a stereoscopic microscope (1000 Stemi; Carl Zeiss, http://www.zeiss.com/) under 2× magnification. The gynoecium was held with forceps, and a sterile scalpel was used to cut longitudinally such that the halves of the silique together with the ovules were immediately exposed to mannitol. Individual live ovules were subsequently collected under an inverted microscope (Axiovert 200M; Carl Zeiss) under sterile conditions, using sterile glass needles (self-made using a Narishige PC-10 puller (Narishige Group, http://www.narishige-group.com), and bent to an angle of approximately 100°) to isolate the ovules from placental tissue. Using a glass capillary (with an opening of 150 μm interior diameter) interfaced to an Eppendorf Cell Tram Vario (http://www.eppendorf.de), the ovules were collected in sterile Eppendorf tubes containing 20 μl of RNA stabilizing buffer (RNAlater; Sigma, http://www.sigmaaldrich.com/). Ten ovules per tube and two samples per accession were collected in this way, frozen directly in liquid nitrogen and stored at −80°C.
RNA isolation, amplification and cDNA generation
RNA from micro-dissected ovule material was isolated using a PicoPure RNA isolation kit (Arcturus Bioscience, http://www.arctur.com) with several modifications. The standard lysis buffer was supplemented with 1% v/v NucleoGuard stock solution and 2 μl N-Carrier (AmpTec, http://www.amp-tec.com). An additional treatment with Turbo DNase (Ambion, http://www.ambion.com) treatment was included to eliminate any contaminating DNA. A second purification step was performed with RNeasy columns (Qiagen, http://www.qiagen.com/) to eliminate contaminating polysaccharides, proteins and the DNase enzyme. RNA integrity and quantity were verified on an Agilent 2100 Bioanalyzer using RNA Pico chips (Agilent Technologies, http://www.agilent.com).
Linear mRNA amplification was achieved using an ExpressArt mRNA amplification kit (AmpTec) with several modifications. Approximately 1 ng of total RNA was used in two independent reactions per sample to eliminate any random methodological effects. The two independent reactions were then pooled prior to cDNA conversion. RNA was converted to cDNA using an anchored oligo(dT)-T7-promoter primer with a mix of the AmpTec reverse transcriptase enzyme and ArrayScript reverse transcriptase (Ambion), and double-stranded cDNA was generated using the box-random-trinucleotide primer included in the kit. The resulting RNA after the first amplification round was purified using RNeasy MinElute columns (Qiagen), followed by a second and third amplification round. RNA integrity and quantity were verified on an Agilent 2100 Bioanalyzer using RNA Nano chips (Agilent). RNA quantity was determined using a Nanodrop ND-1000 spectrophotometer (Thermo Fisher Scientific, http://www.nanodrop.com). Subsequently, 5 μg of DNA-free amplified mRNA was converted into double-stranded cDNA using a 5′-biotinylated primer, and subsequently used for SuperSAGE generation.
SuperSAGE libraries were produced by GenXPro GmbH (Frankfurt am Main, Germany), essentially as described by Matsumura et al. (2006) with the following adaptations: instead of concatemerization and cloning of the ditags, these were directly sequenced using a GS 20 sequencer (454 Life Sciences, Roche, http://www.roche.com). All ditags consisting of the same tag combination were eliminated using the GenXProgram software, which also sorts and counts the 26 bp tags. All tags containing ‘Ns‘ were removed from the analysis, which has been shown to significantly increase the accuracy of GS-20 reads (Huse et al., 2007). The likelihood of differential expression of each SuperSAGE tag was calculated using normalized tag numbers with a correction for multiple tests (Audic and Claverie, 1997), as implemented by Robertson et al. (2007).
Preparation of normalized double-stranded cDNA for 454 transcriptome sequencing
RNA was isolated from pooled flower stages 1–12 (Smyth et al., 1990) from three diploid sexual plants (two B. holboellii and one B. stricta, accessions ES910-2-2K, 105.6-1K and B07261) and three apomictic plants (two diploid B. divaricarpa and one triploid B. divaricarpa, accessions 67.5-K, 300.6.1-1K and 218.2-2K) using TRIzol reagent (Invitrogen, http://www.invitrogen.com/) according to the manufacturer’s protocol. Any remaining genomic DNA contamination was removed by treatment with TURBO DNase (Ambion). The DNase enzyme was removed and the RNA was further purified using a RNeasy MinElute clean-up kit (Qiagen) according to the manufacturer’s protocol, and eluted in 20 μl of RNA storage solution (Ambion). Poly(A)+ mRNA was isolated using a Poly(A)Purist mRNA purification kit according to the manufacturer’s protocol (Ambion).
Two normalized, full-length-enriched cDNA libraries were generated using a SMART cDNA library construction kit (BD Clontech, http://www.clontech.com) and a Trimmer Direct cDNA normalization kit (Evrogen, http://www.evrogen.com), generally following the manufacturer’s protocol but with several modifications. In brief, 1 μg of poly(A)+ mRNA was used for each cDNA library generated. Reverse transcription was performed with a mixture of several reverse transcription enzymes for 1 h at 42°C and 90 min at 50°C. Each step of the normalization procedure was carefully monitored to avoid generation of artefacts, e.g. due to overcycling. The resulting full-length-enriched, normalized cDNAs were linearly amplified, resulting in a total of 15 μg of double-stranded cDNA. The double-stranded cDNA was digested using SfiI enzyme to remove most of the adaptor sequences for 454 sequencing, column-purified, concentrated, and sequenced using 454 technology (GS FLX, Agencourt Genomics Services, http://www.agencourt.com). Sequences were analysed using a standard Newbler assembly (454 Life Sciences).
Eight SuperSAGE tags that were differentially expressed between the apomictic and sexual accessions were selected. The tags were blasted against the 454 sequences of both apomictic and sexual transcriptome libraries to obtain corresponding gene sequences that were analysed using GeneMark (Besemer and Borodovsky, 2005) and GENSCAN (Burge and Karlin, 1997), and compared with Arabidopsis thaliana for prediction of the coding regions (Table S1). Predicted coding regions were used for PCR primer design using the Genome LabTM GeXP system (Beckman Coulter®, http://www.beckmancoulter.com) to generate a multiplex group corresponding to the eight transcripts, plus three pairs of primers for known housekeeping genes: eEF1Bα2 (NM_121956), ACT2 (NM_112764) and ACT11 (NM_112046) (Sturzenbaum and Kille, 2001).
Eleven sets of 20 ovules were isolated by tissue micro-dissection from two accessions (one sexual and one apomictic; Table 1) at two development stages (1I–1II and 2II–2IV, Schneitz et al., 1995) for RNA extraction. Following RNA extraction using a PicoPure RNA isolation kit (Arcturus Bioscience) and quantification on an Agilent 2100 Bioanalyzer using RNA Pico chips, multiplex cDNA synthesis and PCR reactions were performed using the Genome LabTM GeXP system according to the manufacturer’s instructions (Beckman Coulter). The multiplexed PCR products were analysed in a CEQ8000 sequencer using the GenomeLabTM GeXP genetic analysis system (Beckman Coulter). Relative peak area, as measured by the GeXP system, was compared between all genes. Of the housekeeping genes, ACT2 exhibited the least amount of variation across all samples, and hence was used to standardize expression levels of the eight chosen genes.
All SuperSAGE tag sequences for a given subset (four Boechera libraries, in silico SuperSAGE set of Arabidopsis) were compared, the number of positional nucleotide differences was counted, and the ratio of observed versus expected frequencies was plotted in a histogram for each data set. The expected random distribution of k bp differences in the data was calculated using the density function of the binomial distribution where n =22 (tag size minus size of the common restriction site) and is the probability of a nucleotide mismatch given the nucleotide frequency (B. divaricarpa3: A = 0.343, C = 0.169, G = 0.184, T = 0.304; B. divaricarpa8: A = 0.348, C = 0.165, G = 0.177, T = 0.310; B. stricta: A = 0.340, C = 0.154, G = 0.176, T = 0.330; B. holboellii: A = 0.346, C = 0.160, G = 0.174, T = 0.320; A. thaliana: A = 0.295, C = 0.178, G = 0.215, T = 0.312), and thus P = 0.728, 0.724, 0.721, 0.722 and 0.738 for B. divaricarpa3, B. divaricarpa8, B. stricta, B. holboellii and Arabidopsis, respectively. In order to obtain virtual A. thaliana SuperSAGE tags, the predicted transcripts data set (TAIR7_cdna_20070425) was downloaded from the TAIR ftp site (ftp://ftp.arabidopsis.org/). The tags were generated using a modified version of the Perl script published on the TAIR ftp site, which allows extraction of most downstream tags of 26 bp size with a 5′ recognition sequence CATG in a virtual SuperSAGE experiment.
All SuperSAGE tags were blasted to the 454 cDNA database obtained from sexual and apomictic Boechera using the following parameters (blastall -p blastn -m 8 -e 1 -W 7 -r 1 -q -1 –i). Boechera sequences representing significant hits were annotated by Blast2GO using default parameters and the ANNEX annotation augmentation function (version 2.3.1, Conesa and Gotz, 2008). The combined graph function of Blast2GO was used to generate pie charts of the functional annotation of groups of sequences (representing different SuperSAGE tag classes, Table 4) based on gene ontology (GO) categorization, and a Fisher’s exact test (Conesa and Gotz, 2008) was used to analyse Boechera 454 sequence groups corresponding to each SuperSAGE tag class (Table 4) for significant enrichment of particular GO classes. Due to the large numbers of sequences in the sex- and apomeiosis-specific data sets (Table 4), a 5% sequence filter value was used as a cut-off to generate combined graphs (Conesa and Gotz, 2008). No cut-off value was used for any other groups of sequences. Sequences were evaluated for their predicted involvement in biological processes, molecular functions, and cellular localization, and all data are presented at level 3 GO categorization (Table S2 and Figure S2).
We thank I. Schubert, J. Carman (Utah State University) and E. Schranz (Institute for Biodiversity and Ecosystem Dynamics, University of Amsterdam) for valuable comments on the manuscript. We thank the Apomixis Research Group of the Leibniz-Institut für Pflanzengenetik und Kulturpflanzenforschung for enlightening discussions and technical support, and T. Mitchell-Olds (Department of Biology, Duke University), B. Roy (Department of Biology, University of Oregon) and E. Schranz for seed material. This study was supported by additional funding to T.F.S. from the Leibniz-Institut für Pflanzengenetik und Kulturpflanzenforschung for the SuperSAGE analysis, in addition to a ‘Pakt für Forschung und Innovation 2007’ grant from the Leibniz Association.