Duplication and independent selection of cell-wall invertase genes GIF1 and OsCIN1 during rice evolution and domestication

Background Various evolutionary models have been proposed to interpret the fate of paralogous duplicates, which provides substrates on which evolution selection could act. In particular, domestication, as a special selection, has played important role in crop cultivation with divergence of many genes controlling important agronomic traits. Recent studies have indicated that a pair of duplicate genes was often sub-functionalized from their ancestral functions held by the parental genes. We previously demonstrated that the rice cell-wall invertase (CWI) gene GIF1 that plays an important role in the grain-filling process was most likely subjected to domestication selection in the promoter region. Here, we report that GIF1 and another CWI gene OsCIN1 constitute a pair of duplicate genes with differentiated expression and function through independent selection. Results Through synteny analysis, we show that GIF1 and another cell-wall invertase gene OsCIN1 were paralogues derived from a segmental duplication originated during genome duplication of grasses. Results based on analyses of population genetics and gene phylogenetic tree of 25 cultivars and 25 wild rice sequences demonstrated that OsCIN1 was also artificially selected during rice domestication with a fixed mutation in the coding region, in contrast to GIF1 that was selected in the promoter region. GIF1 and OsCIN1 have evolved into different expression patterns and probable different kinetics parameters of enzymatic activity with the latter displaying less enzymatic activity. Overexpression of GIF1 and OsCIN1 also resulted in different phenotypes, suggesting that OsCIN1 might regulate other unrecognized biological process. Conclusion How gene duplication and divergence contribute to genetic novelty and morphological adaptation has been an interesting issue to geneticists and biologists. Our discovery that the duplicated pair of GIF1 and OsCIN1 has experienced sub-functionalization implies that selection could act independently on each duplicate towards different functional specificity, which provides a vivid example for evolution of genetic novelties in a model crop. Our results also further support the established hypothesis that gene duplication with sub-functionalization could be one solution for genetic adaptive conflict.


Background
Gene duplication has long been recognized to be an important way to provide a substrate on which evolution acts. The classical models that predict the most possible fate of one of the duplicate genes is to degenerate to a pseudogene or get lost from the genome due to vagaries of chromosomal remodeling, locus deletion or point mutation [1][2][3][4][5]. A less frequent fate of the duplicate genes is to gain a new function (neo-functionalization) when the other copy still maintains its original function. However, recent studies have indicated that the newly duplicated genes are often sub-functionalized from their ancestral functions held by the parental genes [6][7][8]. The sub-functionalization model (also referred to as duplica-tion-degeneration-complementation model) explains that the duplicate genes are maintained in the genome relying on complementary degenerative changes in a pair of duplicate genes, such that the duplicate genes together retain the original functions of their single ancestor [1][2][3][4][5]9]. During this process, the expression domain shifting is the most common character of duplicate genes. As a consequence, the duplicates acquired sub-functionalization and then were less constrained by selection than the single ancestor, which had to maintain the capacity to fulfill all functions. Therefore, selection could act independently on each duplicate and increase the gene function specificity [10].
Sequence variation plays an essential role in functional renovation of genes, however, the relationship between DNA variation and functional consequence has been enigmatic for the vast majority of genes in plant and animal kingdoms, despite an increasing number of studies have been reported. Crop species and their wild relatives with available genome information are becoming fascinating subjects for study of correlation between cryptic genetic variation and functional evolution, because they have undergone rapid diversification under intense artificial selection [11][12][13][14]. Therefore, investigating crop domestication genes will shed meaningful light on genetic variation that drives cultivation adaptation [15]. Rice was used by human about 11,000 years ago [16,17]. It has been indicated that the divergence of indica and japonica predated rice domestication, suggesting that at least two genetically distinct gene pools of O. rufipogon were cultivated and subsequently domesticated [18][19][20]. During the long-term cultivation and domestication, tremendous diversity in rice has been selected by human, adapting to various ecosystems and agricultural management, in addition to high yielding characteristics, such as grain number and weight [12,17].
Various evolutionary models have been proposed to interpret the fate of paralogous duplicates, but little is known about the mechanisms of evolutionary change in duplicate genes leading to functional novelty. Rice has been recognized as a cereal model for such a study, and recent studies have discovered that some rice genes have undergone adaptive evolution under domestication selection [17,21]. We previously reported that the rice grainfilling gene GIF1 (OsCIN2) encoding a cell wall invertase (CWI) was most likely subjected to domestication selection [22]. Here, we report that GIF1 and another CWI gene OsCIN1 constitute a pair of duplicate genes with differentiated expression and function. Population genetic analysis showed that the two genes have experienced strong domestication selection, and interestingly, the target of selection in the GIF1 gene is the promoter region and that in OsCIN1 is the coding region.

Evolution of GIF1 and OsCIN1 by gene duplication
Rice genome has a CWI family consisting of eight members [23,24]. Our previous study has demonstrated that GIF1 is a member of the gene family and required for assimilated carbon partitioning during early grain-filling [22]. A phylogenetic analysis of the known plant CWI genes and predicted CWI genes from the recently released maize and sorghum genomes showed that OsCIN1, located on chromosome 2, is highly similar to GIF1 located on chromosome 4 ( Figure 1A). Genetic distance based on amino acid substitutions also indicated that OsCIN1 is most closely related to GIF1 (Additional file 1). To gain insight into their evolutionary relationship, the 500-kb flanking sequences of the GIF1 and OsCIN1 regions were compared. The other eight expressed genes flanking the GIF1 gene on chromosome 4 show good colinearity to the eight counterparts of the OsCIN1 region on chromosome 2 ( Figure 1B and Additional file 2). The result indicated that GIF1 and OsCIN1 rose via duplication of a genomic block, which could be as large as 15 Mb (data not shown). As shown in Figure 1A, phylogenetic analysis including cell-wall invertases of Zea mays, Sorghum bicolor, Lolium perenne, Hordeum vulgare, Dendrocalamopsis oldhamii and Oryza sativa showed that GIF1 was closer to cell-wall invertases of Zea mays, Hordeum vulgare and Dendrocalamopsis oldhamii, suggesting that this duplication might occur during the genome duplication of grasses [25]. By directly using synonymous substitution rate between the two paralogs (Ks = 0.57), and assuming the neutral evolutionary rate of rice genes (~6.5 × 10 -9 substitutions per silent site per year) [26,27], we estimated the time of duplication between GIF1 and OsCIN1 about 44 million years ago (MYA), a time much earlier than the genus Oryza diversified from a common ancestor about 15 MYA [28]. However, this estimated duplication age could be invalid because the regions were likely selected during rice domestication (see below).
To investigate the evidence for functional constraint on both copies at the DNA sequence level, we calculated Ka (non-synonymous substitution rate)/Ks ratios between GIF1, OsCIN1 and their homologs in maize, respectively [27]. The respective Ka/Ks value of GIF1 and OsCIN1 are 0.275 and 0.168 (p = 4.13E-24, p = 1.92E-50) ( Table 1), suggesting strong purifying selection.

Sub-functionalization of GIF1 and OsCIN1 by expression differentiation
Duplicate genes can be maintained by sub-functionalization (the duplicate genes perform different aspects of the original gene's function), or neo-functionalization (one of the genes acquires a novel function), and may facilitate adaptation to environmental change [6][7][8][9]. Our previous research has indicated that other CINs, including OsCIN1, are not functionally redundant to GIF1 [22]. Here we further compared the expression patterns of GIF1 and OsCIN1 in different tissues and grain-filling stages. GIF1 transcripts were detected in roots, elongating internodes, shoots and panicles, but not in leaves. In contrast, OsCIN1 was expressed strongly in leaves, but weakly in elongating internodes ( Figure 2A). During the early grain-filling stage, OsCIN1 transcript levels remained high while GIF1 transcript levels decreased after 15 days post-pollination (DAP) ( Figure 2B). In situ hybridization experiments further showed that the GIF1 transcript was only detected in the ovular vascular tissue but not in the pericarp and endosperm [22]; in contrast, the OsCIN1 transcript was detected in both the pericarp and endosperm [29]. Consistent with the difference in their expression pattern, GIF1 was induced in the caryopses supplied with sugars, but OsCIN1 was inducible in the leaves treated with sucrose and pathogen [23]. These results evidently showed that GIF1 and OsCIN1 have differentiated in expression pattern after duplication through altering expression patterns in development and response to environment cues.

Sub-functionalization of GIF1 and OsCIN1 enzymes
Total activity of cell-wall invertases was reduced to 17% of the wild-type in the gif1 mutant [22], indicating that GIF1 contributes to the majority of cell-wall invertase activity in early developing grains, although OsCIN1 was also expressed at a higher level in developing grains (Figure 2B). In support of this observation, the OsCIN1 T-DNA 'knockout' mutant did not show significant defect in grain filling and weight (J. -S. JEON, personal communication). To further determine the functional differentiation of the GIF1 and OsCIN1enzymes, we developed transgenic plants GIF1-OE [22] and CIN1-OE constitutively expressing GIF1 and OsCIN1 driven by the 35S promoter ( Figure 3A). GIF1-OE plants exhibited significantly higher CWI activity than that of CIN1-OE plants although the latter accumulated higher OsCIN1 levels ( Figure 3A and 3B). These results suggested that GIF1 and OsCIN1 could have different kinetics parameters such as Km and Vmax.
The difference in the kinetics parameters of enzymatic activity might result from the amino acid substitutions, in particular the GIF1 and OsCIN1 proteins contain Ala and Gly residues in the NDPNG domain (motif ), respectively ( Figure 3C). Phylogenetic reconstruction ( Figure 1A and 3C) revealed that the Gly-26-Ala substitution occurred after the GIF1 and OsCIN1 duplication. The crystal structure of the Arabidopsis CWI indicated that the conserved NDPNG domain is critical for CWI activity [30][31][32]. However, GIF1 and the homologous maize Mn1 contain a NDPNA motif instead of the NDPNG motif that presents in OsCIN1 and other CWIs, suggesting that the segmental duplication predates maize and rice differentiation. Furthermore, the crystal structure showed that Asp-239 interacted with Lys-242 and both the two amino acids played a crucial role in the transfructosylation process and interacted via H-bonds with the bound substrate [32]. A Thr-241-Arg substitution in the Asp-239/Lys-242 region occurred in GIF1 as well as in Mn1. It is noteworthy that the mutation in the Mn1 gene also caused shrunk grains [33]. The synteny between the GIF1 and Mn1 genome regions suggested that they could be orthologues (Additional file 3). These structure differences might contribute to different enzymatic kinetics of the GIF1 and OsCIN1 proteins. Together, these results suggested that GIF1 and OsCIN1 were subjected to sub-functionalization, or that GIF1 (probable Mn1 too) might have neo- functionalized, albeit we do not know the ancestral function of the CWIs.

Different phenotypes of GIF1-OE and CIN1-OE plants support sub-functionalization of GIF1 and OsCIN1
To further confirm functional divergence of GIF1 and OsCIN1, we determined the phenotypes of GIF1-OE and CIN1-OE plants. In addition to producing shrunken grains ( Figure 4A and 4C), GIF1-OE plants were also dwarfed in comparison with wild-type plants ( Figure 4E). By contrast, CIN1-OE plants did not exhibit any obvious phenotype in grain-filling and plant development ( Figure  4B and 4F). Instead, the CIN1-OE seeds exhibited marked preharvest sprouting in 6 of 10 transgenic lines tested, which expressed high OsCIN1 levels ( Figure 4D and data not shown), a phenomenon never occurring to the wild-type japonica control. These observations suggest that OsCIN1 might indirectly modulating hormone signaling pathways through interfering sugar metabolism in seed germination, leading to preharvest sprouting, since the sugar regulates rice alpha amylase (34). Together, our results demonstrate that GIF1 and OsCIN1 have evolved differentially or most likely sub-functionalized after duplication.

Evidence of OsCIN1 domestication-selection
In the previous study, we analyzed artificial selection using the segment sequences of GIF1. Here we further analyzed the 2-kb promoter region of GIF1 in 25 cultivars and 25 wild rice germplasm (AA genome) ( Table 2). We identified nine types of promoter sequences ( Figure 5A). According to the promoter sequences, nearly all culti-vated rice was classified into type 1, further supporting that the GIF1 promoter was artificially selected during rice domestication. We further sequenced two BACs containing respective OsCIN1 and GIF1 of the BB-genome of wild rice (O. punctata), and found that the coding region of OsCIN1 contains more variation than the coding region of GIF1 in comparison with the reference AA genome sequences (Additional file 4) [28,35], probably suggesting that OsCIN1 and GIF1 might have experienced different selection during Oryza evolution. To carefully investigate the evolution pattern of the OsCIN1 gene, we sequenced the OsCIN1 genome regions of the same set of 25 cultivars and 25 wild rice germplasm ( Table 2). Results  showed that the silent-site nucleotide, θπ, of OsCIN1 in japonica and indica were 0.0025 and 0.0038, respectively, lower than θπ (0.0097) in wild rice; and also much lower than the genome average θπ (0.0052 and 0.0073) in japonica and indica, respectively ( Figure 5C and Additional file 5). Furthermore, the Hudson-Kreitman-Aguade (HKA) test detected a highly significant deviation of OsCIN1 from neutrality for cultivated rice compared with the ADH1 gene (p = 5.89871E-11) [36], using O. punctata as an outgroup ( Table 3). The negative deviation in Tajima's D was also consistent with a selective sweep at the OsCIN1 locus in both japonica and indica, but no such a pattern was observed in wild rice ( Figure 5D and Additional file 5). These results suggest the OsCIN1 gene might also have been artificially selected. We also estimated genetic variation in upstream and downstream regions of OsCIN1 in the cultivars and wild rice genomes, and found that the region under selective sweep may extend as long as ~100-Kb. We further constructed a gene tree using 4.6-kb gene regions of OsCIN1 from the cultivars and wild rice ( Figure 6A). Consequently, all the japonica and indica accessions formed a cluster in the gene tree, in considerable contrast to a genome tree (Figure 6B) established based on SNP data [37,38], suggesting OsCIN1 introgression from one subspecies into another subspecies after domestication-selection, although our data could not rule out the possibility that OsCIN1 was extensively selected during rice domestication indepen-   dently in the two subspecies. All the results strongly support the hypothesis that OsCIN1 was selected during rice domestication. However, how the OsCIN1 gene has played a role in domestication is still unknown.
To narrow down the selection target in the OsCIN1 gene, we examined all the variations in the OsCIN1 genome regions, and found that an amino acid substitution (Arg-212-Leu) almost fixed in the rice cultivars, indicating that, unlike GIF1 which was selected in the promoter region, this site in the coding region could be the target of artificial selection in the OsCIN1 gene (Figure 5B). Further functional characterization of this site will provide more evidence to address how this site has contributed to OsCIN1 function in cultivated rice.

Gene duplication and adaptive conflict
Gene duplication plays a fundamental role in organism evolution by providing genetic materials from which novel functions can arise. Large numbers of duplicate genes were found in genomes, which contributed greatly to the genome structure and function evolution [1][2][3][4][5]39]. In general, the duplicate genes have two fates: first, the duplicate gene lost its function due to chromosome remodeling, deletion, and point mutation (known as nonfunctionalization); second, the duplicate gene retained for the maintenance of ancestral functions [1][2][3][4][5]. According to adaptive conflict model, adaptive mutations could be prohibited in the case of multifunctional genes, or one mutation that can optimize one function, may compromise the other functions, this mutation will be prohibited [6,40]. The adaptive conflict could be solved by sub-functionalization of duplicate genes. In this case, the duplicate genes would be less constrained and be able to evolve new functions under selection [41,42]. With this scenario, it is common that one gene could have multifunction in nature [40,43].

Independent selection of GIF1 and OsCIN1 mutations
The sequenced genomes of Oryza sativa, Arabidopsis thaliana and Populus trichocarpa all contain a family of cell wall invertases [44][45][46], some members of these gene families were reported to be involved in growth and development, disease resistance, stress responses and cell death, suggesting that the CWI gene families might have undergone sub-or neo-functionalized in these species.
Through genomic synteny analysis, we showed that GIF1 and OsCIN1 derived from a segmental duplication from an ancestor, most likely during genome duplication in grass species. After duplication, GIF1 and OsCIN1 have evolved to gain divergent functions with different expression patterns and enzymatic kinetics parameters through accumulating mutations in cultivated rice. In contrast to GIF1 on which domestication selection mainly occurred in the cis-regulatory region ( Figure 5A), the artificial selection occurred mainly in the coding region of OsCIN1 ( Figure 5B). Therefore, both GIF1 and OsCIN1 were most likely subjected to domestication selection, resulting in a cultivated GIF1 locus for better harvest, although the biological importance of OsCIN1 in domestication remains enigmatic. With this scenario, GIF1 and OsCIN1 may provide a good genetic model to demonstrate how duplicate genes could evolve and be artificially selected independently during crop domestication with divergent functions derived from accumulation of mutations in the regulatory and coding regions respectively, adding to those systems reported [47,48].

Differential biological functions of GIF1 and OsCIN1
GIF1 is mainly expressed in seed vascular tissues and controls sucrose unloading for starch synthesis at the early grain-filling stage [22]. Overexpression of the GIF1 gene produced plants with marked defects both in grainfilling and development, indicating that over-activity of the GIF1 enzyme disrupts sugar homeostasis, a process important to normal grain and plant development. In contrast, OsCIN1 has lower CWI activity compared to GIF1 in the transgenic plants ( Figure 3). Consistent with this, no obvious phenotype was observed in CIN1-OE plants except pre-harvest sprouting (Figure 4). Interestingly, OsCIN1 might be involved in pathogen defense and stress response [23]. It has been reported that sugars interact with signaling pathways mediated by phytohormones such as GA and ABA during seed germination and seedling development [34,49], which are also involved in stress responses. Preharvest sprouting of the CIN1-OE seeds may implicate a role for OsCIN1 in sugar-mediated alpha amylases activation [34]. However, detailed experiments are needed to dissect the OsCIN1 function.

Conclusion
Gene duplication and functional divergence contribute greatly to genetic novelty and adaptive evolution. However, molecular basis of selection and functionalization of duplicate genes remains largely unknown. Based on a set of data including population genetic analysis, fine sequencing of wild rice BACs, phenotyping of transgenic plants and analysis of gene expression and enzymatic activity, we provide a line of evidence that the two rice CWI genes GIF1 and OsCIN1 are a pair of duplicate genes and have been subjected to sub-functionalization during evolution or domestication selection. Therefore, duplicate genes could be independently selected towards different functional specificity, either on promoter for

Duplication and synteny analysis
The 500-kb radiuses of the GIF1 and OsCIN1 regions were scanned for homologous pairs. A homolog pair was defined as a single nr-KOME cDNA and its blastn homolog. A total of 18 homologous genes in both sides of the GIF1 and OsCIN1 loci were compared to establish linearity.

Sequencing and evolution analysis
To investigate the selective forces acting on GIF1 and OsCIN1 on the molecular evolution scale, we estimated the statistic Ka/Ks using the re-sequencing data (see below) and the maize Mn1 and Incw1-1 as the outgroup sequence, where Ka was the number of nonsynonymous substitutions per nonsynonymous site and Ks was the number of synonymous substitutions per synonymous site [27]. Ka/Ks values significantly less than 1 were often taken as evidence of constraint. The mean Ks of nine pair homolog genes, including GIF1 and other eight genes (Additional file 2), in the GIF1 and OsCIN1 regions were used to estimate the duplication time. Two BAC clones of O. punctata (BB genome) from the OMAP project http:// www.omap.org/ containing GIF1 and OsCIN1 respectively, were sequenced.

Analysis of OsCIN1 and GIF1 domestication
We deeply analyzed the OsCIN1 and GIF1 sequences from the re-sequenced genomes of 25 rice cultivars and 25 wild rice germplasm (Table 2), which has been done in Dr. Wen Wang's group, using Solexa technology (data not shown). Haplotype diversity was calculated for nucleotide diversity (π), and Tajima's D-statistics were calculated with DnaSP version 4.0. The gene tree was created using MEGA software [50]. The sequences then were aligned. The 2-kb up-/downstream genome sequences and the GIF1, OsCIN1 coding sequences were used for HKA test as described [36]. Sequences from wild rice O. punctata (BB genome) from the OMAP project http:// www.omap.org/ were used as outgroups for the HKA test. The DNA phylogenetic tree was constructed by neighbor-joining method using MEGA. The known or predicted CWI genes with high sequence similarity to GIF1 from Oryza sativa, Lolium perenne, Hordeum vulgare, Dendrocalamopsis oldhamii and the recently released Zea mays and Sorghum bicolor genomes were used in this study.

Development and growth of OsCIN1-OE transgenic plants
The full-length OsCIN1 coding sequence was PCRamplified from ZH11 cDNA by using the primers 5'-TCTAGTACAAAACAATGGGGACTC-3' and 5'-CGGAAAACCTCTTTATTATCTGTA-3'. The amplified fragment was subsequently cloned into the vector 35S-C1301 and transformed into ZH11 to generate 25 independent ectopic expression lines as described [22]. All transgenic materials were assayed in the second (T1) or third (T2) generations with 10-24 sibling plants grown in the paddy field to ensure agronomic traits.

Invertase activity assay
The caryopses were ground in the extraction buffer, and the extraction was centrifuged at 12,000 g for 10 min. The pellet was washed twice then re-suspended in the extraction buffer. Insoluble invertase activity was assayed as described [22].

RNA preparation and analysis
Total RNA was prepared from rice tissues using TRIzol reagent according to the manufacture's protocol (GIBCO BRL). For RT-PCR, 1-5 ug total RNA was used for the first-strand cDNA synthesis with the SuperScript III System (Invitrogen). RT-PCR analysis of GIF1 and OsCIN1 was performed with the primers [22,23].

Accession numbers
All sequences have been deposited in GenBank under accession numbers GU797900-GU798049.